15832-9781462332373.Pdf

IMFscoff papers Robert Flood Editor and Committee Chair Ratna Sahay and Eswar S. Prasad Co-Editors Jeff Hayden Assistant Editor Alvaro Rojas Research Assistant Sheila Kinsella Administrative Coordinator

Editorial Committee Brian Aitken Atish R. Ghosh Tamim Bayoumi Laura Kodres Sharmini Coorey Donald J. Mathieson Tito Cordelia Peter Montiel Liam Ebrill

The objective of IMF Staff Papers is to publish high quality research produced by IMF staff and invited guests on a variety of topics of interest to a broad audience including academics and policymakers in the member countries of the Fund. The papers selected for publication in the journal are subject to an extensive review process using both internal and external referees. IMF Staff Papers also welcomes outside comments, criticisms, and interesting replica- tions of published work. The views presented in published papers are those of the authors and do not necessarily reflect the position of the Executive Board or of the IMF.

Subscription: $U.S.54.00 a volume or the approximate equivalent in the currencies of most countries. Three issues constitute a volume. Single copies may be purchased at $18. Individual academic rate to full-time professors and students of universities and colleges: $27 a volume. Subscriptions and orders should be sent to:

International Monetary Fund Publication Services 700 19th Street, N.W. Washington, D.C. 20431, U.S.A.

Telephone: (202) 623-7430 Fax: (202) 623-7201 E-mail: [email protected] Internet: http://www.imf.org

©International Monetary Fund. Not for Redistribution International Monetary Fund

Volume 46 Number 3 IMFstaffpapers September/December 1999

©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- 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 International Monetary Fund

Volume 46 Number 0 Contents September/December 1999

Determinants and Leading Indicators of Banking Crises: Further Evidence Daniel C. Hardy and Ceyla Pazarbasioglu • 247

Time Series Analysis of Export Demand Equations: A Cross-Country Analysis Abdelhak S. Senhadji and Claudio E. Montenegro • 259

The Uzbek Growth Puzzle Jeromin Zettelmeyer • 274

Monetary Policy and Public Finances: Inflation Targets in a New Perspective Christian H. Beddies • 293

Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union Giovanni Dell'Ariccia • 315

Index of Volume 46 • 335

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

Determinants and Leading Indicators of Banking Crises: Further Evidence

DANIEL C. HARDY and CEYLA PAZARBASIOGLU*

This paper examines episodes of banking system distress and crisis in a large sample of countries to identify which macroeconomic and financial variables can be useful leading indicators. The best warning signs of the recent Asian crises were proxies for the vulnerability of the banking and corporate sector. Full-blown banking crises are shown to be associated more with external developments, and domestic variables are the main leading indicators of severe but contained banking distress. [JEL: E44, G21]

ecent events in East Asia have reminded the world of how rapidly and with what R disruptive force banking crises can erupt, and of how difficult it is to foresee the timing and full ramifications of these dramatic events. Yet financial crises have a long history, and in recent decades many countries have experienced financial sector distress of various degrees of severity, and some have suffered repeated bouts (Lindgren, Garcia, and Saal, 1996, provide a listing and discussion). This history lends importance to the identification of conditions under which banking crises are likely to occur so as to preempt them or prepare for their resolu- tion. In this paper we concentrate on finding robust coincident and leading indicators that might be available in most countries. Since plausibly the causes of banking system distress differ across economies with different structural characteristics, leading indicators are differentiated by region. In particular, the recent Asian crises are shown to differ in several regards from episodes elsewhere. Furthermore, banking

*Daniel Hardy and Ceyla Pazarbasioglu were in the Monetary and Exchange Affairs Department (MAE) when this paper was written. The authors wish to thank William E. Alexander, Timothy Lane, Sunil Sharma, and participants at an MAE seminar for helpful comments and suggestions. Research assistance by Jahanara Begum and Kiran Sastry is greatly appreciated.

247

©International Monetary Fund. Not for Redistribution Daniel C. Hardy and Ceyla Pazarbasioglu sector difficulties may also differ greatly in severity: some may be categorized as severe distress and others as full-blown crises. Results are presented showing that the precursors of crises and coincident economic developments are rather different from those of severe but limited financial system distress. This study is a contribution to the new but growing body of research that attempts to evaluate econometrically the economic precursors and causes of banking sector weakness or crisis. Some studies, such as Cole and Gunther (1995) for the United States and Gonzalez-Hermosillo, Pazarbasioglu, and Billings (1997) for Mexico, have included as explanatory variables primarily bank-specific variables, and looked at the experience of individual institutions. These results are difficult to generalize, however, because for many countries reliable bank-specific data are rarely available to the more general public on a timely basis, if at all, and so cannot be used to make predictions. Gonzalez-Hermosillo (1999) represents an attempt to overcome some of these limitations. Another and more recent group of studies, to which this paper belongs, focuses primarily on macroeconomic variables and other indicators that are available in most countries on a fairly timely basis. A pioneering work in this area is the study by Kaminsky and Reinhart (1996), which examines the behavior of various macroeconomic indicators during episodes of both banking and currency crises. A paper by Demirguc-Kunt and Detragiache (1998) in this journal, which was written concurrently with the research reported here, examines the determinants of the probability of a banking crisis using annual, macroeconomic data. Their sample includes 31 instances of what are judged to be full-fledged banking crises, rather than more moderate distress. They find that low GDP growth, excessively high real interest rates, and high inflation significantly increase the likelihood of systemic problems. They also find weak evidence that adverse terms of trade shocks and rapid credit growth increase the probability of a banking crisis. The size of the fiscal deficit and the rate of depreciation of the exchange rate do not seem to have an independent effect. An interesting finding is that structural characteristics, such as the availability of deposit insurance and the degree of "law and order" achieved by a country, are also relevant. In addition, Demirguc-Kunt and Detragiache present results, albeit from a very small sample, on the determinants of the cost of resolving banking crises.1 Demirguc-Kunt and Detragiache use almost exclusively contemporaneous variables on the right-hand side (only a measure of the growth in bank credit is lagged two periods), and therefore, as the authors acknowledge, the direction of causality is not always unambiguous. By the same token, their findings are of only limited usefulness in predicting crises in advance. On a more methodological issue, their emphasis on coincident indicators hampers the identification of dynamic features of the lead-up to banking crises, such as cyclical turning points. Nor do they distinguish periods in which banking sector difficulties may be incubating but have not yet reached crisis levels from more normal periods of economic activity.

1Eichengreen and Rose (1998) is another related paper, which concentrates on the influence of world- wide economic trends on the incidence of banking crises in developing countries.

248

©International Monetary Fund. Not for Redistribution DETERMINANTS AND LEADING INDICATORS OF BANKING CRISES

Furthermore, they apply a common methodology to the full sample, and do not differentiate by region or severity of banking crisis. Our research complements that of Demirguc-Kunt and Detragiache by addressing these issues.

I. Specification, Estimation, and Sample Definition Our approach to the empirical investigation of banking crises is dictated by the goals set out above.2 The main subject used in estimation will be an indicator, or dummy, variable (denoted y0) taking on the value 2 in a period when banking sector difficulties emerged, 1 in the preceding period, and zero otherwise. The approach of treating the pre-crisis year and the crisis year as separate events has several advantages. First, in many countries intervention defines the start of the crisis, but often the difficulties might have been widely known and been the cause of serious disruption for some time before then. Thus, economic behavior in the run-up to the declared start of an episode may differ significantly from that in more normal times, and the differences may themselves be of interest.3 Second, this approach, rather than using just the crisis as the dependent variable and including lagged values of the explanatory variables, allows one to establish the predictive power of the leading indicators independently of what is known only in the crisis year, and provides a rough indication of the time to crisis. Results will also be reported for an indicator variable (denoted y1) that takes on the value 2 at the start of a full-fledged banking system crisis, 1 at the start of an episode of severe but limited banking system distress, and zero otherwise. Results for this variable will suggest how the determinants of crises differ from those of more contained episodes. The discrete indicator variables will be related to other, usually continuous economic series using a multinomial logit model estimated by maximum likelihood; details of the econometric procedure can be found, for example, in Greene (1990). The definition of a financial crisis, its severity, its onset, and its duration is a matter of judgement and debate. In this study the identification of episodes of banking system distress and their timing follows that provided in Table 2 (pp. 21-35) of Lindgren, Garcia, and Saal (1996). The sample includes all listed cases of crises or banking system distress for which adequate data were available, except for cases in the formerly socialist transition economies, which can be considered sui generis due to their exceptional historical circumstances. Countries suffering hyperinflation were also excluded because of the difficulty in measuring real variables during periods of very high and variable inflation. The experience of countries that have not recently experienced significant banking sector problems should also be relevant, because they constitute a kind of control group. Therefore, data on a number of such non-crises countries were also collected. The full sample eventually obtained covered 50 countries, 38 of which suffered a total of 43

2Details of the approach and some additional results are contained in Hardy and Pazarbasioglu (1998). 3Estimation was also performed for a dependent variable that identified separately crisis years and the two preceding years (i.e., a dummy variable with the values 0,1,2,3). However, finding any significant explanatory variables singling out the periods two years before crises was difficult.

249

©International Monetary Fund. Not for Redistribution Daniel C. Hardy and Ceyla Pazarbasioglu episodes of banking system crisis or significant problems (23 instances of severe problems and 20 crises). At least eight, and usually more, years of annual data on the explanatory variables were obtained from the IMF's International Financial Statistics for each country for each of the explanatory variables, so the sample comprised 323 observations (253 from crisis countries). Most explanatory variables are included in first difference form, and all variables except where noted are in logs and differences (denoted by a prefix D in the acronyms). The prefix Ln denotes the n-th lag relative to that observation. The list of candidate explanatory variables was inspired by the existing empirical and theoretical literature on banking crises, concentrating on those that are widely available on a timely basis. These variables can be split into three groups. The first group relates to the real sector in an attempt to capture the degree of efficient use of credit as well as changes in the repayment capacity of borrowers, and includes the real growth rates of GDP (DRGDP), private consumption (DRPCN) and investment (DRFCF). The incremental capital output ratio (ICOR) is used as a proxy for efficient use of investment. A sharp increase in this ratio may imply the emergence of over-investment in aggregate or in certain sectors such as real estate. The second group of indicators relates to banking sector variables. These include the change in the deposit liabilities of the banking system as a percent of GDP (DRBDL), which may indicate the existence of deposit runs and a loss of confidence in the banking system, or of the shrinkage of banks' balance sheets for other reasons. The growth in the ratio of total bank credit to the private sector to GDP (DRBCP) reflects how extended is the banking sector. The change in the ratio of gross foreign liabilities of the banking system to GDP (DSGFL) is used as a measure of the banking system's reliance on foreign capital to fund its operations, and thus is a proxy for its vulnerability to a sudden withdrawal of capital inflows. The third group includes shocks that may directly or indirectly (through the real sector) affect the health of the banking sector, or which may indicate the advent of such a shock. These include the inflation rate (specifically, the GDP deflator, DPGP), the real deposit interest rate (DRDIR), changes in the real exchange rate (DERR), the growth of imports in real terms (DRIMP), and terms of trade developments (DTOT). Higher real interest rates would likely hurt the nonfinancial corporate sector, in particular companies that are highly indebted. An adverse terms of trade shock and a real exchange rate appreciation may affect the competitiveness of the country and lead to a deterioration in corporate sector prof- itability. A subsequent correction, that is, a sharp depreciation of the exchange rate, may lead to losses for corporations (financial and nonfinancial) indebted in foreign currency. Several countries in the sample suffered repeated financial crises. Possibly, economic behavior will be permanently affected by a banking crisis and economic agents may behave differently when faced with such events a second time. Furthermore, repeated crises may indicate that inherent weaknesses in the banking sector were not adequately resolved. A dummy variable (RPTD) equal one in a repeat crisis and its lead-up, and zero otherwise, was used to capture this effect.

250

©International Monetary Fund. Not for Redistribution DETERMINANTS AND LEADING INDICATORS OF BANKING CRISES

A number of what will be termed "regional" variables were defined. These were constructed by multiplying the macroeconomic explanatory variables with dummy variables that identified the region to which a country belongs (for example, the "Asia dummy" equals unity for Korea, Thailand, Indonesia, etc.). Attention focused on the Asian and African regional variables, which will be denoted by the suffixes A and B, respectively.4 Of course, when such variables are included in the specification, the "nonregional" variables explain events in the remaining countries. A specification search was undertaken to eliminate insignificant terms, starting from a very general specification containing up to two lags of the candidate explanatory variables. The risk of omitted variable bias, and the presence of multicollinearity suggest that variables on the border of significance should not be excluded. However, the dependent variable contains a preponderance of "zeros," that is, the proportion of non-zero terms is low. The danger exists that particular right-hand side variables serve to "explain" only one or two episodes, and results will be spurious or not robust. Hence, parsimony is important. The final specification of the regression equations was determined so as to balance these considerations.

II. Empirical Results Table 1 contains the summary statistics, estimated parameters, and standard errors for the dependent variables y0 and y1. The first two sub-columns report the results for y0 using the same explanatory variables for all countries, and the second pair of sub-columns contain the results taking into account regional effects. Reviewing the results shows that reasonable predictive power has been obtained. For example, when the specification for y0 including regional variables is estimated, more than half of the episodes of banking system distress are predicted correctly, and about one-third of the pre-crisis periods are identified correctly or as a crisis period.5 Predictive power for crisis years (y0 = 2) is usually somewhat better than for pre- crises years (y0 =1), largely because in the former case several contemporaneous variables (such as the change in the real effective exchange rate, DERR) are highly significant. A visual impression of the ability of the model to differentiate crisis, pre- crisis, and calm periods can be obtained from Figure 1, where the estimated probabilities of y0 = 2 and y0 = 1 are plotted. For most countries an upward "spike" in these probabilities in the crisis and pre-crisis years is apparent. The y0 specification excluding regional variables was estimated over a sample that omits four recent East Asian crises (detailed results are available upon request). The estimated coefficients are robust to this change, except that the estimated coefficient on the real effective exchange rate term is somewhat larger in the full sample.

4Eichengreen and Rose (1998) concentrate on banking crises in developing countries, arguing that such crises differ qualitatively from those in industrialized countries. We prefer to single out the newly industrialized countries in Asia and the mostly primary product exporting countries of Africa. 5In a few crisis or pre-crisis years, the estimated probability of y0 = 0 is larger than that of each of the other two possibilities, but still less than 50 percent. Hence, the model predicts either y0 = 1 or y0 = 2 in 41 out of 86 instances where this is the case. Conversely, it predicts either y0 = 1 or y0 = 2 in 14 of 167 instances where in fact y0 = 0.

251

©International Monetary Fund. Not for Redistribution Dah\e\ C.V\avdy and Ceyta Pazarba^og^u

Table 1. Estimation Results

Dependent variable y0 vO : (Excluding (Including (Including regional variables) regional variable regional variables) No. of observations 323 323 323 Constrained log-likelihood -246.79 -246.79 -156.41 Max. log-likelihood -198.53 -169.43 -94.321 Predictions l yj=01 227/5/5 226/7/4 276.921 Predictions \yj-l 31/10/2 28/12/3 16/5/2 Predictions \yj = 2 30/0/13 20/0/23 9/0/11 y0 = 1 y0 = 1 y0 = 2 y0 = 1 y0 = 2 >0 = 2

Explanatory variable

Constant . 118 -2.152 -2.336 -2.623 -3.613 -3,595 0.335** 0.381** 0.392** 0.502** 0 .635** 0.771** DRGP -6.438 -14.585 -8.048 14.303 =5.865 -22.486 4.149 4.306** 4.305+ 4.824** 3.167 6.824** LDRPCN ... 6.562 ... 80610 16.331 8.722 5.017 5.725 6.833* 8.584 0.028 LICOR 0.019 ...... 0.014 0.027 L2IC0R ... 0.019 ... 0.009 0.016 0.027 0.014 0.030 01 0.025 DPGP .453 ... -10.731 ...... 3109** ? •?3.356** LDPGP 10.992 -7.896 12.852 -10.955 -11.185 -9324 2.992** 3.477* 3.235** 3.967** 4.583* 6.593 L2DPGP ... 9.253 ... 14.671 14.770 8.088 3.057** 3.703** 4.201** 5.740 DRBDL 213 -3.626 -4.092 -4.857 -4.335 -0.466 2.110* 2.341 2.281 2.624+ 3.239 3.515 LDRBDL ... 1.578 ... -0.839 -1.437 2.307 1.476 1.793 1.987 3.194 DRBCP -1.526 -3.846 -2.658 -4.329 -4.136 -1.582 1.942 2.064 2.129 2.22 7 2J29** 2.9J2 LDRBCP 1.423 ... 2.066 ... . . , ... 1.467 1.481 L2DRBCP ... 2.262 ... 2.871 3.500 0.915 1.452 1.7A 1J94+ 3.750 DRDIR ... 0.064 ... 0.1( 0.097 0.028 0.026* 0.0 0.038* 0.047 LDRD1R 0.045 ... 0.054 ...... 0.029 0.030+ L2DRDIR 0.600 0.030 0.061 0.057 0.063 -0.010 U.UZJ 0.026 u.uzo 0 0 s 0.030* 0.039

252

©International Monetary Fund. Not for Redistribution DETERMINANTS AND LEADING INDICATORS OF BANKING CRISES

Table 1 (concluded)

y0 y0 y0 (Excluding (Including (Including regional variables regional variables) regional variables) y0 = 1 y0 = 2 y0 = 1 y0 = 2 y1 = 1 y1 = 2 Explanatory variable DERR -2.099 -7.215 -3.796 -6.223 -3.605 -8.155 2.290 1.899** 2.660 2.500* 3.266 3.144** LDERR 4.567 ... 3.630 ...... 2.392+ 2.420 L2DERR ... 4.357 ... 2.133 1.167 2.627 2.419+ 2.690 3.295 4.409 DRFGL -7.765 ... -9.685 ...... 6.782 7.725 LDRGFL 10.241 —7.456 3.673 —7.065 -9.284 -15.758 7.170 7.707 8.790 10.747 13.533 13.480 L2DRGFL ... 16.064 ... 10.703 7.278 22.278 8.886+ 11.454 14.886 15.908 DRIMP -1.028 ... -1.829 ...... 1.402+ 1.751 LDRIMP ... -1.058 ... -4.465 -4.396 -6.321 1.351 1.965* 2.322+ 3.398+ RPTD 1.191 1.130 0.850 1.040 0.653 1.259 0.613+ 0.725 0.664 0.888 1.024 1.426 ADERR ...... 19.421 -22.482 -27.463 -29.477 8.683* 12.372+ 12.615* 13.053* ALDERR ...... 35.048 35.932 30.259 10.775** 12.083** 14.285 ADRGFL ...... -91.762 -92.661 -86.384 33.824**35.410 43.821 ALDRGFL ...... 30.595 ...... 16.830+ AL2DRGFL ...... 38.68938.104 26.424 21.948+25.967 31.605 BDPGP ...... 20.52721.878 19.009 9.151* 9.186* 9.970+ BL2DPGP ...... -23.227-25.317 -26.107 11.428* 12.363 15.241 BDTOT ...... -10.210 ...... 4.033* BDLTOT ...... -9.148 -12.585 -5.206 4.298* 5.552* 5.737 BDRIMP ...... 3.727 ...... 3.211+ BLDRIMP ...... 7.967 9.272 2.382 3.534* 4.007* 6.067 Standard errors in italics; **; significant at 1 percent. *: significant at 5 percent. +: significant at 10 percent.

1Under "Predictions | yj = i" are reported the number of ovservations when the model predicts yj = 0, yj = 1, and yj = 2, respectively, when in fact yj = i, for i = 0, 1, 2, j = 0, 1.

253

©International Monetary Fund. Not for Redistribution 25 4

Figure 1. Estimated Probabilities for yO

i prob (yO = 1) • prob (yO = 2) | crisis and pre-crisis years Danie l C . Hard y an d Ceyl a Pazarba§iogl u

International Financial Statistics country code

©International Monetary Fund. Not for Redistribution DETERMINANTS AND LEADING INDICATORS OF BANKING CRISES

Three of the four East Asian crises are correctly identified out of sample. In only one of these cases was the pre-crisis period identified, however, confirming the impression that these crises were not preceded by typical macroeconomic disturbances. An examination of the lag structure of the estimated equations reveals several regularities. First, the explanatory variables for y0 = 2 (except for contemporaneous terms) tend to be lagged one period relative to those for y0 = 1, which is as one would expect. Second, a number of explanatory variables display a "boom and bust" pattern, with a large positive coefficient lagged one or two years, and a large negative coefficient in the crisis or pre-crisis year. This pattern, which accords with some of the proposed explanations of banking crises, applies to inflation, credit growth, the real effective exchange rate, and banks' gross foreign liabilities. In some instances the interval from "boom" to "bust" is at least two years. Third, variables capturing financial market prices (the real exchange rate and the real effective exchange rate) are the main contemporaneous indicators of banking crises; the variables measuring quantities, such as stocks of financial assets or GDP components, more often enter with a lag. The estimation results for individual explanatory variables largely corroborate the findings of others, including Demirguc-Kunt and Detragiache. Among the first group of explanatory variables, banking distress is associated with a largely contemporaneous fall in real GDP growth, but for at least some countries the fall in GDP growth begins earlier, and this variable has some information content in predicting y0 = 1. The empirical findings also suggest that a consumption boom in the years preceding a crisis (LDPRCN) can be a leading indicator. The estimated coefficient on the lagged incremental capital output ratio (ICOR) is not significant at conventional significance levels, but including the variable improves predictive power, and the estimate is robust to changes in specification. Furthermore, the (positive) sign accords with the theory that overinvestment at decreasing returns often leads to a banking crisis. Turning to the banking-sector variables, deposits at banks (DRBDL) tend to start falling in real terms before a banking crisis is fully acknowledged, possibly due to declining confidence in the domestic banking system, and continue to fall during the crisis. This fall presumably contributes to liquidity problems in the banking sector. There is also a persistent and robust tendency for credit to the private sector (DRBCP and its lags) to follow a boom and bust pattern in advance of crises, with a further decline in credit growth during the crisis. The coefficients of the indicator used to capture the vulnerability of the banking system to private capital inflows (the change in the gross foreign liabilities of the banking sector relative to GDP, denoted by DRGFL) are sometimes significant and contribute to the predictive power for the model. They carry the expected sign, namely positive on a longer lag and negative as the crisis approaches. Among other variables, a rise followed by a sharp fall in inflation seems to be one of the most reliable early indicators of impending banking sector problems. Real interest rates (DRDIR) usually rise in the crisis year, and reliably tend to start increasing already in the preceding years.6 Banking crises are associated with a

6Unfortunately, a measure of interest rate spreads was not available for many countries over most of the sample.

255

©International Monetary Fund. Not for Redistribution Daniel C. Hardy and Ceyla Pazarbasioglu sharp decline in the real effective exchange rate, but an appreciation in this rate often precedes a crisis. A sharp slowdown in the real growth in imports is a good leading indicator of a financial crisis. This contraction may be symptomatic of a general economic slowdown and especially a decline in certain sectors, or of foreign exchange shortage. The estimate of the coefficient on the "repeat crisis" dummy variable is close to significant and relatively robust. Other candidate explanatory variables found not to be systematically significant for this sample included: real gross fixed capital formation, the current account balance, reserve money, credit from the monetary authorities, banks' reserves, banks' net foreign assets, and foreign exchange reserves (relative to imports or deposits). These variables often seem to contain useful information and to have predictive power when used in isolation, but statistical significance is lost when used in conjunction with the other explanatory variables.7 The inclusion of regional variables has a major effect on the estimates, even if most of the qualitative results are preserved. Indeed, some estimated coefficients become larger and more significant when the regional variables are included (e.g., on most of the interest rate terms, or the change in real GDP for y0 = 1); once certain regional factors are accounted for the indicative value of other variables becomes clearer. The importance of regional effects is demonstrated by the improvement in predictive power that is obtained through their inclusion. The banking crises in Asian countries are strongly associated with an appreciation followed by a sharp depreciation in the real effective exchange rate (DERR), and a parallel movement in the gross foreign liabilities of the banking sector (DRGFL). With this specification the estimated coefficients on these terms for the non-Asian countries are lower. These results are consistent with the weight given to capital inflows and real exchange rate movements in accounts of the recent Asian crises. Inclusion of the Asian regional variables also eliminates the significance of the "repeat crisis" dummy (RPTD), which largely serves to identify several of the recent Asian crises. The estimated coefficient on the "Asia dummy" itself (not cross-multiplied with another explanatory variable) was insignificant, however, suggesting that a pure regional reputation effect was small. The results for the African regional variables suggest that banking crises in that region were not closely linked to a rise and sudden fall in inflation or a slowdown in import growth. Rather, a deterioration in the terms of trade seems to have been a major contributing factor in these countries, many of which rely heavily on the export of primary commodities.

7Demirguc-Kunt and Detragiache find that a number of institutional features of the countries in their sample are significant determinants of banking crises. We instead estimated a fixed effects model using a technique from Chamberlain (1980) to capture all persistent institutional or structural differences between countries. The dependent variable took a value of unity at the onset of an episode of banking system distress, and zero otherwise, and all non-crisis countries had to be excluded from the sample. The fixed effects themselves were found to be always jointly highly significant, indicating that country-specific phenomena are indeed important. However, the estimated parameters on the other variables of interest were not greatly affected by the inclusion of fixed effects, and in some instances their significance increased. Detailed results are available on request.

256

©International Monetary Fund. Not for Redistribution DETERMINANTS AND LEADING INDICATORS OF BANKING CRISES

So far all cases of banking system distress have been considered without regard to how profound or pervasive they were, but it is obvious that they differ greatly in these respects, and possibly in their causes. An indication of the importance of these differences can be obtained by considering the estimated coefficients for y1 presented in the last two sub-columns of Table 1, albeit with the caveat that the relatively small number of each type of event may reduce the generality of the results. The differentiation between crisis episodes and those of significant banking system distress reveals important characteristics of these different phenomena. In particular, a decline in growth is an important factor explaining the crisis episodes, but it is not significant for the distress cases. Furthermore, credit expansion funded mainly by capital inflows and leading to over-investment seems to be a critical factor in the crisis cases (significant parameters for L2DRGFL and L2IC0R). Likewise, movements in the real effective exchange rate seem to have been more important in the crises countries. These findings suggest that certain external developments, in particular heavy reliance on external funds, magnify the impact of a negative shock to the system and constrain the policy response to banking system distress, leading to a full-blown crisis. The causation need not be only one way: a very severe banking system crisis may itself precipitate an exchange rate crisis. In contrast, credit expansion seems to have fueled consumption in the cases of significant banking system distress, where movements in the real interest rate on (domestic) deposits is a better indicator. The inclusion of regional variables if anything reinforce these results, implying that they are not merely due to the recent Asian crises.

III. Concluding Remarks This paper concentrates on the role of cyclical movements in macroeconomic, banking sector, and real sector indicators in the lead-up to banking system difficulties. Overall, the empirical findings suggest that banking distress is associated with a largely contemporaneous fall in real GDP growth; boom-bust cycles in inflation, credit expansion, and capital inflows; rising real interest rates and a declining incremental capital output ratio; a sharp decline in the real exchange rate; and an adverse trade shock. Certain of these tendencies seem to have been especially pronounced in the recent Asian crises, which were relatively difficult to predict using traditional macroeconomic indicators. More generally, the results presented are a reminder of the diversity of problems that come under the heading of banking system distress, and how country-specific circumstances need to be recognized in assessing the likelihood of such difficulties. The banking systems of the primary product exporting countries of Africa are vulnerable to a different range of disturbances than those of, say, the Nordic countries, and, as shown, the relevant leading indicators differ likewise. Furthermore, it is recognized in the paper that banking sector difficulties may be severe without reaching the level of a crisis. New evidence is presented to suggest that severe banking problems are more domestic in origin and effect than full-

257

©International Monetary Fund. Not for Redistribution Daniel C. Hardy and Ceyla Pazarbasioglu blown crises. External developments and constraints, such as a heavy reliance on external funds, seem to magnify the impact of a negative shock to the financial system, and full-blown banking crises may contribute to foreign exchange market turbulence. In contrast, cases of significant distress are often preceded by especially rapid credit expansion and growth in consumption, and are associated with a rising domestic real interest rate.

REFERENCES Chamberlain, Gary, 1980, "Analysis of Covariance with Qualitative Data," Review of Economic Studies, Vol. 47, pp. 225-238. Cole, Rebel, and Jeffrey Gunther, 1995, "Separating the Likelihood and Timing of Bank Failure," Journal of Banking and Finance, Vol. 19 (September) pp. 1073-89. Demirguc-Kunt, Ash, and Enrica Detragiache, 1998, "The Determinants of Banking Crises in Developing and Developed Countries," Staff Papers, International Monetary Fund, Vol. 45 (March), pp. 81-109. Eichengreen, Barry, and Andrew K. Rose, 1998, "Staying Afloat When the Wind Shifts: External Factors and Emerging-Market Banking Crises," NBER Working Paper No. 6370 (Cambridge, Mass.). Gonzalez-Hermosillo, Brenda, 1999, "Determinants of Ex-Ante Banking System Distress: A Macro-Micro Empirical Exploration of Some Recent Episodes," IMF Working Paper 99/33 (Washington: International Monetary Fund). Gonzalez-Hermosillo, Brenda, Ceyla Pazarbasioglu, and Robert Billings, 1997, "Determinants of Banking System Fragility: A Case Study of Mexico," Staff Papers, International Monetary Fund, Vol. 44 (September), pp. 295-314. Greene, William, 1990, Econometric Analysis (New York: Macmillan). Hardy, Daniel, and Ceyla Pazarbasioglu, 1998, "Leading Indicators of Banking Crises: Was Asia Different?" IMF Working Paper 98/91 (Washington: International Monetary Fund). Kaminsky, Graciela, and Carmen Reinhart, 1996, "The Twin Crises: The Causes of Banking and Balance of Payments Problems," International Finance Discussion Papers No. 544 (Washington: Board of Governors of the Federal Reserve System). Lindgren Carl-Johan, Gillian Garcia, and Matthew I. Saal, 1996, Bank Soundness and Macroeconomic Policy (Washington: International Monetary Fund).

258

Time Series Analysis of Export Demand Equations: A Cross-Country Analysis

ABDELHAK S. SENHADJI and CLAUDIO E. MONTENEGRO*

The paper estimates export demand elasticities for a large number of developing and industrial countries, using time-series techniques that account for the nonstationarity in the data. The average long-run price and income elasticities are found to be approximately -1 and 1.5, respectively. Thus, exports do react to both the trade partners' income and to relative prices. Africa faces the lowest income elasticities for its exports, while Asia has both the highest income and price elasticities. The price and income elasticity estimates have good statistical properties. [JEL: C22, E21, F14, F41]

n many developing countries that have relatively limited access to international I financial markets, exports play an important role in the growth process by generating the scarce foreign exchange necessary to finance imports of energy and investment goods, both of which are crucial to capital formation. In his Nobel prize lecture, Lewis (1980) pointed out that the secular slowdown in industrial countries will inevitably reduce the speed of development in developing countries unless an alternative engine of growth is found. That engine, he believed, was trade among developing countries. Riedel (1984) challenges Lewis's conclusions by arguing that most developing countries face a downward export demand function and therefore could expand their exports, despite the slowdown in industrial countries, by engaging in price competition. However, Faini, Clavijo, and Senhadji (1992) empirically show that Riedel's reasoning suffers from the fallacy of composition

*Abdelhak Senhadji is an Economist at the IMF Institute and Claudio Montenegro is a consultant at the World Bank. The authors thank Mohsin Khan, Sunil Sharma, and Raimundo Soto for very helpful comments.

259

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro argument in that a country alone can increase its market share through a real devaluation but all countries cannot. A central element in this controversy is the size of the price and income elasticities of developing countries' export demand. Similarly, export and import demand elasticities are critical parameters in the assessment of real exchange rate fluctuations on the trade balance. The higher the income elasticity of the export demand, the more powerful exports will be as an engine of growth.1 The higher the price elasticity, the more competitive is the international market for exports of the particular country, and thus the more successful will a real devaluation be in promoting export revenues. The recent literature is divided on how a real devaluation affects imports and exports. Rose (1990, 1991) and Ostry and Rose (1992) find that a real devaluation has generally no significant impact on the trade balance, while Marquez and McNeilly (1988) and Reinhart (1995) find that it does affect the trade balance. Using much larger samples than previous studies, this paper and its companion paper on import demand elasticities (Senhadji, 1998) offer new evidence on this issue. Section I briefly presents the export demand function and discusses the estimation strategy, and Section II presents the results. Concluding remarks are contained in Section III.

I. The Model The model is derived from dynamic optimization (for details, see Senhadji and Montenegro, 1998). More specifically, the export demand equation has the following form:

log (xt) = y0 + y1log(xt-1) + Y2log(pt) + y3log(gdpxt*) + Et, (1) where xt is real exports of the home country; pt is the export price of the home country relative to the price of its competitors; and gdpxt* is the activity variable defined as real GDP minus real exports of the home country's trading partners. Thus, the model yields an export demand equation that is close to the standard export demand function except that the correct activity variable is real GDP minus real exports of the trading partners, rather than the trading partners' GDP. In the model outlined in Senhadji and Montenegro (1998), four cases are discussed depending on which of the three variables entering equation (1) contains a unit root. The model predicts a cointegrating relationship between the I(1) variables. As will be seen in the next section, most countries cannot reject the unit root for all three variables. Consequently, equation (1) will be estimated by Phillips's Fully Modified estimator (FM), which takes into account the nonstationarity in the data as well as potential endogeneity of the right-hand side variables and autocorrelation of the error term.2 The presence of the lagged dependent variable in the export

1The trade linkage between growth in industrial countries and growth in developing countries is analyzed in detail in Goldstein and Khan (1982). 2For details about the FM method, see Phillips and Hansen (1990), Phillips and Loretan (1991), and Hansen (1992).

260

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS demand equation introduces some econometric issues in the context of a cointegration framework. Pesaran and Shin (forthcoming) show that the autoregressive specification retains its usual properties even in a cointegration framework.3

II. Estimation Results The national account data come from the World Bank national accounts database. The data for the trade shares used to compute the activity variable were taken from United Nations Statistics Office's COMTRADE, a disaggregated trade flow database. The sample includes 75 countries for which the required data are available for a reasonable time span. The list of countries is given in Table 1. In general the data are available from 1960 to 1993, with some exceptions.4 The variables in equation (1) will be proxied by the following: xt will be measured by total exports of goods and services in real terms. The activity variable (gdpx*) is computed as the weighted average of the trade partners' GDP minus exports. The weights are given by the share of the home country exports to each of its partners:

gdpx^ZaiiGDPj-xl), (2) t=\ v where GDP\ and x\ are real GDP and real exports of trade partner / in year t, and CO/ refers to the share of exports to country / in total exports. The choice of a proxy for pt is not straightforward. Ideally, a relative price should be included for all potential competitors of the home country exports, namely the export price of the home country relative to the domestic price of each importing country, as well as the export price of the home country relative to the export price of each potential competitor. Obviously, this strategy cannot be imple- mented econometrically because the equation will contain many highly correlated relative prices leading to the usual multicollinearity problem. Instead, researchers have constructed one relative price that extracts most of the information contained in all the relative prices mentioned above.5 One possibility is to use the weighting scheme for the activity variable, described in equation (2), for the construction of a composite price index that captures closely the potential competitive pressures facing the home country's exports. The home country's exports compete not only with the domestic market of each trading partner, however, but also with other potential suppliers to these markets. The world export unit value, used in this paper, implies that the threat imposed by each country in the world to the home country's exports is measured by each country's share in world exports. The export unit value index has been retained not because it is necessarily the most appropriate one from a theoretical point of view, but because it is readily available.

3See Senhadji and Montenegro (1998) for a discussion. 4The following countries have a shorter data range: Cameroon, 1965-93; Ecuador, 1965-93; Tunisia, 1961-93; and Yugoslavia, 1960-90. 5The reduction of the number of prices included in the equation can be justified from a theoretical point of view by assuming that consumers' preferences are separable leading to multi-stage budgeting. See the discussion in Goldstein and Khan (1985), pp. 1061-63.

261

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Table 1. Augmented Dickey-Fuller Test for Variables Entering the Export Demand Equation

Country x k p k gdpx* k nobs Algeria -5.44** 1 -2.04 1 -1.31 1 34 Argentina -3.16 1 -2.09 1 '. -2J4 , .1 34 Australia -1.96 1 -2.15 1 -2.99 1 34 -0.82 1 "l';y%3l ... 1 -1.33 1 34 Belgium-Luxembourg -1.62 1 -1.89 1 -1.45 1 34 Benin -2.18 1 -2.50 1 -2.44 1 34 Bolivia -2.62 1 -2.52 1 '~v -334* ' 1 34 Brazil -2.21 1 —2.42 2 -5.40** 1 34 Burundi -1.89 1 -2.24 1 -3.26 1 34 Cameroon -1.56 1 -1.59 1 -2.34 1 29 Canada -1.63 1 -2.64 1 -2.86 1 34 Central African Republic -1.46 1 -1.25 1 -2.63 1 34 : Chile -1.49 1 -2.17 1 =' -i2*':'1 34 China -3.06 1 -2.16 1 -2.24 1 34 Colombia -1.63 1 -2.56 1 -2.59 1 34 Costa Rica -1.46 1 -2.60 1 -331 1 34 Cote d'Ivoire -1.49 1 -1.97 1 -3.75** 1 34 Denmark -3.06 1 -1.99 1 -1.74 1 34 Dominican Rep. -3.99 1 -2.63 1 -2.56 1 34 Ecuador -1.75 1 -4.08* 1 -2.53 2 29 1 1 1 1 1 -2.22 Egypt -2.94 1 -1.88 1 34 Finland -1.75 1 -2.13 1 i 34 France -0.95 1 -1.97 1 -1.46-1.45 34 Gambia -2.83 1 -1.95 1 34 Germany -2.08 1 -1.97 1 34 Greece -1.63 1 -1.69 1 -4.21* 1 34 Guatemala -2.55 1 -2.59 1 -3.11 1 34 Haiti -2.29 1 -2.64 1 -3.00 1 34 Iceland -1.67 1 -2.26 1 #;^^5» 1 34 India -0.74 1 -2.35 1 -2.66 1 34 Israel -1.57 1 -2.47 1 -0.88 1 34 Italy -2.23 1 -2.17 1 -1.31 1 34 Jamaica -2.31 1 -2.57 1 -3.32 1 34 Japan -1.26 1 —2.-.265 1 •1.65 1 34 Kenya -1.18 1 -2.17 1 -0.97 1 34 Korea -0.71 1 -2.37 1 -2.66 1 34 Malawi -2.73 1 -2.23 1 -3.60* 1 34 Malaysia -1.27 1 -2.30 1 -2.97 1 34 Malta -1.55 1 -1.37 1 -0.91 1 34 Mauritania 5.41** 1 -1.84 1 -2.14 1 34

262

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS

Table 1 (concluded)

Country x k p k gdpx k nobs Mauritius -2.05 1 -2.66 1 ; -2-32; 1 34 Mexico -1.80 1 -2.62 1 -2.28 1 34 Morocco -3.45 1 -1.78 1 -2.96 1 34 -1.49 1 -1,87 1 -2.13 1 34 New Zealand -3.37 1 -2.24 1 -4.17* 1 34 Niger -2.62 1 -1.77 1 -2.69 1 34 Nigeria -2.07 1 -2,60 1 -0.90 1 34 Norway -2.42 1 -2.14 1 -1.63 1 34 Pakistan -1.49 1 -2.24 1 -3.82 1 34 Panama -2.31 1 -2.59 1 -2.21 1 34

Niger -2.62 1 -1.77 1 -2.69 1 34 Nigeria -2.07 1 -2.60 1 -0.90 1 34 Norway -2.42 1 -2.14 1 -1.63 1 34 Pakistan -1.49 1 -2.24 1 -3.82* 1 34 Panama -2.32 1 -2.59 1 -2.21 1 34

Paraguay -3.00 1 -0.88 1 08 * * 1 34 Peru -2.55 1 -2.30 1 -2.68 1 34 Philippines -2.31 1 -2.42 1 -2.20 1 34 Portugal -2.55 1 -1.97 1 03 1 34 Rwanda -6.12** 1 -2,18 1 -1.87 1 34

Seneai1 -4.40** 1 -1.49 1 -2,55, 1 34 Somalia -2.33 1 -1.99 1 -2.77 1 34 South Africa -2.92 2 -2.13 1 -1.53 1 34 Spain -1.73 1 -1.83 1 -1.72 1 34 ; Sweden -1.99 1 -2.16 1 -L75" ' 1 34 -1.02 1 1 34 Switzerland i 3 -2.01to -2.06 Togo -1.19 1 -1.10 1 -3.53 1 34 Trinidad & Tobago -1.88 1 -2.65 1 -0.83 1 34 Tunisia -1.82 1 -1.31 1 -2.96 1 34 Turkey- -1.87 1 -2.06 1 -2.00 1 34 United Kingdom -1.35 1 -2.20 1 -1.44 1 34 United States -2.69 2 -1.51 1 —Z*oY 1 34 Uruguay -2.40 1 -1.82 1 -2.82 1 34 Yugoslavia -1.74 i -2.28 1 _~) no 1 34 Zaire -2.56 1 -2.73 1 -6.80** 1 34 Note: Variables are as follows: real exports of goods and nonfactor services, .v: a weighted (by the share of exports) average of the trade partners' GDP minus exports, gdpx*; and the real exchange rate, p, computed as the ratio of the exports deflator to the world export unit values index. TUp^p three variables are tested for the existence of a unit root usin? the Augmented Dickev-Puller (ADF) test. The optimal lag selected by the Schwarz Criterion in the ADF regression is given by k. Critical values are a linear interpolation between the critical values for T = 25 and T = 50 given in

Table B.6, case4, in Hamilton (1994) (where T is the is the samle size). Significance at 1 percent and 5 percent indicated by ** and *, respectively. The number of observations is given by mobs.

263

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Unit Root Test

To determine the nature of the relationship described by equation (1), the three variables in the export demand equation—that is, real exports of goods and services of the home country, x; the relative price of exports, p; and the activity variable, gdpx*—must be tested for the presence of a unit root. The unit-root hypothesis is tested using the Augmented-Dickey-Fuller (ADF) test. The lag length, k, in the ADF regression is selected using the Schwarz Criterion (SIC). The results are reported in Table 1. For x, only 6 out of the 75 countries reject the unit root at 5 percent or less (Algeria, Burundi, Mauritania, Rwanda, and Senegal at 1 percent; Dominican Republic at 5 percent). Similarly, the null of a unit root in p is rejected only for one country (Ecuador at 5 percent). Finally, as for gdpx*t, the unit root is rejected for 10 countries (Brazil, Cote d'Ivoire, Paraguay, and Zaire at 1 percent; Bolivia, Gambia, Greece, Malawi, New Zealand, and Pakistan at 5 percent). These results show that for a large number of countries, the unit root hypothesis cannot be rejected at conventional significance levels. This may simply reflect the low power of the ADF test, especially considering the small sample size.

Export Demand Equations The results in Table 1 underscore the presence of nonstationarity in the data. For most countries (53 of the 75) the unit-root hypothesis cannot be rejected for all three variables in the export demand equation, and for the remaining 17 countries the unit-root hypothesis can be rejected for only one of the three variables. The export equation has been estimated for the 75 countries in the sample using both ordinary least squares (OLS) and FM. Table 2 reports the results for the 53 countries that show the correct sign for both the income and price elasticities. Columns labeled x-1, p, and gdpx* give, respectively, the coefficient estimates of the lagged dependent variable (log of exports of goods and nonfactor services in real terms), the short-term price elasticity yl (i.e., the coefficient of the log of the relative price), and the short-term income elasticity y2 (i.e., the coefficient of the log of gdpx*). The long-run price and income elasticities are defined as the short-term price and income elasticities divided by one, minus the coefficient estimate of the lagged dependent variable. These are given by Ep and Ey for the FM estimates. Their variance and hence their t-statistics are computed using the delta method. The column labeled ser reports the standard error of the regression. Finally, column AC gives Durbin's autocorrelation test. For the OLS regressions, AR(1) autocorrelation is detected (at 10 percent or less) for 6 of the 53 countries. Another potential problem with the OLS estimates is the possible endogeneity of pt. The FM estimator corrects for both autocorrelation and simultaneity biases. Even though Table 2 reports both the OLS and FM estimates of the export demand equation, this paper focuses only on the FM estimates, since both estimation methods yield relatively similar results. The short-run price elasticities vary from -0.0 (Peru) to -0.96 (Paraguay), with a sample average (over the first 53 countries) of -0.21, a median of -0.17, and a standard deviation of 0.19. The long-run price elasticities vary from -0.02 (Peru) to -4.72 (Turkey). The sample average is

264

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS

-1.00, the median is -0.76, and the standard deviation is 0.97. Exports are much more responsive to relative prices in the long run than in the short run. The short-run income elasticities vary from 0.02 (Ecuador) to 1.15 (Finland). The sample average is 0.41, the median is 0.33, and the standard deviation is 0.31. Thus, the average short-run income elasticity is significantly less than 1. The long-run income elasticities vary from 0.17 (Ecuador) to 4.34 (Korea). The sample average is 1.48, the median is 1.30, and the standard deviation is 0.85. Thus, exports respond significantly more to both relative prices and income in the long run than in the short run. The columns Ecp and Ecy give the long-run, bias-corrected price and income elasticities. The correction is generally small. As discussed in Senhadji and Montenegro (1998), the bias is negligible when the relative price and the activity variable are either exogenous or weakly endogenous, as is the case for most countries. Since unit-price and unit-income elasticities are widely used as benchmark values, a formal test for long run unit-price and unit-income elasticities is provided in columns labeled Ep = -1 and Ey = 1, respectively. This test uses exact critical values of the t-statistic computed by Monte Carlo methods. Twenty of the 53 countries reject a long-run, unit-price elasticity, and 18 countries reject a long-run, unit- income elasticity at 10 percent or less. The fit as measured by R2 is good. Estimates of price and income elasticities are meaningful only if the I(1) variables are cointegrated. Table 2 shows the results of the Phillips-Ouliaris (P-O) residual test for cointegration. Even with a relatively small sample size (thus low power), the null of non-cointegration is rejected for 51 (at 1 percent in most cases) of the 53 countries. To test whether these elasticities differ significantly across geographical regions, the 53 countries in the sample were classified in five regions—Africa (af), Asia (as), Latin America (la), and Middle East and North Africa (me)—and OLS regressions were run on regional dummies (t-statistics are given in parentheses):

2 \EP\ = 0.79 - 0.02daf + 1.39das - 0.37dla - 0.67dme, R = .07, N = 53; (3) (3.56)(-0.05) (2.38) (1.05) (1.51)

2 Ey = 1.74 - 0.51daf + 0.50das - 0.65dla - 0.22dme, R = .07, N = 53; (4) (9.00)(-1.73) (0.98) (-2.12) (-0.57)

2 \Ep\ = 0.79 - 035dldc, R = .01, N = 53; (5) (3.44) (1.26)

2 Ey = 1.74 - 0.42dldc, R = .04, N = 53; (6) (8.81) (-1.73) where Ep and Ey are the long-run price and income elasticities; and d\ (i = af, as, la, and me) are the regional dummies. The latter take a value equal to one if a country belongs to the region, and zero otherwise. The dummy dldc takes a value equal to one for developing countries, and zero otherwise. Interestingly, Asia has significantly higher price elasticities than both industrial and developing countries, and also has higher income elasticities than the rest of the developing countries.

265

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Table 2. Export Demand Equations

Ordinary Least Squares (OLS) estimates Country x-1 p gdpx* AC set R-

Algeria -0,07 0,99 -0.24 0.08 0.93 0.13 -2.02 6.52 -1.42 0.96 Argentina 0.33 -0.14 0.94 0.12 0.10 0.95 2.15 -2.06 3.90 0.63 Australia 0.82 0.20 0.19 0.13 0.03 0.99 6.57 -1.93 1.20 0.70 Austria 0.67 -0.08 0.88 0.10 0.05 1.00 9.96 -1.41 4.58 0.55 Benin 0.73 -0.29 0.49 0.43 0.19 0.93 5.55 -1.04 1.56 2.76 Burundi 0.04 -0.22 0.98 -0.09 0.16 0.79 0.26 -2.21 4.07 -0.46 Cameroon -0.08 0.94 0.09 0.14 0.96 0.71 -0.50 4.07 0.37 4.96 Chile 0.81 -0.21 0.28 -0.04 0.11 0.99 10.07 -2.39 2.04 -0.21 China 0.69 -0.78 0.46 0.42 0.07 0.99 10.44 -4.30 4.34 2.46 Colombia -2.26 0.54 0.11 0.11 0.96 0.72 1.72 2.01 0.55 Cote d'Ivoire 6.070.64 0.78 -0.05 0.22 0.03 1.00 4.91 10.21 -0.85 1.16 Denmark 0.78 -0.05 0.37 0.22 0.03 1.00 10.21 -0.85 2.56 1.16 Dominican Republic 0*40 -0.47 0.86 0.09 0.14 -.94 3.07 -.11 4.06 0.48 Ecuador 0.57 0.24 0.31 0.14 0.96 0.77 -4.11 0.73 1.54 8.34 Eqypt 0.78 -0.64 0.33 0.26 0.09 0.97 8.84 2.41 2.20 1.34 Finland 038 -0.64 1.30 0.18 0.04 0.99 3.76 -5.05 6.08 1.00 France 0.76 -0.01 0.57 0.37 0.03 1.00 9.97 -0.05 3.09 2.170 Gambia 0.38 -0.51 0.53 1.32 0.18 0.89 2.32 -2.42 3.31 3.46 Greece 0.55 -0.31 1.32 0.18 0.07 0.99 4,44 1.40 3.46 0.94 11.46 0.85 Guatemala -0.12 0.05 0.02 0.09 0.94 0.06 0.40 0.11 Haiti 0.72 -0.02 0.18 0.84 5.69 -0.13 0.37 0.01 1.53 0.06

266

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS

Fully-Modified estimates E2 Etp Etj ser E2=1 X_| P gdpx* Ep R- P-O Er=-\ nobs -0.07 0.83 —0.25 —0.25 -6.62a 1.00 c 0.27 -0.07 0.83 -0.07 0.83 -6.62a 30.42- 2.83 34 2.91 -0.07 7.60 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 0.56 -0.11 0.56 1.75 34 -0.07 0.83 4.33 -1.94 2.73 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.90 -0.17 0.08 34 -0.07 0.83 9.58 -2.13 0.64 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.75 -0.04 0.65 34 -0.07 0.83 11.44 -0.70 3.37 -0.07 —0.25 —0.25 -6.62a 1.00 30.42- 1.75 0.80 0.31 -4.13C 34 -0.07 10.14 -0.07 1.59 0.83 b «1.58 3.00 a 0.28 -0.19 0.74 -0.07 0.83 —0.25 —0.25 -6.62 1.00 1.00 30.42- 1.75 34 -0.07 0.83 1.71 -2.01 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.84 -0.04 34 -0.07 0.83 7.48 -0.30 -0.07 0.91 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.20- 0.85 -0.17 34 -0.07 0.83 14.60 -2.61 0.07 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 0.80 -0.63 0.99--13.33a 2*. f i 0.38 34 -0.07 -0.07 0.83 10.29 -3.08 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.86 -0.21 34 -0.07 -0.07 0.83 7.76 -1.70 -0.07 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.84 34 -0.07 -0.07 -0.07 0.83 7.58 -0.07 -0.07 —0.25 1.00 1.00 30.42- 1.75 0.85 -0.06 -0.41 0.02 34 -0.07 -0.07 11.77 -1.13 !:S. -0.07 -0.07 0.83 a j to -6.62 b e —0.25 1.00 0.56 -0.36 -0.07 -0.07 0.83 o

4.91 -3.41 34 1.00 30.42- 1.75 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.87 -0.43 34 -0.07 -0.07 0.83 8.95 -2.98 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.84 -0.24 34 -0.07 -0.07 0.83 11.33 -2.68 -0.07 -0.07 1.00 30.42- 1.75 0.45 -0.58 -1.20 2.09 0.03 ~r4,5oa 34 -0.07 -0.07 5.36 -5.55 4Q (5 a -0.07 -0.07 0.830.834 30.42- 1.75 0.79 0.00 -0.02 2,8 0.03 1.00 -3.82 34 -0.07 -0.07 0.83 11.42 -0.05 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.49 -0.40 34 -0.07 -0.07 0.83 4.02 -2.59 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.66 34 -0.07 -0.07 0.83 7.08 -0.07 -1.27 -0.07 -0.07 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.90 -0.09 34 -0.07 -0.07 0.83 20.55 -0.76 0.83 —0.25 —0.25 -6.62a 1.00 1.00 30.42- 1.75 0.80 -0.07 34 0.83 10.49 -0.93 -0.07 -0.07 1.82 -0.89

267

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Table 2 (continued)

Ordinary Least Squares (OLS) estimates x-1 R2 Country P gdpx* AC sc 0.82 Iceland 0.82 0.57 0.12 0.07 0.98 0.82 2.60 0.67 -2.02 Italy 0.82 -0.07 0.95 0.18 0.04 1.00 0.82 -0.87 3.25 0.93 Japan 0.82 -0,25 46 0.97 0.06 1.00 10.00 -1.56 0.82 0.97 Kenya 0.62 0.82 0.97 0.59 0.59 0.82 0.82 0.97 '-..• 0.82 0.82 Korea 0.82 0.82 0.59 1.00 0.82 0.82 0.82 1.51 1.51 Malawi 0.34 -0.1-2,058 0.79 1.51 0.11 0.93 1.51 0.82 -1.22 3.38 Malta 0.78 -0.12 0.64 1.51 0.59 0.99 10.80 -0.86 0.82 1.51 Mauritius -0.25 0.82 1.51 0.59 0.59 0.82 1.51 0.82 -1.45 5.96 Morocco 0.82 -0.38 0.82 1.51 0.07 0.97 0.82 0.82 1.51 -2.59 New Zealand 0.82 -0.17 0.21 1.51 0.59 0.99 0.82 -2.16 0.82 1.51 Niger -0.32 0.82 1.51 0.59 0.50 0.82 -1.42 0.82 1.51 4.79 Nigeria 0.82 1.51 0.59 0.85 0.82 0.82 0.82 1.51 6.04 -0.45 Norway 0.82 -0.17 0.82 1.51 0.59 0.59 0.82 -2.10 0.82 1.51 Panama -0.23 0.16 1.51 0.59 0.59 0.82 0.82 0.82 1.51 7.20 Paraguay 0.82 -0.88 0.82 1.51 0.59 0.59 0.82 0.82 1.51 -4.39 Peru 0.82 -0.06 0.82 1.51 0.59 0.59 0.82 0.82 1.51 -O.60 Philippines 0.82 -0.62 0.82 1.51 0.07 0.98 0.82 -6.33 0.82 1.51 Portugal 0.88 -0.25 0.82 1.51 0.59 0.59 0.82 -1.15 0.82 1.51 Senegal 0.82 -0,42 0.82 1.51 0.59 0.59 0.82 -2.58 0.82 1.51 South Africa 0.59 -0,20 0.82 0.59 0.97 0.82 6.59 -AA1 1.51 1.11 Spain 0.60 -0.06 0.82 1.51 0.59 1.00 0.82 1.51 0.82 -0.58

268

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS

Fully-Modified estimates

X_| 2 P gdpx* EP E2 E3 El ser R P-O Ep=-\ Ey=l nobs 0.70 -0.28 0.41 0.59 1.37 -1.11 L39 0.04 0.98 0.59 0.59 0.59 34 9.71 -3.81 3.39 -3.76-' 11.78a 0.65 -0.05 0.80 -0.14 2.26 -0.13 2.25 0.03 1.00 -4.83 a 0.59 13.74- 34 6.55 -0.69 3.27 -0.76 24.63 ^ 0.87 -0.17 0.27 -1.27 2.11 -1.33 2.02 0.03 1.00 -9,74 a -0.48 2.21 - 34 19.70 -1.94 1.81 -2.30 h 4.21" 0.84 -0.33 0.03 -2.07 0.17 -2.36 0.17 0.05 0.94 -7.49;i -0.81 -1.22 34 7.66 -4.71 0.21 -1.56C 0.25 0.76 -0.52 1.04 -2.17 4.34 -2.15 431 0.08 1.00 0.59 -1.51 7.58- 34 10 52 -2.15 2.73 -2.80b 9.85 * 0.50 -0.05 0.63 -0.10 1.25 -0.11 L20 0.06 0.93-10.70;' 5.01a 1.70 34 4.91 -0.55 4.16 -0.57 8.43;i 0.84 -0.04 0.59 -0.22 0.59 0.59 2.79 0.06 0.98 -3.88c 1.18 4.43a 34 13.39 0.59 2.52 -0.34 6.89 a 0.89 -0.21 0.34 -1.92 3.17 -1.67 3,24 0.10 0.94 -6.02a -0.46 1.56 34 10.24 0.59 1.91 -0,96 2.28 c 0.81 0.59 0.22 -1.47 1.12 -1.45 l.U 0.06 0.97 -6,42a -0.45 0.42 34 8.06 -2.19 1.52 -1.41 3.95*> 0.90 -0.13 0.08 -1.25 0.78 -1.62 0,80 0.03 0.59 -9.50a -0.19 -0.41 34 9.33 -2.42 0.64 -0.94 1.49 0.84 -0.28 0.06 -1.74 0.38 0.59 0.59 0.59 0.59 0.59 0.59 -0.82 34 8.60 -1.80 0.47 -1.16 0.50 0.91 -0.04 0.15 -0.50 1.69 0.59 0.59 0.59 0.59 0.59 0.44 0.47 34 9.31 -0.65 1.03 -0.43 1.15 0.90 -0.15 0.17 -1.51 1.65 -1.73 1.65 0.03 1.00 -9.43a -0.46 0.59 34 10.40 -2.32 0.91 -1.36^ 3.67 b 0.85 -0.17 0.07 -1.14 0.47 0.59 0.59 0.59 0.59 0.59 0.59 0.59 34 0.59 0.59 0.41 -1.68 c 0.50 0.64 -0.96 1.11 -2.67 3.08 0.59 0.59 0.59 0.59 0.59 0.59 0.59 34 0.59 -5.70 5.42 -4.19^ 10.66a 0.78 0.00 0.12 -0.02 0.53 -0.02 0.54 0.59 0.59 0.59 0.59 0.59 34 8.17 -0.05 1.64 -0.05 2.30 0.59 -0.51 0.49 -1.24 1.20 -1.22 1.19 0.05 0.98 -4.59 b -1.32 1.57 34 8.64 -6.60 4.20 -6.92a 9.52« 0.93 -0.20 0.09 -2.92 1.30 -2.89 1.29 0.08 0.96 -4.93a -0.33 0.17 34 9.84 -1.21 0,38 -0.50 0.75 0.45 -0.28 0.32 -0.50 0,58 -0.47 0.58 0.08 0.84 -6.64 a 2.54h -2.85 34 3.64 -2.27 2.79 -2.59 b 3.93a 0.65 -0.18 0.23 -0.51 0.66 -0.50 0.65 0.02 0.97 0.59 4.02 » -5.35a 34 8.91 -5.45 5.24 -4.LV' 10.33 » 0.67 -0.06 0.94 -0.18 2.86 -0.19 2.75 0.04 1.00-12.06a 3.80b 11.71" 34 6.05 -0.74 2.64 -0.82 18.01a

269

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Table 2 (concluded)

Ordinary Least Squares (OLS) estimates

Country -V-l E2 gdpx AC ser /?2 Sweden 0.55 -0.13 0.76 0.33 0.03 1.00 0.59 -1.88 3.68 L84 Switzerland 0.31 -0.12 1.18 0.34 0.02 1.00 2.91 -242 6.24 2.04 Togo 0.57 -0.21 0.58 0.59 0.59 0.90 3.20 -1.21 1,22 0.59 Trinidad and Tobago 0.24 -0.29 0.91 0.59 0.59 0.59 1.58 -4.63 4,28 0.59 Tunisia 0.59 -0.17 1.15 —0.09 0.07 0.99 5.62 -1.26 3.67 -^0.47 Turkey 0.59 -0 69 0.31 0.59 0.59 0.59 0.59 -2.50 1.15 0.59 United Kingdom 0.58 -0.16 0.61 0.59 0.59 0.59 0.59 -2.59 4.84 0.59 United States 0.79 —0.19 0.26 0.59 0.59 0.59 8.41 -1.42 2.20 0.59 Uruguay 0.66 0 &St (\ *y\ 0.59 0.59 0.97 0.59 0.59 0.59 0.59 Yugoslavia 0.47 -0.23 0.67 0.59 0.59 0.59 345 . ^3,33 ' ' 2.92 -0.54 Zaire 0.50 -0.15 0.58 0.15 0.59 0.59 3.84 -2.27 2.69 0.59 0.59 0.59 Mean ().: Median 0.59 0.59 0.5 Stdcv 0.19 0.20 0.35 Min 0.04 -0.88 0.05 Max 0.88 -0.01 1.32

aSieniticant at 1 percent ^Significant at 5 oercent cSignigicant at 5 percent. Note: The dependent variablejgsrptst'sgfs'gls'g'sgls'rpw'jtgs'jgs'rewfls'galepw'a'gla'fwe'fagga'jfdlapea'g'a'ga;a'g'ag'a lagged dependent variable, x_f, the real exchange rate, p, computed as the ratio of exports deflator to the world export unit value index, and the weighted (by export shares) average of trade partners' GDP minus exports, gdpx*. The export demand equation is estimated using both OLS and the Phillips-Hansen's Fully Modified estimator. The long-run price ankjlfkgfsd'gr'sdftke'jfdfldpwer'dlre'alda'lfdfpef';aldfa;fj'af;ldfa;epf'afldra'fga''flaga'gla'g'gla'eafla'gala'glag'a'glaalga'g'aglgh corrected for bias (see Table 4 in Senhadji and Montenegro, 1998). For each country, the estimated coefficients and their r-statistic (below the coefficient estimates) are provided. The following statistics are also provided: Durbin's test for autocorrelation, AC; R2\ standard error of the regression, ser, and the number of observations for each country. nobs. Cointegration between the three variables in the export demand equation is tested using the Phillips-Ouliaris

residual test given in column P-O. Finally, the columns labeled Ep = -1 and Ey — 1 report the two-tailed test for unit- price and unit-income elasticities, respectively. The asymptotic critical values for the Phillips-Ouliaris test at 10 percent, 5 percent, and 1 percent are, respectively, -3.84, -4.16, and -4.64. Exact critical values (from Table 8 in Senhadji and j'aefpwjep'adflaerpfajfdaoeajf'ajepwf'afldfeowa;flsalgoaj'gjsdls'dafsdgiaeoataw'fjlfap'sjagds'ajgdalewa'ajfagla'eaa''ga''g'ja'ga.

270

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS

Fully-Modified estimates p x-1 gdpx* E1 E2 E3 E4 ser i?2 P-0 Ep=1 E1=1 nobs 2.802.80 b a -0.77 0.68 -0.09 0,53 -0.29 -0.30 -0.77 -0.77 -0.77 3.20 5.62 34 6.26 2.80 2.80 2.80 2.80 2.80 2.80 2.80 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 0.42 4.36 2.80 2.80 2.80 2.80 0.84 2.80 2.80 2.80 2.80 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 5.82 2.80 2.80 2.80 2.80 0.37 2.80 2.80 2.80 2.80 -0.77 1.22 0.09 0 96 -0.77 -0.77 -0.77 -0.77 2.80 2.80 2.80 2.80 9 41 a 2.80 2.80 2.80 2.80 2.80 -0.77 -0.77 -0.77 2 42 0.05 0.99 4 50b 33 2.80 2.80 2.80 2.80 2.80 2.80 2.80 2.80 2.80 2.80 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.7-0 -0.77 -0.77 -0.77 -0.69 1.04 -0.77 0.99 —3.53 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -1.77 0.59 0.06 0*97 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 —0 37 u.y.z 0 10 0 90 -4.57 b 5.08^ -0.77 -0.77 -0.77 -0.77 ' 2 48 h

-0.77 -0.77 -0.77 -0.77 -0.77 -1.07 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 1.29

-0.77 -0.77 -0.77 -0.77 -0.77 -0.77 &*'/:•' 0.84 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 n 1 f>.::"'" -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 -0.77 :

271

©International Monetary Fund. Not for Redistribution Abdelhak S. Senhadji and Claudio E. Montenegro

Developing countries, except Asia, have significantly lower income elasticities than industrial countries. Developing countries also show lower price elasticities than industrial countries. Finally, the lower income elasticities for developing countries in general, and for Africa in particular, are even more forcefully demonstrated by the following weighted least squares regressions:6

2 Ey = 1.83 - 1.04daf - 0.40das - 0.54dla - 0.62dme, R = .90, N = 53; (7) (25.77) (-6.99) (-1.14) (-2.28) (-3.89)

2 Ey = 1.83 - 0.78dldc, R = .89, N=53. (8) (24.71)(-6.69)

While developing countries' income elasticities are lower, they remain larger than one. Consequently, growth in their partner countries will translate into growth of at least the same magnitude of their exports. Thus trade remains an important engine of growth for all developing countries.

III. Conclusion The paper provides income and price elasticities of the export demand function for 53 industrial and developing countries, estimated within a consistent framework and taking the possible nonstationarity in the data into account. The long-run price and income elasticities generally have the expected sign and, in most cases, are statistically significant. The average price elasticity is close to zero in the short run but reaches about one in the long run. Twenty-two of the 53 countries in the sample have point estimates of long-run price elasticity larger than one, and for 33 countries the unit-price elasticity cannot be rejected. It takes six years for the average price elasticity to achieve 90 percent of its long-run level. A similar pattern holds for income elasticities in that exports react relatively slowly to changes in trade partners' income. The short-run income elasticities are on average less then 0.5, while the long-run income elasticities are on average close to 1.5. Thirty-nine countries have point estimates of long-run income elasticity that are larger than one, and for 35 countries the unit-income elasticity cannot be rejected. Thus, exports do significantly react to both movements in the activity variable and the relative price, though slowly. A comparison with Reinhart (1995), who uses a similar methodology, shows that her estimates of the price elasticities are significantly lower. Her mean estimate (over the 10 developing countries showing the right sign) is -0.44, while it is -1.14 in this paper (where the mean is over the 37 developing countries in the sample). Conversely, her average income elasticity is 1.99 compared to 1.32 in this paper. These differences may simply reflect the difference in the periods of analysis and sample sizes. While developing countries show, in general, lower price elasticities than industrial countries, Asian countries have significantly higher price elasticities

6All the variables in the equations have been weighted by the inverse of the standard error of the corresponding elasticity.

272

©International Monetary Fund. Not for Redistribution TIME SERIES ANALYSIS OF EXPORT DEMAND EQUATIONS than both industrial and developing countries. Furthermore, Asian countries bene- fit from higher income elasticities than the rest of the developing world, corrobo- rating the general view that trade has been a powerful engine of growth in the region. Africa, in contrast, faces the lowest income elasticities.

References Faini, Riccardo, Fernando Clavijo, and Abdelhak Senhadji, 1992, "The Fallacy of Composition Argument: Is It Relevant for LDCs' Manufactures Exports?" European Economic Review, Vol. 36 (May), pp. 865-82. Goldstein, Morris, and Mohsin Khan, 1982, "Effects of Slowdown in Industrial Countries on Growth in Non-Oil Developing Countries," Occasional Paper No. 12 (Washington: International Monetary Fund). , 1985, "Income and Price Effect in Foreign Trade," in Handbook of International Economics, ed. by Ronald Jones and Peter Kenen (Amsterdam: North-Holland). Hamilton, James, 1994, "Time Series Analysis," (Princeton, N.J.: Princeton University Press). Hansen, Bruce, 1992, "Efficient Estimation and Testing of Cointegrating Vectors in the Presence of Deterministic Trends," Journal of Econometrics, Vol. 53 (July-September), pp. 87-121. Lewis, Arthur, 1980, "The Slowing Down of the Engine of Growth," American Economic Review, Vol. 70 (September), pp. 555-64. Marquez, Jaime, and Caryl McNeilly, 1988, "Income and Price Elasticities for Exports of Developing Countries," Review of Economics and Statistics, Vol. 70 (February), pp. 306-14. Ostry, Jonathan, and Andrew Rose, 1992, "An Empirical Evaluation of the Macroeconomic Effects of Tariffs," Journal of International Money and Finance, Vol. 11 (February), pp. 63-79. Pesaran, M. Hashem, and Yongcheol Shin, forthcoming, "An Autoregressive Distributed Lag Modelling Approach to Cointegration Analysis," in Centennial Volume of Ragner Frisch, ed. by S. Strom, A. Holly, and P. Diamond (Cambridge: Cambridge University Press). Phillips, Peter C.B., and Bruce Hansen, 1990, "Statistical Inference in Instrumental Variables Regression with I(1) Processes," Review of Economic Studies, Vol. 57 (January), pp. 99—125. Phillips, Peter C.B., and Mico Loretan, 1991, "Estimating Long-run Economic Equilibria," Review of Economic Studies, Vol. 58 (May), pp. 407-36. Riedel, James, 1984, "Trade as the Engine of Growth in Developing Countries, Revisited," Economic Journal, Vol. 94 (March), pp. 56-73. Reinhart, Carmen, 1995, "Devaluation, Relative Prices, and International Trade," Staff Papers, Vol. 42 (June), pp. 290-312. Rose, Andrew, 1990, "Exchange Rates and the Trade Balance: Some Evidence from Developing Countries," Economic Letters, Vol. 34 (November), pp. 271-75. -, 1991, "Role of Exchange Rates in a Popular Model of International Trade: Does the 'Marshall-Lerner' Condition Hold?" Journal of International Economics, Vol. 30 (May), pp. 301-16. Senhadji, Abdelhak, 1998, "Time Series Estimation of Structural Import Demand Equations: A Cross-Country Analysis," Staff Papers, Vol. 45 (June), pp. 236-268. Senhadji, Abdelhak, and Claudio Montenegro, 1998, "Time Series Analysis of Export Demand Equations: A Cross-Country Analysis," IMF Working Paper 98/149 (Washington: International Monetary Fund).

273

The Uzbek Growth Puzzle

JEROMIN ZETTELMEYER*

After the breakup of the Soviet Union, Uzbekistan's output fell less than in any other former Soviet republic, and growth turned positive in 1996/97. Given the country's hesitant and idiosyncratic approach to reforms, this record has surprised many observers. This paper first shows that a standard panel model of growth in transition systematically underpredicts Uzbek growth from 1992-1996, confirming the view that Uzbekistan's performance constitutes a puzzle. It then attempts to resolve the puzzle by extending the model in a way that encompasses competing hypotheses of what makes Uzbekistan's output path unusual. The main result is that Uzbekistan's performance can be accounted for by a combination of low initial industrialization, its cotton production, and its self-sufficiency in energy. [JEL: O53, P24, P27, P52]

y any measure, the decline in output in Uzbekistan since the beginning of B transition has been relatively mild. According to IMF data based on official statistics, 1997 Uzbek output stood at about 85 percent of its 1991 level, as compared to an average of 60 percent for the Baltics, Russia, and other countries of the former Soviet Union (hereafter BRO; see Table 1). Total cumulative output loss was only 59 percent of 1991 output by 1995 and 89 percent by 1997—as opposed to 126 and 207 percent, respectively, for the BRO average. Output estimates based on electricity consumption—sometimes regarded as preferable because they better capture informal sector output—indicate that

*Jeromin Zettelmeyer is an Economist in the Research Department. The author thanks Peter Keller, Gunther Taube, Adham Bekmuradov, Isaias Coelho, Christoph Rosenberg, three anonymous referees, and seminar participants at the IMF, Tashkent State Economic University, the University of World Economics and Diplomacy (Tashkent), and the Uzbek Institute of Strategic and Interregional Research for valuable comments and suggestions. Mandana Dehghanian and Nada Mora provided outstanding research assistance.

274

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE 112 i V/U 112 1991-95 Cumulative Loss1 112 112 112 112 112 112 112 112 112 112 112 112 112 112 70 71 60 44 68 82 64 62 1995 70 70 70 70 70 70 70 70 70 1991 = 100 88 v 67 v 73 8. 6: 66 1994 88 ll|;f 7C Electricity-Based Data2 Electricity-Based 88 88 88 88 88 88 88 147 141 148 324 279 343 208 269 \: 252 1991-97 :: 202 v 88 88 88 88 88 88 88 3 Loss Cumulative 88 88 169 223 Z3U 126 131 1991-95 88 88 88 88 88 88 88 88 88 88 50 46 SO 74 70 1997 86 60 58 52 52 52 52 52 52 52 52 52 1996 ' Official Data 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 1995 96 96 96 96 96 96 96 96 96 96 96 96 96 96 96 96 96 36 71 73 74 CO 56 eg 1994 96 96 96 96 96 96 96 96 96 96 ) Output Index (1991 = 100 ©International Monetary Fund. Not for Redistribution no 41 60 85 1993 50 50 50 50 50 96 50 50 50 50 50 50 50 50 1992 Table 1. Baltics, Russia, and Other Countries of the Former Soviet Union (BRO): Output Paths (BRO): Output Former Soviet Union Countries of the Russia, and Other Table 1. Baltics, 80 80 80 80 80 80 80 80 96 80 80 80 80 80 80 80 80 3 1Source: IMF' author's calculations. author's calculations. 2Source: Johnson. Kaufmann, and Shleifer (1997); between 1991 level and levels in 1992 through 1995 or 1997). In percent ol 1991 output (sum of differences Azerbaijan Amenia Belarus Estonia Georgia Kazakhstan Kyrgyz Republic Latvia Lithunia Moldova Russia Tajikistan Turkmenistan Ukraine Uzbekistan BRO Average Excl. Uzbekistan

275 Jeromin Zettelmeyer

Figure 1. Output Paths in Transition Time (Pre-Transition Year = 100)1

Output index

Uzbekistan

Average Central and Eastern Europe

Average Baltics, Russia, and other countries of the former Soviet Union

Transition year

•Transition time refers to years since the beginning of transition (defined as transition year 0). This is assumed to be 1992 for the Baltics, Russia, and the other countries of the former Soviet Union; 1990 for Poland, Hungary, and the former Yugoslavia; and 1991 for the remaining Central and Eastern European transition economies. these differences may be exaggerated,1 but they corroborate the finding that Uzbekistan's output decline was far milder than that in the other countries. Uzbekistan appeared to resume positive growth in 1996 and 1997, ahead of other large BRO economies, such as Russia and Ukraine, which continued to decline in 1996 and were at best stagnant in 1997. Finally, it is worth noting that Uzbekistan's transitional recession was mild not only relative to the BRO average but also relative to the average of the Central and Eastern European transition economies (see Figure 1). Observers are often puzzled by Uzbekistan's output performance, typically because they think that the country could have done much worse given its hesitancy to engage in rapid market-oriented reforms and sustained macroeconomic stabilization—policies that have been widely credited with contributing toward milder transitional recessions and quicker and stronger recoveries.2 In Uzbekistan, liberalization has proceeded hesitantly and with occasional reversals—in particular, with regard to

!This is driven by a larger downward bias to official output measurement in the other 14 countries due to faster informal sector growth; see Taube and Zettelmeyer (1998). 2Berg and others (1999); de Melo and others (1997); Havrylyshyn, Izvorski, and van Rooden (1998); Hernandez-Cata (1997); Fischer, Sahay, and Vegh (1996a and b); Sachs (1996); Aslund, Boone, and Johnson (1996); Selowsky and Martin (1997); Wolf (1997); and World Bank (1996).

276

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE its external regime—and structural policies have concentrated on public investments that aimed at substituting energy and industrial imports, along with an extensive system of transfers to the largely state-controlled industrial sector.3 At the same time, the output decline was arrested relatively quickly following limited stabilization, in spite of macroeconomic imbalances that re-emerged in late 1996 and 1997. Puzzling or not, what explains this relatively good performance? The fact that Uzbekistan did not follow standard market-oriented economic reforms makes this question all the more interesting, and poses a challenge to the standard policy prescription. The paper proceeds in two steps. First, it asks if there really is a puzzle. Obviously, structural reforms and macroeconomic policies may not be the only— or perhaps even the main—determinants of output in transition. Other variables, such as initial conditions, also matter. The question is whether Uzbekistan's performance is still puzzling once these variables are taken into account in the context of a standard cross-country regression model. Second, to the extent that standard explanatory variables cannot fully explain Uzbekistan's output path, what are alternative explanations? This is addressed by extending the basic regression model in a way that seeks to encompass competing hypotheses of what could have contributed to Uzbekistan's unusual output path. The main result is that the Uzbek growth puzzle can be "resolved" in an accounting sense after controlling for its low degree of initial industrialization, production of agricultural commodities (including cotton), and the energy balance. Public investment, which has also been cited as a possible reason for Uzbekistan's relative success, seems to have little or no explanatory power. One interpretation of these results is simply that favorable initial conditions, rather than policies, should be credited with Uzbekistan's output performance. An alternative interpretation is that Uzbekistan's policy of subsidizing the official industrial sector was relatively successful in mitigating the output decline, given a low degree of industrialization to begin with, because it could be financed through export proceeds from agriculture and because of the availability of domestic energy. In this view, the combination of go-slow policies with favorable initial conditions achieved a result that eluded other former communist countries that tried similar approaches, but ran into financing constraints much earlier.

I. Is There A Puzzle? This section of the paper is based on a panel regression model of the main determinants of output growth during transition estimated by Berg and others (1999) using data for 26 transition economies.4 The model is flexible in that it has a very general dynamic structure, does not assume that policies and initial conditions necessarily have the same effects on the private and the state sectors, and considers a large number of potential determinants of growth, which are reduced using a general-to-specific methodology. These include macroeconomic variables (fiscal

3For details, see IMF (1998, 1997). 4Berg and others also discuss other variants of the model, which have similar implications as the version used here.

277

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer balance and inflation, instrumented using IMF program targets); structural reform indices (constructed by de Melo and others, 1996a, b, and updated using the reform indices of the EBRD Transition Reports)', initial conditions, including variables capturing initial structure (overindustrialization, initial share of agriculture, trade dependency); initial PPP-adjusted income; initial macroeconomic distortions (as measured by measures of repressed inflation and/or inflation and fiscal balance in the year prior to the beginning of transition); the initial state of reforms; and some other controls, including a dummy for wars. Table 2 compares fitted and actual growth in "transition time" (time since the end of central planning) for (1) an average of 25 transition economies excluding Uzbekistan, (2) an average of the Baltics, Russia, and other countries of the former Soviet Union, again excluding Uzbekistan, and (3) Uzbekistan. "Year zero" is defined as the year in which central planning ended (1992 in Uzbekistan and the BRO and 1990 or 1991 in the remaining transition economies in the sample; see note to Figure 1). In addition to showing the residuals in each group as the difference between fitted and actual growth, the table shows the average of the absolute residuals across countries in the transition and BRO groups, respectively. This per- mits a comparison of the absolute magnitude of the residual for Uzbekistan with that of a "typical" transition country. The main results from the table are as follows. First, the model correctly predicts a higher growth for Uzbekistan in the first two years of transition relative to the average, that is, a smaller output decline. Consequently, we can get some insights into the relatively good Uzbek output performance during 1992-93 by looking into what drives the model's predictions (see below). Second, the model systematically underpredicts Uzbek growth. The underprediction is particularly impressive for 1994 (year 2 in transition time), when the model predicts a large collapse in output that did not materialize. As a result, the total regression residual for Uzbekistan (as measured by the cross sum of the five annual absolute residuals) is much larger than that for the typical transition country or BRO economy (28.7 versus 18.5 and 17.0, respectively). A Chow test for predictive stability confirms that this difference is much larger than what could reasonably be attributed to chance.5 Based on the model by Berg and others, it thus certainly seems justified to speak of an "Uzbek growth puzzle." To resolve this puzzle, one must look beyond this model. Before doing this, however, we seek to understand the variables that drive the existing model's limited capacity to explain Uzbek growth performance, and in particular the differences between the Uzbek fitted path and the average fitted path for the other transition economies (Table 3). Table 3 decomposes the fitted values for Uzbekistan and the group of remaining 14 countries into the contribution of the main groups of explanatory variables.6 To the extent that the standard model can explain Uzbekistan's output path in the first two years, it does not attribute Uzbekistan's relatively favorable performance to its macroeconomic policies and the (slow) pace of its structural reforms. On both fronts,

5The null hypothesis of no structural break is rejected at the 5 percent level (p-value: 2.7 percent). 6This decomposition is possible because the model does not contain lagged dependent variables. Thus, at any point in time, the fitted value of the model can be written as a linear combination of the independent variables. See Zettelmeyer (1998) for a more detailed decomposition.

278

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE

Table 2. Uzbekistan and Transition Economy Average: Fitted and Actual Growth Paths Based on the Model by Berg and Others

(in percent per year) Transition Time 0 l 2 3 4 5 Average of transition countries excluding Uzbekistan Actual growth -21.3 -12.5 -9.8 -1.5 1.6 2.6 Fitted growth -20.9 -12.7 -9.1 -1.6 1.7 3.4 Residual -0.4 0.2 -0.7 0.1 0.0 -0.8 Average of absolute residual 3.3 3.1 4.2 3.1 3.3 2.4 Average of the Baltics, Russia, and other countries of the former Soviet Union Union, excluding Uzbekistan Actual growth -25.8 -14.1 -13.3 -3.9 -0.2 Fitted growth -24.7 -14.6 -12.3 -4.1 0.1 Residual -1.1 0.5 -1.0 0.2 -0.3 Average of absolute residual 4.2 3.2 4.6 2.9 3.7 Uzbekistan Actual growth -11.1 -2.3 -4.2 -0.9 1.6 Fitted growth -15.6 -6.4 -18.9 -4.1 0.0 Residual 4.5 4.1 14.7 3.8 1.6 Absolute residual 4.5 4.1 14.7 3.8 1.6

Uzbekistan performed worse than the average of transition economies, according to Table 3. This is not surprising, since the Berg and others cross-country model associ- ates fast reforms with faster output recovery, based on the experience of most other transition economies. Instead, Table 3 attributes the relatively good performance of Uzbekistan in the first two years of transition to unusually favorable initial conditions, which more than offset the unfavorable impact of slow structural reforms and macroeconomic imbalances in that period. An unbundling of these initial conditions shows that this is mainly driven by one variable, "overindustrialization," which captures the degree of industrialization at the beginning of transition relative to the industrialization typical for a market economy in the same range of GDP per capita.7 According to the dataset of de Melo and others (1997), from which the data documenting initial conditions were taken, Uzbekistan's industry share was actually smaller than what would have been expected based on its GDP per capita. Thus, according to the standard model, Uzbekistan did better than the average transition

7More precisely, "overindustrialization" is defined as the difference between the actual share of industry in the country in 1989 and the share that would have been predicted on the basis of the country's per capital income. The latter is obtained as the fitted value from a regression of industrial share on per capita income in a large sample of market economies. For more details, see de Melo and others (1997) and references cited therein.

279

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

Table 3. Uzbekistan and Transition Economy Average: Decomposition of Fitted Growth (Berg and others Model)

(in percent per Transition Time year) 0 1 2 3 4 Average of Baltics, Russia, and other countries of the former Soviet Union, excluding Uzbekistan Fitted growth -24.7 -14.6 -12.3 -4.1 0.1 Macroeconomic policy -2.6 1.0 -0.7 0.3 0.6 Structural reforms 3.7 4.0 5.6 9.4 10.5 War -3.0 -3.0 -0.7 -0.2 -0.2 Constant -8.9 -8.9 -8.9 -8.9 -8.9 Initial conditions -13.9 -7.8 -7.6 -4.8 -2.0 Trade dependency -8.2 -5.5 -2.9 -0.2 2.5 Overindustrialization -8.5 -5.6 0.0 0.0 0.0 Urbanization + agriculture 4.1 -1.9 -3.7 -3.7 -3.7 Other1 -1.3 5.3 -1.1 -0.9 -0.8 Uzbekistan Fitted growth -15.6 -6.4 -18.9 -4.7 0.0 Macroeconomic policy -A3 0.0 -3.5 0.0 2.2 Structural reforms 0.6 -0.5 -0.6 7.3 7.3 War 0.0 0.0 0.0 0.0 0.0 Constant -8.9 -8.9 -8.9 -8.9 -8.9 Initial conditions -3.0 2.9 -5.9 -3.3 -0.6 Trade dependency -8.0 -5.6 -3.2 -0.8 1.5 Overindustrialization 3.4 2.1 0.0 0.0 0.0 Urbanization + agriculture 5.5 -0.6 -1.3 -1.3 -1.3 Other1 -3.9 7.0 -1.3 -1.1 -0.8 1Initial macroeconomic imbalances (estimated repressed inflation in the five years prior to transition; deficits and inflation in the last year prior to transition), pre-transition structural reforms, and a dummy for the resource-rich countries (Azerbaijan, Russia, Kazakhstan, and Turkmenistan). economy in the first two years mainly because it was less industrialized in the first place, and as such had a smaller share of output that was vulnerable to collapse after the end of central planning. However, Uzbekistan's lack of industrialization would only have retarded, but not eliminated, the output collapse according to the model by Berg and others. Since the destructive effect of "overindustrialization" is concentrated in the first two years, the comparative advantage afforded by Uzbekistan's initial economic structure should mostly have been lost after that period.8 Aside from low

8This is what explains the peculiar time path of the "initial conditions" line of Table 3 for Uzbekistan, which contrasts with the nicely upward-sloping path for the BRO average. In year zero, Uzbekistan's "under- industrialized" initial state mitigates but does not quite offset the negative impact of the remaining initial conditions, whereas in year one the latter is slightly more than offset. In year two, the offsetting effect disappears.

280

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE initial industrialization, the remaining initial conditions measured by Berg and others do not show Uzbekistan in a substantially better position than the other countries. In light of the downward trend to output (reflected in the regression constant), which the model by Berg and others attributes to the transition phe- nomenon over and above what is attributable to individual variables, and Uzbekistan's failure to offset this trend by more vigorous market-oriented reform policies, the model would have predicted the output decline to set in with a vengeance in year three. But this did not happen.

II. Explaining the Uzbek Growth Puzzle: Econometric Findings To shed some light on the remaining "growth puzzle," this section extends the model of the previous section to encompass several "explanations" of the growth puzzle that have been suggested in the past. In particular, it includes variables reflecting the dollar value of cash crops and natural resources (including energy and non-ferrous metals), as well as the energy balance; and capital expenditure of the general government, as a measure of public investment.9 The extension of the basic model to include public investment variables is motivated primarily by the Uzbek government's view that its strategy of diversi- fying economic output away from agriculture and raw materials and toward the industrial sector, with a view toward substituting imports, has been a crucial factor in explaining Uzbekistan's relative success.10 In addition to attracting some foreign direct investment (FDI), much of this import substitution and industrialization strategy took the form of government-directed and financed capital investment. Indeed, capital expenditures of the general government have been relatively high, particularly in the later years (12.5 percent of GDP in 1995 and 11.5 in 1996, according to IMF calculations based on the Uzbek authorities' data). Two stories motivate the extension of the model by agricultural commodities and natural resource variables beyond the proxies already used by Berg and others.11 First, production of these goods, which could either be sold for hard currency or may have reduced the need for hard currency imports, could have allowed Uzbekistan to relax the tight external financing constraint, and corresponding import constraint, that was typical for other economies in the region. As a result, Uzbekistan may have been in a better position to maintain production in traditional industries, by purchasing inputs and capital goods that would otherwise have stopped flowing following the disintegration of the Soviet Union (see IMF, 1997, paragraph three). The second story is closely related, but focuses more on the self-

9This variable was used in spite of problems with cross-country consistency (its exact definition depends on national fiscal authorities, and may vary from country to country) because gross fixed capital formation in the public sector, which is taken from the national accounts, is not available for Uzbekistan and several other transition countries in our sample. See Zettelmeyer (1998) for the exact definition and sources of the new data used in this section. 10See the official publication, "Islom Karimov Steers Uzbekistan on its Own Way" (1997). 11Namely, the share of agriculture in GDP prior to transition and a dummy for large raw material pro- ducers: Russia, Kazakhstan, Azerbaijan, and Turkmenistan. Thus, the natural resource dummy used by Berg and others lumps Uzbekistan with the resource poor countries.

281

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer sufficiency and not so much on the foreign exchange implications of domestic energy production. This view stresses that the centrally planned supplier relationships of the former Soviet Union could often not be quickly replaced by markets and international trade, particularly in the Central Asian republics.12 Bilateral trade and barter arrangements, which were put in place in an attempt to maintain Soviet era goods and materials flows between the former Soviet republics, were unreliable and plagued by inter-republican non-payment problems, especially in the energy sector. In this setting, self-sufficiency in certain inputs, in particular energy, may have played a special role that would gradually fade as markets developed and trade was redirected to countries outside the former East bloc. The remainder of the paper proceeds in two steps. First, the new variables are given a maximum chance of "resolving" the Uzbek growth puzzle by not only adding them to the model by Berg and others used before, but by redoing the general-to-specific model selection methodology in the presence of these variables.13 We see which, if any, of the new variables survive the selection process, and whether or not the "growth puzzle" re-emerges in the context of the revamped model. Second, we test the hypothesis that the improvement in the model's ability to fit the Uzbek experience is due to the fact that the new variables are merely proxying an "Uzbekistan effect," which we still have failed to properly identify. This is achieved by checking the robustness of the earlier results.

The Growth Puzzle Revisited The following compares fitted growth paths for Uzbekistan and the average of other BRO economies based on models derived through an analogous procedure as the model used so far, that is, beginning with a very wide set of variables—which now include the commodity, energy, and investment variables discussed above—and then simplifying (eliminating or restricting variables) in the same basic order as Berg and others.14 To deal with the problem that energy production is probably endogenous to same-year industrial activity, and thus to output, first lags are used, either directly or as instruments. The new variables were simplified last, as they are of special interest in this paper and we want to give them a maximum opportunity of playing a role in the final model. The set of surviving variables was somewhat sensitive to variations in the order of elimination, and in particular, there are two alternative final models with different statistically significant sets of the new variables. The coefficients for these two sets are shown in Table 4 (see Appendix for the full models).

12This is closely related to ideas explored by Blanchard and Kremer (1997), who emphasize the breakdown of specific relationships in the absence of fully developed markets as a main factor behind the output decline. 13The presence of the new series may have a bearing on which other variables (in particular, within the set of initial conditions) enter the final model and how they enter it. Repeating the model selection process rather than simply tacking on the new variables thus allows a more precise estimation of the new coefficients and improves the fit of the model. 14For a complete list and definition of the variables introduced, including those that did not survive the elimination process, see Zettelmeyer (1998). Note that the output growth data was also revised model by Berg and others that was used in Part I is based on April 1997 data. While this had some effect on the estimated coefficients, it does not affect any of the conclusions.

282

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE

Table 4. Energy and Agriculture Coefficients in Two Variants of Extended Model (dependent variable: real output growth, in percent) Model Variables Coefficient t-value A Cotton production value ($ per capita) 0.050 2.394 Energy self-sufficiency index (lag)1 2.727 1.704 Energy exports index (lag)1 -2.878 -2.030 B Cotton production value ($ per capita) 0.062 3.133 Value of non-cotton agricultural commodities ($ per capita) -0.047 -3.246 Energy exports index (lag)2 -3.384 -2.448 Note: A and B also differ with respect to some variables not shown in the table. For the full models, see Appendix, Table A1. 1Defined as the ratio of energy production over energy consumption (both in energy units) if this ratio is smaller than one and as one if the ratio is bigger than one. First lags were used to avoid endogeneity (see footnote 8) 2Deilned as the difference between the ratio of energy production over energy consumption and the energy self-sufficiency index. First lags were used.

Table 4 shows a positive effect of cotton production and a negative effect of non-cotton agricultural production (mainly wheat), although only the former is robust across the two variations of the model. One interpretation could be that cotton was more internationally marketable and/or less subject to barter arrangements than wheat and thus more likely to lead to actual foreign exchange earnings. Also, in many transition economies wheat production went along with subsidies to consumers, while cotton earnings were often used to subsidize industry.15 Energy self-sufficiency has the expected positive sign in model A, but was insignificant and eliminated in model B. In contrast, the model finds a negative effect of energy exports in both variations. The last two findings contradict the view that energy production matters mainly as a way of generating cash, but are consistent with the idea that there may have been a special advantage to having one's own inputs in a period when traditional interrepublican trade patterns were disrupted and new trade patterns had yet to be formed. This said, the negative coefficient on energy exports remains something of a puzzle, though perhaps a puzzle with precedents.16 Public capital expenditure did not survive as a determinant of growth in either version.17 This could be because this variable is truly unrelated to growth in transition, perhaps because the state tends to direct investment to the wrong

15I thank Peter Keller for suggesting this interpretation. 16Two well-known examples for the actual or potential counterproductiveness of resource riches are the Dutch disease and the negative impact of large natural resource endowments in long-term growth regressions. On the latter, see Sachs and Warner (1995). 17Because public investment data was not available for the whole sample, the capacity of this variable to explain growth was explored in the context of a general-to-specific exercise performed on a subsample. After finding that public investment was not significant (even when ordered at the end of the elimination process) the exercise was repeated on the whole sample without controlling for public investment. Models A and B are based on this second exercise.

283

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

Table 5. Uzbekistan and Transition Economy Average: Fitted and Actual Growth Paths

(in percent per year) Transition Time 0 1 2 3 4 Model A Average of Baltics, Russia, and other countries of the Former Soviet Union (BRO), excluding Uzbekistan Actual growth -22.3 -12.9 -13.4 -4.1 -1.0 Fitted growth -22.3 -12.7 -12.5 -3.2 -1.1 Residual 0.0 -0.2 -0.9 -0.9 0.2 Average of absolute residual 2.3 3.2 4.8 3.1 5.2

Uzbekistan Actual growth -11.1 -2.3 -4.2 -0.9 1.6 Fitted growth -10.0 -2.2 -8.9 -0.2 -2.2 Residual -1.1 -0.1 4.7 -0.7 3.8 Absolute residual 1.1 0.1 4.7 0.7 3.8

Model B BRO Average, excluding Uzbekistan Actual growth -22.3 -12.9 -13.4 -4.1 -1.0 Fitted growth -22.2 -13.2 -12.6 -3.9 -1.4 Residual -0.1 0.3 -0.8 -0.2 0.4 Average of absolute residual 2.3 3.1 4.1 2.9 5.3

Uzbekistan Actual growth -11.1 -2.3 -42 -0.9 1.6 Pitied growth -11.6 -0.6 -8.4 0.2 -1.5 Residual 0.5 -1.7 4.2 -1.1 3.1 Absolute residual 0.5 1.7 4.2 1.1 3.1 industries.18 Alternatively, it is possible that the variable is so mismeasured (in the sense of cross-country inconsistencies; see footnote 9) that any positive effect is biased toward zero and undetectable. The next step is to see how well the two models explain the Uzbek output path. Table 5 is the equivalent of Table 2 for models A and B. As Table 5 shows, the ability of the two models to fit the Uzbek growth experience is almost the same, with very similar paths of residuals for Uzbekistan. Both models still have some difficulty in explaining why Uzbek output declined so little in 1994 (transition year 2) and why it began to recover in 1996 (transition year 4).19

18The conventional interpretation that public investment crowds out private investment through a macroeconomic (interest rate) effect is less plausible here, as both models A and B control for the fiscal balance. 19Note that the ability of models A and B to predict the Uzbek recovery in 1996 is slightly worse than that of the model by Berg and others (the latter predicted zero growth; the models above slightly negative growth). As a matter of model mechanics, this is just an artifact of the fact that the ratio between energy production and consumption sharply increases for Uzbekistan in 1995, making Uzbekistan an energy exporter according to the definition used in this paper. From Table 4, it is clear that the latter has a negative impact on fitted growth for 1996. The question what drives the modest turnaround in growth in 1996 can thus not be answered based on the regression model used in this paper, and is addressed in a companion paper (Taube and Zettelmeyer, 1998), by examining sectoral growth patterns.

284

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE

However, the main result from the table is that, based on the criteria used in Section I to decide whether a "growth puzzle" existed, the Uzbek growth puzzle vanishes. First, the residuals for Uzbekistan are no longer all on one side; that is, some are positive and some are negative. Thus, Uzbek growth during transition is no longer systematically underpredicted. Second, as is apparent from comparing the lines showing absolute residuals, the model now actually does somewhat better in fitting the Uzbek path than it does in fitting the path of the average BRO economy. Given that the model was extended by including variables suspected to contribute particularly to explaining the Uzbek experience, this is perhaps not surprising. Note, however, that the ability of the model to explain growth in the BRO economies other than Uzbekistan is still at least as good as in the model used by Berg and others. As one would expect, the much milder output decline in Uzbekistan relative to the average BRO country is now attributable to both the initial conditions group (maintaining the same definition as in Table 3) and the new set of energy and agriculture variables (Table 6). As in Table 3, Uzbekistan's macroeconomic and structural policies would ceteris paribus have lead to a lower output path relative to the average for the Baltics, Russia, and other countries of the former Soviet Union. This is more than offset, however, by the effect of cotton production and (in model A) energy self-sufficiency, as well as by more favorable initial conditions (as before, mainly low industrialization). The relative advantage imparted by the initial conditions is again concentrated in the first two years, but the positive impact attributed to the new variables is much more sustained.

Robustness Before concluding, a methodological caveat needs to be addressed. Suppose that the Uzbek puzzle was in fact attributable to some yet unidentified variable that happened to be correlated with the "new variables" identified in the previous section, merely because they take on unusual values for Uzbekistan. Then, this could generate the results of the previous section. To take an extreme example, suppose that Uzbekistan were the sole transition economy producing cotton. Then, the inclusion of cotton production in the regression model would amount to including an Uzbekistan dummy, which we know would be highly significant and resolve the "puzzle"—even if the mildness of Uzbekistan's output decline had entirely different causes. Fortunately, this possibility can be tested by re-estimating the model after excluding Uzbekistan from the sample and seeing how this affects the outcome (Table 7). Table 7 sends a mixed message. With one exception, all the energy and agriculture coefficients in Table 7 lose their statistical significance when estimated without the Uzbek sample points. They also drop in value. Thus, it is correct to say that the strength of the estimated effect of the energy and agriculture variables is driven by the Uzbek "outlier." But while the coefficients drop in value, they are, in economic terms, still quite close (between 50 and 80 percent of the values based on the full sample). Moreover, the fact that they are estimated too imprecisely to be significantly different from zero cuts both ways—it implies that the old values are well within the standard error of the new values. Thus, the coefficients and t-values

285

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

Table 6. Uzbekistan and Transition Economy Average: Contributions of Major Groups of Variables to Fitted Growth

(in percent per year) Transition Time 0 1 2 3 4 Model A Average of Baltics, Russia, and other countries of the former Soviet Union (BRO), excluding Uzbekistan Fitted growth -22.3 -12.7 -12.5 -3.2 -1.1 Macroeconomic policy -1.3 2.3 2.2 1.6 1.9 Structural reforms 9.7 9.0 11.0 13.6 13.6 Initial conditions + constant -28.5 -21.7 -26.1 -20.1 -17.4 War -3.4 -3.4 -0.8 -0.2 -0.4 New variables 1.2 1.2 1.3 1.9 1.2 Cotton 0.7 0.7 0.9 1.0 0.4 Energy 0.5 0.5 0.4 0.8 0.8

Uzbekistan Fitted growth -10.0 -2.2 -8.9 -0.2 -2.2 Macroeconomic policy -5.8 1.1 0.5 0.8 0.7 Structural reforms 7.8 3.3 7.7 9.8 10.9 Initial conditions + constant -18.3 -13.1 -24.6 -19.7 -19.9 War 0.0 0.0 0.0 0.0 0.0 New variables 6.4 6.5 7.6 8.9 6.2 Cotton 3.9 3.9 5.0 6.2 4.1 Energy 2.5 2.6 2.5 2.7 2.1

Model B BRO average, excluding Uzbekistan Fitted growth -22.2 -13.2 -12.6 -3.9 -1.4 Macroeconomic policy -1.8 2.1 2.2 1.5 1.7 Structural reforms 7.1 6.9 7.4 10.2 11.3 Initial conditions + constant -23.3 -17.6 -20.1 -14.3 -12.0 War -2.7 -2.7 -0.7 -0.2 -0.3 New variables -1.6 -1.9 -1.5 -1.1 -2.1 Cotton 0.8 0.8 1.1 1.3 0.5 Non-cotton agri. commodities -1.5 -1.9 -1.5 -1.7 -1.9 Energy -0.9 -0.9 -1.0 -0.7 -0.7

Uzbekistan Fitted growth -11.5 -0.5 -8.4 0.2 -1.4 Macroeconomic policy -6.8 0.8 0.7 0.8 0.6 Structural reforms 5.0 2.4 2.3 4.5 6.5 Initial conditions + constant -13.6 -7.8 -16.6 -11.5 -11.5 War 0.0 0.0 0.0 0.0 0.0 New variables 3.9 4.1 5.3 6.4 3.1 Cotton 4.8 4.8 6.2 7.8 5.2 Non-cotton agri. commodities -0.9 -0.8 -0.9 -1.3 -1.3 Energy 0.0 0.0 0.0 0.0 -0.8

286

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE

Table 7. Energy/Agriculture Coefficients With and Without Using Uzbek Data

(dependent variable: real output growth, in percent) Excluding Full Sample Uzbek data

Model Variables Coefficient t-value Coefficient t-value A Cotton production value ($ per capita) 0.050 2.394 0.025 0.79 Energy exports index (lag) -2.878 -2.03 -1.651 -0.887 Energy self-sufficiency index (lag) 2.727 1.704 2.186 1.266

B Cotton production value ($ per capita) 0.062 3.133 0.045 1.408 Value of non-cotton agricultural commodities -0.047 -3.246 -0.046 -3.109 Energy exports index (lag) -3.384 -2.448 -2.592 -1.411 shown in Table 7 could well be consistent with the hypothesis that they are alternative estimates of the same underlying coefficient. This is confirmed by a Chow test for predictive stability, which is nowhere near a rejection of the null of structural stability (p-values of 75 and 85 percent for Models A and B, respectively). On this basis, one should be inclined to take the previous results seriously, that is, go with the coefficients that were estimated on the whole sample. However, the possibility remains that the structural stability test might have failed to reject the null merely because of a lack of informative data in the sample that excludes Uzbekistan, and estimation based on the whole sample could thus give misleading estimates of the true coefficients on commodities and energy for the reasons discussed previously. To see what this "worst case" would imply for our ability to explain the Uzbek growth puzzle, consider the fitted values that would arise if the coefficients from the regression on the sample excluding Uzbekistan are used (Table 8). Does the growth puzzle re-emerge when using coefficients estimated on a subsample that excludes the Uzbek experience? It depends. Based on Model A, the finding that the model underpredicts Uzbek growth year after year still holds; based on Model B, this finding is true in four out of five years. However, the sum of absolute residuals for Uzbekistan is only insignificantly higher than that for the average BRO economy in Model A (19.3 versus 18.6), while Model B still does better at fitting the Uzbek growth path than that of the average BRO economy (14.2 versus 17.6). Thus, the capacity of the model to explain the Uzbek experience improves decisively after including agricultural commodity and energy variables in the model even if the coefficients are estimated on a sample that entirely ignores the Uzbek experience.

III. Conclusion This paper has two main findings. The first is that the exceptional mildness of Uzbekistan's transitional recession can be largely accounted for by a combination of its low degree of initial industrialization, its cotton production, and its near self-

287

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

Table 8. Uzbekistan and Transition Economy Average: Fitted and Actual Growth Paths Using Coefficients Estimated Excluding Uzbekistan

(in percent per year) Transition Time 0 1 2 3 4 Model A Average of Baltics, Russia, and other countries of the former Soviet Union (BRO), excluding Uzbekistan Actual growth -22.3 -12.9 -13.4 -4.1 -1.0 Fitted growth -22.2 -12.7 -12.7 -3.4 -1.0 Residual -0.2 -0.2 -0.7 -0.7 0.1 Average of absolute residual 2.4 3.2 4.8 3.1 5.2

Uzbekistan Actual growth -11.1 -2.3 -4.2 -0.9 1.6 Fitted growth -11.9 -43 -12.0 -3.7 -43 Residual 0.8 2.0 7.8 2.8 5.9 Absolute residual 0.8 2.0 7.8 2.8 5.9

Model B. Average of Baltics, Russia, and other countries of the former Soviet Union (BRO), excluding Uzbekistan Actual growth -22.3 -12.9 -13.4 -4.1 -1.0 Fitted growth -22.3 -13.1 -12.8 -4.1 -1.3 Residual -0.1 0.2 -0.6 0.0 0.3 Average of absolute residual 2.3 3.1 4.1 2.9 5.2

Uzbekistan Actual growth -11.1 -2.3 -4.2 -0.9 1.6 Fitted growth -13.0 -1.6 -10.4 -2.0 -2.6 Residual 1.9 -0.7 6.2 1.1 4.2 Absolute residual 1.9 0.7 6.2 1.1 4.2 sufficiency in energy. The relative importance of these factors, in particular the latter two, remains uncertain. Second, it is unlikely that the government's public investment program and import substitution strategy (except where it related to the energy sector) has played an important role in achieving Uzbekistan's favorable output performance. Specifically, no statistically significant effect of public capital expenditure on growth performance could be detected in a wide cross-section of transition economies; and the hypothesis that Uzbek growth obeys the same structural determinants as the other transition economies could not be rejected for a cross-country model that controlled for the agriculture and energy variables mentioned above (along with standard initial conditions and policy indices), but not for public investment and other Uzbek policy idiosyncracies such as import substitution. Several caveats remain. First, the negative results regarding the role of public investment and the failure to reject structural stability in the extended model could be

288

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE attributable to lack of power due to noisy data. Second, even accepting that the findings regarding public investment are correct, there remains an ambiguity in how to interpret the relative roles of policies and initial conditions in explaining the mildness of Uzbekistan's transitional recession. One interpretation is simply that Uzbekistan did relatively well because favorable initial conditions—broadly defined to include energy and cotton production—more than offset the effects of bad macroeconomic and structural reform policies. This interpretation would stress the finding that Uzbekistan's macroeconomic and reform policies are shown to contribute less to growth, ceteris paribus, than in other transition economies, as well as the failure to detect a structural break between the observations for Uzbekistan and the remainder of the sample, which suggests that the assumption of homogeneous policy effects across countries is justified. However, it is possible that the estimated effect of the energy and agriculture variables does not just reflect the availability of natural resources as such, but the impact of sectoral policies that tended to go along with these variables (controlling for macroeconomic stabilization and liberalization). Moreover, it remains true that the effect of energy and agriculture is weaker if Uzbekistan is excluded from the sample. On this basis, an alternative interpretation of the results is that Uzbekistan did relatively well in terms of aggregate output because it managed to mitigate the collapse of the (relatively small) industrial sectors by combining rigid state control with subsidies that were in large part financed by cotton exports and by developing the energy sector for domestic uses. While some other countries tried similar policies, particularly at the beginning of transition, these may have been less viable because they violated financing constraints at an earlier stage. As a result, there is no easy answer to the question of whether Uzbekistan could have done better by pursuing more vigorous liberalization and reform policies from the beginning. In the model used in this paper, faster reform would have led to higher growth through the measured macroeconomic and structural policy variables, reflecting mainly the positive impact of reforms on the newly developing private sector. However, if the interpretation is right that the contribution of the energy and agricultural variables reflect a combination of natural resources and the way in which they were exploited, then taking away part of this package—state control and cross-subsidization, which in the model go along with low structural reform indicators—might have led to a bigger output collapse, at least temporarily. In conclusion, while the results stress the importance of favorable initial conditions in explaining Uzbekistan's relative success, they allow for the possibility that this success was also related to Uzbekistan's sectoral policies, particularly during the early transition years. This need not imply that these policies were optimal given the circumstances,20 and even less that they should be continued. As the economic and social turmoil that resulted from the breakup of the Soviet Union subsides, it becomes ever harder to argue in favor of the extensive state control of economic decisions that has characterized the Uzbek experience so far.

20Given the disincentives to production implicit in Uzbekistan's policy approach (including in the agriculture sector), it is hard to imagine that Uzbekistan's approach was optimal even from the narrow perspective of the aggregate output effects of policies, that is, ignoring environmental and broader welfare issues. However, this is not a conclusion that can be narrowly based on the findings of this paper.

289

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

APPENDIX

Table Al. Models A and B Model A Model B Variable Definition Coefficient t-value Coefficient t-value Constant regression constant -18.99 -5.69 -7.78 -2.14 Fbal fiscal balance, in percent of GDP 0.81 5.37 0.91 6.27 lFbal l*Fbal -1.52 -3.31 -1.66 -3.76 Fbal-ls (first lag of Fbal)*s -0.07 -0.52 -0.06 -0.44 lFbal-ls 1*(first lag of Fbal)*s -0.52 -1.18 -0.64 -1.50 Fbal-2s (second lag of Fbal)*s 0.42 2.93 0.39 2.69 lFbal-2s l*(second lag of Fbal)*s -1.01 -2.73 -0.86 -2.31 Infa natural log of (1+average inflation) 3.20 2.55 3.43 2.70 Unfa l*Infa -5.79 -1.78 -6.03 -1.79 LII internal liberalization index 19.38 5.46 ILII-ls l*(firstlagofLII)*s 38.97 3.02 DLH-ls D[(firstlagofLII)*s] -19.74 -1.90 DILII-ls D[lLII-ls] 54.77 1.73 LIE external liberalization index 33.13 4.97 1LIE 1*LIE -64.84 -3.57 LIP-ls (first lag of private sector conditions index)*s -30.64 -3.21 lLIP-ls l*LIP-ls 48.16 2.54 DLIP-2s D[(second lag of pr. sector conds. index)*s] -30.11 -2.38 ^4.60 -2.84 DlLIP-2s D[l*(second lag of pr. sector conds. index)*s] 50.57 1.73 92.00 2.50 Warupd dummy variable for war or internal conflict -11.81 -6.97 -9.48 -5.58 lGrlniO l*(average pre-transition growth)*d -14.95 -3.32 -18.51 ^.16 dFbal-1 d*Fbal-l 1.68 3.42 1.22 2.63 dlFbal-1 d*lFbal-l -11.51 ^.84 -9.29 ^.16 dlnfa-l d*(first lag of Infa) -38.42 -3.69 -36.92 ^1.00 dllnfa-l d*l*(first lag of Infa) 125.66 2.94 115.50 3.05 RepInfDl pre-transition repressed inflation*Dl 0.84 3.14 IJ04 3.80 IRepInfDl l*RepInfDl -2.65 -2.81 -3.53 -3.79 NatRRD3 (resource-rich country dummy)*D3 -8.81 -4.81 -8.18 -4.91 UrbanDl (pre-transition degree of urbanization)*Dl -0.46 -AM -0.60 •^.64 lUrbanDl l*UrbanDl 2.67 3.45 3.36 4.05 TraddeptD2 (pre-transition trade dependency)*t*D2 -0.10 -3.99 -0.17 -5.65 TraddepO2 (pre-transition trade dependency )*O2 -0.15 -2.99 lUrbantDl l*UrbanDl*t -0.94 -2.18 -1.32 -2.89 AgSh89tD2c (1989 share of agriculture in GDP)*D2*(t-2) -93.76 ^.58 -73.44 -3.75 lAgSh89tD2c !*AgSh89tD2c 478.01 4.71 399.11 3.97 lOverlnd l*(initial over-industrialization index) 20.19 3.24 lOvIndtDlc lOverInd*Dl*(t-l) 177.65 3.97 202.09 4.34 Cotton VPC value of cotton production, $/capita 0.05 2.39 0.06 3.13 nonCononAgVPC value of non-cotton agricultural cash crops, $/cap -0.05 -3.25 Ebal-1 first lag energy balance index -2.88 -2.03 Esuf-1 first lag of energy self-sufficiency index 5.61 2.79 Eexp-1 Ebal-1-Esuf-1 -3.38 -2.45

290

©International Monetary Fund. Not for Redistribution THE UZBEK GROWTH PUZZLE

Notes:

The notation conventions used in variable definitions are as follows:

All variables are implicitly indexed by transition time t and country /. t denotes the transition year (t = -2,-1, 0, 1,... 7/,where 7} is the last transition year in the sample for country /). d denotes a dummy variable that takes the value 0 in transition years (t > 0) and 1 in pre-transition years (t < 0); s = 1-d (for all countries). £>[...] denotes the first difference operator. The prefix 1 denotes the estimated share of the private sector in GDP. Dj denotes a dummy variable that takes the value 1 for t smaller or equal / and 0 else; Oj = I - Dj (for all countries).

For a detailed explanation of the econometric methodology and motivation underlying the variable definitions, see Berg and others (1999). For a discussion of the structural reform indices and initial conditions (pre-transition variables) used in model A and B, their sources and construction, see Berg and others (1999); de Melo, Denizer, and Gelb (1996); and de Melo, Denizer, Gelb, and Tenev (1997). For discussion and sources of the energy variables in the table, see text and Zettelmeyer (1998). The agricultural variables in the table were constructed as follows. CottonVPC is the value of cotton production per capita using cotton lint production data from the FAO Yearbook Production, 1991-1996 volumes, and price data (Liverpool Index) from the IMF's International Financial Statistics. NonCottonAgVPC is the aggregate production value of the following crops: Wheat, Rice, Maize, Sorghum, Soybeans, Groundnuts and Tobacco, using data from the same sources. The standard regression statistics for the two models are as follows:

Model A: R2 = 0.87, DW = 1.66, RSS = 2231.7 for 34 variables and 143 observations Model B: R2 = 0.88, DW = 1.96, RSS = 2070.1 for 36 variables and 143 observations.

REFERENCES Aslund, Anders, Peter Boone, and Simon Johnson, 1996, "How to Stabilize: Lessons from Post- communist Countries," Brookings Papers on Economic Activity: 7, Brookings Institution. Berg, Andrew, Eduardo Borensztein, Ratna Sahay, and Jeromin Zettelmeyer, 1999, 'The Evolution of Output in Transition Economies: Explaining the Differences," IMF Working Paper 99/73 (Washington: International Monetary Fund). Blanchard, Olivier, and Michael Kremer, 1997, "Disorganization," Quarterly Journal of Economics, Vol. 112 (November) pp. 1091-1126. de Melo, Martha, and Alan Gelb, 1996, "Transition to Date: a Comparative Overview," in Lessons from the Economic Transition: Central and Eastern Europe in the 1990s, ed. by Salvatore Zecchini (Dordrecht-Boston-London: Klujwer Academic Publishers). de Melo, Martha, Cevdet Denizer, and Alan Gelb, 1996, "From Plan to Market: Patterns of Transition," World Bank Policy Research Paper No. 1564 (Washington: World Bank). de Melo, Martha, Cevdet Denizer, Alan Gelb, and Stoyan Tenev, 1997, "Circumstance and Choice: The Role of Initial Conditions and Policies in Transition Economies," World Bank Policy Research Working Paper No. 1866 (Washington: The World Bank).

291

©International Monetary Fund. Not for Redistribution Jeromin Zettelmeyer

Fischer, Stanley, Ratna Sahay, and Carlos A.Vegh, 1996a, "Stabilization and Growth in Transition Economies: The Early Experience," Journal of Economic Perspectives, Volume 10 (Spring) pp. 45-66. Havrylyshyn, Oleh, Ivailo Izvorski, and Ron van Rooden, 1998, "Recovery and Growth in Transition Economies 1990-97: A Stylized Analysis," IMF Working Paper 98/141 (Washington: International Monetary Fund). •, 1996b, "Economies in Transition: The Beginnings of Growth," American Economic Review, Papers and Proceedings (May) pp. 229-33. Hernandez-Cata, Ernesto, 1997, "Growth and Liberalization during the Transition from Plan to Market," Staff Papers, International Monetary Fund, Vol 44 (December), pp. 405-29. IMF, 1997, Republic of Uzbekistan—Recent Economic Developments. IMF Staff Country Report No. 97/98 (Washington: International Monetary Fund). i 1998, Republic of Uzbekistan—Recent Economic Developments. IMF Staff Country Report No. 98/116 (Washington: International Monetary Fund). Johnson, Simon, Daniel Kaufmann, and Andrei Shleifer, 1997, "The Unofficial Economy in Transition," Brookings Papers on Economic Activity: 2, Brookings Institution, pp. 159-220. Sachs, Jeffrey D., 1996, "The Transition at Mid Decade," American Economic Review: Papers and Proceedings, Vol. 86, (May), pp. 128-133. Sachs, Jeffrey, and Andrew Warner, 1995, "Natural Resource Abundance and Economic Growth," NBER Working Paper No. 5398 (Cambridge: Massachusetts: National Bureau of Economic Research). Selowsky, Marek, and Ricardo Martin, 1997, "Policy Performance and Output Growth in the Transition Economies," American Economic Review, Papers and Proceedings, Vol. 87 (May) pp. 349-353. Taube, Giinther, and Jeromin Zettelmeyer, 1998, "Output Decline and Recovery in Uzbekistan: Past Performance and Future Prospects," IMF Working Paper 98/132 (Washington: International Monetary Fund). Wolf, Holger C, 1997, "Transition Strategies: Choices and Outcomes" (unpublished manuscript; New York: Stern School of Business). World Bank, 1996, World Development Report 1996 (Washington: World Bank). Zettelmeyer, Jeromin, 1998, "The Uzbek Growth Puzzle," IMF Working Paper 98/133 (Washington: International Monetary Fund).

292

Monetary Policy and Public Finances: Inflation Targets in a New Perspective

CHRISTIAN H. BEDDIES*

This paper examines how the private sector, the monetary authority, and the fiscal authority interact and concludes that unrestricted central bank independence may not be an optimal way to collect seigniorage revenues or stabilize supply shocks. Moreover, the paper shows that the implementation of an optimal inflation target results in optimal shares of government finances—seigniorage, taxes, and the spending shortfall—from society's point of view but still involves suboptimal stabilization. Even if price stability is the sole central bank objective, a positive inflation target has important implications for the government's finances, as well as for stabilization. [JEL: E52, E62]

his paper examines the interplay between monetary and fiscal policies in an infla- Ttion-targeting framework. In this vein, the paper asks the following question: can an inflation target induce an independent central bank to provide the optimal rate of inflation, resulting in optimal seigniorage, taxes, public spending, and output? Does this also lead to optimal stabilization of aggregate supply shocks? The answer to the first question is yes, while the answer to the second is no, and the paper shows why. These issues have been analyzed in various ways. First, a strand of literature has focused on the interaction between monetary policy and the private sector, and thus on the credibility/flexibility trade-off.1 This approach, however, fails to take

*The author wishes to thank Sergio Pereira Leite, Roel Beetsma, Richard Disney, Haizhou Huang, Ivailo Izvorski, Chris Martin, Amlan Roy, and Daniel Trinder for helpful comments and suggestions on an earlier version of this paper. This paper was drawn from the author's Ph.D. dissertation; financial support by the Queen Mary and Westfield College, University of London, is gratefully acknowledged. 1See, for example, Barro and Gordon (1983), Rogoff (1985), Lohmann (1992), and Alesina and Grilli (1992).

293

©International Monetary Fund. Not for Redistribution Christian H. Beddies into account the impact of monetary policy on public finances. Second, other authors have employed a deterministic framework to explicitly model the interaction between monetary and fiscal policy. This approach has the weakness of disregarding the implications of aggregate supply shocks.2 Finally, the inflation- target literature aims at resolving the time inconsistency problem of monetary policy but tends to overlook the fact that inflation targets could be used as a way of providing the optimal level of seigniorage (see, for example, Svensson, 1995). The aim here is to merge these ideas to derive implications for the optimal policy mix and the optimal policy response to a supply shock.3 The paper extends the work by Beetsma and Bovenberg (1997) by allowing for an aggregate supply shock and by investigating the merits of inflation targets for public finances when the government interacts with an independent central bank. Beetsma and Bovenberg (1997), following along the lines of Alesina and Tabellini (1987), stress the importance of public debt and assume a constant ratio of real base money holdings to nondistortionary output, that is, the inverse of velocity.4 Within this framework, they analyze the implications of alternative institutional arrangements—centralization versus decentralization, Nash versus Stackelberg—for society's welfare. Whichever arrangement is preferable depends on society's preferences for inflation, output, and public spending, as well as the structural parameters of the economy, such as real base money holdings and outstanding public debt. This paper extends this analysis in several directions. First, it considers the link between monetary and fiscal policies in a stochastic model, that is, it includes an aggregate supply shock. Second, it provides intuition as to why it makes a difference whether the government faces a constrained optimization problem in which public spending is one of the arguments in the government's objective function, or whether public expenditure is given as a residual by substituting the budget constraint into the policymakers' objective functions. The implication of the constrained problem is that the central bank, when decentralizing its policies, does not automatically internalize the government's budget constraint. Thus it does not make a difference whether the bank cares about public spending, which is in contrast to the existing literature (e.g., Alesina and Tabellini, 1987; and Debelle and Fischer, 1994).5 The paper then shows how an inflation target can bring society closer to the second-best equilibrium by serving as a substitute for the central bank's disregard for the government's budget constraint. The paper's final extension is to analyze an "extreme" interpretation of the Maastricht proposal of price stability as the main objective of the European Central Bank (ECB) on a national

2See, for example, Tabellini (1986 and 1988), Alesina and Tabellini (1987), Jensen (1994), and Beetsma and Bovenberg (1997). Exceptions are Debelle and Fischer (1994), and Beetsma and Bovenberg (1999). 3A large body of literature has also focused on the seigniorage hypothesis as part of an optimal taxation problem. See, for example, Mankiw (1987), Fukuta and Shibata (1994), Froyen and Waud (1995), Gros and Vandille (1995), Evans and Amey (1996), and Click (1998). For an empirical investigation of developing countries, see Ashworth and Evans (1998). 4This parameter is assumed to be unity in the Alesina and Tabellini (1987) and Debelle and Fischer (1994) models. 5Except where the central bank would be "Stackelberg" leader with respect to the government.

294

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES basis. Again, a positive inflation target has interesting implications for smoothing the government's financing requirement over the sources of finance, as well as for stabilization. The analysis is formulated as a game involving the private sector, the monetary authority, and the fiscal authority. The main results can be summarized as follows. A social planner, when in charge of monetary and fiscal policy, can achieve only a second-best equilibrium, as lump-sum taxes are ruled out.6 The social planner then has to use alternative sources of finance—distortionary taxes, seigniorage, and the shortfall of public expenditure from its desired target. The resulting second-best equilibrium involves optimal positive mean inflation. Therefore, depending on the tax base—that is, the size of real base money holdings—raising seigniorage revenues to some extent appears optimal, which is in contrast to the various zero inflation rules studied in the literature. Since discretionary policymaking is ruled out, the optimal positive inflation rate derives from optimal revenue considerations and not from a desire to raise output via surprise inflation. Aggregate supply shocks cause inflation, taxes, spending, and output to fluctuate (second best) optimally around their respective means. The policy outcome under the assumption that a benevolent policymaker is in charge of monetary and fiscal policy serves as a benchmark case. Once policies are decentralized, that is, monetary policy is delegated to an independent but committed central bank, both financing and stabilization are distorted. Since the central bank does not optimize subject to the government's budget constraint and therefore ignores the social value of seigniorage, the entire financing requirement has to be met by the fiscal authority. The central bank does not provide any seigniorage revenues, either through budgetary considerations, or through a desire to boost output closer to its target through surprise inflation. Therefore, the fiscal authority has to rely to a greater extent on taxes—causing output to move further away from its desired target—and a larger expenditure gap. In terms of stabilization, inflation/seigniorage fluctuates less, while output and spending vary more. As a result, the social loss in this scenario is larger than under centralization. The way out of this dilemma is to impose a non-state-contingent inflation target on the central bank. The appealing feature of this target is that it provides the optimal level of expected seigniorage. This result highlights that any output effect in the targeting regime derives from lower taxation, since the amount of taxes necessary to finance a given financing requirement depends on the level of seigniorage provided by the central bank. The optimal inflation target is allowed to vary, depending on the base for the inflation tax. At the limit, where real base money holdings tend to zero, the seigniorage motive vanishes and the optimal inflation target becomes zero. In terms of society's loss, this solution—in which the central bank is independent but subject to an optimal inflation target—dominates the arrangement in which the independent central bank has no inflation target, but is still inferior to the centralized case. The last scenario is one in which controlling inflation is the sole objective of the central bank. While the model's inflation target ensures

6For an analysis with lump-sum taxes in the deterministic case, see Beetsma and Bovenberg (1997), and, in the stochastic case, Beddies (1997).

295

©International Monetary Fund. Not for Redistribution Christian H. Beddies that the means of inflation/seigniorage, output taxes, and spending are at their second-best level, the central bank does not stabilize supply shocks at all, leaving the entire burden of smoothing the supply shock to the fiscal authority. Regarding the social loss, this solution is inferior to the centralized setting and the decentralized setting with the inflation target. Whether this extreme form of central bank independence is preferable to a central bank that cares about output but is not subject to an optimal inflation target depends on the significance of supply shocks. The remainder of the paper is organized as follows. Section I sets up the basic model. Section II considers the social planner's problem as a benchmark case. Section III explores the decentralized setting and the implications of inflation targets. Section IV analyzes an "extreme" form of the Maastricht proposal for monetary policy—a framework in which the central bank only cares about inflation. Section V concludes the paper and gives some ideas of how to extend our model. The appendices provide derivations in support of our findings.

I. The Setup The model has three players, namely, the private sector (represented by a trade union), the monetary authority (central bank), and the fiscal authority (government).7 The trade union seeks to minimize deviations of the real wage rate from a particular target. For convenience and without loss of generality, this real wage target is normalized to zero. Thus, trade unions set the log of the nominal wage rate equal to the expected price level, that is, w = pe. To give the monetary and fiscal authorities an incentive to engage in surprise inflation, nominal wage contracts are assumed to be signed before the policies are selected. Our model is stochastic rather than deterministic, in contrast to Beetsma and Bovenberg (1997) and Alesina and Tabellini (1987). Thus, we allow for the possibility that the economy can be hit by shocks. Given these assumptions, normalized output, y, is given by8

y = n-ne - x + 8, (1) where y is the log of real output; n and ne denote the actual and expected rate of inflation, respectively; x is the tax rate on output; and E is an aggregate supply shock, distributed normally with zero mean and variance oE2. From equation (1), it follows that in a rational expectations equilibrium, where EHt-1(nt) = net, the long-run expected output level, denoted by the unconditional mean E(y), is equal to -t. To achieve E(y) = 0, one has to remove the distortions arising from output taxation. The model also allows for nontax distortions, which are measured by y* > 0.9 Note that y* represents the first-best level of output in the absence of any distortion. Hence the first-best output level y* can be achieved only by removing

7The basic model uses the framework of Beetsma and Bovenberg (1997). Also, see Beddies (1997). 8This is standard in the literature. See, for example, Debelle and Fischer (1994) and Beddies (1997) for the case with shocks, and Beetsma and Bovenberg (1997) and Alesina and Tabellini (1987) for the case without shocks. 9This could be labor market union power and/or goods market monopoly power. See Beetsma and Bovenberg (1997), who consider those nontax distortions as an implicit tax on output.

296

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES both the tax and the nontax distortions. The natural way to achieve the first best and to remove these distortions would be to subsidize output by setting j* = -x, whereby the negative tax represents the subsidy on output. This results in E(y) = y*9 which offsets the implicit tax on output caused by the nontax distortions. The preferences of the society are specified in a social loss function defined over inflation, output, and public spending.10 The social loss function is given by

1 2 2 2 *s = 2 [$5n +ty-y') + \is{g-g-) ]$s.\Ls>0. (2) where ^5 is the weight that the society places on inflation and [is represents the weight that the society places on public spending, both relative to the output objective; n is the rate of inflation; y is the log of real output as defined in equation (1); y* represents the first-best nondistortionary level of output; g is public spending; and g* denotes the spending target. For simplicity, the target inflation rate is assumed to be zero.11 The social loss (equation 2) is assumed to be an increasing function of the deviation from targets. The loss functions of the fiscal and the monetary authority can be defined in a similar way:

1 ^+{y-ff+liF{g-g*f}^F,ilF>0, (3) h= zv J_ 2 LM = [^ Hy-yi+»M(s-s*)%>v >o. (4) i\ M

The weights, corresponding to the respective targets in the social, the central bank's, and the government's loss function, may or may not differ. Within this public finance framework, the government has to choose its policies subject to a budget constraint. This budget constraint in terms of shares of nondistortionary output is given by12

g + (1 + r)b + (1 + r + ne - n)d = x + kn, (5) where g denotes government spending; b is the outstanding stock of indexed single-period government debt; and d represents the initial real value of nonindexed single-period debt. The right-hand side of equation (5) represents the sources of revenue. Thereby x is the revenue from distortionary taxes and kn is the revenue from seigniorage, with k > 0 as the constant ratio of real money holdings

10For an identical treatment of the social loss function, see, for example, Alesina and Tabellini (1987), Jensen (1994), Beetsma and Bovenberg (1997), and Beddies (1997). 11 Sections III and IV explore the situation where the central bank is allowed to have a positive inflation target. 12See Appendix I for details. In deriving this budget constraint, the paper follows Beetsma and Bovenberg (1997), but it does not analyze the case of unlimited access to lump-sum taxes. See Beddies (1997) on this issue considered within a stochastic model.

297

©International Monetary Fund. Not for Redistribution Christian H. Beddies and nondistortionary output.13 The key assumption underlying this budget constraint is that all debt sold at the end of the previous period has to be repaid, while no new debt is issued in the present period. One can interpret this as a two-period game in which the sole focus is the last period. This assumption has the advantage of simplifying the algebra substantially. Hence the issue of the intertemporal allocation of tax distortions, inflation, and public spending is ignored. To ensure that there is a demand for government debt, the return on indexed debt must be at least as high as the real ex ante return on an outside investment opportunity, r. Regarding the nominal debt, investors set expected inflation as a markup on the real ex ante rate r; thus, the nominal interest rate on nonindexed debt is r + ne (see, for example, Dornbusch, 1996). To ensure a clear separation between the government's sources of finance and the expenditures that have to be financed by these sources, that is, the government's financing requirement, the budget constraint (equation 5) is rewritten:14

FEEg*+/ + (l+r)(6 + J) = (T + /) + *7l + fe*-g) + (7C-7C«H (6)

The government has to finance the spending target; the output subsidy to (partly) offset the labor market distortions, y*; and the repayment and servicing costs of the indexed and the nominal debt. The right-hand side of equation (6) accounts for the source of finance: "revenues" from inflating away nominal debt, (71 - ne)d\ revenues via the shortfall of public spending from its target (g*- g); seigniorage revenues, kn; and, finally, revenues from explicit and implicit taxes on output, (x + y*).15 The paper assumes that the financing requirement does not exceed production. Finally, the private sector's expectations are assumed to be rational and hence satisfy (conditional on the information set available in the previous period, t - 1, that is, containing all information up to and including period t - 1) the following:

E,-l(7Cr) = TCf. (7)

II. A Benevolent Policymaker This section shall serve as a benchmark case for judging alternative outcomes in the decentralized policy setting. Suppose that a committed, benevolent policymaker is in charge of setting monetary and fiscal policies. She thus can take account of the private sector's expectations. The optimization problem is

13For simplicity, the potentially distortionary effects of the inflation tax on the demand for real money is not considered, thus k is not defined as a function of expected inflation. See, for example, Calvo and Leiderman (1992) and Calvo and Guidotti (1993) for models incorporating the money demand implications of the inflation tax, and Cagan (1956) in a hyperinflation framework. 14The spending and the output target on both sides in equation (5) are added and terms rearranged. 15Labor market distortions are measured by the deviation of the first-best output level, y*, from the actual output level, y, in the absence of any tax distortions, where E(y) would be zero.

298

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES characterized by minimizing the loss function (equation 2) subject to the budget constraint (equation 6), to the rational expectations constraint (equation 7) and to the supply function (equation 1). Hence, the Lagrangian is

1 L = min ^2+(7r_7C,_T + 8_/)2+|Li^_^2J (g) n,ne,x,gX,§ ^

+ X[F-(x + /)-/:7i-(g*-g)-(7r-^)j] + 5[£?_1(7Ur)-^], where n, x, and g are the instruments; X is the Lagrange multiplier associated with the government's budget constraint; and 8 is the Lagrange multiplier associated with the expectations constraint. Minimizing equation (8) with respect to TE, ne, x, g, X, and 8 yields the following first-order conditions:

e £,sn + (n -n -x + e-y*) + X(-k-d) + E(b) = 0. (9)

-(7i - ne - x + 8 - /) + Xd - 8 = 0. (10)

-(7C-7t*-X + £-/)-X = 0. (11)

[is(g-g*) + X = 0. (12)

F - (x + /) - kn - (/- g) - (n - n*)d = 0. (13)

e Et-i(nt)-n t = 0. (14)

Combining equations (10), (11), and (12) with equation (13), taking rational e expectations (note that Et_ \ (nt) = n t\ see equation 14) into account, and using equation (9), we obtain

E(b) = (F-kne J>e(l +

299

©International Monetary Fund. Not for Redistribution Christian H. Beddies

k%F \l k(l + k + d) kn = s e (16) ^5 + ^5 + M Ss + Ws + W + k + d)

V&sF ^s\(k + d)(l + k + d)] x + f = ei (17) ^5+^5+M ^,s + ^s + ]is(l + k + d)

g*-g = & ^5 B (18) Ss + ^s + M ^s + ^s + ^s(l + k + df

VAsF ^s E. (19) y*-y- * + $ li + y. k2 s s s s ^s+^s+[ls(l+k+d)

Because the model abstracted from the unlimited access to lump-sum taxes, the above equilibrium is second best. In contrast to the literature dealing with policy games between the monetary authority and the private sector, the second-best optimal solution here involves optimal positive mean inflation. Depending on the size of k, taxing real base money holdings to some extent, in order to finance part of the public expenditures, appears to be optimal.16 By inspection of equations (16) through (19), one can verify that a higher government financing requirement raises the means of inflation and explicit taxes, while reducing the mean of public expenditure. Moreover, output moves farther away from its target, because of increased taxes, while the optimal relative variability between inflation, taxes, spending, and output is not affected. Hence, supply-side shocks are smoothed out over output and the three sources of finance, independent of the financing requirement. The social loss necessarily increases with a raise in the financing requirement, as all actual outcomes move farther away from their respective targets (see Appendix II on the social loss). Societies with a lower k (that is, higher velocity) experience a lower optimal mean and a lower variance of seigniorage (see equation 16) because the taxable base is smaller. Thus, in these countries, seigniorage is of less importance than in countries where real base money holdings are higher and, hence, the base for the inflation tax is larger. When regarding output deviations (equation 19) and spending deviations (equation 18) from their respective targets, the opposite is true. Means and variances are higher if k is smaller. Not surprisingly, the mean of implicit and explicit taxes (equation 17) also increases when k becomes smaller, while its variance decreases.17 The consequence of reduced accessibility to

16In Andrabi (1997), seigniorage passively adjusts to the budget constraint as a residual tax, while this paper treats it as an instrument. However, his setup is purely decentralized. 17The effect on the tax variance stems from the reduced inflation variance, which is already putting pressure on output stabilization.

300

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES seigniorage is that the government's financing requirement has to be met by less spending and increased taxes, resulting in a larger gap between actual output and its nondistortionary target. The same argument applies to the task of stabilizing supply shocks. At the limit, where k tends to zero, it is no longer optimal to use seigniorage as a source of finance. Hence, inflation responds only to the supply shock to maintain the optimal relative variability among output, inflation, and the remaining financing sources. As a result, F needs to be financed entirely by implicit and explicit output taxes and by the shortfall of spending from its target. If society views inflation as especially important and consequently increases the weight attached to inflation, £s, it can reduce the mean of inflation as well as its variance. This "gain," however, comes at the cost of higher mean distortionary taxes and, hence, less output, and it also moves public spending away from its target. It further induces output and spending to be more variable, and thus transfers the burden of stabilizing shocks from inflation to output and spending. The impact of different inflation weights on the variance of distortionary taxes depends on the parameters but has a likely negative sign, if society cares sufficiently about spending as well.18 A higher weight on public spending, us, decreases the gap between public spending and its target and reduces its variance. This implies that the means of seigniorage, output, and distortionary taxes have to increase to meet the financing requirement, because being more concerned about public spending diminishes its value as a financing source. The task of smoothing out the supply shock is increas- ingly transferred to inflation and output. The tax variance, however, is decreasing in the spending weight, us, in order not to put additional pressure on output. Higher nominal debt ratios undoubtedly increase the government's financing requirement. As a result, seigniorage/inflation is higher, distortionary taxes increase, and output as well as public expenditures move farther away from their respective targets. The fact that the supply shock is positively related to output and thus to taxation implies that debt is positively related to taxes, reducing the impact on output. The same argument holds for public spending. The impact of higher nominal debts on inflation is ambiguous but has a likely negative sign.19 This implies that the negative inflation response to the supply shock should be smaller. The economics behind this result is that unanticipated inflation is valuable for decreasing the real value of the nominal debt, which has to be repaid. As a positive shock reduces inflation, however, higher nominal debts imply that this response should be smaller.

III. The Impact of Central Bank Independence on Public Finances This section investigates an institutional arrangement in which the central bank is independent of the government. The underlying assumption here is that the monetary and the fiscal authorities move simultaneously; hence, they act in a (noncooperative) Nash fashion. The aim of this section is to investigate the possible advantages of inflation targets in improving society's welfare.

18More precisely, the tax variance is decreasing in the inflation weight if us(k + d) > 1. 19That is, the effect is negative if (1 + k + d)2/Es > 1 + 1/us.

301

©International Monetary Fund. Not for Redistribution Christian H. Beddies

Monetary Commitment: No Inflation Target

Many authors have argued that the inflation bias story is overdone. Why should an independent central bank have an incentive to fool the private sector if doing so does not help anyone?20 In this vein, the central bank is assumed to commit to sticking to the ex ante optimal policy. Thus, its optimization problem is characterized by minimizing the loss function (equation 4), subject to the rational expectations constraint (equation 7) and the supply function (equation 1). The Lagrangian is hence given by 1 L = min \Mn* + (n - if - x + 8 - y*f + nM(g - g*f] + 5[*M(*,)" 4 (20) 71,7^,8 2 1 where n is the central bank's instrument and 8 is the Lagrange multiplier associated with the rational expectations constraint. Minimizing equation (20) with respect to 71, ne, and 5, we obtain the following first-order conditions:

^MK + (n _ ne _ T + e _y) + £(5) = o (21)

- (7C - 71* - X + 8 -/) - 8 = 0 (22)

£,_!(*,)-*? = 0. (23)

Note that the above first-order conditions show that it makes no difference whether the central bank cares about spending or not, that is, it does not matter whether \iM = 0 or not. The government's optimization problem is given by minimizing the loss function (equation 3), subject to the budget constraint (equation 6) and the supply function (equation I).21 Thus the Lagrangian is

l_\ 2 L = min ^ 2 ( _ ,_ _ ) ^_^] (24) n,g,X 2 7C + 7r 7l x + e / +

+ X[F-(T + f)-kn-{g*-g)-{n-Tie)d].

By minimizing equation (24) with respect to x, g, and X, we obtain the following first-order conditions:

-(n-ne-x + E-y*)-X = 0 (25)

liF(g-g*) + X = 0 (26)

F - (x + /) - kn - (g*- g) - (7i - n*)d = 0. (27)

20See, for example, McCallum (1995); Clark, Goodhart, and Huang (1996); and Blinder (1997) on this issue. 21 Note that the government does not choose inflation. Thus it cannot take account of the private sector's rational expectations.

302

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES

The equivalent of equation (15) in Section II is obtained by combining equations (22), (25), and (26) with equation (27), taking rational expectations into e account (note that Et- \(it,) - % t\ see equation 23) and using equation (21):

e E(5) = (F-kn \\J±-) = F (28) .l + |i, J i+lV where equation (28) is the (average) marginal cost of expected inflation. Note that central bank independence implies that the central bank does not internalize the government's budget constraint and thus does not have any temptation to devalue the nonindexed debt d. Technically, the term (1 + d) in equation (15) does not appear in equation (28). Substituting equation (28) into equation (21), using equations (25) through (27), and assuming that [if = [is, one can solve for the policy outcomes:

hi = - M e (29) ^M + ^s + \is(1 + k + d)

lM+[is{k+d) x + f = BL+ + + l e (30) i+[is Z>M ^s Vs( + k + d)

F Zu g*-g = £ (31) l + [ls ^M+^S+^s(l + k + d)

f-y = \±sL ^M e. (32) i+[is SM+^MV-s + Vsil + k + d)

Decentralization here has obvious effects. The central bank does not internalize the budget constraint of the government and hence ignores the social value of seigniorage as a source of finance.22 As is easily seen from equation (29), inflation— and thus seigniorage—merely fluctuates around a zero mean. Hence, the zero inflation rules, as, for example, studied by Rogoff (1985) and Lohmann (1992), fail to consider that, as long as base money holdings are positive, inflation has some social value as a source of taxation. As a result, the entire burden of meeting the government's financing requirement rests on distortionary taxes, leading to a greater output shortfall, caused by insufficient subsidies, and the spending shortfall. This result can also be looked at in a more technical manner. The model is dealing with a constrained optimization problem, in which the government

22Also, see Beetsma and Bovenberg (1997) and Beddies (1997).

303

©International Monetary Fund. Not for Redistribution Christian H. Beddies chooses its policies subject to a budget constraint. In contrast to this view, some other authors, such as Alesina and Tabellini (1987), Jensen (1994), Debelle and Fischer (1994), and Huang and Padilla (1995), transform the constrained optimization problem into an unconstrained one by substituting the budget constraint into the loss function (via spending). As a result, a central bank has to internalize the government's budget constraint if it cares about public spending (UM > 0), that is, if it has the same preferences as society or the government.23 However, in practice, why would the independent central bank optimize subject to the government's budget constraint? Therefore, this paper's definition ensures that there is no need to justify why preferences among society, the government, and the central bank should be different—as opposed to, for example, Debelle and Fischer (1994), who merely assume different preferences.24 Assuming that EM = £>s> the paper finds that stabilization also differs from the second best. With an independent central bank, the variance of inflation/seigniorage is lower, while the output and spending variances are higher [compare equations (16), (18), and (19) with equations (29), (31), and (32)]. The intuition behind this result is as follows. As the central bank does not internalize the government's budget constraint, it fails to account for the effect of unanticipated inflation on the value of repayable nominal debt. For that reason, inflation responds to a lower extent to the supply shock, producing a higher variability of output and spending. Furthermore, the variance of implicit and explicit taxes (x + y*) is only lower with the independent central bank if us(k + d) > 1 (see equations 17 and 30). Thus, it appears that low-debt countries can lower the variability of their tax system by centralizing policies. High-debt countries, however, are better off in terms of the tax variability by decentralizing policies, given that k is equally low. The impact of a change in the structural parameters of the economy, k and d, and the political parameters, <^s and us, on the means and the variances of seigniorage, taxes, the spending shortfall, and the output shortfall are the same as discussed in Section II, except for two important differences. First, the mean sources of finance in the decentralized setting do not depend upon k. However, at the limit, if k tends to zero, that is, seigniorage is of insignificant importance, decentralization "seems" to be attractive. The means of the sources of finance (seigniorage, taxes, and the spending shortfall) coincide with those under centralized commitment (compare equations 16-19 with equations 29-32 for k —> 0). Second, the variance of inflation/seigniorage is strictly decreasing in d, no matter how important society views public spending and output relative to inflation. The reason for this result is that the central bank does not balance the impact of unanticipated inflation on repayable nominal debt against the preferences of society. Regarding the welfare of the society, within this setup benevolent policymaking is preferable to central bank independence (see Appendix II for technical details).

23This argument necessarily disappears in the centralized setting. 24Thus, in their model, with respect to discretionary policy, fiscal parameters enter the inflation out- turn via the tax effect on output and not through revenue considerations.

304

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES

Monetary Commitment: An Optimal Inflation Target

In practice, most central banks, at least in the industrialized countries, are more or less independent of the government. Some governments, however, still have the power to set the targets for their national monetary policy. We will use this observation to examine whether the finance dilemma analyzed above can be solved by the use of an inflation target. Suppose the central bank would choose its policy subject to the following objective function:

1 T (33) h =21U* - * f+(y - ff+M« - rf} SM> ^M > o> where the only difference from equation (4) (and equation 2 if assuming equal weights) is the target inflation rate nT. Since we have shown in the previous section that the central bank does not internalize the government's budget constraint and thus ignores the social value of seigniorage, we examine now whether nT can be chosen in such a way that the central bank provides the optimal level of seigniorage. Going through the same steps as earlier, we can show that by setting nT equal to 2 &|i5F/(£s + ^sP-s + Vsk ) and assuming that (if = [is and t,M = Is, the optimal mean inflation rate resulting from benevolent policymaking (Section II) can be obtained:

k%F kn = 2 M e (34) ^s + ^s\xs + lisk $s+iis)is+\Ls(i+k+d]

x + y* = VAf S5 + Hs(* + <0 2 + l + 1 + + d e (35) ^s+Xs\is + \isk *s + 5s*s M * )

& $5 g*-g = 2 e (36) ^5+^5 + M $s + $JLs + \ls(l + k + d)

V&S V£s e. (37) y*-y = 2 '^ + ^ + \i (l + k + d) 4S+^S+M s s s

The major drawback of this analysis is that the implementation of a positive inflation target results in the optimal mean shares of finance, as if a benevolent policymaker had chosen monetary and fiscal policies. Thus, compared with the centralized case, an independent central bank can be induced to deliver the optimal level of mean seigniorage merely by implementing an optimal inflation target.25 The mechanism behind this result is as follows. With the inflation target, the

25Note that this is different from the Svensson (1995) inflation target that is imposed to remove the inflationary bias that arises from discretionary policymaking.

305

©International Monetary Fund. Not for Redistribution Christian H. Beddies central bank provides the optimal level of financing to the government. Since, as a result, government taxes can be lower than in the case without the inflation target, the output shortfall, as well as the spending shortfall, is smaller. It is important to note here that positive mean inflation does not derive from an incentive to boost output via surprise inflation, since it was assumed that the central bank is committed for the reasons explained above, but from optimal revenue share considerations imposed by the inflation target. The fact that this paper's inflation target is optimal, in the sense that it ensures the optimal share of finances, is best seen by looking at the parameter k. If k = 0, the equations in Section II show that the optimal inflation rate—and, thus, seigniorage—is zero. The above analysis easily verifies that, in this scenario, the optimal inflation target becomes zero, resulting in mean shares of finance, as if the benevolent policymaker had been in charge. Since the inflation target is non-state-contingent, stabilization is still suboptimal compared with the benevolent case. Regarding society's welfare, the targeting regime of this section is still inferior to that of the social planner. However, a comparison between the losses resulting from pure central bank independence and central bank independence with the optimal inflation target shows that the targeting regime is preferable to the regime without the inflation target (see Appendix II for formal details).

IV. Inflation as the Sole Objective of Monetary Policy The desire of some European countries to establish a European Central Bank (ECB) that is especially concerned with inflation—that is, concerned about low and stable prices—is the motivation for this section.26 On a national level, the monetary requirement for participating in the Economic and Monetary Union (EMU) is the establishment of an independent central bank whose main concern is seen to be inflation. Following along these lines, this section considers an extreme central bank that is only concerned with inflation.27 This case is represented by the specification in equation (38) below.28 Technically, this coincides with the assumption of an infinitely conservative central banker.

A General Inflation Target Let the modified objective function of the (conservative) central bank take the following form:29

26To achieve the ultimate goal of price stability, the ECB targets money growth like the Bundesbank, for which authors such as Bernanke and Mihov (1996) and Clarida and Gertler (1996) find evidence that it would be better characterized as an inflation targeter, rather than a monetary targeter. 27Nevertheless, in terms of the loss function, Alesina and Grilli (1992), when examining the ECB, follow Kydland and Prescott (1977), Barro and Gordon (1983), and Rogoff (1985). 28Note that an arrangement such as EMU coincides with one of centralized monetary policy and decentralized fiscal policy. See, for example, Sibert (1994). 29Rankin (1998) also captures the idea of (extreme) conservatism in this way. However, he does not consider the possibility of having a positive inflation target or shocks.

306

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES

1 2 C (38) L M = 2 n-n?] , where nT denotes the central bank's inflation target. Given equation (38), one can immediately establish the solution to the central bank's problem:

T JI = n , (39) which implies that inflation is always at its target. The government still faces the optimization problem of the previous section. Thus, its first-order conditions are still given by the equations (25)-(27). Solving equation (39) and equations (25)-(27) jointly and imposing the condition of rational expectations, equation (7), one arrives at (assuming that \iF = \is) the following:

kn = knT = kne, (40)

T \iskn |ISF 1 x + f = + + £ (41) l + lxs 1 + H5 i + Hs

knT F 1 g*~g = - + e (42) i+^; 1 + HS l + \L5

nT Vs \is f-y = - _hL E. (43) 1+Jl/ i+ns i+^s

These results, characterized in equations (40)-(43), have some interesting implications. Whatever target inflation rate the central bank has in mind, inspection of equation (41) immediately shows that any positive rate of inflation reduces the necessity for distortionary taxes, as long as k is positive. Furthermore, output and public spending are closer to their respective targets, equations (42) and (43). The above-derived solution, however, also reveals that the entire burden of stabilizing the aggregate supply shock lies on fiscal policy and output. The reason for this is that unanticipated inflation is not available. The central bank does not care about output, given the specified loss function (equation 38). Necessarily, a government concerned about meeting fiscal criteria such as those defined in the Maastricht Treaty faces trade-offs among higher taxes, lower expenditures, and lower output subsidies.

A Specific Inflation Target Turning to the central bank's inflation target, 7Cr, if the government can impose an inflation target that provides the desired level of seigniorage, as specified in the second-best equilibrium of Section II, straightforward calculation reveals that

k^F kn = (44) ?S+^5+M

307

©International Monetary Fund. Not for Redistribution Christian H. Beddies

1 HAP e (45) x + f = 2T ls+^s + M l+\ls

l,F 1 e (46) g*-g = 2 ^5+^S + M 1 + ^5

F ^s »s y*-y = 2 8, (47) ^s+^slis+[isk l + Hs which gives the same shares of the mean financing sources as in the case where the benevolent policymaker was in charge. Thus, although the central bank does not have output in its objective function, the imposed inflation target ensures that expected output is at its second best level. Implicitly, the central bank, by implementing the target, acts as if it cared about average output and the government's budget constraint. The intuition behind this result is as before, namely, that by providing the optimal level of seigniorage, the government has to rely to a lesser extent on taxes. The decreased reliance on taxes has a positive effect on output. Since this scenario considers a central bank that is committed to stick to the ex ante optimal policy, this result again highlights that any output effect derives from taxation, which itself depends on the level of seigniorage provided by the central bank. However, as the central bank is assumed to explicitly care only about inflation in its objective function, the inflation target cannot achieve optimal stabilization. Thus, extreme forms of central bank independence might not be optimal.30 Depending on the size of k, seigniorage should be a part of finance. The above-specified target ensures that, if k -» 0 at the limit, the seigniorage motive, as well as inflation, vanishes.31 Thus, the commitment/discretion discussion is not an issue as long as the government sticks to the inflation target. If the government cares about staying in office, why should it then not try to act in the interest of the private sector? In that connection, Goodhart (1993, p. 8) states: "As Lincoln said, you cannot fool all of the people all of the time." The social loss under this arrangement, however, is higher than it would be in the case in which the bank also cares about output stabilization and faces the inflation target. Compared to the case where the central bank does care about output stabilization but does not face an inflation target, the welfare implications are ambiguous and depend on the size of the supply shock variance (again, see Appendix II for a formal derivation).

V. Conclusion This paper has focused on the interplay between monetary and fiscal policies within a stochastic framework. Its contribution to the literature is the examination

30See, for example, Goodhart (1993) for an excellent discussion of the issue of central bank independence. 31 The reason for this result is that the optimal inflation target in this scenario is zero. Note that this has nothing to do with the Svensson (1995) approach, where a "negative" inflation target (given that society's inflation target is zero) is required in order to remove the inflationary bias. The positive inflation target here is designed to provide the government with optimal seigniorage revenues, as in the previous section.

308

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES of the explicit inclusion of issues of stabilization in a public finance framework and the implications of inflation targets for public finances. The paper concludes that unrestricted central bank independence may not be optimal from society's point of view, regardless of whether the bank cares only about inflation or about both output and inflation. In terms of society's welfare, in the absence of a benevolent policymaker, the most appealing solution is to implement an optimal inflation target for the independent central bank, given that the bank also cares about output. Given the paper's way of specifying the preferences of society and the policymakers, neither a fully independent central bank without the above-specified optimal inflation target nor a central bank that cares only about inflation (even if given an optimal inflation target) will generate a preferable outcome from society's point of view, compared with an independent central bank mindful of all arguments in society's loss function and having an optimal inflation target. This paper assumed that all debt has to be repaid within the current period. In light of EMU, the model could be extended to investigate the Maastricht deficit criterion by allowing for debt accumulation. It could also be extended to allow for fiscal policy interactions between sovereign fiscal authorities within the union, which together interact with the centralized monetary authority, the ECB. Thus, one could focus on the public good character of fiscal policy. Since the European Community lacks a powerful federal government, one could investigate a situation in which the decentralized fiscal authorities can build coalitions to minimize the spillover effects of their fiscal decisions into other union countries. To capture the potentially distortionary effects of the inflation tax on the demand for real money balances, one could also endogenize the real base money holdings parameter. These issues are left for future research.

APPENDIX I

Derivation of Government Budget Constraint

As in Beetsma and Bovenberg (1997), real money balances in period t are given by MtIPt = kY*, where Y* is the output level in the absence of any distortions (the antilog of y*) and k > 0 is the constant ratio of real money holdings and nondistortionary output (given by k = Mtl(PtY*). 32 Hence, (Mt- Mt-1)/Mt = (Pt- Pt- 1)/Pt. Under the assumption that the tax distortions are not 33 too large, revenues from distortionary taxes can be approximated by ttPtY*. Lump-sum taxes are given by 0tPtY*. Denoting by Gt the level of government spending, by Bt the amount of indexed single-period debt, by Dt the amount of nonindexed single-period debt sold at the end of the previous period against the price Pt-1 and interest rates rBt and rDt, respectively, and, finally, by (Mt - Mt - 1), the increase in the nominal money supply, the nominal government budget constraint is given by (also, see Jensen, 1994)

PtGt + (1 + rBt)PtBt + (1+ rDt)Pt-1 Dt = ttPtY* + 0tPtY* + (Mt - Mt-1) + Pt(Bt+1 +Dt+1). (A1)

32The requirement for this to hold is that nondistortionary output Y* is independent of the tax rate t. Alesina and Tabellini (1987) or Canzoneri (1985) use an identical simplification. 33See, for example, Alesina and Tabellini (1987), who also use the approximation Y~ Y*.

309

©International Monetary Fund. Not for Redistribution Christian H. Beddies

Dividing equation (Al) by PtY*, using the result nt = (Pt - Pt-i)/Pt, and approximating (1 + fDt)Pt- \IPt by (1 +rot- TC/X the government budget constraint (equation Al) can be rewritten in terms of shares of nondistortionary output:

gt + (1 + rBt)bt + (1 + rDt - nt)dt = Tt + Qt + kKt + bt+{+dt+i. (A2)

To ensure that investors are willing to buy government debt, the interest rate on indexed debt should be as high as the ex ante real rate of return of an outside investment opportunity— r, say. Regarding nominal, nonindexed debt, the investor simply sets expected inflation as a markup on this ex ante real rate r to compensate for any expected inflation during the maturity of the debt. Hence, the nominal interest rate on nonindexed government debt is equal to r + ne. Thus the government budget constraint stated in equation (A2), dropping time subscripts and assuming that no new debt is issued, satisfies

g + (1 + r)b + (1 + r + ne - ri)d = T + 0 + hi, which (without lump-sum taxes) is the budget constraint stated in equation (5) in the text.

APPENDIX II

Comparison of Social Loss in Alternative Central Banking Regimes This appendix compares the social loss (equation 2) under benevolent policymaking, BP; central bank independence, CI; central bank independence with an optimal inflation target, CIT; and central bank independence where the bank is only concerned with inflation but has an optimal inflation target, CCIT. The social loss for the alternative regimes is obtained by substituting the respective policy outcomes into the social loss function (equation 2), assuming that players share the same preferences, that is, £, = ^ = ^F = ^M and [i = \is = \±F = JLIM. It is then straightforward to show that

LBP < LCIT < Lci (A3) LBP < LCIT < LCCIT

LCi g LCCIT

The proofs for these results are straightforward. The social loss for each regime can be cal- culated as

2 BP ^s? 1 L = J_ + 2 + ^s°l (A4) 2 ^s ^s + M 2 $5+^5+M1 + * + <02

2 CI 1 \1SF 1 ^s^s+^+^s)^ L = (A5) = 2T^ +2 2 " (^+yi5+^(i+*+j))

2 2 CIT 1 tsVsF 1__ ^sfe+^s+^K L = + 2 + (A6) 2 ^S+^5 M z ($5+yi5 + Jls(l+ * +

1 1 iisol LCCIT = w (A7) 2 ^5+^5+M2 + 21 + JV

310

©International Monetary Fund. Not for Redistribution MONETARY POLICY AND PUBLIC FINANCES

Therefore, it can be shown that

1 LH

2 1 ScH^ CT LBP_LC1T = 1 ^^S e (A9) + 2 + 2 2 2l=5 ^s+M •-+'t,s]is + \iA\ + k + d)

1 ^.^ 1 ^sfe + ^s+^sVl 2 2 2 2^s + 4#j+M (^s+^5|as+^(l + A: + rf))

1 2 41n|(jfc+tf) (i+n.)qg <0 2 2 •(^+^s+lis(i+*+d) )(^+^J+iis(i+ *+

1 g ^SP _ [CCIT - J_ %&*** ^5^5 e (A10) 2 + 2 2 ^+^5+M s + ^s+Hs(\ + k + df 1 2 1 ii ol tesF 2 s 2tc.+ic.ac + ack ' 21+lis

1 _|#5^L 1 ^5ge ? -^s+^s[is+lis(l + k + df >l+n5

2 2 _!_ ^5(l + fc+ flf) ae 2 <0 2(^+^H5+Hs(l + fc+ rf) )(l+ns)

+ 2 jci _ LOT _ LiiZ! ^s(^+^ ^K (All) zi+n/ 'fe+^+M1+^+'3f))2

i ^^ 1 ^H5^5+H5+^5^ f 2 O5+^5+M 2£s+$slxs + iis(l + k + d)f 1 H,F2 2 = tesF 2i + u, : Ss+^+M u.kzF2 I >0 2(4s+^s+nsF)(l + Hs

311

1 i ^M^+m+^n.W;? LCI_LCCIT = M% (A12) fl + Hs 2 (ks+ksiis+iis(\+k+d)f

1 ^sF2 2 1 Wl 2 %s+%s\ls+\lsk ~21 + ns

1 ^ 2 [2 1 + n5 2"^s+^+M J"

1 W^S+^+W + 1 Hs 3jsa Z: 2' 2 1 + jis (^s+^+H,(l + /: + rf))

Note that the coefficient on F2 is strictly positive while the coefficient on a| is strictly negative. Thus, if the supply shock variance is not too large, central bank independence when the bank has an optimal inflation target but cares only for inflation might be better than a central bank that stabilizes output but does not keep in mind the government's finances.

1 1 5sHcfes + Hc + $cJi>* LCIT _ LCCIT = . ^M* (A13) 2 42 2 5s + ^s+M ' ($s+S5Jl5+ll5(l+ * + £/))

1 ^F 1 M-5a^ L 2 "2 1 + jis 45+5sM's+M "

i ^Ks^+^+^K 1 ^5^e 2 2 1 + JlLn, 2,(£5+^l5+|Ll5(l + * +

References Alesina, Alberto, and Vittorio Grilli, 1992, "The European Central Bank. Reshaping Monetary Politics in Europe," in Establishing a Central Bank: Issues in Europe and Lessons from the U.S., ed. by Matthew B. Canzoneri, Vittorio Gulli, and Paul R. Masson (Cambridge: Cambridge University Press), pp. 49-77. Alesina, Alberto, and Guido Tabellini, 1987, "Rules and Discretion with Noncoordinated Monetary and Fiscal Policies," Economic Inquiry, Vol. 25 (October), pp. 619-30. Andrabi, Tahir, 1997, "Seigniorage, Taxation and Weak Government," Journal of Money, Credit, and Banking, Vol. 29 (February), pp. 106-26. Ash worth, John, and Lynne Evans, 1998, "Seigniorage and Tax Smoothing in Developing Countries," Journal of Economic Studies, Vol. 25 (No. 6), pp. 486-95. Barro, Robert J., and David B. Gordon, 1983, "A Positive Theory of Monetary Policy in a Natural Rate Model," Journal of Political Economy, Vol. 91 (August), pp. 589-610. Beddies, Christian H., 1997, "The Interaction of Fiscal and Monetary Policies in a World with Uncertainty," Queen Mary and Westfield College Discussion Paper No. 372 (London: Department of Economics, Queen Mary and Westfield College, University of London).

312

Beetsma, Roel M, and A. Lans Bovenberg, 1997, "Designing Fiscal and Monetary Institutions in a Second-Best World," European Journal of Political Economy, Vol. 13 (February), pp. 53-79. -, 1998, "Monetary Union Without Fiscal Coordination May Discipline Policymakers," Journal of International Economics, Vol. 45 (August), pp. 239-58. -, 1999, "The Interaction of Fiscal and Monetary Policy in a Monetary Union: Balancing Credibility and Flexibility," in The Economics of Globalization: Policy Perspectives from Public Economics, ed. by Assaf Razin and Efraim Sadka (Cambridge: Cambridge University Press). Bernanke, Ben S., and Ilian Mihov, 1996, "What Does the Bundesbank Target?" NBER Working Paper 5764, (Cambridge, Massachusetts: National Bureau of Economic Research). Blinder, Alan S., 1997, "What Central Bankers Could Learn from Academics—and Vice Versa," Journal of Economic Perspectives, Vol. 11 (Spring), pp. 3-19. Cagan, Phillip, 1956, "The Monetary Dynamics of Hyperinflation," in Studies in the Quantity Theory of Money, ed. by Milton Friedman (Chicago: University of Chicago Press), pp. 25-117. Calvo, Guillermo, and Pablo E. Guidotti, 1993, "On the Flexibility of Monetary Policy: The Case of the Optimal Inflation Tax," Review of Economic Studies, Vol. 60 (July), pp. 667-87. Calvo, Guillermo, and Leonardo Leiderman, 1992, "Optimal Inflation Tax under Precommitment: Theory and Evidence," American Economic Review, Vol. 82 (March), pp. 179-94. Canzoneri, Matthew B., 1985, "Monetary Policy Games and the Role of Private Information," American Economic Review, Vol. 75 (December), pp. 1056-70. Clarida, Richard, and Mark Gertler, 1996, "How the Bundesbank Conducts Monetary Policy." NBER Working Paper 5581 (Cambridge, Massachusetts: National Bureau of Economic Research). Clark, Peter B., Charles A. E. Goodhart, and Haizhou Huang, 1996, "Optimal Monetary Policy Rules in a Rational Expectations Model of the Phillips Curve," LSE Financial Markets Group Discussion Paper No. 247 (London: London School of Economics and Political Science). Click, Reid W, 1998, "Seigniorage in a Cross-Section of Countries," Journal of Money, Credit, and Banking, Vol. 30 (May), pp. 154-71. Debelle, Guy, and Stanley Fischer, 1994, "How Independent Should a Central Bank Be?" in Goals, Guidelines, and Constraints Facing Monetary Policymakers, ed. by Jeffrey C. Fuhrer (Boston: Federal Reserve Bank of Boston), pp. 195-221. Dornbusch, Rudiger, 1996, "Debt and Monetary Policy: The Policy Issues," NBER Working Paper No. 5573 (Cambridge, Massachusetts: National Bureau of Economic Research). Evans, Lynne J., and Michael C. Amey, 1996, "Seigniorage and Tax Smoothing—Testing the Extended Tax-Smoothing Model." Journal of Macroeconomics, Vol. 18 (Winter), pp. 111-25. Froyen, Richard T., and Roger N. Waud, 1995, "Optimal Seigniorage versus Interest-Rate Smoothing." Journal of Macroeconomics, Vol. 17 (Winter), pp. 111-29. Fukuta, Y. and Akihisa Shibata, 1994, "A Cointegration Test of the Optimal Seigniorage Model," Economics Letters, Vol. 44 (April), pp. 433-37. Goodhart, Charles A. E., 1993, "Central Bank Independence," LSE Financial Markets Group, Special Paper Series No. 57 (London: London School of Economics and Political Science). Gros, Daniel, and Guy Vandille, 1995, "Seigniorage and EMU: The Fiscal Implications of Price Stability and Financial Market Integration," Journal of Common Market Studies, Vol. 33 (June), pp. 175-96. Huang, Haizhou, and A. Jorge Padilla, 1995, "Fiscal Policy and the Sub-Optimality of the Walsh Contract for Central Bankers," LSE Financial Markets Group Discussion Paper No. 223 (London: London School of Economics and Political Science).

313

Jensen, Henrik, 1994, "Loss of Monetary Discretion in a Simple Dynamic Policy Game," Journal of Economic Dynamics and Control, Vol. 18 (May), pp. 763-79. Kydland, Finn E., and Edward C. Prescott, 1977, "Rules Rather than Discretion: The Inconsistency of Optimal Plans," Journal of Political Economy, Vol. 85 (June), pp. 473-91. Lohmann, Susanne, 1992, "Optimal Commitment in Monetary Policy: Credibility Versus Flexibility," American Economic Review, Vol. 82 (March), pp. 273-86. Mankiw, N. Gregory, 1987, "The Optimal Collection of Seigniorage: Theory and Evidence," Journal of Monetary Economics, Vol. 20 (September), pp. 327-41. McCallum, Bennett T., 1995, "Two Fallacies Concerning Central Bank Independence," American Economic Review, Papers and Proceedings, Vol. 85 (May), pp. 207-11. Rankin, Neil, 1998, "Is Delegating Half of Demand Management Sensible?" International Review of Applied Economics, Vol. 12 (September), pp. 415-22. Rogoff, Kenneth, 1985, "The Optimal Degree of Commitment to an Intermediate Monetary Target," Quarterly Journal of Economics, Vol. 100 (November), pp. 1169-89. Sibert, Anne, 1994, "The Allocation of Seigniorage in a Common Currency Area," Journal of International Economics, Vol. 37 (August), pp. 111-22. Svensson, Lars E. O., 1995, "Optimal Inflation Targets, 'Conservative' Central Banks, and Linear Inflation Contracts," CEPR Discussion Paper No. 1249 (London: Centre for Economic Policy Research). Tabellini, Guido, 1986, "Money, Debt and Deficits in a Dynamic Game," Journal of Economic Dynamics and Control, Vol. 10 (December), pp. 427-42. , 1988, "Monetary and Fiscal Policy Coordination with a High Public Debt," in High Public Debt: The Italian Experience, ed. by Francesco Giavazzi and Luigi Spaventa (Cambridge: Cambridge University Press), pp. 90-134.

314

Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union

GIOVANNI DELL'ARICCIA*

This paper analyzes the effects of exchange rate volatility on bilateral trade flows. Through use of a gravity model and panel data from western Europe, exchange rate uncertainty is found to have a negative effect on international trade. The results seem to be robust with respect to the particular measures representing exchange rate uncertainty. Particular attention is reserved for problems of simultaneous causality, stemming from the endogenous behavior of monetary authorities. The negative correlation between trade and bilateral volatility remains significant after controlling for the simultaneity bias. [JEL F14, F17, F31]

ne main argument against flexible exchange rates has been that exchange rate Ovolatility could have negative effects on trade and investment. If exchange rate movements are not fully anticipated, an increase in exchange rate volatility, which increases risk, will lead risk-averse agents to reduce their import/export activity and to reallocate production toward domestic markets. This paper provides some estimates of the importance of these effects in the European Union. The trade issue has played an important role in the debate on the European Monetary System (EMS) and the European Monetary Union (EMU). The EMS was established with the intent of controlling exchange rate volatility and avoiding large

* Giovanni Dell'Ariccia is an Economist in the Developing Countries Studies Division of the Research Department. This paper is partly based on work done for the author's Ph.D. dissertation at MIT. He thanks Andrew Bernard, Rudi Dornbusch, Mike Mussa, Karen Swiderski, Jaume Ventura, two anonymous referees, and all the participants in the MIT International Breakfast and the Conference on International Trade and Market Structures in Le Mans for useful suggestions. He is particularly grateful to Dave Riker for extensive and helpful comments, and to Fabio Fornari and Sandro Giustiniani who provided most of the data. Financial support by Banca Nazionale del Lavoro is gratefully acknowledged.

315

©International Monetary Fund. Not for Redistribution Giovanni Dell'Ariccia misalignments among European currencies. One of the stated purposes was to reduce exchange rate uncertainty to promote intra-EU trade and investments. The discussion on the transition to EMU, and in particular the idea of a "two-speed" European Union, where "virtuous" countries would switch to using the euro from the beginning and other countries would join later, involves similar issues. One major concern is that a partial monetary union would have negative effects on the trade flows of the countries joining the single currency at a second stage. The idea is that, as is the case for customs unions, a partial monetary union could divert trade away from nonmem- ber countries. However, there is not strong or unambiguous empirical evidence to support these views. A quite extensive literature has tested the effects of exchange rate regimes on trade, but the results are not always significant and they change across studies.1 Moreover most papers use only cross-sectional or time-series data instead of a panel, and just a few use bilateral data. The analysis in this paper includes only Western European countries, allowing gathering of both trade and financial data across time as well as across countries, instead of using cross-sections only. This enables us to deal in a new manner with some of the problems met in the previous literature. There are other reasons to limit the scope of this study to Europe. The theoretical foundations of the gravity model assume identical and homothetic preferences across countries and rely heavily on the concept of intra-industry trade.2 European countries are relatively homogeneous in terms of technology, factor endowments, and per capita income, so the model seems particularly appropriate for this case. Moreover, as Bayoumi and Eichengreen (1995) note, the relationship between trade and other economic characteristics might be different for industrial and developing countries. Thus restricting the sample to Western European countries minimizes problems due to country-specific factors. Finally, the actual perspective of a single currency regime for the EU makes this set of countries the natural target for this kind of study. The paper tests the effects of exchange rate volatility on trade using different measures and techniques, with particular attention to the simultaneous causality problem that may arise in these kind of studies. If central banks make an effort to stabilize the exchange rate with their main trade partners, a negative correlation between exchange rate volatility and trade would appear from the data, but this should not be construed to mean that trade reacts negatively to exchange rate instability. The use of panel data facilitates dealing with this problem in a way that explicitly takes into account the behavior of the central banks. If the central bank stabilizing strategy does not change over the period considered, it can be treated as a country-pair specific effect and it can be eliminated by using a fixed-effect model.

1For example, Bahmani-Oskooee and Payesteh (1993), Bailey, Tavlas, and Ulan (1986), and Hooper and Kohlhagen (1978), find no evidence of a negative effect of volatility on trade. Wei (1996) in his work on OECD countries finds that volatility coefficients have the wrong sign. Frankel and Wei (1993) and Kenen and Rodrik (1986) find conflicting results. While Kim and Lee (1996), Stokman (1995), Chowdhury (1993), and Peree and Steinherr (1989) find significant evidence of a negative relation. For a discussion see IMF (1984), or European Commission (1990). The existence of conflicting evidence is consistent with Gagnon (1993), who suggests that the likely impact of volatility on trade should be small. 2See Helpman (1987).

316

The empirical evidence in this paper supports the view that exchange rate uncertainty depresses international trade. However, according to the results, the negative effect of exchange rate volatility on trade is very small. The results are robust with respect to the particular measures chosen to represent uncertainty. They also show that the negative correlation between exchange rate volatility and bilateral trade remains significant when one controls for simultaneous causality. However, they reject the hypothesis of the absence of a simultaneity bias.

I. Gravity Models The gravity model has been widely used in empirical work in international economics.3 The microeconomic foundations of this model can be directly linked to the theory of trade under imperfect competition, and more specifically to intra-industry trade theory, but the characteristics of this approach are consistent with most theoretical models of trade.4 In a gravity model the volume of trade between two countries increases with the product of their GDPs and decreases with their geographical distance. The idea is that countries with a larger economy tend to trade more in absolute terms, while distance represents a proxy for transportation costs and it should depress bilateral trade. In general, a per capita income variable is included to represent specialization; richer countries tend to be more specialized, and thus they tend to have a larger volume of international trade for any given GDP level. Models often include a number of dummy variables to control for different factors that might affect transaction costs. For example, a common border, language, or membership in a customs union are suppose to decrease transaction costs and to promote bilateral trade. This paper includes a proxy to represent exchange rate uncertainty. In the actual estimation this variable will take different forms: the standard deviation of the first differences of the logarithmic exchange rate, the sum of the squares of the forward errors, and the percentage difference between the maximum and the minimum of the nominal spot rate. The pooled ordinary least squares (OLS) regression is

log(TRADEijt) = yt + B1log(GDPitGDPjt) + B2log(DISTij) = B3log(popitpopjt) + B4BORDij + B5EUijt + B6LANGij B7vijt + Eijt, where TRADE is the gross bilateral trade (Exports + Imports) between countries i and j at time t. EU represents membership in the European Union (1 when both countries j and i are in the union at time t, 0 otherwise), and BORDER and LANG represent respectively a common border and language. The variable v represents the proxy for uncertainty about the bilateral exchange rate between country i and j at time t. Note that the intercept has to be allowed to change over time. Indeed, following the model in Helpman (1987), any change in world aggregate GDP will be

3See, for example, Bayoumi and Eichengreen (1995), Frankel (1992), and Krugman (1991). 4Helpman (1987) uses a Dixit/Stiglitz imperfect competition model to obtain the relation between gross trade and GDPs. Bergstrand (1989) generalizes this model to include Hecksher-Ohlin trade.

317

©International Monetary Fund. Not for Redistribution Giovanni DelPAriccia captured by the intercept.5 This implicitly imposes a restriction on the "third-country" coefficient—in other words, assuming, for example, that the trade between Germany and Italy reacts in the same way to a change in U.S. or French incomes. A major advantage of using panel data is the ability to control for possibly unobservable country-pair individual effects. Such omitted effects, if correlated with the included regressors, would bias the OLS estimation. This papers considers a standard model assuming that the latent individual effect is a time-invariant random variable. That regression reads

log(TRADEijt) = y, + o^ + ${log(GDPitGDPjt) + p2log(D/S7}/) + $3log(popitpopjt) + ^4BORDU + $5EUijt + fieLANGij + fyv# + eijt9 where ay stands for the individual effect. The use of panel data allows one to control for cultural, economical, and institutional country-pair specific factors that are constant over time and are not explicitly represented in the model. Note that in the fixed-effects specification any time-invariant country-pair specific effect will be captured by the dummy ay.

II. Exchange Rate Volatility Measures If purchasing power parity (PPP) held, domestic and foreign trade would not systematically involve a different degree of uncertainty. However, exchange rates experience significant and persistent deviations from PPP,6 adding an exchange risk component to import/export activities. Then an increase in exchange rate uncertainty may lead risk-averse firms to reduce their foreign activity, reallocating production toward their own domestic markets.7 With regard to this, the relevant type of exchange rate risk will depend on the model of exporting/importing firm that we have in mind. On the one hand, exporting

5Assume two differentiated products X and Y, and homothetic preferences identical in every country. Then, in the completely specialized case, imports of country k from country j would be

IMPkj = sk(pxXj + PyYj), where sk is country &'s share in world spending (and its share of world income in the absence of trade imbalances) and A} and Y-} are the outputs of goods X and Y produced in country j (the time index is omitted here). The symmetric is true for the imports of country j from country k. Thus the total gross trade is

Tkj = sk (pxXj + PyYj) + Sj(pxXk + PyYk) = skGDPj + SjGDPk.

Rewriting, GDP. Tkj = skSjGDPworU+sjSkGDPworU = 2GDPj GDPwoM

And, when one takes logs, any change in the world GDP will be captured by the constant. 6See Froot, Kim, and Rogoff (1995). 7This result holds under certain conditions; see De Grauwe (1988). When those conditions are violated, the sign of the elasticity of trade flows with respect to exchange rate volatility is ambiguous. Exchange rate volatility creates a positive option value for firms that have the opportunity to choose whether to sell on the domestic or on foreign markets.

318

©International Monetary Fund. Not for Redistribution EXCHANGE RATE FLUCTUATIONS AND TRADE FLOWS firms may sign short-term export contracts in foreign currency. Then, assuming that costs in the firm's own currency are known at t - 1, the only uncertainty arises from the nominal exchange rate: the firm does not know its revenue in domestic currency at t- I.8 In this situation forward exchange rate markets represent an effective way to hedge against uncertainty. Short-term contracts are available for all the major currencies and they are relatively cheap.9 On the other hand, firms might have some sort of long term commitment to the export activity. These kind of firms have to sustain sunk costs to enter particular foreign markets and are interested in the relationship between their costs and the price that they can charge on those markets. In this case what matters is the real exchange rate: firms are interested in the evolution of their revenues relative to their costs.10 To hedge against this kind of uncertainty is much more difficult. Forward markets are not complete in terms of maturity, and the future exchange needs might not be known precisely at the moment of the decision. Hence, real exchange rate uncertainty may play an important role in determining firms' import/export choices.11 The first problem in estimating the effects of exchange rate uncertainty on trade is choosing an appropriate variable to represent instability.12 The literature has used a number of measures of exchange rate volatility and variability as a proxy for risk. Some papers used the standard deviation of the percentage change of the exchange rate or the standard deviation of the first differences of the logarithmic exchange rate.13 This latter measure has the property of being zero in the presence of an exchange rate that follows a constant trend, and it gives a larger weight to extreme observations (consistently with the standard representation of

8The expected utility from profit at time t - 1 for the exporting firm will be

£,_,£/(n,) = £,_,£/((

E0UO:tnt(l + r)-') = E0U(Xt(pUt - Ct(pt))(l + r))~0 and Hi {P>,-C,(P,)) EOU\1, -(l + r)-'j = £ot/I, p, P, '-(1 +rT I

1'These considerations suggest that the next step in this kind of study should be to look at more disaggregated data. It seems important to be able to discriminate the effects of exchange rate volatility across industries characterized by different import/export structures. 12For a discussion of exchange rate volatility measures, see Brodsky (1984), Kenen and Rodrik (1986), and Lanyi and Suss (1982). 13See Brodsky (1984), Kenen and Rodrik (1986), and Frankel and Wei (1993).

319

©International Monetary Fund. Not for Redistribution Giovanni Dell'Ariccia risk-averse firms).14 Others consider the average absolute difference between the previous period forward rate and the current spot to be the best indicator of exchange rate risk. The advantage of this measure is that, under a target zones regime, or under pegged but adjustable exchange rates, it would pick up the effect of the presence of a "peso problem" or the lack of credibility of the official parity. Another possibility is to use the percentage difference between the maximum and the minimum of the nominal spot rate over the t years preceding the observation, plus a measure of exchange rate misalignment. This index stresses the importance of medium-run uncertainty. The idea is that large changes in the past generate expected volatility.15 It is worth noting that the measures proposed as proxies for risk are backward-looking, the assumption being that firms use past volatility to predict present risk. Then, even if one could restrict the choice to a particular measure, there would still be many options: daily, weekly, or monthly changes; which temporal window; etc. Consequently, this paper tests the model using different variables: the standard deviation of the first difference of the logarithmic exchange rate, the sum of the squares of the forward errors, and the percentage difference between the maximum and the minimum of the nominal spot rate.16 Moreover, it uses different temporal windows, and both real and nominal exchange rates. A problem of simultaneous causality may arise using some of these measures. Central banks could systematically try to stabilize the bilateral exchange rate with their most important trade partners. In this case exchange rate volatility cannot be treated as an exogenous variable. Exchange rate volatility and trade would be negatively correlated, but the direction of causality would be uncertain, and OLS would provide a biased estimation. In other words, with an OLS regression it would not be possible to distinguish between the effects of investors' risk aversion and the effects of central bank policies. This concern is confirmed by Bayoumi and Eichengreen (1998), who find that monetary authorities are more likely to intervene on the exchange rate when trade links are strong. Instrumental variable estimators represent a solution to this problem. Frankel and Wei (1993) use the standard deviation of the relative money supply as an instrument for the exchange rate volatility. Their justification is that relative money supplies and bilateral exchange rates are highly correlated, but monetary policies are less affected by trade considerations than exchange rate policies. Unfortunately, this solution presents the problem that for many European countries exchange rate stability has been an important determinant of the monetary policy.17 However, the forward error is not a target of central banks' policies and somehow reflects exchange rate uncertainty. The sum of the squares of the forward errors (defined

14The underlying assumption is that a constant trend would be perfectly anticipated and would not affect uncertainty. An alternative variable some authors have used is the standard deviation of the level of the nominal exchange rate. This measure relies on the underlying assumption that the exchange rate moves around a constant level. In the presence of a trend this index would probably overestimate exchange rate uncertainty. For similar measures see Akhtar and Hilton (1984), Bailey, Tavlas, and Ulan (1986), and Hooper and Kohlhagen (1978). 15See Peree and Steinherr (1989). 16All these variables are constructed using end-of-period exchange rate monthly data from the IMF's International Financial Statistics (IFS). 17This is especially true for the countries participating in the ERM.

320

©International Monetary Fund. Not for Redistribution EXCHANGE RATE FLUCTUATIONS AND TRADE FLOWS as the difference between the log of the three-month forward rate and the log of the spot rate three months later, using "end-of-the-month" data) is correlated with the standard deviation of the spot rate and thus it represents an instrument for exchange rate volatility. The availability of panel data allows a different approach to solving the simultaneous causality problem. The idea behind the simultaneity bias is that central banks try to stabilize the bilateral exchange rate against their countries' main trade partners. If that is the case, the exchange rate volatility becomes a function of the share of the bilateral trade between the two countries over their total trade (T..\ + T1.. v* = VPaVi<)u {TJ'J hjt, where the terms p and y represent the stabilization effort functions of the two central banks. In this context, if the bilateral trade shares were constant over time, one could write

vijt = kijt + By + r\iJt.

In that case the central bank factor could be treated as a country-pair fixed effect. Then the central bank effect would be captured by the country-pair dummy, and the fixed effects specification of Regression (2) would give unbiased estimates. One can imagine central banks following a more general and less accurate rule, in which the stabilization effort depends on the order of magnitude of the bilateral shares, and not on their exact value. In such a case the trade shares would not need to be perfectly constant, but only more or less stable over time. In other words, countries would only need to maintain their relative importance as trade partners. This is actually the case for the sample in this paper: trade shares are not strictly constant over time, but for every country the relative size of its trade partners remains more or less the same over the period considered.

III. Empirical Evidence The sample period covers 20 years from 1975 to 1994. The countries included are the current 15 EU countries (with Belgium and Luxembourg taken as a whole)18 and Switzerland, for a total of 2,100 observations. The source for the trade data is the OECD database: bilateral data for both import and export flows are available. The GDP data are also from the OECD. The original data were expressed in current prices and different currencies. In order to be used in a multiperiod gravity model they had to be deflated and converted to a common currency.19 There were two possible ways to proceed. One could first convert the data into a common currency and then use the

18Austria, Belgium and Luxembourg, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain, Sweden, and the United Kingdom. 19For the conversion PPP values from the OECD series were used; very similar results were also obtained by converting all the data to U.S. dollars.

321

©International Monetary Fund. Not for Redistribution Giovanni Dell'Ariccia deflator for that currency to express the data in constant prices, or, alternatively, one could first deflate the data with each country deflator and then convert them to a common currency. If PPP applied, the two procedures would be equivalent. However, since PPP often fails, the second procedure seems superior. Indeed, as different countries have different consumption baskets, the second procedure has the advantage of applying the right deflators to each country's data. For similar reasons the paper uses only export data to compute the gross bilateral trade flows.20 The available export (import) deflators are based on a basket that reflects a country's total export (import).21 However, with this paper's data the correct deflator should use baskets reflecting the bilateral flows between each pair of countries. It seems reasonable to assume that the bias introduced by using the "aggregate" deflator is smaller for export data than for import data. The idea is that, for each country, the goods it exports to different countries are more homogenous than the goods it imports from different countries. Distances are represented by air distances between capital cities.22 This paper uses different proxies to represent exchange rate uncertainty: the standard deviation of the first differences of the logarithm of the monthly average bilateral spot rate, the sum of the squares of the forward errors, and the percentage difference between the maximum and the minimum of the nominal spot rate. Exchange rate data are end-of-month observations from the IFS. Analogous measures are used for the real rate that is constructed using CPI indexes from the IFS.23 The dummy EU is included to control for the progressive enlargement of the union: this variable has value one for country pairs and years for which both countries are EU members. An additional dummy LANGUAGE represents country pairs with a common language. Table 1 describes the results of Regression (1) using various measures to represent exchange rate uncertainty. The intercept was allowed to change over time and robust standard errors were estimated. All coefficients have the expected sign and are significant at the 1 percent level. Moreover, the results seem to be robust. Most coefficients are similar for the different regressions, suggesting that the four measures of exchange rate uncertainty are in some way equivalent (the regression using the sum of the squares of the forward errors as exchange rate volatility measure is on a subsample of countries that does not include Portugal). It is worth noting the relative importance of having a common language in determining trade flows. Even after controlling for GDP, population, membership in the EU, and a common border, countries speaking the same language trade between each other 24 percent more than those that do not share a common language. The exchange rate volatility coefficient is small, but not irrelevant. From the nominal exchange rate standard deviation coefficient, a total elimination of exchange rate volatility in 1994 would have determined a 12 percent increase in trade,24 a 13 percent increase using the real exchange rate mea-

20Note that, at least in theory, country j's imports from country k is equal to country &'s exports to country j, so import and/or export data could be used to compute the bilateral gross trade. 21These are IFS data. 22Exceptions are Frankfurt for Germany and Milan for Italy. The source for all distance data is Alitalia. 23There is no monthly price index for Ireland. The monthly real exchange rate was constructed using the quarterly price index and assuming the inflation rate constant within the quarter. 24The average standard deviation of the monthly nominal exchange rate change in 1994 was about 0.55 percent.

322

Table 1. Regression (1): Pooled Regression

Nominal Standard Real Standard Forward Variable Deviation Deviation Error Range GDP 0.94 0.89 0.97 0.93 (0.026) (0.026) (0.039) (0.028) POPULATION -0.20 -0.17 -0.26 -0.19 (0.029) (0.029) (0.041) (0.031) DISTANCE -0.32 -0.32 -0.19 -0.23 (0.027) (0.027) (0.029) (0.030) COMMON BORDER 0.27 0.27 0.33 0.29 (0.017) (0.017) (0.018) (0.021) COMMON LANGUAGE 0.21 0.22 0.22 0.24 (0.025) (0.025) (0.029) (0.026) EU 0.24 0.23 0.34 0.29 (0.014) (0.014) (0.015) (0.015) EX. RATE VOLATILITY -19.52 -21.67 -0.74 -0.87 (1.204) (1.219) (0.076) (0.105) Note: All coefficients are significant at the 1 percent level. Standard errors are in parentheses. Sources: OECD; IFS. sure, and a 10 percent increase using the forward error.25 It is interesting to note that the results for nominal exchange rate volatility are very close to the results for real volatility. This outcome is not particularly surprising given that in the sample there is a strong correlation between nominal and real exchange rate volatility (see Figure 1). The results of Table 1 are statistically significant and seemingly do not depend on the variable chosen to represent exchange rate uncertainty. Nonetheless, the validity of these results could be questioned for the presence of simultaneity bias in Regression (1) when using the standard deviation of the exchange rate change. Central banks are likely to try to stabilize the exchange rate vis-a-vis their main trading partners. In such a case, even if exchange rate uncertainty had no negative effect on trade flows, there would be a negative correlation between exchange rate volatility and trade at a bilateral level. To solve this problem the forward error can be used as an instrument for exchange rate volatility: in particular, the sum of the squares of the three-month logarithmic forward error as an instrument for the standard deviation of the first differences of the logarithmic spot rate. This variable is not controlled by central banks and it is positively correlated with this paper's measure of exchange rate volatility. Note that the forward exchange rate was not available for Portugal, so the regression with instrumental variables uses only a subsample of 14 countries (1,820 observations).26 Also here the constant

25This compares with an average bilateral trade annual growth rate of 3.5 percent for the sample period. 26For all the other countries it was possible to construct a forward rate using short-term interest rates. The source was IFS.

323

Figure 1. Real and Nominal Exchange Rate Volatility (as from Standard Deviation of Exchange Rate Change)

devjreal

dev_nom was allowed to change over time and errors were estimated controlling for heteroscedasticity and autocorrelation. Table 2 describes the results of the regression using instrumental variables (two-stage generalized least squares) and the results of the standard regression on the same countries (without Portugal). All coefficients still have the right sign, they are significant at the 1 percent level, and their size does not change with respect to the results of Table 1. For the instrumental variable estimation the results are more or less the same, suggesting that the negative correlation between exchange rate volatility and trade is not determined solely by the simultaneous causality bias. In other words, the negative correlation between exchange rate variability and trade does not depend, or at least does not depend entirely, on central banks' policies. It is possible to test the null hypothesis of absence of simultaneous causality using a Hausman specification test. If the hypothesis is verified, OLS are unbiased and consistent, but they are biased in the presence of simultaneous causality, while the instrumental variable (IV) estimator is unbiased and consistent under both the null and the alternative hypothesis. From the results of the Hausman test we can reject at the 10 percent level the hypothesis that the estimator in Table 1 is unbiased. This result is thus consistent with the presence of a simultaneity bias. Nevertheless, the results obtained with the instrumental variable estimation are still valid and confirm the existence of a negative relation between bilateral exchange rate volatility and trade flows. The existence of unobserved country-pair specific effects may bias the results of Regression (1). Then, to further test the robustness of these findings, one can use the simple model proposed in Section II. In the fixed effect model any individual

324

Table 2. Regression (1): Instrumental Variables

Variable Nominal Nominal IV Real Real IV GDP 1.04 1.03 0.98 0.98 (0.033) (0.034) (0.034) (0.035) POPULATION -0.32 -0.31 -0.28 -0.27 (0.036) (0.036) (0.036) (0.037) DISTANCE -0.30 -0.30 -0.29 -0.30 (0.030) (0.032) (0.030) (0.032) COMMON BORDER 0.28 0.28 0.29 0.29 (0.019) (0.020) (0.019) (0.020) COMMON LANGUAGE 0.22 0.22 0.22 0.23 (0.024) (0.023) (0.024) (0.024) EU 0.27 0.26 0.27 0.27 (0.015) (0.016) (0.015) (0.017) EX. RATE VOLATILITY -20.36 -21.47 -21.32 -22.17 (1.295) (2.147) (1.327) (2.210) Note: All coefficients are significant at the 1 percent level. Standard errors are in parentheses. Reduced sample excluding Portugal. Sources: OECD; IFS. effect will be captured by the country-pair dummy. Then, to the extent that the trade shares are stable over time, the fixed effect estimator will also take care of the simultaneity bias.27 The "central bank effect" has to be constant over time in order to be captured by the country-pair specific dummies. This paper considers both fixed-effect and random-effects estimations. The random-effect model has the obvious advantage of allowing the estimation of the coefficients of time- invariant variables. However, if individual effects are not drawn from the same distribution, the random effect estimates are not consistent. Table 3 reports the results of Regression (2). In Table 3 the sample is the complete set of 15 countries for the first four columns and the subset without Portugal for the regression with the forward errors. These results seem to confirm the previous findings. The GDP and population coefficients have the right sign and are still positive at the 1 percent level with all three measures of exchange rate volatility. The EU dummy coefficient is positive and statistically significant at the 1 percent level. The Hausman test rejected the unbiasedness of the random-effect estimator at the 5 percent level. Hence, the random-effect coefficients could be biased, and one should rely solely on the fixed-effects estimator. However, the main focus of this

27Trade shares are very stable in the sample. The only big change is in Spain/Portugal share. For each country, trade partners were ranked by their share in the country's total trade and then the rankings for 1975 and 1994 were compared. They were very similar for all countries. The overall average place change between rankings was 0.9 places. No change had taken place in 42 percent of the cases, and the maximum change had been five places.

325

Table 3. Regression (2): Random and Fixed Effects Estimations

Nominal Standard Real Standard Forward Deviation Deviation Errors Random Fixed Random Fixed Random Fixed Variable Bffecti Effects Effects Effects Effects Effects GDP 1.27* 1.69* 1.25* 1.64* 1.19* 1.41* (0.062) (0.098) (0.062) (0.098) (0.075) (0.105) POPULATION -0.50* -0.66* -0.48* -0.67* -0.42* -0.49* (0.068) (0.132) (0.068) (0.132) (0.079) (0.138) DISTANCE -0.07 — -0.08 -0.16 (0.094) (0.094) (0.106) BORDER 0.36* 0.36* — 0.35* (0.073) (0.072) (0.081) LANGUAGE 0.19** — 0.19** — 0.18*** — (0.093) (0.093) (0.102) EU 0.15* 0.14* 0.15* 0.14* 0.14* 0.13* (0.009) (0.010) (0.010) (0.010) (0.011) (0.012) EX. RATE VOLATILITY -3.21* -2.84* -4.68* -4.15* -0.27* -0.25* (0.616) (0.608) (1.384) (0.645) (0.034) (0.034) Note: One asterisk signifies significance at the 1 percent level; two at the 5 percent level; three at the 10 percent level. Sources: OECD; IFS. paper is on the exchange rate volatility coefficient that is very similar for fixed- effect and random-effect estimations. The exchange rate volatility coefficient is still negative. It is significant at the 1 percent level for all three different measures and for both fixed-effect and random-effect estimations. However, according to these estimates the size of the effect of volatility on trade is very small. A total elimination of exchange rate volatility in 1994 would have increased trade only by 3 or 4 percent (equivalent to the average annual growth rate of bilateral trade in the sample). Nevertheless, these results are consistent with the idea that a negative correlation between exchange rate volatility and trade exists and that at least a part of it is not spurious correlation caused by central bank stabilization policies. They also suggest that country-specific effects play an important role, advising against the use of pooled OLS estimations. To test the efficacy of this method in eliminating simultaneous causality, a Hausman test was performed. Also in this case the instrumental variable was represented by the forward error measure. The test could not reject the hypothesis of unbiasedness of the OLS fixed-effect estimator. The result is then consistent with the assumption that the central banks factor is stable over time and is eliminated by using the fixed-effect model. As noted earlier there is no "right" measure of exchange rate volatility. Accordingly, this paper further tests the robustness of the previous results using a

326

©International Monetary Fund. Not for Redistribution EXCHANGE RATE FLUCTUATIONS AND TRADE FLOWS different time window for the measures. Table 4 reports the results of Regression (1) using a two-year window to compute the various exchange rate volatility variables. The results are consistent with the previous ones, confirming a negative effect of volatility on trade. Note that an instrumental variable estimation is used given the outcome of the Hausman test on the previous results. All coefficients have the expected sign and are significant at the 1 percent level. Finally, some analysis is conducted on the effects of third-country volatility on trade; for example, what happens to trade flows between France and Italy when the volatility between the franc and the deutsche mark increases? However, multicollinearity problems meant that the contribution of third-country volatility could not be isolated. As in Wei (1996), the coefficient was not significant and had the wrong sign.28 The evidence in this section shows a negative correlation between exchange rate volatility and trade flows. With the results presented here the hypothesis that the behavior of the central banks has no role in determining the negative correlation between volatility and trade can be rejected. However, the results of estimations that are robust to simultaneous causality bias support the hypothesis that firms, reacting negatively to volatility on foreign currencies markets, determine a decrease in the volume of international trade when the exchange rate becomes more volatile.

IV. The ERM Effect Most observers viewed the 1992/93 crisis of the EMS (or more precisely, of the Exchange Rate Mechanism) as a stop in the process of economic integration of the European countries. The purpose of the EMS was to reduce exchange rate volatility among member currencies to promote trade and economic convergence, and the ERM was actually successful in reducing both nominal and real exchange rate volatility (this is especially true for the period 1987-92).29 Thus, following the results from the previous section, the ERM should have had a positive effect on the bilateral trade between EU member countries. If the end of the ERM meant a diminished exchange rate stability, a reduction in intra-EU trade could be expected. In this section the framework presented in the previous sections is used to try to estimate the effects of

28A variable representing the exchange rate volatility of the two currencies with respect to all the others was included

log(TRADEijt) = yt+ aij + B1log(GDPitGDPjt) + B2log(DISTij) + B3log(popitpopjt) + B4BORDERij + B5EUijt + B6LANGij + B7vijt + B8mijt + Eijt, where miJt = Ei=jvijtwijt + Ej=iwijt, with weights wijt represented by relative GDPs. If the trade diversion hypothesis is valid the sign of (B8 should be negative. Table 5 reports the results for Regression (4) with real and nominal exchange rate volatility. Most coefficients have more or less the same values as in Regression (1). However, for both cases there is probably a multicollinearity problem. The correlation between the bilateral exchange rate volatility and the volatility with the rest of the countries in the sample is above 0.9. Then it is not possible to determine the contribution of the two variables separately. Indeed, the "third country" volatility coefficient is not significant and has the wrong sign. 29See, for example, Figure 2. For a detailed analysis see De Grauwe and Verfaille (1988).

327

Table 4. Regression (1): Two-Year Window

Variable Nominal IV Real IV Forward Error GDP 1.02 0.95 0.94 (0.038) (0.040) (0.040) POPULATION -0.29 -0.24 -0.23 (0.040) (0.041) (0.042) DISTANCE -0.36 -0.35 -0.22 (0.037) (0.036) (0.032) COMMON BORDER 0.24 0.25 0.29 (0.022) (0.022) (0.021) COMMON LANGUAGE 0.25 0.25 0.24 (0.026) (0.026) (0.026) EU 0.25 0.26 0.34 (0.019) (0.019) (0.016) EX. RATE VOLATILITY -13.01 -13.12 -0.46 (1.311) (1.324) (0.046) Note: All coefficients significant at the 1 percent level. Standard errors are in parentheses. Sources: OECD; IFS.

Table 5. Regressions (4): The "Third Country" Effect

Nominal Standard Deviation Real Standard Deviation Variable Random Effects Fixed Effects Random Effects Fixed Effects GDP 1.27* 1.69* 1.25* 1.64* (0.062) (0.099) (0.062) (0.098) POPULATION -0.50* -0.66* -0.48* -0.67* (0.068) (0.132) (0.068) (0.132) DISTANCE -0.07 — -0.08 (0.095) — (0.095) — BORDER 0.36* — 0.36* (0.073) — (0.073) — LANGUAGE 0.19** — 0.19** — (0.094) — (0.094) — EU 0.15* 0.14* 0.15* 0.13* (0.010) (0.010) (0.010) (0.010) EX. RATE VOLATILITY -3.22* -2.85* -4.70* -4.17* (0.617) (0.609) (0.651) (0.646) "THIRD-COUNTRY" -0.24 -0.13 -0.37 -0.27 VOLATILITY (0.451) (0.444) (0.468) (0.462) Note: One asterisk signifies significance at the 1 percent level; two at the 5 percent level; three at the 10 percent level. Standard errors are in parentheses. Sources: OECD; IFS.

328

Figure 2. Lira/Deutsche Mark Exchange Rate Volatility with and without ERM

Percentage change

September 1992: Italy exits ERM

March 1979: ERM starts

the ERM on trade. A dummy was constructed equal to 1 for pairs in which both countries are members of the ERM and 0 otherwise.30 The resulting equation is

log(TRADEijt) = yt + c^ + ^log(GDPitGDPJt) + fclogP/S^) + ^og(popupopjt) + fitfORDij + fcEUijt + fieLANGij + ^ERMijt + eijt.

In this way the ERM dummy captures the stabilizing role that the ERM had on the currencies of member countries. On the other hand, if one is interested in the effect that the ERM had per se, not only through the reduction of exchange rate volatility, the equation becomes

log(TRADEijt) = Yr + ay + ${log(GDPitGDPjt) + fclogCDZST^) + $3log(popitpopjt) + faBORDij + fcEUijt + peLANGij + frERMijt + p8v//, + eijt.

A negative sign on the ERM dummy coefficient would mean that the mechanism's role in reducing uncertainty went beyond the induced reduction in volatility. The results of both regressions are presented in Table 6. All the usual coefficients still have the right sign and are still significant. The ERM coefficient has the wrong sign. For the fixed-effect model it is significant at the 5 percent level when

30This approach has the advantage of avoiding the simultaneous causality problem. The decision to enter the ERM concerns a country's general policy more than simply its trade policy.

329

Table 6. Regressions (3a) and (3b): The ERM Effect

Nominal Standard Real Standard Forward ERM Deviation Deviation Errors Only Random Fixed Random Fixed Random Fixed Random Fixed Variable Effects Effects Effects Effects Effects Effects Effects Effects GDP 1.27* 1.71* 1.24* 1.66* 1.19* 1.44* 1.33* 1.72* (0.062) (0.099) (0.061) (0.099) (0.075) (0.106) (0.066) (0.099) POPULATION -0.50* -0.66* -0.47* -0.67* -0.43* -0.50* -0.55* -0.64* (0.067) (0.132) (0.067) (0.132) (0.078) (0.138) (0.072) (0.133) DISTANCE -0.08 — -0.09 — -0.16 — -0.03 — (0.093) — (0.092) — (0.105) — (0.107) — BORDER 0.36* — 0.35* — 0.35* — 0.37* — (0.071) — (0.071) — (0.079) — (0.084) — LANGUAGE 0.19** — 0.19** — 0.18**' 0.19*** (0.091) — (0.090) — (0.100) — (0.107) — EU 0.15* 0.14* 0.15* 0.14* 0.15* 0.14* 0.15* 0.14* (0.010) (0.010) (0.010) (0.010) (0.012) (0.012) (0.010) (0.010) EX. RATE -3.31* -2.%* -4.88* -4.36* -0.27* -0.26* — — VOLATILITY (0.620) (0.610) (0.657) (0.649) (0.034) (0.034) — — ERM -0.01 -0.02** -0.02** -0.02** -0.02** -0.02** -0.01 0.02*** (0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.10) (0.10) Note: One asterisk signifies significance at the 1 percent level; two at the 5 percent level; three at the 10 percent level. Standard errors are in parentheses. Sources: OECD; IFS. controlling for exchange rate volatility, and at the 10 percent level when alone. For the random-effect estimation it is significant at the 5 percent level in the regression with the real volatility measure and with the forward-errors measure. It is not significant in the regression with nominal volatility and when alone. On the one hand, this result seems surprising and conflicts strikingly with the findings in Section III. Indeed, ERM membership should decrease uncertainty and thus increase trade. On the other hand, a large literature addressed the issue of the credibility of the ERM and rejected the full credibility hypothesis for most cases.31 From that point of view, the result in this section can be reconciled with those in the rest of this paper. If, for most periods and countries, the exchange rate target zones were not credible, one should not expect a significant effect of the ERM dummy on trade flows. At the same time, a non-credible ERM would generate expectations of relatively large realignments, to which agents might react particularly negatively.32 In other words,

31See Giovannini (1990), Svensson (1991), and Frankel and Phillips (1992). 32A way to address this issue might be to control for the credibility of the bilateral target zones and construct a "credible ERM" dummy. One would first have to define a measure of credibility, and then could construct a variable taking the value 1 when the commitment to the bilateral parity is credible, and 0 otherwise. The quoted literature relies on tests based on forward rates (or interest rate differentials) first proposed in Svensson (1991). The basic idea is that if the forward rate is outside the band, the target zone cannot be fully credible.

330

©International Monetary Fund. Not for Redistribution EXCHANGE RATE FLUCTUATIONS AND TRADE FLOWS agents might find a system of discrete changes, that are typically large over a short period, more harmful than similar, but more gradual changes under a system of flexible rates. An alternative, but not very appealing, explanation is provided by political economy. Brada and Mendez (1988) suggest that countries with fixed exchange rate regimes are more likely to use trade restrictions to defend their trade balance. They find some evidence that countries with fixed rates trade less than countries with floating rates. However, in our context this effect seems very unlikely because most countries in the sample (all countries in the ERM) are EU members.

V. Conclusions This paper tests the relationship between exchange rate uncertainty and trade with data from Western European countries. The analysis uses different variables as proxies for uncertainty, all of which gave consistent results. There was evidence of a small but significant negative effect of bilateral volatility on trade. The problem of a possible simultaneity bias was addressed in two different ways, and both instrumental variables and fixed effects over time gave results consistent with the hypothesis of a negative effect of exchange rate uncertainty on trade. Nevertheless, a Hausman specification test rejected the hypothesis that no simultaneity bias exists. Further research in this area should look at more disaggregated data. It is more difficult to find financial instruments to hedge against exchange rate risk when the time horizon becomes longer. Then EMU might have a different impact across industries. In sectors where the export activity requires large investments, trade should prove more sensitive to exchange rate volatility than in sectors characterized by "short-term" exports.33 For the same reasons, exchange rate stability might be more important for foreign direct investments than for trade flows.34

Appendix 1. EU-EMS Chronology Apr. 1951 European Coal and Steel Community—Treaty of Paris Mar. 1957 European Economic Community—Treaty of Rome (6 countries) Aug. 1971 End of the Bretton Wood System Mar. 1972 Introduction of the Snake (Belgium, France, Germany, Italy, Netherlands) May 1972 Denmark, the UK, and Norway join the Snake. Jun. 1972 Denmark and the UK exit the Snake. Oct. 1972 Denmark rejoins the Snake. Jan. 1973 Denmark, Ireland, and the UK become members of EEC Feb. 1973 Italy exits the Snake. Jan. 1974 France exits the Snake. Jul. 1975 France rejoins the Snake.

33Stokman (1995) uses disaggregated, but not bilateral, data to estimate the effects of exchange rate volatility on the intra-EU exports of five European countries. 34See Campa and Goldberg (1995) or Goldberg and Kolstad (1995) for some evidence on the relationship between exchange rate volatility and foreign direct investment.

331

Mar. 1976 France exits the Snake. Mar. 1979 EMS starts (Belgium, Denmark, France, Germany, Ireland, and Netherlands with 2.25 percent margins, Italy with 6 percent). Jan. 1981 Greece joins EEC. Jan. 1986 Portugal and Spain join EEC. Jun.1989 Spain joins the EMS with 6 percent margins. Jan. 1990 The margin for the Italian lira is narrowed to 2.25 percent. Oct. 1990 Unification of Germany. The UK joins the ERM with 6 percent margins. Feb. 1992 Maastricht Treaty on European Union. Apr. 1992 Portugal joins ERM with 6 percent margins. Sep. 1992 Italy and the UK suspend participation in the ERM. Jan. 1993 Single European Market. Aug. 1993 ERM margins widened to 15 percent. Jan. 1995 Austria, Finland, and Sweden join the EU.

REFERENCES Akhtar, M.A., and R. Spence Hilton, 1984, "Effects of Exchange Rate Uncertainty on German and U.S. Trade," Federal Reserve Bank of New York Quarterly Review, Vol. 9 (spring) pp. 7-16. Arize, Augustine, 1996, "Real Exchange-Rate Volatility and Trade Flows: The Experience of Eight European Economies," International Review of Economics and Finance, Vol. 5 (No. 2), pp. 187-205. Bahmani-Oskooee, Mohsen, and Sayeed Payesteh, 1993, "Does Exchange Rate Volatility Deter Trade Volume of LDCs?" Journal of Economic Development, Vol. 18 (December), pp. 189-205. Bailey, Martin, George Tavlas, and Michael Ulan, 1986, "Exchange Rate Variability and Trade Performance: Evidence from the Big Seven Countries," Weltwirtschaftliches-Archiv, Vol. 122 (No. 3), pp. 466-77. Bayoumi, Tamim, and Barry Eichengreen, 1995, "Is Regionalism Simply a Diversion? Evidence from the Evolution of the EC and EFTA," CEPR Discussion Papers, No. 1294, (London: Centre for Economic Policy Research). , 1998, "Exchange Rate Volatility and Intervention: Implication of the Theory of Optimum Currency Areas," Journal of International Economics, Vol. 45 (August), pp. 191-209. Bergstrand, Jeffrey, 1989, "The Generalized Gravity Equation, Monopolistic Competition, and the Factor-Proportions Theory in International Trade," Review of Economics and Statistics, Vol. 71 (February), pp. 143-53. Brada, Josef, and Jose Mendez, 1988, "Exchange Rate Risk, Exchange Rate Regime, and the Volume of International Trade," Kyklos, Vol. 41 (No. 2), pp. 263-80. Brodsky, David, 1984, "Fixed Versus Flexible Exchange Rates, and the Measurement of Exchange Rate Instability," Journal of International Economics, Vol. 16 (May), pp. 295-306. Campa, Jose, and Linda Goldberg, 1995, "Investment in Manufacturing, Exchange Rates and External Exposure," Journal of International Economics, Vol. 38 (May), pp. 297-320. Chowdhury, Abdur, 1993, "Does Exchange Rate Volatility Depress Trade Flows? Evidence from Error-Correction Models," Review of Economics and Statistics, Vol. 75 (November), pp. 700-706.

332

De Grauwe, Paul, 1988, "Exchange Rate Variability and the Slowdown of Growth in International Trade," Staff Papers, Vol. 35 (March), pp. 63-84. , and Guy Verfaille, 1988, "Exchange Rate Variability, Misalignment, and the European Monetary System," in Misalignment of Exchange Rates: Effects on Trade and Industry, ed. By Richard C. Marston, NBER Project Report Series (Chicago: University of Chicago Press), pp. 77-100. Dornbusch, Rudiger, and Jeffrey Frankel, 1988, "The Flexible Exchange Rate System: Experience and Alternatives," in International Finance and Trade in a Polycentric World, ed. by Silvio Borner, (Basingstoke, England: MacMillan in association with International Economic Association). European Commission, 1990, "One Market, One Money: An Evaluation of the Potential Costs and Benefits of Forming an Economic and Monetary Union," European Economy, Vol. 44 (October), pp. 3-347. Frankel, Jeffrey, 1992, "Is Japan Creating a Yen Bloc in Asia and the Pacific?" NBER Working Papers No. 4050 (Cambridge, Massachusetts: National Bureau of Economic Research). , and Steven Phillips, 1992, "The European Monetary System: Credible at Last?" Oxford Economic Papers, Vol. 44, (October), pp. 791-816. , and Shang-Jin Wei, 1993, "Trade Blocs and Currency Blocs," NBER Working Papers, No. 4335 (Cambridge, Massachusetts: National Bureau of Economic Research). Froot, Kenneth A., Michael Kim, and Kenneth Rogoff, 1995, "The Law of One Price Over 700 Years," NBER Working Papers, No. 5132 (Cambridge, Massachusetts: National Bureau of Economic Research). Gagnon, Joseph, 1993, "Exchange Rate Variability and the Level of International Trade," Journal of International Economics, Vol. 34 (May), pp. 269-87. Giovannini, Alberto, 1990, "European Monetary Reform: Progress and Prospects," Brooking Papers on Economic Activity: 2, pp. 217—74. Goldberg, Linda, and Charles Kolstad, 1995, "Foreign Direct Investment, Exchange Rate Variability and Demand Uncertainty," International Economic Review, Vol. 36 (November), pp. 855-73. Helpman, Elhanan, 1987, "Imperfect Competition and International Trade: Evidence from Fourteen Industrial Countries," in International Competitiveness, ed. by A. Michael. Spence and Heather A. Hazard, (Cambridge, Massachusetts: Ballinger). Hooper, Peter, and Steven Kohlhagen, 1978, "The Effect of Exchange Rate Uncertainty on the Prices and Volume of International Trade," Journal of International Economics, Vol. 8 (November), pp. 483-511. IMF, 1984, The Exchange Rate System: Lessons of the Past and Options for the Future, IMF Occasional Paper No. 30 (Washington: IMF). Kenen, Peter, and Dani Rodrik, 1986, "Measuring and Analyzing the Effects of Short-Term Volatility in Real Exchange Rates," Review of Economics and Statistics, Vol. 68 (May), pp. 311-15. Kim, Kiheung, and WooRhee Lee, 1996, "The Impact of Korea's Exchange Rate Volatility on Korean Trade," Asian Economic Journal, Vol. 10 (March), pp. 45-60. Krugman, Paul, 1991, "The Move Toward Free Trade Zones," Federal Reserve Bank of Kansas City Economic Review, Vol. 76 (November/December), pp. 5-26. Lanyi, Anthony, and Esther Suss, 1982, "Exchange Rate Variability: Alternative Measures and Interpretation," Staff Papers, Vol. 29 (December), pp. 527-60.

333

Peree, Eric, and Alfred Steinherr, 1989, "Exchange Rate Uncertainty and Foreign Trade," European Economic Review, Vol. 33 (July), pp. 1241-64. Pozo, Susan, 1992, "Conditional Exchange-Rate Volatility and the Volume of International Trade: Evidence from the Early 1900s," Review of Economics and Statistics, Vol. 74 (May), pp. 325-29. Stokman, A.C.J., 1995, "Effect of Exchange Rate Risk on Intra-EC Trade," De Economist, Vol. 143, No. 1, pp. 41-54. Svensson, Lars E. O., 1991, "The Simplest Test of Target Zones Credibility," Staff Papers, Vol. 38 (September), pp. 655-65. Viaene, Jean-Marie, and Casper de Vries, 1992, "International Trade and Exchange Rate Volatility," European Economic Review, Vol. 36 (August), pp. 1311-21. Wei, Shang-Jin, 1996, "Intra-National versus International Trade: How Stubborn Are Nations in Global Integration?" NBER Working Paper No. 5531 (Cambridge, Massachusetts: National Bureau of Economic Research).

334

IMFstaffpapers

Volume 46, 1999

Washington, D.C.

Volume 46 Index

Volume 46 (1999) comprises three issues, as follows: March, pages 1-106 June, pages 107-246 September/December, pages 247-341

Authors

Aylward, Lynn. Countries' Repayment Performance Vis-a-Vis the IMF: A Response to Backer 242 Backer, Arno. Countries' Repayment Performance Vis-a-Vis the IMF: A Comment on Aylward and Thome 238 Baig, Taimur, and Ilan Goldfajn. Financial Market Contagion in the Asian Crisis 167 Barajas, Adolfo, Roberto Steiner, and Natalia Salazar. Interest Spreads in Banking in Colombia, 1974-96 196 Bayoumi, Tarnim, and Ronald MacDonald. Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions 89 Beddies, Christian H. Monetary Policy and Public Finances: Inflation Targets in a New Perspective 293 Berg, Andrew, and Catherine Pattillo. Are Currency Crises Predictable? A Test 107 Carranza, Luis, and Chorng-Huey Wong. Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results 225 Crafts, Nicholas. East Asian Growth Before and After the Crisis 139 Dell' Ariccia, Giovanni. Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union 315 Goldfajn, Ilan, and Taimur Baig. Financial Market Contagion in the Asian Crisis 167 Hardy, Daniel, and Ceyla Pazarbasioglu. Determinants and Leading Indicators of Banking Crises: Further Evidence 247

336

MacDonald, Ronald, and Bayoumi, Tamim. Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions 89 Mauro, Paolo, and Antonio Spilimbergo. How Do the Skilled and the Unskilled Respond to Regional Shocks? The Case of Spain 1 Montenegro, Claudio E., and Abdelhak S. Senhadji. Time Series Analysis of Export Demand Equations: A Cross-Country Analysis 259 Pattillo, Catherine, and Andrew Berg. Are Currency Crises Predictable? A Test 107 Pazarbasioglu, Ceyla, and Daniel Hardy. Determinants and Leading Indicators of Banking Crises: Further Evidence 247 Ramaswamy, Ramana, and Robert Rowthorn. Growth, Trade, and Deindustrialization 18 Rowthorn, Robert, and Ramana Ramaswamy. Growth, Trade, and Deindustrialization 18 Salazar, Natalia, Adolfo Barajas, and Roberto Steiner. Interest Spreads in Banking in Colombia, 1974-96 196 Spilimbergo, Antonio, and Paolo Mauro. How Do the Skilled and the Unskilled Respond to Regional Shocks? The Case of Spain 1 Steiner, Roberto, Adolfo Barajas, and Natalia Salazar. Interest Spreads in Banking in Colombia, 1974-96 196 Tamirisa, Natalia. Exchange and Capital Controls as Barriers to Trade 69 Vamvakidis, Athanasios. Regional Trade Agreements or Broad Liberalization: Which Path Leads to Faster Growth? 42 Wong, Chorng-Huey, and Luis Carranza. Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results 225

Titles

Are Currency Crises Predictable? A Test 107 Countries' Repayment Performance Vis-a-Vis the IMF: A Comment on Aylward and Thorne 238 Countries' Repayment Performance Vis-a-Vis the IMF: A Response to Backer 242 Determinants and Leading Indicators of Banking Crises: Further Evidence 247 Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions 89 East Asian Growth Before and After the Crisis 139 Exchange and Capital Controls as Barriers to Trade 69 Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union 315 Financial Market Contagion in the Asian Crisis 167 Growth, Trade, and Deindustrialization 18 How Do the Skilled and the Unskilled Respond to Regional Shocks? The Case of Spain 1 Interest Spreads in Banking in Colombia, 1974-96 196

337

Monetary Policy and Public Finances: Inflation Targets in a New Perspective 293 Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results 225 Regional Trade Agreements or Broad Liberalization: Which Path Leads to Faster Growth? 42 Time Series Analysis of Export Demand Equations: A Cross-Country Analysis 259

Subjects To facilitate electronic storage and retrieval of bibliographic data, IMF Staff Papers has adopted the subject classification scheme of the Journal of Economic Literature (Nashville, Tennessee). c Mathematical and Quantitative Methods C12 Hypothesis Testing Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions. By Tamim Bayoumi and Ronald MacDonald 89 C2 Econometric Methods: Single Equation Models Time-Series Models C22 Time Series Analysis of Export Demand Equations: A Cross-Country Analysis. By Abdelhak S. Senhadji and Claudio E. Montenegro 259 Models with Panel Data C23 Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions. By Tamim Bayoumi and Ronald MacDonald 89 C32 Time-Series Models Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results. By Chorng-Huey Wong and Luis Carranza 225 E Macroeconomics and Monetary Economics Consumption; Saving Time Series Analysis of Export Demand Equations: A Cross-Country Analysis. E21 By Abdelhak S. Senhadji and Claudio E. Montenegro 259 Employment; Unemployment; Wages E24 How Do the Skilled and Unskilled Respond to Regional Shocks? The Case of Spain. By Paolo Mauro and Antonio Spilimbergo 1 Determination of Interest Rates; Term Structure of Interest Rates E43 Interest Spreads in Banking in Colombia, 1974-96. By Adolfo Barajas, Roberto Steiner, and Natalia Salazar 196 Financial Markets and the Macroeconomy E44 Determinants and Leading Indicators of Banking Crises: Further Evidence. 338 By Daniel C. Hardy and Ceyla Pazarbasioglu 247

E52 Monetary Policy (Targets, Instruments, and Effects) Monetary Policy and Public Finances: Inflation Targets in a New Perspective. By Christian H. Beddies 293 E61 Policy Objectives; Policy Designs and Consistency; Policy Coordination Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results. By Chorng-Huey Wong and Luis Carranza 225 E62 Fiscal Policy; Public Expenditures, Investment, and Finance; Taxation Monetary Policy and Public Finances: Inflation Targets in a New Perspective. F By Christian H. Beddies 293 International Economics F1 Global Outlook Growth, Trade, and Deindustrialization. By Robert Rowthorn and Ramana Ramaswamy... 18 F13 Commercial Policy; Protection; Promotion; Trade Negotiations Exchange and Capital Controls as Barriers to Trade. By Natalia T. Tamirisa 69 F14 Country and Industry Studies of Trade Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union. F17 By Giovanni Dell'Ariccia 315 Trade Forecasting and Simulation Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union. By Giovanni Dell'Ariccia 315 Time Series Analysis of Export Demand Equations: A Cross-Country Analysis. F21 By Abdelhak S. Senhadji and Claudio E. Montenegro 259 International Investment; Long-Term Capital Movements Policy Responses to External Imbalances in Emerging Market Economies: Further F3 Empirical Results. By Chorng-Huey Wong and Luis Carranza 225 F30 International Finance General F31 Financial Market Contagion in the Asian Crisis. By Taimur Baig and Ilan Goldfajn ... 167 Foreign Exchange Are Currency Crises Predictable? A Test. By Andrew Berg and Catherine Pattillo 107 Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions. By Tamim Bayoumi and Ronald MacDonald 89 Exchange and Capital Controls as Barriers to Trade. By Natalia T. Tamirisa 69 Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union. By Giovanni Dell'Ariccia 315

339

F32 Current Account Adjustment; Short-Term Capital Movements Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results. By Chorng-Huey Wong and Luis Carranza 225 F4 Macroeconomic Aspects of International Trade and Finance F40 General Financial Market Contagion in the Asian Crisis. By Taimur Baig and Ilan Goldfajn ... 167 F41 Open Economy Macroeconomics Policy Responses to External Imbalances in Emerging Market Economies: Further Empirical Results. By Chorng-Huey Wong and Luis Carranza 225 Time Series Analysis of Export Demand Equations: A Cross-Country Analysis. By Abdelhak S. Senhadji and Claudio E. Montenegro 259 F43 Economic Growth of Open Economies Growth, Trade, and Deindustrialization. By Robert Rowthorn and Ramana Ramaswamy... 18 Regional Trade Agreements or Broad Liberalization: Which Path Leads to Faster Growth? By Athanasios Vamvakidis 42 F47 Forecasting and Simulation Are Currency Crises Predictable? A Test. By Andrew Berg and Catherine Pattillo 107 G Financial Economics G15 International Financial Markets Financial Market Contagion in the Asian Crisis. By Taimur Baig and Ilan Goldfajn ... 167 G21 Banks; Other Depository Institutions; Mortgages Determinants and Leading Indicators of Banking Crises: Further Evidence. By Daniel C. Hardy and Ceyla Pazarbasioglu 247 Interest Spreads in Banking in Colombia, 1974-96. By Adolfo Barajas, Roberto Steiner, and Natalia Salazar 196 J Labor and Demographic Economics J61 Geographic Labor Mobility; Immigrant Workers How Do the Skilled and Unskilled Respond to Regional Shocks? The Case of Spain. By Paolo Mauro and Antonio Spilimbergo 1 L Industrial Organization L13 Oligopoly and Other Imperfect Markets Interest Spreads in Banking in Colombia, 1974-96. By Adolfo Barajas, Roberto Steiner,

N and Natalia Salazar 196 N1 Economic History N15 Macroeconomics and Monetary Economics; Growth and Fluctuations Asia including Middle East East Asian Growth Before and After the Crisis. By Nicholas Crafts 139

340

N2 Financial Markets and Institutions N25 Asia including Middle East East Asian Growth Before and After the Crisis. By Nicholas Crafts 139 N3 Labor, Demography, Education, Income, and Wealth N35 Asia including Middle East East Asian Growth Before and After the Crisis. By Nicholas Crafts 139 O Economic Development, Technological Change, and Growth Economic Development O1 Growth, Trade, and Deindustrialization. By Robert Rowthorn and Ramana Ramaswamy... 18 Technological Change O3 Growth, Trade, and Deindustrialization. By Robert Rowthorn and Ramana Ramaswamy... 18 Macroeconomic Analyses of Economic Development O11 East Asian Growth Before and After the Crisis. By Nicholas Crafts 139 Measurement of Economic Growth; Aggregate Productivity O47 East Asian Growth Before and After the Crisis. By Nicholas Crafts 139 Asia including Middle East 053 The Uzbek Growth Puzzle. By Jeromin Zettelmeyer 274 Economic Systems P Socialist Systems and Transitional Economies P2 P27 Performance and Prospects The Uzbek Growth Puzzle. By Jeromin Zettelmeyer 274 P52 Comparative Studies of Particular Economies The Uzbek Growth Puzzle. By Jeromin Zettelmeyer 274

341

dots (...) indicate that the data are not available;

a dash (—) indicates that the figure is zero or less than half the final digit shown, or that the item does not exist;

a single dot (.) indicates decimals;

a comma (,) separates thousands and millions;

"billion" means a thousand million; and "trillion" means a thousand billion;

a short dash (-) is used between years or months (for example, 1998-99 or January-June) to indicate a total of the years or months inclusive of the beginning and ending years or months; a slash (/) is used between years (for example, 1998/99) to indicate a fiscal year or a crop year;

a colon is used between a year and a number indicating a quarter within that year (for example, 1998:1); and

components of tables may not add to totals shown because of rounding.

The term "country," as used in this publication, may not refer to a territorial entity that is a state as understood by international law and practice; the term may also cover some territorial entities that are not states but for which statistical data are maintained and provided internationally on a separate and independent basis.

Design: Luisa Menjivar-Macdonald and Sanaa Elaroussi