03362-9781463961855.Pdf

International Monerary Fund IMFstaff papers

Volume 47 IMFAnnual ResearcConference 2001

©International Monetary Fund. Not for Redistribution EDITOR'S NOTE

The Editor invites from contributors outside the IMF brief comments (not more than 1,000 words) on published articles in IMF Staff Papers. These comments should be addressed to the Editor, who will forward them to the author of the original article for reply. Both the comments and the reply will be considered for publication. The data underlying articles published in IMF Staff Papers are available on the journal's website (http://www.imf.org/staffpapers). Readers are invited to use these data to expand on the material in the articles, and the journal will consider publishing such work.

This serial publication is catalogued as follows: International Monetary Fund IMF staff papers — International Monetary Fund. v. 1- Feb. 1950- [Washington] International Monetary Fund. v. tables, diagrs. 26 cm. Three no. a year, 1950-1977; four no. a year, 1978-1998; three no. a year, 1999-

Indexes: Vols. 1-27, 1950-80, 1 v. ISSN 1020-7635 = IMF staff papers — International Monetary Fund. 1. Foreign exchange—Periodicals. 2. Commerce—Periodicals. 3. Currency question—Periodicals. HG3810.15 332.082 53-35483

©International Monetary Fund. Not for Redistribution Inrernorionol Monetary Fund

Volume 47 Special Issue Contents 2001

International Macroeconomics: Beyond the Mundell-Fleming Model Maurice Obstfeld • 1

Do Monetary Handcuffs Restrain Leviathan? Fiscal Policy in Extreme Exchange Rate Regimes Antonio Fatas and Andrew K. Rose • 40

Exchange Rate Regimes and Economic Performance Eduardo Levy-Yeyati and Federico Sturzenegger • 62

The Interest Rate-Exchange Rate Nexus in Currency Crises Gabriela Basurto and Atish Ghosh • 99

Consumption and Income Inequality During the Transition to a Market Economy: Poland, 1985-1992 Michael Keane and Eswar Prasad • 121

Bail-Ins, Bailouts, and Borrowing Costs Barry Eichengreen and Ashoka Mody • 155

Crisis Resolution and Private Sector Adaptation Gabrielle Lipworth and Jens Nystedt • 188

Notes on the History of the Mundell-Fleming Model Robert Mundell • 215

©International Monetary Fund. Not for Redistribution This page intentionally left blank

Mundell-Fleming Lecture International Macroeconomics: Beyond the Mundell-Fleming Model

MAURICE OBSTFELD*

This paper presents a broad overview of postwar analytical thinking on international macroeconomics, culminating in a more detailed discussion of very recent progress. Along the way, it reviews important empirical evidence that has inspired alternative modeling approaches, as well as theoretical and policy considerations behind developments in the field. The most recent advances in model building center on the "new open-economy macroeconomics," which synthesizes Keynesian nominal rigidities, intertemporal approaches to open-economy dynamics, and the effects of market structure on international trade. [JEL F41, F33]

odern international macroeconomics progresses in two main ways. First, M techniques or paradigms developed in mainstream micro or macro theory have been applied in an international setting. Second, researchers probe more deeply, using both theoretical and empirical methods, into the classic issues that define international economics as a distinct field—the implications of sovereign governments and national monies, of partial or complete cross-border factor immobility, of transport costs and cross-border information asymmetries that

'Maurice Obstfeld is the Class of 1958 Professor of Economics at the University of California, Berkeley. This paper formed the basis of the Mundell-Fleming Lecture, presented by the author at the First Annual Research Conference of the International Monetary Fund, November 9—10, 2000, in Washington, D.C. The author thanks Jay Shambaugh for excellent research assistance, Charles Engel for providing data, and the National Science Foundation for its support. Robert Flood supplied helpful comments. The discussion draws extensively on joint work with Kenneth Rogoff. The author dedicates the lecture to the memory of Mundell's student Douglas D. Purvis, a friend, mentor, and role model to many.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld impede or even prevent trade. Frequently, prominent international policy problems, even crises, provide the inspiration for new explorations. Enduring contributions typically reflect both modes of progress. For example, an application of new techniques developed elsewhere to an international monetary problem may consist merely of relabeling "households" as "countries." But the most productive deployments of new techniques throw additional light on the specific problems of imperfect international economic integration at the heart of the field. In words that Robert A. Mundell (1968, p. Ill) used to describe the rise of pure trade theory, advances in general economics, when applied with skill and sense in settings that capture salient empirical features of international economic data, have allowed "constant refinement and extension" in open-economy macroeconomics, "permitting analytical developments surpassing the possible achievements of unaided intuition." The contributions of Mundell and of J. Marcus Fleming (1962) exemplify the most successful interactions of method and subject. No wonder this body of work has now been honored through the award to Mundell of the 1999 Nobel Prize in Economic Sciences. By merging Keynesian pricing assumptions and international market segmentation within a simple yet illuminating model, Mundell and Fleming provided the basic template for much subsequent research in both theory and policy.! This paper presents a broad overview of postwar analytical thinking on international macroeconomics, culminating in a more detailed discussion of very recent progress. Along the way, I will review some important empirical evidence that has inspired alternative modeling approaches, as well as theoretical and policy considerations behind developments in the field. My general topic is, in fact, too large and diverse for comprehensive coverage, given my limited space and scholarship. The account that follows therefore should be interpreted as my own, somewhat impressionistic, view. A testament to the lasting influence of their work is that much of the discussion can be framed with reference to what Fleming and especially Mundell accomplished in their work of the 1960s and 1970s—and left undone.2 The discussion is organized as follows. Section 1 focuses on postwar advances, discussing how, successively, the work of Mundell and Fleming, the monetary approach to the balance of payments, and the intertemporal approach to the current account placed international capital mobility and dynamics at center stage in open-economy macromodels. Section II is an overview of the behavior of international prices, and presents the strong evidence that prices are sticky in

'The most elaborate exposition-cum-interpretation of the Mundell-Fleming framework is offered by Frenkel and Razin (1987). 2The postwar period is coterminous with the history of the International Monetary Fund, and it is no accident that a number of important postwar intellectual developments—including contributions by Mundell and Fleming—originated in the research functions of the Fund. In this paper, however, I will not attempt systematically to review research done by the Fund or by any other institution. Retrospective eval- uations ot Fund research can be found in Blejer, Khan, and Masson (1995) and Polak (1995). I will focus on nuts-and-bolts modeling of open-economy macrostructure and policy effects, to the exclusion of the very interesting work done in recent years on regime switches, policy credibility, and the like. Even my discussion of macrostructure will be partial, because I have space only to mention in passing research on international asset pricing and on the different functions of international capital flows.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL nominal terms and that international markets for "tradable" goods remain highly segmented. Section III discusses newer modeling approaches that reconcile the breakthroughs in dynamic open-economy theory through the 1980s with the sticky-price setup pioneered by Mundell and Fleming. The section includes a rather detailed example of a stochastic "new open-economy macroeconomics" model in which the expenditure-switching effects of exchange rates central to the Mundell-Fleming model coexist with extreme market segmentation for tradable consumption goods and pricing to market. This class of models allows us to address rigorously, for the first time, a number of issues that have long been central to informal policy discussions in international macroeconomics. Section IV concludes.

I. Disequilibrium, Capital Mobility, Dynamics The classical paradigm that dominated international macroeconomic thinking until the First World War depicted a self-regulating global economy based on the gold standard. The classical world is a stable dynamical system in which adjustment of national price levels and the free flow of specie can be relied on to swiftly restore both full employment and equilibrium in national balances of payments. This picture of near-frictionless adjustment, with sovereign economic interventions limited by rules of the game, is a gross exaggeration even of the conditions existing under the Victorian Pax Britannica. But the classical paradigm, despite its positive shortcomings, serves as a useful theoretical benchmark. And it reminds us that dynamic considerations have been at the heart of international macroeconomics at least since the days of Gervaise and Hume. During the interwar period, extreme economic dislocations promoted an overly idyllic view of prewar conditions. These dislocations included nationalistic measures that sharply reduced economic integration among countries: fiat monies trading at floating exchange rates, preferential trade arrangements, direct state trading, default on foreign debts, and exchange controls. In earlier years economists had regarded such nationalistic experiments, when they occurred in Latin America or other areas of the periphery, with "fascinated disgust."3 Now, however, similar economic arrangements were painfully relevant among the advanced countries themselves. The classical paradigm had become largely irrelevant to actual international conditions. Moreover, it could not explain the economic cataclysm that had brought those conditions into being.

Keynesian Approaches: From Metzler and Machlup to Mundell The stage was set for Keynes's "revolution" and for the adaptation of its central ideas to international questions at the hands of Metzler, Machlup, and others. The new models they developed dealt with an essentially static world characterized by rigid wages and prices, unemployment, and limited financial linkages between countries. Key contributions elucidated the effects of trading relations

3The apt phrase is that of Bacha and Diaz-Alejandro (1982, p. 3).

©International Monetary Fund. Not for Redistribution Maurice Obstfeld on Keynesian multipliers, international repercussions, the effects of devaluation, the determination of floating exchange rates, and the role of the terms of trade in the Keynesian consumption function. The role of monetary factors, so central to the classical approach, was downplayed if not ignored. Metzler's (1948) survey for the American Economic Association was quite explicit in repudiating "the central role which [the classical mechanism] attributes to the monetary system" (p. 212). However, the new approach gave no guide to the alternative mechanisms that would eliminate external imbalances over time, and implicitly assumed instead that sterilization policies could be followed indefinitely.4 Meade's magisterial treatise The Balance of Payments (1951), published exactly a half-century ago, attempted to embed the Keynesian developments within a much broader framework that also embraced monetary factors. Meade's book is remarkable both in its ambitions and in its accomplishments. It summarized and (to a lesser extent) synthesized international monetary thinking as it had developed over the centuries. Meade sought to present a systematic account of economic problems and their solutions in an open economy, entertaining a wide variety of assumptions on price flexibility and international payments arrangements with the goal of guiding policy choice. In so doing, he discovered much that would be rediscovered later and set a large portion of the research agenda for the subsequent decades. Indeed, it became fashionable for a time to dismiss much of subsequent international economic research with the remark "It's all in Meade"—an assertion difficult to refute given the work's nearly impenetrable expository style. The forbidding facade, however, repelled would-be readers and greatly reduced its impact. Mundell (1968, p. 113) rated the work as a "landmark in the theory of international trade and economic theory in general," much underestimated by contemporary reviewers, but he lamented the "defects of its organization and presentation." Meade himself (1951, p. viii) pointed to one major gap in his theoretical treatment of international macroeconomics:

But I must confess frankly that there is one piece of modem technique in economic analysis which is very relevant to the problems discussed in this volume, but of which I have made no use. I refer to the analysis of the dynamic process of change from one position of equilibrium to another. The method employed in this volume is first to consider a number of countries in at least partial or temporary equilibrium, domestically and internationally; then to consider the new partial or temporary equilibrium which the economies will attain when the direct and indirect effects of the disturbing factor have fully worked themselves out; and finally to compare the new position of equilibrium with the old. In other words, this is a work not on dynamics, but on comparative statics, in economics.

Meade criticized the lack of any explicit mathematical account of how the economy gets from one "partial or temporary" equilibrium to another. A related problem, already noted above and highlighted by Meade's "partial or temporary"

4For further discussion, see Obstfeld (1987).

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL qualifier, was even more fundamental to much of the Keynesian theorizing of the 1940s and 1950s. Situations of external imbalance necessarily imply that stocks of domestic wealth—money and perhaps other assets—are not stationary and, thus, that the economy's temporary equilibrium must evolve over time, even in the absence of exogenous impulses. But what intrinsic dynamics would operate in a world seemingly at odds with the assumptions driving Hume's analysis of the economy's movement toward a long-run equilibrium? In particular, would this dynamical process be a stable one, and how would its nature depend on the activist economic policies that might be in play? Despite Alexander's important formulation of the absorption approach in the early 1950s (Alexander, 1952), an approach that brought to the fore the role of desired wealth changes in generating international imbalances, surprisingly little progress on these questions was made until Mundell's work in the early 1960s. In a pathbreaking series of articles, Mundell took up the challenge of filling the gap that Meade's omission of dynamics had left. By so doing, he reintroduced the idea of a self-regulating adjustment mechanism that had been central to the classical framework. In line with the evolution of world financial markets since Meade's book, Mundell put private international capital flows at center stage in his dynamic analysis. Had his achievement been entirely technical, it might have had little impact. Instead, through a rare combination of analytical power and Schumpeterian "vision," Mundell distilled from his mathematical formulations important lessons that permanently changed the way we think about the open economy. Mundell followed Meade in emphasizing the monetary sector, using a liquidity- preference theory of money demand to tie down the short-run equilibrium. Metzler (1968), in work done around the same time, took a similar tack, but he was less successful, whether his work is judged by its theoretical elegance or by immediate policy relevance. Fleming (1962), working in parallel, developed a model quite similar to Mundell's basic short-run equilibrium framework, and the two justly share credit for this contribution. Fleming did not, however, formally address the long-term adjustment process implicit in Keynesian models; he confined himself to some prescient remarks on the long- versus short-term responsiveness of the capital account. Mundell focused squarely on the dynamic effects of payments imbalances in his paper on "The International Disequilibrium System" (Mundell, 1961a). Even in a world of rigid prices, Mundell argued, an "income-specie-flow mechanism" analogous to Hume's price-specie-flow mechanism ensures long-run equilibrium in international payments. An increase in a country's money supply, for example, would depress its interest rate, raise spending, and open an external deficit that would be settled, in part, through money outflows. For a small economy, the end process would come only when the initial equilibrium had been reestablished. Mundell clarified the role of sterilization operations, showing that they can be at best a temporary response to permanent disturbances affecting the balance of payments. This work was influential in indicating the ubiquity of self-regulating mechanisms of international adjustment and, as a corollary, the limited scope for monetary policy with a fixed exchange rate, even under Keynesian conditions.5

5Of course, earlier writers, such as Keynes (1930, p. 309), had recognized some of the limitations that fixed exchange rates and capital mobility place on national monetary policies.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

While Mundell's "disequilibrium system" argument (1961a) showed how the income-specie-flow mechanism would restore balance of payments equilibrium under Keynesian conditions, the analysis did not delineate automatic forces tending to restore full employment (internal equilibrium). To address that issue, Mundell pursued the idea of a "policy mix" in which fiscal policy would play a central role. Mundell (1962) applied a dynamic approach to the joint use of monetary and fiscal policy to attain internal and external targets under a fixed exchange rate. He observed that thanks to capital mobility, policy dilemma situations that might arise under fixed exchange rates could be solved. Mundell showed that by gearing monetary policy to external balance (defined as a zero official settlements balance) and fiscal policy to internal balance (full employment), governments could avoid having to trade off internal against external goals in the short run. The key to his argument was the claim that either monetary or fiscal expansion can raise output, but they have opposite effects on interest rates. Thus, for example, a country simultaneously experiencing unemployment and an external deficit could couple fiscal expansion with monetary contraction in a way that lifts aggregate demand while attracting a sufficiently large capital inflow to close the foreign payments gap. Without capital mobility, however, this approach could not succeed. Mundell went on to argue that, when capital is mobile and the exchange rate pegged, a stable policy mix requires assigning fiscal policy to internal balance and monetary policy to external balance. A new and subtle insight in this work was that dynamic stability conditions might differ for alternative policy assignments and could therefore be used to assess the appropriateness of the policy mix. Related work by Mundell (1960) investigated the relative efficacy of fixed and flexible exchange rates in helping countries adjust to economic shocks. Mundell showed that the answer depended on government policy rules, the speed of domestic price-level adjustment in the face of excess or deficient demand, and the degree of capital mobility. Mundell's emphasis on the role of differential sector adjustment speeds in determining an economy's dynamic behavior proved influential in other contexts—for example, in Dornbusch's (1976) Mundellian model of exchange-rate overshooting. In one of his most celebrated contributions, Mundell (1963) took the speed of capital market adjustment to an extreme. With perfect capital mobility, he showed, only fiscal policy affects output under fixed exchange rates; monetary policy serves only to alter the level of international reserves. In contrast, fiscal policy might be dramatically weakened under floating rates. One implication of this analysis, which was not seen right away, was that the balance of payments might be a misleading indicator of external balance in a world where central banks could in principle borrow reserves in world capital markets. A more relevant concept of external balance would have to focus on the long-run solvency of the private and public sectors, taking into account vulnerabilities that might expose a country to a liquidity crisis. The Mundellian idea of the policy mix was a major conceptual advance and seemingly offered an elegant way to avoid unpleasant tradeoffs. But the approach had at least two theoretical drawbacks. First, Mundell's theoretical specification of the capital account as a flow function of interest rate levels (a formulation used by

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Fleming (1962) as well) was theoretically ad hoc. It implied, implausibly, that capital would flow at a uniform speed forever even in the face of a constant domestic-foreign interest differential. The second problem, already mentioned, was the definition of external balance in terms of official reserve flows, rather than in terms of attaining some satisfactory sustainable paths for domestic consumption and investment. As a medium-term proposition, it would be unattractive, perhaps even infeasible, to maintain balance of payments equilibrium through a permanently higher interest rate. The results of such a policy—crowding out of domestic investment and an ongoing buildup of external debt—would eventually call for a sharp drop in consumption.6 While Mundell's framework was perhaps useful for thinking about very short-run issues (such as the need to maintain adequate national liquidity), it failed completely to bridge the gap from the short run to the longer term. Indeed, the theory of the policy mix had little practical significance under Bretton Woods. In his detailed study of nine industrial countries' policies during the postwar period to the mid-1960s, Michaely (1971, p. 33) found only two episodes in which the prescription of the Mundellian policy mix was consistent with the official measures authorities actually took. Most of the time, Michaely concluded, fiscal policy simply was excluded from the list of available instruments.7

Classicism Redux: Monetary and Portfolio Approaches By the mid-1960s, Mundell's dissatisfaction with his own early rendition of monetary dynamics led him to pursue the monetary approach to the balance of payments. The approach is also associated, in differing forms, with Harry G. Johnson and with the IMF's Research Department under Jacques J. Polak (see Frenkel and Johnson, 1976, and International Monetary Fund, 1977).8 If one thought of the University of Chicago style of monetary approach as being primarily a retrogression to the classical paradigm, one might reckon its intellectual impact as being rather transitory. I think that conclusion would be wrong, however, and that the monetary approach in reality made three enduring contributions. Along with research of the late 1960s on closed-economy models of money and growth, it helped drive home to the profession key distinctions between stocks and flows in dynamic international macroeconomic analysis. Furthermore, it provided a set of consistent long-run models that, aside from their intrinsic theoretical interest, could serve as benchmarks for more realistic analyses. Finally, with its formal elegance and the extravagance of its claims, the monetary approach breathed new life and brought new blood into a field that was becoming a bit tired. I argued earlier that Mundell's treatment of capital flows in his work on the policy mix emphasized the monetary component of wealth at the expense of other

6Meade (1951, p. 104n) recognized clearly that in choosing between monetary and fiscal policy, "[tjhe question of the optimum rate of saving is involved." Mundell (1968) briefly discusses problems of the composition of the balance of payments, likening them in Chapter 10 to problems of the proper division of national product between consumption and saving. Purvis (1985) includes a nice discussion of fiscal deficits and the external debt burden from the perspectives of the Mundell-Fleming and subsequent models. 7For a more detailed discussion of practical problems in deploying fiscal policy, see Obstfeld (1993). 8For a perspective on alternative interpretations of the monetary approach, see Polak (2001).

©International Monetary Fund. Not for Redistribution Maurice Obstfeld forms, notably net external assets. As a result, the distinction between stock equilibrium in asset markets and flow equilibrium in output and factor markets was blurred. These failings would have appeared even more glaring had Mundell's models of the early 1960s been applied to longer-term issues, rather than to the short-run Keynesian stabilization questions for which they were designed. Mundell presumably intended to include his own earlier work when he remarked, in his "Barter Theory and the Monetary Mechanism of Adjustment" (Mundell, 1968, p. 112):

Innovations in the field since the 1930s have stressed the application of Keynesian economic concepts to the international sphere, rather than the integration of Keynesian international economics with classical barter theory or classical international monetary economics, creating a weakness in the area.

Indeed, the model Mundell developed here as a "start on the problem" did allow for sticky domestic prices, slowly adjusting to a carefully specified underlying classical equilibrium. Later work, by both Mundell and his students at Chicago, was to take the classical assumptions more literally (see, e.g., Mundell, 1971). Although most of the work of the monetary approach school oversimplified wildly in abstracting from assets other than money, a major element, as noted above, was the careful attention paid to the equilibrating role of output and factor prices in the transition from temporary to long-run equilibrium. While the approach lacked realism, it clarified the precise mechanics of Humean international adjustment, demonstrated the longer-run links between growth and the balance of payments, and showed how a focus on money supply and demand could help one to quickly ascertain the balance of payments effects of various disturbances. However, because of the ongoing growth of world financial markets, the appropriateness of relying on the balance of payments money account as an indicator of external balance was becoming increasingly questionable as the monetary approach developed, especially for industrial countries with reliable access to world capital markets. Models incorporating a broader spectrum of assets were being developed concurrently. The work of McKinnon and Gates (1966) is an early attempt along these lines, as is Chapter 9 of Mundell (1968). Tobin's (1969) seminal contribution to monetary theory set off a surge of research on multi-asset portfolio-balance models better equipped than those of the monetary school to describe international adjustment in a world of mobile capital. Foley and Sidrauski's (1971) elegant dynamic closed-economy rendition of the portfolio-balance approach provided a model for intertemporal applications in open-economy settings. The monetary and portfolio-balance approaches essentially merged in the mid-1970s, producing useful descriptive models of the long-run adjustment of monetary flows, current accounts, goods prices, and, somewhat later, floating exchange rates.9 Following

'For a more detailed discussion, see Obstfeld and Stockman (1985). A notable intellectual landmark here is the May 1976 conference issue of the Scandinavian Journal of Economics. The monetary approach to exchange rate determination is one strand in this literature; generally that approach builds on a version of the flexible-price monetary model in which the law of one price holds and money supplies rather than exchange rates are exogenous while exchange rates rather than money supplies adjust to equilibrate simultaneously the goods and asset markets. See, for example, the essays in Frenkel and Johnson (1978).

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Black's (1973) lead, many of these floating-rate models took from macroeconomics proper the then-novel modeling assumption of exchange rate expectations that are rational, that is, consistent with the economy's underlying structure (including the statistical distribution of the relevant exogenous forces). Of course, the rational expectations assumption also figured prominently in the many extensions of the Mundell-Fleming framework that continued to be fruitfully pursued, starting with Dornbusch's (1976) landmark "overshooting" version of Mundell- Fleming, which incorporated output-price, but not wealth, dynamics. These modern exchange rate models share a view of the exchange rate as an asset price (the relative price of two currencies), determined so as to induce investors willingly to hold existing outside stocks of the various assets available in the world economy. Mussa (1976) offers what is perhaps the classic exposition of this asset view of exchange rate determination. The enduring insight of the exchange rate's asset-price nature was obscured in versions of the Mundell- Fleming model that modeled the capital account analogously to the current account, as a flow function of the level of relative interest rates. By postulating that the exchange rate is determined by the condition of a zero net balance of payments, those models missed the exchange rate's role in reconciling stock demands and supplies that are normally orders of magnitude greater than balance of payments flows. Thus, the Mundell-Fleming model, in its earlier incarnations, offered no account of high exchange rate volatility.

Intel-temporal Approaches to the Current Account A final important branch in these dynamic developments was the application of optimal growth theory, in the style of Ramsey, Cass, and Koopmans, to open economies. Notable contributions along these lines were made early on by Bardhan (1967), Hamada (1969), and Bruno (1970). Building on these approaches in the early 1980s, a number of researchers developed an intertemporal approach to the current account in which saving and investment levels represent optimal forward-looking decisions.10 The new approach contrasted with the Keynesian approaches in which net exports are determined largely by current relative income levels and net foreign interest payments are, for the most part, ignored. These new models, unlike the earlier open- economy growth models, were applied to throw light on short-run dynamic issues—such as the dynamic effects of temporary and permanent terms-of-trade shocks—and not just the transition to a long-run balanced growth path. They could also be used to think rigorously about the policy implications of national and government intertemporal budget constraints. The intertemporal approach, unlike the Keynesian or monetary approaches, provided a conceptual framework appropriate for thinking about the important and interrelated policy issues of external balance, external sustainability, and equilibrium real exchange rates (for a recent example, see Montiel, 1999). All of these concepts are intimately connected with the intertemporal tradeoffs that an

10For a more extensive survey of the area, see Obstfeld and Rogoff (1995a).

©International Monetary Fund. Not for Redistribution Maurice Obstfeld economy faces. Another major advantage of the intertemporal approach was its promise of a systematic welfare analysis of policies in open economies—an analysis on a par, in rigor, with those already applied routinely to intertemporal tax questions. The approach shifts attention from automatic adjustment mechanisms and dynamic stability considerations to intertemporal budget constraints and transversality conditions for maximization, although those perspectives may well, of course, be mutually consistent.

II. Models vs. Reality Like the monetary approaches to the balance of payments and to exchange rates, the intertemporal mode of current account analysis developed in the 1980s generally assumed perfectly flexible domestic prices. It thereby abstracted from short- run price rigidities and the concomitant disequilibriums in goods and factor markets. The issues at the heart of the Mundell-Fleming model and its successors, such as the Dornbusch model, were simply put aside. The monetary and intertemporal models also, in general, assumed a rather high degree of economic integration among the economies being modeled. Presumably, a high level of integration among economies might justify abstraction from price rigidities, because these could not survive long in the face of international goods arbitrage. Was this level of abstraction justified? In a well-known paper published after nearly a decade of floating exchange rates, McKinnon (1981) argued that the world economy had moved far away from the "insular" economic pattern of the 1950s and 1960s, in which countries carried out some foreign trade but were otherwise largely closed to external influences. In the new world of more open economies, the earlier Keynesian (and elasticities) approach to open-economy macroeconomic questions had become outmoded. Instead, the type of assumption underlying the monetary approach—highly open economies with goods and capital markets open to the forces of international competition—was a better approximation of reality.'' Twenty years farther on in the process of postwar globalization, shouldn't the McKinnon arguments, if valid when they were advanced, carry even greater force?

Evidence on Insularity The answer that seems to come resoundingly from the data is "no." Even today, the world's large industrial economies (along with many smaller economies) remain surprisingly insular, to use McKinnon's term. McKinnon rightly identified a trend of increasing openness, but jumped the gun in declaring the age of insularity to be over. There are several well-known manifestations of persistent insularity. For many goods, transport costs are sufficiently high that a large proportion of GDP can be

"McKinnon differed from the monetary approach in questioning the existence of a stable national money demand function, in the absence of which, he argued, fixed exchange rates were preferable to floating rates.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL considered effectively nontradable. But as Meade (1951, p. 232) pointed out in an even more insular era, nontradability is only an extreme consequence of trade costs, which attach to all goods in varying degrees:

Products range with almost continuous variation between those for which the cost of transport is negligibly low in relation to their value and those for which the costs of transport are so high as to be in all imaginable circumstances prohibitive.

One consequence of pervasive transport costs (and other costs of trade such as official impediments) is the existence of a fairly sizable "transfer effect" due to changes in countries' net foreign assets.12 This transfer effect can be seen both in countries' terms of trade and in their real exchange rates. Regarding the latter, Lane and Milesi-Ferretti (2000), using the most comprehensive cross-country panel on national net foreign asset positions developed to date, estimate that a 50 percent of GDP fall in a country's net foreign asset position (corresponding, perhaps, to a 2 percent of GDP fall in the long-run current account balance) would be associated with a 16 percent real currency depreciation. They find that this effect is potentially much larger for bigger and less open economies. In addition, we see fairly large home biases in consumption and trade. Of course, standard versions of the Mundell-Fleming model assume a transfer effect arising from (generally unexplained) home consumption preference. We also observe puzzling symptoms of capital-market segmentation, notably the Feldstein-Horioka cross-section correlation of saving and investment and the home bias in equity portfolios. While imperfections intrinsic to capital markets lie in part behind these capital-market puzzles, costs of trade in goods can go a long way in generating "insular" capital-market behavior, as argued by Obstfeld and Rogoff (2000b). Some of the most dramatic evidence on insularity—certainly the hardest to rationalize in terms of internationally divergent consumer tastes or other devices—is the evidence on international price discrepancies for supposedly tradable goods. Because of all the evidence on segmentation, which on international prices bears most directly on the appropriateness of competing open macromodels and on the role of the exchange rate, I will focus on price behavior for the balance of this section.

Evidence on International Pricing A large body of empirical work weighs in against the proposition that international goods-market arbitrage is effective in quickly eliminating price differentials. The most convincing evidence began accumulating in the 1970s. Not coincidentally, the timing coincides with the availability of evidence on international pricing relationships under floating exchange rates. Isard's (1977) classic study was one of the first to present evidence against the Law of One Price (LOOP), showing that the

l2See Mundell (1991) for a discussion of theory relating to the transfer problem. As Mundell points out, even with nontradable goods, there need be no transfer problem—see also Chapter 4 in Obstfeld and Rogoff (1996) for a model along these lines. But the conditions for this result are very stringent and unrealistic.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld common currency unit values for similar tradable goods categories typically have diverged widely. These types of results continue to be confirmed in later data.13 Therefore, at the micro level, there is evidence that either arbitrage is inoperative or it operates with significant lags. At the macro level, some of the most striking evidence on international prices was assembled in an important paper by Mussa (1986). Mussa documented that, systematically and across a wide range of industrial country episodes, real exchange rates become much more variable when the nominal exchange rate is allowed to float. Real exchange rate variability tends to be almost a perfect reflec- tion of nominal rate variability, with changes in the two rates highly correlated and independent movements in price levels playing a minor, if any, role. The over- whelming evidence that a key real relative price depends systematically on the monetary regime amounts to a powerful demonstration that domestic price levels are quite sticky. Some of the regime changes Mussa examined—such as Ireland's switch from its currency board sterling link to the European Monetary System— amount virtually to natural experiments, and it therefore is hard to argue that the regime switches themselves are endogenous responses to underlying shifts in the volatility of international prices. When the exchange rate is cut loose, its variability is vastly accentuated while nominal goods prices continue to move sluggishly. The sole exceptions seem to be episodes in which domestic inflation is extremely high, in which case prices, like the exchange rate, come unhinged and the correlation between real and nominal exchange rate changes drops (see Obstfeld, 1998). The Mussa results also have a bearing on hypotheses about cross-border goods market integration. If arbitrage works strongly to keep international goods prices in line, then very large real exchange rate fluctuations would be ruled out. At the very least, large fluctuations would be rather temporary, as profit-maximizing traders quickly move to reap extra-normal profits, driving real exchange rates back into line. In fact, the evidence contradicts this idea as well. As shown by many studies (see Rogoff, 1996, for a survey), real exchange rate movements are highly persistent, so much so that it has been difficult to reject the statistical hypothesis that real exchange rate processes contain unit roots—a violation of even a weak form of long-run purchasing power parity that allows for deterministic trends. The best current estimates of real exchange rate persistence suggest that under floating nominal exchange rate regimes, the half-lives of real exchange rate shocks range from 2 to 4.5 years. Such macrolevel evidence may not be viewed as entirely convincing. For one thing, real exchange rates are calculated with respect to the overall CPI including nontradables (or some other comparably broad price index). Thus, it could be the case that there is actually a considerable degree of arbitrage among the most tradable goods entering these price indexes. Furthermore, there is the theoretical possibility that much of the persistence in real exchange rates is due to real shocks, shocks that alter international relative prices permanently without necessarily creating opportunities for arbitrage. Work based on disaggregated price data for CPI compo-

13For more complete references, see the valuable surveys by Rogoff (1996) and Goldberg and Knetter (1997).

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL nents, however, suggests that both of these hypotheses are not very relevant. As Engel (1993) shows, Mussa-style results also hold for international comparisons of the consumer prices of similar tradable goods. Regarding real shocks as a source of persistence in real exchange rates, it is much harder to argue that real shocks are responsible for changes in the relative international prices of very similar, supposedly tradable goods. Yet movements in these relative prices are just as persistent as movements in real exchange rates. Rogers and Jenkins (1995), for example, perform unit root tests on the relative United States-Canada prices of 54 narrowly defined categories of goods and services. They find that they can reject the null hypothesis of a unit root at the 10 percent level for only 8 of the 54 categories (all food products). But even for food products, not all relative prices seem to be detectably mean reverting. Obstfeld and Taylor (1997) document similar persistence at a disaggregated level. This, it seems to me, is persuasive evidence against the theory that real shocks are the main source of real exchange rate persistence. What recurrent real shocks do we imagine would be buffeting the relative supermarket price of Canadian and U.S. flour? The weight of evidence suggests that whatever factors allow the LOOP deviations documented by Isard (1977) and many others also lie behind the big and persistent swings in countries' real exchange rates.

Accounting for Real Exchange Rate Changes In recent important work, Engel (1999a) offers a striking way of illustrating just how pervasive the segmentation between countries' consumer markets is. Engel suggests decomposing a CPI real exchange rate change into (1) the component due to changes in international differences in two countries' relative prices of nontradable to tradable goods and (2) the component due to changes in the countries' relative consumer price of tradables.14 Under classic theories of real exchange rate determination, such as the Harrod-Balassa-Samuelson account (see Obstfeld and Rogoff, 1996, chap. 4), tradables as a category closely obey the LOOP and so all variability in real exchange rates is attributable to factor (1) above. Engel's work shows that the opposite is true for real exchange rates against the United States. Even for real exchange rate changes over fairly long horizons, nearly all variability can be attributed to component (2), the relative consumer prices of tradables. This is a striking contradiction of the Harrod-Balassa-Samuelson theory. International divergences in the relative consumer price of "tradables" are so huge that the theoretical distinction between supposedly arbitraged tradables prices and completely sheltered nontradables prices offers little or no help in understanding U.S. real exchange rate movements, even at long horizons. Apparently, consumer markets for tradables are just about as segmented internationally as consumer markets for nontradables.15

!4As Engel (1999a) explains, this is not an orthogonal decomposition. However, I do not believe that issue is central in interpreting his results, and I henceforth follow Engel in leaving it aside. Rogers and Jenkins (1995) earlier provided extensive evidence pointing to the same kinds of results that Engel emphasizes. 15Of course, the results are subject to the caveat made above that tradability is a matter of degree. Notwithstanding that fact, one can still view them as evidence that goods commonly considered to be "tradable" due to relatively low transport costs appear to be not very tradable at all if they are defined to include the services that bring them from the point of production to the consumer.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

Figure 1. United States and Japan, 1973-95

Engel's findings can be summarized by graphs such as Figures 1-4. These figures are based on monthly 1973-95 data from Engel (1999a, sec. I). They show bilateral comparisons based on three floating exchange rates, dollar/yen, dollar/French franc, and deutsche mark/Canadian dollar. Let e denote the nominal exchange rate, P the overall CPI, and PT and PN, respectively, the CPIs for tradables and nontradables.16 In Figures 1-4, the behavior of the overall CPI-based real exchange rate, Q = eP*/P, is compared with that of the tradable and nontradable real exchange rates, £P^/PTand eP^/PN. Each panel shows a plot against time, /, of the correlation between /-period percentage changes in the two variables shown. As suggested by Engel's findings, the data reveal no significant difference between short-term and long-term correlations, consistent with the hypothesis that mean reversion in the relative international price of consumer tradables is extremely slow. Strikingly, it seems not to matter much whether tradables or nontradables are used to compute real exchange rates. Indeed, it is remarkable that relative tradables prices consistently tend to display a higher correlation with real exchange rates than do relative nontradables prices, though the discrepancies are small and statistically insignificant. For the countries shown, the behavior of relative international tradables prices is so similar to that of relative international nontradables prices, even for horizons out to five years, that one is led to question whether the distinction is even meaningful in discussing how real exchange rates behave under floating. At the consumer level, one sees no evidence here of greater cross-border price coherence for tradables.

16Engel (1999a, app. A) outlines his methodology for constructing the CPI subindexes. Engel's tradables comprise the OECD's "all goods less food" and "food," his nontradables, "services less rent," and "rent."

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Figure 2. United States and France, 1973-95

Figure 3. Germany and Canada, 1973-95

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

Figure 4. Germany and Japan, 1973-95

High exchange rate volatility is what makes the preceding results so dramatic. Given the sluggish behavior of nominal consumer prices and the lack of operational arbitrage between consumer markets, the behavior of nominal exchange rates dominates these data. The role of exchange-rate volatility in masking domestic relative price movements is illustrated in Figures 5-7, which use the monthly 1972-97 data from section IV of Engel (1999a), in which the producer price index (PPI) is used to proxy the price of tradables. The figures are constructed as follows. For horizons t - 1 through t = 18, I average the mean squared error (MSB) of the 18 /'-period changes in the "traded goods" component of the CPI, log(e x PPI*IPPI), each expressed as a fraction of the MSB of the f-period change in log Q. The resulting ratios, measured by the vertical axes of Figures 5-7, are plotted against the standard deviation of month-to-month changes in log £, which can be read off the horizontal axis. A more nuanced picture now emerges. Figure 5 shows that, in pairings against the United States, Canada is an outlier, showing both the lowest nominal exchange rate variability and the lowest share of real exchange rate variability explained by tradables, just under 70 percent.17 When we examine all non-U.S. pairings in Figure 6, however, we see that there are other instances in which low nominal exchange rate variability is associated with relatively low shares of tradables in real exchange rate variability, shares that can be as low as 50 percent. Putting all the pairings together in Figure 7, U.S./Canada no longer appears as an outlier. Indeed, when nominal exchange rate volatility is suppressed, factors other than changes in the relative inter-

17The countries paired with the United States in Figure 5 are Austria, Canada, Denmark, Finland, Germany, Greece, Japan, the Netherlands, Spain, Sweden, Switzerland, and the United Kingdom.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Figure 5. Pairings Against the United States

Figure 6. Non-U.S. Pairings

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

Figure 7. All Industrial Country Pairings

national price of consumer tradables appear to play a significant role in determining real exchange rates. They are no longer swamped by huge nominal exchange rate changes, as in Figures l^-.18 Figures 5-7 nonetheless confirm the main conclusions we have already reached. Even for fixed exchange rate pairings such as Austria/Germany, relative tradables prices still play a very big role (over 50 percent) in explaining real exchange rate changes. This number still suggests considerable market segmentation at the consumer level. That nominal exchange rate changes alone seemingly can induce large, persistent changes in relative tradables prices, without setting into motion any prompt, noticeable arbitrage mechanisms, is a further indication of segmentation. Finally, the fact that nominal volatility plays so strong a role in driving relative international price changes is more evidence of stickiness in domestic nominal prices. Indeed, the data behave as if all consumer prices are sticky in domestic currency terms, even the prices of goods that may be imported from abroad.

Direct Evidence on Pricing to Market The preceding evidence leads to the finding of both sticky domestic prices for consumer goods and high barriers between countries' consumption-goods markets. Assumptions on consumer prices are key components of any macromodel

18These types of results also occur in the bilateral comparisons carried out by Rogers and Jenkins (1995), but they do not systematically compare the correlation between nominal volatility and the importance of LOOP deviations in determining real exchange rates.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL because consumer decisions are based on those prices and will influence the economy's response to disturbances. Because the ultimate consumer is several steps removed from the port of entry of import goods, however, findings such as Engel's (1999a) have only an indirect bearing on the height of barriers to international trade between firms, which accounts for most of international trade. Consumer prices, in addition to incorporating the import prices of the underlying commodities purchased, also reflect the costs of nontradable components such as retailing costs, internal shipping, and the like. These nontraded inputs make it difficult to assess the specific role of barriers to international trade in the underlying "tradable" commodities. They also make it harder to draw conclusions about the stickiness of import prices. Further ambiguities come from the impossibility of ensuring that even disaggregated CPI data from different countries represent in any sense the prices of identical goods. Not only does index composition differ across countries with respect to subcategories of the tradable, there is also no way to ensure that even very similar tradables included in two indexes originate from the same producer. Recent evidence on "pricing to market" (PTM) in international trade has helped to resolve these difficulties. The term PTM, popularized by Krugman (1987), refers to third-degree price discrimination by an exporter, based on the location of importers.19 Using detailed disaggregated data on the different prices that individual firms set for the same good exported to different locations, the econometrician can control for unobserved marginal production cost while ascertaining how exchange rate changes relative to different destinations affect prices charged. Under perfectly competitive conditions and costless international trade, an exporter's home currency price is determined entirely by domestic marginal cost; thus, given marginal cost, the exporter will raise price in any destination country fully in proportion to an appreciation in the source country's nominal exchange rate against the destination country. In other words, the extent of exchange rate pass- through is unitary. When pass-through is incomplete, however, there is evidence of PTM. The extent of PTM differs across source countries and products, of course. In their excellent survey of studies on international pricing, Goldberg and Knetter (1997, p. 1244) conclude that "a price response equal to one-half the exchange rate change would be near the middle of the distribution of estimated responses for shipments to the U.S." The markup adjustment following an exchange rate change, according to them, generally occurs within a year.

''Another seminal contribution is that of Dornbusch (1987). Under first-degree price discrimination, a monopolistic producer can perfectly exploit the heterogeneity among consumers by quoting each individual buyer a different price (of course, resale must be prevented). Under second-degree price discrimination, the monopolist does not observe any signal of consumer type and can exploit heterogeneity only by inducing self-selection on the part of consumers. Under third-degree price discrimination, the monopolist bases price on an observable signal about consumer type—in the present context, national location. (For further discussion see Tirole, 1988, p. 135.) As under first-degree price discrimination, trade impediments must rule out arbitrage between different consumer types. That is, pricing to market cannot take place unless there is international market segmentation.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

On the Need for a Synthesis

Perhaps the most salient feature of the data is that nominal exchange rate changes are associated with virtually commensurate fluctuations in real exchange rates. Moreover, relative international consumer prices of even tradable manufactured goods seem to move virtually one-for-one with the nominal exchange rate. At the level of consumers, the pass-through from exchange rates to import prices is virtually zero in the short run. Further up the distribution chain, where imports first enter a country, the pass-through of exchange rate changes to prices generally is positive, but substantially below one. Market segmentation allows exporters to change destination-specific markups in response to exchange rate movements. The accumulation of such evidence highlighted how some basic assumptions shared by the monetary and intertemporal approaches contradict central policy- relevant facts. Both approaches fail to incorporate short-run price rigidities, and both assume complete pass-through of exchange rate changes to import prices. Despite building in nominal rigidities, the Mundell-Fleming model, taken literally, also assumes unitary pass-through from the exchange rate to import prices. That assumption is central to the expenditure-switching effect of exchange rates, which is the key feature giving monetary policy its efficacy under flexible exchange rates, or allowing exchange rate changes to counter nation-specific shocks in Mundell's (1961b) optimum currency area analysis. The empirical failures of the standard models developed before 1990 were crippling not only from a positive point of view, but also because they reduced applicability to questions of policy. The Mundell-Fleming model, for example, makes no distinction between retail and wholesale import prices, and it would have to be modified to capture both the near-zero pass-through of the exchange rate to consumer prices of imported goods and the partial (but positive) pass- through of the exchange rate to the prices exporters charge. The precise way this is done, however, has critical policy implications. If one abstracts from partial exchange-rate pass-through at the wholesale level, as Engel (2000) does, then the expenditure-switching effect of the exchange rate disappears entirely: exchange rate shifts do not alter the prices consumers face, and there is no firm-to-firm trade. On the other hand, if partial pass-through by exporters does affect economic decisions, the expenditure-switching effect is present, albeit perhaps in a muted form. As another example, its failure to incorporate price rigidities dramatically reduces the policy utility of the intertemporal approach (except, perhaps, for very long-run issues). How can one accurately evaluate an economy's departure from external balance without some notion of its distance from internal balance—that is, a state of zero output gap and near-target inflation? How can one evaluate the equilibrium real exchange rate without some notion of the kind and height of the barriers separating national markets? How can one conduct welfare analysis without an attempt to model the economy's distortions, including those due to price rigidity? By the late 1980s and early 1990s, the clash of theory and data suggested the need for a coherent synthesis of the Keynesian and intertemporal approaches within an empirically oriented framework.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

III. The New Open-Economy Macroeconomics

The most recent synthesis of earlier approaches combines monopolistic producers with nominal rigidities in a dynamic context with forward-looking economic actors. As proponents of the New Keynesian approach to closed-economy macroeconomics argued in the late 1980s, monopoly is a natural assumption in any context with price setting, and imperfect competition helps rationalize the idea that output is demand determined over some range, because price exceeds marginal cost in the absence of unexpected shocks. Moreover, the literature on sunk costs and hysteresis makes it plausible that market power should play some role in explaining the persistence of international relative price movements.20 Svensson and van Wijnbergen (1989) provided an early prototype model along these lines, in which output is produced at zero marginal cost up to some limit and asset markets, as an aid to modeling, are assumed to be complete. Obstfeld and Rogoff (1995b, 1996, 1998, 2000a) proposed a more tractable intertemporal monopolistic competition framework with sticky (that is, preset) nominal output prices and incomplete asset markets. The new approach addresses the positive questions asked of the Mundell-Fleming model as well as normative questions that could previously be addressed only within the empirically incomplete intertemporal approach. There have been numerous extensions, which I do not have the space to discuss in any systematic way. Luckily, Lane (2001) has provided a comprehensive survey.

Price Setting in New Open-Economy Models One of the most important extensions has been to allow for PTM, in contrast to the original Obstfeld-Rogoff model, which assumed complete pass-through of exchange rates to import prices. Belts and Devereux (1996) made the initial contribution along these lines. In an open economy, nominal price rigidities can take a variety of forms because producers can choose to preset product prices in domestic or foreign currency. In the Betts-Devereux setup, some producers preset home prices in home currency and export prices in foreign currency—what Devereux (1997) refers to as "local currency pricing" (LCP)—although home and foreign markups over marginal cost are initially the same. After a shock is realized, however, the home and foreign prices of the good can diverge widely as the exchange rate moves, with international market segmentation preventing arbitrage by consumers. This approach implies zero exchange rate pass-through in the short run for PTM goods and complete pass-through for goods priced in the exporter's currency. It is only one of several possible ways to model PTM, which of course can occur in a flexible-price world. Because the Betts-Devereux setup (like that of Obstfeld and Rogoff, 1995b) assumes constant ex ante markups, it is not amenable to an analysis of possible ex ante variability on markups.21

2('Cheung, Chinn, and Fujii (1999) present some industry-level evidence supporting this idea. 2lThe assumption of constant markups also precludes an analysis of the choice of invoice currency, which of course should be endogenous. See Friberg (1998).

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

The PTM-LCP framework with constant demand elasticities has been widely adopted, however, because it provides tractability in model-solving while repro- ducing (for PTM goods) the virtually proportional effect of nominal exchange rate changes on the relative international prices of similar traded goods. Furthermore, the assumption of market segmentation at the consumer level certainly rings true and helps to rationalize why even big exchange rate changes seem to have so little impact on the economy in the short run (see Obstfeld and Rogoff, 2000b, on the "exchange rate disconnect" puzzle). Recent PTM models assume that all imports are priced in the local currency. Leading examples in this literature include Devereux and Engel (1998, 2000), who adapt the stochastic models in Obstfeld and Rogoff (1998, 2000a) to a PTM- LCP environment with complete nominal asset markets, and Kollmann (2001), Chari, Kehoe, and McGrattan (2000), and Bergin and Feenstra (2001), who incorporate dynamic Calvo (1983) nominal contracts and investigate the ability of the resulting models to replicate a number of business cycle covariances.22 These models, however, while capturing the apparent zero pass-through of exchange rates to retail prices, do not reflect the partial pass-through to wholesale import prices implied by the micro studies on pass-through. Indeed, taken literally—as models in which the prices consumers pay for imported goods correspond to the import prices used to define a country's terms of trade—these PTM-LCP models imply that when a country's currency depreciates, its terms of trade improve. The reason is simply that import prices are given in domestic currency, whereas export prices, which are rigid in foreign-currency terms, rise in home- currency terms when the home currency depreciates. That prediction is wildly at odds with the data, as Obstfeld and Rogoff (2000a) show. The discrepancy is illustrated in Figures 8-11, which plot 12-month percent changes in bilateral exchange rates against 12-month changes in relative export prices, using monthly 1974-98 data on four country pairs. The relative export price is defined as eP^/Px, an increase in which reflects an increase in the home country's overall export competitiveness, and if export prices are somewhat sticky in the exporters' currencies, we would expect this relative price to display a positive correlation with the nominal exchange rate. On the other hand, if exporters practice LCP on their foreign sales, the correlation could well be negative. The extent of positive correlation is strikingly apparent, although there is a suggestion of PTM for some episodes in which relative export prices respond in a damped fashion to exchange rate changes. These relationships are consistent with a model in which domestic marginal cost (consisting mostly of wages) is sticky in domestic currency terms, and export prices are set as a (perhaps somewhat variable) markup over marginal cost.23 Results such as those in Figures 8-11 illustrate the likelihood of quite divergent behavior by wholesale and retail import prices. A natural question, however,

22Obstfeld and Rogoff (1998, 2000a) assume that full exchange-rate pass-through to import prices is a reasonable approximation of reality, although their basic results hold qualitatively with partial (but positive) pass-through. 23That wages exhibit substantial stickiness in domestic currency terms strikes me as indisputable. For some U.S. evidence, see Altonji and Devereux (1999).

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Figure 8. U.S. Dollar/Yen Exchange Rate and Relative Export Competitiveness (12-month proportional changes)

is whether these correlations are central to understanding the macroeconomic implications of exchange rate changes. Most current PTM-LCP models effectively shut down any mechanism by which exchange rate changes might redirect expenditure internationally. In essence, when all the prices domestic actors face are preset in domestic currency, there is no room for the exchange rate to change the relative prices they face. Thus, the implicit assumption is that price changes such as those shown in Figures 8-11 are economically unimportant. Devereux and Engel (1998, 2000), following Engel (2000), argue that because of PTM-LCP, the size of the expenditure-switching effect central to the Mundell-Fleming model is likely to be very small. Devereux, Engel, and Tille (1999) explicitly model the distinction between wholesale and retail import prices, but in a model where retailers hold no inventories and simply respond passively to consumer demand, so that once again, exchange rates are assumed not to alter any relative prices relevant to economic agents' decisions. This line of reasoning takes the implications of some very restrictive models much too seriously. In reality, there is copious evidence that exchange rate changes do indeed redirect global expenditure, though perhaps with lags—see Krugman (1991) for an overview of evidence through around 1990. That effect should be built into our macroeconomic models because it is central for conclusions about the efficacy of macroeconomic policies and the performance of alternative exchange rate regimes. There is no reason models characterized by PTM-LCP cannot simultaneously imply strong expenditure-switching effects, as I now illustrate. Indeed, in the model I describe below, PTM-LCP is a rather incidental feature; despite it, the

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

Figure 9. U.S. Dollar/Pound Sterling Exchange Rate and Relative Export Competitiveness (12-month proportional changes)

Figure 10. U.S. Dollar/Mark Exchange Rate and Relative Export Competitiveness (12-month proportional changes)

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Figure 11. Canadian Dollar/Pound Sterling Exchange Rate and Relative Export Competitiveness (12-month proportional changes)

model is isomorphic in its essential implications to the non-LCP model of Obstfeld and Rogoff (1998). The main action occurs in firm-to-firm trade that is driven by exchange rate fluctuations.

A Stochastic Model with Local Currency Pricing to Market Multiperiod dynamics are not central to the effects I wish to explore here, so I specify a model in which all economic action save the setting of nominal wages and consumer prices takes place within a single period. Nominal wages and consumer prices are set before the market period. The model combines PTM-LCP for final consumption goods with marginal cost, source-country currency pricing for intermediate goods. It can, alternatively, be interpreted as a model in which firms produce some intermediate imports at foreign subsidiaries, rather than importing them from separate entities located abroad. Rangan and Lawrence (1999) stress sourcing decisions as a major channel through which exchange rate movements influence trade flows and, hence, aggregate demands for countries' outputs. There are two countries, Home and Foreign. Home contains a unit interval [0,1] of workers indexed by i. Each supplies, in a monopolistic fashion, a distinctive variety of labor services. Home and Foreign have identical preferences,

©International Monetary Fund. Not for Redistribution Maurice Obstfeld where p > 0, v > 1, and the real consumption index C is a composite of a continuum affinal commodities indexed by j e [0,1],

In what follows P is the usual CPI function of the individual nominal commodity prices P(j). Foreign is symmetric, but with prices and quantities denoted by asterisks. Importantly, K is a stochastic shock to labor supply that the monetary authorities may be able to respond to after sticky nominal prices are set. Home produces a homogeneous intermediate good y//. Its nominal domestic currency price is PH, and with free trade in intermediates, PH = P#/e, where £, the nominal exchange rate, is the Home currency price of Foreign currency. The Home intermediate good is produced under competitive conditions out of all of the distinctive varieties of domestic labor services. (Similarly, Foreign produces the intermediate YF, with foreign currency price Pp, etc.) The assumption of full and immediate pass-through for intermediates is not realistic, but it simplifies the model and as long as there is an economically significant pass-through, as suggested by Figures 8-11, the main results will go through. The production function for either intermediate is

Nominal wages W(f) (W*(i) in Foreign) are set in advance to maximize expected utility, EU'. The demand for labor variety i in this setting is

(in Home or Foreign), where

Because intermediates markets are perfectly competitive, PH - W and PF = W*. A Home final-goods producer (or distributor) j "manufactures'" final consumption Y(j) out of the Home and Foreign intermediates according to the production function (1)

where YH(j), for example, is input of the Home-produced intermediate into production of final consumption goody. But, as noted above, substitution between production inputs can be viewed as a sourcing decision [for Foreign, U2 m 7*0') = 2YH(j) YF(j) ]. The Home (Foreign) distributor distributes exclu-

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL sively in Home (Foreign), so that in a sense, final consumption goods are nontradables and in equilibrium, C(j) = Y(j) and C\j) = Y\j). I assume that consumers have no way to arbitrage across international markets. The final goods firm must hold domestic money in order to transform intermediates into retail consumer goods. Specifically, a Home firm j will choose

while a Foreign firm chooses

(Note that P^2 P^2 is the minimal Home currency cost of producing a unit of final consumption good, given the production function of equation (1).) This "money in the production function" formulation is one way of imposing a demand for money on the model; Henderson and Kim (1999) describe a related device in the context of consumer money demand. Aggregate money supplies M and M*, like K and K*, are random variables with realizations that do not become known until after wages and final product prices are set. The model assumes an extreme home bias in equity ownership, such that Home (Foreign) residents own all claims to the profits of the Home (Foreign) distributor. The (ex post) budget constraint of representative Home resident / is

where Tl(j) denotes nominal profits of firm j, and his/her first-order condition for the preset nominal wage, W(i), is

(2)

Recall that in the present setup with the CPI preset in domestic currency terms, P is deterministic and can be passed through the expectations operator £{•}. Aside from that modification, this first-order condition is the same as in Obstfeld and Rogoff(1998, 2000a). The more novel element is the first-order condition for the final goods firm, which presets consumer prices in local currency. Let T denote real government transfers to Home firms. Consider the maximization problem of Home firm j, which chooses P(j) in advance to maximize the expected value (to its domestic owners) of profits:

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

subject to

That is, the firm (in effect) maximizes

with respect to P(f), a problem that yields the first-order condition

Of course, the Foreign CPI is preset at the level

I have written the preceding equation for P* in terms of C rather than C* because the model implies that C = C* in equilibrium, PTM notwithstanding. To see why, notice that the Home budget constraint (in equilibrium, after eliminating the government budget constraint) gives

(3)

For Foreign, one finds that

Clearing of the markets for Home and Foreign intermediates gives

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

Substituting for C and C* in either of these yields the equilibrium relative price

from which it follows that

Let us now assume that all of the exogenous variables driving the economy are log-normally distributed. It then becomes relatively straightforward to combine the preceding first-order conditions to obtain log-linear price and consumption equations that parallel precisely those derived in Obstfeld and Rogoff (1998, 2000a). These equations, importantly, involve the second as well as the first moments of the (random) endogenous variables. The log-linearized equations for CPIs (with lower- case letters denoting natural logarithms and e = log e) are

which together imply

This is the equation for the ex ante CPI real exchange rate. (The intuition turns on how production costs — the exchange rate term — and hence profit covaries with demand and with the marginal utility of consumption; see Obstfeld and Rogoff, 2000a.) To derive the (log) expected terms of trade, use equation (3) to rewrite equation (2) (in a symmetric equilibrium) as

©International Monetary Fund. Not for Redistribution Maurice Obstfeld in Foreign,

Let us assume that the first and second moments of K s log K and K* = log K* are the same. Log-linearize the preceding equations in the usual way, and combine them with the equations for p and p* above. The result is the very familiar pair of equations for the expected terms of trade and expected consumption (see Obstfeld andRogoff, 1998, 2000a):

(4)

(5) where

We can derive ex post levels of consumption and the exchange rate from the monetary equilibrium conditions. The Home and Foreign money demand conditions imply that in equilibrium,

From these one derives (6) and

(7)

These expressions can be used to express the variances and covariances affecting the economy's equilibrium in terms of the model's exogenous shocks.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

A couple of immediate points can be made about the model. First, because p and p* are predetermined, the CPI real exchange rate will be perfectly correlated with the nominal exchange rate e, close to what one sees in the data. Nominal exchange rate fluctuations induce commensurate changes in the international prices of supposedly tradable consumption goods. Second, the model includes a substantial expenditure-switching effect of exchange rate changes, one that operates at the firm rather than at the consumer level. Solving the model, one can show that

An increase in a country's money supply depreciates its currency and leads to a proportional rise in output and employment in this model. If Home raises w, Foreign output is unaffected (a knife-edge result) because the negative expenditure- switching effect of the rise in e on y* is exactly offset by the accompanying rise in world consumption spending c. Nonetheless, domestic monetary policy can differ- entially affect domestic output through its impact on the nominal exchange rate.24

Optimal Policy and Alternative Exchange Rate Regimes One major advantage of stochastic models in the new open-economy macroeconomics is that they allow a complete analysis of welfare under alternative exchange rate regimes in the presence of sticky prices and uncertainty. That enter- prise goes well beyond what was possible before and opens up the prospect of a rigorous evaluation of alternative systems. In the example of this section, countries clearly gain if they can gear monetary policy reaction functions toward offsetting shocks to K and K*, as in Obstfeld and Rogoff (2000a). That factor increases the relative attractiveness of national monetary policy autonomy, for reasons that Mundell (1961b) spelled out. It is straightforward to illustrate optimal policies and evaluate alternative monetary regimes in the present model. One simplifying feature of this model, which also holds in my 1998 paper with Rogoff but not in our 2000a paper, is that, regardless of the policy rules the two countries adopt, Home and Foreign always enjoy equal expected welfare levels:

EU = EU*.

(See the Appendix for a derivation.) An implication is that there are no potential divergences of interest between the countries; any development that aids or hurts one automatically affects the other identically. This implication of the model is not

24Were wages flexible instead of sticky, the impact of domestic money on domestic output would be reduced to mlv < m.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld realistic; however, it greatly simplifies the analysis of policy regimes because coordination issues can be ignored. What is the preceding common level of global welfare? Making use of another result derived in the Appendix, we see that

so that expected utility is proportional to

A further simplification is to express the (log) "productivity" shocks in terms of global and idiosyncratic components,

which are mutually orthogonal and satisfy 0^ = a£. = a^., + 0^. Equation (5), finally, implies that we can fully understand the ex ante welfare implications of policy rules by considering their effects on the expression

(8)

Optimal policy rules take the form

These rules, combined with equations (6) and (7) above, allow us to express the world welfare criterion V defined by equation 8 in terms of the parameters 8, 6*, y, and y*. For example, the term 0^ in V can be written as

By setting 3W38 = 0 (similarly for the other policy parameters) and solving, we find that the optimal policy rules entail

These monetary rules have an intuitive interpretation. As one can verify, they allow the global economy to attain the flexible-wage and price real resource allocation ex post, notwithstanding nominal price rigidities. A similar inteipretation

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL characterized optimal policies in the non-PTM, traded-nontraded goods model in Obstfeld and Rogoff (2000a). In that paper, Rogoff and I worked through the properties of a number of possible exchange rate regimes, showing how to carry out exact welfare calculations and rankings. The paper showed that an "optimal float," in which countries commit to optimal monetary rules analogous to those just derived, dominates an "optimal fix," in which national money supplies are perfectly correlated but the world money supply responds optimally to the global shock KW. Of course, the latter regime dominates a "global monetarist" regime in which exchange rates are fixed and the world money supply is held constant in the face of the real shocks. All of those results extend directly to this section's model. An alternative conceptual experiment considers an asymmetrical one-sided peg regime. Suppose that one of the countries (Home, say) can choose the monetary policy rule it prefers on nationalistic grounds while Foreign is required to peg its currency's exchange rate against Home's. One can show in that case that the center country, Home, will choose 8 = 0 and y = 1/fv - (1 - p)] as under an "optimal fix," while the passive country, Foreign, must likewise choose 8* = 0 and y* = i/[v- (1 -p)] in order to maintain a fixed exchange rate against Home. Global welfare thus is the same here as under an optimal cooperative fixed-rate regime. Why? Under one-sided pegging by Foreign, a Home monetary policy rule that responded in a contractionary fashion to Kd would force Foreign to shrink its money supply when hit by favorable idiosyncratic real shocks. That would be harmful to Foreign, leading to a contractionary expected output response there that would reduce Home and Foreign ex ante consumption equally. As a result, it is in Home's self-interest (given Foreign's monetary behavior) to weight domestic and Foreign real shocks equally in its monetary policy response rule. This is not a general result. It need not hold, for example, when countries' expected welfare levels can differ as a result of a direct utility effect of the expected terms of trade, Ex, as in my 2000a paper with Rogoff.

Other Applications The class of model that I have just illustrated can reconcile the "old-time religion" of Mundell and Fleming with the evidence on international prices. But the range of applications goes much further. One of the most striking implications of the model is that uncertainty can affect the first moments of endogenous variables such as the terms of trade and consumption. For example, a rise in Home monetary variability raises the prospects that Home workers will be called on to supply unexpectedly high levels of labor when consumption is high and the desire for leisure greatest. That effect tends to raise domestic relative wages and lower worldwide consumption; see equations (4) and (5). This natural incorporation of uncertainty under price rigidity suggests that we may finally be close to understanding, at an analytical level, some of the gains monetary unification confers by eliminating exchange rate uncertainty. Those gains are fundamental to the affirmative case for monetary union, as outlined so briefly in Mundell's (1961b) optimum currency area paper, yet progress in understanding

©International Monetary Fund. Not for Redistribution Maurice Obstfeld them had been so slow that Krugman (1995) was led to identify progress in this area as the major challenge for international finance. What form is the new research on stochastic dynamic Keynesian models taking? New developments have occurred on several fronts. • We now have some rigorous general-equilibrium models allowing us to investigate the links between exchange-rate variability and international trade; see, for example, Bacchetta and van Wincoop (2000). • We have the prospect of understanding the characteristics of the risk premiums in interest rates and forward exchange rates within dynamic Keynesian settings; see Obstfeld and Rogoff (1998) and Engel (1999b). • We can begin to contemplate a full integration of classical international macroeconomic questions with issues of liquidity and financial constraints (see, e.g., Aghion, Bacchetta, and Banerjee, 2000). • We can now reformulate the analysis of international policy coordination in terms of coordination on policy rules, a topic that has become quite important as institutional reform of monetary institutions has proceeded at the national level; see Obstfeld and Rogoff (2001).

IV. Conclusion In this paper I have outlined major challenges facing international monetary analysis at the start of the postwar era, as well as some of the remarkable strides forward that have been made over the half century since the publication of Meade's (1951) landmark work, The Balance of Payments. I end on an upbeat note—I think we can now claim to have progressed far beyond the point where anyone can seriously argue that "It's all in Meade," or in Mundell and Fleming, for that matter. But I do not want to minimize, either, the work that remains to be done, both in the areas I have discussed in this paper and in others that I have not touched upon. Though there is much to learn still, I believe that the promising recent developments will promote continued growth in our understanding of international macroeconomic relations.

APPENDIX

Expected Utilities in the PTM-LCP Model The text established that the first-order condition for nominal wage setting (by worker 0 is

(9) that the "retailers" set the nominal price of final consumption goods so that (for every retailer j)

(10)

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL and that C= C* always, where C is final consumption. For the Foreign price P*, the last equality in equation 10 holds with the sole modification that E1/2 is replaced by e~1/2. Using this information, one can evaluate

from which it will follow that EU = EU*. To do so, start by invoking the budget constraint,

where FI (as before) denotes the profits of the domestic retailing firms (owned entirely by domestic residents). The preceding equation implies that

so that equation (9) becomes (after multiplying through by the predetermined wage and suppressing the;' index)

(Remember that P is predetermined, too.) The next step is to substitute equation 10 for P on the left-hand side above. The result is

One now concludes that

as claimed in Section III of this paper.

REFERENCES

Aghion, Philippe, Philippe Bacchetta, and Abhijit Banerjee, 2000, "A Simple Model of Monetary Policy and Currency Crises," European Economic Review, Vol. 44 (May), pp. 728-38. Alexander, Sidney S., 1952, "Effects of a Devaluation on a Trade Balance," Staff Papers, International Monetary Fund, Vol. 2 (April), pp. 263-78. Altonji, Joseph G., and Paul J. Devereux, 1999, "The Extent and Consequences of Downward Nominal Wage Rigidity," NBER Working Paper No. 7236 (Cambridge, Massachusetts: National Bureau of Economic Research).

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

Bacha, Edmar Lisboa, and Carlos F. Diaz-Alejandro, 1982, International Financial Intermediation: A Long and Tropical View, Essays in International Finance No. 147 (Princeton, New Jersey: International Finance Section, Department of Economics, Princeton University). Bacchetta, Philippe, and Eric van Wincoop, 2000, "Does Exchange Rate Stability Increase Trade and Welfare?" American Economic Review, Vol. 50, No. 5 (December), pp. 1093-109. Bardhan, Pranab K., 1967, "Optimum Foreign Borrowing," in Essays on the Theory of Optimal Economic Growth, ed. by Karl Shell (Cambridge, Massachusetts: MIT Press). Bergin, Paul R., and Robert C. Feenstra, 2001, "Pricing-to-Market, Staggered Contracts, and Real Exchange Rate Persistence," Journal of International Economics, Vol. 54 (August), pp. 333-59. Belts, Caroline, and Michael B. Devereux, 1996, "The Exchange Rate in a Model of Pricing- to-Market," European Economic Review, Vol. 40 (April), pp. 1007-21. Black, Stanley W., 1973, International Money Markets and Flexible Exchange Rates, Princeton Studies in International Finance No. 32 (Princeton, New Jersey: International Finance Section, Department of Economics, Princeton University). Blejer, Mario I., Mohsin S. Khan, and Paul R. Masson, 1995, "Early Contributions of Staff Papers to International Economics," Staff Papers, International Monetary Fund, Vol. 42 (December), pp. 707-33. Bruno, Michael, 1970, "Capital, Growth, and Trade" (unpublished; Cambridge, Massachusetts: Massachusetts Institute of Technology). Calvo, Guillermo A., 1983, "Staggered Prices in a Utility-Maximizing Framework," Journal of Monetary Economics, Vol. 12 (September), pp. 383-98. Chari, V.V., Patrick J. Kehoe, and Ellen R. McGrattan, 2000, "Can Sticky Price Models Generate Volatile and Persistent Real Exchange Rates?" NBER Working Paper No. 7869 (Cambridge, Massachusetts: National Bureau of Economic Research). Cheung, Yin-Wong, Menzie D. Chinn, and Eiji Fujii, 1999, "Market Structure and the Persistence of Sectoral Real Exchange Rates," NBER Working Paper No. 7408 (Cambridge, Massachusetts: National Bureau of Economic Research). Devereux, Michael B., 1997, "Real Exchange Rates and Macroeconomics: Evidence and Theory," Canadian Journal of Economics, Vol. 30 (November), pp. 773-808. , and Charles Engel, 1998, "Fixed vs. Floating Exchange Rates: How Price Setting Affects the Optimal Choice of Exchange-Rate Regime," NBER Working Paper No. 6867 (Cambridge, Massachusetts: National Bureau of Economic Research). , 2000, "Monetary Policy in the Open Economy Revisited: Price Setting and Exchange Rate Flexibility," NBER Working Paper No. 7665 (Cambridge, Massachusetts: National Bureau of Economic Research). -, and Cedric Tille, 1999, "Exchange Rate Pass-Through and the Welfare Effects of the Euro," NBER Working Paper No. 7382 (Cambridge, Massachusetts: National Bureau of Economic Research). Dornbusch, Rudiger, 1976, "Expectations and Exchange Rate Dynamics," Journal of Political Economy, Vol. 84 (December), pp. 1161-76. , 1987, "Exchange Rates and Prices," American Economic Review, Vol. 77 (March), pp. 93-106. Engel, Charles M., 1993, "Real Exchange Rates and Relative Prices: An Empirical Investigation," Journal of Monetary Economics, Vol. 32, No. 1 (August), pp. 35^1-0. , 1999a, "Accounting for U.S. Real Exchange Rate Changes," Journal of Political Economy, Vol. 107 (June), pp. 507-38.

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

, 1999b, "On the Foreign Exchange Risk Premium in Sticky-Price General Equilibrium Models," in International Finance and Financial Crises: Essays in Honor of Robert P. Flood, Jr., ed. by Peter Isard, Assaf Razin, and Andrew K. Rose (Boston, Massachusetts: Kluwer). 2000, "Local-Currency Pricing and the Choice of Exchange-Rate Regime," European Economic Review, Vol. 44 (August), pp. 1449-72. Fleming, J. Marcus, 1962, "Domestic Financial Policies under Fixed and under Floating Exchange Rates," Staff Papers, International Monetary Fund, Vol. 9 (November), pp. 369-79. Foley, Duncan K., and Miguel Sidrauski, 1971, Monetary: and Fiscal Policy in a Growing Economy (London: Macmillan). Frenkel, Jacob A., and Harry G. Johnson, 1976, eds., The Monetary Approach to the Balance of Payments (London: Allen & Unwin). , eds., 1978, The Economics of Exchange Rates: Selected Studies (Reading, Massachusetts: Addison-Wesley). Frenkel, Jacob A., and Assaf Razin, 1987, "The Mundell-Fleming Model: A Quarter Century Later," Staff Papers, International Monetary Fund, Vol. 34 (December), pp. 567-620. Friberg, Richard, 1998, "In Which Currency Should Exporters Set Their Prices?" Journal of International Economics, Vol. 45 (June), pp. 59-76. Goldberg, Pinelopi Koujianou, and Michael M. Knetter, 1997, "Goods Prices and Exchange Rates: What Have We Learned?" Journal of Economic Literature, Vol. 35 (September), pp. 1243-72. Hamada, Koichi, 1969, "Optimal Capital Accumulation by an Economy Facing an International Capital Market," Journal of Political Economy, Vol. 77, pp. 684-97. Henderson, Dale W., and Jinill Kim, 1999, "Exact Utilities under Alternative Monetary Rules in a Simple Macro Model with Optimizing Agents," in International Finance and Financial Crises: Essays in Honor of Robert P. Flood, Jr., ed. by Peter Isard, Assaf Razin, and Andrew K. Rose (Boston, Massachusetts: Kluwer). International Monetary Fund, 1977, The Monetary Approach to the Balance of Payments: A Collection of Research Papers by Members of the Staff of the International Monetary Fund (Washington: IMF). Isard, Peter, 1977, "How Far Can We Push the 'Law of One Price'?" American Economic Review, Vol. 7 (December), pp. 942-48. Keynes, John Maynard, 1930, A Treatise on Money, Vol. 2 (London: Macmillan). Kollmann, Robert, 2001, "The Exchange Rate in a Dynamic-Optimizing Current Account Model with Nominal Rigidities: A Quantitative Investigation," Journal of International Economics, Vol. 55 (December), pp. 243-62. Krugman, Paul R., 1987, "Pricing to Market When the Exchange Rate Changes," in Real- Financial Linkages Among Open Economies, ed. by Sven W. Arndt and J. David Richardson (Cambridge, Massachusetts: MIT Press). , 1991, "Has the Adjustment Process Worked?" in International Adjustment and Financing: The Lessons from 1985-1991, ed. by C. Fred Bergsten (Washington: Institute for International Economics). 1995, "What Do We Need to Know about the International Monetary System?" in Understanding Interdependence: The Macroeconomics of the Open Economy, ed. by Peter B. Kenen (Princeton, New Jersey: Princeton University Press). Lane, Philip, R., 2001, "The New Open Economy Macroeconomics: A Survey," Journal of International Economics, Vol. 54 (August), pp. 235-66.

©International Monetary Fund. Not for Redistribution Maurice Obstfeld

, and Gian Maria Milesi-Ferretti, 2000, "The Transfer Problem Revisited: Net Foreign Assets and Real Exchange Rates," CPER Discussion Paper No. 2511 (London: Centre for Economic Policy Research). McKinnon, Ronald I., 1981, "The Exchange Rate and Macroeconomic Policy: Changing Postwar Perceptions," Journal of Economic Literature, Vol. 19 (June), pp. 531-57. , and Wallace E. Gates, 1966, "The Implications of International Financial Integration for Monetary, Fiscal and Exchange-Rate Policy," Princeton Studies in International Finance No. 16 (Princeton, New Jersey: International Finance Section, Department of Economics, Princeton University). Meade, James E., 1951, The Balance of Payments (London: Oxford University Press). Metzler, Lloyd A., 1948, "The Theory of International Trade," in A Survey of Contemporary Economics, ed. by Howard S. Ellis (Philadelphia, Pennsylvania: Blakiston). , 1968, "The Process of International Adjustment under Conditions of Full Employment: A Keynesian View," in Readings in International Economics, ed. by Richard E. Caves and Harry G. Johnson (Homewood, Illinois: R. D. Irwin). Michaely, Michael, 1971, The Responsiveness of Demand Policies to Balance of Payments: Postwar Patterns (New York: Columbia University Press). Montiel, Peter J., 1999, "Determinants of the Long-Run Equilibrium Real Exchange Rate," in Exchange Rate Misalignment: Concepts and Measurement for Developing Countries, ed. by Lawrence E. Hinkle and Peter J. Montiel (Washington: World Bank). Mundell, Robert A., 1960, "The Monetary Dynamics of International Adjustment under Fixed and Flexible Exchange Rates," Quarterly Journal of Economics, Vol. 74 (May), pp. 227-57. , 1961a, "The International Disequilibrium System," Kyklos, Vol. 14 , pp. 154-72. , 1961b, "A Theory of Optimum Currency Areas," American Economic Review, Vol. 51 (November), pp. 509-17. , 1962, "The Appropriate Use of Monetary and Fiscal Policy under Fixed Exchange Rates," Staff Papers, International Monetary Fund, Vol. 9 (March), pp. 70-79. , 1963, "Capital Mobility and Stabilization Policy under Fixed and Flexible Exchange Rates," Canadian Journal of Economics and Political Science, Vol. 29 (November), pp. 475-85. , 1968, International Economics (New York: Macmillan). , 1971, Monetary Theory: Inflation, Interest, and Growth in the World Economy (Pacific Palisades, California: Goodyear Publishing). 1991, "The Great Exchange Rate Controversy: Trade Balances and the International Monetary System," in International Adjustment and Financing: The Lessons from 1985-1991, ed. by C. Fred Bergsten (Washington: Institute for International Economics). Mussa, Michael, 1976, "The Exchange Rate, the Balance of Payments, and Monetary and Fiscal Policy under a Regime of Controlled Floating," Scandinavian Journal of Economics. Vol. 78 (May), pp. 229-48. , 1986, "Nominal Exchange Rate Regimes and the Behavior of Real Exchange Rates: Evidence and Implications," Carnegie-Rochester Conference Series on Public Policy, Vol. 25 (Autumn), pp. 117-214. Obstfeld, Maurice, 1987, "International Finance," in The New Palgrave: A Dictionary of Economics, ed. by John Eatwell, Murray Milgate, and Peter Newman (New York: Stockton Press). , 1993, "The Adjustment Mechanism," in A Retrospective on the Bretton Woods System: Lessons for International Monetary Reform, ed. by Michael David Bordo and Barry Eichengreen (Chicago, Illinois: University of Chicago Press).

©International Monetary Fund. Not for Redistribution INTERNATIONAL MACROECONOMICS: BEYOND THE MUNDELL-FLEMING MODEL

, 1998, "Open-Economy Macroeconomics: Developments in Theory and Policy,"' Scandinavian Journal of Economics, Vol. 100 (March), pp. 247-75. , and Kenneth Rogoff, 1995a, 'The Intertemporal Approach to the Current Account," in Handbook of International Economics, Vol. 3, ed. by Gene M. Grossman and Kenneth Rogoff (Amsterdam: North-Holland). , 1995b, "Exchange Rate Dynamics Redux," Journal of Political Economy, Vol. 103 (June), pp. 624-60. , 1996, Foundations of International Macroeconomics (Cambridge, Massachusetts: MIT Press). , 1998, "Risk and Exchange Rates," NBER Working Paper No. 6694 (Cambridge, Massachusetts: National Bureau of Economic Research). Forthcoming in Contemporary Economic Policy: Essays in Honor of Assaf Razin, ed. by Elhanan Helpman and Efraim Sadka (Cambridge; New York: Cambridge University Press). , 2000a, "New Directions for Stochastic Open Economy Models," Journal of International Economics, Vol. 50 (February), pp. 117-53. , 2000b, "The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?" in NBER Macroeconomics Annual 2000, ed. by Ben S. Bernanke and Kenneth Rogoff (Cambridge, Massachusetts: MIT Press). , 2001, "Global Implications of Self-Oriented National Monetary Rules," Quarterly Journal of Economics, forthcoming. Obstfeld, Maurice, and Alan C. Stockman, 1985, "Exchange-Rate Dynamics," in Handbook of International Economics, Vol. 2, ed. by Ronald W. Jones and Peter B. Kenen (Amsterdam; New York: North-Holland). Obstfeld, Maurice, and Alan M. Taylor, 1997, "Non-Linear Aspects of Goods-Market Arbitrage and Adjustment: Heckscher's Commodity Points Revisited," Journal of the Japanese and International Economies, Vol. 11 (December), pp. 441-79. Polak, Jacques J., 1995, "Fifty Years of Exchange Rate Research and Policy at the International Monetary Fund," Staff Papers, International Monetary Fund, Vol. 42 (December), pp. 734-61. , 2001, "Two Monetary Approaches to the Balance of Payments: Post-Keynesian and Johnsonian," IMF Working Paper 01/100 (Washington: International Monetary Fund). Purvis, Douglas D., 1985, "Public Sector Deficits, International Capital Movements, and the Domestic Economy: The Medium-Term Is the Message," Canadian Journal of Economics, Vol. 18 (November), pp. 723^2. Rangan, Subramanian, and Robert Z. Lawrence, 1999, A Prism on Globalization: Corporate Responses to the Dollar (Washington: Brookings Institution). Rogers, John H., and Michael Jenkins, 1995, "Haircuts or Hysteresis? Sources of Movements in Real Exchange Rates," Journal of International Economics, Vol. 38 (May), pp. 339-60. Rogoff, Kenneth, 1996, "The Purchasing Power Parity Puzzle," Journal of Economic Literature, Vol. 34 (June), pp. 647-68. Svensson, Lars E. O., and Sweder van Wijnbergen, 1989, "Excess Capacity, Monopolistic Competition and International Transmission of Monetary Disturbances," Economic Journal, Vol. 99 (September), pp. 785-805. Tirole, Jean, 1988, The Theory of Industrial Organization (Cambridge, Massachussets: MIT Press). Tobin, James, 1969, "A General Equilibrium Approach to Monetary Theory," Journal of Money, Credit and Banking, Vol. 1 (February), pp. 15-29.

Do Monetary Handcuffs Restrain Leviathan? Fiscal Policy in Extreme Exchange Rate Regimes

ANTONIO FATAS and ANDREW K. ROSE*

This paper is an empirical study of fiscal policy in countries with extreme monetary regimes. We study members of multilateral currency unions, dollarized countries that officially use the money of another country, and countries using currency boards. We find that belonging to an international common currency area is not associated with fiscal discipline; if anything, spending and taxes are higher inside currency unions. This effect is especially pronounced for dollarized countries that unilaterally adopt the currency of another country. Currency boards are associated with fiscal restraint. [JEL F33, H30]

• his paper studies fiscal policy in countries that have chosen an extreme mone- TI t;tary stance. We think of a country as having an extreme monetary policy if it is in either a currency board or a common currency area. In much of our analysis, we distinguish between multilateral currency unions (such as the East Caribbean Currency Area, or ECCA) and countries that have unilaterally adopted the currency of an anchor country (such as Panama).

'Antonio Fatas is Associate Professor of Economics and Area Coordinator for Economic and Political Sciences at INSEAD and a Centre for Economic Policy Research (CEPR) Research Fellow. Andrew K. Rose is Rocca Professor of International Business, National Bureau of Economic Research (NBER) Research Associate, and CEPR Research Fellow at the University of California at Berkeley. Rose thanks INSEAD, the Federal Reserve Bank of New York, and the Board of Governors of the Federal Reserve System for hospitality while this paper was written. The opinions expressed and all errors are those of the authors alone. A current version of this paper and the data set are available at both authors' websites. The authors thank Michael Bleaney, Eduardo Borensztein, and Atish Ghosh for allowing access to their data on exchange rate regime classifications, and Tamim Bayoumi, IMF conference participants, and an anonymous referee for comments.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

It is possible to motivate our analysis in several ways. A number of countries are considering whether to abandon national monetary sovereignty and unilaterally adopt the money of another country, including Mexico and Argentina; Ecuador, Guatemala, and El Salvador are already proceeding with dollarization. In Europe, 12 countries have already abandoned national monetary discretion within the Economic and Monetary Union (EMU). More generally, there has been much discussion of the "disappearing center" of exchange rate regimes; countries are said to have a choice of either freely floating or going to an extreme monetary stance. A tight monetary regime might be expected to be associated with a smaller fiscal presence because it reflects generally conservative economic policies. It also might induce conservative fiscal policy to harmonize policy, avoid fiscal externalities, and enhance the sustainability of the monetary regime, as is the (partial) intent of the "Growth and Stability Pact" (Eichengreen and Wyplosz, 1998). More generally, if one interprets an extreme monetary regime as a credible commitment device to improve credibility by limiting discretionary economic policy, then one might expect a smaller fiscal presence in extreme monetary regimes. On the other hand, a tight monetary regime makes fiscal policy a more potent tool of policy in a variety of models. For instance, the classic Mundell-Fleming logic dictates that fiscal policy grows in importance when monetary independence is abandoned. The role of fiscal policy might therefore be expected to be large in countries with extreme monetary regimes. The purpose of this paper is to explore if there is in fact any systematic difference between fiscal policy in extreme monetary regimes and fiscal policy in typical countries that retain monetary sovereignty. In our analysis we consider the issue of endogeneity. Some countries have experienced episodes of hyperinflation associated with loose fiscal policy that have in turn led toward tighter monetary regimes. This is very relevant in practice for currency boards; one thinks of Argentina as the quintessential example. Hence, one might expect to see very loose fiscal policy preceding the adoption of a currency board and much tighter policy after the date of adoption. We argue below that this endogeneity problem is not nearly so relevant for currency unions. Currency unions have not been adopted as a result of episodes of macroeconomic instability, and indeed most of the currency unions in the data remain as such for the whole sample period. Still, our results are best viewed as correlations rather than causal statements, especially in the case of currency boards. We find that currency boards and multilateral currency unions are characterized by conservative fiscal policies. Their governments are smaller, and on average they have kept a larger budget surplus when compared with either all the other countries in our sample or a restricted sample of countries with fixed exchange rates. Unilateral currency unions, on the other hand, are characterized by governments that spend more, as a percentage of GDP. This result supports the view that the implementation of fiscal policy in currency boards is dominated by the goal of adding credibility to the monetary regime. In multilateral currency unions, the restrictions on fiscal policy might originate in the possible externalities associated with loose national fiscal policies. This type of reasoning has

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose recently led to explicit restrictions on budget deficits in both the EMU and the proposed West African Economic and Monetary Union (WAEMU). The results of unilateral currency unions are in line with Rodrik (1998), who shows that countries exposed to larger external risk are associated with a larger safe government sector in order to stabilize economic fluctuations. Currency unions, where governments have already tied their hands by adopting the currency of some other country, use fiscal policy to ensure against the additional risk imposed by the extreme monetary regime. This logic also appears when we look at the composition of government spending and the type of taxes used by currency boards and multilateral currency unions. Even though they have smaller governments and larger surpluses, the composition of their budgets is biased toward direct taxes on the revenue side and social spending and transfers on the expenditure side. These components of fiscal policy are generally associated with the role of automatic stabilizers. The paper is structured as follows. Section I provides a brief discussion of the theory of fiscal policy under different exchange rate regimes. Section II describes the econometric methodology, and Section III presents the data set used. The empirical analysis starts with some preliminary statistics in Section IV, the main results appear in Section V, and these are followed by some sensitivity analysis in Section VI. Sections VII and VIII extend the analysis to different measures of fiscal policy. Section IX discusses the findings. Section X concludes.

I. Theory Although there is a large literature on the effects of the exchange rate regime on macroeconomic variables (volatility, trade), not much attention has been paid to the interaction between the exchange rate regime and the way fiscal policy should operate using modern techniques. This is even more true of the empirical relationship between the exchange rate regime and fiscal policy, about which little is known. One way to rectify the empirical deficiencies in this literature would be to estimate the relationship between fiscal policy and the exchange rate regime for typical choices of the latter. Because most countries are in fixed, intermediate, or floating rate regimes, such an investigation would have to classify countries into exchange rate regimes and search for systematic differences in fiscal policy between, say, fixers and floaters. We choose to focus instead on the small number of countries that have chosen extreme monetary regimes. From a methodological perspective, we hope that these extreme regimes can shed light on the interaction that is blurred by other considerations when one compares fixers and floaters. Of course, there is no guarantee that looking at extreme data points will clarify the situation because outliers are fundamentally . . . outliers. Essentially there are three theoretical channels through which fiscal policy is related to the exchange rate regime: (1) fiscal policy as a credibility device, (2) fiscal policy as a stabilizing tool, and (3) the externalities associated with loose fiscal policies in multilateral currency unions. A standard view of the connection between exchange rate regimes and fiscal policy is that fixed exchange rate regimes are associated with stricter

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN? fiscal policy because of the credibility role of economic policies. Because many exchange rate devaluations are associated with fiscal deficits and severe problems of credibility for governments and central banks, tighter fiscal policy becomes a required element in any exchange rate-based stabilization. Also, the external visibility and impact of devaluations in a fixed exchange rate regime raises the cost associated with irresponsible fiscal policy. Flexible exchange rates, on the contrary, not being subject to large realignments, do not provide the type of punishment that will discourage governments from running irresponsible fiscal policies. This argument has recently been challenged by Tornell and Velasco (2000), who use the same credibility logic to argue that flexible exchange rates in fact provide more discipline. The reason is that movements in the currency reflect the excesses of fiscal policy faster and in a more transparent way. Under fixed exchange rates, the indicators of future crises, such as foreign reserves, are not transparent enough to reveal unsustainable paths of fiscal policy. In fact, one can think of the difference between flexible and fixed rates as being reflected in the intertemporal allocation of the inflation tax burden. Under flexible exchange rates, the excesses of fiscal policy are paid immediately. Thus, if the fiscal authority is impatient enough, there will be more adjustment under flexible than fixed exchange rates. A second way of establishing a relationship between fiscal policy and exchange rate regimes is to think about fiscal policy as a stabilizing tool for business cycles.1 Different exchange rate regimes are associated with different types of risks, and in an environment where economic policy is designed optimally, we should expect different exchange rate arrangements leading to different design of fiscal policies. When governments abandon monetary policy by fixing the exchange rate, they eliminate an important stabilization tool. The result is a greater need to make use of the other available tools, such as fiscal policy. Fiscal policy thus might be larger and be more responsive to business cycles under fixed exchange rates.2 Along these lines, there is strong evidence (Rodrik, 1998) that openness and the additional risk that it imposes through terms of trade volatility are associated with larger governments (as a mechanism to stabilize fluctuations).3 The third connection between fiscal policy and extreme exchange rate regimes originates from the need to overcome the externality associated with the irresponsible fiscal policy of partners in multilateral currency unions. In the case of multilateral currency unions, countries might want to impose limits on fiscal policy because of the fear that partners in the currency union, having abandoned monetary policy, opt for fiscal policy that is too loose and imposes externalities on their

'A discussion on the evidence that fiscal policy is an effective stabilizing tool can be found in Fatas and Mihov (1999). 2This possibility might be especially relevant if a credible conservative monetary policy anchors the public's expectations, allowing a strong stabilizing role for fiscal policy. Also, following the standard text- book Mundell-Fleming model, fiscal policy is much more effective as a stabilizing tool under fixed than under flexible exchange rates. -'How exchange rate mechanisms relate to this evidence is not straightforward from a theoretical point of view. One could argue that fixed exchange rates provide a more stable environment in terms of exchange rate volatility, and thus they will be associated with smaller governments.

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose neighbors. This is, for example, the principle behind the Growth and Stability Pact of the EMU and the fiscal restrictions set out by the proposed WAEMU.4 The absence of such strictures clearly played an important role in the disintegration of the ruble zone in the former Soviet Union in the early 1990s. In summary, the theoretical arguments are divided between those who put their emphasis on credibility and suggest that fixed exchange rates may be characterized by conservative fiscal policy and those who predict more active fiscal policy under fixed exchange rates, given that it is the only tool available to smooth out economic fluctuations. This theoretical ambiguity can only be resolved by an examination of the data. We next turn to that task.

II. Econometric Methodology Our methodology consists of regressing different variables that characterize fiscal policy against dummy variables for the countries with extreme exchange rate regimes (currency unions or currency boards). We control for a set of variables that we expect to be related to both fiscal policy and the exchange rate regime. Our goal is to assess whether fiscal policy in these countries is significantly different from that in the rest of our sample. We also perform a narrower comparison between currency boards, currency unions, and countries with fixed exchange rates. We estimate equations of the form (1) where v is one of our measures of fiscal policy, the subscripts / and / denote countries and time periods respectively, {A} is a comprehensive set of time dummy variables that we usually include, X is a set of control regressors that we discuss further below, UniCU denotes a dummy variable for countries that have unilaterally adopted the money of another country, MultiCU is a dummy variable for membership in a multilateral currency union, CB is a dummy variable for countries in currency board arrangements, and e is a well-behaved residual term denoting all other influences on fiscal policy. We estimate this equation with OLS and robust standard errors. We are really interested only in the § coefficients (the other coefficients are essentially nuisance terms). Positive § estimates indicate a larger fiscal presence for countries in extreme monetary regimes.

III. The Data Set We use a large data set with a broad range of countries. The fact that some of the countries in the sample we are interested in are small might raise questions about how general the results can be and the extent to which they can be applied to other countries. The advantage of this sample is that by focusing our analysis on extreme

4For a detailed discussion of the theoretical arguments behind the Growth and Stability Pact of the EMU, see Eichengreen and Wyplosz (1998).

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN? monetary regimes, we have countries where exchange rate regimes are well-defined and where the issues of endogeneity are minimized. The alternative would be to focus on a narrower sample of richer and better-known countries, excluding small countries or those for which data availability is an issue. But if we were to follow this strategy, we would struggle with both the measurement and the endogeneity problems associated with the classification of exchange rate regimes. At the same time, by looking at "obscure" and small countries we are subjecting our theories to a strong test, because it might be more difficult to find any significant effect. Our data set is data taken from the World Bank's World Development Indicators CD-ROM. The data set is annual, and it includes observations for 206 countries from 1960 through 1998 (though there are many missing observations). The countries in our sample are listed in Table 1. In this data set, there are 1,915 country-year observations (24 percent of the sample) on countries that are members of currency unions and 223 observations (3 percent of the sample) on countries in currency boards. Members of common currency areas are shown in Table A1 in the Appendix. Most currency unions occur where one of the geographic units does not issue its own currency and uses that of another. A few occur where there is considerable currency substitution (also known as "dollarization") between two currencies with a long-term peg at 1:1 (to make price comparison trivial).5 In some of our work below, we distinguish between countries that have unilaterally chosen to surrender monetary sovereignty and countries that are members of the ECCA and the colonies franchises d'Afrique (CFA) franc zone, multilateral currency unions.6 Currency boards are listed in Table A2 in the Appendix. We note parenthetically that both currency unions and currency boards are associated with reduced exchange rate volatility. The effects of an extreme monetary regime on exchange rate volatility are both economically and statistically significant. In particular, nominal effective exchange rate volatility (the standard deviation of the percentage change in the trade-weighted effective exchange rate) is about 13 percentage points lower for members of unilateral currency unions, 9 points lower for members of multilateral currency unions, and 17 points lower for currency boards.7 These reductions are somewhat lower for measures of real effective exchange rate volatility, and they are independent

5We do not include countries that are informally or unofficially dollarized, German unification in 1990. or the reintegration of Okinawa with Japan in 1972. 6Of the total currency union observations, 725 are for multilateral currency unions, and 1,190 are for unilateral membership in a common currency area. We exclude the East African countries, given the analysis in Cohen (2000). The ECCA consists of Anguilla and Montserrat (British territories), Antigua and Barbuda. Dominica, Grenada, St. Kitts and Nevis, St. Lucia, and St. Vincent and the Grenadines. The CFA franc zone includes Benin; Burkina Faso; Cameroon; Central African Republic; Chad; (Republic of) Congo; Comoros using the Comorian franc; Cote d'lvoire, Equatorial Guinea, and Gabon using the franc of the Cooperation Financiere Africaine; Guinea-Bissau; Mali; Niger; Senegal; and Togo using the franc of the Communaute Financiere Africaine (Equatorial Guinea and Mali joined in 1984). There are some technical issues of little interest; for instance, the Banque Centrale des Etats de 1'Afrique Equatoriale et du Cameroun (BCEAC) of the central region issues currency with similar appearance and identical name identifiable by member, while the Banque Centrale des Etats de 1'Afrique de 1'Ouest (BCEAO) of the western region issues a single currency. 7The average level of exchange rate volatility for the entire sample (including extreme regimes) is 16 percent a year.

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose

Table 1. Countries in the Data Set

Afghanistan Dominica Lebanon Rwanda Albania Dominican Republic Lesotho St. Kitts and Nevis Algeria Ecuador Liberia St. Lucia American Samoa Egypt Libya St. Vincent and the Andorra El Salvador Liechtenstein Grenadines Angola Equatorial Guinea Lithuania Samoa Antigua and Barbuda Eritrea Luxembourg Sao Tome' and Principe Argentina Estonia Macao China Saudi Arabia Armenia Ethiopia Macedonia, Former Senegal Aruba Faeroe Islands Yugoslav Republic of Seychelles Australia Fiji Madagascar Sierra Leone Austria Finland Malawi Singapore Azerbaijan France Malaysia Slovak Republic Bahamas, The French Polynesia Maldives Slovenia Bahrain Gabon Mali Solomon Islands Bangladesh Gambia, The Malta Somalia Barbados Georgia Marshall Islands South Africa Belarus Germany Mauritania Spain Belgium Ghana Mauritius Sri Lanka Belize Greece Mayotte Sudan Benin Greenland Mexico Suriname Bermuda Grenada Micronesia, Federated Swaziland Bhutan Guam States of Sweden Bolivia Guatemala Moldova Switzerland Bosnia and Herzegovina Guinea Monaco Syrian Arab Republic Botswana Guinea-Bissau Mongolia Tajikistan Brazil Guyana Morocco Tanzania Brunei Darussalam Haiti Mozambique Thailand Bulgaria Honduras Myanmar Togo Burkina Faso Hong Kong SAR Namibia Tonga Burundi Hungary Nepal Trinidad and Tobago Cambodia Iceland Netherlands Tunisia Cameroon India Netherlands Antilles Turkey Canada Indonesia New Caledonia Turkmenistan Cape Verde Iran, Islamic Republic New Zealand Uganda Cayman Islands of Nicaragua Ukraine Central African Iraq Niger United Arab Emirates Republic Ireland Nigeria United Kingdom Chad Isle of Man Northern Mariana United States Channel Islands Israel Islands Uruguay Chile Italy Norway Uzbekistan China Jamaica Oman Vanuatu Colombia Japan Pakistan Venezuela, Republica Comoros Jordan Palau Bolivariana de Congo, Democratic Kazakhstan Panama Vietnam Republic of Kenya Papua New Guinea Virgin Islands (U.S.) Congo, Republic of Kiribati Paraguay West Bank and Gaza Costa Rica Korea, Democratic Peru Yemen, Republic of C61e dTvoire People's Republic of Philippines Yugoslavia, Federal Croatia Korea, Republic of Poland Republic of Cuba Kuwait Portugal (Serbia/Montenegro) Cyprus Kyrgyz Republic Puerto Rico Zambia Czech Republic Lao People's Qatar Zimbabwe Denmark Democratic Republic Romania Djibouti Latvia Russian Federation

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN? of whether our effective exchange rates are constructed with import, export, or total trade weights.8 The measures of fiscal policy that we use are proxies for the theoretical arguments developed in Section I. We first need to capture the insurance that governments provide through fiscal policy and automatic stabilizers. The government share of GDP and the overall budget surplus are two indicators of the importance of automatic stabilizers. If we think of the government sector as being a safe sector that is less volatile than the private component of GDP, a larger share of government spending on GDP provides smoother business cycles.9 Also, the ratio of taxes to GDP is highly correlated to the responsiveness of taxes and transfers to business cycles, a common indicator of automatic stabilizers.10 Both variables can also provide information on the second dimension of fiscal policy that we are after, namely credibility. Smaller budget deficits and restrained spending are always behind the standards recommendations to countries with fixed exchange rates in order to gain the necessary credibility. We therefore start our analysis by focusing on five key measures of fiscal policy: (1) total expenditures (WDI mnemonic GB.XPD.TOTL.GD.ZS); (2) current revenue (excluding grants, GB.RVC.TOTL.GD.ZS), a measure of the budget balance; (3) the overall budget surplus (including grants, GB.BAL.OVRL.GD.ZS); (4) general government consumption (NE.CON.GOVT.ZS); and (5) tax revenue (GB.TAX.TOTL.GD.ZS). All of these variables are expressed as percentages of GDP; to ease interpretation we do not transform the regressands further.[ l

IV. Descriptive Statistics We begin our analysis with some descriptive statistics. Table 2 tabulates means of our five key fiscal variables for the observations without extreme monetary regimes (i.e., non-currency union/board observations) and tabulates the additional effects of both currency unions and currency boards. P-values for two tests are also tabulated in the last two columns of the table. The first tests the hypothesis that the currency union effect equals that of currency boards; the second tests the hypothesis that the two effects are jointly zero. Table 2 provides mild evidence that extreme monetary rales are associated with tighter fiscal policy in terms of budget deficits. Both currency boards and

8Real effective exchange rate volatility is 4 percentage points lower for members of unilateral currency unions, 2 points lower for members of multilateral currency unions, and 10 points lower for currency boards. Again, these results are all statistically significant at conventional levels. 'Rodrik (1998) formalizes these arguments when looking at the relationship between openness and government size. 10Van den Noord (2000) shows that, in a sample of OECD countries, government size is positively correlated to the cyclical elasticity of taxes and transfers. In the next section we also look at direct measures of the cyclical elasticity of taxes and the deficit. 1 'We have two measures of both government expenditures and revenues to confirm the robustness of our results across different definitions of fiscal policy. While government expenditures is a broader measure than government consumption, it can be subject to more measurement problems associated with certain categories of spending. Rodrik (1998), for this reason, favors the use of government consumption as the measure of government size.

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose

Table 2. Fiscal Policy and Extreme Monetary Regimes: Descriptive Statistics Test Test Mean for Non- Currency Union Currency CU = CB CU=CB = 0 Unions/Boards Effect Board Effect (p-value) (p-value) Spending 28.2 1.3 -0.1 0.48 0.12 (0.29) (1.9) (0.3) Revenue 23.9 0.5 1.6 0.54 0.44 (0.2) (0.6) (1.7) Budget surplus/deficit -3.7 0.8 1.8 0.18 0.00 (0.1) (0.3) (0.7) Taxes 19.2 1.2 2.6 0.27 0.01 (0.2) (0.5) (1.2) Government consumption 15.3 1.5 -0.3 0. 1 1 0.00 (0.1) (0.3) (1.1) Note: Standard errors are in parentheses. currency unions are characterized by smaller budget deficits than other countries in the sample. The difference with the other observations in the sample varies from 0.8 percent (in the case of currency unions) to 1.8 percent (for currency boards); both effects are statistically significant. In the case of government size (measured by either spending or revenue), the differences are economically and statistically smaller. Larger governments characterize currency unions, despite the evidence of tight fiscal policy as measured by the budget deficit. Currency board countries have slightly lower spending but higher taxes. One has to be careful interpreting these unconditional means because of the presence of variables that can be correlated with both fiscal policy and the exchange rate regime. For example, and following Rodrik (1998), openness is positively correlated with government size, and we also know that the exchange rate regime is directly related to openness; currency unions tend to be more open to trade. Similarly, GNP per capita is related to government size, and it can be argued that it could influence or be influenced by the exchange rate regime. For this reason, we now turn to a more sophisticated statistical analysis where controls are introduced for those variables that can be useful in explaining cross-country differences in fiscal policy.

V. Benchmark Results In Table 3 we report benchmark regressions for our five key fiscal variables. The top panel uses two key controls (Xs in equation 1 above): the natural logarithm of real GDP per capita and the log of openness (trade as a percentage of GDP). This panel, in turn, is split into three parts. At the extreme left we record our benchmark results, which do not include time effects (so that we impose (3 = 0 in equation 1).

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

In the middle of the table we allow for time effects. At the right we estimate our benchmark regressions but include only data from extreme monetary regimes and fixed exchange rate observations. Thus extreme monetary regimes are compared only to fixes, not to floaters or countries in intermediate exchange rate regimes.12 The bottom panel of Table 3 is an analogue that adds four additional controls, the logarithms of population, land area, the urbanization rate, and the dependency rate. In all cases, we report point estimates for 5i through 63. In parentheses under- neath we record absolute values of ^-statistics; these test the hypothesis that the relevant coefficient is zero. Although not reported in Table 3, the basic controls in the regressions generally come out with the expected sign. Government size (measured by either spending or taxes) increases with openness. The size of the coefficient is similar in magnitude to the estimates reported in Rodrik (1998).13 We find, unlike Rodrik (1998), that government size is positively related to GDP per capita (Wagner's Law). The dependency ratio is always significant and, as expected, is positively correlated to government size. The coefficients on the exchange rate regime dummies differ in magnitude and sign depending on the arrangement considered. When it comes to currency unions, the overall picture that emerges from Table 3 is that there is no evidence that currency unions restrain fiscal policy as measured by government size. In fact, in the case of unilateral currency unions, the regressions support the view that, on the contrary, these countries seem to have larger governments (measured by total expenditures or total taxes). This is also true when we compare them with countries with fixed exchange rate regimes (the right-hand block of columns). On average, unilateral currency unions have governments that spend (as a percentage of GDP) 5 percentage points more than countries with fixed exchange rates. In the case of multilateral currency unions, the evidence is not clear-cut. Most of the coefficients are not significant and are not robust to the introduction of additional controls, as in the bottom panel of Table 3. Overall, according to the evidence, multilateral currency unions tend to have smaller governments. For example, in the bottom panel of Table 3, with the enlarged list of controls, multilateral currency unions have governments that, compared with other countries with fixed exchange rates, spend about 2.8 percentage points less (and this coefficient is significant). In the case of currency boards, the results are in line with those of multilateral currency unions, but the economic effects are larger. The coefficients on government size are consistently negative and large. For example, in the bottom panel of Table 3, relative to all countries with fixed exchange rate regimes, currency boards have governments that spend 8 percentage points (of GDP) less.

12To measure the exchange rate regime, we rely on the classification from Ghosh and others (1996), whom we thank for providing us with their data sets. We have experimented with other exchange rate regime classifications such as the IMF's official Annual Report on Exchange Restrictions and Exchange Arrangements, and found similar results. 13If, for comparison purposes, we introduce government size in logs, the estimated elasticity is about 0.29, close to the estimates reported in Rodrik (1998) of 0.2.

©International Monetary Fund. Not for Redistribution Table 3. Fiscal Policy and Extreme Monetary Regimes: Regression Analysis

Results with Income and Openness Controls Benchmark Results With Time Effects Against Fixes Multi- Currency Multi- Currency Multi- Currency Unilateral lateral board S2 Unilateral lateral board R2 Unilateral lateral board R2 Total expenditures 3.02 -1.81 -6.41 0.26 2.80 -1.94 -5.97 0.26 2.49 -3.30 -5.82 0.25 (2.4) (2.4) (8.6) (2.3) (3.0) (7.6) (2.1) (4.6) (7.0) Revenue 2.22 -0.73 -3.20 0.41 2.08 -0.74 -3.21 0.41 1.95 -1.40 -3.25 0.39 (2.5) (1.6) (3.7) (2.3) (1.6) (3.6) (2.1) (2.8) (3.4) Budget surplus/deficit 1.65 1.98 3.13 0.04 1.73 2.05 2.24 0.08 1.87 2.40 2.63 0.08 (2.5) (4.3) (6.8) (2.7) (4.4) (4,6) (2.8) (5.0) (5.2) Taxes 2.33 0.27 -0.67 0.37 2.30 0.30 -0.78 0.37 2.83 0.09 0.31 0.33 (2.5) (0.7) (0.7) (2.4) (0.8) (0.8) (2.9) (0.2) (0.3) Government consumption 2.61 0.10 -6.31 0.15 2.41 -0.01 -5.94 0.16 0.76 -1.80 -6.85 0.20 (3.4) (0.3) (5.0) (3.2) (0.0) (4.4) (1.0) (4.4) (5.1) Notes: Controls indueled in each reg;ression are n;itural logarithms of real GDP per capita and log of trade/GDP ratio (openness). Absolute values of (-statistics (calculated with robust standard errors) are recorded in parentheses. Results with Additional Controls

Benchmark Results With Time Effects A{gainst Fixes Multi- Currency Multi- Currency Multi- Currency Unilateral lateral board R2 Unilateral lateral board R2 Unilateral lateral board R2 Total expenditures 5.14 -0.34 -6.44 0.25 4.67 -0.50 -5.82 0.27 5.39 -2.83 -8.07 0.25 (3.2) (0.4) (7.9) (2.9) (0.6) (7.0) (3.7) (2.9) (7.8) Revenue 3.72 0.61 -3.23 0.40 3.40 0.54 -2.87 0.41 4.28 -0.53 -4.73 0.40 (3.3) (1.1) (3.7) (3.0) (1.0) (3.2) (4.0) (0.9) (4.5) Budget surplus/deficit 1.00 1.88 2.95 0.04 0.71 1,91 2.18 0.07 1.02 2.77 3.00 0.10 (1.2) (3.2) (5.5) (0.9) (3.2) (4.0) (1.2) (4.1) (4.7) Taxes 2.84 1.04 -0.84 0.35 2.80 1.04 -0.57 0.36 3.72 0.22 1.03 0.32 (2.6) (2.3) (0.9) (2.5) (2.4) (0.6) (3.6) (0.4) (1.0) Government consumption 0.89 -1.77 -3.67 0.27 0.77 -1.75 -3.68 0.28 -0.18 -3.33 -3.81 0.32 (1.5) (4.7) (4.3) (1.3) (4.6) (4.2) (0.3) (7.0) (4.5) Notes: Controls included in each regression are natural logarithms of real GDP per capita and openness, urbanization, dependency, population, and land area. Absolute values of r-statistics (calculated with robust standard errors) are recorded in parentheses.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

It is clear that some of the regressions, especially those of currency boards, are difficult to interpret because of the problems of endogeneity. Do currency boards lead to restrictive fiscal policy? Or are countries with the potential for restrictive fiscal policy more likely to adopt a currency board? We cannot distinguish here between these two explanations, but we still find that the results shed light on the behavior of fiscal policy under extreme exchange rate regimes. We can confirm that while currency boards and multilateral currency unions are characterized by restrictive and conservative fiscal policies, unilateral currency unions are not. In fact, unilateral currency unions display governments that are significantly larger in comparison to either all countries in the sample or a restricted sample of those that have fixed exchange rate regimes. The third row in Table 3 uses the budget surplus as the indicator of fiscal policy. In this case, there is consistency across the three exchange rate regimes considered. The coefficient is always positive and significant. The only exception is the case of unilateral currency unions. In this case, and after the introduction of additional controls, the coefficient is not significant. This result confirms our previous conclusion that while there is evidence of restrictive fiscal policy for the cases of currency boards and multilateral currency unions in the form of smaller governments and larger budget surpluses, there is no such evidence for unilateral currency unions.

VI. Sensitivity Analysis The top panel of Table 4 is an analogue to Table 3 that removes all country-year observations where CPI inflation is either below 0 percent or above 100 percent.14 The bottom panel is an analogue that removes all countries with volatile nominal effective exchange rates.15 The motivation of excluding these countries is twofold. First, we want to make sure that outlier observations are not driving any of our results. Second, we want to eliminate one possible source of endogeneity. Countries that have gone through exchange-rate-based stabilizations following hyperinflation could display a pattern where fiscal deficits take place before stabilization and fiscal discipline follows a successful stabilization. Also, large changes in inflation could have consequences for tax collection and budget deficits. The estimates of Table 4 are comparable to those of Table 3 and thus confirm our basic results. For example, we confirm that, while currency boards and multilateral currency unions display smaller governments, unilateral currency unions have governments that spend more than those of the other countries in the sample. The size of the effects is similar to the ones found in Table 3. Regarding the budget surplus, even after we remove these outliers, there is clear evidence that for all three exchange rate regimes, governments tend to keep healthier budget finances. We have done additional sensitivity analysis by using different sets of controls, adding a control for OECD members (and appropriate interactions with the other controls), and adding country fixed effects. In all cases, there was little

14This removes some 343 country-year observations. l5We define a nominal effective exchange rate as one where the standard deviation exceeds 100 percent a year, using trade-weighted effective exchange rates. This removes some 117 observations.

©International Monetary Fund. Not for Redistribution Table 4. Fiscal Policy; Sensitivity Analysis Results with All Inflation Observations (OJOO) Benchmark Results With Time Effects Against Fixes Unilateral Multilateral Currency board Unilateral Multilateral Currency board Unilateral Multilateral Currency board Total expenditures 3.31 -1.69 -6.46 3.13 -1.80 -6.16 2.73 -3.07 -6.03 (2.6) (2.3) (8.5) (2.5) (2.5) (7.7) (2.3) (4.0) (7.1) Revenue 2.35 -0.59 -3.29 2.21 -0.57 -3.40 2.08 -1.15 -3.47 (2.6) (1.2) (3.6) (1.2) (1.2) (3.6) (2.2) (2.2) (3.5) Budget surplus/deficit 1.51 1.73 2.82 1,56 1.85 1.98 1.80 2.19 2.41 (2.2) (3.4) (5.9) (2.4) (3.6) (3.8) (2.6) (4.2) (4.6) Taxes 2.44 O.SO -0.54 2.38 0.52 -0.68 2.70 0.26 0.28 (2.5) (1.2) (0.5) (2.4) (1.3) (0.7) (2.8) (0.6) (0.3) Government spending 2.92 0.33 -6.03 2.75 0.26 -5.65 1.04 -1.55 -6.72 (3.8) (0.9) (4.6) (3.6) (0.7) (4.0) (1.3) (3.6) (4.7)

Results Without Countries with Highly Volatile Effective Exchange Rates Benchmark Results With Time Effects Against Fixes Unilateral Multilateral Currency board Unilateral Multilateral Currency board Unilateral Multilateral Currency board Total expenditures 3.00 -1.86 -6.42 2.81 -1.98 -6.05 2.58 -3.17 -5.76 (2.4) (2.8) (8.6) (2.3) (3.0) (7.7) (2.2) (4.5) (7.0) Revenue 2.12 -0.89 -3.26 2.00 -0.89 -3.31 1.93 -1.42 -3.31 (2.4) (2.0) (3.8) (2.2) (2.0) (3.7) (2.1) (2.8) (3.4) Budget surplus/deficit 1.62 1.92 3.09 1.67 1.98 2.25 1.77 2.26 2.52 (2.4) (4.2) (6.8) (2.6) (4.3) (4.7) (2.6) (4.7) (4.9) Taxes 2.30 0.20 -0.69 2.28 0.23 -0.83 2.86 0.13 0.31 (2.4) (0.5) (0.8) (2.4) (0.6) (0.9) (2.9) (0.8) (0.3) Government spending 2.88 0.37 -6.16 2.69 0.28 -5.76 1.13 -1.42 -6.67 (3.8) (1.1) (4.9) (3.5) (0.8) (4.3) (1.5) (3.6) (5.0) Notes: Controls included in each regression are natural logarithms of real GDP per capita and log of trade/GDP ratio (openness). Absolute values of (-statistics (calculated with robust standard errors) are recorded in parentheses.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN? change in the results reported above. We have also introduced the currency dummies interacting with the degree of openness as in Rodrik (1998), and the interaction terms are always highly significant and of the same sign as the ones reported in Table 3.16

VII. Analysis of Elasticities The analysis above has focused on only two dimensions of fiscal policy: government size and the budget surplus. When discussing the stabilizing role of fiscal policy, more attention is normally paid to the cyclical elasticities of taxes and expenditures.17 These elasticities are used as a direct indicator of the smoothing properties of fiscal policy.18 In Table 5, we use fiscal elasticities as regressands in place of the key fiscal ratios that we employed in Tables 2-4. The regressions we report are pure cross- sections; we use country-specific period averages of the regressors as our controls. We estimate our fiscal elasticities by using the coefficient estimate from a regression of the change in the fiscal ratio against the growth rate of real GDP. That is, we use the point estimate £ in the time-series regression:

where A denotes the first-difference operator, u denotes a well-behaved residual term, and each of the N regressions is run over time for an individual country /. We require at least 15 observations to estimate an elasticity. To estimate the effect of extreme monetary regimes on fiscal elasticities, we use (1')

where ^(y), denotes an estimated fiscal elasticity for country i for fiscal variable y, Xt denotes the period-averages of the controls regressors for country i, and so forth. Note that, for example, Cfi, = ~L,CBih the period average membership in a currency board; this averaging is necessary because most countries were not currency boards over the entire period. At the extreme right of the table, we report probability values for the hypothesis Ho: 8] = 82 = 83 = 0. Again, the top panel contains results with two key controls (real income and openness), and the bottom panel includes four other controls. The results are essentially insignificant regardless of the fiscal variables used and the exchange rate regime analyzed. The only exception is the coefficient on

16If we introduce both the dummies and the interaction terms, then the interaction terms always come out with the opposite sign from the dummies themselves, a result of the collinearity among the three variables. 17For example, the growth and stability pact signed by EMU members emphasizes cyclical elasticities when imposing limits on fiscal policy. 18At the same time, although it is true that cyclical elasticities of taxes and expenditures can provide a direct measure of the smoothing properties of fiscal policy, the difficulties in measuring them, and the fact that they are correlated to the overall size of the budget, has moved the debate from these elasticities to measures of government size (e.g., the Commission of the European Communities (1977) report on the need for a fiscal federation in EMU talks about the required size of a Europe-wide budget).

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose

Table 5. Fiscal Policy Elasticities and Extreme Monetary Regimes Results with Income and Openness Controls

Unilateral Multilateral Currency Boards p Coefficient = 0

Total expenditures -0.01 0.06 0.44 0.02 (0.1) (1.2) (3.1) Revenue -0.03 -0.11 -0.14 0.52 (0.3) (1.1) (1.0) Budget surplus/deficit 0.03 -0.17 -0.00 0.52 (0.2) (1.4) (0.0) Taxes -0.03 -0.12 -0.06 0.39 (0.3) (1.6) (0.8) Government consumption -0.09 -0.07 0.63 0.51 (1.3) (1.3) (1.0) Notes: Controls included in each regression are natural logarithms of real GDP per capita and log of trade/GDP ratio. Absolute values of f-statistics (calculated with robust standard errors) are recorded in parentheses.

Results with Additional Controls

Unilateral Multilateral Currency Boards p Coefficient = 0

Total expenditures 0.01 0.05 0.22 0.47 (0.1) (0.8) (0.15) Revenue 0.03 -0.09 -0.11 0.55 (0.5) (1.0) (0.8) Budget surplus/deficit 0.10 -0.15 0.31 0.31 (0.6) (1.5) (0.6) Taxes 0.03 -0.10 -0.04 0.43 (0.4) (1.5) (0.5) Government consumption -0.12 -0.07 0.67 0.70 (1.1) (1.1) (1.0) Notes: Controls included in each regression are natural logarithms of real GDP per capita and openness, urbanization, dependency, population, and land area. Absolute values of f-statistics (calculated with robust standard errors) are recorded in parentheses. the elasticity of government expenditures for the case of currency boards. This coefficient is positive and significant, which indicates that government expenditures in these countries are more reactive to cyclical conditions. One reason for the lack of significance of all the other coefficients might be that we have only one observation per country and that the estimation of these elasticities is not very precise because of the few data points used in some of the regressions. Because of all the measurement problems associated with cyclical elasticities, we turn now to an indirect method to learn about these elasticities by exploring differences in the composition of expenditures and budget revenues. Different components of the budget can be more or less responsive to economic conditions, and therefore evidence of differences in the relative size of these components can provide additional information on the cyclical responsiveness of fiscal policy.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

VIII. Disaggregated Analysis

Table 6 contains analogues to Table 3, which examine fiscal policy at a more disaggregated level. We estimate equation (1) looking at subcomponents of government spending and revenue, as well as some alternative measures of aggregate fiscal policy. We examine five components of revenue generation: (1) nontax revenue, (2) goods and services taxes, (3) trade taxes, (4) social security taxes, and (5) taxes on income and profits, all measured as percentages of revenue. We also examine five components of spending: (1) goods and services expenditures, (2) interest, (3) subsidies and transfers, (4) wages and salaries, and (5) capital, all measured as percentages of total spending. Finally, we examine two interesting adjuncts to our five aggregated fiscal measures: central government debt and foreign financing, both measured as percentages of GDP. The reason for looking at different components of spending and taxes is that they play different stabilizing roles. For example, direct taxes are generally more progressive and therefore more likely to help smooth out business cycle fluctuations. On the spending side, spending on welfare and transfers are more accurate measures of the insurance provided by governments. Along these lines, Rodrik (1998) shows that spending in social security and welfare is more sensitive than other components of the budget to measures of risk (i.e., countries that are exposed to more risk because of a higher degree of openness tend to have more spending in social security and welfare). On the revenue side, the first two rows, which represent measures of direct taxes, indicate that currency boards and multilateral currency unions are more likely to use direct taxes. In both cases, social security taxes are responsible for this result. In the case of currency boards, although taxes on income and profits are smaller, social security taxes are larger enough to make the sum of the two higher than in other countries in the sample. In the case of unilateral currency unions, the evidence is mixed. Although social security taxes represent a higher share of total taxes, they are compensated by lower taxes on income and profits so that total direct taxes are not significantly higher. On the spending side, currency boards are the ones that make more use of subsidies and transfers relative to other forms of spending, while in the case of currency unions there is no significant pattern that distinguishes their expenditure components from those of other countries in the sample. Therefore, and in line with the results on the use of direct taxes, currency boards appear as countries where the composition of the budget is more biased toward expenditures and revenues that are better suited to provide insurance against economic fluctuations. How does the above result relate to our previous results on government size? Rodrik (1998) convincingly shows that more open countries face higher external risk (because of, e.g., terms of trade volatility), and thus tend to choose large governments. Our findings in Section VI on the smaller size of governments under currency boards (and also to some extent on multilateral currency unions) go somewhat against Rodrik's finding. In currency boards, as in any other form of extreme exchange rate regime, monetary policy is absent as a stabilizing tool;

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose

Table 6. Disaggregated Analysis of Fiscal Policy and Extreme Monetary Regimes Benchmark Results With Time Effects Multi- Currency Multi- Currency Unilateral lateral board Unilateral lateral board

Social security taxes 3.23 4.07 18.0 3.44 4.16 17.6 (% revenue) (4.3) (6.9) (9.7) (4.5) (7.0) (9.2) Taxes on income and profits -4.52 0.99 -9.62 -3.85 0.97 -8.09 (% revenue) (4.2) (0.9) (3.9) (3.5) (0.9) (3.4) Goods and services taxes -7.32 -8.41 7.57 -7.90 -8.39 5.54 (% revenue) (6.8) (9.9) (3.3) (7.2) (9.9) (2.5) Nontax revenue 5.26 -2.06 -8.87 4.37 -2.20 -9.04 (% revenue) (2.6) (1.8) (5.5) (2.2) (1.9) (5.7) Trade taxes 6.19 3.88 -4.89 6,73 3.90 -4.12 (% revenue) (2.5) (3.1) (3.1) (2.7) (3.1) (2.5) Subsidies and transfers -1.88 0.14 6.90 -1.09 0.47 6.63 (% spending) (1.3) (0.1) (4.2) (0.7) (0.4) (3.4) Goods and services -0.18 -0.13 0.78 -0.20 -0.09 0.81 expenditures (% spending) (0.1) (0.1) (0.4) (0.1) (0.1) (0.4) Capital (% spending) 4.50 0.81 -3.66 5.28 1.08 -2.91 (2.8) (0.5) (4.3) (3.3) (0.6) (3.1) Interest (% spending) 0.99 -2.53 -3.79 -0.49 -3.26 -4.30 (1.5) (2.9) (2.7) (0.7) (3.8) (2.7) Wages and salaries 2.21 0.91 -2.24 1.98 0.82 -2.31 (% spending) (1.5) (0.7) (2.0) (1.4) (0.7) (2.0) Central government debt -15 19 -39 —23 12 -30 (% GDP) (2.6) (1.9) (6.5) (3.7) (1.2) (2.1) Foreign financing 0.35 0,94 -0.32 0.49 0.94 -0.12 (% GDP) (0.7) (1.7) (0.9) (1-0) (1.7) (0.3) Notes: Controls included in each regression are natural logarithms of real GDP per capita and openness, urbanization, dependency, population, and land area. Absolute values of r-statistics (calculated with robust standard errors) are recorded in parentheses. thus one might expect a large fiscal presence. Our findings suggest that, on the contrary, currency boards and multilateral currency unions have smaller governments. The evidence of Table 6 on the components of the budget might explain some of this apparent contradiction. Although currency boards display conservative fiscal policies from the perspective of the size of the government and the budget deficit, the composition of their budgets is tilted toward some of the components that can provide more social insurance (e.g., transfers and subsidies, social security taxes). This is confirmed by the fact that the cyclical elasticity of government expenditures was larger under currency boards. Fiscal policy faces a trade-off between credibility and the need to provide stabilization, and the way it is resolved is by limiting the size of the government and the budget deficit and putting emphasis on components that ensure the functioning of automatic stabilizers.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

IX. Discussion

We have found that there are significant differences in fiscal policy among the three types of extreme exchange rate regimes considered. These differences are only present when we look at government size and some of the components of the budget; they do not appear when we look at the cyclical elasticities of taxes, spending, or the budget deficit. Do these differences correspond to the standard recommendations of economic policy given to these countries? Do they conform to conventional wisdom? In the case of currency boards, the results favor the hypothesis that these countries show more fiscal discipline relative to all countries in the sample and even relative to countries characterized by fixed exchange rate regimes. This fits the common advice given to governments to make currency boards sustainable. Part of the fiscal discipline can come from the reform of the central bank and the impossibility of direct monetary financing of government expenditures. But much of the fiscal impact can be interpreted as an attempt to give economic policy as much credibility as possible. A few examples of recent currency boards can illustrate this argument. A recent joint assessment of the economic policy priorities of the Republic of Estonia by the European Commission (2000) and the Estonian government makes clear that although "fiscal policy remains the main tool of macroeconomic policy to foster the emergence of the right conditions for strong and balanced growth . . . maintaining strict fiscal policy and prudent debt management are key conditions to ensure the full benefits provided by the currency board." Indeed, the consolidated budget was balanced in Estonia during the period 1984-98. Also in the cases of Lithuania and Bulgaria, significant fiscal efforts have been made during the years that the currency board has been in place.'9 Argentina presents a similar scenario. The introduction of the currency board has been associated with unprecedented reductions in public spending. Total public sector spending fell from 39.3 percent of GDP in 1989 to less than 28 percent in 1996. Regarding the composition of the fiscal adjustment, the case of Argentina presents some interesting insights that corroborate several of our results. The fall in total spending has taken place despite the fact that social spending has increased faster than GDP.20 This anecdotal evidence confirms our results in Table 6 that currency boards are associated with larger subsidies and transfers (as well as larger social security taxes). One could argue that the restriction on the general level of spending is the pillar of credibility needed by the currency board, while social spending is kept at levels that guarantee social insurance and the operation of fiscal automatic stabilizers. What is different about currency unions? First of all, the results in Table 3 suggest interesting differences between the behavior of unilateral and multilateral currency unions. While in the case of multilateral currency unions there is some

19A detailed discussion on these cases and an analysis of their advantages in the run-up to EMU membership can be found in Guide. Kahkonen, and Keller (2000). 20See Traa and others (1998) for a detailed analysis of these figures.

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose mild support for the idea that fiscal policy is restrictive, in the case of unilateral currency unions there is clear evidence that fiscal policy is bigger as measured by a larger government size. This result on unilateral currency unions is close in spirit to the results of Rodrik (1998) on the connection between external risk and government size. Rodrik found that more open economies choose larger governments as a form of insurance against the additional risk imposed by terms of trade volatility combined with a higher degree of openness. Our result could be justified by arguing that in the case of unilateral currency unions, the lack of monetary and exchange rate policies impedes a stabilizing mechanism against shocks that needs to be compensated by a larger government that can provide the required insurance. For this analysis to be correct, it has to be the case that the issues of credibility are not so relevant for the unilateral currency unions as they are for currency boards. This is plausible, given the origin of these currency unions and the fact that most of them have remained as such for the whole sample period. Why are multilateral currency unions different? The multilateral arrangement of these currency unions can have two effects that might explain why they are different. First of all, the multilateral nature of the agreement can make the currency union less stable, and therefore more subject to the problems of credibility. As we argued in the case of currency boards, this tilts fiscal policy toward a more restrictive stance. Second, and more important, multilateral currency unions suffer from the externalities that loose fiscal policy can impose on other members of the union. The recent Stability Pact adopted by the EMU and its analogue in the proposed WAEMU are good examples of this type of behavior.21 It is interesting to see that fiscal restrictions of the type set in these multilateral currency unions are not commonly observed in countries that decide to adopt unilaterally another country's currency.

X. Summary and Conclusion This paper has studied the role that fiscal policy plays in extreme monetary regimes (currency unions and currency boards). Our analysis is empirical and relies on a large cross-country panel data set that includes almost 40 years of data for some 200 countries. Our analysis is nonstructural; while we are most interested in the impact of extreme monetary regimes on fiscal policy, fiscal policy may well affect the choice of monetary regime (especially in the case of currency boards). From a theoretical point of view, there is ambiguity about the nature of fiscal policy in extreme monetary regimes. When fixed exchange rate regimes are viewed from the perspective of countries that are in the process of establishing the credibility of their economic policies, it is to be expected that fiscal policy should be more restrictive (compared with that of other countries) to add to the credibility of tight monetary policy. On the other hand, if we abstract from the issue of credibility, countries that have abandoned monetary policy under fixed exchange rate regimes might be more likely to use fiscal policy (rather than floaters) to stabilize

2'In the WAEMU, even before the adoption of the Convergence, Stability, and Solidarity Pact of December 1999, there were restrictions (during the period 1994-98) on the composition and level of certain components of the fiscal budget. See Dore and Nachega (2000) for details on these arrangements.

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN? business cycles. If this is the case, we should see larger governments and more responsive fiscal policy under extreme monetary regimes. The evidence that we present offers partial support for both views. In the case of unilateral currency unions, the fact that larger governments characterize them lends support to the view that fiscal policy has to grow in importance as the need for stabilization increases once monetary policy has been abandoned. However, in the case of currency boards, the effect goes in the opposite direction, as we find that governments of currency boards are smaller in size, giving support to the idea that credibility issues are more important in the case of these countries. Does this mean that in currency boards fiscal policy does not address the lack of policy flexibility imposed by the exchange rate regime? No. Looking at the composition of government spending and revenues, we find that currency boards tend to favor direct taxes and spending on transfers, which are associated with automatic stabilizers and insurance.

APPENDIX

Table Al. Currency Unions in the Data Set

American Samoa Lesotho Andorra Liberia Antigua and Barbuda Liechtenstein Bahamas, The (after 1970) Luxembourg Benin Mali (after 1984) Bermuda (after 1968) Marshall Islands Bhutan Mayotte Brunei Darussalem Micronesia, Federated States Burkina Faso Monaco Cameroon Namibia Central African Republic New Caledonia Chad Niger Channel Islands Northern Mariana Islands Comoros (before 1994) Palau Congo, Republic of Panama Cote d'lvoire Puerto Rico Dominica St. Kitts and Nevis Equatorial Guinea (after 1984) St. Lucia Faeroe Islands St. Vincent and the Grenadines Gabon Senegal Greenland Swaziland Grenada Tanzania (before 1973) Guam Togo Guinea-Bissau (after 1971) Tonga (before 1971) Ireland (before 1979) Uganda (before 1973) Isle of Man Virgin Islands (U.S.) Kenya (before 1973) West Bank and Gaza Kiribati (before 1971 and after 1973)

©International Monetary Fund. Not for Redistribution Antonio Fatas and Andrew K. Rose

Table A2. Currency Boards in the Data Set Argentina (after 1990) Bahrain (before 1974) Bosnia and Herzegovina (after 1997) Bulgaria (after 1996) Cayman Islands (after 1971) Djibouti Estonia (after 1991) Fiji (before 1976) Gambia, The (before 1972) Hong Kong (before 1975 and after 1982) Lithuania (after 1993) Oman (before 1975) Qatar (before 1974) Tonga (after 1970 and before 1975) Western Samoa (before 1974) Yemen, Republic of (before 1974)

REFERENCES

Cohen, Benjamin J., 2000, "Beyond EMU: The Problem of Sustainability," in The Political Economy of European Monetary Unification, ed. by Barry J. Eichengreen and Jeffry A. Frieden (Boulder, Colorado: Westview Press). Commission of the European Communities, 1977. "Report of the Study Group on the Role of Public Finance in European Integration," Economic and Financial Series Studies No. 13A-B (Brussels). Dore, Ousmane, and Jean-Claude Nachega, 2000, "Budgetary Convergence in the WAEMU: Adjustment Through Revenue or Expenditure?" IMF Working Paper No. 00/109 (Washington: International Monetary Fund). Eichengreen, Barry J., and Charles Wyplosz, 1998, "The Stability Pact: More Than a Minor Nuisance?" in EMU: Prospects and Challenges for the Euro (London: Blackwell). Fatas, Antonio, and Ilian Mihov, 1999. "Government Size and Automatic Stabilizers: International and Intranational Evidence," CEPR Discussion Paper No. 2259 (London: Centre for Economic Policy Research). Ghosh, Atish R., Anne-Marie Guide, Jonathan D. Ostry. and Holger Wolf, 1996, "Does the Exchange Rate Matter for Inflation and Growth?" Economic Issues, No. 2 (Washington: International Monetary Fund). Government of the Republic of Estonia and European Commission, 2000, "Joint Assessment of the Economic Policy Priorities of the Republic of Estonia," March 29, http://www.estemb.be/English/Estonia%20and%20EU/offdocs/JointAssessment/JA%20fm al_2303.htm Guide, Anne-Marie, Juha Kahkonen, and Peter Keller, 2000, "Pros and Cons of Currency Board Arrangements in the Lead-Up to EU Accession and Participation in the Euro Zone," IMF Policy Discussion Paper No. 00/1 (Washington: International Monetary Fund). International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions (Washington: IMF, various issues).

©International Monetary Fund. Not for Redistribution DO MONETARY HANDCUFFS RESTRAIN LEVIATHAN?

Rodrik, Dani, 1998, "Why Do More Open Economies Have Bigger Governments?" Journal of Political Economy, Vol. 106 (October), pp. 997^1032. Tornell, Aaron, and Andres Velasco, 2000, '"Fixed Versus Flexible Exchange Rates: Which Provides More Fiscal Discipline?" Journal of Monetary Economics, Vol. 45 (April), pp. 399^36. Traa, Bob, and others, 1998, 'Argentina: Recent Economic Developments," Staff Country Report, No. 98/38 (Washington: International Monetary Fund). van den Noord, Paul, 2000, "The Size and Role of Automatic Fiscal Stabilizers in the 1990s and Beyond," OECD Working Paper (Paris: Organization for Economic Cooperation and Development).

Exchange Rate Regimes and Economic Performance

EDUARDO LEVY-YEYATI and FEDERICO STURZENEGGER*

This paper studies the impact of exchange rate regimes on inflation, nominal money growth, real interest rates, and GDP growth. We find that, for nonindustrial economies, "long" pegs (lasting five or more years) are associated with lower inflation than floats, but at the cost of slower growth. A similar trade-off between inflation and growth is still present in the case of "hard" pegs (currency boards and economies without separate legal tender), whose growth performance does not differ significantly from that of conventional pegs. In contrast, "short" pegs clearly underperform floats, as they grow slower without providing any gains in terms of inflation. [JEL E31, E52, F41, F43J

he proper assessment of the costs and benefits of alternative exchange rate T:regimes has been a hotly debated issue and remains perhaps one of the most important questions in international finance. The theoretical literature has concentrated on the trade-off between monetary independence and credibility implied by different exchange rate regimes, as well as in the insulation properties of each arrangement in the face of monetary and real shocks.1 Recent episodes of financial distress have refocused the discussion by introducing the question of which

'Eduardo Levy-Yeyati is Director of the Center for Financial Research and Professor at the Business School of Universidad Torcuato Di Telia and Federico Sturzenegger is Secretary of Economic Policy of Argentina and Professor at the Business School of Universidad Torcuato Di Telia. The authors would like to thank Eduardo Borensztein, Jorge Carrera, Sebastian Edwards, an anonymous referee, participants at the First Annual Research Conference at the IMF, and seminar participants at Centre de Estudios Macroeconomicos (CEMA) for useful comments, and Iliana Reggio for her outstanding research assistance. 'References on this issue would be too numerous to cite here. A general discussion on some of these issues is given in Obstfeld and Rogoff (1996).

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE exchange rate regime is better suited to deal with increasingly global and unstable world capital markets.2 In particular, given the increasing importance of international capital flows and the predominance of external over domestic monetary shocks, the traditional trade-off has narrowed down to a price stability-growth dilemma, according to which fixes are expected to enhance the credibility of noninflationary monetary policies, reducing inflation and the volatility of nominal variables, while floats are seen as allowing the necessary price adjustments in the face of external (real and financial) shocks, reducing output fluctuations and improving growth performance. The terms of the debate about exchange rate regimes and the views prevalent in policy circles have evolved over time, as they have rarely been independent from the characteristics of international financial markets. In the 1980s, in a context of relatively closed capital markets, external shocks were less relevant and, with many countries struggling with disinflation policies, monetary aspects appeared to be much more important than today. The issues stressed in the academic literature have changed accordingly: while economists in the 1980s concentrated on studying the implications of exchange rate regimes as stabilization instruments (or as credibility enhancers), today the debate focuses on how different regimes may act as absorbers of external shocks or provide a shield against speculative attacks.3 The lack of consensus on the subject has been paralleled by recent developments in the real world. Recent years have witnessed an unprecedented number of changes of exchange rate regimes, in a way that seems to provide partial support to almost any view about the long-run trends in the choice of regimes. Thus, while the inherent vulnerability of intermediate exchange rate arrangements to sudden aggregate shocks revealed by the notorious collapses of pegs or managed floats in Southeast Asia and Latin America have suggested to some observers the conve- nience of more flexible regimes, a number of countries have taken the opposite path, moving toward monetary unions or unilateral dollarization, as was the case in Europe in the aftermath of the European Monetary System (EMS) crisis of 1992, or in Ecuador with the recent adoption of the U.S. dollar as legal tender. The debate is further complicated by another important consideration: Characterizing the exchange rate regimes actually in place in different countries is not a trivial task. Calvo and Reinhart (2000), for example, have pointed out that many countries that claim to be floaters intervene heavily in exchange rate markets to reduce exchange rate volatility, suggesting a mismatch between de jure and de facto regimes. Similarly, Levy-Yeyati and Sturzenegger (2000a) highlight the recent increase in what could be labeled "fear of pegging": countries that run a de facto peg but avoid an official commitment to a fixed parity.4

2Recent contributions include Calvo (1999), Eichengreen (1994), Frankei (1999), Larrafn and Velasco (1999), and Rose (2000). 3Compare, for example, the literature on the role of exchange rates tor stabilization following the seminal contribution of Calvo and Vegh (1994) with more recent papers like Broda (2000) on the relevance of exchange rate regimes as a shock absorber, or Domac and Martinez Peria (2000) about the impact of regimes on the likelihood of banking crises. 4These mismatches between de jure and de facto regimes have been pointed out repeatedly in the literature. See, for example, Frankei (1999) and Quirk (1994). Frieden, Ghezzi, and Stein (2001) and Ghosh and others (1997) make partial attempts at correcting this problem in their empirical work.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

With all this in mind, in this paper we revisit the inflation-growth trade-off, using an extensive database that includes 154 countries and covers the post-Bretton Woods era. We deliberately ignore the Bretton Woods period in which fixes were dominant, largely for political reasons, to concentrate in the recent period of increasing financial integration, in which, we believe, the linkage between exchange rate regimes and the real economy better reflected the choice of individual countries' monetary authorities. Several new aspects are introduced in our analysis. First, we use a de facto classification, described in detail in Levy-Yeyati and Sturzenegger (2000a) (henceforth denoted LYS), that groups exchange rate regimes according to the actual behavior of the main relevant variables, as opposed to the traditional classification compiled by the IMF based on the de jure (i.e., legal) regime that the countries' authorities declare to be running.5 By doing this, we refine the analysis substantially. On the one hand, we avoid the misclassification of pegs that pursue independent monetary policies (and eventually collapse) and floats that subordi- nate their monetary policy to smooth out exchange rate fluctuations, which may bias the statistics of the tests toward lack of significance or incorrect interpretations. On the other hand, the new classification makes a distinction between high and low volatility economies, providing a natural way to discriminate the impact of the regime in tranquil and turbulent times. Second, we distinguish between "long" and "short" pegs; long pegs are defined as those in place for five or more consecutive years and short pegs as those in place for less than five years. We find the distinction useful at least in two respects. On the one hand, it allows us to determine whether the impact on macroeconomic variables is a product of the regime in place or rather the result of the short-run effect of a regime switch. On the other hand, our focus on long pegs addresses the concern that the poor showing of many conventional pegs may be mainly attributable to countries with weaker macroeconomic and political fundamentals that are forced to implement ultimately unsustainable fixed exchange rate regimes. Third, in addition to looking at the inflation-growth trade-off, the paper examines the impact of exchange rate regimes on the cost of capital, as measured by the real interest rate, something that has not been done yet in the literature, to our knowledge. The issue has important policy implications inasmuch as lower interest rates are typically invoked as a key argument in favor of fixed exchange rates and, more recently, of the full adoption of a foreign currency as legal tender. Fourth, we conduct a "deeds vs. words" comparison that makes use of both the LYS and the IMF-based classification, which sheds light on a number of issues. For example, it allows us to test the extent to which economic performance is determined by the actual (as opposed to the reported) exchange rate policy, as well as the "announcement" value of a de jure peg, above and beyond the actual behavior of the regime. Finally, we test whether fixed exchange rate arrangements that imply a harder commitment, such as currency boards or currency unions (a group usually referred

5A detailed description of both classifications is provided in the next section.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE to as "hard" pegs), are different from (and better than) conventional fixes and other regimes in general. This increasingly popular hypothesis stresses that the stronger commitment embedded in a hard peg reduces the vulnerability of the regime to speculative attacks (thus enhancing growth) while reaping all the benefits in terms of lower inflation.6 The main findings discussed in the paper are the following: 1. For industrial countries, we find no significant link between regimes and economic performance. 2. For nonindustrial economies, a robust association between fixed regimes and lower inflation rates appears only when we focus on long pegs. This link seems to work both through its influence on monetary growth and through its impact on expectations. Moreover, deeds rather than words matter for inflation: The announcement of a fixed exchange rate regime has an impact on inflation only in the case of long pegs. 3. Real rates appear to be lower under fixed exchange rate regimes than under floats only according to the de jure classification, suggesting that the result is mostly due to the role of unanticipated devaluations. Interestingly, for de facto pegs, we find that the announcement of a fixed regime has a negative effect on real interest rates only for short pegs, possibly because short pegs, while effective in reducing inflation expectations (and thus nominal interest rates), are not effective in reducing actual inflation (point 2 above). 4. Within the group of nonindustrial countries, pegs (both short and long) are significantly and negatively related to per capita output growth. Thus, the inflation-growth trade-off implicit in the choice between fixed and floating regimes seems to apply only to long pegs. In contrast, short pegs clearly underperform floats: they grow more slowly without providing significant gains in terms of inflation. 5. Hard pegs deliver better inflation results than conventional pegs, but they do not eliminate the inflation-growth trade-off, as they still display significantly smaller growth rates than floating exchange rate arrangements. 6. Compared with de facto floats, de facto pegs that shy away from legally committing to a fixed exchange rate benefit from higher growth performance, providing a justification for the "fear of pegging." The plan of the paper is as follows. Section I succinctly describes the LYS and IMF classification used in the econometric tests. Section II presents the data. Section III shows the main empirical findings for inflation and money growth. Section IV discusses the impact of regimes on real interest rates. Section V looks at the relation between regimes and growth. Section VI explores whether hard pegs behave differently from conventional pegs. Section VII outlines some areas for future research and concludes.

6Besides proponents of the bipolar view like Eichengreen (1994), Summers (2000), and Fischer (2001), who regard intermediate regimes (and, in particular, conventional pegs) as inherently unsustainable in a context of integrated international capital markets, hard peg advocates include, most notably, supporters of full dollarization like Calvo (1999) and Eichengreen and Hausmann (1999). For a thorough presentation of the full dollarization debate, see also Levy-Yeyati and Stur/enegger (forthcoming).

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

I. Exchange Rate Regime Classification

LYS Classification

The LYS de facto classification7 that we used in this paper is based on three variables closely related to exchange rate behavior: (1) exchange rate volatility (ae), measured as the average of the absolute monthly percentage changes in the nominal exchange rate during the year; (2) volatility of exchange rate changes (OA<>), measured as the standard deviation of the monthly percentage changes in the exchange rate; and (3) volatility of resen>es (

7This section borrows from Levy-Yeyati and Sturzenegger (2000a), which provides a detailed explanation of the classification procedure. "Computing the change in reserves relative to the monetary base is a way of assessing the monetary impact of the exchange rate intervention. However, external liabilities and government deposits need to be netted out from the reserves data, to capture only those changes that have a counterpart in monetary aggregates. More precisely, the variable is calculated using line 11 from the International Financial Statistics (IPS), net of lines 16c and 16d, and dividing its change by line 14 (or 14a if line 14 was not available) lagged one month. 'Within this group, the classification distinguishes between dirty floats and crawling pegs, the latter corresponding to the case of significant changes in the nominal exchange rate coupled with relatively stable increments and active intervention. In the empirical analysis conducted in this paper, however, both types are subsumed in the intermediate group. '"For a discussion of cluster analysis techniques, see, for example, Anderberg (1973) and Norusis (1993). ""Inconclusives" from the second-round classification are left unclassified. We use them, however, in several robustness checks below.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 1. LYS Classification Criteria

e C , Flexible High High Low Intermediates Medium/High Medium/High Medium/High Fixed Low Low High Inconclusive Low Low Low first- and second-round regimes, in turn associated with high and low volatilities in the underlying classification variables.12

IMF Classification As we mentioned above, we also conduct our tests using an IMF-based classification for the purpose of comparison with previous work, as well as to address issues related to the announcement value of an exchange rate regime, particularly in the case of pegs.13 The IMF has changed the way it classifies exchange rate regimes over the years. Before 1998, the IMF grouped countries into three basic categories: pegs, limited flexibility, and more flexibility, in turn divided into several subgroups. After 1998, the IMF moved to an eight-way classification: no separate legal tender, currency boards, conventional fixed, horizontal bands, crawling pegs, crawling bands, dirty float, and free floats. In general, however, the categories can be readily mapped on a simpler grouping that includes different forms of pegs (to a single currency, or to a disclosed or undisclosed basket), intermediate regimes (crawling pegs, bands, managed floats, cooperative arrangements), and pure floats. Levy-Yeyati and Sturzenegger (2000a) discuss at length the nature of the mismatches between both classifications. In particular, they show that their number for any given year hovers around 50 percent of all cases. The IMF has recently started to acknowledge the difference between deeds and words by reporting, in some cases, countries with a formal regime and a different de facto one. These regimes are identified by the superscript 6 in IMF (1999). In what follows, we deliberately ignore this distinction when considering the IMF classification and assign countries according to their "legal" arrangement.

II. The Data Our sample covers annual observations for 154 countries over the period 1974-99. A list of countries, as well as the definitions and sources of the variables used in the paper, is presented in Appendix I. With the exception of the political instability and secondary school enrollment variables used in the growth regressions, all of

12The complete database is available at http://www.utdt.edu/~ely or http://www.utdt.edu/~fsturzen. 13The details of the classification appear in the IMF's Annual Report on Exchange Rate Arrangements and Exchange Restrictions. A summary of the classification is included in the International Financial Statistics volumes.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table 2. Distribution of Exchange Rate Regimes Regime First Round Second Round Total IMF Float 473 186 659 459 Intermediate 261 334 595 801 Fix 418 512 930 924 Total 1,152 1,032 2,184 2,184 our data come from the IMF and the World Bank. Data availability varies across countries and periods, so the tests in each subsection were run on a consistent subsample of observations (which is reported in each case along with the results). The LYS de facto classification covers a sample of 2,825 observations, of which 637 are labeled inconclusive in the second round. Table 2 shows the distribution of the remaining 2,188 observations, along with the alternative IMF-based classification for the same group of observations.

III. Inflation and Money Growth

A First Pass at the Data The typical association of fixed exchange rates with lower inflation rates is based primarily on the belief that a peg may play the role of a commitment mechanism for monetary authorities, inasmuch as an expansionary monetary policy is inconsistent in the long term with a fixed exchange rate, and that the failure to comply with the commitment entails some political cost to the authorities (Romer, 1993, and Quirk, 1994). To this effect, which should work entirely through the behavior of the monetary aggregates, the literature adds the potential impact of a credible peg on inflation expectations, which might stabilize money velocity and reduce the sensitivity of prices to temporary monetary expansions. In this way, a fixed exchange rate regime is expected to affect the link between money and prices. Similarly, particularly in those cases in which dollar indexation is widespread, a credible peg may help reduce inertial inflation by placing a limit to devaluation expectations. Table 3 provides a first pass at the data. The table shows the means and medians of inflation for each of our control groups, namely, the floating, intermediate, and fixed exchange rate regimes according to the IMF and the LYS classification (the latter being further disaggregated into first and second rounds). For consistency, the sample of 1,925 observations comprises all countries and years classified by LYS (see Table 2) for which data on inflation and monetary growth are available. Because the sample includes many countries that exhibit extraordi- narily high inflation, it seems reasonable to concentrate the analysis in the medians, which are less affected by such extreme values. For both classifications, the intermediate regimes are the ones that fare the worst in terms of inflation. However, important differences emerge when comparing fixes and floats. Whereas the IMF index seems to indicate, quite

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 3. inflation and Money Growth

IMF LYS LYS (First Round) LYS (Second Round) FLOAT INT FIX FLOAT INT FIX FLOAT INT FIX FLOAT INT FIX

Observations 425 740 760 610 548 767 434 236 356 176 312 411

INFLATION Means 22.3 20.2 16.7 14.2 38.3 9.7 16.1 75.3 11.8 9.4 10.3 7.9 Medians 8.3 9.9 8.7 9.4 12.7 7,4 10.3 40.1 8.5 8.0 7.6 6.4 AM2 Means 24.9 26.3 20.4 19.1 40.6 15.1 21.0 72.3 17.8 14.4 16.7 12.8 Medians 13.8 16.9 14.6 14.9 20.2 12.9 16.3 41.9 14.6 14.0 14.9 11.8

Source: IMF's International Financial Statistics. Note: Exchange rate classifications: IMF de jure from IPS. LYS de facto from Levy-Yeyati and Sturzenegger (2000a). surprisingly, that fixes are associated with slightly higher inflation levels, the result reverses when we group observations according to the LYS classification. This is a logical consequence of the fact that the IMF classification does not distinguish between successful and collapsing pegs, and thus includes within the fix group countries that displayed high inflation levels as a result of inconsistent monetary policies that eventually led to a currency crisis.14 The table also shows that, as expected, second-round observations correspond to lower inflation rates, indicating that this group captures country observations with relatively less volatility. Within this group, inflation decreases monotonically as we move to regimes with less flexibility. As mentioned above, one way a regime (and particularly, a peg) may influence inflation is by imposing discipline on the dynamics of money creation. As expected, the numbers for money growth presented in the table mirror those for inflation. While the IMF index, if anything, seems to indicate that the rate of money growth (AM2) tends to increase more rapidly under fixed than under floating exchange rates, the LYS classification finds the opposite result. Again, in both cases intermediate regimes stand out as the most expansionary, which is consistent with the numbers for inflation.

Inflation These results have to be confirmed by a more careful analysis where we control for relevant additional variables that may also be affecting both inflation and money growth. We start from a standard money demand equation to obtain (1)

Here, n represents the inflation rate, Am is the rate of growth of broad money, &GDP is real output growth, i is the nominal interest rate, v is money velocity, and a and |3

14Note also that the ranking between fixes and floats under the IMF classification changes according to whether the analysis focuses on means or medians.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger are positive constants. As mentioned, the exchange rate regime may affect inflation indirectly through its disciplinary effect on Am, as well as directly through lower inflation expectations. While it is not completely clear how this last channel may be modeled, a first assessment of this "credibility" effect may be obtained by including regime dummies in the money demand equation (1). More precisely, we use a dummy IMFINT (IMFFIX) that takes the value of one when an observation is classified as an intermediate (fixed) regime by the IMF. The dummies LYSINT and LYSFIX are constructed in a similar way from the LYS classification. As additional explanatory variables, we include a measure of the openness of the economy (OPEN) to control for the potential disciplinary effect elicited by international arbitrage, three regional dummies corresponding to Latin American (LATAM), sub-Saharan African (SAFRICA), and transition economies (TRANS), and year dummies.15 Finally, we add the lagged dependent variable (INF1) as a regressor to capture for the effect of past policies on current expectations, as well as to control for the possibility of backward-looking indexation. To reduce the influence of outliers in the econometric test, the sample excludes high-inflation countries, defined as those with annual inflation rates above 50 percent. The results, presented in Table 4, are largely consistent with those sketched in the previous discussion.16 The coefficients for real GDP, money, openness, and interest rate growth (respectively, AGDP, AM2, OPEN, and &INTRATE), as well as lagged inflation, are all highly significant and of the expected sign. Regarding the regime effect, both classifications yield the same result when applied to the whole sample (columns 1 and 2, indicating no significant difference between fixes and floats in terms of inflation rates. However, once we exclude high-inflation countries (defined as those with annual inflation rates above 50 percent), the fix dummy becomes negative and significant (and under the de facto classification, highly so), as shown in columns 3 and 4.17 This finding is confirmed when we exclude intermediates from the sample (column 5). Both results seem to imply that, for low- to moderate-inflation countries, fixed regimes appear to be associated with inflation rates about 1.8 percent lower than floats. Intermediates, on the other hand, display significantly higher inflation. This association, however, does not apply evenly to the sample. In particular, the beneficial influence of fixed regimes on inflation appears to be significant only for low-volatility and nonindustrial countries (columns 7 and 8).18 In short, while there is some evidence of a link between regimes (in particular, pegs) and the inflation rate, this link appears to be more limited than is typically assumed.

15On the inclusion of openness, see Romer (1993). 16Here, as well as for the tests in the remaining sections of the paper, the coefficient of the year dummies are omitted for conciseness. Standard errors reported in the paper are corrected by heteroscedasticity, whenever a White-test rejects the null hypothesis of homoscedasticity. 17The result disappears when we use the alternative (and less stringent) cut-off points for outliers of 100 and 200 percent annual inflation. This may be due to the potential nonlinearities in the relationship between variables. 18See Appendix I for a list of industrial countries. The previous finding is confirmed by splitting the first- and second-round samples into industrial and nonindustrial countries: the regime is significantly and negatively related to inflation only for second-round nonindustrial countries. The results, omitted here, are available from the authors on request.

©International Monetary Fund. Not for Redistribution Table 4. Inflation

(10) LYS« (5) (6) (7) (8) (9) High 0) (2) (3) (4) LYS" LYS" LYS" LYS" LYS" credibility IMF LYS IMF" LYS" (LYSINT =0) First round Second round Industrials Nonindustrials nonindustrials

&GDP -0.76*" -0.75*** -0.22** -0.20** -0.22* -0.08 -0,15* -0.10 -0.26"* 0.29*** 0,14 0,14 0.09 0.09 0.10 0.13 0.08 0.09 0./0 0.09 AM2 0,66*" 0.65**' 0.14" 0.14** 0.14* 0.10 0.18*** 0.02 0.12* 0.14* 0.08 0.08 0.07 0.07 0.08 0.07 0.04 0.03 0.07 0.07 A1NTRATE 5.4!*" 5.20** 3.45** 3.23** 2.84" 2.18* 3.30*** 1.22** 3.13* 2.71 1.98 1.96 1.36 1.35 1.38 1.15 0.98 0.49 1.87 1.96 INF1 0.31*"* 0.31*** 0.20** 0.19" 0.16** 0.17 0.21"" 0.84*** 0.17** 0.19" 0.07 0.07 0.08 ao« 0.08 0.10 0.05 0.04 0.08 0.08 OPEN -1.58 -0.03 •^,28*** -5.30*** -4.20*** -2.93 -6.22*** -0.02 -7.73"* -7.14*** 1.77 1.98 1,50 1.39 1.39 1.78 1.33 0.73 1.91 1.74 IATAM 0.36 0.93 5.16*" 5.15*** 4.37*** 9.12*" 2.89*" 4.79*** 3.16"* 1.07 1.04 0.88 0.83 0.78 1.56 0.67 0.82 0.68 SAFRICA 0.40 1.42 6.43*** 6.15*** 5.35*** 6.83*** 3.56*" 5.65*** 3.90*" 1.49 1.20 7.19 1.12 1.04 1.49 1.02 1.05 O.S9 TRANS 3.29 3.07 3.19 4.00 4.52 9.44" -5.59 5.14 7.78" 3.91 3.95 4.38 4.26 3.83 .3.77 7.28 3.80 3.16

IMFINT -0.59 1.12** 1.03 0.48 1MFFIX 0.30 -1.62* 1.53 0.85 LYSINT 2.12" 1.75"* 7.36"* 0.09 -0.04 1.98" 1.71** 0.92 0.60 1.65 0.51 0.27 0.84 0.83 LYSFIX -1.30 -1.83*** -1.77*** -0.66 -2.03*** 0.06 -2.95*** -2.77*" 0.85 0.48 0.47 0.71 0.61 0.25 0.74 0.71 LYSFIX'INCONC -0.54 0.78

Observations 1,269 1,269 997 997 733 504 493 368 629 778 R2 0.901 0.902 0.518 0.526 0.510 0.582 0.561 0.877 0.475 0.476

Notes: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics. "Sample includes only low- to moderate-inflation countries {annual inflation below 50 percent).

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

The final column of Table 4 addresses an additional issue raised by our exchange rate classification procedure. The de facto methodology leaves unclassified a number of countries that display very little variability in both the nominal exchange rate and reserves. It could be argued that credible fixes are less likely to be tested by the market (hence exhibiting a lower volatility of reserves) and, possibly for the same reason, more likely to exhibit lower inflation rates. If so, by leaving out the so-called "inconclusives," we would be ignoring this credibility dimension and discarding "good pegs," thus biasing the results toward underesti- mating the beneficial effects of fixed regimes on inflation. A natural way to address this concern is to include these "high credibility" pegs in our regressions. Because the de facto approach is silent as to the regime to be assigned to these observations, we simply added to the group of fixers all those de facto inconclusives that did not exhibit changes in their exchange rates. The last column of Table 4 reports the results of our baseline regression, where LYSFIX now represents the expanded group of pegs, and the dummy INCONC takes the value of one whenever an observation was originally classified as inconclusive. A simple comparison of these results with those in regression (4) indicates that the introduction of high credibility pegs does not alter the previous conclusions: all coefficients remain virtually unchanged and, in particular, the coefficient of INCONC, which should capture any additional credibility effect associated with the new pegs, is not significant. Along the same lines, we further refine the tests in Table 4 by distinguishing between long and short pegs, according to whether or not they have been in place for at least five consecutive years. More precisely, we rerun regressions (4)-(9) of Table 4, splitting the fix group into long and short pegs (respectively, dummies LONG and LYSFIX-LONG). As mentioned in the introduction, this enables us to isolate the short-run impact of the implementation of a peg from the effects associated with the permanence of the regime, as well as to focus our attention on those countries capable of implementing sustainable pegs. As Table 5 shows, the distinction is highly relevant. A significant link with low inflation is found only for the group of long pegs, with the exception of industrial countries, for which, as before, regimes exhibit no significant impact.

Endogeneity Underlying the previous tests was the presumption that the adherence to a fixed regime may lead to a lower average inflation rate. However, it is easy to conceive a different argument by which countries with greater price stability have better chances to implement a sustainable peg and, for this reason, are more likely to choose one in the first place. Thus, the finding that pegs are associated with lower inflation, at least for some groups of countries, is subject to a potentially serious endogeneity problem. To address this issue we use a feasible generalized two-stage IV estimator (2SIV) suggested by White (1984), which allows us to correct simultaneously for endogeneity and heteroscedasticity.19 The results are presented in Table 6, where

19The methodology and the additional controls used in the first stage of the estimation procedure are described in detail in Appendix II.

©International Monetary Fund. Not for Redistribution Table 5. Inflation: Long vs. Short Pegs (2) (3) (4) (5) (6) (7) (1) LYSINT^Q First Round Second Round Industrials Nonindustrials High Credibility AGDP -0.21" -0.22" -0.07 -0.15* -0.10 -0.27"* -0.23"* 0.08 0.70 0.13 0.08 0.09 0.09 0.08 AM2 0.14" 0.14* 0.10 0.18*" 0.02 0.12* 0.16" 0.06 0.07 0.07 0.05 0.03 0.07 0.06 MNTRATE 3.18" 2.82" 2.21* 3.23*" 1.22" 3.09* 2.97" 1.33 1.33 1.16 0.99 0.49 1.82 7.40 1NF1 0.19** 0.15" 0.16 0.21*" 0.84*** 0.17" 0.20" 0.08 0.07 OJO 0.05 0.05 0.07 0.08 OPEN -4.06*" -2.46** -1.60 -5.53*" 0.11 -6.00"* -3.98*" 1.25 1.20 7.65 7.47 0.74 7.70 7.75 LATAM 5.45*** 4.76*** 8.83*** 3.34*" 5.28*** 4.28"* 0.85 0.79 1.47 0.75 0.86 0.76 SAFRICA 6.33*** 5.55*** 6.75*** 3.80"* 5.96*** 5.22*" 1.12 1.02 7.42 1.02 7.08 7.00 TRANS 3.43 3.87 8.73" -5.69 4.55 7.32" 4. IS 3.75 3.80 7.21 3.75 3.37

LYSINT 1.67*** 7.36*** 0.07 -0.05 1.94" 1.58"* 0.59 7.65 0.57 0.27 0.84 0.59 LYSFIX-LONG 0.47 0.79 1.90 -0.71 0.24 0.13 0.67 0.81 0.83 1.34 0.76 0.43 7.79 0.72 LONG -3.25™* -3.45"* -2.26"* -2.96*" -0.08 -4.76*** -3.16*" 0.62 0.59 0.77 0.80 0.21 O.S8 0.64 Observations 997 733 504 493 368 629 1146 R2 0.537 0.533 0.593 0.568 0.878 0.491 0.528 Notes: *", **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics. Sample includes only low- to moderate-inflation countries (annual inflation below 50 percent).

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table 6. Inflation: Accounting for Endogeneity0 (iy (1)* Long Pegs &GDP -0.26"* -0.25*** 0.10 0.09 AM2 0.18** 0.17" 0.08 O.OS &INTRATE 3,48" 3.07** 1.49 1.47 INF I 0.19" 0.18** 0.09 O.OS OPEN -5.99*** -3.59*** 1.56 1.18 IATAM 2.07'" 2.84"* 0.63 0.72 SAFR1CA 3.22*** 4.02*" 0.87 0.96 TRANS 6.12* 6.80** 3.30 3.20

LYSFIX 1.06 0.87 LONG -2.13" 0.97 Observations 851 851 Note: ***, **, and * represent 99,95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics. "The sample includes high credibility pegs. ^Instruments: FIXFIT, where FIXFIT is the estimate of LYSFIX in a logit model over the sample excluding intermediates. "Instruments: LONOFIT, where LONGFIT is the estimate of LONG in a logit model over the sample excluding intermediates. we apply this correction, in turn, to assess the inflation effect of conventional fixed regimes and long pegs. As can be seen from the table, only the impact of long pegs on inflation survives the endogeneity correction. This confirms that the negative link between inflation and pegs is weaker than casual observation seems to reveal, and appears to be largely confined to the case of long-standing pegs.

Money Growth At the beginning of this section we mentioned that a typical argument supporting the connection between pegs and inflation points to the presence of a disciplinary effect on monetary policy. According to this view, de jure pegs, inasmuch as failing to comply with the legal commitment entails a significant political cost, should result in lower rates of money growth. The same can be said of de facto pegs. To test this hypothesis, we run cross-section regressions of money growth on the regime dummies and the following additional explanatory variables: real GDP

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 7. Money Growth (3) (4) (5) (1) (2) LYS LYS LYS IMF LYS Long pegs Industrials Nonindustrials SUPODP -29.59" -28.65* -26.88* -17.74 -20.21 14.14 14.75 14.75 11.04 77.40 AGDP1 0.44*" 0.44"* 0.44*" 0.50" 0.09 0.11 0.12 0.77 0.24 0.72 AM21 -3.90 -3.89 -4.32 28.97*" -11.25 14.32 14.40 74.39 8.38 14.87 OPEN -9.12"* -9.69*** -8.48*** 5.53 -18.42"* 3.14 2.46 2.80 5.80 4.22 LATAM 10.91"* 10.86"* 11.07*" 7.10*" 1.88 2.13 2.10 7.76 SAFRICA 8.54*" 7.71"* 7.83"* 3.13" 1.67 1.45 7.42 7.57 TRANS 13.17" 14.59" 13.88" 12.11* 6.64 6.67 6.87 6.54 IMFINT 4.05"* 7.52 IMFFIX -0.10 1.66 LYSINT 2.75"* 2.67*" 0.34 2.41* 1.00 7.07 7.45 7.42 LYSFIX -1.74 7.29 LYSFIX-LONG 0.40 0.81 -0.71 1.94 1.64 2.68 LONG -3.05** -0.80 -3.77* 7.47 1.38 2.09 Observations 991 997 997 368 629 J?2 0.141 0.137 0.141 0.247 0.149 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics. growth (AGDP7, lagged to reduce potential endogeneity problems), openness (OPEN), the ratio of the fiscal surplus to GDP (SUPGDP), the three regional dummies (LATAM, SAFRICA, and TRANS), and the lagged dependent variable (AM27).20 Our results, reported in Table 7, offer partial support for the hypothesis of the existence of a disciplining effect on money growth. Using either the IMF classification (column 1) or the de facto classification (column 2), the fixed regime dummy has the expected negative sign but is not significant. However, a significant relationship is detected when we look at long pegs separately (column 3), a result driven, once again, by the group of nonindustrials (column 5). Thus, for the group of long pegs, the regime has an effect on inflation through both enhanced credibility and a disciplining effect on monetary policy.

20Additional tests were run including the change in interest rates (&1NTRATE), lagged inflation (INFl), the change in government consumption (GOVI), and the ratio of government consumption to GDP (GOVGDP), with similar results.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Deeds vs. Words

The mismatch between the IMF and the LYS classification, and in particular the fact that in the past numerous countries repeatedly adopted de jure fixed regimes without implementing consistent monetary policies, opens the question of whether, for a given monetary policy, the announcement of a peg brings by itself a benefit in terms of lower inflation, thus providing a potential motivation for this seemingly inconsistent behavior. In the "deeds" regression (Table 8, column 1), we control for the announced (de jure) regime, including the dummy FIXFIX that takes the value of one for observations identified as pegs by both classifications. In this way, we test whether the actual behavior of the economy (deeds) has any additional effect on inflation, above and beyond that resulting from the announcement of a peg. The coefficient of FIXFIX is highly significant and negative, suggesting that countries that announce a peg but in practice let the exchange rate fluctuate exhibit higher inflation levels, an unsurprising result that simply confirms the inflationary impact of (partially) unanticipated devaluations. Regarding words, in addition to controlling for the de facto regime (distinguishing between long and short pegs), we include two interaction terms that identify observations within each group that are also classified as de jure pegs. This allows testing whether the actual announcement of a peg (words) has any additional effect on inflation, above and beyond that resulting from the actual behavior of the economy. As can be seen in column (2) of Table 8, the announcement only lowers inflation rates for the case of long pegs. The "deeds vs. words" comparison indicates that, for inflation, deeds appear to play a more important role. De jure pegs that do not behave as real pegs are obviously associated with higher inflation, because the announcement of a peg has no value in the face of recurrent devaluations. On the other hand, within de facto pegs, words appear to have no value in terms of inflation unless the country behaves in a manner consistent with maintaining a peg.

IV. Interest Rates While much has been said of the impact of exchange rate regimes on real wages and employment, there is surprisingly little work on their effect on the cost of capital, which accounts for a larger share of production costs in most countries. Quite possibly, one reason for the scarcity of research on the issue is the difficulty in obtaining reliable interest rate data for a reasonably large number of countries, as in many cases interest rates were largely administered and thus unrepresentative of actual market rates.21 On the other hand, episodes of very high inflation are typically characterized by negative real rates, as the banking sector sometimes is not allowed to fully accommodate extremely high inflation expectations. Moreover, in a context of rapidly changing expectations, a small mismatch between the time at which inflation and the nominal interest rate are measured may derive in sizable distortions.

21This problem is particularly acute before the 1990s.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 8. Inflation: Deeds vs. Words (1) (2) Deeds Words AG£>P -0.22*** -0.20** 0.09 O.OS AM2 0.13** 0.13** 0.06 0.06 MNTRATE 3.23** 3.23** 7.50 7.27 INFl 0.20** 0.19** O.OS O.OS OPEN -2.74** -2.74** 1.37 7.22 LATAM 6.43*** 6.37*** 1.00 0.99 SAFR1CA 7.25*** 7.00*** 7.25 7.20 TRANS 4.07 4.24 4.10 4.08

IMFINT 1.15** 0.47 IMFFIX-FIXFIX 0.52 0.85 FIXFIX -4.60*** 7.09 LYSINT 1.59*** 0.5S LYSFIX-WNG 0.92 0.88 LONG -1.20** 0.57 IMFFIX*(LYSFIX-LONG) -3.09 2.29 {MFFrX*LONG -3.91 *** 7.00 Observations 997 997 R2 0.539 0.547 Note: ***, **, and * represent 99,95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics. The regression sample includes only low- to moderate-inflation countries (annual inflation below 50 percent).

Measurement errors aside, the channels through which the exchange rate regime may influence the real rate are by no means obvious. Legal pegs are a case in point. Whereas they are prone to exhibit a "peso problem" that increases real interest rates, pegs on the other side may reduce inflation expectations and thus nominal (and real) rates. More in general, the real rate should depend on the same fundamentals that determine the level of country risk that typically represents a lower bound for all domestic lending rates. Thus, a relatively larger amount of liquid international

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger reserves, a buoyant economy, or a low level of indebtedness should help reduce the cost of capital for the economy inasmuch as open international markets tend to equalize the funding cost of countries of the same risk class. On the contrary, as long as there is some imperfect substitutability between domestic and foreign assets, increases in the government financing needs may crowd out domestic resources, pushing the domestic real rate higher. Alternatively, sluggish growth may provide incentives for short-term expansionary monetary policies with a view to lowering domestic financing costs. With all these caveats in mind, we attempted to explore the issue using a relatively broad specification that captures some of the factors mentioned above. Thus, we include lagged GDP growth rate (AGDP7) to control for incentives to use monetary policy to lower the real rate, the ratio of net interest payments over GDP (INETGDP) as a proxy for the level of debt, the degree of openness (OPEN) to control for international arbitrage constraints, and the ratio of fiscal surplus over GDP (SUPGDP) as an (inverse) measure of government crowding out. We also included current inflation (INF) to control for potential measurement error due to differences in the sampling time or to financial repression, as well as the three regional dummies.22 Table 9 presents the results for the IMF classification. Given that this exercise has not been undertaken with either classification, we study alternative specifications for both. Using the IMF classification, we found that real rates are significantly lower under pegs, while both intermediate and floating regimes do not differ from each other (column 1). These results are even stronger during the 1990s, both for the whole sample and for nonindustrial countries (columns 3 and 5). However, the regime dummies are not significant during the 1970s and 1980s, either for the whole sample or for nonindustrial countries (columns 2 and 4), probably reflecting substantial measurement errors during those years. The previous results may be a consequence of the inclusion of failed pegs among the fixers in the IMF classification. More precisely, unexpected devaluations may induce negative real interest rates in the aftermath of the realignment of the nominal exchange rate. This hypothesis is consistent with a series of findings. First of all, the result applies to first-round (high volatility) observations (column 6) but not to second-round (low volatility) observations (column 7). Moreover, no systematic link is detected when using the LYS classification (Table 10).23 This hypothesis is further confirmed by the results of Table 11, which indicate that while de jure fixes display significantly lower interest rates in general, the impact appears to be stronger for those of them that in practice let the exchange rate fluctuate (as shown by the coefficient of the regime dummy IMFFIX-FIXFIX in column 1). Alternatively, when looking at de facto regimes, we find an effect on real interest rates only for the case of short pegs (columns 2 and 3). Note that the last result is consistent with the presence of an announcement effect on inflation expectations, combined with the failure of short pegs to lower inflation as revealed in the previous section.

"Other proxies for country risk, such as the ratio of reserves over GDP and over the monetary base, and interest debits over exports, were also tested and found to be not significant. 23A negative link appears in the second-round regressions (columns 4 and 6).

©International Monetary Fund. Not for Redistribution Table 9, Real Interest Rates: IMF Classification (4) (5) (2) (3) Nonindustrials Nonindustrials (6) (7) (1) 1974-89 1990s 1974-89 1990s First Round Second Round INETGDP 1.54 -3.93 0.16 0.02 0.05 0.83 -2.74 2.33 73.72 2.78 75.66 2.27 3.32 2.32 AGDP7 0.21*** 0.13 0.23*** 0.17 0.29*** 0.27"* 0.12** 0.06 0.70 0.07 0.72 0.08 0.09 0.06 INF -0.08*** -0.08*** -0.08*** -0.08*** -0.08*** -0.06*** -0.45*** 0.02 0.02 0.03 0.02 0.03 0.07 0.05 OPEN 2.23 -0.12 2.37 1.75 3.24 2.69 -2.65** 7.37 2.42 7.83 2.89 2.73 2.37 7.32 SUPGDP -2.41 -6.21 4.56 -8.41 3.95 14.10 -17.28*** 5.60 7.69 7.68 9.54 70.20 9.34 4.63 LATAM -0.24 -1.27 0.20 -0.55 0.39 -1.92" 3.57*** 0.57 0.97 0.72 7.07 0.80 0.89 0.56 SAFR1CA -2.94*** -5.02*** -0.83 _4.41*» -0.57 -3.73*** -0.44 0.70 0.94 7.07 7.07 7.05 0.94 0.59 TRANS -4.07* -3.43 -3.98* -3.37 -5.11 2.76 2.78 2.70 2.3/ 3.37

IMFINT -0.10 0.06 0.43 0.42 0.39 -1.37* 2.54*** 0.48 0.65 0.65 7.80 0.98 0.73 0.53 IMFFIX -2.20*** -0.71 -2.85*** -0.55 -2.84*" -4.23*** -0.92 0.62 0.95 0.85 7.87 7.07 1.04 0.67 Observations 981 493 488 319 351 497 484 #2 0.272 0.304 0.191 0.283 0.220 0.301 0.484 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics.

©International Monetary Fund. Not for Redistribution Table 10. Real Interest Rates: LYS Classification (5) (6) (2) (3) (4) Second Round Second Round (i) Long Pegs First Round Second Round 1974-89 1990s INETGDP 1.68 2.00 1.03 ^t.65** -30.32*** -1.50 2.34 2.37 3.40 2.34 8.47 2.69 AGDP1 0.20*** 0.20*** 0.25*" 0.10* 0.03 0.19** 0.06 0.06 0.09 0.06 0.08 0.09 INF -0.07*™ -0.07*" -0.05*** -0.39*** -0.47*** -0.30*** 0.02 0.02 0.07 0.05 0.05 0.08 OPEN -0.26 -0.49 -1.83 -4.04*** -10.01*" -0.07 7.24 7.25 2.75 7.29 7.74 2.04 SUPGDP -1.34 -1.60 15.38 -17.81*** -24.56*** -15.01** 5.68 5.73 9.87 4.74 6.70 7.75 LATAM -0.71 -0.76 -2.29*** 3.15*" 2.81** 3.14"* 0.57 0.57 0.88 0.6S 7.72 0.87 SAFRICA -3.55"* -3.60*** -3.70*** -1.14* -1.33 -2.14*** 0.68 0.69 0.90 0.66 0.90 0.70 TRANS ^.40* -4.32* -3.18 -5.02 -6.70 2.28 2.27 2.54 4.32 4.73

LYSINT -0.53 -0.52 -3.09** 0.29 0.19 0.62 0.50 0.50 7.24 0.46 0.58 0.73 LYSFIX 0.16 0.47 LYSFIX-LONG -0.48 1.01 -1.72" -1.76* -1.07 0.75 1.24 0.80 0.97 7.79 LONG 0.53 0.31 -1.42" -0.54 -1.88* 0.51 0.77 0.68 0.87 7.02 Observations 981 981 497 484 272 212 R2 0.259 0.260 0.294 0.438 0.551 0.271 /7-values (Wald) 0.209 0.592 0.745 0.258 0.564 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 11. Interest Rates: Deeds vs. Words 0) (2) (3) Deeds Words Words II INETCDP 2.30 1.33 1.78 2.42 2.37 2.36 AGDPI 0.21"* 0.20"* 0.19*** 0.06 0.06 0.06 INF -0.08"* -0.08"* -0.08"* 0.02 0.02 0.02 OPEN 1.95 0.38 -0.26 7.37 1.30 7.25 SUPGDP -2.62 -1.58 —2.32 5.65 5.62 5.72 LATAM -0.49 -0.41 -0.42 0.60 0.67 0.60 SAFRICA -3.12"* -3.33"* -3.32*" 0.77 0.77 0.70 TRANS -4.30" -4.18* -3.91* 2.77 2.24 2.72

IMFINT -0.10 0.48 IMFFIX-FIXFIX -2.70*" 0.69 LYSINT -0.57 -0.53 0.50 0.50 LYSFK-FIXFIX 0.77 0.54 FIXFDL -1.43* -0.68 0.77 0.69

LYSF1X-LONG 1.06 0.74 LONG 0.55 0.66 fMFFIX*(LYSFIX~LONG) -6.89"* /.76 IMFFDCLONG -0.21 0.84 Observations 981 981 981 7?2 0.274 0.261 0.275 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics. The regression sample includes only low- to moderate-inflation countries (annual inflation below 50 percent).

VI. Growth The literature has not considered the exchange rate regime as an important deter- minant of growth performance. This is probably due to the fact that we tend to associate only nominal effects to the choice of nominal variables. However, several arguments have been advanced to suggest a link between the two.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

On the one hand, by reducing relative price volatility, a peg is expected to foster growth through its positive effect on investment and trade. Moreover, lower price uncertainty should lead to lower real interest rates, contributing to the same effect. On the other hand, the lack of exchange rate adjustments under a peg, coupled with some degree of short-run price rigidity, may result in price distortions and high unemployment in the face of external shocks. More important, the need to defend a peg in the event of negative external shocks entails a significant cost in terms of real interest rates, as well as increased uncertainty as to the sustainability of the regime. Calvo (1999) has suggested that the external shocks faced by a country are not independent of the exchange rate regime. Not surprisingly, as pointed out in Fischer (2001), all the countries that suffered from a currency crisis had fixed exchange rate regimes. However, while both the lack of adjustment argument and the frequent external shocks that characterize a peg imply a higher expected output volatility, their consequences in terms of long-run growth are less straightforward. At an empirical level this relationship has been studied in a series of recent papers. Mundell (1995), for example, examines the growth performance of industrial countries before and after the demise of Bretton Woods, finding that the earlier period, characterized by the prevalence of fixed exchange rates, was associated with faster average growth. Ghosh and others (1997), using all IMF reporting countries for the period 1960-90, fail to find systematic evidence of an impact of the type of regime on growth. However, these results are challenged by Rolnick and Weber (1997), who find, using long-term historical data, that output growth was higher under fiat standards compared with commodity (e.g., gold) standards. A similar conclusion is reached by Levy-Yeyati and Sturzenegger (2000b), who explore the relationship between exchange rate regimes and growth using annual data covering the period 1974-99. In a nutshell, their main findings are the following: 1. Fixed exchange rate regimes are associated with a lower per capita output growth rate. The estimates range from 0.7 percent to 1 percent a year according to the specification. This result remains robust to alternative specifications of the model, including a correction for endogeneity. The previous result is driven by nonindustrial economies. For industrial economies the exchange rate regime is not related to growth performance. 2. Similarly, fixed exchange rate regimes are associated with higher output volatility only in the case of nonindustrial countries; they have no significant impact on volatility within the group of developed economies. Levy-Yeyati and Sturzenegger (2000b) cast a relatively negative light on pegs. If policymakers worry about inflation and exchange rate stability because of their potential negative impact on economic growth, it appears that the beneficial effect of a peg in terms of price stability does not translate in the end into a stronger growth performance. In this paper, we build on these results to explore two additional issues. First, we evaluate the announcement value of regimes in terms of growth performance, in light of the argument that legal pegs are more vulnerable to external shocks and

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE speculative attacks that ultimately undermine their growth performance. In turn, in the next section, we examine whether the negative growth performance found to be associated with fixed regimes remains even when we focus on the subgroup of hard pegs as opposed to conventional fixes.

Fear of Pegging The experience of the financial crisis of the 1990s has placed in the forefront of the exchange regime discussion the increasing vulnerability of fixed regimes to speculative attacks and financial contagion. This may be behind a phenomenon that could be labeled "fear of pegging," namely the practice of de facto running a peg while avoiding a commitment to a fixed parity and the potential vulnerability to attacks that a legal peg may introduce.24 This issue is addressed in Table 12, where we test the consequences of the announcement of a peg on growth performance, after controlling for the de facto regime. Interestingly, when we split the group of pegs into short and long, we find that only the former are negatively affected by the announcement. This result seems to suggest that a commitment to a fixed parity increases the vulnerability of the country, except in those cases in which the regime has been in place long enough to strengthen its credibility. Thus, the evidence provides some support for the view that the adoption of a legal peg entails increased vulnerability, a fact that may underscore the finding of "fear of pegging."

VII. Are Hard Pegs Different? The LYS classification works on the basis of facts, distinguishing between the broadly defined groups of fixed exchange rates, intermediate regimes (crawling pegs and dirty floats), and pure floats. However, some analysts, notably Eichengreen (1994) and, more recently, Fischer (2001), have argued in favor of the relative merits of extreme exchange rate regimes, drawing a line between conventional fixes and "hard pegs" that exhibit a stronger commitment to a fixed parity (as in a currency board) or directly relinquish control over their own currency (as in the case of countries with no separate legal tender). More precisely, it has been argued that, if the benefits of pegging accrue from increased credibility, conventional fixes may fall short on this ground and the stronger commitment that char- acterizes hard pegs may be necessary. In this section, we test whether and in which way this hypothesis is consistent with the data. To do so, we have a closer look at the group of hard pegs, defined as those countries classified by the IMF as having either a currency board or no separate legal tender.25

24The case of El Salvador before the recent attempt at full dollarization is a clear example of a country that claims to be running a flexible regime while keeping the exchange rate constant. On this, see Levy- Yeyati and Sturzenegger (2000a). 25The only exception to this rule are the countries within the African franc zone in 1994, the year in which they devalued their currency by 100 percent and, as a result, are classified as intermediates according to LYS. Thus, all observations in our group of hard pegs are de facto classified as either fixes or inconclusives. See Appendix I for a list of countries in this category.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table 12, GDP Growth: Fear of Pegging INVGDP 8.96*** 1.92 POPGR -0.43" 0.15 GDPPC74 -0.36*** 0.11 GOV1 -1.15*" 0.40 SECB -0.89 0.95 CIVIL -0.25* 0.12 ATI 5.17*** 1.03 OPEN -0.10 0.82 LATAM -0.87" 0.34 SAFRICA -0.83* 0.47 TRANS -1.35 1.75

LYSINT -1.01*** 0.29 LYSFIX-LONG -0.61* 0.35 LONG -0.22 0.45 IMFFI^LYSFIX-LONG) -1.11* 0.66 IMFFDCLONG -0.83 0.63 Observations 1,349 Ri 0.206 p-value (Wald) 0.023 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics.

Before we present the econometric results, a few comments are in order. First, as many authors before us have stressed, with the exception of the European Economic and Monetary Union (EMU), and possibly Argentina and Hong Kong SAR, countries with hard pegs are relatively small. Moreover, the biggest countries in the group (Argentina, Bulgaria, Estonia, Ecuador, El Salvador, Lithuania, and countries in the euro area) have adopted a hard peg relatively recently, and in many cases there is not sufficient data to test the impact of the new regime empirically. Second, most of the hard pegs for which there are data to conduct econometric tests have been around for a long enough time to dispel concerns about potential endogeneity problems. With the exception of Argentina and Bulgaria, all of the remaining countries in the list have had the same regime in place over the whole

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE period included in our sample. On the other hand, long-standing currency boards include only Hong Kong SAR, Djibouti, and Brunei, which seriously limits the possibility of conducting meaningful tests of this type of regime in a separate way. In view of the above, in what follows we treat hard pegs as a single group without discriminating according to their different varieties, bearing in mind the limits of extrapolating the experience of small countries and island economies to the rest of the sample. Moreover, because EMU observations are excluded from the LYS classification, no industrial country in our sample is classified as a hard peg. Therefore, we restrict our tests to the subgroup of nonindustrial economies. Tables 13 and 14 offer a rough pass at the data by comparing the means and medians of the inflation rate, money growth (AM2), and the rate of growth of real per capita GDP (AGDPPC), for nonindustrial countries as a whole and for the subgroup of hard pegs.26 Simple inspection indicates that, while hard pegs appear to exhibit much lower inflation and money growth levels, their growth performance does not differ from the group of nonindustrial pegs as a whole. To assess the relative merits of hard pegs in terms of inflation, we run econometric tests similar to those in Tables 4 and 5, this time including a hard peg dummy (HARDPEG), To compare with previous results in the literature, see Ghosh and others (2000), we first run the inflation regression including only the hard peg dummy. Column 1 in Table 15 confirms the negative correlation between hard pegs and inflation present in Table 13. Furthermore, as column 2 shows, hard pegs have an additional disinflationary effect relative to conventional fixes, a result that remains once we expand the sample to include high credibility pegs (column 3).27 In line with the results of Section III, we replicate the previous three regressions dividing pegs into three mutually exclusive groups: short pegs, long conventional pegs (i.e., long pegs that are not hard pegs), and hard pegs (identified by the dummies LYSFIX-LONG and LONG-HARDPEGS).2* As columns 4-6 show, we can conclude in all cases that, while both hard and long (conventional) pegs reduce inflation, the former have a significant additional disinflationary effect, as indicated by the p-value of a Wald test of the equality of the coefficients of both dummies reported in the last row of the table. Short pegs, as before, appear to be ineffectual. Turning to growth, we run regressions for pooled annual data as well as for a cross section of countries. In both cases, the specification, taken from Levy-Yeyati and Sturzenegger (2000b), includes the following additional controls:29 the investment to GDP ratio (INVGDP), the degree of openness (OPEN), the growth of government consumption (GOV1; lagged to avoid endogeneity problems), per capita GDP at the beginning of the period (GDPPC74; computed as the average over 1970-73), the degree of initial secondary enrollment (SEC), population growth (POPGR), and a measure of political instability (CIVIL). The literature suggests a

26The sample of Tables 13 and 14 comprises all nonindustrial observations for which data are available. 27Because the LYSFIX dummy already includes the hard pegs, the new dummy captures the differential effect associated with the presence of a hard peg. 28Note that all hard peg observations also belong to the group of long pegs, with the sole exception of the Bulgarian currency board (1998-99), which is therefore excluded from these regressions. 29Controls were chosen from those variables suggested by the growth literature as the most systematically significant determinants of growth. Several additional control variables were also tested, with similar results.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table 13. Inflation and Money Growth: Conventional and Hard Pegs (Nonindustrial countries) LYS FLOAT INT FIX HARDPEGS LONG Observations 409 445 650 363 491 INFLATION Means 17.6 45.2 10.6 5.7 8.2 Medians 11.5 19.3 8.3 4.0 7.3 &M2 Means 23.1 47.3 16.3 9.7 13.7 Medians 18.3 24.2 14.5 8.5 12.7 Source: IMF's International Financial Statistics. Note: Exchange rate classifications: IMF de jure from IPS, LYS de facto from Levy-Yeyati and Sturzenegger (2000a).

Table 14. GDP Growth: Conventional and Hard Pegs (Nonindustrial countries) LYS FLOAT INT FIX HARDPEGS LONG Observations 413 458 744 433 569 DGDPPC Means 1.9 0.6 1.4 1.5 1.4 Medians 2.1 1.1 1.3 1.3 1.3 Source: IMF's International Financial Statistics. Note: Exchange rate classifications: IMF de jure from IPS, LYS de facto from Levy-Yeyati and Sturzenegger (2000a). positive sign for investment, openness, and education variables, and a negative sign for the government consumption (associated with a less productive use of resources), the measure of freedom (where a higher number implies less freedom), and population growth. A negative sign of the coefficient of initial GDP would be consistent with the presence of conditional convergence (Barro and Sala-i-Martin, 1995). We also control for changes in the terms of trade (AT/), as another source of variation in GDP, which is usually absent in cross section analysis but may play a role in annual data (Broda, 2000). Finally, we include regional and year dummies. Table 16 shows the results for pooled annual data. The control variables behave largely as expected: Real growth is positively correlated with investment and negatively correlated with government consumption, population growth, and (albeit weakly) the political instability variable. The link is less clear in the case of openness and initial secondary enrollment, in contrast with what is usually suggested in the literature.30 Changes in the terms of trade display the correct sign and are highly significant. Finally, the sign for the initial per capita GDP is negative, indicating the presence of conditional convergence.

30Levine and Renelt (1992) have already cautioned on the robustness of the coefficients of these variables.

©International Monetary Fund. Not for Redistribution Table 15. Inflation: Are Hard Pegs Different? (Nonindustrial countries) (3) (6) (1) (2) LYS (4) (5) LYS" LYS LYS High credibility LYS« LYS" High credibility &GDP -0.24" -0.23" -0.26"* -0.26*** -0.25"* -0.26*" 0.10 0.10 0.09 0.09 0.70 0.09 AM2 0.13* 0.12* 0.14* 0.11* 0.11* 0.13* 0.07 0.07 0.07 0.07 0.07 0.07 MNTRATE 3.45* 3.05* 2.58 1.78 1.77 1.14 1.87 1.84 7.90 1.35 7.33 7.35 1NF1 0.18** 0.17" 0.19" 0.19" 0.19" 0.21** O.OS 0.08 O.OS 0.09 0.09 0.70 OPEN -9.37"* -7.81"* -6.89*** -5.46"* -5.83"* -5.55"* 1.98 1.88 1.61 7.67 7.73 7.57 LATAM 4.19*** 5.03*** 3.82*** 4.98"* 5.16*" 4.00*** 0.78 0.82 0.72 0.92 0.92 0.82 SAFRICA 5.06*™ 5.96*** 4.72*** 5.68"* 5.94*** 4.76*" 1.01 1.06 0.94 7.76 7.75 1.03 TRANS 5.27 6.39* 10.33*** 7.05"* 7.78*** 9.98"* 3.69 3.65 3.16 2.52 2.57 2.59

LYSJNT 2.05** 1.85" 1.97** 1.78" 0.84 0.83 0.84 O.S3 LYSFIX -2.27*** -1.99*** 0.79 0.68 LYSFIX-LONG 0.25 0.09 7.76 0.93 LONG-HARDPEGS -5.07*** ^t.26*" -3.59"* 1.04 0.98 0.89 HARDPEGS -5.88*** -4.15"* -5.01*" -6.86*** -6.17"* -6.83"* 1.00 7.05 0.88 7.04 0.98 1.02 Observations 629 629 778 628 628 776 R2 0.459 0.483 0.494 0.499 0.505 0.513 p- values (Wald) 0.04 0.03 0.00 Notes; *", **, and * represent 99,95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics. "Excluding Bulgaria.

©International Monetary Fund. Not for Redistribution Table 16. GDP Growth: Are Hard Pegs Different? (Nonindustrial countries)

(7) (4) (5) (6) High Credibility (3) Fixed vs. Floats vs. High Credibility Pegs (1) (2) LYSINT =0 Hard Pegs Hard Pegs Pegs LYSINT =0 INVGDP 10.30**' 10.35"* 6.64" 7.16" 8.04 1 1 .77*** 9.82"* 2.45 2.45 2.66 2.99 5.04 2.75 2.27 POPGR -0.43" -0.43** -0.42" -0.20 -0.50" -0.40*** -0.35** 0.17 0.17 0.77 0.28 0.20 0.76 0.76 GDPPC74 -0.52' -0.53* -0.33 -1.39"* 0.35 -0.77** -0.80" 0.28 0.28 0.33 0.57 0.43 0.32 0.40 GOV1 -1.35*** -1.23*** -0.36 -0.49 -0.47 -1.22*** -0.31 0.41 0.42 0.79 7.27 0.96 0.47 0.73 SEC 0.52 0.18 -0.50 2.00 -2.08 1.08 1.21 1.31 1.32 1.61 2.36 2.75 7.30 7.55 CIVIL -0.24* -0.24* -0.14 -0.11 -0.07 -0.13 -0.01 0.14 0.14 0.7(5 0.23 0.20 0.72 0./3 ATI 5.22'" 5.27*™ 5.47"* 6.08*" 5.69*** 5.51*** 5.78*** 1.08 /.OS 1.28 1.46 2.01 7.02 7.77 OPEN -1.93* -1.28 -0.06 -0.01 -0.57 -2.16" -1.78 1.16 7.78 7.3S 7.02 2.33 0.82 1.21 LATAM -1.18*™ -0.97*" -0.71* -2.16" -0.48 -0.82** -0.51 0.35 0.36 0.42 0.89 0.52 0.34 0.40 SAFRICA -0.97™ -0.79 -0.26 -2.02" 0.11 -1.09" -0.73 0.48 0.50 0.52 0.99 0.72 0.46 0.49 TRANS -1.97 -2.17 -0.23 0.39 -2.11 -0.30 0.82 0.82 0.82 0.82 1.74 0.82

LYSINT -1.17"* -1.17*** 0.39 0.39 LYSFIX-HARDPEGS -1.05*" -1.25*** -0.85" -0.94*** 0.40 0.41 0.35 0.36 HARDPEGS -0.16 -0.97 -1.46" -0.19 -1.68** -1.00* -1.19" 0.70 0.73 0.72 0.76 0.83 0.53 0.53 Observations 962 962 652 353 386 1,185 875 2 ff 0.191 0.200 0.191 0.261 0.172 0.203 0.203 p-values (Wald) 0.909 0.762 0.767 0.632 Note: "*, **, and * represent 99, 95, and 90 percent significance, respectively. Heteroscedasticity-consistent standard errors are in italics.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Regarding specifically the impact of exchange rate regimes, we start again by comparing the growth performance of hard pegs against the rest of the sample, without discriminating among different regimes.31 As can be seen, the results seem to suggest that hard pegs are no different from (and in particular, do not trail) other regimes. However, once we refine this specification to discriminate between particular regimes, a different picture emerges: Hard pegs cannot be distinguished from either conventional pegs or flexible regimes (column 2).32 Moreover, repeating the regression on a sample that excludes intermediates (column 3), we find that hard pegs, while still similar to other pegs, grow significantly more slowly than floats. To resolve this apparent ambiguity, in columns 4 and 5 we compare hard pegs, separately, with conventional pegs and with the group of floats. The results confirm those in column 3: hard pegs appear to be similar to conventional pegs and to significantly trail floats. The same conclusion is reached when we rerun the regressions on an extended sample that includes "high credibility" pegs (columns 6 and 7). In view of these findings, it is not surprising to find that hard pegs do not differ significantly from other long pegs and that both groups are associated with slower growth than floats (Table 17). The results of the regressions presented in Table 18 point in a similar direction, where the dependent variable is now the average growth rate throughout the whole period. In this exercise, investment, population growth, government consumption, civil liberties, and openness are now averaged over the sample period, while initial GDP levels and secondary enrollment are again measured at the beginning of the period.33 The construction of a dummy to represent the "average" regime for each country is problematic given that many countries changed regimes during the period under study. As a compromise solution, we use the dummy FIX50 to identify countries that are classified as pegs more than 50 percent of the time. We define dummies LONG50 and HARD50 similarly. As before, this leaves us with pegs divided into three mutually exclusive groups: those characterized by recurrent but short-lived pegs (FIX50-LONG50), those that implemented long (but not hard) pegs during most of the period (LONG50-HARD50), and hard pegs (HARD50).34 The results, presented in Table 18, largely mirror those in the previous table. A preliminary specification with HARD50 as the only regime dummy fails to find any difference between hard pegs and the rest. However, a more complete specification that includes the variable F1X50 to control for the presence of conventional pegs reveals that hard pegs trail in a statistically significant manner the growth performance of floats by approximately 1 percent a year (column 2). Moreover, a Wald test does not reject the null that the coefficient of HARDPEGS is equal to that of conventional pegs. Even when long pegs are singled out

31This exercise intends to replicate the specification tested in Ghosh and others (1999) for the case of currency boards. 32The last line of Table 16 shows the p-value of a Wald test of the null that the coefficient for hard pegs (i.e.. the sum of the coefficients of the FIX and HARDPEGS dummies) is equal to zero. 33The specification excludes the annual change in the terms of trade. 34Note that countries that implemented hard pegs relatively recently are not identified as fixes for the purpose of this exercise.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table 17. GDP Growth: Long vs. Hard Pegs (Nonlndustrlal countries) (1) (2) Fixed Float vs. Long INVGDP 7,29** 6.38" 2.99 2.95 POPGR -0.21 -0.27* 0.25 0.16 GDPPC74 -0.13" 0.26 0.05 0.40 GOV1 -0.33 -0.80 1.23 0.90 SEC 2.38 -1.19 2.41 1.91 CIVIL -0.12 -0.09 0.25 0.17 &TI 6.12*** 6.96'** 1.46 1.52 OPEN -0.69 -0.33 1.76 1.62 IATAM -2.45*** -0.77 0.87 0.48 SAFRICA -2.05" 0.07 0.97 0.5S TRANS 0.61 7.27

LONG-HARDPEGS 0.71 -1.12* 0.59 0.61 HARDPEGS 0.13 -1.91** 0.76 0.76 Observations 353 529 «2 0.263 0.196 Note: ***, **, and * represent 99,95, and 90 percent significance, respectively. Heteroscedasticity- consistent standard errors are in italics.

(column 3), hard pegs are still associated with slower growth than floats. However, the coefficients of the regime dummies seem to suggest that the stronger the commitment to a peg, the weaker its cost in terms of growth. Indeed, while a Wald test fails to distinguish hard pegs from other long pegs, it does indicate that the former exhibit significantly stronger growth than countries with frequent short-lived pegs.35

35More precisely, the p- value of the null that the coefficients for hard and long pegs are equal is 0.45, while the p-value corresponding the null that hard pegs and short pegs do not differ is 0.076.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table 18. GDP Growth: Are Hard Pegs Different? (Nonlndustrial countries, period averages) (2) (3) Conventional vs. Long and (1) Hard Pegs Hard Pegs INVGDP 8.66" 7.52™ 7.24" 3.72 3.2S 3.2S POPGR -0.81"* -0.72"* -0.76*" 0.23 0.20 0.2; GDPPC74 -0.13 -0.09 -0.11 0.26 0.23 0.23 GOV1 -1.15 -1.08 -1.19 1.01 0.89 0.89 SEC -0.01 -0.03" -0.02" 0.0 1 0.01 0.01 CIVIL -0.13 -0.08 -0.10 0.24 0.21 0.21 OPEN -0.26 0.02 0.02 1.36 0.01 0.01 LATAM -1.20™ -0.69 -0.64 0.53 0.48 0.4S SAFRICA -1.51*** -1.23" -1.14" 0,55 0.49 0.50 TRANS -0.84 -0.08 -0.23 1.69 1.49 1.50

F1X50-LONG50 -2.20*" 0.50 FIX50-HARD50 -1.95"* 0.44 LONG50-HARD50 -1.60*** 0.56 HARD50 -0.19 -1.14" -1.09" 0.56 0.54 0.54 Observations 74 74 74 R2 0.456 0.587 0.595 Note: ***, **, and * represent 99, 95, and 90 percent significance, respectively. Standard errors are in italics.

In sum, bearing in mind all the provisos mentioned at the beginning of this section, we could conclude that the evidence presented here provides only partial support to the hypothesis that hard pegs behave differently from conventional pegs. On the one hand, they appear to deliver better results on inflation. On the other hand, they do not eliminate the inflation-growth trade-off usually involved in the choice of exchange rate regimes. More precisely, they do not improve significantly on the growth performance of conventional fixed arrangements, particularly when compared with countries that displayed stable fixed regimes for a long period of time.

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

VIII. Conclusions

This paper explored the implications for macroeconomic variables of choosing a particular exchange rate arrangement by assessing the impact of exchange rate regimes on inflation, money growth, real interest rates, and real output growth. Surprisingly, there are relatively few references on these issues, possibly because of the lack of an appropriate exchange rate regime classification. Indeed, the paper illustrates how the use of a de facto classification that relies solely on actual behavior delivers new results. Even at this exploratory level, we believe there is substantial evidence that regimes indeed matter in terms of real economic performance. On inflation, the data seem to suggest a negative correlation between fixed exchange rate regimes and inflation. However, a more careful examination revealed that this link, far from being a general finding, is mainly attributable to long pegs in low- to moderate-inflation developing countries. This distinction, combined with the results in Levy-Yeyati and Sturzenegger (2000b) showing that nonindustrial fixes grow more slowly than their more flexible counterparts, indicates that the regime choice involves an inflation-growth trade-off only for the case of long pegs. Short- lived pegs, in contrast, appear to be clearly inferior to floats, exhibiting a poorer growth performance with no substantive inflation gain. The combined use of the de jure and the de facto classification made in this paper allowed us to test the relative value of announcements (words) as opposed to actual behavior (deeds). In this regard, while we find deeds to be the relevant dimension for inflation, words seem to be important for reducing inflation expectations and real interest rates. In contrast, among short de facto pegs, those that openly announce a fix are shown to grow more slowly than those that do not. Thus, words matter in terms of growth, albeit in a "negative" way, somehow rational- izing the concept of "fear of pegging." Two additional distinctions introduced in this paper merit some attention: high and low volatility countries (first- and second-round observations according to LYS) and industrial and nonindustrial economies, both of which play an important role in our tests. In particular, the finding that exchange rate regimes have virtually no impact on the performance of nonindustrial economies deserves a more careful look. At any rate, these distinctions should inform future research on the topic. Recently, Fischer (2001) has suggested that the relevant grouping of exchange rate regimes should involve hard pegs, intermediate regimes (including conventional pegs) and floating regimes (including dirty floats). We find only partial support for this bipolar view. While hard pegs are indeed associated with lower inflation rates than their more conventional counterparts, they are far from eliminating the inflation-growth trade-off mentioned above. Moreover, long pegs appear to be closer to hard pegs than to short pegs. Thus, in the end, the distinctive line seems to hinge not on the legal definition of what constitutes a hard peg but rather on whether the peg, conventional or not, attains some degree of perdurability.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Appendix I. Description of the Data

Table Al. Definitions and Sources Variable Definitions and sources AGDP Rate of growth of real GDP (Source: World Economic Outlook [WHO]) AGDWC Rate of growth of real per capita GDP (Source: WHO) MNTRATE Change in the interest rate (Source: IMF's International Financial Statistics [IMF]) AM2 Rate of growth of M2 (Source: IMF) A 77 Change in terms of trade—exports as a capacity to import (constant Local Currency Units) (Source: World Development Indicators [WDI]; variable NY.EXP.CAPM.KN) CIVIL Index of civil liberties (measured on a 1 to 7 scale, with 1 corresponding to highest degree of freedom) (Source: Freedom in the World—Annual survey of freedom country ratings) DCREDIT Net domestic credit (current LCU) (Source: WDI, variable FM.AST.DOMS.CN). GDPPC74 Initial per capita GDP (average over 1970-73) (Source: WEO) GOV1 Growth of government consumption (lagged one period) (Source: IMF) INETGDP Ratio of net interest payments over GDP (Source: IMF) INF Annual percentage change in the Consumer Price Index (Source: IMF) INVGDP Investment to GDP ratio (Source: IMF) LATAM Dummy variable for Latin American countries OPEN Openness (ratio of [export + import]/2 to GDP) (Source: IMF) POPGR Population growth (annual percent) (Source: WDI; variable SP.POP.GROW) QMM Ratio quasi-money/money (Source: IMF) SAFRICA Dummy variable for sub-Saharan African countries SEC Total gross enrollment ratio for secondary education (Source: Barro, 1991) SIZE GDP in dollars over U.S. GDP (Source: IMF) SUPGDP Ratio of fiscal surplus to GDP (Source: IMF) TRANS Dummy variable for transition economies

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Table A2. List of Countries Industrial Australia Finland Ireland Portugal Austria France Italy Spain Belgium Germany Japan Sweden Canada Greece Netherlands Switzerland Denmark Iceland New Zealand United Kingdom Norway United States

Nonindustrial

Albania Egypt Madagascar Senegal Antigua and Barbuda El Salvador Malawi Seychelles Argentina Equatorial Guinea Malaysia Sierra Leone Armenia Estonia Maldives Singapore Azerbaijan Ethiopia Mali Slovak Republic

Bahamas, The Gabon Mauritania Slovenia Bahrain Gambia, The Mauritius South Africa Bangladesh Georgia Mexico Sri Lanka Barbados Ghana Moldova Sudan Belize Grenada Mongolia Suriname

Benin Guatemala Morocco Swaziland Bhutan Guinea Mozambique Syrian Arab Republic Bolivia Guinea-Bissau Myarunar Tanzania Brazil Guyana Namibia Thailand Bulgaria Haiti Nepal Togo

Burkina Faso Honduras Netherlands Antilles Tonga Burundi Hong Kong SAR Nicaragua Trinidad and Tobago Cambodia India Niger Tunisia Cameroon Indonesia Nigeria Turkey Central African Iran, Islamic Republic of Oman Uganda Republic Ukraine Israel Pakistan Chad Jamaica Papua New Guinea United Arab Emirates Chile Jordan Paraguay Uruguay Colombia Kazakhstan Peru Venezuela, Republica Comoros Kenya Philippines Bolivariana de Congo, Democratic Yemen, Republic of Republic of Korea Poland Zambia Kyrgyz Republic Qatar Zimbabwe Congo, Republic of Lao People's Romania Costa Rica Democratic Republic Russia C6te d'lvoire Latvia Rwanda Croatia Lebanon Cyprus St. Kitts and Nevis Lesotho St. Lucia Czech Republic Libya St. Vincent and the Djibouti Lithuania Grenadines Dominica Luxembourg Sao Tome and Principe Dominican Republic Macedonia, former Saudi Arabia Ecuador Yugoslav Republic of

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

Table A3. Hard Pegs Antigua and Barbuda Estonia Argentina: 1992-99 Gabon* Benin0 Grenada'' Bulgaria: 1998-99 Guinea-Bissau*r 1989-90 and 1997-99 Burkina Paso" Hong Kong Cameroon* Lithuania: 1995-99 Central African Republic* Mali*'' Congo, Republic of* Niger* Cote d'lvoire0 St. Kitts and Nevis d Chad* St. Luciarf Djibouti St. Vincent and the Grenadinesd Dominica"* Senegal0 Equatorial Guinea* Togo" Note: Members of WAEMU and CAEMC are classified as hard pegs except in 1994, when their currency was devalued 100 percent against the French franc. "West African Economic and Monetary Union (WAEMU). ^Central African Economic and Monetary Community (CAEMC). 'Mali became a member of the WAEMU in 1984 and Guinea-Bissau in 1997. ''Eastern Caribbean Currency Area (ECCA).

Appendix II. White's Efficient 2SIV Estimates36 Consider the following structural equation for variable I:

The matrix X includes both endogenous and exogenous variables. In our specification, y, corresponds to the inflation rate and X includes the exogenous regressors in the inflation equation as well as the endogenous regime dummy. Let V = V(e,) denote the (nonspherical) variance covariance matrix (VCV) of the residuals. We can estimate consistently our parameter of interest, 8, by finding the value of 8 that minimizes the quadratic distance from zero of Z'(y—XS)~, that is,

where Z indicates a set of instrumental variables and R corresponds to any symmetric positive definite matrix. R must be chosen appropriately, however, in order to achieve asymptotic efficiency. The estimator corresponding to the minimization problem is (1)

It can be shown that the limiting distribution of 8 is

where (2)

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Proposition 4.45 in White (1984) proves that choosing R = V~{ provides the asymptotically efficient IV estimator, distributed according to (3)

If we use R to obtain the asymptotically efficient estimator, we need an estimator of V. However, because the es are not observable, we need consistent estimators of the errors in order to construct a feasible estimator for the VCV. The procedure is as follows. We first construct a regime index denoting the regimes included in the regression. We then estimate a standard multinomial logit regression of the regime index on all the variables included in the inflation regression, plus the following additional controls: the ratio of domestic credit over GDP (DCREDIT), the ratio of the country's GDP over that of the United States (SIZE), and a measure of financial deepening (the ratio of quasi-money over narrow money, QMM). Frankel and Rose (1996) find that DCREDIT is significantly and positively associated with the collapse of an exchange rate regime. The size variable is potentially related to the regime choice by the usual argument that smaller countries tend to be more open and favor fixed exchange rate regimes. Finally, other authors, notably Kaminsky and Reinhart (1999), have shown that the degree of financial deepening may be associated with the probability of a currency collapse, thus motivating the inclusion of QMM in our model.37 Once we obtain the estimated regime from the multinomial logit, we use them as instruments for the regime dummies in the original specification of the growth regression. This provides consistent estimates of the error terms that allow us to estimate White's efficient covariance matrix.38 From this simple IV regression we obtain a consistent estimate for the e's, which are then used to compute a consistent estimate of V, V, as

which allows for heteroscedasticity. Thus, we can readily implement our estimators as suggested in (1) and compute its VCV matrix as in (3).

REFERENCES Aizenman, Joshua, 1991, Foreign Direct Investment, Productive Capacity and Exchange Rate Regimes, NBER Working Paper No. 3767 (Cambridge, Massachusetts: National Bureau of Economic Research). Anderberg, M.R., 1973, Cluster Analysis for Applications (New York: Academic Press). Barro, Robert J., 1991, "Economic Growth in a Cross Section of Countries,'' Quarterly Journal of Economics, Vol. 106 (May), pp. 407-43. , and Xavier Sala-i-Martin, 1995, Economic Growth (New York: McGraw-Hill). Broda, Christian, 2000, "Terms of Trade and Exchange Rate Regimes in Developing Countries" (unpublished; Cambridge, Massachusetts: Massachusetts Institute of Technology). Calvo, Guillermo A., 1999, "Fixed versus Flexible Exchange Rates: Preliminaries of a Tum- of-Millennium Rematch" (unpublished; College Park, Maryland: University of Maryland).

36This Appendix follows White (1984). 37Other instruments tried in the first stage model (lagged inflation and the ratios to GDP of reserves, government deficit, and M2) yielded identical conclusions but entailed the loss of some observations and were therefore excluded. 38We thank Jerry Hausman for suggesting this procedure to us.

©International Monetary Fund. Not for Redistribution EXCHANGE RATE REGIMES AND ECONOMIC PERFORMANCE

, and Carmen M. Reinhart, 2000, "Fear of Floating," NBER Working Paper No. 7993 (Cambridge, Massachusetts: National Bureau of Economic Research). -, and Carlos Vegh, 1994, "Credibility and the Dynamics of Stabilization Policy: A Basic Framework," in Advances in Econometrics, Sixth World Congress, Vol. II, ed. by Christopher A. Sims (New York: Cambridge University Press). Domac, Ilker, and Maria Soledad Martinez Peria, 2000, "Banking Crises and Exchange Rate Regimes: Is There a Link?" World Bank Policy Research Working Paper, No. 2489 (Washington: World Bank). Eichengreen, Barry J., 1994, International Monetary Arrangements for the 21st Century (Washington: Brookings Institution). , and Ricardo Hausmann, 1999, "Exchange Rates and Financial Fragility," NBER Working Paper No. 7418 (Cambridge, Massachusetts: National Bureau of Economic Research). Fischer, Stanley, 2001, "Exchange Rate Regimes: Is the Bipolar View Correct?" Finance & Development, Vol. 38 (June), pp. 18-21. Frankel, Jeffrey A., 1999, "No Single Currency Regime Is Right for All Countries or at All Times," NBER Working Paper No. 7338 (Cambridge, Massachusetts: National Bureau of Economic Research). , and Andrew K. Rose, 1996, "Currency Crisis in Emerging Markets: Empirical Indicators," NBER Working Paper No. 5437 (Cambridge, Massachusetts: National Bureau of Economic Research). -, 2000, "Estimating the Effect of Currency Unions on Trade and Output," NBER Working Paper No. 7857 (Cambridge, Massachusetts: National Bureau of Economic Research). Frieden, J., P. Ghezzi, and E. Stein, 2001, "The Political Economy of Exchange Rate Policy in Latin America," in The Currency Game, ed. by Jeffrey Frieden and Ernesto Stein (Washington: Inter-American Development Bank). Ghosh, Atish R., and Paolo Pesenti, 1994, "Real Effects on Nominal Exchange Rate Regimes," IGIER Working Paper (Milan, Italy: Bocconi University). Ghosh, Atish R., Anne-Marie Guide, Jonathan D. Ostry, and Holger Wolf, 1997, "Does the Nominal Exchange Rate Regime Matter?" NBER Working Paper No. 5874 (Cambridge. Massachusetts: National Bureau of Economic Research). Ghosh, Atish R., Anne-Marie Guide, and Holger Wolf, 2000, "Currency Boards: More Than a Quick Fix?" Economic Policy, Vol. 31 (October), pp. 270-335. International Monetary Fund, 1999, International Financial Statistics (Washington). Kaminsky, Graciela, and Carmen Reinhart, 1999, "The Twin Crises: The Causes of Banking and Balance of Payments Problems," American Economic Review, Vol. 89 (June), pp. 473-500. Larrafn, Felipe, and Andres Velasco, 1999, Exchange Rate Polic\ for Emerging Markets: One Size Does Not Fit All (Princeton, New Jersey: International Finance Section, Department of Economics, Princeton University). Levine, Ross, and David Renelt, 1992, "Sensitivity Analysis of Cross-Country Growth Regressions," American Economic Review, Vol. 82 (September), pp. 942-63. Levy-Yeyati, Eduardo, and Federico Sturzenegger, 2000a, "Classifying Exchange Rate Regimes: Deeds vs. Words," GIF Working Paper No. 02/2000 (Buenos Aires: Universidad Torcuato Di Telia). Available via the Internet: http://www.utdt.edu/~ely/Regimes_final.pdf. , 2000b, To Float or to Trail: Evidence on the Impact of Exchange Rate Regimes," GIF Working Paper No. 01/2001 (Buenos Aires: Universidad Torcuato Di Telia). Available via the Internet: http://www.utdt.edu/~ely/growth_final.pdf. , forthcoming, Dollarization (Cambridge, Massachusetts: MIT Press).

©International Monetary Fund. Not for Redistribution Eduardo Levy-Yeyati and Federico Sturzenegger

Mundell, Robert A., 1995, "Exchange Rate Systems and Economic Growth," Revista di Politico Economica, Vol. 85 (June), pp. 3-36. Norusis, M., 1993, SPSS for Windows Professional Statistics, Release 6.0 (Englewood Cliffs, New Jersey: Prentice Hall). Obstfeld, Maurice, and Kenneth Rogoff, 1996, Foundations of International Macroeconomics (Cambridge, Massachusetts: MIT Press). Quirk, Peter J., 1994, "Fixed or Floating Exchange Regimes: Does It Matter for Inflation?" IMF Working Paper 94/134 (Washington: International Monetary Fund). Rolnick. Arthur J., and Warren E. Weber, 1997, "Money, Inflation, and Output Under Fiat and Commodity Standards," Quarterly Review, Federal Reserve Bank of Minneapolis, Vol. 22 (Spring), pp. 11-17. Romer, David, 1993, "Openness and Inflation: Theory and Evidence," Quarterly Journal of Economics, Vol. 108 (November), pp. 869-903. Rose, Andrew, 2000, "One Money, One Market? The Effect of Common Currencies on International Trade," Economic Policy, Vol. 30 (April), pp. 9^-5. Summers, Lawrence H., 2000, International Financial Crises: Causes, Prevention and Cures," American Economic Review, Papers and Proceedings, Vol. 90 (May), pp. 1-16. White, Halbert, 1984, Asymptotic Theory for Econometricians (New York: Academic Press).

The Interest Rate-Exchange Rate Nexus in Currency Crises

GABRIELA BASURTO and ATISH GHOSH*

Sharp exchange rate depreciations in the East Asian crisis countries (Indonesia, Korea, and Thailand) raised doubts about the efficacy of increasing interest rates to defend the currency. Using a standard monetary model of exchange rate determination, this paper shows that tighter monetary policy was in fact associated with an appreciation of the exchange rate in these countries and during the Mexican currency crisis. Moreover, there is little evidence of higher real interest rates contributing to a widening of the risk premium. [JEL F31, G15, E40]

ne of the more controversial elements of the stabilization programs in the oEast Asian crisis countries (Indonesia, Korea, and Thailand) was the stance of monetary policy. With the sharp exchange rate depreciations experienced at the onset of the crisis, standard policy prescriptions called for an immediate tightening of monetary policy. But continued depreciation of the exchange rates—well into the stabilization programs—began to raise doubts about the efficacy of raising interest rates to defend the currency.1 Some commentators, indeed, started suggesting that raising

*Gabriela Basurto is Financial Markets Consultant at the Inter-American Development Bank. Atish Ghosh is a Deputy Division Chief in the IMF's Policy Development and Review Department. This paper was prepared for the First Annual Research Conference of the International Monetary Fund, November 9-10, 2000. The authors would like to thank Robert Flood, Philip Lane, Timothy Lane, and participants at the Policy Development and Review Department Seminar Series and at the Research Conference for many useful comments on an earlier version of this paper. 'IMF-supported programs began in August 1997 in Thailand, November 1997 in Indonesia, and December 1997 in Korea, while the most depreciated exchange rates were in January 1998 in Korea and Thailand and in July 1998 in Indonesia. Lane and others (1999) provides a useful summary.

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh interest rates, far from stabilizing the exchange rate, could actually prove counterproductive: further depreciating the exchange rate instead of appreciating it. The mechanism of this "perverse" effect is straightforward. High (presumably, real) interest rates, by causing widespread bankruptcies (or the expectation thereof), result in larger country risk premiums—so much so that the expected return to investors actually declines as interest rates increase, thus prompting even more capital flight and generating greater downward pressure on the exchange rate.2 Establishing whether tighter monetary policy—often taken to mean an increase in nominal interest rates—appreciates or depreciates the currency turns out to be a surprisingly difficult task. Such studies as do exist typically use regressions or vector autoregressions to correlate exchange rate movements to changes in nominal interest rates. This approach, however, runs into two main problems.3 First, the level of nominal interest rate is simply not a good measure of the monetary stance. To give but the starkest example, in January 1998 interest rates in Indonesia reached almost 60 percent per year (far higher than the interest rates witnessed in the other Asian crisis countries) at a time when the money supply was expanding at a monthly rate of 30 percent—scarcely a tight monetary stance. Second, simple time series correlations or vector autoregressions provide very little structure on the model, and their empirical performance in explaining exchange rate movements—even in the absence of a crisis—is, at best, limited. It is difficult to know what to make of a statement such as "higher interest rates are not correlated with exchange rate appreciations during the East Asian crisis" when the model is mute on what is driving the exchange rate. In this paper, we propose an alternative approach to examining whether high real interest rates resulted in exchange rate depreciations. We start from the simple proposition that, as the relative price of two monies, the exchange rate should appreciate in response to a contraction of the domestic money supply. This, together with the empirical observation that in the Asian crisis countries there is a somewhat better correspondence between the exchange rate and the money supply (than between the exchange rate and interest rates), suggests that a standard monetary model may be useful for explaining the bulk of the exchange rate dynamics. This allows us to isolate the risk premium, controlling for changes in monetary policy, and permits a direct test of whether higher real interest rates are associated with a larger risk premium—and thus, ceteris paribus, downward pressure on the exchange rate.

2An important proponent of this school of thought is Joseph Stiglitz; see, for example, Furman and Stiglitz(1998). 3 A third problem is that of policy endogeneity and causality. Interest rates were raised in East Asia precisely because the exchange rate was depreciating. The issue goes beyond finding appropriate instruments for interest rate policy (itself no easy task): In an environment in which policies are being set in anticipation of reactions of the exchange market, and the market-determined exchange rate embodies expectations of future policy, it becomes virtually impossible to disentangle cause from effect. Kraay (1999) reports results using an instrumental variable technique.

100

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

By measuring the monetary stance by the money supply, and by using an explicit model of exchange rate determination, our approach goes at least part of the way in addressing the methodological problems identified above. Of course, even if higher real interest rates are correlated with a larger risk premium, it does not necessarily follow that tightening monetary policy is counterproductive for stabilizing the exchange rate. The magnitude of the effect on the risk premium may be small. And, of course, there may be third factors (such as adverse political news) affecting both the real interest rate and the risk premium on the exchange rate. Nonetheless, if the findings suggest no correlation between real interest rates and the risk premium, then the possibility of the perverse effect (of tight monetary policy causing an exchange rate depreciation) can be ruled out. We apply our methodology to the 1997 currency crises in the three Asian countries and, by way of comparison, to the 1994 Mexican crisis. Our results may be summarized briefly. We find that the pure monetary model does credibly well in explaining much of the observed exchange rate movements (though the stringent cross-equation constraints are rejected). Augmenting this framework to allow for a time-varying risk premium, we find little evidence that high real interest rates are correlated with a larger risk premium in any of the countries except Korea. Once a simple contagion variable is added to the explanatory variables of the risk premium, moreover, the significance of the real interest rate diminishes even in the case of Korea. We conclude that there is little evidence of a "perverse" effect of a monetary tightening on the exchange rate. The remainder of the paper is organized as follows. Section I provides a brief review of the literature and an overview of exchange rate developments during the crisis. Section II lays out the methodology. Section III reports the parameter estimates of the monetary model. Section IV turns to the behavior of the risk premium. Section V concludes.

I. Background Perhaps the most dramatic aspect of the East Asian crisis was the sharp depreciation of exchange rates. Before the crisis, the nominal exchange rates in these countries had, to varying degrees, been de facto pegged against the U.S. dollar. In July 1997, the Thai baht depreciated by 18 percent, eventually going from baht 25 to baht 54 per U.S. dollar (at its most depreciated rate, in January 1998). The initial (sharp) depreciations in Indonesia and Korea were somewhat smaller, around 12 percent (in August 1997 and November 1997, respectively), though the maximum depreciations—from, 2,400 rupiah to 15,000 rupiah per U.S. dollar; and 850 won to 1,700 won per U.S. dollar (January 1998)—were, if anything, even more spectacular.4 Confronted by sharply depreciating exchange rates, monetary policy had to tread warily between two objectives. Under the assumption that tighter monetary

4Exchange rates in Indonesia and Korea were not. in fact, pegged, and the exchange rate had already started depreciating, so the onset of the crisis in each country is not precisely defined. Below, we use August 1997 and November 1997, respectively, as the start dates of the "floating period" in Indonesia and Korea.

101

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh policy would stabilize the exchange rate, there was an obvious need to limit the currency depreciation, not least because of the large foreign currency debt exposures of the banking and corporate sectors (particularly in Thailand and Indonesia). Against this was the danger of an excessive contraction that could severely weaken economic activity. In the event, this dilemma resulted in stop-go policies, with significant declines in money growth rates occurring only in early 1998 in Korea and Thailand. In Indonesia, a deepening banking sector crisis necessitated massive liquidity injections, and the money supply grew rapidly until mid-1998. The continued exchange rate depreciations despite (generally) rising interest rates began to raise doubts about the efficacy of raising interest rates to defend the currency. On the other hand, at least to date, direct evidence that higher interest rates—brought about by raising risk premiums—resulted in further depreciation of the exchange rate (whether in East Asia or elsewhere) has been scant. Furman and Stiglitz (1998) identify a set of 13 episodes, in nine emerging markets, of "temporarily high" interest rates. Using a simple regression analysis, they find that both the magnitude and the duration of such interest rate hikes are associated with exchange rate depreciation. Though they caution that this evidence is not definitive and that its interpretation is fraught with difficulties concerning endogeneity, they conclude that it at least questions the usefulness of raising interest rates to defend the exchange rate. Conversely, Goldfajn and Baig (1999), using daily data to analyze the relationship between nominal interest rates and nominal exchange rates during the Asian crisis, find no evidence that higher interest rates resulted in weaker exchange rates—if anything, they find support for the "orthodox" relationship. Finally, Kraay (1999) uses a large panel data set to examine whether higher interest rates helped stave off speculative attacks. Importantly, he instruments for the policy endogeneity of interest rates, though he notes the difficulties in finding adequate instruments. He finds very little association (positive or negative) between raising interest rates and the outcome of the speculative attack.5 Overall, perhaps the most robust finding of these papers is that the interest rate-exchange rate nexus does not lend itself very easily to econometric analysis, particularly in the East Asian crisis context. Figures 1-4 show why. Take the case of Thailand (Figure 1). Until May 1997, interest rates hovered between 8 and 15 percent, while the exchange rate remained virtually constant (the currency was de facto pegged against the U.S. dollar). From May 1997 to September 1997, higher interest rates were generally associated with continual exchange rate depreciation (the "perverse" effect), but from September 1997 to December 1997, interest rates fell and the exchange rate depreciated (the "orthodox" relationship). Interest rates then rose (with continued exchange rate

5Goldfajn and Gupta (1998) take another tack and study the behavior of nominal exchange rates in the aftermath of a speculative attack and, in particular, whether higher interest rates are associated with the reversal of the overshooting of the real exchange rate taking place through a nominal exchange rate appreciation rather than through higher inflation. They find that higher interest rates are indeed associated with the real appreciation taking place through the nominal exchange rate, with the important caveat that this result does not apply in countries that also suffered a banking crisis.

102

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Figure I.Thailand: Interest Rate and Exchange Rate

Figure 2. Indonesia: Interest Rate and Exchange Rate

103

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh

Figure 3. Korea: Interest Rate and Exchange Rate

Figure 4. Mexico: Interest Rate and Exchange Rate

104

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES depreciation) until January 1998, and from January to March 1998, higher interest rates were associated with an exchange rate appreciation (the orthodox effect). Finally, since June 1998, interest rates have fallen steadily—with few ill effects on the exchange rate. It is hard to know what to make of all this, with neither the orthodox school ("high interest rates appreciate the exchange rate") nor the "perverse" school ("high interest rates depreciate the exchange rate") receiving unequivocal support. A quick check of the other countries likewise shows periods during which interest rate and exchange rate movements were positively correlated, but also periods when higher interest rates were associated with exchange rate appreciations. Of course, it is always difficult to know the counterfactual, and many factors other than interest rates—the availability of official external financing, debt deals with creditors, and political uncertainty—must have been impinging on the exchange rate as well. A further difficulty, alluded to above, is that interest rates often reflect risk premiums and expectations of inflation and/or depreciation and, as such, do not provide a very clear indication of the monetary stance of the country.6 Do monetary aggregates tell a better story? Figures 5-8 show corresponding time plots for the exchange rate and broad money supplies in these countries. For Thailand and Indonesia, the orthodox relationship—greater monetary expansion is associated with an exchange rate depreciation—comes through reasonably clearly. For Korea, the time plot is more difficult to interpret: the exchange rate obviously overshot in late 1997 and then appreciated back, but taking the period as a whole, looser monetary policy is associated with an exchange rate depreciation. Finally, for Mexico, there is again a relatively clear positive relationship between the expansion of the money supply and the depreciation of the exchange rate. This (comparatively) stronger relationship between monetary aggregates and the exchange rate suggests an alternative approach to studying whether higher interest rates contributed to an exchange rate weakening via the risk premium, using an explicit monetary model of exchange rate determination.

II. Methodology The basic idea, which follows Ghosh (1992) in a similar context, is to calculate a "benchmark" exchange rate, based on a pure monetary model that abstracts from any (non-constant) risk premium.7 The difference between the actual exchange rate and this benchmark exchange rate therefore captures the risk premium. The risk premium thus identified can then be correlated to explanatory variables, such as those capturing political events, contagion from other countries, and, in

6In fact, finding pure "policy" interest rates in these countries is not easy. In Korea, for instance, the so-called Bank of Korea discount rate barely moved during the crisis, and actually fell from 5.0 percent to 3.0 percent. In Indonesia, the market-determined interest rate rose to 60 percent even as broad money was expanding at a monthly rate of 30 percent, while Bank Indonesia's discount rate remained constant at 20 percent per year. 7The essential econometric methodology was developed by Campbell and Shiller (1987) in a somewhat different context.

105

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh

Figure S.Thailand: Broad Money and Exchange Rate

'Excludes valuation effects of foreign currency deposits.

Figure 6. Indonesia: Broad Money and Exchange Rate

106

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Figure 7. Korea: Broad Money and Exchange Rate

'Excludes valuation effects of foreign currency deposits.

Figure 8. Mexico: Broad Money and Exchange Rate

'Excludes valuation effects of foreign currency deposits.

107

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh particular, the level of real interest rates. If indeed high real interest rates are expected to cause widespread bankruptcies—and through this mechanism to exert downward pressure on the exchange rate—then they should be positively correlated with the risk premium. To fix ideas, consider the case of Thailand (Figure 5). Between January 1997 and January 1998, broad money (excluding valuation effects) expanded by roughly 20 percent. The simplest "monetary" model (abstracting from changes in money demand, the foreign money supply, or expectations) would suggest a roughly commensurate depreciation of the exchange rate—say, 20 percent as well. This, then, would be the "benchmark" exchange rate. Meanwhile, the actual depreciation (change in the log) of the exchange rate was on the order of 75 percent—suggesting an "excess" depreciation of about 55 percent. This excess depreciation may be attributed (defmitionally) to a widening risk premium. Our methodology thus consists of first calculating a benchmark exchange rate (using a somewhat more sophisticated monetary model) and then tiying to correlate the excess depreciation to the rise in real interest rates (as suggested by Furman and Stiglitz). Two points bear noting. First, the risk premium that this methodology identifies probably comes relatively close to the "credit" risk premium emphasized by Furman and Stiglitz. In particular, and as explained below, under the null hypothesis that the model is correct, the risk premium identified here controls for expectations of future monetary growth based on agents' entire information set. For instance, if there is an expectation of looser monetary policy (perhaps because of adverse political developments or the onset of a banking crisis), this is controlled for in identifying the risk premium.8 Second, the underlying framework is in the spirit of the monetary model of exchange rate determination. This model, a workhorse of exchange rate economics, fell into disuse after its relatively poor predictive performance in the 1970s and 1980s (Meese and Rogoff, 1983). In fact, however, the model has generally performed well in times of high nominal volatility (indeed, much of the early work on this model is based on the high inflation experience of the 1920s and 1930s) and, in its modern incarnation, actually performs rather well even for low-inflation, industrialized countries (as documented in Woo, 1985, and Ghosh, 1992, among others). Ultimately, of course, the proof of the model lies in its empirical fit and, as shown in Section III below, for the crisis countries considered here, the simplicity of the model notwithstanding, its fit is remarkably good. However, we are less interested in the fit of the monetary model than in using it as a filter—to take out the influence of the expansion of the money supply on the exchange rate and to see whether, once this has been controlled for, higher real interest rates are correlated with the "excess" depreciation of the currency.9

'However, other shocks, such as negative shocks to money demand, will be included in the risk premium. Thus, the test proposed here is probably conservative in the sense of being more likely to find a "perverse" effect of higher interest rates on the exchange rate. 9In this sense, the test proposed here is similar in spirit to variance bounds tests where the precise model is of less interest than the excess movement of the variable relative to the benchmark bound.

108

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

The model consists of three basic building blocks. Real money demand is assumed to depend positively on income and negatively on the nominal interest rate:10

where rn is the log of money, p, the log of the domestic price index, y, the log of output, and /, the domestic interest rate. Domestic and foreign interest rates are linked by an interest parity condition:

where s is the log of the exchange rate (an increase in s is a depreciation), and TI is the risk premium. Finally, the real exchange rate is given by1 '

Solving forward for the (first difference of the) nominal exchange rate yields'2

(1) where Equation (1) is merely a statement of the monetary model of exchange rate determination. According to equation (1), faster money growth in the home country (relative to the rest of the world) leads to a depreciation of the exchange rate, while faster output growth, achieved by raising money demand, results in an appreciation. (A widening risk premium, of course, depreciates the exchange rate.) Current movements of (any component of) x, affect the exchange rate directly, while expected future movements are discounted at the rate [p/(l + P)}'. Notice that, in this model, a monetary contraction in the home country necessarily appreciates the exchange rate (via the first term in xi) — unless higher real interest rates happen to cause a sufficiently large increase in the (present-value) risk premium, n.

"'Correspondingly, for the foreign country (the United States) in" — p" = ay* - |3/*. "There are some subtle issues concerning the treatment of the real exchange rate. Clearly, the real exchange rate was not constant following the currency crises in these countries, so purchasing power parity (PPP) cannot be imposed. On the other hand, to the extent that the real exchange rate is driven entirely by movements of the nominal exchange rate, the "fundamentals" zVv will spuriously be correlated with the nominal exchange rate movement. In both Mexico and the Asian countries, however, real exchange rate changes were large and persistent, without a return to the pre-crisis level either through nominal appreciation (once the float began) or inflation—suggesting that real factors were also at play. Because our inten- tion here is to create a benchmark model to filter out fundamentals, we include the real exchange rate in x. As a robustness check, we report results instrumenting for Ax with its lagged value. 12Ghosh (1992) works with a lagged adjustment money demand function and shows that the quasi-first difference, s, — A.s,_i, should be stationary (where 0 < A, < 1 is a quadratic root that depends on the money demand parameters). The estimated value of A, is typically very close to unity, however; as a simplifying approximation, therefore, we use the first difference directly.

109

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh

It is useful to define the benchmark exchange rate (excluding the risk premium) by

(2)

Then, conceptually, our test consists of correlating the difference between As- As* to the variable of interest, rt, such as the level of real interest rates (as suggested by Furman and Stiglitz). The actual test is somewhat different and follows Campbell and Shiller (1987), who study such present value relations extensively in a somewhat different context, and Ghosh (1992), who studies the risk premium in a monetary model of the exchange rate. The first step in estimating the pure monetary model is to create the projection of the expected future discounted monetary policy, Ax,. The simplest approach would be to use a univariate autoregression. However, in general, agents have much more information about the evolution of future monetary policy than would be contained in past values of A*,. For instance, agents may be expecting looser monetary policy based on news about political events or adverse developments in the banking sector. In general, it is difficult to identify and capture the additional information that is being used by agents to determine the exchange rate, and the econometrician's information set, 0, will be only a small subset of the agent's information set, Q. As shown by Campbell and Shiller (1987), however, it is possible to include all the relevant information in the econometrician's information set because, under the null, the exchange rate itself embodies this additional information.13 (As discussed by Campbell and Shiller, one implication of this is that As should Granger-cause Ax.) Therefore, rather than use a univariate autoregression in Ax, a vector autoregression (VAR) is estimated in z - {Ax,, As,}: z, = 3>z,-i + £,. This turns out to be particularly convenient for computing the infinite sum on the right-hand side (RHS) of (2) k because Et(zt+k) = *& zt so that the cross-equation constraint (2) becomes

where g - [0 1] and h - [1 0] so that the "benchmark" exchange rate (excluding the effects of higher real interest rates on the risk premium) may be computed as14

(3)

13The issue is important because otherwise expectations of looser monetary policy are shifted to the risk premium term (since it is the residual), and the risk premium would be capturing not only credit risk, but also the risk associated with looser monetary policy. l4 J To see this, note that E, A,r,+J•= [1 OJ'^z,. Therefore, |l/(l+(3)] I E, A*,+,-(P/(l+p)) '= [l/(l+p)|

110

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Writing out equation (3) explicitly yields As* = Y\Ax, + T^Ast. This "benchmark" exchange rate can be compared to the actual exchange rate. Within the framework adopted here, to the extent that the actual depreciation, As, exceeds the benchmark exchange rate, As*, it reflects a widening risk premium. Next, in order to examine whether the risk premium depends on some variable r (such as real interest rates), the VAR is augmented to include it: z = {A*,, As,, Ar(}. Proceeding exactly as above yields an implied exchange rate (including the effects, if any, of higher real interest rates on the risk premium):

(4)

Again, writing out equation (4) explicitly yields Under the null hypothesis that higher real interest rates have no effect on the risk premium, F^ = 0, which can be tested using the appropriate Wald statistic. Under the 15 alternative, that rt is (positively) correlated with a currency depreciation, F^ > O.

III. Empirical Results Applying the methodology outlined above raises some practical issues. Because we use broad money as our monetary aggregate, part of the expansion of the money supply may be endogenous to the exchange rate because of foreign currency deposits. To address this, we remove the valuation effect of the stock of foreign exchange deposits on the money supply (numerically, it turns out to be important only in Indonesia, where foreign exchange deposits are substantial and the exchange rate depreciation was large). A trickier issue concerns the choice of sample period. A natural choice would be the post-float period (i.e., once the fixed exchange rate regimes were abandoned). There are two drawbacks to this choice, however. First, except in the case of Thailand and Mexico, formal pegged exchange rate regimes were not in place. In Indonesia and Korea, for example, there was a lengthy period of successive depreciations before the very sharp depreciation at end-1997 and early 1998. Hence, the post-float period is not always clearly defined. Second, the post-float period may yield to few observations for reliably estimating the VARs. Without the passage of time, there is essentially no way around this problem. We proceed by reporting results both for the period 1990:Q1-2000:Q6 and for the post-float period.16 We begin by making some preliminary parameter estimates for the money demand function. With high rates of monetization and substantial financial innovation

l5The pure monetary model also has implications for the other parameters; to wit. F] = 0, FT = I. "Thailand, July 1997 onward; Indonesia, August 1997 onward. Korea, November 1997 onward. For Mexico there are enough observations to use only the post-float period (December 1994 onward). Monthly data are taken from International Financial Statistics: exchange rate (line af); money plus quasi- money (lines 34+35); consumer price index (line 64); lending interest rate (line 60p); and industrial production (line 66). For Thailand, industrial production was taken from the Bank of Thailand Monthly Bulletin, and for Indonesia, quarterly data from Biropustat Statitistik are interpolated. Data on foreign exchange deposits to adjust the broad money figures are taken from the central bank bulletins or websites.

111

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh in the years preceding the crises, it is often quite difficult to obtain stable parameter estimates for the money demand functions. The estimates given in Table 1, however, are of plausible magnitude and statistically significant of the expected sign (a positive income elasticity and negative interest elasticity of money demand). Moreover, as discussed in the robustness section below, the main findings turn out not to be terribly sensitive to the exact parameter values of the money demand function. For instance, the interest elasticity (usually the most difficult parameter to estimate) only enters the expression for the exchange rate as the discount factor. We therefore proceed on the basis of the estimates given in Table 1 and, in the robustness section, test the sensitivity of our main results to variations in the money demand parameter values. Next, we check the order of integration of s and x. As indicated in the bottom panel of Table 1, the augmented Dickey-Fuller test cannot reject the null hypothesis of a unit root for the levels of s and x, but readily do so for their first differences; therefore it is appropriate to work with Ast and Axt. With these preliminary transformations, we estimate the vector autoregres- 17 sion, zt = <&Zt-[ + £t. Table 2 reports the VAR parameters for a first-order system. As the model would suggest, Ast Granger-causes subsequent movements in Axt in each case except Indonesia, where the ^-statistic on Ast-\ is marginal (about 1.20). Before turning to the formal test of whether higher real interest rates are associated with a widening of the risk premium, we can gauge the usefulness of the monetary model as a "filter" by comparing the benchmark exchange rate (3) to the actual exchange rate. Here, the model performs credibly well, with the correlation coefficient during the float period ranging from 0.67 to 0.97, and the simple time series plots given in Figures 9-12 show that the model correctly captures much of the movement in the exchange rate. These figures are also useful for identifying periods for which the pure monetary model does not work—which, in the framework adopted, means periods during which there were changes in the risk premium. In Indonesia, there seems to be little left to explain. Essentially, the massive liquidity injection in December 1997/January 1998 so swamps any other developments that the pure monetary model can account for nearly all of the exchange rate depreciation. In February 1998, however, the small re-appreciation of the actual exchange rate falls short of what the pure monetary model would predict—suggesting that the risk premium widened. In Thailand, from July 1997 to January 1998, the actual exchange rate depreciated more than the monetary model would predict, suggesting a widening risk premium, with a decrease in the risk premium starting in March 1998. In Korea the story is much the same: a very large increase in the risk premium in December 1997, which starts reversing around April 1998. Finally, in Mexico, the risk premium widened in January 1995 and again significantly in March 1995, before narrowing in May 1995. To summarize, the pure monetary model seems to characterize movements of the exchange rates reasonably well, and it provides a credible framework to control for the direct impact of monetary aggregates on the exchange rate.

17The order of the VARs was chosen using the Schwartz-Bayes criterion.

112

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Table 1, Money Demand Parameter Estimates and Unit Root Tests

Indonesia Korea Thailand Mexico

Parameter estimates'7 a 0.26** 0.32** 1.33** 1.78** (-statistic 2.59 4.87 20.78 8.91 P 0.04** 0.15** 0.13** 0.12** f-statistic 2.96 5.28 2.38 2.13 K2 0.87 0.99 0.82 0.49 Unit root tests*

X -0.66 -0.39 -1.28 —2 29 s -0.98 -1.11 -1.03 -0.52 At -3.63** ^.70** -5.06** -3.92** As -3.38** -4,73** -4.16** -4.17** Md residual -3.54** -3.66** -3.93** -3.50**

"OLS estimates; asterisks denote 10 (*) and 5 (**) percent significance levels, respectively. 'Augmented Dickey-Fuller tests with six lags.

Table 2. VAR Parameters

Indonesia Korea Thailand Mexico Coefficient /-statistic Coefficient /-statistic Coefficient f-statistic Coefficient r-statistic

Dependent variable: Av A.v(-l) -0.287 -0.99 0.088 0.55 -0.204 -2.04** -0.340 -2.31** M-l )« 0.485 1.21 0.300 1.80* 0.515 2.44** 0.929 4.09** AK-1) -0.183 -0.21 -1.074 -1.53 1.999 1.61 0.064 0.04 Constant 0.019 1.71* 0.010 2.63** 0.006 0.74 -0.01 1 -1.13 Dependent variable: As A;t(~l) -0.178 -0.85 0.252 1.80* 0.020 0.43 0.205 2.18** As(-l) 0.460 1.59 0.235 1.51 0.174 1.73* 0.069 0.47 Ar(-l) -0.548 -0.86 -0.418 -0.68 1.103 1.86* 2.323 2.27** Constant 0.010 1.29 0.000 -0.03 0.003 0.72 0.012 1.82* Dependent variable: &r Ax(-l) -0.144 -5.81** -0.067 -3.18** -0.007 -1.08 0.012 1,12 As(-l) 0.108 3.16** 0.049 2.10** -0.012 -0.85 -0.056 -3.23** Ar(-l) -0.132 -1.76* -0.155 -1.67 -0.334 -3.92** -0.177 -1.46 Constant 0.001 1.47 0.001 1.28 0.000 0.25 0.001 0.95

Note: Asterisks denote 10 (*) and 5 (**) percent significance levels, respectively. "Model implies that A* should Granger-cause AJC.

However, it is also clear that the risk premium was not constant. In the next section, therefore, we relax this assumption and, in particular, allow the risk premium to depend upon real interest rates.

IV. Determinants of the Risk Premium Many factors account for the widening risk premiums as the crisis deepened in each country—political uncertainties, contagion effects, corporate bankruptcies, banking system problems, prospects of possible capital controls, and indeed a seemingly never-ending stream of bad news. Most of these factors are difficult to

113

©International Monetary Fund. Not for Redistribution Gabrleld Basurto and Atish Ghosh

Figure 9. Thailand: Benchmark and Actual Exchange Rate (log first difference)

Figure 10. Indonesia: Benchmark and Actual Exchange Rate (log first difference)

114

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Figure 11. Korea: Benchmark and Actual Exchange Rate (log first difference)

Figure 12. Mexico: Benchmark and Actual Exchange Rate (log first difference)

115

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh capture econometrically, but to the extent that rising real interest rates contributed to widespread bankruptcies, part of the widening risk premiums may be correlated to higher real interest rates. As discussed above, conceptually our test simply consists of regressing the difference between the actual and theoretical exchange rates on the real interest rate. Econometrically, however, it is preferable to do the estimation in a single step by augmenting the VAR to include the (change in) the real interest rate and then testing whether FS = 0 (the null), or F3 > 0 (the alternate hypothesis, that higher real interest rates are associated with a widening premium). Table 3 reports the empirical results. First, the top panel gives Wald test statistics on the pure "monetary" component of the model, based on the implied F coefficients from the estimated VAR parameters.18 Although the estimates of FT are significantly different from zero (except for Indonesia), they are also significantly different from unity. Moreover, the FI coefficients are also significantly different from zero.19 Turning to the correlation between real interest rates and the risk premium, panel 2 of Table 3 reports the estimates of FV For Indonesia, there is a positive relation between real interest rates and the risk premium, though the coefficient is not significantly different from zero. For Thailand and Mexico, the coefficients are actually negative (suggesting, especially in the case of Mexico, that the risk premium went down only when investors saw higher real interest rates). Only for Korea do we find a positive and statistically significant relationship (^-statistic: 1.53), essentially because real interest rates started increasing around end-1997, when the risk premium also widened significantly. But of course, this correlation between real interest rates and the risk premium does not prove that tighter monetary policy caused a widening of the risk premium. One possibility is that some other variable affected the risk premium. An obvious candidate is the contagion effect from other crisis countries in the region. To capture this, for each Asian crisis country, we simply use the unweighted average of the contemporaneous exchange rate movements in the other two countries. Once this variable is added to the explanatory variables of the risk premium, the real interest rate loses its significance even in the case of Korea, with the ^-statistic falling to 1.33 (Table 3, panel 3). Beyond their purely statistical significance, the effects are relatively small in economic terms as well. From Table 3, a 1 percentage point increase in real interest rates would be associated with a 0.1 percent depreciation of the currency. At their peak, real lending rates rose by about 10 percentage points in Korea (relative to the pre-crisis levels). Based on these estimates, the rise in real interest rates could account for less than a 1 percent depreciation of the won—a paltry effect relative to the observed depreciation.

18That is, we compute [1 ,then PI is the resulting coefficient on Ax,

P2 is the coefficient on As, and P? is the coefficient on Ar. l9Recall that the model implies PI = 0 and P2 = 1. Standard errors were computed numerically as VQ'LVQ, where VQ is the gradient of P with respect to the VAR parameters, and I, are the White- consistent standard errors.

116

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES

Table 3. Cross-Equation Constraints Indonesia Korea Thailand Mexico Monetary model" Fl 0.91** 0.89** 0.88** 0.87** Standard error 0.02 0.02 0.0 1 0.01

F2 0.03 0.03* 0.05** 0.08** Standard error 0.03 0.02 0.02 0.01 Real interest rate effect on risk premium6 F3 0.01 0.11* -0.18 -0.04 /-statistic 0.23 1.53 -1.66 -3.28 Real interest rate and contagion effect on risk premium' F3 0.01 0.10* -0.18 f-statistic 0.26 1.33 -1.62 F4 -0.16** 0.01 0.02 (-statistic -7.27 1.53 1.48 Correlation (A,v,A5*) Full sample 0.86 0.95 0.50 Float period only 0.96 0.97 0.74 0.67 Note: Asterisks denote coefficients that are different from zero at the 10 (*) and 5 (**) percent significance levels, respectively. "Pure monetary model, null hypothesis Fl =0, F2 = 1. 6If higher real interest rates result in larger risk premium, F3 > 0. clf contagion results in larger risk premium, F4 > 0.

Robustness

To check for robustness regarding our main finding—that higher real interest rates are not particularly associated with a widening risk premium—we consider a number of alternative specifications. As noted above, one issue concerns the choice of sample period. Specifications 1 and 2 of Table 4 repeat the analysis, but restrict the sample to the post-float period (for the Asian countries). This makes little difference to the results. One possibility is that the perverse effect of tighter monetary policy was purely a crisis phenomenon. In specification 3, therefore, we restrict the sample to the one-year period following the collapse of the fixed regime and the adoption of the float; again, the results differ little. Specifications 4-7 vary the money demand parameters to within one standard error of the point estimates given in Table 1 above, while specifications 8 and 9 use an alternative interest rate (usually the deposit or money market rate rather than the lending rate) or an alternative deflator (the WPI instead of the CPI); specification 10 instruments for AJC using its lagged value; specification 11 includes the estimated residuals from the money demand functions in the forcing variable Ax. The estimated T^ coefficient is generally not significantly positive (except for some of the specifications for Korea, which are borderline significant with ^-statistics around 1.40). Finally, beyond the risk premium effect emphasized by Furman and Stiglitz (1998), higher real interest rates could also affect the exchange rate by depressing

117

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh

Table 4, Robustness Checks0

Indonesia Korea Thailand Mexico f 1] Float period, excl. contagion T3 0.03 0.19 -0.54 (-statistic 0.22 0.94 -2.20

[2] Float period, incl. contagion F3 0.03 0.10 -0.52 f-statistic 0.27 0.49 -2.13

[3] Crisis year only F3 -0.01 0.04 -0.24 -0.71 f-statistic -0.08 0.11 -0.88 -28.71

[4] Low a n 0.01 0.09 -0.18 -0.04 f-statistic 0.19 1.26 -1.66 -3.28 [5] High a F3 0.01 0.10* -0.18 -0.07 f-statistic 0.26 1.39 -1.58 -5.63 [61 Low p F3 0.01 0.09* -0.15 -0.03 /-statistic 0.27 1.34 -1.62 -3.14 [7] High (5 F3 0.01 0.10* -0.21 -0.05 f-statistic 0.26 1.33 -1.62 -3.63

[8] Deposit or money market interest rate F3 0.02 0.10* -0.18 -0.01 f-statistic 0.37 1.40 -1.65 -0.92 [9] Real interest rate deflated by wholesale price index F3 0.00 0.06 -0,18 -0.19 f-statistic 0.71 1.12 -2.86 -11.64

[10] Lagged Ax instrument F3 0.01 -0.01 0.03 0.00 f-statistic 0.58 -0.25 0.31 -0.49

[11] Including money demand residuals F3 0.07 0.18 -0.08 -0.14 f-statistic 1.30 1.28 -1.47 -8.45

[12] Not controlling for Ay F3 0.00 0.06 -0.14 -0.14 f-statistic 0.10 0.87 -2.38 -9.20

Memorandum item

Baseline model n 0.01 0.10* -0.18 -0.04 f-statistic 0.26 1.33 -1.62 -3.28 Note: Asterisk denotes significance at the 10 percent level (for a one-sided test for F3 > 0). "Coefficient on real interest rate, F3. Contagion variable included (except in [1]), but not reported.

118

©International Monetary Fund. Not for Redistribution THE INTEREST RATE-EXCHANGE RATE NEXUS IN CURRENCY CRISES output. In the methodology adopted here, this would not be captured because the benchmark exchange rate controls for (actual and expected) changes in output. The issue is addressed readily enough, however, by simply dropping Av from the definition of Ax in equation (4). As Table 4, specification 12, suggests, this makes little difference to the results.

V. Conclusions One of the most controversial elements of the East Asian crisis was the stance of monetary policy. With continued depreciation of the exchange rate, some commentators suggested that higher interest rates, far from defending the currency, were having a perverse effect. In this view, higher interest rates, by creating the expectation of widespread bankruptcies, were widening the risk premium and resulting in downward pressure on the exchange rate. In this paper, we argue that nominal interest rates are not a good gauge of the monetary stance—particularly in a crisis environment, where the nominal interest more likely reflects fears of inflation and of currency depreciation—and propose a method of testing whether higher real interest rates indeed contributed to a weakening of the currency. We use a simple monetary model to filter out the effects of the monetary expansion on the exchange rate and to identify the risk premium. Turning to the determinants of this risk premium, we find little evidence that higher real interest rates contributed to a widening premium and hence, ceteris paribus, to a weakening of the exchange rate. Only for Korea is the coefficient even occa- sionally positive, and even there, it is generally statistically and economically insignificant once contagion effects are accounted for. We conclude that the perverse effect of higher interest rates on the exchange rate remains largely a theoretical curiosus.

REFERENCES

Campbell, John Y., and Robert Shiller, 1987, "Cointegration and Tests of Present Value Models," Journal of Political Economy, Vol. 95 (October), pp. 1062-88. Furman, Jason, and Joseph E. Stiglitz, 1998, "Economic Crises: Evidence and Insights from East Asia," Brookings Papers on Economic Activity: 2, pp. 1-35. Ghosh, Atish R., 1992, "Is It Signalling? Exchange Intervention and the Dollar-Deutschemark Rate," Journal of International Economics, Vol. 32 (May), pp. 201-20. Goldfajn, Ilan, and Taimur Baig, 1999, "Monetary Policy in the Aftermath of Currency Crises: The Case of Asia," IMF Working Paper 98/170 (Washington: International Monetary Fund). Goldfajn, Ilan, and Poonam Gupta, 1998, "Overshootings and Reversals: The Role of Monetary Policy" (unpublished; Washington: International Monetary Fund). Kraay, Aart, 1999, "Do High Interest Rates Defend Currencies During Speculative Attacks'?" Policy Research Working Paper No. 2267 (Washington: World Bank). Lane, Timothy, and others, 1999, IMF-Supported Programs in Indonesia, Korea, and Thailand: A Preliminary Assessment, IMF Occasional Paper No. 178 (Washington: International Monetary Fund).

119

©International Monetary Fund. Not for Redistribution Gabriela Basurto and Atish Ghosh

Meese, Richard A., and Kenneth Rogoff, 1983, "Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?" Journal of International Economics, Vol. 14 (February), pp. 3-24. Woo, Wing T., 1985, "The Monetary Approach to Exchange Rate Determination Under Rational Expectations: The Dollar-Deutschemark Rate," Journal of International Economics, Vol. 18 (February), pp. 1-16.

120

Consumption and Income Inequality During the Transition to a Market Economy: Poland, 1985-1992

MICHAEL KEANE and ESWAR PRASAD*

This paper challenges the conventional wisdom that income and consumption inequality in Poland increased substantially following the economic transition in 1989-90. Using microdata from the 1985-92 Household Budget Surveys, we find that overall income inequality increased in 1989 but subsequently declined to pre- transition levels. The distribution of consumption reveals a similar pattern. Social transfers are shown to have played an important role in mitigating increases in overall income inequality during the transition. However, the relative well-being of different socioeconomic groups was altered and, despite the reasonably good targeting of transfers, there were clear winners and losers in the transition process. [JEL D31, J31, O15]

(oland experienced a sudden economic transformation in late 1989 and early p 1990 that has become known as the "big bang." The noncommunist government that took power in 1989 ended food price controls in August 1989 and ended price controls on most other products in January 1990. This led to substantial inflation and changes in relative prices. Other aspects of the reforms, including reductions in state orders for manufactured goods and restraints on credit for state-owned enterprises,

'Michael Keane is Professor in the Department of Economics at Yale University. Eswar Prasad is Assistant to the Director in the Research Department of the IMF. The authors thank the staff at the Polish Central Statistical Office, especially Wiestaw Lagodzinski and Jan Kordos, for assistance with many inquiries about the Polish Household Budget Survey data used in this study. They also thank Krzystof Przybylowski and Barbara Kaminska for excellent assistance with the translations of the survey instruments from Polish, and Vincent Koen for providing some of the Central Statistical Office (CSO) aggregate data. They are grateful to Mark De Broeck, Vincent Koen, Thomas Krueger, Branko Milanovic, Michal Rutkowski, Adam Szulc, and numerous other colleagues and seminar participants for helpful discussions and comments.

121

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad along with external shocks such as increased import competition and the collapse of the Council for Mutual Economic Assistance (CMEA) trade bloc, also contributed to large declines in real GDP (of 11.4 percent in 1990 and 7.0 percent in 1991 according to IMF estimates).1 The conventional wisdom is that the process of transition to a market economy has been accompanied by great increases in income inequality, both in Poland and in most of the other formerly centrally planned economies of Eastern Europe. For instance, in a cross-country study, Milanovic (1998) reports that, between 1987-88 and 1993-95, the Gini coefficient for household per capita income rose in 17 of 18 Eastern bloc countries. He notes that the average Gini increased from 0.24, a level similar to that in the Scandinavian and Benelux countries, to 0.33, a level similar to that in Canada and the United Kingdom. To put such an increase in historical perspective, it is roughly three times as great as the increase reported for the United States in the 1980s by Atkinson, Rainwater, and Smeeding (1995). For Poland, the Organization for Economic Cooperation and Development (OECD, 1997) reports that the Gini increased from 0.249 in 1989 to 0.290 in 1993, after which it stayed relatively flat through 1996.2 In this paper, we provide new evidence on changes in inequality in Poland during the transition. The main difference between our work and that of previous authors (reviewed in Section I) is that we have obtained for the first time direct access to the detailed microdata of the Polish Household Budget Survey (HBS) conducted by the Polish Central Statistical Office (CSO)3 for the years 1985-92.4 Prior work on inequality in transition economies has been based primarily on aggregate data about income distributions that are published by the statistical bureaus of the various countries. But, as we discuss in Section II, the published aggregate income data for Poland and other transition economies do not correspond to conventional economic measures of household income. However, at least for Poland, meaningful income measures can be constructed using the household level microdata. Using the HBS microdata, we find no evidence that income inequality increased in Poland in the first three years following the big bang. For instance, we find that Gini coefficients actually declined from 1989 to 1992. Interestingly, while our Ginis for 1992 are quite similar to those reported by the CSO and OECD, we obtain much higher Ginis for the pre-1990 period. We conclude that

'There is some controversy over the relative importance of various factors in generating the output decline in Poland. Calvo and Coricelli (1992) and Commander and Coricelli (1992) stress the contraction of credit to state enterprises. Because most of these enterprises are loss making, a contraction in credit would force them to reduce their scale of operation. On the other hand, Berg and Blanchard (1994) argue that an aggregate demand contraction was the more important cause of the output decline. The basis for this claim is the finding that finished goods inventories increased after the big bang. 2Milanovic (1998) reports that the Gini for Poland increased from 0.256 in 1987 to 0.284 in 1993 (first half). This is somewhat smaller than the increase implied by the OECD figures, but nevertheless substantial by historical standards. 3Or, in Polish, the Gtowny Urzad Statystyczny, commonly referred to as GUS. 4At the time we began our study, the Polish CSO had never before released the HBS microdata. A long negotiation process by the first author during 1992-93 led to its release. Subsequently, the microdata for just the first six months of 1993 were released to the World Bank, and these data are used in World Bank (1995) and Milanovic (1998).

122

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992 the published aggregate statistics seriously understate the degree of inequality that existed before the big bang. As a result, most of the post-big bang increase in inequality that is present in the aggregate statistics appears to be spurious. In the HBS microdata we are able to distinguish between pre- and post- transfer income. We find that inequality in pre-transfer income did in fact increase substantially in the transition. Thus, it appears that transfer programs were quite successful in mitigating any increases in inequality. We find that these programs are well targeted in the sense that most transfers go to those at the low end of the income distribution. This is true even though transfer programs in Poland, as in other transition economies, tend to be based on class rather than income. Another important difference between our work and that of previous authors is that we examine consumption inequality as well as income inequality. To the extent that households can smooth consumption over time, consumption inequality is certainly a more interesting measure. It is again our access to the detailed microdata that allows us to examine consumption inequality in a meaningful way. As we discuss in Section II, the aggregate consumption figures that were published by the Polish CSO, as well as by other former communist countries, did not correspond to conventional economic measures of consumption. After constructing reasonable consumption measures from the microdata of the HBS, we again find no evidence of increased inequality during the transition. One reason for the interest in the changes in inequality that may be occurring in transition economies is that, to the extent that inequality has been increasing, it may create social unrest and political pressures that could stall the transition process. Our results suggest that, at least in Poland, such concerns may have been exaggerated. The existing social safety net appears to have done an adequate job of limiting the impact of transition on inequality. Although we find no evidence of increases in overall inequality, our access to the HBS microdata enables us to examine whether certain socioeconomic groups have been relative winners or losers in the transition. We find that the transition did have significant distributional impacts across broadly defined socioeconomic groups. Some groups also experienced large increases in within-group inequality. For instance, among households for which labor income is the primary source of income, income differentials by education level of household head increased rapidly after the big bang. Gorecki (1994) previously noted such a pattern in the aggregate data released by the CSO. Before the transition, the wage structure in Poland was highly compacted, with wages of college-educated white collar workers little different from those of manual workers. Soon after the big bang, those with a college degree became much more concentrated in the upper quantiles of the income distribution, while those with only primary education became much more concentrated in the lower quantiles. Such a widening of across-group income differentials is to an extent desirable, as it implies an enhanced incentive for human capital investment. But it also raises concerns that dissatisfaction and social unrest may be a problem among those groups that have fared poorly. In the next section, we describe the prior research on income inequality in Poland during the transition in more detail. Then, in Section II, we describe the HBS data. As we explain there, the Polish HBS is of higher quality and was

123

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad collected according to a more consistent methodology over the transition period than the microdata for any of the other former communist countries. Thus, while the Polish case is interesting for its own sake, an analysis of the HBS data also provides the best hope for arriving at conclusions about the effects of transition on consumption and income distributions that may be generalizable. In comparing the relative welfare of households with different levels of income or consumption, an important consideration is that an adjustment needs to be made for household size and, more generally, for the demographic composition of households. Most previous studies on inequality in transition economies have used per capita measures or equivalence scales constructed using industrial country data. An additional contribution of this paper is the construction of a full set of equivalence scales for Poland, which differ in some important respects from those based on industrial country data. Section III describes our procedure for constructing equivalence scales. Section IV presents our main empirical results on the evolution of inequality. Section V analyzes income and consumption mobility. Section VI concludes.

I. Review of Prior Research Several other studies have examined income inequality in Poland during the transition. But they report rather contradictory results, even though they all use income data from the HBS. For instance, OECD (1997, Figure 22, p. 86) reports that the Gini based on household per capita income for Poland is 0.25 for 1989, drops to 0.23 in 1990, and then rises substantially to 0.26, 0.27, and 0.29 over the period 1991-93. In contrast, Gorecki (1994) also finds a drop in inequality from 1989 to 1990, but finds no evidence of a subsequent increase in 1991. Similarly, Milanovic (1993) reports Gini values of 0.260, 0.255, and 0.247 for 1989-91. Thus, the OECD figures imply a very large increase in income inequality in 1991, while the Milanovic and Gorecki figures do not show this. The OECD (1997) and Milanovic (1998) figures are consistent, however, in implying that large increases in inequality had occurred by 1993. The prior studies were based on aggregate statistics published by the CSO, with the exception of Milanovic (1998), who had access to the microdata for just the first six months of 1993.5 The Gini values in the studies cited above were thus approximated using aggregate data on the income distribution published by the CSO in the annual publication Budzety Gospodarstw Domowych, which we henceforth refer to as the Surveys.^ The accuracy of these approximations is certainly subject to question.

5A more recent paper by Torrey, Smeeding, and Bailey (1999) uses a sample that constitutes about 45 percent of the full HBS sample and that is available through the Luxembourg Income Survey (LIS), but only for selected years. Using the LIS data, these authors report income Gini coefficients of 0.217 for 1987, 0.248 for 1990, and 0.242 for 1992. 6The Surveys report the number of households in each of several per capita income ranges, along with the average per capita income within each range and the average number of persons per household within each range. The number of income ranges reported differs by year. This difference in reporting may itself account for some change in the Gini over time.

124

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

A more important point is that the aggregate income statistics reported by the CSO, as well as those reported in household budget surveys done in other former communist countries, differ in a number of important ways from measures of income that would be considered economically meaningful in the West. For example, for farmers, income includes gross farm revenues, rather than net revenues. This is an important problem, because approximately one-fourth of Polish households are either farm households or mixed farmer/worker households. In light of this, one must question any results on income inequality based on the aggregate data. Because we have access to the detailed microdata, we are able to make important adjustments to income in order to obtain a meaningful measure (in this example, by calculating net farm income).7 Furthermore, the aggregate consumption figures published by the Polish CSO, as well as by other former communist countries, do not correspond to Western- style measures of consumption. Rather, they correspond to something like total money outflows. For instance, for farm households, consumption includes farm investment and purchases of supplies. An indication of the strange nature of the aggregate consumption data is provided by Milanovic (1998, p. 41), who reports that for 1993 the Gini for consumption is 0.31, which substantially exceeds the Gini of 0.28 that he calculates for income. Also, on page 33 he reports that for 1993 the ratio of consumption to income is 1.30, an unreasonably high figure. It is again our access to the detailed microdata that allows us to examine consumption inequality in a meaningful way. Once we make necessary adjustments to the categories that are included in consumption, we find the more plausible results that consumption Ginis are generally smaller than income Ginis and that the aggregate consumption to income ratio falls in the 0.894 to 0.955 range during 1985-92. Note that previous research on inequality in Poland and other transition economies has relied almost exclusively on Gini coefficients to measure inequality. In this paper, we provide a more detailed characterization of changes in the income and consumption distributions. We examine alternative entropy measures besides the Gini, we examine quantile ratios, and we examine kernel density estimates of the income and consumption distributions. In addition, prior studies have generally used household per capita income rather than accommodating household economies of scale by using equivalence scales. We examine the sensitivity of our results to choice among a number of alternative equivalence scales.

II. The Household Budget Surveys The Polish Central Statistical Office has been collecting detailed microdata on household income and consumption at least since 1978, using fairly sophisticated sampling techniques. In the Polish HBS, the primary sampling unit is the household. A two-stage geographically stratified sampling scheme is used, where the first-stage

7It is possible to make some (but not all) of the necessary adjustments to income using information in the aggregate data on categories of income. Inconsistencies in the set of adjustments actually made may account for some of the discrepancies in Gini values reported in previous studies.

125

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad sampling units are the area survey units and the second-stage units are individual households. Households are surveyed every month for a full quarter in order to monitor their income and spending patterns, and supplementary information is collected from these households once every year. A certain fraction of the households interviewed in a quarter are interviewed in the same quarter of the following year, thereby adding a limited panel aspect to the data. The typical sample size is about 25,000 households per year (6,250 per quarter). The CSO uses the data obtained from these household surveys to create aggregate tabulations that are then presented in its monthly and annual Statistical Bulletins, or Surveys. The HBS contains very detailed information on consumption. We have aggregated across many of the very detailed consumption categories provided in the surveys to classify total household expenditure into these 16 categories: (1) food, (2) alcohol and tobacco, (3) clothing and footwear, (4) house purchases, (5) house construction, (6) household nondurables (including energy), (7) household durables (including furnishings, appliances), (8) rent, (9) health, (10) hygiene, (11) education, (12) "cultural" durables (radio, TV, sporting goods, etc.), (13) recreation and tourism, (14) vehicles, (15) transportation, and (16) other expenditures. In this paper, we use a coarser breakdown in which the nondurable components of categories 4 through 16 are aggregated into two categories: nonfood commodities and services. Information on sources and amounts of income is available for both households and individuals within each household. Total income is broken down into four main categories: (1) labor income (including wages, salaries, and nonwage compensation), (2) pensions, (3) social security and other transfers, and (4) other income. For farm households, farm income and expenditures, as well as consumption of the farm's produce, are also reported. Finally, the HBS also contains information on characteristics of the dwelling, stocks of durables, and demographic characteristics of all household members. Using information obtained from other CSO publications and IMF databases, we have also extracted time series on prices corresponding to each of our 16 expenditure categories, as well as the nonfood commodities and services group- ings mentioned above. Hence, we have been able to construct disaggregated measures of real consumption for each year. To put the quality of the Polish HBS data in context, it is useful to discuss the limitations of the data sources available for other former communist countries. As discussed by Cornelius and Weder (1996), the Family Budget Surveys (FBSs) collected in the Soviet Union suffer from a number of severe problems. First, the data are not a representative sample of the population (because families were selected mainly on the basis of the industrial affiliation of their wage earners). Second, the income data are grouped, so only the fraction of the sample with income in various intervals is known. Thus, the FBSs do not provide true household- or individual-level income data. After the breakup of the Soviet Union, some of the former Soviet states maintained the same primitive data collection methods, while others (including all of the Baltic states) adopted improved sampling methods in which individuals were chosen from the population register, with gender, age, and household size used as

126

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992 stratifying criteria. In either case, looking at changes in distributions over the transition period is problematic—in the former case because the data are poor throughout, in the latter because the improved data from after the breakup are not comparable to the Soviet-era data. Similarly, the Hungarian income data suffered from a substantial change in methodology in the early 1990s. And based on Flanagan (1995), it appears that data collection efforts in the Czech Republic have been sporadic over time. In contrast, the Polish CSO remained well funded throughout the transition period, and collection of the HBS data using a fairly consistent methodology continued throughout the transition and continues today. For this reason, the HBS offers the highest quality and most consistent microdata available for any of the former communist countries. Thus, it provides the best hope for arriving at conclusions about the effects of economic transition on consumption and income distributions that may be generalizable. This is the first study based on micro-level data from the HBS for years both before and after the big bang. Other researchers who previously used the data (such as Gorecki, Milanovic, and Szulc) had to either work with the aggregated information published by the CSO in the Surveys, submit requests for the CSO to calculate certain statistics for them, or work onsite at the CSO. This greatly limited the kind of analysis that was feasible, for obvious reasons.

Basic Statistics We begin by presenting some basic statistics for Poland in the 1985-92 period. Table 1 reports changes in aggregate GDP, imports, exports, and consumption, as taken from the IMF International Financial Statistics, along with average household income and consumption, as taken from the HBS. A striking aspect of the aggregate data is that per capita consumption actually fell more than GDP in 1990 (-23.8 percent vs. -11.4 percent). Thus, there was no aggregate smoothing of the

Table 1. Selected Macroeconomlc Indicators for Poland (Annual percentage changes) 1986 1987 1988 1989 1990 1991 1992

Aggregate Data Real GDP 4.2 2.1 4.0 0.3 -11.4 -7.0 2.6 Real consumption per capita 5.0 1.2 3.8 -0.3 -23.8 4.6 3.0 Import volumes 4.9 4.5 9.4 1.5 -17.9 37.7 13.9 Export volumes 4.9 4.8 9.1 0.2 13.7 -2.4 -2.6 Consumer price index 16.5 26.4 60.2 251.1 585.8 70.3 43.0 Employment (year end) 0.3 0.0 -1.0 -0.8 -6.2 -3.9 -3.1

Household Survey Per capita real income 1.7 -2.4 7.4 7.7 -25.9 3.3 -2.1 Per capita real total consumption 0.2 -0.2 4.6 3.6 -23.8 3.0 0.0 Notes: Aggregate data were obtained from various publications of GUS and the IMF. Aggregate consumption is deflated by the CPI. Net income and total consumption from household surveys are also deflated by the CPI.

127

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad adverse income shock, as is reflected by the large decrease in net imports. But, in 1991, consumption begins to bounce back (+4.6 percent) even as GDP continues to fall (-7.0 percent). This change is reflected in the very large increase in net imports. It is comforting that the HBS data show a similar pattern of consumption in 1990-91. Table 2 reports a list of variables that we use extensively in our analysis, along with their overall means in the HBS. The total number of observations across all eight years from 1985 to 1992 is 203,620. Note that the mean of total real consumption is 149,610, and the mean of total real income is 161,574, where both variables are deflated by the aggregate CPI. The ratio is 0.926 (not shown in table), which seems reasonable. The sample is 50 percent urban, and 57 percent of the household heads are males in the 31-60 age range. Fifty-four percent of the households include a married couple (not shown in table), and the mean household size is 3.22. There are seven education categories reported, and the most common education levels for the household heads are primary school (35 percent), basic vocational training (31 percent), and high school or equivalent vocational training (19 percent). An interesting feature of the HBS data is that they contain information on whether households own each of a list of 21 durable goods at the start of the inter- view period, and whether their house or apartment possesses each of 5 fixtures. In Table 2, we list the percentage of households with each of the 5 fixtures. Overall means for the durable stocks are not very meaningful because many of them change drastically over time.

III. Equivalence Scales As noted above, most of the prior work on income distributions in Poland has simply looked at per capita household income, and has not attempted to account for household economies of scale by employing equivalence scales. The exception is the work by Szulc (1994, 1995), which analyzes poverty rates. He calculates equivalence scales based on estimating a demographically flexible Almost Ideal Demand System (Deaton and Muellbauer, 1980) for four categories of consumption using the microdata from several years of the HBS.8 We were concerned about estimating a complete demand system under conditions when rationing of certain commodities was probably an issue in some years, but where we do not observe the rationing regimes.9 Thus, we choose to adopt the simpler Engel (1895) method, the basic idea of which is to assume that two households with the same food share are equally well off. Thus, implementation requires only the estimation of the food share equation, rather than a complete demand system. We examine food shares out of total nondurable consumption, because in Poland rationing was far more prevalent for durables than for other goods.

8The years are 1980-82, 1984-89, and 1990-92. The categories are (1) food, alcohol, tobacco. (2) clothing, footwear, hygiene, medical services, (3) house expenses and energy, and (4) transportation, education, entertainment, and other. 9Deaton (1981) discusses estimation of demand systems with known rationing regimes.

128

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

If durables are weakly separable from other goods in the utility function, then expenditure on durables only has an income effect on other demands, and this procedure is appropriate (see Pollak, 1971). Given the food share equation estimates, we obtain the equivalence scale as the relative expenditure necessary for a household of any given composition to achieve the same food share as a base household. The almost ideal demand-type food share equation is (1) where w/, is the food share of household h, x/, is nondurable consumption, &/, is the equivalence scale, and the PJ are the prices of food (7=1) and other goods. The other goods categories that we include are alcohol and tobacco, nonfood commodities, and services. Note that the y\j must sum to zero to satisfy zero degree homogeneity in prices and total expenditure. The P is an aggregate price index defined by lnP = oco+£afcln/7/t +(l/2)ZZyi:/ln/7iln/7/. But, as noted by Deaton and Muellbauer (1980), share-weighted aggregate price indices will tend to be highly correlated with P. Thus, we estimate equation (1) by replacing P with the aggregate price index for nondurable commodities (P*) obtained from the Surveys. Imposing zero degree homogeneity and substituting the aggregate price index, we obtain (2)

We then specify -(3 In £/, = Z, §jD/,j, where the D/y are dummy variables indicating whether household h has characteristic j, where j indexes the set of demographic categories listed in Table 2. A base household consisting of a married couple with no children or other adults present, and where both the husband and wife are in the 31-60 age range, forms the omitted category. We estimated equation (2) including quarter dummies. Given the estimated food share equation, we estimate the equivalence scale k/, as (3)

Note that the equivalence scale k/, equals one for the base-type household. A potential problem with estimation of equation (2) is that denominator bias is present if nondurables consumption is measured with error. Thus, we have estimated equation (2) using both OLS and 2SLS. In 2SLS the instruments for logC/, are (1) the set of 18 household demographic dummies, (2) the education-level dummies for the household head, along with age and age squared of the household head, (3) an urban dummy, (4) the 21 durable holding dummies, (5) the five household fixture dummies, and (6) quarter dummies (to capture seasonals in tastes for food consumption). The first-stage regressions were run separately by year, and their R2 values range from 0.64 to 0.84. Table 3 reports OLS and 2SLS estimates of the food share equation. Note that the coefficient on log real nondurable consumption changes from -0.195 to -0.263 when we instrument. In the 2SLS regression, the three relative price terms taken together imply a coefficient of 0.190 on the log price of food. This implies that a

129

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 2. Summary Statistics Mean Standard Deviation Real household income Total 161,574 127,026 Labor income 81,910 82,632 Transfers 39,728 36,906 Farm income 30,724 109,567 Other income 9,212 39,766

Real household consumption Total 149,610 102,273 Durables 19,035 59,106 Nondurables 130,575 69,207 Food 71,369 33,290

Household characteristics Urban 0.50 0.50 Number of persons in household 3.22 1.62

Primary income source of household Workers 0.49 0.50 Farmers 0.11 0.32 Farmers/workers 0.12 0.32 Pensioners, others 0.29 0.45

Household head characteristics Male, 18-30 0.11 0.31 Male, 31-60 0.57 0.49 Male, > 60 0.15 0.35 Female, 18-30 0.01 0.09 Female, 31-60 0.08 0.28 Female, > 60 0.08 0.28

Age 48.35 15.15 College degree 0.06 0.24 Some college 0.00 0.07 High school 0.19 0.39 Some high school 0.02 0.12 Basic vocational training 0.31 0.46 Primary school 0.35 0.48 Primary not completed 0.07 0.25

Demographic characteristics of other members of household Wife, 18-30 0.37 0.48 Wife, 3 1-60 0.09 0.29 Wife, > 60 0.09 0.28

Child, 0-7 0.42 0.76 Child, 8-12 0.30 0.61 Male, 13-17 0.14 0.40 Female, 13-17 0.14 0.39 Male, 18-30 0.12 0.37 Female, 18-30 0.16 0.40 Male, 31-60 0.01 0.11 Female, 31-60 0.22 0.44 Male, > 60 0.05 0.23 Female, > 60 0.12 0.33

130

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 2. (concluded) Mean Standard Deviation Fixtures Running water 0.83 0.37 Water closet 0.72 0.45 Bathroom 0.70 0.46 Gas 0.63 0.48 Central heating 0.56 0.50

Number of observations (households) Total 203,620 1985 21,560 1986 25,475 1987 29,510 1988 29,287 1989 29,366 1990 29,148 1991 28,632 1992 10,642

1 percent increase in the price of food, holding real expenditure (on nondurables) fixed, increases the food share by close to two-tenths of one percentage point. This is quite comparable to other estimates in the consumption literature. For instance, Deaton and Muellbauer (1980) obtained a value of 0.186 for the own food price coefficient using annual British data for 1954-74.10 This agrees with our estimate to the second decimal place. Their estimate of the real nondurable consumption coefficient was -0.160, which is smaller than ours but still in the ballpark. As a sign of the quality of the HBS data, it is again comforting that we obtain estimates that look reasonably similar to ones in the established consumption literature. In Figure 1, we examine how well our food share equation is able to mimic the actual changes in the average food share for Polish households over the 1985-92 period. The performance of the equation is strikingly good. We then break down the equation to examine the food share changes predicted by each of its four components (changes in real expenditure, relative prices, demographics, and seasonals), holding the other components fixed at their respective sample means. The average food share over the whole sample period is 0.58." Now consider the effect of varying only real expenditure, holding other factors fixed. The model predicts an increase in the food share of 11 percentage points, from 0.55 to 0.66, between 1989Q4 and 1990Q1. This is the immediate impact of the drop in real incomes following the big bang.

10Deaton and Muellbauer (1980) examined allocation of expenditure across eight nondurable commodity categories. Thus, like us, they treat durables as weakly separable. "The food share is so high largely because expenditures on housing are very small. During our sample period, the government provided heavily subsidized housing, and housing was rationed. Because there was no properly functioning market for housing (either rental or owner occupied), we cannot impute the true level of housing consumption.

131

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 3. Food Share Equation (Dependent variable: expenditure on food as a ratio to total expenditure on nondurables) OLS 2SLS log csmn. -0.195* (0.001) -0.263* (0.002) log Pz- log PI 0.024* (0.002) 0.029* (0.003) log P3 - log P, 0.005 (0.003) -0.054* (0.004) log P4 - log P, -0.117* (0.001) -0.165* (0.002) urban -0.038* (0.000) -0.033* (0.001) hdmale, 18-30 -0.003* (0.001) -0.003* (0.001) hdmale, >60 0.002* (0.001) -0.010* (0.001) hdfem, 31-60 -0.020* (0.001) -0.024* (0.001) hdfem, 18-30 -0.029* (0.003) -0.027* (0.003) hdfem, >60 -0.041* (0.001) -0.075* (0.002) couple, 18-30 0.045* (0.002) 0.072* (0.002) couple, 31-60 0.060* (0.001) 0.089* (0.001) couple, >60 0.064* (0.001) 0.081* (0.002) child, 0-7 0.021* (0.000) 0.026* (0.000) child, 8-12 0.029* (0.000) 0.036* (0.001) male, 13-17 0.034* (0.001) 0.042* (0.001) fern, 13-17 0.025* (0.001) 0.033* (0.001) male, 18-30 0.036* (0.001) 0.049* (0.001) fern, 18-30 0.033* (0.001) 0.048* (0.001) male, 31-60 0.046* (0.002) 0.056* (0.003) fern, 31-60 0.054* (0.001) 0.076* (0.001) male, >60 0.042* (0.001) 0.054* (0.001) fern, >60 0.053* (0.001) 0.068* (0.001) qrtrdum2 -0.001 (0.001) 0.000 (0.001) qrtrdumS 0.043* (0.001) 0.047' (0.001) qrtrdum4 0.014* (0.001) 0.022* (0.001) constant 2.680* (0.007) 3.538* (0.018) Note: Standard errors are reported in parentheses. An asterisk indicates statistical significance at the 5 percent level.

However, immediately following the big bang and proceeding through 1992, there was a substantial drop in the relative price of food. Figure 2 presents the price indices used in our analysis. Notice that the relative price of food rose substantially during 1989. Thus, holding other factors fixed, our model predicts that changes in relative prices would have sent the food share from 0.55 in 1989Q1 up to 0.70 in 1989Q4, and that it would have then plummeted to 0.62 in 1990Q1 and further to 0.49 in 1992Q4. In fact, by 1992 the relative price effect clearly dominates the real expenditure effect, and the food share is predicted to have dropped into the 50 percent range (as it in fact did). The two other factors in the model are seasonals and demographics. The quarterly dummies are quite significant and generate a predicted seasonal pattern of 0.56, 0.56, 0.61, and 0.58. But changes in household demographics over the sample period had little effect on food shares.

132

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Figure 1. Actual and Predicted Food Shares

103

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Figure 2. Aggregate CPI and Relative Prices, 1985-92

Note: Lower panels show price indexes relative to aggregate CPI.

Table 4 reports, for representative household types, the values of the household equivalence scales we obtain using the Engel method. For comparison, we also report equivalence scales used by the CSO, the OECD scale, and the scale constructed by McClements (1977), which is widely used in the United Kingdom. Note that our equivalence scales imply somewhat greater household economies of scale than do these other scales. We also ran the second stage food share regression separately by year and constructed equivalence scales separately for each year of the sample. We do not report the results here but note that, for each type of household, the values of the scales changed little over time. This suggests that the changes in relative prices over the sample period had little effect on the relative cost of maintaining different types of households.

-134

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 4. Equivalence Scales as a Function of Household Composition Food-Share Equations Household Type OUS OECD McClements OLS IV Single person households 1 HD = Male, 31-60 0.54 0.59 0.55 0.74 0.71 2 HD = Male, 18-30 0,54 0.59 0.55 0.72 0.70 3 HD = Male, >60 0.54 0.59 0.55 0.74 0.68 4 HD = Female, 31-60 0.46 0.59 0.55 0.66 0.65 5 HD = Female, 18-30 0.46 0.59 0.55 0.63 0.64 6 HD = Female, >60 0.46 0.59 0.55 0.60 0.53 Married couples 7 HD = Male, 31-60; Female, 31-60 1.00 1.00 1.00 1.00 1.00 8 HD = Male, 18-30; Female 18-30 1.00 1.00 1.00 0.91 0.92 9 HD = Male, >60; Female >60 1.00 1.00 1.00 1.03 0.92 Married couples with one child HD = Male, 31-60; Female, 31-60 10 Male/Female, < 7 1.23 1.29 1.17 1.12 1.10 11 Male/Female, 8-12 1.32 1.29 1.24 1.16 1.14 12 Male, 13-17 1.46 1.29 1.29 1.19 1.17 13 Female, 13-17 1.41 1.29 1.29 1.14 1.13 Married couples with older dependents HD = Male, 31-60; Female, 31-60 14 Male,>60 1.54 1.41 1.40 1.24 1.23 15 Female, > 60 1.46 1.41 1.40 1.32 1.29 16 Male, >60; Female, >60 2.00 2.00 1.80 1.63 1.59 Notes: HD indicates the head of household.

IV. Inequality

This section contains our main results on the evolution of inequality in Poland over the period 1985-92. It is worth emphasizing that our measures of inequality focus on the cross-sectional distributions of income and consumption and that, unless noted otherwise, our unit of analysis is the individual. After adjusting household income (or consumption) by an equivalence scale, we assign the same level of income (or consumption) to each individual in the household. In Table 5 we report on the behavior of several alternative inequality measures over the 1985-92 period. The top panel reports Gini coefficients for household income based on four alternative equivalence scales. These are the food share-based, OECD, and McClements scales reported in Table 4, along with the simple per capita scale obtained by dividing household income by household size. Note that the three scales that allow for economies of scale all produce very similar Ginis, typically differing only in the third decimal place. The Ginis based on all four scales indicate that inequality grew from 1985 to 1988 and that inequality actually fell from 1989 through 1992. The Gini based on the food share

135

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 5. Poland: Measures of Inequality, 1985-92

1985 1986 1987 1988 1989 1990 1991 1992

Ibtal income Gini Coefficients Food share-based equivalence scale 0.252 0.254 0.246 0.256 0.263 0.250 0.235 0.230

McCiements equivalence scale 0,249 0.253 0.246 0.254 0.261 0.249 0.238 0.234 OECD equivalence scale 0.253 0.257 0.250 0.256 0.264 0.253 0.242 0.238 Per capita 0.270 0.274 0.270 0.272 0.278 0.271 0.266 0.264

Urban 0.201 0.203 0.198 0.202 0.223 0.217 0.213 0.210 Rural 0.317 0.307 0.287 0.302 0.296 0.278 0.249 0.249

Income excluding transfers 0.373 0.375 0.368 0.385 0.384 0.389 0.404 0.416

Nondurables consumption Food share-based equivalence scale 0.196 0.200 0.205 0.211 0.219 0.209 0.208 0.205

McCleraents equivalence scale 0.197 0.202 0.208 0.214 0.220 0.210 0.213 0.212 OECD equivalence scale 0.200 0.207 0.212 0.217 0.224 0.214 0.218 0.217 Per capita 0.222 0.229 0.236 0.239 0.242 0.235 0.245 0.249

Total consumption 0.230 0.234 0.239 0.244 0.258 0.241 0.233 0.227

Half the Square of Coefficient of Variation (variables adjusted by food share based equivalence scales)

Total income 0.085 0.090 0.085 0.091 0.105 0.086 0.079 0.077 Nondurables consumption 0.066 0.068 0.070 0.074 0.081 0.068 0.072 0.068 Income excluding transfers 0.184 0.190 0.186 0.203 0.210 0.207 0.230 0.244

Mean Log Deviation (variables adjusted by food share based equivalence scales)

Total income 0.075 0.079 0.077 0.078 0.087 0.075 0.071 0.069 Nondurables consumption 0.060 0.062 0.064 0.067 0.074 0.062 0.064 0.064 Income excluding transfers 0.224 0.214 0.213 0.221 0.244 0.247 0.268 0.278

Notes: The inequality measures shown here are for the individual distributions of income and consumption. Household income and consumption are adjusted using the food share-based equivalence scale (unless indicated otherwise) and allocated equally to individuals in the household. scale implies a somewhat sharper decline in inequality in 1989-92 (from 0.260 to 0.230) than do the Ginis based on the other three scales. The Ginis based on simple per capita household income are consistently about 0.015 to 0.030 greater than those based on per equivalent income. Nevertheless, they show the same pattern of inequality growing from 1985 to 1988 and declining from 1989 to 1992. We noted earlier that OECD (1997) reports that the Gini based on per capita income grew from 0.25 in 1989 to 0.27 in 1992. In contrast, we obtain a decline from 0.278 to 0.264 when we use the per capita scale. Thus, the choice of equivalence scale is clearly not the cause of this difference in results. What does account for the difference between our Ginis and those reported by the OECD, or, for that matter, by Milanovic (1993, 1998) for this same period?

136

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

One potential source of difference is that prior studies approximated Ginis based on grouped income data. Consider the year 1987. In the Survey for that year, the CSO published data on the number of people in each of eight per capita household income intervals. Based on those data, Milanovic (1998) calculates an approximate Gini of 0.252. Using the same data, we obtain a similar Gini value of 0.248.12 This is comparable to the value of 0.270 that we calculate from the HBS microdata. Thus, prima facie, it appears that use of grouped data does lead to downward bias in the Gini. However, if we take the HBS microdata for 1987, group households into the same eight per capita income intervals, and approximate the Gini based on that information, we obtain a value of 0.265. Hence, it appears that use of grouped data does bias down the Gini—but not by nearly enough to account for the substantially lower 1987 Gini value reported in earlier studies. The same pattern holds for other years. Another potential source of difference between our Gini estimates and those in earlier studies is that prior studies used different definitions of income. As we noted earlier, the CSO includes gross rather than net farm income in its household income measure. If we do the same, then for 1987 we obtain a Gini of about 0.260. Thus, it appears that this difference in income definitions cannot account for much of the difference in Gini values. Consider next the years 1989 and 1990. For those years, OECD (1997, Figure 22, p. 86) reports Gini values based on household per capita income of 0.25 and 0.23. Similarly, Milanovic (1993) reports Gini values of 0.260 and 0.255. All these figures are based on various aggregate income decile data provided by the CSO, and we are not certain of the sources of the (minor) discrepancies. Our Ginis based on per capita household income for those two years are much higher, at 0.278 and 0.271, respectively. If we group our data into deciles and then approximate the Ginis, we get slightly lower Ginis. And if we leave in gross farm income instead of net farm income, we get marginally higher Ginis. Thus, neither grouping nor the difference in income definition accounts for our much higher Gini values in 1989-90. Strangely, in 1991—92 the discrepancies between our results and those in prior studies largely disappear. In those years our Gini values drop substantially, while those reported by the OECD rise substantially, and all the values fall in the 0.26-0.27 range. At this point, we have been unable to determine why we obtain higher Gini values for years before 1991 than do prior studies based on aggregate income data from the CSO. But, at least mechanically, this difference explains why we find that inequality fell after the big bang while prior studies found that it increased: Essentially, our calculations suggest that income inequality was far higher before the big bang than the aggregate statistics from the CSO would indicate.

l2We are unsure of the reason for the slight difference between our calculation and that of Milanovic (we tried to replicate his approximation method). The grouped data provided in the Surveys contain the number of households within certain ranges of household per capita income, the average of per capita income among households in each interval, and the average size of households within each interval. We approximate Ginis based on these data by assuming that all households in an interval are at the mean of per capita income for that interval.

107

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

We would argue that the inequality measures that we have calculated directly from the HBS microdata are more reliable than those calculated from the aggregate CSO statistics. Hence, we now leave off the comparison of our statistics with those from previous studies and go on to analyze our statistics in more detail. Because the choice of equivalence scale appears to make little difference to our results, we will henceforth report results using the scale based on food shares unless otherwise noted. Consider now the rows of Table 5 that report separate Gini coefficients for the urban and rural populations. The Ginis for the rural population are consistently much greater than those for the urban population. Neither group shows any clear pattern of change in inequality during 1985-88. During 1989-92, there is a decline in inequality for both groups, but it is far greater among the rural population. We next examine the role of transfer payments in reducing inequality. Strikingly, the Gini based on income excluding transfers increased from 0.384 to 0.416 during 1989-92. Thus, we find that actual income did grow more unequal after the big bang.13 Yet, the transfer system more than compensated for this, as the decline in the Gini for total income from 0.263 to 0.230 during 1989-92 indicates. Nevertheless, it is possible that the growth in inequality of earned income has led at least in part to the general perception that inequality has risen. These results contradict the received wisdom that transfers in Poland have been regressive and have thus contributed to the increase in inequality (see, e.g., Milanovic, 1998, p. 49). We will explore this in more detail below. Now we turn to examination of changes in consumption inequality. Again, the Ginis based on the three equivalence scales that allow for household economies of scale all show a similar pattern. Inequality grows from 1985 to 1989 and then declines from 1989 to 1992. The decline from 0.219 to 0.205 indicated by the food share based scale is again sharper than for the other scales. Similarly, the Gini for nondurable consumption declines from 0.258 in 1989 to 0.227 in 1992. It is also worth noting that, as expected, Ginis for nondurable consumption are well below those for income. The Gini coefficient is sensitive to changes in a distribution near the median (see Atkinson, 1970). The coefficient of variation is more sensitive to changes at the high end of a distribution, while the mean logarithmic deviation is more sensitive to changes near the low end. We report these other inequality measures in the bottom panels of Table 5, in order to determine if they tell a consistent story. These measures echo the results based on the Gini coefficients. Both measures also reveal a sharp increase in inequality based on income net of transfers.

Kernel Density Estimates for Income and Consumption To obtain a visual representation of changes in the shape and features of the entire distribution, we now examine kernel density estimates of the income and consumption distributions. Figure 3 (top panel) contains kernel density estimates

13One cannot conclude from this that earnings potential grew more unequal. For instance, the labor earnings we observe are accepted rather than offered earnings. The accepted earnings distribution can grow more unequal even if the offered earnings distribution does not, simply because the nature of the selection into the pool of those who accept wage offers can change over time.

138

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992 for real household income for the years 1988, 1989, 1990, and 1992.14 An Epanechnikov kernel with a bandwidth of 4,000 is used. The density is calculated at the same 200 points for all four years, and the first 125 are plotted in the figure. This covers at least 96 percent of the households in all four years. Figure 3 (lower panel) contains kernel density estimates for real household nondurable consumption for the same four years. Reflecting the more compact distribution of consumption, the first 75 points cover more than 99 percent of the households. The change in the shape of the densities between the years 1988-89 and the years 1990-92 is striking. Much of the change simply reflects the decline in mean income and consumption following the big bang. However, the change in shape observed in Figure 3 is not due simply to a contraction of the mean. To see this, consider taking the distribution for 1992 and multiplying all the income figures by the ratio of mean income in 1988 to that in 1992. Such a transformation will preserve relative inequality measures, while equating mean income in 1992 with that in 1988. The 1988 income density and the transformed density for 1992 are plotted together in Figure 4. The most prominent features of Figure 4 are that, in moving from 1988 to 1992, the mass in the left tail is reduced, and the distribution becomes more peaked around the mode. This accounts for the declines in the various inequality measures noted above. A key aspect of what happened becomes apparent if one compares Figure 3 (top panel) with Figure 4. As the overall income distribution shifted left, there was a support area at about 34,000 to 58,000 zlotys (in 1992 fourth-quarter zlotys) below which household income tended not to fall. Because of the drop in mean real income from 1988 to 1992, the ratio of this support level to mean income increased. In Figure 4, this has the effect of shifting to the right the fat part of the left tail of the scale-adjusted income distribution. We investigated the income sources of households with real income in the 34,000 to 58,000 zloty range, and found that these households receive more than 80 percent of their income from pensions (80.5 percent in 1988, 82.2 percent in 1992). The percentages drop off quickly as household income rises above the 58,000 zloty level. The proportion of total household income for all households coming from pensions was 16.8 percent in 1988 and 26.8 percent in 1992. Thus, the households with income in the support area of about 34,000 to 58,000 zlotys received a far higher share of income from pensions than the typical household. Furthermore, it is important to note that, while mean real household income fell from 178,969 zlotys in 1988 to 131,563 zlotys in 1992, the mean real pension actually rose from 29,811 to 35,258 zlotys. This resulted from legislation that took effect in 1991 that made pensions substantially more generous. Hence, our results suggest that the new pension law helped shift the fat part of the left tail of the income distribution to the right, and that this contributed significantly to the reductions in inequality measures noted above.15

14No adjustment is made for household size. 15It is also worth noting that the fraction of households headed by pensioners increased from approximately 15 percent in the 1985-90 period to roughly 25 percent in 1992. Opting for the more generous pensions was apparently an attractive option for workers who did not fare well in the transition.

139

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Figure 3. Kernel Density Estimates

Quantile Ratios and Shares Another common way to summarize changes in inequality is to examine quantile ratios. Unlike the scalar inequality measures considered above, examination of a

140

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Figure 4. Kernel Density Estimates

set of quantile ratios allows one to consider changes in inequality at various different points in the distribution. Figure 5 reports values of the 0.10, 0.25, 0.50, 0.75, and 0.90 income and consumption quantiles for each quarter over the sample period, as well as the 90/10 and 75/25 quantile ratios. The values are for real household income and nondurable consumption, adjusted using the food share-based equivalence scale. There are upward blips in both quantile ratios in late 1988 and early 1989, but there is little evidence of any trends over the sample period as a whole. If anything, the 90/10 ratio for income appears to drift slightly downward after 1989. In Table 6 we report the shares of income and consumption going to each quintile of the respective distributions. Note that the share of total income going to the bottom quintile rose slightly over the 1989-92 period, while the share going to the top quintile declined. But for income net of transfers the pattern is reversed, again indicating that transfers served to reduce inequality after the big bang. For consumption the share of the bottom quintile also rose over 1989-92, while that of the top quintile fell.

Income and Consumption Patterns Categorized by Source of Income, Education, and Age We have found no evidence of an increase in inequality in Poland in the first three years following the big bang, regardless of which of several inequality measures we consider. However, this does not mean that there were not winners and losers

141

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Figure 5. Real Income and Consumption, 1985-92 (millions ofzloty, 1992 Q4 prices)

Note: Household income and consumption are adjusted by equivalence scales. in the transition. In this section, we turn to an examination of how different groups fared in terms of income and consumption. In Figure 6 we report how median income and consumption moved for four types of households differentiated by main income source of the household head: workers, farmers, mixed farmers/workers, and pensioners. A notable feature of the results is that the use of equivalence scales is important. The per capita household income and consumption plots in the top panel suggest that pensioner-headed households moved from a middle position to being clearly better off than other households after the big bang. According to Milanovic (1998, p. 49), who looks at

142

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 6. Quintile Shares of Income and Consumption

1985 1986 1987 1988 1989 1990 1991 1992

Total income Quantile range

<20 9.1 8.4 8.5 8.6 8.8 9.2 9.9 10.3 21-40 15.1 14.9 15.0 14.7 14.3 14.8 15.0 15.1 41-60 18.5 18.3 18.4 18.0 17.8 18.2 18.3 18.3 61-80 22.6 22.6 22.5 22.1 22.2 22.5 22.5 22.6 >80 34.7 35.8 35.7 36.6 37.0 35.4 34.3 33.7

Income net of transfers

<20 2.5 2.0 2.0 2.0 2.5 1.9 1.6 1.2 21-40 13.3 13.2 13.3 12.7 12.4 12.5 11.8 10.7 41-60 18.9 18.7 18.7 18.1 17.9 18.4 18.3 18.2 61-80 24.7 24.5 24.5 24.1 24.2 24.9 25.4 26.0 >80 40.6 41.6 41.5 43.2 43.1 42.4 42.9 43.9

Total consumption

S20 11.0 10.8 10.6 10.4 9.9 10.6 10.8 10.8 21-40 14.8 14.6 14.4 14.4 14.0 14.5 14.7 14.8 41-60 18.0 17.9 17,7 17.7 17.5 17.8 18.0 18.1 61-80 22.0 22.1 21.9 22.1 22.1 22.1 22.2 22.3 >80 34.2 34.6 35.4 35.5 36.4 35.1 34.4 34.0

Nondurables consumption

<20 11.9 11.7 11.5 11.2 10.9 11.4 11.5 11.4 21-40 15.6 15.5 15.4 15.2 15.1 15.4 15.4 15.5 41-60 18.7 18.7 18.6 18.5 18.5 18.5 18.5 18.6 61-80 22.4 22.4 22.4 22.5 22.7 22.5 22.4 22.5 >80 31.5 31.7 32.1 32.6 32.8 32.2 32.2 32.0

Note: Each column indicates the share of aggregate income or consumption accounted for by persons in different quin- tiJe ranges for that variable. per capita income, "pensions thus contributed strongly to increase inequality." But the per equivalent unit results in the middle panel tell a very different story.16 They indicate that pensioner-headed households had very low income and consumption relative to other groups during the 1985-89 period, and that their relative position improved dramatically after the big bang so as to bring their income and consumption up to almost the same level as the next lowest group (farmers). As a result, we find that pensions contributed importantly to a reduction in inequality.17 The main impetus behind the improved relative position of pensioners was a substantial increase in pension levels that took place in 1991.

l6The reason for the difference in the scales is that the mean numbers of persons in worker, farmer, farmer/worker, and pensioners households are 3.59, 3.64, 4.55, and 1.88, respectively, while the mean numbers of equivalent units are 1.69, 1.77, 2.08, and 1.19, respectively. 17Milanovic (1998, p. 54) concludes that transfers in Poland were regressive overall and contributed to increased inequality. This also contradicts our findings above about the impact of transfers on the Gini coefficient.

143

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Figure 5. Real Income and Consumption, 1985-92 (millions ofzloty, 1992 Q4 prices)

142

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 7 reports the fractions of households that fall in each quantile of the income distribution, conditional on education or age of the household head. For example, in 1989, 45.8 percent of households in which the head had a college degree were in the top quantile. By 1992 the fraction rose to 58 percent. In contrast, in 1989, among households in which the head had only a primary school education, 14.9 percent were in the top quantile, but by 1992 the number had fallen to 9.5 percent. Another striking feature is the improvement of conditions for the old, which resulted from more generous pensions. Among households in which the head was older than 60, 39.2 percent were in the bottom quantile in 1989, but the number dropped to 24.3 percent by 1992. In contrast, the probabilities that a household with a young (18-30) or middle aged (31-60) head would fall in the bottom quantile of the income distribution grew over the same period.

Quantile Regressions We now examine how changes in the overall well-being of households were influenced by the education level of the household head, using quantile regression techniques.18 This enables us to characterize in a parsimonious way the changes in the entire conditional distribution of income, as opposed to looking only at changes in the conditional mean. We ran quantile regressions of log real quarterly household per equivalent income on demographic characteristics of household heads. These characteristics were dummies for the six education categories (with primary school being the omitted category), labor market experience (i.e., age minus years of education minus 6), experience squared, location (urban/rural), and gender. Table 8 reports conditional quantiles of log real household income based on the education level of the household head. Note that log income for all education groups drops substantially at all quantile points from 1989 to 1990. The drops tend to be larger at the higher quantile points (e.g., at the 0.90 quantile they range from 30 to 33 percent, while at the 0.10 quantile they range from 25 to 28 percent). Thus, we see a decline in inequality within education groups as measured by quantile ratios from 1989 to 1990. The interesting thing the table reveals is that for the vocational, primary, and some primary groups, income rises slightly in 1991 but then declines (below the 1990 levels) in 1992, with the drops much more pronounced at the higher quantiles. In contrast, for households headed by college graduates, the 0.10, 0.25, and 0.50 quantiles recover a bit, while the 0.75 and 0.90 quantiles hold steady. Hence, in 1991 and 1992, we see a further drop in inequality within each education group and an improvement in the relative position of the households headed by a college graduate. High school graduates hold steady at all quantiles in 1991-92, except that the 0.90 quantile falls. Thus, the only group of households that experiences a slight recovery in earnings from 1990 to 1992 is the group with a college-educated head.

l8For an analysis of changes in the labor income of individual workers during the Polish transition, see Keane and Prasad (2001).

145

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 7. Fractions of Various Groups in Different Income Quantile Ranges (Based on education and age of household head)

1985 1986 1987 1988 1989 1990 1991 1992

Quantile range College degree or some college <20 3,4 3.3 3.4 3.5 3.9 3.9 2.6 2.2 21-40 10.4 8.4 7.5 9.2 9.0 7.5 5.4 6.0 41-60 16.1 16.1 16.0 17.8 14.8 14.4 11.2 11.5 61-80 26.7 27.7 27.3 26.9 26.5 25.5 21.0 22.3 >80 43.4 44.5 45.8 42.5 45.8 48.8 59.7 58.0 Fraction of annual sample 7.8 7.0 6.3 6.3 6.0 6.2 6.8 8.4

High school S20 9.7 9.9 9.2 10.6 11.0 11.0 9.5 9.9 21-40 17.8 16.7 15.4 16.8 16.8 16.2 12.9 15.1 41-60 22.6 22.2 22.3 21.7 21.4 20.9 20.7 19.2 61-80 25.4 26.9 27.0 26.3 25.2 25.0 25.9 27.6 >80 24,5 24.3 26.0 24.6 25.6 26.9 31.0 28.2 Fraction of annual sample 21.1 18.7 17.7 18.2 17.6 19.0 20.0 22.8

Some high school or vocational training <20 12.9 13.8 14.1 13.0 14.2 17.5 17.5 19.1 21-40 19.8 19.5 20.0 19.6 19.2 19.9 20.8 21.2 41-60 23.8 23.1 23.2 23.1 23.0 21.5 22.2 22.9 61-80 23.5 22.9 22.4 23.5 23.3 21.7 22.6 20.4 >80 20.0 20.8 20.3 20.7 20.4 19.5 16.9 16.3 Fraction of annual sample 29.1 30.3 30.9 31.9 33.0 35.2 34.1 34.0

Primary school

S20 30.4 28.7 28.1 28.6 27.8 27.6 29.2 30.4 21-40 23.0 23.3 23.1 23.0 23.1 23.3 25,0 25.5 41-60 17.7 18.3 18.4 18.2 18.7 19.3 19.5 20.0 61-80 14.5 14.8 15.6 14.7 15.5 16.2 15.5 14.7 >80 14.4 14.9 14.8 15.5 14.9 13.5 10.9 9.5 Fraction of annual sample 35.3 36.5 37.5 36.6 36.8 33.9 33.5 30.3

Within- and Between-Group Decompositions of Inequality

In this subsection, we address the question of the extent to which inequality is within versus between groups, and the extent to which each type of inequality changed over the transition. The single parameter generalized entropy measures of inequality can be additively decomposed into within- and between-group components (see Shorrocks, 1984). This family includes the mean log deviation and half the square of the coefficient of variation, but not the Gini coefficient. Hence, in Table 9, we report decompositions of the former two inequality measures for both income and consumption, grouping households by main income source of the household head. Notice that the vast majority of inequality is within group, rather

146

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 7. (concluded)

1985 1986 1987 1988 1989 1990 1991 1992

Quantile range Less than primary school

<20 47.8 43.6 43.3 45.8 44.4 37.8 38.1 41.4 21-40 23.0 25.0 25.8 24.1 25.1 27.4 28.6 24.4 41-60 11.9 13.8 12.7 13.0 13.3 18.2 17.8 18.3 61-80 9.2 9.0 9.1 9.2 9.1 9.7 9.0 9.9 >80 8.0 8.5 9.2 8.0 8.2 6.9 6.5 6.1 Fraction of annual sample 6.7 7.5 7.6 7.1 6.6 5.7 5.7 4.5

Age 18-30 <20 10. 1 12.6 13.0 11. 1 11.7 15.6 14.1 16.0 2 1^0 17.7 17.3 17.4 17.4 18.0 17.0 17.3 18.9 41-60 22.7 22.8 22.2 22.1 22.2 19.7 21.7 21.7 61-80 25.1 22.6 22.6 24.2 22.9 22.0 23.4 20.4 >80 24.3 24.7 24.9 25.2 25.2 25.7 23.5 23.0 Fraction of annual sample 12.8 13,0 13.0 11.4 10.4 10.9 10.7 10.2

Age 31-60 S20 14.3 14.8 16.1 15.1 14.5 16.8 17.9 19.0 21-40 18.1 17.6 17.9 18.0 17.2 17.4 17.4 18.3 41-60 21,7 21.1 20.7 21.1 21.2 20.1 19.7 19.3 61-80 22.6 23.0 22.4 22.5 23.4 22.3 21.9 21.0 >80 23.3 23.5 23.0 23.3 23.8 23.5 23.0 22.4 Fraction of annual sample 66.5 65.2 65.1 66.3 66,2 65.2 65.3 64.7

Age >60 £20 44.4 40.1 35.8 39.2 39.2 30.8 28.3 24.3 21-40 27.5 28.9 27.9 27.4 28.8 28.4 28.3 24.7 41-60 13.0 15.0 16.6 15.7 15.8 20.0 20.0 21.1 61-80 8.4 9.5 11.4 10.3 9.2 12.9 13.3 17.2 >80 6.6 6.6 8.3 7.4 7.0 8.0 10.1 12.6 Fraction of annual sample 20.7 21.7 21.8 22.3 23.5 23.9 24.0 25.1

Note: Each column indicates the share of aggregate income or consumption accounted for by persons in different quantile ranges for that variable. than between groups, which is not surprising given the coarse nature of the grouping. Both within- and between-group inequality rose in 1989. The total decline in inequality from 1989 to 1992 is attributable in about equal part to declines in both the within- and between-group components. The lower panel of Table 9 shows the evolution of within-group inequality, as measured by the Gini coefficient, for different socioeconomic groups. The interesting finding is that the changes in within-group inequality are very different for the different groups. For instance, for households headed by workers, the Gini coefficient increased slightly after the big bang. In contrast, for households headed by farmers and farmers/workers, the Gini coefficient declined substantially from 1988 to 1992.

147

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 8. Conditional Quantiles of Real Quarterly Household Income (in logs) (Based on educational attainment of head of household)

Quantile = 0.10 Quantile = 0.25 Quantile = 0.50 COL HS VOC PS COL HS VOC PS COL HS VOC PS 85 11.20 11.05 10.96 10.82 11.44 1 1 .28 11.19 11.08 11.66 11.51 11.41 11.33 86 11.23 11.07 10.98 10.85 11.47 11.31 11.20 11.10 11.70 11.54 11.44 11.36 87 11.24 11.07 10.98 10.85 11.47 11.31 11.21 11.09 11.70 11.53 11.42 11.34 88 11.28 11.12 11.06 10.92 11.50 11.34 11.26 11.16 11,73 11.57 11.48 11.40

89 11 .36 11.16 11.06 10.95 11.61 11.40 11.29 11.20 11.84 11.64 11.53 11.46 90 11.09 10.89 10.78 10.68 11 .32 11.12 10.99 10.93 11.57 11.36 11.23 11.17 91 11.21 10.95 10.82 10.72 11 .42 11.20 11.03 10.95 11.68 11.43 11.26 11.17 92 11.13 10.89 10.76 10.66 11..36 11.13 10.98 10.90 11.61 11.37 11.20 11.11

Quantile = 0.75 Quantile = 0.90 COL HS VOC PS COL HS VOC PS

85 11.90 11.74 1 1.65 11.60 12,.13 11.99 11.91 11.86 86 11 .94 11.77 11.69 11.62 12..20 12.03 1 1 .93 11.89 87 11.93 11.76 11.65 11.59 12..19 12.01 11.90 11.85 88 11.99 11.82 11.72 11.66 12..23 12.07 11.97 1 1 .93

89 12.11 11.91 11.78 11.73 12.41 12.18 12.04 12.02 90 11.84 11.61 11.48 11.43 12.08 11.88 11.72 11.69

91 11.92 11.67 11.49 11.42 12. 17 11.90 11.71 11.66 92 11 .84 11.59 11.43 11.35 12.07 11.81 11.65 11.59

Notes: COL represents college degree; HS represents high school degree; VOC represents basic vocational training; PS represents primary school. The regressors in the quantile regressions included an urban dummy and the following variables based on household head attributes: experience, experience squared, a dummy for sex, and six education dummies. To generate the predicted quantiles. all independent variables except for the education dummies were set to their means over the full sample.

V. Income and Consumption Mobility The Polish HBS also contains a limited panel aspect. A certain fraction of the households interviewed in each year are interviewed again the next year. We now exploit this panel aspect of the data to examine whether income and consumption mobility (e.g., the probabilities of moving across quintiles of the distribution) have changed. That is, in the transition to a free market economy, to what extent do household income and consumption become more variable over time? In Table 10 we report quintile transition rates for total household income, adjusted using the food share-based equivalence scales. The transition matrices are presented for five pairs of years (1987-88, 1988-89, 1989-90, 1990-91, and 1991-92). Farmers and farmers/workers are excluded from the analysis because the income figures are reported quarterly and there are strong seasonals in farm income. There is a strong tendency for farm households who are interviewed during the harvest period to have high incomes in both years, while farm households interviewed in other periods tend to have low incomes in both years. As a

148

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 9. Decomposition of Inequality Measures

1985 1986 1987 1988 1989 1990 1991 1992

Half the square of the coefficient of \iiriation ( X 100) Total 8,5 9.0 8.5 9.1 10.5 8.6 7.9 7.7

Between-group 1.4 1.2 0.8 1.2 1.8 0.9 0.6 0.6 Within-group 7.1 7.8 7.7 8.0 8.7 7.8 7.3 7.0

Mean log deviation (X 100)

Total 7.5 7.9 7.7 7.8 8.7 7.5 7.1 6.9

Between-group 0.8 0.6 0.4 0.6 1.0 0.5 0.3 0.3 Within-group 6.7 7.3 7.3 7.2 7.7 7.0 6.8 6.6

Gini coefficients

Workers 0.186 0.192 0.191 0.189 0.208 0.211 0.208 0.211 Farmers 0.475 0.483 0.478 0.496 0.440 0.420 0.366 0.321 Mixed, farmers/workers 0.272 0.279 0.276 0.285 0.271 0.253 0.229 0.220 Pensioners, other 0.211 0.212 0.203 0.205 0.214 0.206 0.210 0.203

Note: Socioecortomic groups are defined on the basis of the household's primary source of income. result, inclusion of farm households will lead to an exaggeration of the degree of persistence in income. The own transition rates for the bottom three quintiles decline noticeably from 1987-88 to 1991-92. For instance, the sample probability of staying in the bottom quintile drops from 0.64 to 0.57, while the probability of moving from the bottom up to the second quintile increases from 0.21 to 0.25. For the fourth and fifth quintiles the transition rates are little changed. Overall, the results seem to show some increase in mobility, but it does not appear to be dramatic. Because Table 10 is rather hard to digest, we have attempted to summarize it using the simple regression in Table 11, in which the elements of the transition matrices are regressed on a set of dummy variables and time effects. The most interesting parameters are the coefficients on the interactions between calendar time and the dummies for whether the matrix element is one of the diagonals. These coefficients for income mobility (left column) show that the diagonals are trending significantly downward. The largest and only significant coefficient for an off-diagonal with time interaction is that for two off the diagonal. This coefficient indicates that the probability of a transition across two quintiles (either up or down) has trended upward. Table 10 also reports transition rates for consumption. These transition matrices appear to be remarkably stable over time. This visual impression is confirmed in Table 11, which also reports regression results for the consumption transition matrix elements. Notice that there are no significant trends in the diagonal elements. Thus, we find that while income mobility has increased during the transition, consumption mobility has not. A striking feature of the results is that the transition matrices for income and consumption look almost identical in 1992. Note, however, that this does not mean

149

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Table 10. Quintlle Transition Rates Income Consumption Quintile group in 1988 I II III IV V I II III IV V I 0.64 0.21 0.08 0.05 0.02 I 0.58 0.25 0.10 0.06 0.02 Quintile II 0.22 0.40 0.21 0.12 0.05 n 0.25 0.33 0.24 0.12 0.06 group in in 0.08 0.24 0.34 0.24 0.10 m 0.10 0.25 0.30 0.23 0.12 1987 IV 0.04 0.10 0.27 0.35 0.24 IV 0.05 0.12 0.25 0.33 0.24 V 0.02 0.05 0.10 0.25 0.59 V 0.02 0.05 0.11 0.26 0.56

Quintile group in 1989

I II HI IV V I II III IV V i 0.61 0.25 0.08 0.03 0.02 I 0.57 0.25 0.11 0.06 0.02 Quintile n 0.24 0.35 0.24 0.12 0.04 II 0.25 0.32 0.24 0.14 0.06 group in m 0.09 0.24 0.31 0.25 0.11 m 0.11 0.24 0.29 0.24 0.12 1988 IV 0.04 0.11 0.25 0.35 0.25 IV 0.05 0.13 0.25 0.32 0.26 V 0.02 0.05 0.10 0.25 0.58 V 0.03 0.06 0.12 0.25 0.54

Quintile group in 1990

I II III IV V I H HI IV V i 0.57 0.27 0.10 0.04 0.03 I 0.55 0.25 0.12 0.06 0.03 Quintile n 0.25 0.34 0.21 0.14 0.06 n 0.25 0.32 0.22 0.14 0.07 group in in 0.11 0.22 0.29 0.27 0.11 m 0.11 0.23 0.30 0.25 0.12 1989 IV 0.04 0.12 0.26 0.31 0.26 IV 0.06 0.14 0.23 0.31 0.26 V 0.03 0.05 0.14 0.24 0.54 V 0.02 0.06 0.14 0.25 0.53

Quintile group in 1991

I II III IV V I II III IV V i 0.58 0.25 0.10 0.06 0.02 i 0.53 0.26 0.13 0.06 0.02 Quintile ii 0.24 0.33 0.23 0.15 0.05 n 0.27 0.30 0.22 0.15 0.06 group in HI 0.11 0.24 0.33 0.21 0.11 HI 0.13 0.25 0.30 0.20 0.12 1990 IV 0.05 0.14 0.21 0.33 0.26 IV 0.05 0.14 0.24 0.31 0.27 V 0.02 0.05 0.13 0.25 0.56 V 0.02 0.06 0.11 0.29 0.53

Quintile group in 1992

I II III IV V I II HI IV V I 0.57 0.25 0.11 0.05 0.02 I 0.57 0.23 0.13 0.06 0.02 Quintile II 0.25 0.34 0.25 0.11 0.05 II 0.25 0.33 0.24 0.13 0.05 group in III 0.11 0.25 0.29 0.25 0.10 III 0.12 0.25 0.27 0.25 0.11 1991 IV 0.05 0.12 0.26 0.33 0.23 IV 0.04 0.14 0.24 0.32 0.25 V 0.02 0.04 0.09 0.25 0.59 V 0.02 0,05 0.13 0.24 0.57

Note: Only households surveyed in consecutive years are used for calculating income and consumption quintiles and quintile transition rates. Households with farm income or income from work on farms as the primary income source were excluded. Household income and nondurables consumption were deflated by the aggregate CPI and adjusted using food share based equivalence scales.

150

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 11. Quintlle Transition Rate Regressions Nondurables Income Consumption trend X dumllSS -0.008* -0.003 (0.003) (0.003) trend X dum2244 -0,008* -0.003 (0.003) (0.003) trend X dum33 -0.009 -0.005 (0.005) (0.004) trend X dumloff 0.002 0.000 (0.002) (0.001) trend X dum2off 0.004* 0.004 (0.002) (0.002) trend x dumSoff 0.001 -0.001 (0.002) (0.002) trend X dum4off 0.001 -0.001 (0.003) (0.002) dumllSS 0.268* 0.257* (0.020) (0.016) dum2244 0.029 0.021 (0.020) (0.016) dumloff -0.101* -0.060* (0.017) (0.014) dum2off -0.241* -0.193* (0.017) (0.014) dumSoff -0.295* -0.248* (0.018) (0.015) dum4off -0.319* -0.282* (0.020) (0.016) constant 0.339* 0.306* (0.016) (0.013) Notes: The dependent variable consists of income (or consumption) quintile transition rates stacked into one vector with 125 observations (25 transition rates based on 5 quintiles, for 5 pairs of years). Let qij denote the transition rate from quintile i in year 1 to quintile/ in year 2. The dummies in the regression are then defined as follows: dumllSS (qll, q55); dum2244 (q22, q44); dum33 (q33); dam/ojf indicates transition rates one off the diagonal off the 5X5 transition matrix; dum2off indicates transition rates two off the diagonal, and so on. These dummies were also interacted with a time trend. The excluded dummy in the regression is dum33. Standard errors are reported in parentheses below the coefficients. An asterisk indicates statistical significance at the 5 percent level.

151

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad that consumption is as variable as income or that consumption closely tracks income. Table 12 reports variances of real household income and consumption over two-year periods, using the subsample of households that were interviewed in consecutive years. Note that the variance of consumption is only half that of income in 1992. Part of that difference is a scale difference, because mean consumption is 84 percent of mean income in that year. But the coefficient of variation for consumption is also less than for income. The explanation for how the quantile transition matrices for income and consumption can look so similar in 1992, despite the fact that consumption is less variable than income in absolute terms, is that the consumption quantiles are more compactly grouped together than are the income quantiles. Finally, notice in Table 12 that the variance of income spikes up substantially in 1988-89 and 1989-90, the two most turbulent periods of the transition. The increase in variance for consumption is far less pronounced, suggesting some ability of households to smooth consumption over this period. We are exploring this issue further in ongoing work.

VI. Conclusions We conclude by comparing our evidence on changes in inequality in Poland with the evidence from previously available aggregate statistics. Table 13 reports the statistics that are germane to this comparison. The first row of the table reports the Gini coefficients we have calculated from the HBS microdata for the 1985-92 period. These Ginis are for the distribution of individuals' incomes, using the per capita income of the household in which they reside. Although we calculated many alternative inequality measures, this is the one most comparable to published aggregate statistics. The second row of the table shows the Ginis calculated by the CSO for the OECD, and published in OECD (1997). Observe that we obtain similar values in 1991 and 1992 (0.267 vs. 0.270). But our calculations imply that a much higher level of inequality was present in Poland in 1989 (before the big bang) than do the CSO-OECD figures (0.278 vs. 0.249). Thus, we conclude that the increase in inequality in Poland for the first three years after the big bang that is implied by the CSO-OECD figures is spurious, resulting from serious understatement of the degree of inequality that existed before 1990. We also obtained a number of other interesting findings. Social transfers appear to have played an important role in preventing increases in inequality during the transition. In fact, inequality in pre-transfer income did increase during the transition, but transfers more than counteracted this. An increase in the generosity of pensions was a particularly important factor in preventing increased inequality. There were important differences across socioeconomic groups in how inequality has changed. In particular, income and consumption inequality grew in households headed by workers, but declined in households headed by farmers and farmer/workers. A key factor in increasing inequality among workers was a substantial rise in education premiums. In fact, we find that only households headed by college graduates experienced a substantial recovery in incomes following the big bang.

152

©International Monetary Fund. Not for Redistribution CONSUMPTION AND INCOME INEQUALITY IN POLAND, 1985-1992

Table 12. Variability of Household Consumption and Income 1987-88 1988-89 1989-90 1990-91 1991-92 Var (d 5.57 6.18 6.67 3.35 2.87 Var (y) 8.30 16.10 22.40 6.32 5.73 Var (c-Vvar (v) 0.67 0.38 0.30 0.53 0.50 Coeff. var (c) 0.270 0.277 0.323 0.255 0.231 Coeff . var (y) 0.278 0.364 0.478 0.288 0.275 Coeff. var (c)l Coeff. var 0') 0.97 0.76 0.68 0.89 0.84

Observations 8.471 8,874 4,775 4,313 4,224 Notes: Variances of real nondurables consumption and real income were computed over two-year periods for each household surveyed in consecutive years. Sample means of these variances (divided by 10E+8) are reported here. Both variables are measured at 1992Q4 prices and were adjusted by equivalence scales. The coefficient of variation of nondurables consumption is computed as the square root of the variance of consumption shown in the table divided by the sample average of the two-year mean of consumption for each household surveyed in consecutive years; likewise for income.

Table 13. Comparisons with Per Capita Ginis Based on CSO Methodology 1985 1986 1987 1988 1989 1990 1991 1992 HBS microdata — full distribution 0.270 0.274 0.270 0.272 0.278 0.271 0.266 0.264

CSO-OECD Ginis ______0.249 0.230 0.260 0.270 Notes: The first row shows per capita Ginis calculated using the HBS microdata. The second row shows Gini coefficients calculated by the CSO for the OECD. Documentation from the CSO suggests that, for 1989-92, Ginis were computed using income decile groups based on per capita income. These Gini coefficients for 1989-92 were obtained from the OECD.

REFERENCES

Atkinson, Anthony B., 1970, "On the Measurement of Inequality," Journal of Economic Theory, Vol. 2, pp. 244-63. , Lee Rainwater, and Timothy M. Smeeding, 1995, "Income Distribution in OECD Countries: Evidence from the Luxembourg Income Study." OECD Social Policy Studies No. 18 (Paris: OECD). Berg, Andrew, and Olivier J. Blanchard, 1994, "Stabilization and Transition: Poland, 1990-1991," in The Transition in Eastern Europe, Vol. 1: Country Studies, ed. by Olivier Blanchard. Kenneth Froot, and Jeffrey Sachs (Chicago, Illinois: University of Chicago Press). Calvo, Guillermo A., and Fabrizio Coricelli, 1992, "Stabilizing a Previously Centrally Planned Economy: Poland 1990," Economic Policy, Vol. 14 (April), pp. 176-226. Commander, Simon, and Fabrizio Coricelli, 1992, "Output Decline in Hungary and Poland in 1990-91: Structural Change and Aggregate Shocks," Policy Research Working Paper No. 1036 (Washington: World Bank)."

153

©International Monetary Fund. Not for Redistribution Michael Keane and Eswar Prasad

Cornelius, Peter K., and Beatrice S. Weder, 1996, "Economic Transformation and Income Distribution: Some Evidence from the Baltic Countries," Staff Papers. International Monetary Fund, Vol. 43 (September), pp. 587-604. Deaton, Angus, 1981, "Theoretical and Empirical Approaches to Consumer Demand under Rationing," in Essays in the Theory and Measurement of Consumer Behavior, ed. by Angus Deaton (Cambridge: Cambridge University Press). , and John Muellbauer, 1980, "An Almost Ideal Demand System," American Economic Review, Vol. 70 (June), pp. 312-26. Engel, E.,1895, "Die Lebenskosten belgischer Arbeiter-Familien fruher und jetzt," Bulletin de I'lnstitut International de Statistique, Vol. 9, pp. 1-129. Flanagan, Robert, 1995, "Wage Structures in the Transition of the Czech Economy," Staff Papers, International Monetary Fund, Vol. 42 (December), pp. 836-54. Gorecki, Brunon, 1994, "Evidence of a New Shape of Income Distribution in Poland," Eastern European Economics, Vol. 32 (May-June), pp. 32-51. Keane, Michael P., and Eswar S. Prasad, 2001, "Changes in the Structure of Earnings During a Period of Rapid Technological and Institutional Change: Evidence from the Polish Transition," IMF Working Paper, forthcoming (Washington: IMF). McClements, Leslie, 1977, "Equivalence Scales for Children," Journal of Public Economics, Vol. 8, pp. 191-210. Milanovic, Branko, 1993, "Social Costs of the Transition to Capitalism: Poland 1990-91," Policy Research Working Paper No. 1165 (Washington: World Bank). , 1998, "Income, Inequality and Poverty During the Transition from Planned to Market Economy," in World Bank Regional and Sectoral Studies (Washington: World Bank). Organization for Economic Cooperation and Development, 1997, OECD Economic Surveys: Poland 1997 (Warsaw, Poland). Pollak, Robert, 1971, "Conditional Demand Functions and the Implications of Separable Utility," Southern Economic Journal, Vol. 37, pp. 423-33. Shorrocks, Anthony F., 1984, "Inequality Decomposition by Population Subgroups," Econometrica, Vol. 52, pp. 1369-85. Szulc, Adam, 1994, "Poverty in Poland During the Transition Period: 1990-1992 Evidence," Statistics of Transition, Vol. 5, pp. 669-82. , 1995, "Measurement of Poverty: Poland in the 1980s," Review of Income and Wealth, Vol. 41 (June), pp. 191-205. Torrey, Barbara B., Timothy M. Smeeding, and Debra Bailey, 1999, "Income Transitions in Central European Households," Economic Development and Cultural Change, Vol. 47 (January), pp. 237-57. World Bank, 1995, Understanding Poverty in Poland (Washington: World Bank).

154

Bail-Ins, Bailouts, and Borrowing Costs

BARRY EICHENGREEN and ASHOKA MODY*

We examine how IMF programs and collective action clauses affect the terms of market access. We find different effects on different types of borrowers. Fund programs appear to create additional commitment for bond borrowers with "intermediate" credit, who consequently are able to borrow at lower cost. Moreover, Fund programs with limited structural conditions have the most favorable effects on investor confidence. Easier restructuring through collective action clauses raises the costs for borrowers with low credit ratings but lowers them for those with high ratings. Thus, for the relatively high quality creditors, own commitment and voluntary restructuring work best. [JEL F21, F33, F34]

variety of ideas for governing the flow of portfolio capital have been put on A the table in the course of discussions of how to strengthen the international financial architecture. The goal in each case is to reduce the frequency and severity of crises and limit investor moral hazard by creating alternatives to large-scale financial rescues, which are increasingly seen as ineffective in catalyzing private capital flows. While the ideas falling under this heading are diverse, one thing they have in common is their uncertain impact on borrowing costs. The argument is the same whether the proposal is to add collective action clauses to loan contracts, to mandate the inclusion of universal-debt-rollover options in loan agreements, to empower the IMF to sanction a standstill on payments, or to have the Fund lend into arrears to private creditors. Each of these practices, their critics warn, would weaken the

"This paper was prepared for the first annual IMF Research Conference. November 9-10, 2000. Barry Eichengreen is at the University of California at Berkeley, and Ashoka Mody is in the Research Department of the IMF. The authors thank Galina Hale and Antu Murshid for research assistance, Olivier Jeanne and Jeromin Zettelmeyer for help with data, and Charles Adams for comments.

155

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody bonding role of debt and raise the cost of funds for emerging markets. Friedman (2000) has put the point forcefully. The proponents of these ideas maintain, in contrast, that by preventing avoidable crises and more efficiently resolving those that still arise, these innovations may in fact reduce borrowing costs. As is frequently the case in economics, models can be constructed supporting both points of view. Miller and Stiglitz (1999) and Dooley (2000) illustrate the point. While Dooley lays out a model of international lending and borrowing in which measures that weaken the bonding role of debt do not just raise borrowing costs but actually reduce lending to the vanishing point, Miller and Stiglitz show that easier restructuring may render foreign lending more attractive by minimizing interruptions of debt service and preventing avoidable losses. Both arguments are logically consistent. Unfortunately, they have opposing implications for the cost of market access. In the absence of evidence, it is impossible to know which one dominates. A precisely analogous debate surrounds the IMF programs that these measures are intended to replace. The IMF and the official sector generally are fond of refer- ring to the "catalytic role" of the Fund. The news that the Fund is providing essential finance and that the government has agreed to pursue the reforms upon which that assistance is conditioned is supposed to restore investor confidence, catalyze an inflow of private capital, and reduce borrowing costs. Others are more critical of the effects of IMF programs on market conditions. The fact that a country has been forced to approach the Fund may be perceived as a signal of the depth of its difficulties that swamps any positive effect of prospective policy reforms. Moreover, governments have a decidedly mixed record of actually implementing the reforms on which IMF assistance is conditioned; if reform is not likely to follow, then investors will use the provision of official finance as an opportunity to get out rather than getting in. And, in the wake of the Asian crisis, the long lists of structural conditions that the Fund has attached to its loans have been criticized for doing more to undermine investor confidence, by emphasizing the existence of problems that no government can solve in short order, than they do to support it. Again, all of these arguments are consistent, but they have quite different implications for the impact of IMF assistance on borrowing costs. In the absence of evidence, it is impossible to know which one dominates. This paper is a modest contribution to the evidence on this subject. Section I reviews proposals for changing the way that capital flows are governed ex post and ex ante. Section II describes our data set and methods. We then provide evidence on the effects on the terms of market access of two types of interventions: IMF programs (Section III) and the adoption of collective action clauses (Section IV).1 Section V concludes. In part our message is pedagogic. There is an imbalance in the literature between theory and evidence. More energy has been devoted to modeling these interventions analytically than to identifying their effects empirically. This is an imbalance that should be redressed.

'We also provide some evidence on the effects of official guarantees and credit enhancements, which have been suggested by Corrigan (2000) as a way of encouraging the private provision of new credit to crisis countries. That evidence is limited, however, by the relatively small number of such guarantees in our data set.

156

©International Monetary Fund. Not for Redistribution BAIL-INS, BAILOUTS, AND BORROWING COSTS

In addition we have a substantive message, namely, that existing practices and prospective reforms have different effects on different types of borrowers. For example, the commitment effects associated with IMF loans appear to do little for either borrowers with very poor credit, who are least likely to be able to stick to the terms of their programs, or borrowers with very good credit, who are presumably able to commit to reform in their absence. In contrast, Fund programs do appear to enhance the market access of borrowers with "intermediate" credit, who, as a result of the additional commitment that these mechanisms entail, are able to borrow at lower cost. It follows that general statements about the effects of IMF programs run the risk of misleading if they are applied to emerging markets as a class. Moreover, there is a suggestion that not all Fund programs are created equal. Those accompanied by limited structural conditions have the most favorable effects on investor confidence, while those accompanied by elaborate structural conditions of a sort that governments are not likely to be able to deliver in short order have a negative effect and undermine confidence rather than inspiring it. Similar messages emerge with regard to changes in the provisions of loan contracts such as the more widespread adoption of collective action clauses. These clauses appear to raise the costs of market access for borrowers with low credit ratings but lower them for borrowers with high ratings. They support Dooley's prediction for the first group, in effect, but Miller and Stiglitz's for the second. The tendency for renegotiation-friendly contractual provisions to weaken the bonding role of debt and create moral hazard dominates for borrowers with poor credit, in other words, while for borrowers with good credit, who are less likely to act opportunistically, greater ease of restructuring dominates the increase in moral hazard. Again, it is likely to be misleading to apply statements about the impact on borrowing costs to emerging markets as a class.

I. A Review of the Proposals In this section we summarize and critique the proposals under review.

Collective Action Clauses Mechanisms to promote orderly restructuring are seen as opening up an alternative to large-scale financial rescues and the moral hazard they create. According to the advocates of collective action clauses, restructuring is difficult and costly under present institutional arrangements, rendering it impossible for the international financial institutions to stand aside if the markets refuse to roll over maturing claims or provide new money. The international financial institutions are then placed in an untenable position of having to back down on their previous commitment not to provide resources that can be used to finance exit by foreign investors. And because investors are aware of these facts, their behavior is not likely to be modified by the international financial institutions' less-than-credible statements of intent. The crux of the matter is thus the difficulty of debt restructuring, sovereign debt restructuring in particular, under prevailing institutional arrangements. American-

157

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody style instruments governed by New York state law typically require the unanimous consent of the bondholders to the terms of a restructuring.2 Many do not even provide for a bondholder assembly. This contrasts with bonds governed by U.K. law, which include provisions enabling the holders of debt securities to convene an assembly empowered to pass resolutions addressing issues relating to the settlement of defaults and other modifications to the original bond covenant, subject to the consent of bondholders holding a clear majority of the outstanding principal.3 Their resolutions are binding on all bondholders as long as the requisite majority agrees. And in contrast to American-style bonds, which contain no prohibition on legal action by dissident bondholders, U.K.-style bonds shelter the issuer from lawsuits by "vultures" seeking to hold up the restructuring.4 Amending loan contracts to include sharing, majority voting, and collective representation clauses is designed to address these problems and make restructuring a viable option. Majority voting and sharing clauses would discourage maverick creditors from resorting to lawsuits and erecting other obstacles to a settlement beneficial to the debtor and the majority of creditors. Clauses speci- fying who represents the bondholders and making provision for a bondholders committee or assembly would allow orderly solutions to be reached.5 The restructuring of problem debts could then be left to the consenting adults involved and reduce the pressure on the IMF to extend bailout loans.6 This mechanism for orderly restructurings could actually make emerging-market issues more attractive by minimizing acrimonious disputes, unproductive negotiations, and extended periods when no service is paid and growth is depressed by a suffocating debt overhang.7 The objection is that renegotiation-friendly provisions would make it too easy for countries to walk away from their debts. Collective action clauses would

2There are exceptions: a few U.S.-style bonds provide for amendments of payment terms with the approval of a qualified majority of bondholders. And most do not require unanimous consent to changes in their nonfmancial terms, as the case of Ecuador has recently demonstrated. We return to this case below. 3Typically a majority is 75 percent. Some covenants provide for lowering the necessary quorum to 25 percent if 75 percent of the bondholders cannot be reached. 4U.K. bonds governed by trustee deed agreements, but not those involving fiscal agents, generally prohibit individual bondholders from initiating litigation. The power to do so is vested with the trustee, acting on the instruction of creditors holding a specified fraction, typically at least 25 percent of the principal, who is required to distribute any funds recovered in proportion to the principal amount. -This was suggested in 1996 by the Group of Ten (G-10) in its post-Mexico report (G-10, 1996) and echoed in a series of recent Group of Twenty-Two (G-22) and Group of Seven (G-7) reports and declara- tions (G-22, 1998; G-7, 1999). The G-7 then placed the issue on its work program for reforming the international financial system with the goal of reaching a consensus by the Cologne Summit in June 1999. Two recent discussions of the operation of such provisions are Yanni (1999) and Drage and Mann (1999). 6Countries that attach a high value to maintaining market access would be free to take the extreme measures needed to keep current on their debts, while restructuring would now be viable for countries that lack the same capacity to adjust and that therefore attach priority to obtaining a reduction in debt-servicing costs. Countries that value a well-defined seniority structure could choose to restructure junior debts while leaving senior debts untouched. Limited IMF lending into arrears would become feasible if renegotiation- friendly provisions in loan contracts could be used to avert a major financial drain and extended loss of market access. 7As The Economist put it in a leader, "the prospect of an orderly renegotiation rather than a messy default might actually make some bonds more attractive" (Economist, 1999, p. 21).

158

©International Monetary Fund. Not for Redistribution BAIL-INS, BAILOUTS, AND BORROWING COSTS weaken the bonding role of debt, create moral hazard, disrupt credit market access, and raise borrowing costs.8 By compounding the difficulties that low- and middle-income countries already face when attempting to access international capital markets, the more widespread use of collective action clauses could be welfare reducing, so the conclusion follows. This dispute is empirical, not analytical. There is no disagreement that the more widespread use of collective action clauses would have two offsetting effects. By weakening the bonding role of debt, they would raise borrowing costs, but by facilitating debt restructuring and limiting efficiency losses ex post, they would reduce the costs of market access. Disagreement centers not on these analytics but on the quantitative importance of the two effects. In the wake of the exchange offers put on the table by Pakistan, Ukraine, and Ecuador, this debate has taken a different turn. It is now argued (by, for example, Roubini, 2000) that collective action clauses are superfluous because what cannot be accomplished by restructuring under New York law is still possible via exchange offers. Pakistan restructured its debt by completing a voluntary exchange without using the collective action clauses in its bonds. Ukraine restructured by completing a voluntary debt exchange without calling for a bondholders meeting. Ecuador restructured by completing a voluntary debt exchange of bonds that did not include collective action clauses, binding in reluctant bondholders through the use of "exit consents."9 Proponents of collective action clauses argue that this new view is too simple (see, for example, Kahn, 2000). The mere existence of such clauses helps to concentrate the minds of investors. The mechanism is analogous to corporate debt restructuring in the shadow of the court. While firms prefer to work out problems with their creditors directly, voluntary settlements are facilitated by the knowledge that the debtor will seek the shelter of the bankruptcy court if the effort to achieve an out-of-court settlement fails. Similarly, in the international context, everyone knows that if the voluntary exchange fails, the debtor will invoke the collective action clauses in his debt instruments, which allow the majority to bind in rogue creditors (and, not incidentally, provide the issuer with shelter from legal action). This increases the likelihood of voluntary action. In fact, this is precisely how collective action clauses have been used in recent restructurings. Pakistan warned that they might have to be invoked if a critical mass of investors did not accept its exchange offer. In Ukraine, where three of the four debt securities involved had collective action clauses, bondholders tendering their bonds for exchange were required to assign their voting rights to the trustee, who could then vote to bind in other bondholders if the need arose. In this view, collective action clauses are not the first recourse, nor should they be. But they are the final recourse, and a useful one.

8As William Rhodes has put it, "Approaches by the official sector to force the insertion of bankruptcy clauses into sovereign bond issues could limit the demand for these instruments, and generally inhibit market access for those emerging market countries implementing correct reform policies" (Institute of International Finance, 1999a, p. 2). '"Exit consents" involve changes in the nonfmancial terms of the bond to make it less attractive for maverick investors to hold out.

159

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody

These two views of collective action clauses have different empirical implications. If all the same things can be accomplished through exchange offers of debt securities lacking collective action clauses as through the restructuring of debt securities featuring them, then the contractual provisions in question will have no discernible impact on borrowing costs. Evidence of an impact, in contrast, is inconsistent with the view that collective action and representation provisions are superfluous.

Universal Debt Rollover Options Buiter and Sibert (1999) propose that a clause providing for a universal debt rollover option with penalty (UDROP) be added to all foreign-currency- denominated loans and credits as a way of dealing with the creditor panic problem. The borrower would then have the option of extending a maturing debt for a specified period (say, three months). While the regulatory authorities would mandate the inclusion of this option in all debt instruments, its precise terms could be negotiated between the debtor and its creditors when the loan agreement was written. To prevent the borrower from exercising the option under orderly market conditions and thereby avoid moral hazard, Buiter and Sibert propose requiring a debtor invoking the option to compensate the lender at a penalty rate and allowing the option to be invoked only once. Hence, a borrower who was insolvent would not be sheltered from the need to restructure its debts at the end of the rollover period.10 The effect on borrowing costs of mandating the inclusion of such options in international loan agreements is predictably ambiguous. On the one hand, there is the argument that, if repatriating their funds is made more difficult for foreign investors, they will demand an additional premium before they lend in the first place. On the other hand, if the presence of these provisions reduces the risk of bank-run-like liquidity crises, then that risk reduction will make lending to emerging markets more attractive, reducing rather than increasing spreads."

LOUDROPs should probably be thought of as complements to rather than substitutes for collective action clauses. UDROPs are designed for liquidity crises whose resolution requires only a temporary breathing space until confidence returns, and collective action clauses are designed for solvency problems that require write-downs and restructurings. 1 'Just as it may be in the collective interest of the depositors to sit tight but in their individual interest to queue up at the bank as soon as they see a line being formed—given the bank's rule of first come, first served—it can be in the collective interest of a country's creditors to roll over their maturing claims but in their individual interest to scramble for the exits if they see other creditors doing likewise—given that the limited availability of foreign reserves similarly creates a sequential service constraint. In the domestic bank run context, deposit insurance and, historically, temporary suspensions of the convertibility of deposits into currency are designed to alter these incentives and attenuate their effects. Deposit insurance minimizes the incentive for depositors to run. Temporary suspensions of convertibility allowed banks to avoid having to close down as a result of depositor runs and therefore of having to incur the associated costs. UDROPs are designed to mimic this function in the international setting. They would give the debtor a breathing space of, say, three months, a period of time assumed to be sufficient for the restoration of investor confidence and the resumption of business as usual. They obviate the need to declare a costly default.

160

©International Monetary Fund. Not for Redistribution BAIL-INS, BAILOUTS, AND BORROWING COSTS

Standstills

Another proposal is for resort to a payments standstill endorsed or sanctioned by the IMF. Williamson (1992) and Sachs (1994) first mooted this idea in the context of broader proposals for an international bankruptcy court. Eichengreen and Fortes (1995) appraised the proposal in a background paper to the Rey Report (G-10, 1996), where the official community offered a discussion of the idea. Following the Asian crisis, Canadian Finance Minister Paul Martin put before the Interim Committee a proposal that would have compelled IMF member countries to enact legislation requiring that all bonds and short-term borrowing instruments carry a covenant allowing the IMF to declare and/or approve a debt moratorium.12 If the problem is panic, then a payments standstill, like a universal debt rollover, would allow investors to collect their wits. It would give them time to reflect and to agree on mutually beneficial actions. It would allow the authorities to contact the creditors and encourage them to recognize their collective interest. Even if there are problems with fundamentals, the imposition of a temporary standstill would provide the government a chance to signal its commitment to the policy reforms needed for the restoration of confidence. In analogy to domestic bankruptcy procedures, a temporary stay could ensure that restructuring remains orderly—that it is not undermined by attempts to seize assets. By preventing the creditors from scrambling for the country's limited foreign exchange and from shutting off external finance for future economic activity (which would otherwise be unavailable for fear that this too will be garnished by the creditors), the country will then have the finance to grow out of its temporary problem. Total payments to the creditors, in present value terms, could be greater than if investors engaged in a grab race. Moreover, standstills, according to their proponents, would reduce the moral hazard associated with IMF bailouts by creating a viable alternative. At present, the IMF finds the extension of rescue loans irresistible because it does not want to see a country with good growth prospects strangled by a creditor panic that strips it of its reserves and denies it access to new external credits. This creates investor moral hazard, however, which can be avoided if the country and the Fund can invoke a standstill instead. Governments can already declare debt standstills unilaterally, and the IMF can already signal its approval verbally or by lending into arrears.13 But governments imposing standstills are still vulnerable to legal action by

12Canadian officials have subsequently elaborated somewhat more modest versions of the proposal (Murray. 2000), as have their British counterparts. '•'In their 1999 report on strengthening the international financial architecture, G-7 governments argued that "in exceptional cases, countries may impose capital or exchange controls as part of payments suspensions or standstills, in conjunction with IMF support for their policies and programs, to provide time for an orderly debt restructuring" (G-7, 1999, para. 50).

161

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody disaffected creditors, who may resort to litigation to seize assets.14 Legislation requiring that all bonds and short-terra borrowing instruments carry a covenant allowing the IMF to declare and/or approve a debt moratorium or an IMF-sanctioned stay that was binding on the creditors, as suggested by Martin, would shelter the government from such legal disruptions. Another option would be to amend the IMF Articles of Agreement to give the institution the power to officially "sanction" a stay on payments and to give that amendment effect in the relevant national courts (through national legislation or judicial precedent).15 These are not high-probability events. The most basic objection is that standstills would weaken creditor rights and raise borrowing costs.16 As the Institute of International Finance (1996, app. A. p. 29) has written, "the more investors perceive that institutional arrangements are trending towards 'no-fault default,' with minimal pain for the borrower and substantial risk of the politicization of debt, the less willing they will be to supply capital to the emerging markets." It has made the point specifically in connection with IMF-approved stay-of-litigation proposals: "Far from contributing to 'orderly' solutions, official statements [advocating IMF approval of stays of litigation by creditors] ... are raising doubts among market participants about the official community's commitment to upholding private contracts" (Dallara, 1999, p. 7). It follows that, as the Institute has put the point in another document: "Further consideration of stays is likely to have a significant dampening effect on the willingness of private creditors to provide cross-border financing . . ." (Institute of International Finance, 1999b, p. 12). These objections are telling. The IMF does not have the powers of a bankruptcy judge who can replace the management of the company in receivership, reorganize its financial affairs, and impose a settlement on uncooperative creditors. That recourse to a standstill might discourage adjustment by the government of the crisis country is a serious objection. As Summers (1996, p. 4) has put it, the analogy with corporate bankruptcy is flawed because "the safeguards against moral hazard built into domestic bankruptcy codes cannot be applied to sovereign debtors."

14The IMF has long suggested that fears of legal action may be exaggerated, both because it is costly and because the assets that can be seized are limited. But while the absence of legal action in connection with the recent debt exchanges of Ukraine, Pakistan, and Ecuador confirms that litigation is not inevitable, the recent Elliott case against Peru (in which an investment fund attempted to attach a government's payments to other creditors) suggests that neither the excessive-cost nor the inadequate-assets-to-seize argument is necessarily correct. ''Article VIII.2(b) presently provides that "exchange contracts which involve the currency of any member and which are contrary to the exchange control regulations of that member maintained or imposed consistently with this Agreement shall be unenforceable in the territories of any member." In other words. Article VIII.2(b) gives sanction to certain types of exchange controls. The shareholders in the Fund could agree to amend Article VIII.2(b) to make clear that it applies to capital as well as exchange controls, when the former were applied under extenuating circumstances. This would require approval by at least 50 percent of the member countries, which together held 85 percent of the total member countries. Alternatively, the Executive Board could give the article a new. definitive interpretation consistent with this broader coverage without taking a formal vote. "Other objections include the danger of contagion, if the declaration of a standstill by one country raises fears of the imposition of a standstill by another, and that the prospect of a standstill might only incite creditors to run to the exits more quickly in anticipation of its imposition. See Frankel and Roubini (2000).

162

Miller and Zhang (2000) argue, in contrast, that the IMF could limit moral hazard by attaching conditions to its approval or activation of the measure. If the cooling-off period provided by the standstill prevented the creditors' panicked rush for the exits from compounding the country's financial crisis, its economic problems would be lessened and its ability to pay enhanced. In principle, borrowing costs could fall rather than rise.17

Guarantees and Other Enhancements Official guarantees and similar credit enhancements have been suggested as another device for encouraging the private sector to provide new liquidity to crisis countries. The enhancements in question could range from plain vanilla guarantees of interest payments for a fixed period on new loans or capital market placements, to structured, one-time capital market placements that might have a put feature for a limited period at a market price below the issue price.18 As the IMF (1999, p. 63) has explained, such guarantees would be "intended to 'leverage' official capital, allowing a limited amount of official capital to support a larger amount of financing, while lowering the costs of private financing for emerging market borrowers. In general, the proposals aim to encourage a renewal of relations between governments and their private creditors, enhance the creditworthiness of the borrower, allow a speedier restoration of market confidence, and help to address concerns about burden sharing among official and private creditors." Corrigan (2000) motivates the case similarly: Enhancements could be used to get critically needed private money into the crisis country and limit the need for official financing. They would "provide incentives for private participation and incentives for reasonably rapid turn- around in the troubled country such as we have seen in Mexico in 1995 and Brazil, Korea and Thailand in 1998-99." Analytically, it is not clear that plain vanilla interest guarantees and more complex enhancements are really any different from IMF stand-by loans. In one case the Fund lends to the government, which then uses its monies to pay off investors if they choose to exit. In the other case the Fund provides resources directly to investors if the borrower halts payments or the market price of the security falls to the strike price. While the funds pass through the accounts of the government in one case but not in the other, their ultimate disposition is the same. Investors have the same incentive to purchase claims on the crisis country in the expectation that multilateral resources will be made available if further financial difficulties arise. The pressure for adjustment by the crisis country is

l7This is most plausible if the crisis is of the pure-liquidity variety, in which case the country's fundamental ability to pay would be unimpaired. Thus, Miller and Stiglit/ (1999) and Miller and Zhang (2000) suggest that the existence of a "Super Chapter 11" that prevented creditors from engaging in a grab race in response to macroeconomic shocks might prevent adverse balance sheet effects from transforming the disturbance into a full-blown crisis. And knowing that a crisis is less likely to result, investors would be less inclined to launch the grab race in the first place. Borrowing costs, rather than rising, might fall. 18That is, the (international financial institution) guarantor would agree to purchase the debt if the price fell to some set (strike) price below the issue price.

163

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody not obviously different in the two cases. Nothing else having changed, it is not clear that the magnitude of the official finance needed to aid the crisis country will be any different, that fewer investors will panic, or that more contrarians will bottom fish. It does not obviously follow that official enhancements "help to address concerns about burden sharing among official and private creditors" (to repeat the quotation above). It has been suggested that official enhancements of claims on the crisis country may encourage investors to coordinate on a cooperative equilibrium when providing new money is in their collective but not their individual interest.19 But again, it is not clear that guarantees are superior to conventional IMF loans for encouraging investors to coordinate on the more efficient equilibrium. This argument for enhancements is the same as that made for IMF stand-by loans, in other words, and it is subject to the same objections. In particular, just as with an IMF program, it is possible that the provision of an enhancement will be perceived as an adverse signal. If taken as a sign that the borrower is in dire straits, it may damage rather than enhance credit market access. It is not certain that the availability of external finance will rise and that its cost will fall, rather than the other way around.

Recapitulation There is no shortage of ideas for changing the arrangements governing international capital flows and providing alternatives to IMF rescue loans. One of the few things all such proposals have in common is that there is no consensus regarding their impact on the cost and availability of private credit to emerging markets, one of the key criteria used by diverse observers to judge their efficacy. This lack of consensus is symptomatic of a lack of evidence. It is to the development of such evidence that we now turn. A caveat is important before we proceed. Evidence that a particular reform is likely to raise or reduce borrowing costs does not, by itself, tell us whether that reform is desirable. To put the point another way, while everyone agrees that the cost of emerging-market finance is distorted, they do not agree on the direction. The dominant presumption in low- and middle-income countries is probably that asymmetric information and inadequate contract enforcement inflate the cost of capital for emerging markets, so anything that reduces those costs is efficiency and welfare enhancing. But observers situated in high- income countries argue that expectations of official bailouts have created spread compression and encouraged overlending to emerging markets (see Dooley, 1997, and McKinnon and Pill, 1997). If this effect dominates, then higher spreads will indicate a more efficient allocation of resources. In this paper we do not attempt to determine whether higher or lower spreads are better (whether an increase or a reduction moves the cost of borrowing toward its efficient, first-best level) but focus on the prior question of whether spreads are likely to go up or down.

19That is, when there is an "After you, Alphonse" problem.

164

II. Data and Methods

Our data set is essentially the universe of international bonds issued by emerging markets in the course of the 1990s.20 We assembled information on bonds issued between 1991-1 and 1999-IV from Capital Bondware.2[ Hence, the evidence developed here is exclusively for the bond market; we do not analyze the transactions in the international bank market.22 Throughout this paper, the basis for our empirical analysis is a two-equation model of the supply and demand for international debt—equivalently, a probit for the issue decision (the supply-of-debt decision) and a spreads equation (which indicates investor demand). The spreads equation is a linear relationship of this form: (1)

where the dependent variable is the logarithm of the spread; X is a vector of issue, issuer, and period characteristics; and MI is a random error. The spread (and its relationship to issuer and issuer characteristics) will be observed only when the decision to borrow and lend is made. Assume that spreads are observed when a latent variable B crosses a threshold B' defined by (2)

where X' is the vector of variables that determines the desire of borrowers to borrow and the willingness of lenders to lend, and u^ is a second error term. If the error terms in equations (1) and (2) are bivariate normal with standard deviations S] and 52 and covariance si22/siS2, this is a sample selection model, and equations (1) and (2) can be estimated simultaneously. They can be identified by the nonlinearity of the fitted probabilities in the selection equation or by the inclusion of elements in X' that are not also in X.

20Where information on characteristics of the issue or the issuing country was missing, we were forced to drop the bond in question. Out of an initial sample of 2.913 bonds, we were forced to drop 408 bonds because spreads are not reported (generally, these are bonds issued very early in the sample period) and 114 bonds because complementary information is not available. 2'Among the variables thereby obtained are the spread, maturity, and amount of each issue; whether it was privately placed; whether it was subject to a guarantee; whether the issuer was a private or govern- mental entity; whether the issue was denominated in dollars, yen. or deutsche marks; the industry of origin; whether the issuer was a sovereign, (other) public entity, or private sector issuer; and whether the interest rate was fixed or floating. We supplemented this with information on national and global macroeconomic variables drawn from the IMF's International Financial Statistics, The World Bank's World Development Indicators, and national sources. 22In an earlier paper (Eichengreen and Mody, 2000a), we analyzed the determinants of spreads in the international market for syndicated loans. One could imagine extending the analysis here to those instruments as well. Early critiques of proposals to bail in the private sector by, inter alia, pushing for the inclusion of collective action provisions in bond covenants (e.g., Institute of International Finance, 1996), were critical on the grounds that bank loans remained the principal vehicle for portfolio capital flows to emerging markets. With the subsequent decline of syndicated bank lending and the continued progress of securiti/.a- tion, this objection has clearly lost some of its force.

165

Throughout, we focus on primary-market (launch) spreads. We include the following variables as measures of creditworthiness in the spreads equation: external debt relative to GNP, debt service relative to exports, whether a debt restructuring agreement has been concluded within the previous year with either private or official creditors, the growth rate of real GDP, the variance of export growth, the ratio of reserves to short-term debt, and the ratio of domestic private credit to GDP. We also use a subjective measure of political risk constructed from country credit ratings provided by Institutional Investor.23 We include the log of the 10-year U.S. Treasury rate (to capture the opportunity cost of lending to emerging markets), the swap rate (the market measure commonly used to measure investor risk tolerance), and the difference between the 10-year and 1-year U.S. Treasury rates (to represent the slope of the yield curve and hence the appetite for difference maturities).24 We estimate the determinants of spreads and the probit for the borrowing decision as a system, by maximum likelihood. Estimating the determinants of market access requires information on those who did not issue bonds. For each country we consider three categories of issuers: sovereign, (other) public, and private. For each quarter and country where one of these issuers did not come to the market, we record a zero, and where they did we record a one.25

III. Bailouts IMF programs are the benchmark against which reform proposals should be judged. We therefore start by analyzing the association of Fund programs with the cost and availability of finance. Do countries approaching the IMF for assistance find it easier or harder to access financial markets subsequently? Do they pay higher or lower spreads? We address these questions by including indicator variables for IMF programs in the issue and spreads equations specified above, which allows us to consider the impact on both quantities and prices. There is a large literature on the effects of IMF programs, much of which is rendered inconclusive by methodological problems.26

23The advantage of the Institutional Investor data over the Moody's/S&P ratings used by most previous authors is more complete country coverage and more regular publication (the data are biannual). Because Institutional Investor's country credit rating is correlated with other issuer characteristics, to include it in the spreads equation, where many of these other issuer characteristics also appear, compli- cates interpretation. We therefore substituted the residual from a first-stage regression in which the credit rating was regressed on the ratio of debt to GNP, the debt rescheduling dummy, the ratio of reserves to GNP, the rate of GDP growth, and the variance of export growth. In addition to entering these variables in levels, we interacted them with a dummy variable for Latin America. Representative results are reported in Eichengreen and Mody (2000b); we suppress them here in the interest of space. 24Ten-year rates are appropriate insofar as the term to maturity of the underlying asset is broadly similar to that on the bonds in our sample. 25To conserve space, we do not discuss the estimator, the results of estimating the probit, or other coefficients in the spreads equation. See Eichengreen and Mody (2000b, 2000c) for details. 26Among the prominent contributions to this literature are the following. Hajivassiliou (1987) finds a negative relationship between IMF involvement and the subsequent supply of new loans for a cross section of developing countries in the period 1970-82. Faini and others (1991) and Killick (1995) rind the same for the 1980s. Rowlands (1994) disaggregates public and private flows and finds that private flows respond negatively, public- flows positively. A recent study by Bird and Rowlands (1997) finds that the effect is unstable but, when significant, strongly negative. The one study of which we are aware of the impact of IMF programs not on gross or net capital flows but on spreads, by Ozler (1993), finds a consistent positive impact, again inconsistent with the idea of a catalytic role. But Ozler's analysis is of syndicated bank loans, not bonds, and it is for an earlier period.

166

For example, there is the potential endogeneity of IMF programs. Fortunately, this problem is less serious in our context than others. Our dependent variable is the individual bond issue, not a macroeconomic aggregate like the growth rate of aggregate capital flows. It is less likely that the decision to approach the IMF and the Fund's decision to help are affected by the success of an individual bond issue (given that a number of the program countries in our sample floated multiple issues in periods when they were involved with the Fund) than that they are affected by the growth rate or the overall level of capital flows. Hausman tests do not allow us to reject the null that IMF programs are exogenous with respect to individual security issues. Other problems remain. For example, if one finds a certain association between IMF programs and subsequent economic performance or market outcomes, it is not possible to say whether that association reflects the impact of Fund programs per se or the implications of other unobservable characteristics of the country that both prompt it to approach the Fund and shape the performance of its economy and the reaction of the markets. These alternatives are sometimes referred to as the "commitment" and "signaling" interpretations—that Fund programs are a commitment device that enhances the government's dedication to reform, on the one hand, and that they send a signal of the depth of the country's problems, on the other. Fancy econometric fixes for this problem are no more convincing than the restrictive assumptions on which they are based. Our presumption is that the commitment and signaling effects operate to different degrees on different borrowers. Specifically, we are interested in the differential impact of IMF programs across countries with different credit ratings—in the hypothesis that the market will view Fund programs differently depending on the credit quality of the government. We are also interested in the differential effects of different types of conditionality. In column 1 of Table 1, we show the results of the probit relating the decision to issue a bond to a vector of country characteristics (the growth rate, the level of indebtedness, and so on), a vector of market conditions (the U.S. treasury bond rate, the yield curve, and the log swap rate), and whether or not the country has negotiated an IMF program. The estimated probit includes interaction terms for Latin America. As such, the estimated coefficients for Latin America, when the dummy for Latin America takes on the value 1, are the sum of the coefficients in columns la and Ib. In columns 2 through 6, we relate many of these same variables to the determination of spreads.27 It appears that IMF programs have a positive impact on market access, other things being equal, and a negative impact on spreads,

27As explained in Section III, the probit equation for borrowing and the spreads equation are estimated simultaneously. We identify the borrowing equation by omitting the size of the issue, its maturity, and whether it is privately placed. The probit has variables not in the spreads equation: debt service/exports, short-term/total debt, and reserves/imports. Note that identification is also provided by the nonlinearity in the probit. These equations can be thought of as part of a larger system in which the decision to borrow, the amount borrowed, the maturity of the obligation, and its price are simultaneously determined. We have taken a step toward estimating that larger model in Eichengreen, Hale, and Mody (2001), where we analy/e the determinants of issuance, maturity, and spreads as a system. As shown there, the other results reported in Tables 1 and 2 are largely unaffected by this extension.

167

Table 1. Full-Sample Estimates of Impact of IMF Programs on Market Access Probability of Bond Issuance0 Log of Spread at the Time of Issue (la) db) (2) (3) (4) (5) (6) Non-Latin Latin American American All issuers issuers interactions Log amount -0.006 -0.005 -0.005 -0.004 -0.008 (-0.36) (-0.32) (-0.34) (-0.22) (-0.47) Maturity 0.003 0.003 0.003 0.003 0.002 (1.24) (1.28) (1.28) (1.28) (0.97) Private placement 0.073 0.073 0.075 0.074 0.068 (2.87) (2.87) (2.95) (2.91) (2.70) Log of 10-year 0.032 -0.504 -0.356 -0.351 -0.347 -0.359 -0.324 U.S. Treasury rate (0.43) (-3.55) (-2.40) (-2.38) (-2.35) (-2.42) (-2.19) Log (10-year-l-year) -0.097 0.061 0.030 0.026 0.028 0.031 0.056 U.S. treasury rate (-7.83) (2.53) (1.24) (1.08) (1.17) (1.29) (2.24) Log swap rate -0.221 -0.273 0.668 0.654 0.670 0.677 0.657 (-7.42) (-4.72) (9.68) (9.56) (9.71) (9.79) (9.52) Credit rating residual 0.015 0.002 -0.050 -0.050 -0.050 -0.050 -0.050 (18.50) (0.002) (-27.86) (-27.86) (-27.99) (-27.92) (-27.88) Debt/GNP -0.411 -0.991 1.325 1 .303 1.305 1.304 1.276 (-8.20) (-9.43) (14.41) (14.24) (14.40) (14.14) (13.83) Debt rescheduled -0.010 -0.024 0.195 0.181 0.204 0.205 0.213 in previous year (-0.30) (-0.51) (4.27) (4.05) (4.39) (4.39) (4.58) GDP growth 5.405 -4.961 -9.022 -8.68 -9.024 -9.199 -9.717 (8.97) (-2.70) (-4.86) (-4.71) (-4.87) (-4.95) (-5.23) Standard deviation -0.909 -0.412 2.271 2.223 2.226 2.265 2.209 of export growth (-10.22) (-2.48) (11.16) (10.84) (11.10) (10.98) (10.73) Reserves/short-term -0.033 -0.107 -0.038 -0.036 -0.038 -0.038 -0.042 debt (-6.60) (-5.40) (-3.12) (-3.03) (-3.15) (-3.16) (-3.51) Ratio of domestic 0.029 -0.058 -0.035 -0.033 -0.034 -0.036 -0.024 credit to GDP (4.34) (-2.81) (-2.61) (-2.48) (-2.59) (-2.67) (-1.76) IMF program 0.103 0.093 -0.055 (4.63) (2.53) (-1.53) Stand-by arrangement -0.004 -0.030 -0.085 (-0.11) (-0.75) (-1.99) Extended Fund Facility -0.069 -0.080 -0.153 (-1.75) (-1.90) (-3.34) Enhanced Structural 0.449 0.379 Adjustment Facility (1.53) (1.30) Number of quarters in 0.009 an IMF program (4.03) Constant 1.000 3.465 3.483 3.440 3.439 3.427 (7.69) (8.39) (8.45) (8.33) (8.32) (8.32) Lambda -0.618 -0.613 -0.615 -0.620 -0,611 (-26.35) (-25.87) (-26.29) (-26.54) (-26.13) Number of 7,355 7,355 2,381 2,381 2,381 2,381 2,381 observations/bonds Pseudo/?2 0.400 0.400 Log of likelihood -2,916.16 -2,916.16 -4,861.86 -4,863.04 -4,861.51 -4,859.93 ^t,851.81

Notes: Dummy variables for private and public issuers, industrial sectors, currency of issue, and whether the bond was issued at a fixed or floating rate were included in the regressions but are not reported here. Figures in parentheses are f-statistics. Number of observations are reported for probits, and number of bonds are reported for spreads equation. 4JThe coefficients for the probit are normalized to the partial derivative of the probability distribution function with respect to a small change in the independent variable, evaluated at the average values of the independent variables. Additional variables in the probit include debt-service/exports, short-term/total debt, and reserves/imports. Note that column Ib lists only additional interaction effects for Latin borrowers.

168

©International Monetary Fund. Not for Redistribution BAIL-INS, BAILOUTS, AND BORROWING COSTS although the first effect is more robust than the second.28 In contrast to most of their predecessors, these results are more readily reconciled with the (positive) commitment interpretation than the (negative) signaling effect. They also provide some evidence of the catalytic role of IMF programs—or of the moral hazard created by IMF lending if one prefers to interpret them that way. There are some interesting differences by program type and history. We distinguish stand-by arrangements, support through the Extended Fund Facility (EFF), and Structural Adjustment Facility (SAF) and Enhanced Structural Adjustment Facility (ESAF) arrangements, in contrast to previous quantitative analyses which clump all Fund programs together. Stand-by arrangements and EFF loans have a negative impact on spreads, although only for the latter does that effect approach significance at the standard 5 percent level (i.e., when the absolute value of the f-statistic is 2 or more) when each type of arrangement is included in the model by itself. For ESAF loans, there is no evidence of lower spreads: the coefficient in question is positive (although it is again indistinguishable from zero at conventional confidence levels). When we include measures of all three types of arrangements at the same time, these findings are reinforced: now the negative impact of stand-by and EFF loans is significantly less than zero at conventional confidence levels, and EFF loans have roughly twice the impact of stand-by arrangements on borrowing costs. The effect of ESAF loans on spreads continues to be positive but insignificantly different from zero. Interestingly, we also find that the impact on spreads weakens with the length of IMF involvement.29 The magnitudes are not insignificant. The coefficient on stand-by arrangements when all three types of Fund programs are entered simultaneously (in 6) implies that their presence reduces spreads by 24 basis points, which is 8.5 percent (relative to a sample average of 280 basis points). For EFF arrangements, spreads fall by 15.3 percent, which is 43 basis points. These benefits are eliminated once a country has been in a stand-by arrangement for 10 quarters and in an EFF arrangement for 18 quarters. It is tempting to interpret these patterns in terms of differences in policy condi- tionally and in the perceived probability of country compliance. Goldstein (2000) identifies three types of conditionality: macro, macro with a light emphasis on structural reform, and macro with a heavy structural emphasis. Stand-by arrangements tend to be of the first type—that is, they are accompanied mainly by macroeconomic conditions. EFF programs tend to have more structural conditions but a somewhat lower compliance rate. ESAF loans have the most structural conditions and, again, a mixed record of compliance. This suggests the following interpretation of the patterns in Table 1. The conditions associated with stand-by arrangements enhance market access by strengthening the commitment of the government to removing the (mainly) macroeconomic imbalances that led it to approach the Fund. The conditions associated with EFF arrangements, which are in the nature of "first generation reforms," further enhance

28The positive effect on access appears to hold whether we assume that the determinants of borrowing are the same across regions (as in the left-hand side of column 1) or allow them to differ between Latin America and other parts of the world (as on the right). 29We attempt to pick up the effects of serial borrowing by continuing to count previous quarters of IMF involvement if a country drops out of and back into an IMF program.

169

©International Monetary Fund. Not for Redistribution Barry Eichengreen and Ashoka Mody market access by strengthening the commitment of the authorities to addressing not just macroeconomic imbalances but also some of the structural causes of the crisis. That these effects weaken with program length suggests that the markets see chronic involvement with the Fund as indicative of compliance problems. The fact that ESAF loans do not appear to have a favorable impact on terms of market access may reflect doubts that the authorities can effectively push through by executive fiat the more ambitious "second generation reforms" demanded by the Fund. The lack of an effect (and perhaps even an unfavorable effect) on borrowing costs is not obviously consistent with the official view that more comprehensive and ambitious conditionality of the sort that the Fund has attached to some of its recent programs is necessary or desirable for restoring market access. All that these extensive requirements may do is to alert the markets to the existence of structural problems (accounting for the positive coefficient on ESAF loans on our spreads equations). These results can be read as supporting the view that second-generation reforms that take more time and are more difficult to implement are best left until the crisis passes. In Table 2, we disaggregate by Institutional Investor credit rating, placing countries into four groups: 0-30, 30-50, 50-70, and 70-1OO.30 Not surprisingly, we observe no bond issuer from a country with a rating of 70 or higher with an IMF program, and no country in the 50-70 range with an EEF or ESAF loan. For countries in the lowest rating category (0-30), the coefficients on IMF programs are small and insignificant statistically (column 1). It would appear that the country's economic problems and difficulty in making a credible commitment to policy reform are already factored into spreads and that the arrival of the IMF is not seen as making a difference in this regard. For borrowers in the next rating category (30-50), in contrast, IMF programs tend to be associated with a reduction in spreads, other things being equal (column 2), especially if they are provided through the EFF. An interpretation is that the bad news about these countries is already known, but the arrival of IMF support has a commitment-strengthening effect.31 The opposite is true for countries in the next rating category (50-70), where IMF programs tend to be associated with an increase in spreads (column 3). The main effect of approaching the Fund in this case would appear to be to send bad news to the market about these otherwise creditworthy countries.32

-"-'The countries that fall into the four groups are listed in the Appendix, for four different years. Many of the lowest rated countries (0-30) are not associated with "emerging market" status, though some notable ones, such as Russia and Ukraine, are important borrowers in international markets. We distinguish these four rating groups because disaggregating in this way sheds light on the effects of the other initiatives we consider below and because we used this disaggregation in previous work (Eichengreen and Mody, 2000b, 2000c), which facilitates comparisons with the results of this paper. We do not report the associated probit for the decision to borrow, although, as implied by the presence of the Inverse Mills Ratio, we again estimate the decision to borrow and pricing equations as a system. 3'Again, prolonged involvement with the Fund causes this benefit to evaporate, but at a slower pace for this category of borrowers than for the same as a whole. 32We also examined whether these effects differed by the dollar value of the IMF commitment, normalized by the amount of debt service due in that year. The results, in columns 4-6, do not show consistent impact of Fund programs on borrowing costs. We interpret this as suggesting that it is the news of the program, and its implication for prospective policy reform, rather than the exact amount of financial assistance associated with it, that matters for the majority of countries.

170

Table 2. Subsample Estimates of Impact of IMF Programs on Spreads Log of Spread at the Time of Issue

Credit rating Credit rating Credit rating Credit rating Credit rating 30 or more but 50 or more but Credit rating 30 or more but 50 or more but less than 30 less than 50 less than 70 less than 30 less than 50 less than 70 (1) (2) (3) (4) (5) (6) Log amount -0.141 -0.043 0.035 -0.144 -0.041 -0.001 (-2.90) (-2.39) (1.06) (-3.01) (-2.28) (-002) Maturity 0.001 0.001 0.012 0.005 0.002 0.009 (0.08) (0.66) (2.72) (0.38) (0.80) (2.02) Private placement -0.051 0.065 0.090 -0.058 0.067 0.067 (-0.81) (2.40) (1.62) (-0.93) (2.49) (1.12) Log of 10-year -0.930 -0.089 -0.205 -0.974 -0.055 0.075 L'.S. treasury rate (-2.17) (-0.55) (-0.66) (-2.31) (-0.34) (0.18) Log (10-year minus 1-year) 0.075 -0.002 0.014 0.012 -0.037 0.004 U.S. treasury rate (0.67) (-0.08) (0.22) (0.12) (-1.45) (0.05) Log swap rate 0.466 0.448 0.813 0.414 0.426 1.25 (2.02) (6.42) (5.12) (2.06) (6.20) (6.40) Credit rating residual -0.025 -OO30 -0.093 -0.022 -0.027 -0.092 (-1.84) (-9.20) (-13.74) (-1.92) (-8.52) (-9.40) Debt/GNP 0.260 0.659 1.287 0.282 0.590 0.881 (0.73) (4.61) (7.26) (0.81) (4.14) (2.14) Debt rescheduled 0.171 0.120 -0.093 0.160 0.054 0.881 in previous year (1.34) (2.49) (-13.74) (1.27) (1.08) (2.14) GDP growth -12.361 -3.797 -27.696 -14.05 -1.775 -27.82 (-2.63) (-1.84) (-5.23) (-3.12) (-0.88) (-4.30) Standard deviation -0.145 1.355 3.764 -0.312 1.184 3.327 of export growth (-0.21) (6.36) (5.71) (-0.51) (5.66) (4.51) Reserves/short-term -0.056 -0.127 -0.017 -0.056 -0.115 -0.042 debt (-2.10) (-7.39) (-0.69) (-2.15) (-6.60) (-1.23) Ratio of domestic -0.157 -0.148 -0.102 -0.148 -0.152 -0.096 credit to GDP (-3.45) (-6.14) (-3.78) (-3.71) (-6.37) (-2.46) Stand-by arrangement -0.087 -0.065 0.364 (-0.96) (-1.49) (2.12) Extended Fund Facility -0.189 -0.114 (-1.55) (-2.61) Enhanced Structural 0.261 -0.010 Adjustment Facility (0.71) (-0.03) Stand-by arrangement -0.001 0.0002 -0.00002 amount (-1.55) (0.83) (-0.03) Extended Fund Facility -0.0001 0.0001 amount (-1.24) (1.61) Enhanced Structural -0.0001 0.0002 Adjustment Facility (-0.20) (0.30) amount Number of quarters 0.010 0.005 0.027 in an IMF program (0.81) (2.10) (3.32) Constant 5.793 4.317 3.343 6.238 4.313 1.458 (1.12) (9.34) (3.92) (6.47) (9.34) (1.22) Lambda -0.034 -0.465 -0.695 -0.022 -0.452 -0.725 (-0.49) (-17.62) (-16.98) (-0.34) (-16.11) (0.06) Number of bonds 275 1,245 588 275 1.245 385 Log of likelihood -515.11 -2,057.45 -1,084.63 -514.65 -2.059.91 -768.50

171

Thus, there is some evidence in the reaction of financial markets to the news of IMF programs that Fund loans enhance market access. They do so mainly when they have macroeconomic and modest structural conditions attached, but not when they require more far-reaching and difficult-to- implement structural reforms. They do so mainly for countries of intermediate creditworthiness, for whom involvement with the Fund can be seen as strengthening the commitment to reform, and not for countries with poor credit (where program compliance is viewed as unlikely) or good credit (where the commitment to reform is already strong). These results will hearten official observers concerned to document the "catalytic effect" of IMF programs, although they also suggest cautions about the type of IMF programs that exercise this effect. To the extent that investors rush in and spreads are compressed following the announcement of IMF programs, these results can also be interpreted as evidence of moral hazard, although the differential pattern of effects is not easily reconciled with the moral hazard view.

IV. Bail-Ins In a previous paper (Eichengreen and Mody, 2000c), we analyzed the impact on spreads of collective representation clauses, comparing bonds issued under U.K. governing law with comparable bonds subject to U.S. law in the period 1991-98.33 Earlier contributions had failed to find an impact on borrowing costs (e.g., see Deutsche Bundesbank, 1999; Tsatsaronis, 1999; and Griffith-Jones, Ocampo, and Cailloux, 1999). We suggested that this resulted from their failure to control for other characteristics of the issue and the issuer, to adjust for the selectivity associated with the decision to borrow, to allow for the endogeneity of the choice of governing law, and to permit different effects for more and less creditworthy borrowers. Upon doing so, we found that opting for collective action clauses raised spreads for borrowers with low credit ratings (below 50 on the Institutional Investor scale) but reduced them for borrowers with high credit ratings (above 50 by this measure). The impact was particularly strong at the extremes—for countries with Institutional Investor ratings below 30 and above 70. The obvious interpretation is as follows. More creditworthy borrowers value their capital market access and are not likely to walk away from their debts. Including collective action clauses in their loan contracts does not significantly aggravate moral hazard. In the exceptional circumstance that such borrowers have difficulties in servicing their debts, the fact that investors can avail themselves of provisions that allow them to restructure their claims in an orderly way is viewed positively by the markets. For less creditworthy borrowers, in contrast, the presence of collective action clauses aggravates moral hazard and increases borrowing costs. The two effects work in opposite directions, resulting in a small and insignificant overall impact on borrowing costs, but in noticeable net effects, opposite in sign, for different credit rating classes. The existence of these effects

33The latter typically means New York state law. Note that we also have some bonds subject to other, mainly German and Japanese, governing laws.

172

©International Monetary Fund. Not for Redistribution BAIL-INS, BAILOUTS, AND BORROWING COSTS is hard to reconcile with the new conventional wisdom that the same things can be accomplished through exchange offers under U.S. law, at the same cost, as can be accomplished in the presence of collective action clauses. This analysis covered the period through 1998, ending before collective action clauses became the subject of debate and scrutiny, making it worthwhile to ask whether the results carry over when we update the data. In addition, we did not ask whether the results were different for sovereign and other borrowers. The argument for collective action clauses on efficiency-of-resolution grounds is strongest for sovereign borrowers because unlike corporations, sovereigns cannot resort to court proceedings to resolve their financial difficulties in an orderly way. The argument against them on moral hazard grounds is also strongest because there is no court to reach into governments' financial affairs and impose sanctions for opportunism. Finally, we did not analyze the determinants of choice of governing law in detail. We relax these limitations in what follows. As before, we control for selectivity and for the endogeneity of the governing law. We now include bonds subject to Luxembourg law under the "U.K. law" heading because the former also include collective action provisions.34 We instrument the choice of governing law because there is good reason to think that it should be regarded as a choice variable.35 And, as noted above, we now use data through the end of 1999. If our interpretation is appropriate, then the advantages of bonding (which should attract borrowers to U.S. law) and easy recontracting (which should attract them to U.K. law) should be reflected in their choice of contracting terms. Table 3 shows the number of bonds issued under the provisions of U.K. law, U.S. law, and other governing laws for each of our four credit-rating categories.36 The bulk of issuance by borrowers with the lowest credit ratings (0-30 on the Institutional Investor scale) is subject to U.K. law. This plausibly reflects the value to borrowers and lenders of having renegotiation-friendly procedures in place on loans for which restructuring is in any case a relatively high-probability event. Borrowers in the next higher rating category (30-50) do more than half of their borrowing under U.S. law. For this class of borrowers there is a lower probability of having to restructure, and issuers are apparently willing to accept greater difficulty of restructuring in return for the greater commitment provided by U.S. law. Finally, the bulk of issuance by borrowers with high credit ratings (50 and above on the Institutional Investor scale) is again subject to the provisions of U.K. law, presumably reflecting the ability of these borrowers to commit to repay without the addition of inflexible contract terms,

34See Dixon and Wall (2000). This is not something we did in our previous study. 35In addition, the possibility of measurement error makes it important to instrument our measure of the presence or absence of collective action clauses. Recall that we have information on the governing law (U.S., U.K., or other) but not the presence or absence of these clauses. In fact, there are a few instances where provisions for collective representation of the bondholders are included in bonds governed by U.S. law and where no such provisions are included in bonds governed by U.K. law. The appropriate treatment for this kind of measurement error is to instrument the variable in question, which is what we do in the empirical analysis above. If the governing law measures the presence of collective action clauses with random error, then the use of instrumental variables will provide a reliable indication of the magnitude and significance of the impact of the latter. 36Recall that Luxembourg law is included under the U.K. heading, here and in what follows.

173

Table 3. Characteristics of Borrowers by Rating Category and Type of Governing Law

Credit Rating Credit rating Credit rating Credit rating Credit rating 30 or more 50 or more more than 70 less than 30 but less than 50 but less than 70 but less than 100 Total Governing Laws (1) (2) (3) (4) (5)

United Kingdom Number of bonds issued 176.00 549.00 379.00 197.00 1,301.00 Average spread paid 484.22 340.02 122.71 55.66 253.15 Log of amount 4.44 4.70 4.33 4.42 4.52 Maturity in years 4.04 5.42 4.79 4.55 4.92 Share of private issuers 0.62 0.53 0.72 0.68 0.62

United States Number of bonds issued 85.00 633.00 198.00 39.00 955.00 Average spread paid 449.34 367.65 225.78 76.35 335.89 Log of amount 4.27 5.06 5.28 5.44 5.05 Maturity in years 3.79 8.13 11.89 11.13 8.65 Share of private issuers 0.75 0.63 0.64 0.54 0.64

Other Number of bonds issued 45.00 352.00 198.00 55.00 650.00 Average spread paid 384.81 306.26 119.39 86.07 235.18 Log of amount 3.75 5.00 4.57 4.74 4.76 Maturity in years 3.44 5.81 6.38 5.20 5.77 Share of private issuers 0.73 0.21 0.41 0.47 0.33

Total Number of bonds issued 306.00 1,534.00 775.00 291.00 2,906.00 Average spread paid 462.89 346.22 150.22 63.43 278.98 Log of amount 4.29 4.92 4.64 4.62 4.75 Maturity in years 3.89 6.63 7.01 5.55 6.33 Share of private issuers 0.67 0.50 0.62 0.62 0.56 something that allows them the option value of easy renegotiability. Who borrows under what kind of governing law is thus explicable in terms of our conceptual framework. These patterns are harder to rationalize if one believes that governing law is irrelevant (because restructuring can be accomplished as easily in the presence and absence of collective action clauses) or that borrowers and lenders have historically failed to pay close attention to these provisions. Conceivably, these patterns reflect technical characteristics of the market on which the issuer borrows, correlated with the choice of governing law but omitted from these bivariate comparisons in Table 3. In other words, the market that borrowers approach is determined by the level of interest rates and other conditions that make borrowing there attractive, and the governing law is an incidental product of that choice (Becker, Richards, and Thaicharoen, 2000). This is a testable proposition: if the choice of governing law depends on characteristics of the borrower and not merely characteristics of the market, then this is evidence for our interpretation and against the view of governing laws as a purely incidental characteristic of the choice of market.

174

As reported in Table 4, where we relate the choice of governing law to the state of the market, technical characteristics of the loan, and the same credit rating measure as before, we continue to find that the characteristics of the issuer are important, inconsistent with the aforementioned critique. Issuers from countries with the highest and lowest ratings continue to be most inclined to issue under U.K. law.37 The probability of issuing under U.S. law peaks at a rating of approximately 50. Again, it appears that for issuers with the lowest ratings and the highest probability of having to restructure, borrowers and lenders prefer having in place a contractual mechanism for restructuring, while for issuers with the highest ratings and no need for the additional bonding that U.S. law provides, having recourse to renegotiation-friendly provisions in the event of an extraordinary contingency has option value. The key point is that this association between borrower credit quality and choice of governing law is consistent with our emphasis on the trade-off between commitment and ease of renegotiation. It is hard to imagine an explanation for these patterns grounded purely in the technical characteristics of the relevant markets.38 Finally, Table 5 reports the results for spreads, separately for countries in different Institutional Investor rating categories. Many of the coefficients are less precisely estimated than for the full set of bonds (as shown in Eichengreen and Mody, 2000c), reflecting the now smaller sample size.39 But the results of particular interest, on U.K. governing law, are relatively robust. The presence of collective action clauses, so measured, raises borrowing costs for countries with poor credit while reducing costs for countries with relatively good ratings. The coefficients on U.K. governing law shift smoothly from large positive to small positive, to small negative, to large negative, as we move up the credit rating gradient (columns 1-4). Figure 1 presents a diagrammatic representation of the results. Spreads decline with improved credit rating at a higher rate for U.K. law bonds than for U.S. law

37This parsimonious specification is designed to highlight the importance of credit quality for choice of governing law. In Eichengreen and Mody (2000b, 2000c) we show that in more elaborate specifications, a number of other economic characteristics of the country and the borrower also have a significant impact on the choice of governing law. 38We experimented with a variety of other variables in efforts to probe further whether the governing law should really be regarded as a purely technical characteristic of the market on which debtors choose to borrow (for reasons unrelated to extent of commitment or ease of restructuring). For example, we added U.K., German, and U.S. interest rates to the multinominal logit as determinants of the preference for British as opposed to German and U.S. markets. None of these changes significantly reduced the importance of credit quality on the choice of law (or eliminated the impact of governing law on borrowing costs, as we will see below). It is always possible, of course, that other characteristics of borrowers that we find difficult to observe and measure determine the choice of market (e.g.. South Africa has issued very few bonds in the United States because borrowers there have long-standing ties to issuing banks in London, while Latin American countries issue bonds in New York because they have long-standing ties with U.S. banks and the ultimate buyers are disproportionately American). Again, however, if this is the dominant factor, it is hard to imagine how we would find such a consistent and intuitive association between credit quality (and other borrower characteristics) on the one hand and the choice of market (and borrowing costs) on the other. 39We have 356 sovereign bonds for countries with credit ratings below 50, but only 37 bonds for sovereigns with ratings above 50.

175

Table 4. Determinants of Choice of Governing Law U.K. Governing Law" Other Governing Laws" Log amount -0.438 -0.649 (-6.02) (-6.90) Maturity -0.122 -0.136 (-7.74) (-6.53) Private placement -0.536 -1.214 (-5.07) (-7.38) Log of 10-year U.S. treasury rate -0.081 0.142 (-0.15) (0.19) Log (1 0-year- 1 -year) U.S. treasury rate 0.227 0.921 (2.44) (7.05) Log swap rate -0.469 0.761 (-1.95) (2.38) Credit rating -0.148 0.113 (-4.36) (2.18) Credit rating squared 0.001 -0.001 (3.66) (-1.96) IMF program interactions with: Rating less than 30 -0.870 0.458 (-3.29) (1.13) Rating 30-50 -0.198 -0.089 (-1.41) (-0.46) Rating 50-70 -0.033 -0.777 (-0.09) (-1.46) Foreign guarantee -0.228 -1.574 (-0.71) (-2.39) Domestic guarantee 0.130 -0.509 (0.82) (-2.14) Number of bonds 2,893 2,893 Pseudo R2 0.3497 0.3497 Log of likelihood -1,992.53 -1,992.53

Notes: Dummy variables for private and public issuers, industrial sectors, currency of issue, and whether the bond was issued at a fixed or floating rate were included in the regressions but are not reported here. Figures in parentheses are ^-statistics. "The base category is the U.S. governing law. bonds. When we do not instrument for the choice of law (and for the measurement error that exists because these laws do not perfectly represent the presence or absence of collective action clauses), the U.K. law bonds are seen to have somewhat higher spreads than U.S. law bonds at low ratings, but the differences are not statistically significant. However, when we replace the dummy variable for law by the probability of issuance under the law, the U.S. law line, in effect, swivels, creating a larger positive difference at the lower end and a larger negative difference at the higher end. The implication at the lower end of the rating spectrum is that unobserved factors that lead to higher spreads also lead to choice of the U.S. law. The opposite is the case at the higher end of the rating scale.

176

In columns 5-6 of Table 5, we report similar equations for sovereigns alone. The effects are similar—higher spreads for borrowers with poor credit ratings, lower spreads for more creditworthy borrowers. Levels of statistical significance are lower, however, reflecting the still smaller sample size.40 We can raise the precision of the estimates if we impose the additional assumption that the other determinants of emerging market spreads are the same for sovereigns and other borrowers—that is, by estimating the model on the full sample but adding interaction terms for sovereign status and governing laws.41 The interaction terms then tell us whether there are significantly different effects for sovereigns than for other borrowers. It turns out that this is not the case: none of the interaction terms enters with a coefficient that differs significantly from zero, while the dummy for U.K. governing law continues to enter positively for borrowers with low credit ratings and negatively for borrowers with high credit ratings. It would appear that the market's pricing of these provisions is no different for sovereigns than for other borrowers (reported in Eichengreen and Mody, 2000d). Again, the results are robust to a number of additional checks.42 For example, to check that we are picking up the characteristics of the bond rather than the market in which it is issued, we added interest rates in additional financial centers (London and Tokyo) to our spreads equation.43 There is a slight reduction in the size of the positive coefficient on U.K. law for borrowers with high credit ratings but no discernible change in the point estimate for less creditworthy borrowers, and there are no changes in statistical significance. A further check is whether IMF programs affect the connection between governing laws and spreads. Recall that above we found that Fund programs had the strongest tendency to enhance creditworthiness, presumably by buttressing the borrower's commitment to reform, in countries with credit ratings in the 30-50 range. If IMF programs serve as commitment-enhancing mechanisms for these countries, then we should expect those that opt for U.K. law (which would otherwise have to trade commitment for ability to restructure) to now suffer less of a penalty in terms of higher spreads. This is what

40This is true especially for sovereign borrowers with relatively high credit ratings, where we have a very limited number of observations. 41Note, as before, that the governing law variables may also be interacted with credit rating category. 42In addition to the sensitivity analysis discussed in the text, we estimated the model using Heckman's two-stage procedure rather than maximum likelihood to test the sensitivity of the results to the specification of the selectivity correction, because specification errors affecting one equation are more likely to contaminate the other equation when the model is estimated by maximum likelihood. We dropped the variables with insignificant coefficients from the first-stage probit to further test the sensitivity of the results to implementation of the selectivity correction. We substituted the raw credit rating for the credit rating residual in the spreads equation, on the grounds that any misspecification in the equation we estimate to derive the credit rating residual would contaminate the other results. We entered the explanatory variables in levels rather than logs. We respecified the dependent variable in the spread as a proportion of the riskless rate. We eliminated influential observations, such as Panama, which has a low credit rating but apparently enjoys a "halo" effect as a result of its special relationship to the United States, and countries that had undergone Brady Plan restructurings. In all cases the results for the effect of choice of governing law were basically unchanged. 43Recall that U.S. interest rates were already included. Given its high correlation with U.S. rates, we entered the variable as the difference between U.S. and foreign rates.

177

Table 5. Impact of Governing Laws by Credit Rating Category Log of Spread at the Time of Issue All issuers Sovereign issuers Credit rating Credit rating Credit rating 30 or more 50 or more 30 or more Credit rating but less but less Credit rating but less Credit rating less than 30 than 50 than 70 70 or more than 50 50 or more (1) (2) (3) (4) (5) (6)

Log amount -0.069 -0.050 -0.066 -0.114 0.005 0.084 (-1.04) (-2.42) (-1.68) (-1.46) (0.17) (0.61)

Maturity 0.015 0.004 0.003 -0.025 0.006 0.021 (0.98) (1.39) (0.726) (-1.25) (1.97) (1.40)

Private placement -0,019 0.046 -0.063 -0.118 -0.035 -0.300 (-0.20) (1.49) (-0.99) (-0.83) (-0.66) (-1.19)

Log of 10-year -0.839 -0.063 -0.115 0.560 -0.370 0.306 U.S. treasury rate (-1.99) (-0.40) (-0.38) (0.43) (-1.69) (0.33)

Log (10-year-l-year) 0.093 0.036 0.057 -0.761 0.008 -0.124 U.S. treasury rate (0.82) (1.26) (0.875) (-2.15) (0.19) (-1.13)

Log swap rate 0.426 0.561 0.929 0.681 0.511 0,186 (1.89) (7.82) (5.58) (1.04) (5.84) (0.59)

Credit rating residual -0.016 -0.027 -0.093 -0.092 -0.036 -0.082 (-1.10) (-7.66) (-13.86) (-1.32) (-8.97) (-3.41)

Debt/GNP -0.118 0.880 1.194 1.121 0.683 0.482 (-0.32) (6.30) (6.91) (0.35) (4.21) (0.65)

Debt rescheduled 0.174 0.064 -0.048 in previous year (1.35) (1.35) (-0.69)

GDP growth -18.471 -4.871 -25.809 -165.67 -7.571 -9.650 (-3.79) (-2.43) (-4.93) (-2.10) (-3.32) (-0.67)

Standard deviation -0.248 1.081 3.970 6.460 0.626 6.451 of export growth (-0.36) (5.07) (6.17) (1.61) (2.80) (6.06)

Reserves/short-term -0.045 -0.121 -0.027 0.142 -0.111 debt (-1.74) (-7.12) (-1.11) (0.61) (-6.02)

Ratio of domestic -0.164 -0.153 -0.096 -0.402 -0.059 credit to GDP (-3.57) (-6.58) (-3.60) (-1.07) (-1.72)

Stand-by arrangement 0.099 -0.012 0.282 0.034 (0.54) (-0.19) (1.67) (0.41)

Extended Fund Facility -0.048 -0.058 -0.049 (-0.24) (-0.91) (-0.59)

Enhanced Structural 0.434 0.077 0.150 Adjustment Facility (1.15) (0.20) (0.44)

178

Tables, (concluded) Log of Spread at the Time of Issue All issuers Severe!;gn issuers Credit rating Credit rating Credit rating 30 or more 50 or more 30 or more Credit rating but less but less Credit rating but less Credit rating less than 30 than 50 than 70 70 or more than 50 50 or more (1) (2) (3) (4) (5) (6) Number of quarters in 0.008 0.006 0.018 0.004 an IMF program (0.66) (2.70) (2.15) (1.47)

U.K. governing law 1.054 0.602 -0.829 -1.667 0.520 -0.937 (1.98) (3.70) (-2.99) (-1.98) (2.74) (-0.83)

Other governing laws -0.039 -0.613 -1.415 -0.526 0.066 0.307 (-0.04) (-2.70) (-3.72) (-0.55) (0.30) (0.47)

U.K. governing lawx -0.105 -0.296 -0.120 IMF program (-0.36) (-2.56) (-1.39)

Other governing lawsx 0.127 0.083 0.077 IMF program (0.28) (0.85) (0.73)

Foreign guarantee -1.080 -0.456 -0.264 0.587 -0.516 (-3.39) (-5.09) (-2.57) (1.06) (-3.25)

Domestic guarantee -0.086 -0.094 0.167 0.107 -0.057 (-0.82) (-2.14) (2.23) (0.56) (-0.26)

Constant 4.680 3,446 4.048 7.032 4.075 2.783 (3.48) (6.66) (4.30) (1.35) (6.44) (1.22)

Lambda -0,007 -0.418 -0.659 -0.070 -0.349 0.159 (-0.11) (-16.03) (-15.39) (-0.37) (-6.41) (0.68)

Number of bonds 275 1,245 588 273 431 43

Log of likelihood -506.80 -1,996.44 -1,067,49 -387.45 -770.88 -118.55

Notes: Dummy variables for private and public issuers, industrial sectors, currency of issue, and whether the bond was issued at a fixed or floating rate were included in toe regressions but are not reported here. Figures in parentheses are /-statistics. we find when we interact the existence of Fund programs with U.K. governing law. The story is different, in plausible ways, for countries within the 0-30 range. Recall that Fund programs do less to enhance the commitment of these countries to meet their commercial obligations. It follows that we should find less of a change in the effect of governing law on spreads when we interact IMF programs with choice of law. This is what we find. Again, it is hard to imagine an explanation for these patterns grounded in the technical characteristics of the relevant markets, while it is straightforward to interpret them in terms of the ability of alternative contractual provisions to facilitate bonding and restructuring. Before concluding, we can also present a little bit of evidence on proposals to use official enhancements and guarantees to encourage the private sector to

179

Figure 1. Spreads Over Risk-Free Rate, Credit Ratings, and Governing Law

extend new credits in periods of shaky investor confidence.44 Among the objections to this idea, as noted above, is that guarantees may do more to signal problems than to reassure investors. Table 5 shows that guarantees reduce spreads on issues from low-rated countries (0-30 on the Institutional Investor scale), as if the bad news about these countries is already known and the main effect of guarantees is to reduce risk. This is the same group of countries, it will be recalled, for which IMF programs have no discernible impact on spreads. Thus, it would appear that guarantees can do what IMF programs cannot in reducing risk for investors in these sub-investment-grade issues.45 The impact on spreads is of the same order of magnitude as the premium paid by U.K. law issuers relative to U.S. law issuers, as if the guarantee offsets the risk to investors resulting from any additional borrower moral hazard under this contractual arrangement. The magnitude of the effect declines as one moves up the credit-rating scale and even turns positive, in some cases, for borrowers with good credit. For these borrowers, if not others, guarantees may do more to signal problems than to solve them.

•^This evidence is indirect in that private guarantors provide the majority of guarantees in the sample, although we also have nine multilateral guarantees (five of which are for a series of issues by the Republic of Argentina). 45Whether this is efficiency enhancing or merely redistribution to creditors is a separate question, as explained in Section II.

"180

V. Conclusion

This paper has considered the impact of IMF involvement and of some alternative approaches to crisis management and resolution on the cost of borrowing for emerging markets. While the official community has held out hope that both Fund programs and these alternatives can reduce the cost of borrowing and improve the availability of private credit, both have been accused of doing more to damage than enhance market access. How can sensible people subscribe to such different views? The answer, in part, is that different borrowers are affected differently by these interventions. Where one stands on these issues consequently depends on where one sits. A further implication is that, in discussions of new approaches to governing capital flows and managing crises, it is not likely to be helpful to speak of the impact on emerging markets as a class. Table 6 summarizes our findings and indicates how they lead us to think about the world. For countries with poor credit, including many whose institutions of financial and economic governance remain underdeveloped, there is only limited evidence that IMF programs enhance private market access. In many of these countries, it would appear, the commitments entailed by IMF conditionality are simply not credible. The same is true of provisions in private contracts that imply a strong commitment to service debts on an inflexible schedule (in other words, we do not see these countries borrowing under U.S. law). If the official community is serious about enhancing the market access of these countries in the short run, then there may be no alternative to official guarantees as a mechanism to reassure investors. More generally, the role of the multilaterals in these countries should be thought of as supporting the kind of structural and developmental changes that can only be accomplished in the long run, not as somehow using their resources to immediately catalyze private finance.46 For countries with a somewhat stronger capacity for reform, both IMF programs and relatively inflexible private contracts are ways for borrowers to send a credible signal of commitment to the market. Fund programs can play something of a catalytic role, or so it would appear, which supports the case for IMF lending in times of crisis. Countries with superior credit ratings have a greater capacity to commit without outside intervention. Multilateral intervention—whether it takes the form of IMF programs or official guarantees—runs the risk of raising questions about that capacity. It is no surprise that we see such countries hesitating to approach the Fund. For such countries, a preferable solution to financial difficulties may be to approach their creditors on a bilateral basis, assuming that they have in place the institutional mechanisms necessary for negotiations to succeed. Our findings also may have implications for the design of Fund programs, although here we move beyond what is explicit in our results. There is a suggestion that the most effective Fund programs are conditioned on a combination of macro-

46There may also be a role for debt reduction for such countries, but this is not the subject of the paper, nor is it something on which we have evidence.

181

Table 6. Broader Implications of the Results Credit Role of Multilateral Bailout Bail-In Guarantee

Poor Little ability to exercise Only weak evidence Measures that ease Strong effect. catalytic effects. Role that EFF reduces rescheduling raise of multilaterals mainly borrowing costs. borrowing costs. long-term and IMF presence does developmental. little to mitigate these costs.

Middling Stronger ability to play EFF reduces Measures that ease Modest effect. catalytic role. Crisis borrowing costs. rescheduling again intervention may have raise borrowing costs, significant effects on but to a lesser extent terms of private market when IMF is present. access.

Good While short-term, IMF intervention Measures that ease Possibly perverse targeted crisis signals trouble and rescheduling do not effect. management can be appears to raise raise borrowing costs. useful, execution may borrowing costs. IMF role is superfluous. be problematic. Own commitment and voluntary restructuring work best.

economic adjustments and limited structural reforms. Programs of this character do more to enhance investor confidence and improve the crisis country's terms of market access than programs that pay little heed to structural problems, but they also have more positive effects than programs conditioned on very extensive structural reforms. The first comparison is consistent with the view that policy reforms beyond the narrowly macroeconomic are essential for credibility and stability in today's financially integrated world and that the Fund ignores them at its peril. At the same time, demanding deep, ambitious structural reforms at the height of a crisis is not likely to be productive. Conditionality needs to strike a balance.

182

APPENDIX

Table Al. Country Classification by Credit Rating Category

1991 1994 199? 1999

Angola Algeria Algeria Algeria Argentina Angola Angola Angola Bangladesh Bangladesh Bangladesh Bangladesh Bolivia Bolivia Bolivia Bolivia Brazil Brazil Bulgaria Bulgaria Bulgaria Bulgaria Croatia Dominican Republic Costa Rica Costa Rica Dominican Republic Ecuador Dominican Republic Croatia Ecuador El Salvador Ecuador Dominican Republic El Salvador Ethiopia Egypt, Arab Republic of Ecuador Ethiopia Ghana El Salvador Egypt, Arab Republic of Ghana Guatemala Ethiopia El Salvador Guatemala Iran, Islamic Republic of Guatemala Estonia Iran, Islamic Republic of Jamaica Indonesia Ethiopia Jamaica Kazakhstan Iran, Islamic Republic of Ghana Kazakhstan Kenya Jamaica Guatemala Kenya Liberia Kenya Iran, Islamic Republic of Latvia Nigeria Lebanon Jamaica Lebanon Pakistan Liberia Kazakhstan Liberia Romania Morocco Kenya Lithuania Russian Federation Nigeria Lebanon Nigeria Senegal Pakistan Liberia Pakistan Seychelles Panama Lithuania Panama Ukraine Paraguay Nigeria Russian Federation Vietnam Peru Pakistan Senegal Zambia Philippines Panama Seychelles Zimbabwe Poland Paraguay Ukraine Romania Peru Zambia Senegal Philippines Seychelles Poland Sri Lanka Romania Trinidad and Tobago Russian Federation Uruguay Senegal Zambia Seychelles Zimbabwe Slovak Republic Slovenia Sri Lanka Trinidad and Tobago Ukraine Vietnam Zambia Zimbabwe

183

Table AT. (continued) 1991 1994 1997 1999

Rating Between 30 and 50

Argentina Algeria Argentina Argentina Barbados Bahrain Barbados Bahrain Brazil Barbados Brazil Barbados Bulgaria Bulgaria Colombia Brazil Chile Costa Rica Costa Rica Colombia Colombia Czech Republic Croatia Costa Rica Czech Republic Egypt, Arab Republic of Egypt, Arab Republic of Croatia Former Czechoslovakia Hungary Estonia Dominican Republic Hungary India Ghana Ecuador India Kuwait Hungary Egypt, Arab Republic of Indonesia Mauritius India El Salvador Kenya Mexico Latvia Estonia Kuwait Morocco Lebanon Ghana Mauritius Papua New Guinea Lithuania India Mexico Paraguay Mexico Indonesia Oman Philippines Morocco Latvia Pakistan Poland Panama Lebanon Papua New Guinea Slovak Republic Papua New Guinea Lithuania Qatar Slovenia Paraguay Mexico Romania South Africa Peru Morocco Russian Federation Sri Lanka Philippines Panama South Africa Trinidad and Tobago Poland Papua New Guinea Trinidad and Tobago Tunisia Romania Paraguay Tunisia Turkey Slovak Republic Peru Turkey Uruguay Slovenia Philippines Uruguay Venezuela South Africa Romania Venezuela Sri Lanka Russian Federation Trinidad and Tobago Slovak Republic Tunisia South Africa Turkey Sri Lanka Uruguay Thailand Venezuela Trinidad and Tobago Vietnam Tunisia Zimbabwe Turkey Uruguay Venezuela Vietnam Zimbabwe

184

Table Al. (concluded)

1991 1994 1997 1999

Rating Between 50 and 70

Bahrain Bahrain Chile Bahrain China Chile China Chile Czech Republic China Czech Republic China Former Czechoslovakia Czech Republic Hong Kong SAR" Czech Republic Hong Kong Hong Kong Indonesia Hong Kong SAR Indonesia Indonesia Korea, Republic of Hungary Korea, Republic of Korea, Republic of Kuwait Korea, Republic of Kuwait Kuwait Malaysia Kuwait Malaysia Malaysia Mauritius Malaysia Oman Oman Oman Oman Qatar Qatar Poland Poland Russian Federation Saudi Arabia Qatar Qatar Saudi Arabia Thailand Saudi Arabia Saudi Arabia Thailand United Arab Emirates Slovenia Slovenia United Arab Emirates Thailand United Arab Emirates Tunisia

1991 1994 1997 1999

Rating More than 70

Singapore Korea, Republic of Korea, Republic of Singapore Taiwan Province of China Singapore Singapore Taiwan Province of China Taiwan Province of China Taiwan Province of China

Source: Institutional Investor. "The territory was referred to as Hong Kong before July 1, 1997.

REFERENCES Becker, Torbjorn, Anthony J. Richards, and Yunong Thaicharoen. 2000, "Collective Action Clauses for Emerging Market Bonds: Good News for Lower Rated Borrowers Too." IMF Seminar No. 2000-59 (Washington: International Monetary Fund). Bird, Graham, and Dane Rowlands, 1997, "The Catalytic Effect of Lending by the International Financial Institutions," World Economy, Vol. 20 (November), pp. 967-91. Buiter, Willem H., and Anne C. Sibert, 1999, "UDROP: A Small Contribution to the New International Financial Architecture," CEPR Discussion Paper No. 2138 (London: Centre for Economic Policy Research). Corrigan, E. Gerald, 2000, "Two International Financial Stability Issues: Asset Price Inflation and Private Sector Participation in Financial Crisis Stabilisation," Financial Stability Review, Bank of England, Issue 8 (June), pp. 136-41. Available via the Internet: http://www.bankofengland.co.uk/fsr/articles.htmttissue08 Dallara, Charles, 1999, "Letter to the Chairman of the Interim Committee." September 15. Available via the Internet: http://www.iif.com/data/public/icdc099.pdf.

185

Deutsche Bundesbank, 1999, "Recent Approaches to Involving the Private Sector in the Resolution of International Debt Crises." Monthly Report (December), pp. 33-48. Dixon, Liz, and David Wall, 2000, "Collective Action Problems and Collective Action Clauses," Financial Stability Review, Bank of England, Issue 8 (June), pp. 142-51. Available via the Internet: http://www.bankofengland.co.uk/fsr/articles.htmttissue08. Dooley, Michael, 1997, "A Model of Crises in Emerging Markets," NBER Working Paper No. 6300 (Cambridge, Massachusetts: National Bureau of Economic Research). , 2000, "Can Output Losses Following International Financial Crises Be Avoided?" NBER Working Paper No. 7531 (Cambridge, Massachusetts: National Bureau of Economic Research). Drage, J., and F. Mann, 1999, "Improving the Stability of the International Financial Sector," Financial Stability Review (June), pp. 40-77. Economist, 1999, "Sovereign Policy," The Economist, February 13, p. 21. Eichengreen, Barry, and Ashoka Mody, 2000a, "Lending Booms, Reserves, and the Sustainability of Short-Term Debt: Inferences from the Pricing of Syndicated Bank Loans," Journal of Development Economics, Vol. 63, pp. 5-44. , 2000b, "What Explains the Changing Spreads on Emerging-Market Debt? Fundamentals or Market Sentiment?" in Capital Flows and the Emerging Economies: Theory, Evidence, and Controversies, ed. by Sebastian Edwards (Chicago: University of Chicago Press). , 2000c, "Would Collective Action Clauses Raise Borrowing Costs?" NBER Working Paper No. 7458 (Cambridge, Massachusetts: National Bureau of Economic Research). 2000d, "Would Collective Action Clauses Raise Borrowing Costs? An Update and Additional Results," Policy Research Working Paper No. 2363 (Washington: World Bank). Eichengreen, Barry, Galina Hale, and Ashoka Mody, 2001, "Flight to Quality: Investor Risk Tolerance and the Spread of Emerging Market Crises," in International Financial Contagion: How It Spreads and How It Can Be Stopped, ed. by Stijn Claessens and Kristin Forbes (Boston, Massachusetts: Kluwer). Eichengreen, Barry, and Richard Fortes, 1995, Crisis? What Crisis? Orderly Workouts for Sovereign Debtors (London: Centre for Economic Policy Research). Faini, Ricardo, Jaime de Melo, Abdel Senhadji-Semlali, and Julie Stanton, 1991, "Macro Performance Under Adjustment Lending." in Restructuring Economies in Distress: Policy Reform and the World Bank, ed. by V. Thomas, A. Chibber. M. Dailami, and J. de Melo (New York: Oxford University Press). Fischer, Stanley, 1999, "Learning the Lessons of Financial Crises: The Roles of the Public and Private Sectors, speech to the Emerging Market Traders' Association Annual Meeting, New York, December. Available via the Internet: http://www.imf.org/externaVnp/speeches/1999/120999.htm. Frankel, Jeffrey, and Nouriel Roubini, 2000, "The Role of Industrial Country Policies in Emerging Market Crises," paper prepared for the NBER Conference on Economic and Financial Crises in Emerging Market Economies. Woodstock, Vermont, September. Friedman, Benjamin, 2000, "How Easy Should Debt Restructuring Be?" in Managing Financial and Corporate Distress: Lessons from Asia, ed. by Charles Adams, Robert Litan, and Michael Pomerleano (Washington: Brookings Institution Press). Goldstein, Morris, 2000, "IMF Structural Programs" (unpublished: Washington: Institute for International Economics). Griffith-Jones, Stephany. Jose Antonio Ocampo, and Jacques Cailloux, 1999, "The Poorest Countries and the Emerging International Financial Architecture" (unpublished; Swedish Ministry of Foreign Affairs). Group of Seven, 1999, Strengthening the International Financial Architecture (Washington: Group of Seven).

186

Group of Ten, 1996, Resolving Sovereign Liquidity Crises (London: Group of Ten). Group of Twenty-Two, 1998. Report of the Working Group on International Financial Crises (Washington: Group of Twenty-Two). Hajivassiliou, Vassilis, 1987, "'External Debt Repayments Problems of LDCs: An Econometric Model Based on Panel Data," Journal of Econometrics, Vol. 36 (September/October), pp. 205-30. Institute of International Finance, 1996, Resolving Sovereign Financial Crises (Washington: IIP). , 1999a, "Global Private Finance Leaders Stress the Importance of Voluntary Approaches to Crisis Resolution in Emerging Markets," June 24. 1999b, "Summary Report on the Work of the IIF Steering Committee on Emerging Markets Finance" (Washington: OF). International Monetary Fund, 1999, Involving the Private Sector in Forestalling and Resolving Financial Crises (Washington: IMF). Kahn, Robert, 2000, "The Role of the Private Sector in the Prevention and Resolution of International Financial Crises," presentation to the Conference on the Governance of the Global Capital Market. Montreal, October 23. Killick, Tony. 1995, IMF Programmes in Developing Countries: Design and Impact (London: Routledge/Overseas Development Institute). McKinnon, Ronald, and Huw Pill, 1997, "Credible Economic Liberalizations and Overborrow- ing," American Economic Review, Papers and Proceedings, Vol. 87 (May), pp. 189-93. Miller, Marcus, and Joseph Stiglitz, 1999, "Bankruptcy Protection Against Macroeconomic Shocks: The Case for a 'Super Chapter 11"" (unpublished; Washington: World Bank). Miller, Marcus, and Lei Zhang, 2000, "Sovereign Liquidity Crisis: The Strategic Case for Payments Standstill," Economic Journal, Vol. 110, No. 460, pp. 335-62. Murray, John, 2000, "The Role of the Private Sector in the Prevention and Resolution of International Financial Crises," presentation to the Conference on the Governance of the Global Capital Market, Montreal, October 23. Ozler, Sule, 1993, "Have Commercial Banks Ignored History?" American Economic Review, Vol. 83 (June), pp. 608-20. Roubini, Nouriel, 2000, "Bail-In, Burden Sharing, and Private Sector Involvement in Crisis Resolution" (unpublished; New York: Stern School of Business, New York University). Rowlands, Dane, 1994, "The Response of New Lending to the IMF," Developing Studies Working Paper No. 7 (Ottawa: Norman Paterson School of International Affairs). Sachs, Jeffrey, 1994, "Do We Need an International Lender of Last Resort?" (unpublished; Cambridge, Massachusetts: Harvard University). Summers, Lawrence, 1996, "The Right Kind of IMF for a Stable Global Financial System," text as prepared for delivery to the London School of Business, December 14. Available via the Internet: http://www.ustreas.gov/press/releases/ps294.htm. Tsatsaronis, Costas, 1999, "The Effects of Collective Action Clauses on Sovereign Bond Spreads," BIS Quarterly Review (November), pp. 22-23. Williamson, John, 1992, "International Monetary Reform and the Prospects for Economic Development," in Fragile Finance: Rethinking the International Monetary System, ed. by Jan Joost Teunissen (The Hague: FONDAD). Yanni, A., 1999, "Resolution of Sovereign Financial Crisis-Evolution of the Private Sector Restructuring Process," Financial Stability Review (June), pp. 78-84.

187

Crisis Resolution and Private Sector Adaptation

GABRIELLE LIPWORTH and JENS NYSTEDT*

Efforts at crisis resolution that succeed in reducing potential inefficiencies and instability in the international financial system are in the interest of both the private and the public sectors. This paper focuses on how the private sector is likely to adapt to the recent initiatives by the official sector to further involve the private sector in the resolution of crises. The key conclusion is that recent experiences with payment suspensions and bond restructurings are limited as guides to determining the future success or failures of these initiatives, because the private sector most likely has adapted in order to minimize any unwanted involvement, f JEL F34]

I rivate sector involvement (PSI) has been an integral part of all crisis resolu- p tion efforts, and it is not new. At the time of the resolution of the Latin American debt crisis of the 1980s, for example, the official community had many of the same objectives it has today: limiting the size of official packages, reducing moral hazard in the private sector's lending decisions, and restoring the external viability of the country in crisis. Some academics (e.g., Dooley, 1994) see the lending preceding the debt crisis of the 1980s as raising charges of moral hazard. By the late 1980s, however, the improved financial positions of major international banks (as measured by their developing country loan exposures relative to their capital) and the continuing poor economic performance of many emerging markets led to the adoption of the Brady plan, which involved substantial write- downs (measured in net present value terms) of developing country syndicated

*Both authors belong to the Global Markets Division of the IMF's International Capital Markets Department. The authors would like to express their thanks to Torbjorn Becker, Bankim Chadha, Olivier Jeanne, Andrew Haldane. Arend Kapteyn, Ma/.iar Minovi, and an anonymous referee for helpful comments and suggestions.

188

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION loans. Indeed, the losses experienced by banks on medium-term syndicated lending to developing countries in the 1980s are regarded as a key factor in the decision of banks to shift away from syndicated lending to sovereigns toward shorter-term interbank lending in the 1990s. Large official financing packages in the 1990s, starting with Mexico (1994-95) and then in Asia (1997), were also seen by many observers as increasing the private sector's expectation of being rescued should it be confronted by an imminent credit event. This sentiment likely peaked in the run-up to the Russian default in August 1998 because Russia was widely viewed as "too big or too nuclear to fail" and would therefore receive the support of the official community no matter what. Others have seen the crises of the 1980s and 1990s as arising out of a much more complex set of macroeconomic and financial factors and have argued that there needs to be a more nuanced view of the extent and potential sources of moral hazard. It has also been argued that moral hazard in the international financial system can potentially arise from a number of sources, including the official safety net that underpins all banking systems and the lending activities of international financial institutions (IFIs). The official safety net underpinning the banking system is typically designed to ensure the overall stability of the domestic financial system and to protect the domestic payments system. It is widely recognized that the knowledge that a bank is "too big to fail" can lessen the incentives to impose both market and manage- rial discipline. Domestic bank bail-outs costing the sovereign the equivalent of 10-20 percent of GDP have not been uncommon, and they clearly have an impact on the expectations for future bail-outs by the domestic banks as well as the expectations of international banks providing financing to domestic banks. While the moral hazard effects of the official safety net underpinning national banking systems are a constant feature of the global financial system, the potential moral hazard effects of lending by IFIs will be influenced by both the scale and the timing of such lending. As noted above, market participants regard such lending as having had its most significant effect on creditors' expectations during the run-up to the Russian default in August 1998. Nonetheless, there remains considerable disagreement between those who see lending by IFIs as having a "first-order" effect in creating moral hazard and those who view such lending as having a much smaller and episodic effect (Lane and Phillips, 2000). Whatever the conclusion on the likely significance of moral hazard arising from official international support to countries facing external financing difficulties, the fact is that the scale of such support is limited. When there is a meaningful risk that a country may be insolvent and therefore incapable of timely repayment of emergency official assistance, the official community typically refrains from providing such assistance except on the condition that other claims against the country be rescheduled and written down to an extent that ensures that emergency official assistance can be repaid. These are situations where, like it or not, the creditors of a country's debtors (its sovereign, its banking system, or its private sector) will unavoidably be "involved" in the resolution of the country's financial difficulties. More broadly, when a country faces a huge outflow of capital that threatens to swamp that country's own resources plus any plausible level of emergency assis-

189

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt tance from the official community, and when efforts to resolve the crisis through policy adjustments, limited official assistance, and a spontaneous restoration of confidence fail, the creditors of that country will also face "involvement" in the resolution of that country's financial difficulties on terms and conditions not contemplated in their credit instruments. In these situations, private sector involvement in crisis resolution is, and always has been, a fact of life. In designing and implementing policies on private sector involvement, the official sector has—and is perceived in private markets to have—several, not necessarily consistent, objectives. One is burden sharing. Because of concerns about moral hazard and for other reasons, the official community wants to keep its emergency support limited. It also wants to ensure that private creditors play—and are seen to play—an appropriate role in resolving crises. When losses need to be absorbed—especially in situations of insolvency—the official sector wants to ensure that private creditors do not escape by imposing losses they should bear onto others. A second broad objective is limiting the damage done by the crisis, both to the country primarily involved and to the world economy more generally. Sometimes, especially in cases of insolvency, this may again mean that creditors should absorb losses (also part of burden sharing). It also means, especially in cases of illiquidity, seeking to restore external viability and market access as rapidly as possible following the resolution of a crisis—something that may not be facilitated by efforts to impose substantial losses on creditors. The third broad objective of the official community is to preserve integrity and reasonable efficiency in the functioning of international credit markets. This means that debtors should not be allowed to escape from servicing their obligations when they have the capacity to do so. It also means that creditors who undertake risks should expect to see those risks sometimes materialize into actual losses. These policies also interact dynamically, as the private sector reacts to the policies of the official sector for future financing flows and the official sector, in turn, adapts its policies. Debt restructurings are repeated games, and while addressing the current crisis the official sector is already affecting the conditions of the next debt crisis. This phenomenon is clearly apparent in the evolution of international credit arrangements over the past two decades. Medium-term loans from large syndicates of commercial banks to developing country sovereigns and public sector entities were a dominant form of international capital flows before the debt crisis of the 1980s. An important part of the mechanism that the official sector used to deal with that crisis involved the concerted rollover and subsequent restructuring and write-down (in present value terms) of syndicated bank loans. Bonded debts of affected sovereigns generally escaped restructuring on the grounds that the amounts were small and that these instruments (held by widely diversified creditors) were difficult to restructure. The market adapted. Medium- term syndicated bank loans to developing country sovereigns largely disappeared in the 1990s. Banks shifted to interbank loans of much shorter maturity. International borrowing by sovereigns predominantly took the form of bonded debts. The shifts in the form of international credit flows posed new challenges in efforts to resolve the financial crises of the 1990s. Lenders to emerging markets

190

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION were either thousands of individual bondholders whose actions were difficult to concert or banks with short-term facilities that could easily "cut and run" in a crisis. Mechanisms for private sector involvement in the crises of the 1990s have adapted to these new realities. The fact that the private sector will adapt, taking losses or gains on existing debt while changing the level or structure of its future lending, to the official sectors' policies and practices with respect to private sector involvement is not necessarily negative. For example, although unwelcome to potential debtors, policies that raise the cost and diminish the availability of international credits to some emerging market borrowers may be desirable if they reflect a more appropriate pricing of risks and serve properly as a deterrent to imprudent borrowing, that is, reduce debtor moral hazard. Policies that encourage longer-term securitized borrowing (which is presumably limited by available collateral) may contribute to the avoidance of more efficient resolution of crises because such loans are hard to restructure. Longer-term loans are likely to be less dangerous in a potential crisis than an equivalent volume of short-term loans, and creditors who believe they have secure collateral should be less prone to panic than those that do not. On the other hand, a country that has already encumbered most of its liquid assets and a good deal of its future export earnings may find itself in a very difficult situation in the event of a financial crisis. The point is that in considering various policies and practices with respect to private sector involvement, it is critical to be aware of how the private sector is likely to adapt to these policies and practices and to the difficulties or opportunities that these reactions will generate. The analysis in this paper is based on some fairly simple observations and assumptions of the behavior of creditors and debtors in the international capital markets. It takes as given that debtor moral hazard is a key concern for a creditor's lending decision, and the analysis presented here tries to fill a gap in earlier papers' focus solely on creditor moral hazard. Section I of this paper presents a framework that examines how a sovereign debtor and a private sector creditor value debt and therefore how they are likely to be affected when key variables change, such as the cost of default and the recovery value, as they may have done in the context of the recent official sector crisis resolution initiatives. Two forms of impact are discussed: first how future lending will be affected as the debtor and creditor adjust to a new equilibrium, and second how already existing debt will be affected in their price. Section II discusses how this framework can be used to analyze how private creditors will adapt to recent official community initiatives with regard to debt payment suspensions. Section III addresses how the private sector is likely to adapt to the recent string of successful bond exchanges. It is argued that the conclusion that bonds will be as easy to restructure in the future as they have been recently is premature, because it does not take into account the fact that private sector creditors will adjust their expectations and lending instruments going forward. Section IV ends the paper with some broad conclusions and raises the question whether the private sector is already adapting to the recent official sector crisis resolution initiatives.

191

I. Determinants of Debt Restructuring: A Model

Consider the simplest two-period sovereign borrowing model. A sovereign debtor has the choice to borrow an amount, D, from an international lender, in the form of either bonds or loans, and invests in a risky project with an uncertain return of rp. If rp exceeds the coupon, c, which is what the sovereign is obligated to pay on the debt, the sovereign can service his debt and avoid default. However, if the return on the risky project is less than the required debt service, the sovereign has to enter a default, which is costly to both the debtor and the creditor, and in default the creditor instead receives only the recovery value RV instead of c. The recovery value is commonly defined as the expected present value of the future debt service that the instrument in default is expected to generate for the creditor, for example, through the issuance of an exchange bond at some point in the future with a reprofiled or lower debt service schedule. Clearly, the recovery value following default will be affected not only by the debt servicing capacity of the debtor but also by the respective bargaining strength of the creditors of a particular instrument vis-a-vis the debtor as well as other creditors. The recovery value in default will vary from one debt instrument to another. Some debt instruments are collateralized with foreign assets and designed in such a way that should default occur, the recovery value is very high (an example would be bonds collateralized by future export revenue flows).

The Sovereign Debtor If strategic default is allowed for, the debtor's decision function becomes slightly more complex. Now, even if the sovereign does have the money to repay, that is, the "ability to pay," he will only repay, that is, "willingness to pay," if by doing so he can avoid an excessively costly default. In other words, the net benefit of servicing his debt has to be larger than the benefit from a strategic default. If this is not the case, no lender will ever extend credit to a sovereign, because default is a certainty. In a model, one can express the sovereign's decision as, given that rp > c, the sovereign will service his debt, FI = 1, if

Therefore, a strategic default will be avoided as long as the cost, A,, which is instrument-specific, together with the recovery value, is larger than the interest rate on the loan, that is,

(1)

Note that the suggested description in (1) breaks up the cost of default for a debtor in two parts: first the recovery value, which directly benefits the creditor, and second A,, which is a pure loss and benefits no one in particular. Many different motivations and specifications have been proposed with regard to A, (see Cline, 2000, or Dooley, 2000, for an overview). The costs of default could include lower reputation, loss of market access for both trade and long-term financing, the

192

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION potential to become a target of costly litigation, and output losses while in default because the debtor lacks foreign capital for investments. The output losses will naturally be greater the longer the country is in default, and this time period is under the control of the creditor and the debtor when the original loan contract is drawn up. For the creditor the importance of the ease of renegotiating the contract is a function of the value of other disciplining devices as well as whether the cost of renegotiating the contract is materially different from the cost of renegotiating other contracts. That is, when there is more than one debt contract, the debtor will default on the least costly debt first.1 In a world where strategic default is possible but there is an associated cost to default, the utility function of a sovereign borrower is denoted Us and the probability of a good outcome and hence the ability to pay is denoted p:

(2)

Unfortunately, in the real world as well as in many theoretical models (see Dooley, 2000), often a lender can observe only the default itself and not really the reasons behind the decision. Because in some states of the world the outcome of the investment may be low enough to actually infringe on the recovery value for the creditor,2 it is in the creditor's interest to distinguish between "bad-luck" defaults, where preferably the default cost is zero to maintain as high a recovery value as possible, and strategic defaults, where the default cost should be very high to discourage them. Because a distinction is not possible, by assumption, a creditor would have to impose a default cost by, for example, making the debt contract hard to restructure, leading to large output losses, irrespective of whether the default occurred as a result of the inability to pay or the unwillingness to pay.3

The Creditor For any risk-neutral creditor, any debt contract, irrespective of whether it is a loan or a bond, is valued ex ante, depending on (1) the probability of default by the debtor on that contract; (2) the recovery value if the default occurs; and (3) the maturity, principal, and coupon paid on the debt if the debtor does not default. This can be expressed, in a two-period model, as

'See Nystedt (2001) for a more elaborate model involving two types of creditors extending credit through two different instruments: loans with a low cost of default and bonds with a high cost of default. A clear implication of the model is that if the cost of default is suddenly reduced for bonds, the borrower will shift to borrow relatively more through loans in the new equilibrium. 2 This assumes that in a bad-luck default rp - A - RV > 0 and hence RV = rf - A, which may be equal to zero if A is large enough and the outcome bad enough. 3Other authors have suggested that the IMF, or some other multilateral institution, could play the role of an outside monitor that could distinguish between bad-luck defaults and strategic defaults. The IMF could signal what type of default had occurred by lending-into-arrears and thereby, at least partly, offset the cost of default for the debtor, in that state of the world, and lead to a higher recovery value for the creditor ex post, leading to a lower coupon ex ante.

190

(3)

Here, r/ simply reflects that on a risk-adjusted basis a creditor, through a no arbitrage argument, expects to earn a yield of at least the risk-free rate. However, when (2) and (3) are compared, the interrelationship between the two expressions is through the coupon rate. If the coupon is high, the probability of nonpayment will increase and the creditor will have to charge an even higher coupon up front to compensate. Given some assumption about the probability distribution, an equilibrium coupon and probability of payment can easily be derived, but the resulting coupon has to be compatible with the strategic default constraint presented in equation (1) to represent a true achievable equilibrium.

The Equilibrium Coupon

To derive the equilibrium coupon rate, it is assumed that rp is uniformly distributed between rm;n and rmax. Therefore, the probability that rp is larger than the coupon rate is simply

(4)

Given a certain probability, equation (4) can also be rearranged in terms of the coupon that is compatible with such a payment probability for the debtor:

(5)

Similarly, by manipulating equation (3), for the creditor, the coupon compatible with a certain repayment probability is equal to:

(6)

An equilibrium coupon is obtained when ccrecntor = c debtor- There is no guarantee of an equilibrium, however, because it depends on the parameter values chosen (see Figure 1). At the point(s) where the coupon demanded by the creditor (supply) intersects the line that is compatible with the debtor's fundamental coupon paying ability (demand), an equilibrium is reached. However, if the repayment probability is reduced, for example, by a reduction in the maximum return to 15 from 21 percent of the original amount borrowed, as shown in Figure 1, no equilibrium is possible and no lending would occur. Algebraically the equilibrium coupon solves the following equation:

194

Figure 1. Creditor's and Lender's Supply and Demand Coupons

Note: This depicts the supply and demand coupons and probabilities of payment, given a risk-free rate of 10 percent, a minimum return from the project of 2.5 percent, and a recovery value of 6.25 percent.

(7)

As depicted in Figures 1 and 2, for some parameter values there may be two equilibrium coupons. In those cases it is clear that for the debtor the lower coupon dominates the higher coupon outcome, to maximize the debtor's utility and minimize the probability of an actual default (the lower path in Figure 2).

Cost of Default and the Equilibrium Coupon Any equilibrium outcome will have to satisfy the no-strategic-default condition, which depends on A, and RV, as shown in Figure 2. A "true" equilibrium, to avoid a strategic default, has to be below the line described by the sum of the recovery value and the cost of default. Figure 2 indicates that there is a non-negative cost of default that is a necessary requirement for any equilibrium with risky lending to occur. A zero cost of default, that is, A = 0 percent, would be compatible only with a 1 0 percent coupon and a constant recovery valve that is equal to the risk-free rate — that is, the creditor would recover 10 percent return whether there was a strategic default or not.

195

Figure 2. Equilibrium Probability of Payment, and the Maximum Return

Hence, if the cost of default is low, only the relatively "safe" projects—that is. high-quality sovereign debtors—would in equilibrium receive financing. Projects that are risky would not receive any funding, because the coupon that the creditor would need to charge the debtor, to cover the creditor's risk, would encourage the debtor to enter into a strategic default. However, as the cost of default increases, for example with A = 7 percent, and the equilibrium shifts to B, riskier projects are able to receive some funding. Their equilibrium repayment probability will have to be at least 53 percent or greater in this equilibrium for financing to occur. If the cost of default is increased even further to A, = 10 percent and hence equilibrium A, the riskiness of the projects that can now be financed in equilibrium is increased and the required probability of repayment reaches its global minimum of 41 percent. Because all risky projects are positive net present value projects, the increased availability of financing is welfare enhancing.4 Moreover, as Figure 2 shows, not only is there a minimum cost of default, to allow risky projects, there is also a maximum cost of default. When A, is larger than 10 percent, as in this example, no benefit is gained from opening up the opportunity of financing additional risky projects. Instead welfare is lost, to the detriment of both the debtor and in some states the creditor, because if there is a bad-luck default a larger socially inefficient cost is incurred.

4See Nystedt (2001) for a more elaborate discussion on the welfare effects.

196

Probability of Payment and the Recovery Value A remaining question is how the equilibrium probability of payment relates to the recovery value. As shown in Figure 3, using the negative root in equation 7, the equilibrium coupon is strictly decreasing with the recovery value on a particular debt instrument. In much the same way the equilibrium coupon was derived, the equilibrium payment probability can be derived from

solving for p where

(8)

Plotting the equilibrium payment probability in a figure (see Figure 4), it is clear that the higher the recovery value, the higher the probability of payment, because the coupon the creditor demands in equilibrium can be lowered. If default would occur, abstracting from the possibility of a strategic default, a debt instrument with a higher recovery value would be preferable to the creditor and he would feel more comfortable with extending credit to risky projects. For example, in this model the use of collateral to back a new loan to a sovereign would actually increase the probability that the sovereign remains solvent, because the sovereign's debt service burden would decrease.

Effect on Secondary Market Price of Already Existing Debt While the equilibriums derived and discussed above reflect the effects of changes in key variables on newly contracted debt, already existing debt will also post significant price changes with changes in RV and X. If, for example, the debt instrument under consideration is near the strategic default frontier and the cost of default is reduced, that particular debt instrument would see its price fall from its par value to RV, even though the fundamental repayment probability of the sovereign remains unchanged. In the case of changes in RV, it would affect the price of the debt instrument by rearranging equation (3) to

where RVo represents the old equilibrium recovery value and A/?V the change in recovery value. The impact on the price of the existing debt instrument by changing the recovery value can then be simply expressed as

197

Figure 3. Equilibrium Coupon, Strategic Default Costs, and Recovery Value

Figure 4. Equilibrium Probability of Payment and Recovery Value

198

(9) with a positive price change if the recovery value increases and a negative price change (a loss to the creditor) if the recovery value is reduced.

Implications When considering recent initiatives to enhance PSI in crisis resolution, it could be worthwhile to consider how these initiatives affect the debt instrument specific recovery value and cost of default. For creditors, being able to discipline debtors is a crucial aspect of international debt flows. If the cost of default is lowered, there may be an effect on the availability and structure of the future financing provided by creditors—that is, they will adapt. Moreover, it would cause sharp price falls for those debt instruments and debtors for which the cost of default frontier was binding. As Figure 2 shows, the decision on whether, for example, to make a debt contract easier to restructure, and hence reduce the cost of default, depends on whether the cost of default is already seen as too high and hence a reduction would on net reduce welfare losses, or whether the cost of default is in an intermediate area (in the above example between 0 and 10 percent) where it is binding and actually reduces the amount of risky lending. A reduction in the cost of default in the latter case would actually reduce the lending to risky borrowers, while less risky debtors would be unaffected by small enough changes. Moreover, if default costs are very high, only countries that suffer from extreme bad luck and are in very bad shape will default. By definition this means that the average recovery value for all creditors will be very low. If, on the other hand, default costs are low, such that debtors with plenty of spare debt servicing resources choose to default, recovery values for a while will be higher, because the debtor and creditor can agree on a debt service profile that could be serviced at a higher level than in the case of a truly bankrupt debtor. An implication, therefore, from the framework presented here is that if enhanced PSI in crisis resolution means the net lowering of the costs of default, and lending occurs with the cost of default in the intermediate area, recovery values following an increase in sovereign defaults should be expected to go up, not down. It should also be noted at the outset that if creditor moral hazard plays an important role in world capital markets, it will also affect the key parameters for the debtor. Multilateral financing packages that include no private sector involvement, and hence maintain creditor moral hazard in the system, will lower the cost of default for the country (by mitigating output losses), lower the probability of default (an up-front lower interest can be charged by the creditor because the creditor has some "insurance"), and increase the recovery value if default still occurs. Interestingly, therefore, initiatives that increase the cost of default and reduce the recovery value may be optimal in a world where the level and structure of financing flows has already been affected by creditor moral hazard. However, if

199

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt the creditor moral hazard inherent in the global financial system is believed to be limited, there is less room to make the argument that the creditor's position needs to be weakened, because most likely the current state already reflects an equilibrium outcome with debtor moral hazard but no creditor moral hazard. What the model shows, therefore, is that the balance between creditor and debtor "rights" is a subtle one, and great care needs to be taken to not shift the balance too far in either direction. Moreover, whether the current cost of default in the international capital markets is in the intermediate area or too high would critically determine whether efforts to reduce a debtor's default costs would increase or decrease fe-global welfare.

ll. Debt Payment Suspensions: Private Sector Reaction and Adaptation

During a financial crisis the capital outflows from a country may be so over- whelming and broad-based, reducing the country's foreign exchange reserves, that there is no time to enter into negotiations on how to restructure the debt, and a suspension on debt payments and other capital transactions, including the freezing of domestic bank deposits, may have to be announced. While the outflows can originate in the corporate sector, a massive outflow may lead to the socialization of the debt in order to salvage the local financial system and also avoid the wholesale bankruptcy of a country's corporate sector. In such a situation a sovereign will be faced with the decision of which debt payments to suspend first, a pecking order of debt, and for what amount of time. As discussed in the previous section, this default decision will depend on the relative costs of defaulting on some debts and not others. Moreover, the sovereign needs to consider how comprehensive the suspension should be, which asset classes will be affected (bonds, loans, equity, domestic or external, etc.), whether the suspension will be voluntary or coercive, whether it is intended to be done with the implicit sanctioning of the official community, and what the implications of the suspension are for future market access. Therefore, the coverage and the magnitude of debt payments suspended can vary between delaying one payment on Paris Club debt, leaving the external private sector debt intact, full-scale imposition of capital controls, declaration of a corporate and sovereign standstill, and the freezing of bank deposits. The private sector creditors will react and adapt to all of these decisions for which debt payments are suspended, both ex ante and ex post. Unilateral suspensions of domestic or foreign debt payments form a natural, but unwanted, part of all debt restructuring. They are and have been a fact of life for debt instruments. Frequently, the actual time it takes for a sovereign or a corporate to launch a restructuring offer for either its bonds or its loans means it may have to actually enter default and risk collateral damage in so doing, before a successful restructuring has been completed. Collateral damage in this case could result in an inability to borrow internationally and domestically, the triggering of cross-default clauses with the debtor losing control over the debt restructuring process, a grab race for assets prompting a fire sale, loss of investor and domestic confidence,

200

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION collapse in the secondary market value of the debt, disruptive litigation, and a broad-based rush to the exit, triggering large capital outflows by both domestics and foreigners. These potential side effects of entering into default by declaring unilateral suspension have led a number of observers in both the official and academic communities (see Eichengreen, 2000, for an overview) to suggest a more predictable framework to manage suspensions, taking as an example the U.S. corporate bankruptcy context (how a sovereign can be forced to accept the court's decision, unlike in the corporate context, is a nearly insurmountable problem). There have been many suggestions (see IMF, 2000c, for a brief overview). Among them are setting up an international bankruptcy court replicating U.S. Chapter 11 bankruptcy protection for sovereign debtors; giving the IMF the right to declare a stay on litigation and declare temporary payment suspensions; or officially, explicitly or implicitly, sanctioning debt payment suspensions by allowing the IMF to lend into private sector arrears. While all of these proposals take the current state of play with potentially damaging ad hoc payment suspensions as given, few of the recent studies consider that for some types of creditors there is an advantage in having an unpredictable and uncontrolled payment suspension procedure in place, because it increases the debtor's cost of default. Furthermore, few observers have considered what is the potential impact on the structure and level of international capital in a world where the official sector is increasingly willing to sanction debt payment suspensions.5 As indicated in Section I, a more orderly framework for payment suspensions may have a similar effect on private sector adaptation, such as that of reducing the cost of default in Figure 2. Whether a more orderly framework will be welfare increasing or welfare reducing clearly depends on whether one believes that the current cost of default imposed on debtors is beyond the range where more risky projects can receive new financing if the cost of default is increased.

Issues Regarding Orderly Payment Suspensions The advantage of payment suspensions for the debtor in crisis, at least in theory, is that they can lock in both foreign and domestic private sector capital in a given country for a limited time period by restricting net capital outflows. From the official community's point of view, the main policy challenge has been viewed as whether and how payment suspensions should be officially sanctioned and whether a predictable framework for their imposition is needed. There are several variables that will affect the impact of more frequent payment suspensions. Four are described here.

Can Payment Suspensions Be Voluntary? It is likely that payment suspensions will have the least effect on the future structure of international capital flows if they are voluntary—that is, forbearance by a certain creditor group in return for some "sweetener." The inherent problem with the voluntary approach is that it is similar to raising new money. If a country suffers from only

5Some exceptions are IMF (2000b) and IMF (2000c).

201

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt a temporary debt service hump and asks some creditors to forgo debt service payments during that hump in return for a larger debt service payment later, the debtor could decide to issue a new bond rather than declare a payment suspension for a certain creditor. Hence, sovereign debtors will not declare a payment suspension in cases where they are facing a pure liquidity problem, because raising new debt is a realistic option and a better choice. For example, there would be no reputational loss and no risk that the debtor opens itself up to, say, litigation threats. If market access has been lost, and a payment suspension may be the only option, it is hard to see how it can be completely voluntary. At a minimum there is at least a need for a credible threat of default to encourage some creditors to agree to a payment suspension. In a situation when solvency concerns as well as liquidity concerns are important, the value a debtor can offer a creditor in a voluntary payment suspension negotiation would be through arbitrarily making that creditor's claim de facto senior to that of other creditors.6 Hence, in this scenario the sweetener needed to reach a voluntary agreement with one creditor group will occur at the expense of another creditor group. However, this is analogous to issuing either secured debt or very short-term debt. Voluntary payment suspensions also allow for the asymmetric treatment of creditors with limited potential ramifications down the road. Partially voluntary or coercive payment suspensions may have to be comprehensive across at least all foreign currency creditors, whether domestic or senior, in order to avoid the creation of new seniority structures.

Coordination Problem? In the cross-border emerging corporate borrowing context, voluntary payment suspensions have been quite common for both bonded debt and syndicated lending. The presence of a well-established bankruptcy procedure and creditor protection, in order to avoid asset stripping, makes voluntary payment suspensions more credible and easier to negotiate. In the sovereign context there is no bankruptcy court, and therefore it is natural to expect that voluntary payment suspensions will be harder and costlier to negotiate. The direct costs of reaching a payment suspension agreement with a group of creditors will depend on both the maturity of their claims and how other creditors are being dealt with. • Longer-term creditors will be unwilling to agree to a payment suspension if their forbearance is being used to bail out shorter-term creditors. Unlike the situation in the 1980s, when the long-term and short-term creditors were the same group of banks and could be more easily coordinated, the current creditor dispersion will make coordination much harder. Moreover, the very nature of the causes of the crisis will have an effect on the feasibility of reaching a payment suspension agreement. For example, longer-term creditors may be more willing to accept a pause in debt service if they believe that the pause will be used to enact reforms that increase the likelihood of better debt service performance in the future.

6For example, in return for its forbearance, a creditor will either receive ancillary business if it is a bank or have its claim's obligor status be upgraded to that of the sovereign rather than that of a corporate (see, for example, the description of the Korean crisis experience in IMF 2000b).

202

However, to agree, these longer-term creditors will ask for some sort of sweetener. The attractiveness of the sweetener, however, will be a function of whether the country is illiquid or insolvent—that is, it will depend on the magnitude of the creditor concession required. In cases where the size of the sweetener demanded by creditors to agree to a voluntary payment suspension is larger than the country's medium-term debt servicing capacity, a voluntary payment suspension could not be arranged unless it is expected that the "residual," the gap between the sweetener and the debt service capacity not covered, is supplied by a third party—for example, the official community. • Furthermore, the size of the sweetener is likely to be affected by the compre- hensiveness of the payment suspension. If only long-term investors are asked to participate, for the implicit benefit of the short-term investor, then the sweetener will most likely be upsized to reflect this. • Shorter-term creditors will be costlier to persuade to participate voluntarily in a payment suspension. In situations where this group of investors is fairly small, they will face the decision of either running now and receiving the full par value of their claims, or running later and receiving either a recovery value (because a possibility exists that the sovereign will be in default even after the agreed upon payment suspension period ends) or the par value plus some sweetener. Hence, short-term creditors could potentially agree to a payment suspension if the sweetener is large enough, and in contrast to longer-term creditors, for them the present value of the sweetener offered has to exceed the par value of the short-term credit.

What If Payment Suspensions Are Not Voluntary? Partially voluntary or coerced noncomprehensive payment suspensions are more likely to have an impact on the structure and level of international debt flows because, by definition, various groups of creditors are more or less likely to be amenable to moral suasion or coordination. If, for example, one group of creditors is more likely to be successfully subject to moral suasion and hence persuaded to participate in a payment suspension for a nonvoluntary sweetener, that creditor group is likely to adapt and in the future make sure to reduce its vulnerability by either charging a higher fee up front—that is, incorporating the moral suasion risk in its initial lending decision—or reducing the amount of capital it actually lends and thereby its exposure (similar to the analysis behind Figure 2). Hence, a partially voluntary payment suspension may be successful in resolving one particular crisis, but it may soon lose its usefulness as an instrument because credit flows are likely to be increasingly channeled through the creditor groups that are less amenable to moral suasion. In a worst-case scenario, where a sovereign debtor is insolvent and broad- based capital flight is occurring, the debtor may have no choice but to resort to declaring a unilateral payment suspension. Extending the scope of payment suspensions to most private creditor groups, as well as freezing, in this example, bank deposits, may give some breathing space to the sovereign debtor and also avoid an asymmetric treatment of creditors. Thus, making the payment suspension

200

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt comprehensive would close all potential capital outflow channels and treat all private creditors symmetrically. By not favoring one asset class over another, this approach limits the incentives for the private creditors affected by the payment suspension to create senior debt farther down the road.

Should Some Private Creditors Be Treated as Senior? Arguments have been made that some private sector creditors should receive preferential treatment when a payment suspension is imposed. Notably this discussion has focused on trade credits or new money extended to the sovereign during a crisis. However, the experience of the 1980s provides some useful examples on how the private sector will adapt if certain debts are treated preferentially. The example of the impact of the 1980s restructuring experiences on sovereign syndicated lending was mentioned earlier. Another lesson from the 1980s was that in many cases trade financing was excluded from loan renegotiations, giving creditors and debtors a natural incentive to structure future general purpose credits in the form of trade credits (Buchheit, 1991). With respect to new money, the motivation for treating it as senior is often said to be that it is in all creditors' interest to make sure that the country can still get access to capital in order to limit the output losses that occur during the crisis. This is not necessarily true. As mentioned in Section I, output losses may be the only effective disciplining device creditors have available to ensure repayment. New money shifts the bargaining strength somewhat toward the debtor. Hence, it is not clear whether all private creditors necessarily favor the approach of making new money senior.

Payment Suspensions and the Composition of Cross-Border Debt Flows As mentioned above, partially voluntary or coercive payment suspensions that are not comprehensive—that is, one creditor group is treated more favorably than another—are very likely to lead to a change in the composition and level of debt financing flows to emerging markets in the future. While this primarily affects the recovery value of debt instruments, some have argued that a benefit of more frequent imposition and official sanctioning of payment suspensions is to make short-term debt less attractive and long-term debt more attractive. This line of reasoning assumes that the incentives of longer-maturity debt are more similar to those of the sovereign debtor and of the official community. For example, temporary payment suspensions on only short-term debt could perhaps still allow the sovereign to service its long-term debt. Unfortunately, however, when a country starts moving toward a crisis situation, the maturity of its debt stock will most likely shorten, for three reasons. • One reason is that foreign creditors will almost by definition be unwilling to take long-term exposure toward the country as previously long-term debt falls due. • Second, this unwillingness by foreign creditors to extend new long-term credit will lead to an upward sloping yield curve, making long-term borrowing for

204

the country very costly. This would encourage a shift by the debtor to borrow through shorter-term instruments. As more and more creditors offer shorter-term loans at increasingly higher interest rates, the yield curve of the sovereign is likely to widen as well as flatten. Eventually, when a debt service interruption becomes imminent, the yield curve of the sovereign debtor will invert, with markets pricing in a very high risk of default in the short end. • The third reason is that a payment suspension on debt amortizations, where the sovereign still makes interest payments, may lead to default on longer- term bonds or loans that are amortizing or to the nonpayment of principal of initially long-term debt that matures within the payment suspension period. In such cases, the payment suspension could trigger cross-default clauses on other longer-term debt as well. Longer-term creditors will then have to decide whether they want to accelerate their debt and be treated the way the short-term creditors in the resolution are treated, or whether they want to keep their acceleration option open as long as their debt is still being serviced. Shorter-term creditors may also demand, as a condition to agree to a restructuring, that longer-term creditors be brought to the negotiation table.

Implications The higher likelihood of a payment suspension, and thereby the reduction in both the cost of default and perhaps also the recovery value, may encourage a shift in the maturity structure of debt toward the shorter end of the spectrum, as creditors move to protect themselves. This can be done, for example, by providing financing only in the overnight market, to ensure a withdrawal before a payment suspension becomes operational. Hence, if the main benefit of having more orderly and officially sanctioned payment suspensions is to reduce the reliance on short-term debt, it is worth noting that it may create the reverse result, encouraging creditors to lend at even shorter maturities to ensure that they get their money out. Indeed, one interpretation of the 1980s debt crisis resolution is that a similar dynamic was at work when bank lending shifted, following the resolution of the crisis, from mainly medium- and long-term maturities to short-term interbank debt. Maturities shorter than that of interbank lines would severely restrict the effectiveness of a payment suspension.

III. Bond Exchanges and Private Sector Adaptation Facing a situation where a payment suspension, whether comprehensive or not, looks likely, a sovereign debtor could try to approach its bondholders to discuss reprofiling the bond payments to avoid the actual imposition of a payment suspension and a subsequent default. As discussed before, truly voluntary agreements will be uncommon because true liquidity crisis situations will be hard to distinguish from solvency ones, and the outcome of any bond restructuring or reprofiling discussion after market access has been lost will be determined by the

205

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt credibility of the sovereign's default threat. During the past year or so there have been defaults on both sovereign eurobonds and Brady bonds,7 and bond exchanges of defaulted or nearly defaulted bonds have been completed by Pakistan, Ukraine, Ecuador, and Russia (IMF, 2000b). The ease and success of the recent bond exchanges have surprised many observers. The reason for the surprise was that eurobonds and Brady bonds were believed to be technically hard to restructure, both for legal reasons (for example, it requires unanimity and it raises the threat of litigation) and for practical reasons (for example, restructuring requires the identification and coordination of thousands of bondholders). It may, however, be too early to draw conclusions about the applicability of recent bond exchanges for future bond restructurings because the cost of default, A,, may have been lower than previously expected and creditors will seek new forms of debt instruments with a higher A, and/or RV, shifting emerging market borrowing and lending decisions to a new equilibrium where financing may be reduced or more costly. Moreover, the impact on already issued bonds will also reflect a certain adaptation to the new circumstances of lower or higher A, or RV. The following are three examples. • Private creditors will adjust upward the price at which they are willing to participate in future bond exchanges if they view recent experience as indicating that emerging market bond recovery value estimates were too low. As shown in equation (9), a higher RV would increase the price of existing debt, making it harder to offer large sweeteners. For future lending decisions, according to the model, a higher recovery value will reduce in equilibrium the probability of default, for the same debtor, and would be beneficial for both debtors and creditors because more risky sovereign debtors would receive financing. • While the success of recent bond exchanges has put into question creditors' cost of default estimates—that is, the cost of default may be lower than previously anticipated—a more broad-based introduction of collective action clauses or the use of exit amendments may reduce the cost of default even further. However, recent successful litigation (e.g., Elliott Associates against Peru) could indicate that the threat of litigation has been underestimated. It may actually make bond exchanges more costly in the future and work in the opposite direction from the initiatives of making bonded debt easier to restructure. • Private creditors are likely to adapt the structure of their lending vehicles in ways such that restructuring and creditor coordination may be more difficult in the future. The next subsection discusses the impact on existing sovereign bondholders of the recent experiences of easier than expected bond exchanges. Changes in recovery value and cost of default estimates by these bondholders will have a key bearing on whether bond restructurings will be as easy to complete in the future. After this discussion, and using the basic model as a background, the second subsection will discuss the impact of collective action clauses, exit amendments, and the renewed

7As several authors have pointed out, for example Eichengreen (1999) and Petas and Rahman (1999), defaults on international bonds have been frequent, historically speaking, especially in the 1930s. In 1999, Ecuador was the first sovereign to default on both eurobonds and Brady bonds.

206

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION litigation threat on both existing bonds and the future lending/borrowing equilibrium. The last subsection will go further toward reflecting over how the bondholders could adapt in a new equilibrium by changing the structure or terms of their lending.

Impact on Outstanding Bonds and Bond Restructurings In evaluating bond exchanges, different investor groups will have different reaction functions. Mark-to-market investors, having borne the full brunt of the fall in secondary market price of the to-be-exchanged bonds, tend generally to compare the net present value (NPV) value of the exchange offered (at some discount rate) with the current market price of the to-be-exchanged bonds and with the recovery value of not participating in the exchange. In the simplest case, if the NPV of the exchange bond is higher, taking into account the likelihood that the exchange will succeed and the haircut in terms of a potential debt write-off, then the holder of the to-be- exchanged bond has an incentive to tender its bonds in the exchange. This is the way most fund managers would rationally respond. For commercial banks a similar response function is less likely because they generally do not mark-to-market all of their investment portfolios, but make a reasonable approximation. The response function of retail investors is much more uncertain; their tender decision may be based less on an NPV comparison than on whether they have to participate in a debt write-off. The recent successful bond exchanges have involved some form of sweetener (see Table 1). For example, the post-announcement price has been 10-30 percent higher than the pre-announcement price, with the exception of Pakistan. In Russia, the sweetener was enhanced with the upgrade in the obligor to that of the sovereign. Large sweeteners have also reduced the incentive in many cases for at least mark-to- market bondholders to litigate, because capturing the sweetener provides an immediate gain, while the outcome from litigation is uncertain and time-consuming. The limited experience to date with external bond exchanges required that several key parameters of the debt pricing equation presented in equation (3) and the cost of default were "guesstimated." For example, without much empirical basis the recovery value had consistently been estimated at 18-20 cents on the dollar for sovereign eurobonds. If the recovery value is actually much higher—for example, close to the average recovery value in the U.S. high-yield sector (47 cents on the dollar with an average 2.1 years for collection)8—market prices of bonds that were in default would shoot up, as reflected in equation (9), to reflect the new revised estimate of recovery values. It may be that low recovery value estimates were one reason that prices post- announcement could credibly be substantially higher than prices pre-announcement in all of the recent bond exchanges with the exception of Pakistan. In Pakistan, the actual number of bondholders was limited and fairly well known in advance, and the high share of locally connected investors reduced the necessary size of the sweetener. In addition to the size of the sweetener, enhancing recovery values further is the speed of curing the default (this would also reduce the cost of default to the debtor,

8See Rappoport (2000) and Rappoport and Xu (2000) for a much more detailed discussion on the sensitivity of emerging market and U.S. high-yield bonds to default probability, recovery value, and other assumptions.

207

©International Monetary Fund. Not for Redistribution Table 1. Market Reactions to Recent Bond Exchanges (in percent) Price Price Week Week Post Assitance- Price Day Before After Face Completion/ then/ Comptetion/ Date of Exchange Exchange Listing After Announce- Announce- Price at Value Listing Pre-announce- Pre-announce- Default Announced Completed Price Default ment ment Completion Write-off Price ment Price ment Price Pakistan 6% USD 2002 eumbond no default 11/15/99 12/6/99 100.0 58.3 60.4 63.6 0 -36 4 9

Ukraine 16% DEM 2001 eumbond 1/20/00 2/4/00 4/7/00 100.0 50.0 55.5 67.0 81.5 0 -19 212 47

Russia Prins 12/2/98 2/11/00 8/17/00 57.8 6.5 18.5 24.4 32.0 -38 -45 32 73 Ions 6/2/99 2/11/00 8/17/00 67.2 10.0 21.1 25.0 32.8 -33 -51 19 55

Ecuador Brady Par Bond 11/28/99 7/27/00 8/1 1/00 27.4 33.8 35.5 39.5 39.5 -60 44 11 11 Brady Discount Bond 8/28/99 7/27/00 8/11/00 47.0 34.4 37.0 46.7 46.7 -42 __l 26 26 Brady PD1 Bond 2/28/00 7/27/00 8/11/00 27.0 23.5 25.3 30.5 30.5 —22 13 21 31 Note: Bonds presented are not all of those restructured. Prices are ask and include accrued interest, when bond is in default. For Pakistan accrued interest is estimated. Default is the date of the first missed coupon payment, except in the case of Ukraine, where the default reflects the missed amortization on its October 2000 eurobond.

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION so the net effect in equilibrium may be ambiguous). The experience from recent bond exchanges suggests that they were completed much faster than expected (one to two years compared with the lost decade of the 1980s debt restructurings), and in one case the exchange was even completed before the actual default occurred (Pakistan). Bond exchanges are not guaranteed to succeed, however, and significant deal risk may remain even after they are announced, but overall the probabilities of success for the recent exchanges once announced have been very high,9 in part reflecting the large mark-to-market gains inherent in the offers for tendering bondholders (see Table 1). Recent exchanges, such as the one in Ecuador and Ukraine, have taken a "carrot and stick" approach to encourage bondholders to tender into the exchange. The adaptation of private creditors' expectations of recovery will make it more difficult to offer similar carrots (i.e., large mark-to-market gains) in the future. For example, if the recovery value following default is now believed to be consistently higher than before, secondary market prices will actually not fall as low as they did, for example, in the Russian case, and hence offering as attractive a sweetener may not be credible from a debt-service capacity point of view. Therefore, it remains to be seen whether the experience with recent bond exchanges is useful in determining whether future bond exchanges will be equally successful. These bond exchanges have managed to address very successfully some of the issues with regard to the technical aspects of how to restructure sovereign bonds that were unknown before. By doing so, these exchanges have firmly pierced the "halo" that was previously believed to surround eurobonds and Brady bonds. It remains to be seen whether the private sector adapts by switching to harder-to-restructure lending instruments.

Cost of Default, Legal Innovations, and the Threat of Litigation The recent bond exchanges by Ecuador, Pakistan, Russia, and Ukraine are generally regarded as successful in terms of having obtained a high degree of investor participation while avoiding creditor litigation. The keys to success of the voluntary exchanges have been the sweeteners provided. As discussed above, when the market adjusts its expectations for recovery values it will become more difficult to offer such sweeteners in the future, and the threat of litigation may increase. The experience of recent bond exchanges and their success, together with the official community initiatives to increase the prevalence of collective action clauses10 (CACs), raises the ques-

9This can be determined by looking at the differential between the price of the new exchange bond when issued (basically forward contracts on the new bond) and the price of the old to-be-exchanged bond. A small differential implies that investors believe the exchange will succeed. When-issued markets existed in all four bond exchanges. 10The clauses commonly referred to as CACs are as follows: a majority action clause (allowing a qualified majority of bondholders to bind a minority); a sharing clause (stating that any funds received through, for example, litigation by one bondholder have to be shared with the other bondholders based on their share of the outstanding bond); and a collective representation clause (allowing a trustee, for example, to represent bondholders, facilitating majority actions). CACs can facilitate creditor-debtor negotiations because they reduce both the threshold needed for achieving a restructuring agreement (the majority action clause) and the potential threat of litigation from "holdout" creditors (reducing their incentive to litigate through the sharing clause). Bonds issued under English law typically include CACs. These clauses are not regularly contained in bonds issued under New York law.

209

©International Monetary Fund. Not for Redistribution Gabrielle Lipworth and Jens Nystedt tion of whether these events have an impact on the cost of default and how it will affect the future emerging market borrowing/lending equilibrium.

Collective Action Clauses and Exit Consents As discussed in IMF (2000b), the actual usefulness of CACs in bond exchanges has been limited.'' For example, it appears to have been less difficult than anticipated to organize and locate individual bondholders and to offer them an exchange. Moreover, the value of CACs as a threat to encourage bondholders to tender in an exchange seems secondary. As discussed earlier, it was the large carrot (sweetener), rather than the stick, that clinched the exchange deals. However, there is at least one example of CACs being used as a stick. Ukraine dealt with potential holdout creditors within the context of its voluntary exchange offer by making use of the CACs embedded in three of the four affected bonds, after a substantial majority had agreed to the exchange. Experience would therefore suggest that the main usefulness of CACs is that they effectively limit litigation, while exit consents12 can be used to increase the success and ease of any bond restructuring. The more widespread use of both CACs and exit consents has the potential of reducing the cost of default for the debtor, increasing the possibility of a strategic default, with a subsequent shift in the lending/borrowing equilibrium (from A to B in Figure 2). Again, as was the case with the payments suspension discussed earlier, the actual outcome on the level of financing flows will depend on whether the preceding cost of default was binding. For example, if creditor moral hazard is ignored, and if the cost of default would be binding and allow for the maximum amount of financing to emerging market sovereign in equilibrium (in Figure 2 this would be consistent with a X = 10 percent), any reduction in the cost of default would lead to a reduction in financing available to more risky emerging market sovereigns, thereby reducing welfare. Because more than 60 percent of the J.P. Morgan EMBI+ benchmark index is compromised by issuers below investment grade, it is likely that these are also the debtors most sensitive to changes in the cost of default. The impact on flows to emerging markets could be substantial. Note, however, that this effect, according to the simple model, would not show up in the spreads or coupons the debtors have to pay. If strategic default is becoming increasingly likely for a debtor, no creditor would extend financing at any realistic interest rate. Moreover, higher quality debtors would be unaffected. In terms of empirical evidence (see IMF, 2000a), there is some evidence that bond financing

"Empirical work, however, suggests that collective action clauses tend to reduce the cost of borrowing for some borrowers. See, for example, Eichengreen and Mody (2000), and for a more critical assessment Becker, Richards, and Thaicharoen (2000). l2The use of exit consents is a strategy that, similar to CACs, can be employed to provide the stick in a bond exchange and induce bondholders to participate in the exchange by changing key nonpayment terms and thereby reducing the value of the old bond after the exchange has been completed. The main advantage of exit consents is that they can be used in bonds that otherwise would require unanimity to change any of their payment terms, and they can thereby replicate some of the features of CACs. Exit consents in a sovereign New York law bond were first used in Ecuador. See Buchheit and Gulati (2000) for an insightful discussion.

210

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION to emerging market sovereigns remains ample, and only the lowest, such as single- B minus creditors, are shut out of the market.

The Litigation Threat One of the major deterrents to creditor litigation has been the difficulty of attaching sovereign assets. First, many assets can quickly be hidden if the sovereign believes them to be at risk. Second, U.S. case law suggests it may be difficult to attach significant assets even in the case of a favorable ruling. However, the prevailing belief that litigation remained a rather limited threat may change because of the recent success of Elliott Associates in its case against Peru (see Lindenbaum and Duran, 2000). If sovereign litigation as an alternative becomes feasible and economical for creditors, they will have a further tool to counteract any reduction in the cost of default. Surprisingly, as shown in Figure 2, the subsequent increase in the cost of default may be welfare enhancing for some parameter values. Some litigation may therefore be a good thing, depending on what the overall cost of default is, and may put a limit to other forms of creditor adaptation by changing the form of the lending instalment.

Changing the Structure of the Eurobond Market In the same way creditors adapted to the restructuring experiences of the 1980s by changing the structure of their lending vehicles through increasing the cost of default or reducing their lending by charging a higher interest rate up front, following the recent bond exchanges there may be incentives for private sector creditors to again adapt so as to avoid the shift to a low financing equilibrium (B). Increasing the seniority of the lending instruments by increasing A, or RV also has value to avoid default or to have an improved claim in a restructuring situation. A creditor can increase the seniority of its claim in a number of ways—through requesting up-front collateral, seeking private or public sector guarantees, or changing the ownership and voting structure of the creditor (many more are possible). • As mentioned in the introduction, there may be an increased demand among creditors for lending vehicles that are collateralized by the borrower with cash flows or real assets that can be captured outside the debtor country's borders and hence could easily be transferred into the control of the creditors in the event of a default. The two main types of collateral that a sovereign could pledge would be its foreign exchange reserves or future export revenues13 of its state-owned companies, such as oil or telecom. By definition the amount of collateral available to be easily pledged limits the use of collateralized borrowing for the debtor. While there are significant amounts of quasi- sovereign collateralized bonds outstanding, none of these collateralized bonds has yet been restructured. Specifically, it is not certain how one would actually

13Rating agencies assign future flow securitizations substantially higher ratings (up to four notches in some cases; see Standard & Poor's, 1999) as the political and transfer risk of these types of debt is miti- gated by their structure. In emerging markets, major examples of issuers using future flow securitizations are PdVSA and PEMEX (Venezuela's and Mexico's oil companies).

211

go about restructuring these bonds and asking the creditor to take a significant haircut. In the case of Pakistan, one eurobond that was not involved in the restructuring was a government-guaranteed Pakistani Telecom bond, which was collateralized by the future export earnings on incoming international tele- phone calls. According to Standard & Poor's (1999), the structure of the Pakistani Telecom bond showed the resiliencies of these types of transactions, and if default occurs, recovery values will be very high because creditors can access the collateral, which is usually placed in an offshore trust. • An alternative would be lending guaranteed by either official or private insurance entities. For example, by making its lending conditional on the presence of a World Bank "policy-based guarantee," a lender should be able to make itself more immune to risks of unwanted PSI. Argentina, for example, issued policy- based guarantee bonds in October 1999, and Colombia is considering raising part of its financing in 2001 through the use of such World Bank-guaranteed bonds. Other public guarantee programs such as the International Finance Corporation's B-loan program, and that of other international finance institutions, effectively extend the preferred creditor status of these institutions to private creditors. In such cases, it is not clear how these bonds would be restructured. • By definition the use of collateralized borrowing is limited to corporations and could be relevant to the sovereign in the case of large export-oriented state- owned companies, which the sovereign can use for its borrowing purposes. Hence, there would be incentives to find additional ways of lending money where the security from trying to avoid PSI stems from how the lending trans- action is structured. It would be possible, for example, to repackage bonds and loans through the use of collateralized bond obligations (CBOs) or collateralized loan obligations (CLOs) and separate the legal ownership of the debt instrument, and hence the power to vote to participate in a debt exchange, from the stream of payments accruing in such a way that it would be extremely difficult to find a representative of the actual creditor lender with the right to decide on participation in the exchange on behalf of the CBO/CLO. The point is not that it would be impossible to restructure a CBO and CLO, but that it is more difficult, adding to the cost of default for the debtor if it decided to default on debt instruments that are predominantly held by CBOs and CLOs. These are some of the potential adaptations that bondholders could opt for. The simplest one, of course, is to increase the borrowing cost for the debtor. But it may be in both the creditors' and sovereign debtors' interest to race up the seniority ladder in order for more debtors to find cheaper financing and for the creditors to protect themselves. It is not clear whether such a seniority race is in the interest of official creditors, both bilateral and multilateral.

IV. Implications and Conclusions Efforts at crisis resolution that succeed in reducing potential inefficiencies and instability in the international financial system are in the interest of both the private and the public sectors. In the absence of clearly established rules of the game, the approaches adopted toward crisis resolution, and the extent to which

212

©International Monetary Fund. Not for Redistribution CRISIS RESOLUTION AND PRIVATE SECTOR ADAPTATION they are interpreted by market participants as setting a precedent, can have profound implications for the workings of the international financial system and the nature and structure of international capital flows. A key lesson from the 1980s is that as particular lending instruments are involved in restructurings, the private sector will seek out new instruments that increase the probability of repayment and are insulated from restructurings. The large-scale restructuring of syndicated bank loans in the aftermath of the 1980s debt crisis, while leaving eurobonds untouched, provided an important impetus to the use of the international bond market for emerging market borrowers. The experience with concerted interbank rollovers in Korea, Indonesia, and Brazil has shed some light on how the private sector will adapt to an increase in the frequency of payment suspensions. The expectation by the market that this will be the case will likely lead at least some international banks to cut their lines and run early in the face of an imminent crisis. Experience following the most recent string of crises has firmly pierced the halo surrounding international bonds. This experience with bond restructurings will likely lead bondholders to update their estimates of key variables, such as the cost of default and the recovery value, making recently successful bond exchanges less useful as predictors of the success or failure of future bond exchanges. A key question about the welfare effects of recent official sector crisis resolution initiatives is whether the current cost of default is too high, and whether welfare can be gained by reducing it, or whether it is in the intermediate region where a reduction would lead to a loss in welfare. Moreover, it is natural to expect that, as bond restructurings become more common, private sector creditors will increasingly try to structure debt so that it is harder to restructure, for example, by issuing securitized or guaranteed debt or adopting investor holding structures that are difficult to negotiate with. Such private sector adaptation to official sector crisis resolution initiatives, as discussed above, may not be costless, and they will lead either to increased borrowing costs for the debtor or to more short-term and rigid debt structures, as the sovereign debtor tries to offset increases in borrowing costs by agreeing to borrow through ever more advanced and renegotiation-proof structures. Either way, the net effect on emerging market capital flow may be negative, especially for the lowest quality sovereign borrowers.

REFERENCES

Becker, Torbjorn, Anthony J. Richards, and Yunyong Thaicharoen, 2000, "Collective Action Clauses for Emerging Market Bonds: Good News for Lower Rated Borrowers Too," IMF Seminar No. 2000-59 (Washington: International Monetary Fund). Buchheit, Lee, 1991, "The Pari Passu Clause Sub Specie Aeternitatis," International Financial Law Review (December), pp. 11-12. , and G. Mitu Gulati, 2000, "Exit Consents in Sovereign Bond Exchanges," UCLA Law Review, Vol. 48 (October), pp. 59-84. Cline, William, 2000, "The Role of the Private Sector in Resolving Financial Crises in Emerging Markets," Institute of International Finance (unpublished; Washington: IIP, October).

213

Dooley, Michael P., 1994, "A Retrospective on the Debt Crisis," NBER Working Paper No. 4963 (Cambridge, Massachusetts: National Bureau of Economic Research). , 2000, "Can Output Losses Following International Financial Crises Be Avoided?" NBER Working Paper No. 7531 (Cambridge, Massachusetts: National Bureau of Economic Research). Eichengreen, Barry, J., 1999, Toward a New International Financial Architecture: A Practical Post-Asia Agenda (Washington: Institute for International Economics). , 2000, "Can the Moral Hazard Caused by IMF Bailouts Be Reduced?" Geneva Reports on the World Economy Special Report 1 (Geneva: International Center for Monetary and Banking Studies). -, and Ashoka Mody, 2000, "Would Collective Action Clauses Raise Borrowing Costs?" NBER Working Paper No. 7458 (Cambridge, Massachusetts: National Bureau of Economic Research). International Monetary Fund, 2000a, "Emerging Market Financing: Quarterly Report on Developments and Prospects," Research Department (Washington). , 2000b, International Capital Markets: Developments, Prospects, and Key Policy Issues, World Economic and Financial Surveys (Washington). 2000c, "Involving the Private Sector in the Resolution of Financial Crises— Standstills—Preliminary Considerations," Policy Development and Review Department and Legal Department (Washington). Lane, Timothy D., and Steven Phillips, 2000, "Does IMF Financing Result in Moral Hazard?" IMF Working Paper 00/168 (Washington: International Monetary Fund). Lindenbaum, Eric, and Alicia Duran, 2000, Debt Restructuring: Legal Considerations—Impact of Peru's Legal Battle and Ecuador's Restructuring on Nigeria and Other Potential Burden Sharing Cases (New York: Merrill Lynch & Co.). Nystedt, Jens, 2001, "Sovereign Debt Restructuring: A Pecking Order of Debt and Private Sector Adaptation" (unpublished; Washington: International Monetary Fund). Petas, Peter, and Rashique Rahman, 1999, "Sovereign Bonds—Legal Aspects That Affect Default and Recovery," Global Emerging Markets—Debt Strategy (London: Deutsche Bank). Rappoport, Peter, 2000, "Valuing Credit Fundamentals: Rock-Bottom Spreads" (New York: J.P. Morgan Securities Inc.). , and David Xu, 2000, "Emerging Markets versus High Yield: Credit Fundamentals Revisited" (New York: J. P. Morgan Securities Inc.). Standard & Poor's, "Lessons from the Past Apply to Future Securitizations in Emerging Markets" (New York).

214

On the History of the Mundell-Fleming Model Keynote Speech

ROBERT MUNDELL

t is a great pleasure for me to speak at this opening of the first IMF Annual II Conference on International Macroeconomics and a special honor to be here at the inauguration of the annual Mundell-Fleming Lecture. Quite apart from the flattering distinction it confers on Marcus Fleming and me, it serves to commemorate a very special period in the Research Department at the Fund when, under the able leadership of Jacques Polak, it made an enduring contribution to the development of the standard international macroeconomic model. I think the best service I could provide tonight would be to give some of the background behind the development of the model, some of the influences that were important, and, yes, some of the defects of the model. If what I have to say seems too autobiographical, I can only respond that I wish Marcus Fleming could have been here to fill in the blanks from his point of view and redress the balance.

I I have read interpretations of my work that have made stylistic points about the "early" and the "late" Mundell, the first being a Keynesian, the second a classicist. Such periods may be relevant to painters, but are they really applicable to economists? I am not myself aware of any basic shift of direction. I did write on different subjects and use different models at different points in time, but why not? I worked on what came to be called the Mundell-Fleming model mainly over the years 1960-64, but before, after, and during this period, I was also publishing my work on the pure theory of trade, monetary theory, optimum currency areas, the public debt, the monetary approach to the balance of payments, customs unions,

215

©International Monetary Fund. Not for Redistribution Robert Mundell and the theory of inflation. The agenda, models, and information changed, but the periodization doesn't ring true. If there was an "early" Mundell, it was a classical one. Let me start as close to the beginning as seems necessary. After graduating from the University of British Columbia (UBC) in 1953,1 went as a teaching fellow and graduate student to the University of Washington, where I had my first real brush with macroeconomics, mathematical economics, and international trade. My next stop was the Massachusetts Institute of Technology (MIT) in 1954-55, where I was of course especially influenced by Samuelson and Kindleberger. After completing my doctorate exams at MIT in the spring of 1955, I made use of a Mackenzie King Traveling Scholarship from Canada to study at the London School of Economics (LSE). I had a special interest in Lionel (later Lord) Robbins and (later Sir) James Meade. I got a nice letter of acceptance directly from Meade, and he agreed to "super- vise" my thesis for MIT up until March 1956, when he was to leave for New Zealand. I want to discuss my relations with Meade. I saw him in his office about once a week, and also participated in, besides the Robbins theory seminar, the Meade-Robson (Robson was a political scientist) seminar on international economics, as well as lectures by Harry Johnson, who came up from Cambridge once a week to give a course in which he read—yes, read—his latest papers. In those two terms I wrote two papers, "Transport Costs in International Trade Theory" (Mundell, 1957b), and "International Trade and Factor Mobility" (Mundell, 1957a), which were two of five chapters of my MIT Ph.D. thesis. The latter article I presented in the Meade-Robson seminar, and I got helpful comments on it from Tadeusz Rybczynski, Dick Lipsey, Max Corden, and Steve Ozga, as well as James Meade and Harry Johnson. Throughout that year and the following summer in Boston, my work was entirely on aspects of the classical or Heckscher-Ohlin theory of trade, and I had no discussions about macroeconomics with Meade or anyone else. I had "read" Meade's (1951b) Mathematical Supplement. In June 1998 Max Corden stayed with me in Siena a few days, and reminded me of a conversation we had at the time. When asked whether I had read Meade's (195 la) Balance of Payments, I replied, "No, but I have read his Mathematical Supplement.'" This gave me the reputation (along with the prestige of coming from MIT), quite unmerited, that I was a mathematician. I never told anyone that when I began graduate work, I had zero knowledge of even rudimentary calculus. But my reply to Corden was not quite accurate. One didn't read the Mathematical Supplement. It was almost as tedious as the main book. What was exasperating was the taxonomy, roundly criticized by Harry Johnson in his review.' Meade has a very amusing footnote on combinations at the bottom of page 33, where he contributes the information, confirmed by William Baumol, that there were precisely 28,781,143,379 possible solutions to his model.

'This negative, even harsh, review of Meade's book cost Harry Johnson a friendship with Lionel Robbins, who was tenaciously loyal to his friends, and who only agreed to speak to Harry again on the occasion of Al Harberger's wedding in London in 1958.

216

Much later, in 1970, during a walk in the foothills of Mount Fuji, Meade told me that he had a mind like Pigou's—a "meat-grinder's mind," he said. He told a story about Pigou on his way out after a lecture being asked by a student if he had not made an error in the sign of an elasticity, at which point Pigou marched back up to the podium to his notes (presumably left for his assistant to return), looked up the relevant section, and simply replied "no." Meade said that he wrote down the equations, differentiated them, and reported the results in his book. It wasn't very exciting, but his two volumes and their appendices were nevertheless land- marks in the development of international economic theory.2 I learned a lot from Meade, of course—not macroeconomics, but his brilliant contributions to the classical model. This influence can be seen all through my "Pure Theory of International Trade" article (Mundell, 1960b), which was an expansion (and contraction) of two of the five chapters of my thesis. When you asked a question like "How much will a tariff, or unilateral transfer, or productivity change alter the terms of trade (or some other variable)?," you would find that Meade had produced the first definitive answer to that question. I was able to develop his work in some new areas, develop some of the dynamics, and gener- alize the model, following up on Mosak, in a multicountry framework. There are nevertheless in Meade's Mathematical Supplement (1951b) the equations of an international macroeconomic model. But when I was doing my work on this subject a few years later, I didn't make any connection with Meade's work. The reason, I think, is that my approach came through a Walrasian-like general equilibrium theory, which was at best only implicit in Meade's analysis. But there was one insight in Meade's work that I used extensively, and this came through in my "Pure Theory . . ." article. This was the role of "domestic expenditure," called "absorption" in Sydney Alexander's 1952 article in the IMF's Staff Papers. Of course Metzler and Machlup had used expenditure functions depending on income in their international multiplier work, Laursen and Metzler (1950) had made them dependent on income and the real exchange rate in their famous joint article in the Review of Economics and Statistics, and Chipman, Goodwin, and Metzler had used them in their treatments of the matrix multiplier. But Meade's equations in the Mathematical Supplement broke new ground by making domestic expenditure a function of income, interest rates, exchange rates, some prices, and all kinds of policy variables. He did not develop the implications of this emphasis. In his introduction to the Mathematical Supplement, Meade says he hopes his "model may somewhat further . . . the marriage between the 'classical' and 'Keynesian' analysis of the mechanism of the balance of payments. . . . What we need for balance-of-payments theory is a marriage of the Keynesian and the Hicksian type of analysis; and our model constitutes such an attempt." I think that does explain what he attempted to accomplish, and I think he was partly successful in doing so. It was not, however, what I was trying to do in my international macroeconomic model.

2You have to understand Meade's remark in the context of his own innate, self-effacing modesty. You wouldn't want to take too literally John Stuart Mill's statement in his Autobiography that he wasn't smarter than his contemporaries, only that he started a generation ahead of them.

217

Meade had been, since 1950, an ardent advocate of flexible exchange rates, and it was a still a hot subject at LSE. He had suggested that the signers of the Treaty of Rome achieve balance of payments equilibrium for each country by letting exchange rates float. I didn't have a position on this at the time but could not see why countries that were in the process of integrating with a common market should saddle themselves with a new barrier to trade in the form of uncertainty about exchange rates, or how economic theory could prove that flexible rates were preferable to fixed rates or a single currency.

II My interest in macroeconomics in that year, 1955-56, in London was veiy much beneath the surface, as I was writing a thesis that was entirely a development of the classical and Hecksher-Ohlin models. I spent the following year, 1956-57, as the post- doctoral fellow in political economy at the University of Chicago, and here I became especially interested in the work of Lloyd Metzler in theory and Milton Friedman in policy. Metzler's (1951) architectonic "Wealth, Savings, and the Rate of Interest" started me thinking about that model as a more suitable paradigm for macroeconomics than the Keynesian model. By 1955, Patinkin's work had appeared and the Metzler-Patinkin general equilibrium approach to the closed macroeconomy provided a more classical full-employment counterpart to the standard IS-LM framework. It was around this time that I moved on from writing about the pure classical model and started to think about the way to write down the general equilibrium equations for an open economy taking into account monetary variables, exchange rates, and capital movements. The facts that Canada had a flexible exchange rate and capital flows between Canada and the United States were significant background influences, but there was absolutely no model that was capable of dealing with the subject. I had a few fruitful conversations with Metzler that year that were important. His powers were much reduced after his brain surgery, however, and I remain convinced that had he remained healthy, he would have pioneered the international macroeconomic model. After Chicago, I returned to UBC for the year 1957-58. It was here that 1 presented the first discussion of "Optimum Currency Areas" (Mundell, 1961c) at a faculty seminar. That explains the North American flavor of the article. At the same time I wrote an expository piece for a government publication on macroeconomic developments in Canada, and this exercise led me into putting together the basic equilibrium equations for the open-economy macroeconomic model with capital mobility. I was still thinking along these lines when I left UBC for Stanford for the year 1958-59. It was at Stanford that the model really came together. I was teaching the graduate course in international economics and taught my new equations in it; Jeffrey Williamson probably remembers that class. Equally important was a faculty seminar, attended by Bernard Haley (editor of the American Economic Review), Ken Arrow, Lorie Tarshis, Ed Shaw, Melvyn Reder, and also Tibor Scitovsky and Abba Lerner who had come up from Berkeley. I had titled the talk "A Theory of Optimum Currency Areas," but most of it was the Mundell-Fleming model, and it

218

©International Monetary Fund. Not for Redistribution ON THE HISTORY OF THE MUNDELL-FLEMING MODEL made a big hit. Afterward, Lerner chided me for not talking enough about optimum currency areas, but I was able to give him the gist of the basic argument in a few minutes after the seminar. All these ideas were put in a single paper, tied together by the general equilibrium link. It included not only optimum currency areas but also much of the comparative statics of the Kyklos (Mundell, 1961b) and Canadian Journal (Mundell, 196la) papers, and some of the macrodynamics that became my Quarterly Journal of Economics article (Mundell, 1960a). I sent it to the Economic Journal and was disap- pointed when (later Sir Roy) Harrod rejected it—partly on the grounds that I hadn't referred to his treatment of some of the subjects. But the rejection turned out to be a blessing in disguise. It led to a much more sensible separation of the article into different parts, to become "Monetary Dynamics of International Adjustment Under Fixed and Flexible Exchange Rates" (Mundell, 1960a), "Optimum Currency Areas" paper (Mundell, 1961c), article on "Flexible Exchange Rates and Employment Policy" (Mundell, 1961a), and the Kyklos paper, "The International Disequilibrium System" (Mundell, 196 Ib). Ever since, I have suggested to students and colleagues the merits of a variant of Tinbergen's Rule: one idea, one paper. Later, when I became friends with Harrod, I teased him about his rejection of my paper, and he explained that he had been going through a very stressful situation at the Journal, sorting out a controversy between Harry Johnson and Don Patinkin over the definition of real marginal cost. He gave up the editorship soon after. It is necessary now to distinguish between two strains of my models. What is called the "Mundell-Fleming model" is usually taken to refer to that group of articles that includes Mundell, 1961b, 1961a, 1962, and 1963a-—that is, Chapters 15, 17, 16, and 8, of my International Economics (Mundell, 1968), including the appendix to Chapter 8, which was published in the Canadian Journal (Mundell, 1964). Also relevant is my article in the Banco Nazionale del Eavoro Quarterly Review, "The Nature of Policy Choice" (Mundell, 1963c). This article was, I think, the first fully developed global empirical model of the world economy in a Keynesian framework, a precursor of the forecasting models used by professional forecasting companies like Otto Eckstein's Data Resources and Laurence Klein's WEFA.3 One of the few references I've seen to this article is by Egon Sohmen in his paper "The Assignment Problem" in the Mundell-Swoboda book (Sohmen, 1969, pp. 183 and 186). These articles, usually thought about as the Mundell "half" of the Mundell-Fleming model, are more or less in the tradition of the inter- nationalized IS-LM model. It could also be thought of as an international multiplier model generalized to incorporate the securities and money markets.

Ill When I first heard the expression "Mundell-Fleming model," I supposed it included all my papers on international macroeconomics, including the first one. It took me some time before I realized that some economists did not count the model in "The Monetary Dynamics of International Adjustment Under Fixed and

3Originally, WEFA was an acronym for Wharton Econometrics Forecasting Associates.

219

Flexible Exchange Rates" (Mundell, 1960a, ch. 11), as part of the Mundell- Fleming model. Yet in some respects this first in the series was the most important and set the methodology for the others. Its purpose was to find a way to analyze the difference between an economy with fixed exchange rates and flexible prices, and an economy with flexible exchange rates with fixed prices. I needed a coherent and plausible international macroeconomic model that was consistent with a full-employment economy. There was no such model in the literature. The paper introduced an internal balance schedule for an open economy and a foreign balance schedule (for the first time in the literature). The variables were the interest rate (representing monetary policy) and the real exchange rate (or the relative prices of home and foreign goods). The comparative statics of the model could show the effects of expenditure changes on interest rates and the relative prices. The two schedules demar- cated four zones of disequilibrium, and this made possible an examination of the dynamics relevant to two different policy situations: an economy in which monetary policy was directed at fixing the exchange rate, compared with an economy in which monetary policy was directed at price-level stabilization—in modern language, the choice between exchange rate and inflation targeting. To me this formulation—the diagram with the FF and XX curves in a plane depicting the rate of interest on one axis and the real exchange rate (or some other relative price) on the other—fits the world of today better than the variable output versions. Of course it has to be updated to make a distinction between nominal and real interest rates, growth curves along the lines depicted in my Monetary Theory (Mundell, 1971) and a more explicit treatment of the relation between capital movements and domestic expenditure to produce Ohlin-type transfer effects. The model found a new application for economic dynamics. Meade, who was at heart a Marshallian, had not been concerned at all with dynamics. There were, of course, precedents in the dynamics. Samuelson had formulated the dynamics of the Walrasian system, and Lange, Metzler, Arrow, and others had added more theorems on its dynamic stability. Laursen and Metzler (1950) had analyzed flexible exchange rates, including a dynamic appendix, in the context of a multiplier model; Hicks had developed dynamics of trade theory. Metzler (1951) had an appendix on dynamics in his "Wealth, Savings and the Rate of Interest." Patinkin had followed in Metzler's footsteps. And Polak had analyzed some dynamics of an international general equilibrium model. But theorems about dynamic stability had not before been used to settle the choice between economic policy alternatives, and that was one of the novelties of my paper. When I started writing it, I had no idea what conclusions would emerge. I didn't make the model to elucidate or make appealing to the reader conclusions I had already reached by other means. I used the model as an engine of discovery. I wanted to find out what the mathematical dynamics of the model could teach me. To differ- entiate the dynamics of fixed and flexible rates, I used the same static model for both. The comparative statics of fixed and flexible exchange systems were essentially the same. But what about the dynamics? At first I thought that the different dynamics of the two systems (fixed and flexible rates) didn't really matter much.

220

From the diagrammatic analysis, it was apparent that the business cycle sequences were inverted. But why should that matter? Nevertheless, as a good student of Samuelson, I routinely derived the stability conditions for the two systems. It turned out that, under normal assumptions, both systems were stable. But that was not the end of it. It was with great excitement— and I remember the very moment on that Sunday afternoon in November 1958 in my Menlo Park apartment, just a month before the birth of my first son—that 1 noticed that while the stability conditions for fixed and flexible exchange rates were both satisfied, they were different. In particular, the terms under the discrim- inant determining whether the roots were real or imaginary were different. They could be positive or negative, giving rise to either asymptoticity or cyclicity in the path to equilibrium, depending on the sizes of some parameters or slopes. There suddenly spread before me now a whole new world of implications including the "principle of effective market classification." I was so taken with the idea—elated might be a better word—that I put pencil and paper down, to prolong the enjoy- ment of the suspense about what would, with a little more work, unfold. One implication of the model was that a domestic boom (shift up and right of the XX curve) would raise interest rates, attract capital inflows, appreciate the real exchange rate, and worsen the balance of trade, a conclusion that would hold under either fixed or flexible exchange rates. This was very relevant to an understanding of the economy of Canada, which was the only major country with a flexible exchange rate in the 1950s, and of course later very relevant for understanding the Reagan boom in the early 1980s and the German unification boom in the context of the exchange rate mechanism crisis in the early 1990s. Under the old Keynesian model, which typically assumed capital immobility, it was generally assumed that domestic expansion would weaken the currency. After the article appeared, I had a nice letter from Harry Johnson, saying something to the effect that it carried the subject to a different level.

IV In 1959-61, I taught at the Johns Hopkins School of Advanced International Studies Bologna Center, where I finalized several articles for publication: "The Pure Theory of International Trade" (Mundell, 1960b), "Optimum Currency Areas" (Mundell, 1961c), the Kyklos article (Mundell, 1961b), and the Canadian Journal article (Mundell, 196la). I spent two years in Bologna and thought it was time to get back into the mainstream. The offer from the International Monetary Fund was particularly appealing. When I came to the Fund in September 1961, Marcus Fleming, chief of the Special Studies Division in the Research Department, was away, and Jacques Polak, head of the department, suggested that I work on a problem that had come up in economic policy circles in the United States. There was a great debate going in the U.S. government about the use of monetary and fiscal policy, with different approaches suggested by the Chamber of Commerce, the Council of Economic Advisers, and the Keynesians. The Keynesians wanted expansionary monetary and fiscal policies; the Chamber of Commerce wanted tight monetary and fiscal policies; and the Council of Economic Advisers (CEA),

221

©International Monetary Fund. Not for Redistribution Robert Mundell strongly influenced by Paul Samuelson (President Kennedy's first choice as chairman of the CEA) and James Tobin, a member of the CEA, wanted to use monetary and fiscal policy in different directions, with low interest rates to spur growth and a budget surplus to siphon off the excess liquidity. The theory behind the policy mix was called the Samuelson-Tobin "neoclassical synthesis." When Polak asked me to work on this problem, I replied, "But I already solved that problem in my Kyklos article." Polak replied that "not enough people have got the message" and that I should try again. So I took up what was essentially a selling job. The problem was to make the case succinctly, and I hit on the idea of using the two equations representing policy goals—internal and external balance—in target space, with monetary policy on one axis and fiscal policy on the other. Thus was born "The Appropriate Use of Monetary and Fiscal Policy for Internal and External Stability" (Mundell, 1962). I wrote it in a week, and it was on Marcus Fleming's desk when he returned from his vacation. David Meiselman, then working in the Office of the Comptroller of the Currency, came over to the Fund to introduce himself and asked what I had been working on. I told him and he asked me what I thought of what I had written. I said that I felt like Bizet after he had written the Toreador Song to Carmen: "If it's trash they want, I'll give it to them!" Fleming approved the paper, and it circulated as a departmental memorandum, which meant that it went to the governments of all the member countries, but most important, of course, to the U.S. government. It was an immediate candidate for publication in the IMF's Staff Papers, but it created quite a fuss. All kinds of objections to it were made: It was "contrary to U.S. policy," it would have a "bad influence on developing countries," there was "no difference between monetary and fiscal policy," the "use of monetary and fiscal policy in opposite directions would cancel out," and so on. Graeme Dorrance, on the editorial board, told me he was initially against it for Staff Papers, but when he heard the other objections, he changed his mind. What saved it for Staff Papers was that the editorial board couldn't reach agreement on reasons for rejecting it. The article provided a new way of thinking about macroeconomic policy. At first it wasn't popular. This was to be expected because it recommended a complete reversal in the prevailing policy mix. The Samuelson-Tobin neoclassical synthesis might have had some merits in a closed economy, but it was completely indefensible in an open economy on fixed exchange rates. Fortunately for the United States (and me), President Kennedy reversed the policy mix to that of tax cuts to spur growth in combination with tight money to protect the balance of payments. The result was the longest expansion ever (up to that time) in the history of the U.S. economy, unmatched until the Reagan expansion of the 1980s. Meanwhile, however, the Federal Reserve Board of Governors had mounted an attack on my paper. Herbert Furth (Gottfried Haberler's brother-in-law) and Robert Solomon wrote a sharp critique. Instead of answering it point by point, I wrote the'^Canadian Journal paper (Mundell, 1963b) that is usually cited as the locus classicus of my half of the Mundell-Fleming model.

222

In my IMF paper, monetary policy had a comparative advantage in correcting the balance of payments. The critical assumption was that capital flows were responsive to interest rates. I decided to reply to the Federal Reserve critique by upping the ante, assuming complete capital mobility. This made the opposite policy mix even more absurd, because it showed that under fixed rates and perfect capital mobility, monetary policy was completely impotent. Open market operations to buy Treasuries would result in equivalent gold losses or buildup of dollar balances. The paper was presented at the spring meetings of the Canadian Economic and Political Science Association in Quebec, and published in the November 1963 issue of the Canadian Journal (Mundell, 1963b). This is the article that, as I said, has been so frequently reproduced and is usually cited in the Mundell-Fleming literature. A critical comment on it published the following year provoked me into extending the model to the two-country global context.

V Meanwhile, Marcus Fleming had been writing his paper, "Domestic Financial Policies Under Fixed and Flexible Exchange Rates," published in Staff Papers (Fleming, 1962). This article was later published again in his collected papers on international economics, just following a paper written in 1958 on "Exchange Depreciation, Financial Policy and the Domestic Price Level" (Fleming, 1958). The 1958 paper is entirely in the Bickerdike-Robinson-Metzler-Meade tradition and shows no traces of what came to be called international macroeconomics. But his 1962 paper is an almost fully mature international macroeconomic model, and this constitutes Fleming's contribution to the Mundell-Fleming model. The question arises as to the relation between the two models. He had probably been working on his model before I arrived at the Fund, and of course my papers owed nothing to his. He had certainly read my Quarterly Journal of Economics (Mundell, 1960a), Kyklos (Mundell, 1961b), and Canadian Journal (Mundell, 1961a) papers, as well as the paper on the policy mix (Mundell, 1962) that I wrote at the Fund and that he approved. When he was putting the finishing touches on his own paper in the spring of 1962, he asked me which of my articles I thought he should refer to. I said, why not them all? But he said, "No, I am only going to refer to one of them." That's exactly what he did. Curiously, he chose the least relevant article to his or my topic—my 1961 "Employment Policy and Flexible Exchange Rates" (Mundell, 196la). Even more curiously, he repeated the reference to this paper alone years later when he published his article on "Wider Exchange Margins" as Chapter 13 in his collection, Essays in International Economics (Fleming, 1971). What must have been going through his mind to single out that paper (which showed that commercial policy was ineffective or counterproductive under flexible exchange rates but no capital mobility) as the most relevant of my papers on monetary and fiscal policy? There is a difference between our articles that gets Marcus into trouble. On the second page of his article, he examines the effect of an expansionary shift in fiscal policy in the form of an increase in public expenditure under (1) fixed and (2) flexible exchange rates. The increase in expenditure leads, he says, to a

220

©International Monetary Fund. Not for Redistribution Robert Mundell deterioration in the current account. Then he writes: "In order to isolate the effect of a change in budgetary policy, it is necessary to assume that monetary policy remains, in some sense, unchanged. In this essay, that is taken to mean that the stock of money is held constant. . . ." But this assumption is not consistent with fixed exchange rates. As I showed in my Kyklos paper (Mundell, 1961b), sterilization policy is incompatible with fixed exchange rates and leads to a "disequilibrium system." Here is the problem. With a stock of money constant, the increase in government expenditure will increase interest rates, which will check expenditure and lead to an increased net capital inflow. While the trade balance worsens, the capital account improves, and this means that the balance of payments may improve or worsen depending on certain coefficients (in my framework, it will worsen or improve depending on whether the LL curve has a flatter or steeper slope than the FF curve). Fleming now has to conclude with ". . . if the policy of budgetary expansion results in a deterioration of the balance of payments, shortage of reserves may ultimately lead the authorities to abandon the policy and to renounce the associated expansion in income and employment." His system has no mechanism of adjustment for the balance of payments. In my earliest works on the model I identified monetary policy with interest rate policy. That was certainly true in my Canadian Journal paper (Mundell, 196la) and probably explains why Marcus chose that paper to refer to. It makes a starker contrast between our models. Later, however, when I made the assumption of perfect capital mobility, monetary policy had to be redefined and was correctly treated as an open market operation, or a change in domestic credit. The money supply is an endogenous variable under fixed exchange rates. In my Kyklos paper (Mundell, 1961b), I showed that the balance of payments can be kept in disequilibrium under fixed exchange rates only if automatic effects of reserve changes on the money supply are sterilized, a temporary solution. Had Fleming used constant domestic assets (no open market operations) as the criterion of a constant monetary policy, he would have been able to complete his analysis of the effects of an increase in government expenditure.

VI I am not quite sure when the term "Mundell-Fleming model" first appeared in the literature. At a conference in March 1997 in Claremont, California, I was objecting to the use of the misleading term "Marshall-Lerner condition," a term that originated with Charles Kindleberger. The relevant Marshall here is the writer of the Pure Theory of Foreign Trade (Marshall, 1879), and Lerner refers to the Economics of Control (Lerner, 1944). Marshall had, of course, died (1924) several years before Lerner became an economist (early 1930s), and their themes were quite different. Marshall was talking about changes in relative prices (the terms of trade), while Lerner was talking about the exchange rate. Marshall would have been absolutely horrified at the connection, when he took such careful pains to distinguish between the terms of trade and the exchange rate and to reject any hint of a connection between the stability of his barter model (based on Mill) and the stability of

224

©International Monetary Fund. Not for Redistribution ON THE HISTORY OF THE MUNDELL-FLEMING MODEL exchange rates. He explicitly made clear that the reader should not confuse the exchange rate with the terms of trade. Max Corden then asked me why, if I objected to that connection, did I object to the name "Fleming-Mundell" model rather than "Mundell-Fleming model." I pointed out what I have said above—that his work, if not dependent on mine, at least followed mine, whereas mine was completely independent of his. He had read my earlier papers. That was one of the reasons he wanted me to come to his division in the Fund. I am not suggesting Fleming's work wasn't in an important sense independent of mine. It was certainly to a large extent subjectively (to use Schumpeter's phrase) original. You can see a connection in his model to a paper he wrote on macroeconomics in the late 1930s, analyzing a closed economy in a quasi-general equilibrium framework. The problem was already "in the air" at the Fund, and it was natural that he would have tried his hand at solving it when it had become such a bone of contention in the United States. The assumptions, style, and notation are characteristic of Fleming. The notation is completely anti-mnemonic. Marcus Fleming was a gifted and original economist. He was a "purist" in many senses. Sometimes this trait, combined with his integrity, would get in the way. When he was working at the U.K. Treasury in the 1940s, he was aghast, Lionel Robbins told me, to find that the government was accepting the Treasury's recommendations for the wrong reasons. He would rather be right than president. He could be exasperating to people in his division. A couple of stories, called up from the far recesses of the mind, can be mentioned. 1 used to go into the office quite early and stay late, partly to avoid the rush hour. But for an hour or two after lunch I was not to be seen. I was at the nearby Washington Athletic Club. Long after I left the Fund, Ann Romanis told me that Marcus would frequently come to see me after lunch and get in a frightful stew when I was not to be found. At the same time, Ann would come into my office tearing her hair after an intensive discussion with Marcus, usually about "incomes policy." Despite his predisposition for precision, Marcus was a Keynesian. In the spring of 1963, I presented my "return to the classics" paper, "Barter Theory and the Monetary Mechanism of Adjustment" (Mundell, 1963a) at a Fund seminar. This paper would later start a kind of Mundell-Dornbusch literature. It was then that Fleming made his humorous comment that the only two Keynesians left at the Fund were himself and the managing director (Per Jacobbsen). Marcus was best at developing and refining fine points and details in abstract theory rather than in the rough-and-tumble and necessarily inexact world of forging new systems. He really disliked that paper, and in his written comments on it, he penciled in "lament for economics." It never saw the light of day as a Fund paper, and Marcus had the chance to critique it in detail (but unsuccessfully) when he was its discus- sant at the 1965 World Bank Conference where I first presented it outside the Fund. It is interesting to note that the literature that came from that paper thus also originated at the Fund, as did my earliest Journal of Political Economy papers on inflation theory. There was no Mundell-Fleming paper. We never collaborated on macroeconomics. But there is a Fleming-Mundell paper, "Official Intervention on the

225

Forward Exchange Market," published in Staff Papers (Fleming and Mundell, 1964). Marcus wrote the first draft of this paper and it was his idea to treat the forward market as a stock, rather than a flow, market. It's a great idea, and it's a pity the article has been somewhat neglected. I developed the diagrams and the explanations. In the exchanges between us, relating to our two-country framework, I replaced his "A" and "non-A" with "A" and "5." We went through this exchange a couple of rounds, but he had the last word. That was my first and (almost) last experience with collaboration. I am proud of the fact that our names will be linked together in the Mundell- Fleming Lecture to commemorate a very exciting and fruitful period of interaction and collaboration in the Fund.

REFERENCES

Alexander, Sidney S., 1952, "Effects of a Devaluation on a Trade Balance," Staff Papers International Monetary Fund, Vol. 2 (April), pp. 263-78. Fleming, J. Marcus, 1958, "Exchange Depreciation, Financial Policy and the Domestic Price Level," Staff Papers, International Monetary Fund, Vol. 6 (April), pp. 289-322. , 1962, "Domestic Financial Policies under Fixed and Floating Exchange Rates," Staff Papers, International Monetary Fund, Vol. 9 (November), pp. 369-79. , 1971, Essays in International Economics (London: Allen and Unwin). -, and Mundell, Robert A., 1964, "Official Intervention on the Forward Exchange Market: A Simplified Analysis," Staff Papers, International Monetary Fund, Vol. 11 (March), pp. 1-19. Laursen, Svend, and Lloyd A. Metzler, 1950, "Flexible Exchange Rates and the Theory of Employment," Review of Economics and Statistics, Vol. 32 (November), pp. 281-99. Lerner, Abba P., 1944, The Economics of Control; Principles of Welfare Economics (New York: Macmillan). Marshall, Alfred, 1879, "The Pure Theory of Foreign Trade," paper reprinted in The Pure Theory of Foreign Trade. The Pure Theory of Domestic Values (Clifton, New lersey: A. M. Kelley, 1974). Meade, lames E., 195 la, The Balance of Payments (London; New York: Oxford University Press). , 195 Ib, The Balance of Payments: Mathematical Supplement (London; New York: Oxford University Press). Metzler, Lloyd A., 1951, "Wealth, Savings and the Rate of Interest," Journal of Political Economy, Vol. 59, No. 2, pp. 93-116. Mundell, Robert A., 1957a, "International Trade and Factor Mobility," American Economic Review, Vol. 47 (June), pp. 321-35. , 1957b, "Transport Costs in International Trade Theory," Canadian Journal of Economics and Political Science, Vol. 23 (August), pp. 331-48. , 1960a, "The Monetary Dynamics of International Adjustment Under Fixed and Flexible Exchange Rates," Quarterly Journal of Economics, Vol. 84 (May), pp. 227-57. , 1960b, "The Pure Theory of International Trade," American Economic Review, Vol. 50 (March), pp. 68-110.

226

, 196la, "Flexible Exchange Rates and Employment Policy," Canadian Journal of Economics and Political Science, Vol. 27 (November), pp. 509-17. , 1961b, "The International Disequilibrium System," Kyklos, Vol. 14, No. 2, pp. 154-72. , 1961c, "A Theory of Optimum Currency Areas," American Economic Review, Vol. 51 (November), pp. 509-17. , 1962, "The Appropriate Use of Monetary and Fiscal Policy for Internal and External Stability," Staff Papers, International Monetary Fund, Vol. 9 (March), pp. 70-79. , 1963a, "Barter Theory and the Monetary Mechanism of Adjustment," later published as Chapter 8 of Mundell (1968). Also published as "International Disequilibrium and the Adjustment Process," in Capital Movements and Economic Development: Proceedings of a Conference held by the International Economic Association, ed. by John H. Adler with the assistance of Paul W. Kuznets (London: Macmillan; New York: St. Martin's Press, 1967), pp. 441-68. , 1963b, "Capital Mobility and Stabilization Policy Under Fixed and Flexible Exchange Rates," Canadian Journal of Economics and Political Science, Vol. 29 (November), pp. 475-85. , 1963c, "The Nature of Policy Choice," Banca Nazionale del Lavoro Quarterly Review, Vol. 66 (September). , 1964, "A Reply: Capital Mobility and Size," Canadian Journal of Economics and Political Science, Vol. 30 (August), pp. 421-31. , 1968, International Economics (New York: Macmillan). 1971, Monetary Theory: Interest, Inflation and Growth in the World Economy (Pacific Palisades, California: Goodyear). Sohmen, Egon, 1969, "The Assignment Problem," in Monetary Problems of the International Economy, ed. by Robert A. Mundell and Alexander K. Swoboda (Chicago, Illinois: University of Chicago Press).

227