DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? Oriana Bandiera, Gerard Caprio, Patrick Honohan, and Fabio Schiantarelli*

Abstract—The effect of financial liberalization on private saving is may differ substantially from the impact effect. Lastly, theoretically ambiguous, not only because the link between interest rate levels and saving is itself ambiguous, but also because financial liberaliza- financial liberalization is a process rather than a one-shot tion is a multidimensional and phased process, sometimes involving event. reversals. Using principal components, we construct 25-year time-series The purpose of this paper is to provide an empirical indices of financial liberalization for each of eight developing countries: Chile, Ghana, Indonesia, Korea, Malaysia, Mexico, Turkey, and Zim- examination of the total effect of the financial reform on babwe. These are employed in an econometric analysis of private saving in aggregate private saving based on eight case studies: Chile, these countries. Our results cannot offer support for the hypothesis that Ghana, Indonesia, Korea, Malaysia, Mexico, Turkey, and financial liberalization will increase saving. On the contrary, the indica- tions are that liberalization overall—and in particular those elements that Zimbabwe. These countries have all significantly liberalized relax liquidity constraints—may be associated with a fall in saving. their financial sector policies, but they differ in the nature and phasing of financial liberalization, in other aspects of I. Introduction their policy reform program, and in the macroeconomic context in which liberalization took place. This variety WAVE OF liberalization of financial markets has swept allows us to explore the degree to which the saving response over much of the developing world, especially since the A differs from country to country, as well as to test whether the mid-1980s. This liberalization has been characterized by greater scope being granted to market forces in determining response is a common one. interest rates and in allocating credit (Caprio, Atiyas, and Financial reform typically comprises several key phases Hanson (1994)). Although this has occurred under the that are often separated by several years. Reform measures pressure of increased globalization of financial markets and are introduced in a number of different dimensions: interest following the example of many industrial countries, there rates, credit allocation, bank ownership, prudential regula- has been an expectation that financial liberalization would tion, security markets, and openness of the capital account. help economic development. In particular, the early litera- The best sequencing of these various elements is frequently ture on financial repression, following McKinnon (1973) debated. In practice, reform has not been a monotonic and Shaw (1973), stressed the potential role of higher process: In some cases, setbacks have involved temporary interest rates in mobilizing savings that could be put to policy reversals. productive use. A thorough quantitative assessment of the impact of such But it is far from clear that financial liberalization actually a process must take account of its gradual and reversible does increase private saving. One obvious and important nature. Based on an analysis of the historical evolution in consideration is that the effect of interest rates on saving is each case, we have identified the timing of major moves on itself ambiguous, as the income effect might offset substitu- eight different dimensions towards a more liberalized sys- tion effects. In addition, one must recognize that financial tem. Using the principal components of the resulting matri- liberalization involves more than just a change in interest ces of 0–1 variables (1’s correspond to the years after a rates. Other dimensions of financial liberalization, such as particular reform is introduced), we obtain a single continu- increased household access to consumer credit or housing ous financial liberalization index for each of our countries. finance, might also work to reduce private saving (Muell- As an alternative, we also construct for each country a pair bauer and Murphy (1990), Jappelli and Pagano (1994)).1 of subindices that capture different aspects of the liberaliza- Furthermore, the long-term effect of liberalization on saving tion process. Our data extends over a quarter of a century—a period long enough to allow us to model the response to liberalization in each country separately—but in addition to Received for publication February 2, 1999. Revision accepted for the country-by-country results, we also present panel data publication November 29, 1999. evidence. * London School of Economics, World Bank, World Bank and CEPR, Visual inspection of the time series of the main relevant and Boston College, respectively. We would like to thank Pierre-Richard Age´nor, Craig Burnside, Bruce variables—the financial liberalization index, the real interest Hansen, Tullio Jappelli, Peter Pedroni, Deborah Wetzel, Stephen Zeldes, rate, monetary depth (either M2 or total credit to the private participants in seminars at Boston College and the World Bank, and sector expressed as a percentage of GNDI), and the private especially Klaus Schmidt-Hebbel, Luis Serve´n and John Campbell as well as three anonymous referees for useful suggestions. The findings, interpre- saving ratio—reveals little evidence of a clear-cut relation- tations, and conclusions expressed in this paper are entirely those of the ship between saving and liberalization. authors. They do not necessarily represent the views of the World Bank, its We estimate an econometric relationship expressing the executive directors, or the countries they represent. 1 There is also the view, stressed in the neostructuralist contributions of private saving ratio as a function of the real interest rate and Taylor (1983) and Van Wijnbergen (1982) that the effect of reduced of the index of financial liberalization (or its subcompo- taxation on formal financial intermediaries might actually reduce the flow nents), along with income, inflation, and government saving. of credit to the private sector to the extent that reserve requirements captured funds for the government that had been substituted away from the In addition to directly measuring the contribution of liberal- curb market. ization to the volume of aggregate saving, our procedure

The Review of Economics and Statistics, May 2000, 82(2): 239–263 ௠ 2000 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 240 THE REVIEW OF ECONOMICS AND STATISTICS

improves on earlier estimates of the saving-interest relation, The process of financial liberalization also unleashes a which omitted any role for financial-sector liberalization series of short-run effects. Not only can the process of other than the real interest rate channel. domestic portfolio adjustment lead to transitory changes in Although they cannot be solved-out for a net effect on the the volume of domestic saving, but (especially when com- level of saving, Euler equations can be helpful in detecting bined with liberalization of the foreign exchange market) it the extent of credit rationing. In this spirit, we also assess the may also induce large capital inflows, largely but not impact of financial reform on the extent of liquidity con- exclusively attributable to a return flow of past flight capital. straints by estimating an augmented Euler equation for If not sterilized, such inflows can result in a credit boom consumption, in which it is assumed (in an extension of the leading to real income surges, which in turn have a direct, model of Campbell and Mankiw (1989, 1991)) that the but transitory, effect on the volume of saving. Modeling of fraction of the consumers are liquidity constrained varies the effect of financial liberalization on saving needs to take with the degree of financial liberalization. account of these short-run effects, as well as the long-run The structure of the paper is as follows. Section II effect. It is also important to recognize that some of the describes the main channels through which financial liberal- overall effects can come through the effect of income on ization may affect saving and briefly reviews the relevant saving. empirical literature. Section III describes the financial reform process as it occurred in each of the eight countries A. Steady-State Effects being studied here. This section also explains and graphs our index of financial liberalization and examines the prima If financial liberalization improves the rate of return for facie evidence about its effects. Sections IV and V present savers, then knowledge of the interest elasticity of saving the econometric results based on the saving function and on can help predict the long-term impact of liberalization on the augmented Euler equation for consumption, respec- saving. However, because of the wealth and current income tively. Section VI concludes. effects that will generally be present, there is no presumption as to the direction of the aggregate saving response to an II. Financial Liberalization and Saving: Theoretical exogenous interest rate change. Despite many studies, this Background and Review of the Empirical Evidence remains an empirically controversial area, partly because of a surprising shortage of reliable and comparable cross- Although financial liberalization can enhance the effi- country data on retail interest rates. Recent reviews by ciency with which saved resources are channeled into Balassa (1990), Srinivasan (1993), and Fry (1995) conclude productive use, the effect on the quantity of saving is that more studies have found a positive interest elasticity of theoretically ambiguous.2 saving than a negative one, but the coefficients have The mechanisms at work here include both long-term and generally been small and often insignificant.4 Furthermore, short-term effects. Once it has settled down, a competitive, the theoretical impact on saving of the improved opportuni- liberalized financial system will typically be characterized ties for hedging and risk reduction that can also become by improved saving opportunities, including higher deposit available as a result of financial liberalization is equally interest rates, a wider range of savings media with improved ambiguous. risk-return characteristics, and, in many cases, more banks Possibly of greater importance for aggregate saving than and bank branches, as well as other financial intermediaries. deposit interest rates may be the availability of a variety of Bank lending rates will typically be higher for those alternative, nonfinancial assets, the return on which may not borrowers who had privileged access in the restricted be captured by deposit interest rates. While the use of real regime, but access to borrowing should be wider. These interest rates implicitly acknowledges that goods inventories long-term effects of liberalization on aggregate private are an alternative to financial assets, it would be very useful saving will be felt through changes in rates of return and in in principle to take explicit account of alternative investment the degree of credit restrictions. Moreover, if financial opportunities, notably the rate of return on owner-occupied liberalization also has a favorable effect on the allocation of housing and other real-estate investment. Many developing resources, this will generate increases in income that will in countries have experienced property booms, and household turn increase saving.3 saving may have been very sensitive to the after-tax rate of

2 It should be stressed at the outset that our evidence is based chiefly on promote human capital formation, though this will normally be measured National Accounts definitions of saving. These need to be distinguished as consumption in the National Accounts. from intermediated saving or from capital flows. Dornbusch and Reynoso 4 The effect of interest rates on saving could be nonlinear, perhaps (1989) observe that capital flight through misinvoicing of trade serves to involving threshold effects. Reynoso (1989) presents some evidence that conceal saving that is being hidden abroad: An apparent increase in saving the response of savings to the interest rate may be represented by a may really be a reduction in capital flight. Furthermore, they note that, as parabola, with savings increasing most significantly when interest rates go durable goods purchases are usually treated as consumption in the data, a from sharply negative to just below zero, then leveling off, and finally shift from these to accumulation of financial assets tends to be mislead- declining as real interest rates become very large and positive, in which ingly recorded as saving. case they may reflect political uncertainty, peso-effects, bank insolvency, 3 It should be noted that increased household borrowing may not all go to and the like. For a general review of saving decisions see Browning and consumption or housing. A relaxation of borrowing constraints could Lusardi (1996).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 241

return on investment in real estate. (See, for example, The large capital inflows that have been associated with Koskela and Vire´n (1994).) Unfortunately, in most cases, recent liberalizations have also had complex, short-term, data on such rates of return are not available for developing macroeconomic consequences. Liberalization of the domes- countries. tic financial system has typically been only one element of a Measured interest rates may not reflect capital market package of reforms that have been associated with these realities if households and small enterprises are constrained inflows, and the inflows have proved to be easily reversible. from borrowing what they would wish because of financial The impact on saving comes through the associated changes repression or for other reasons. To the extent that liberaliza- in availability and cost of credit, revised expectations of tion reduces these borrowing constraints, saving ratios could income growth, and increases in financial wealth, especially be lowered (Jappelli and Pagano (1989, 1994)). Two mecha- due to upward movements in property prices. All this may nisms are at work here. First, when the borrowing constraint lead to consumption booms and to a fall in the saving rate. binds, it induces the individual to consume less. Second, even when the constraints are not binding in the current period, the expectation that they may bind in the future C. Quantifying the Effects of Financial Liberalization reduces today’s consumption. on Saving A very large literature, in response to Hall’s (1978) Most empirical examinations of the effects of financial original contribution, has attempted to gauge the importance liberalization or, more generally, of financial development of borrowing constraints by inferring that any dependence of on saving have involved adding one or more variables to the change in consumption on income might reflect the established econometric specifications either of saving or of inability of households to smooth the intertemporal pattern the rate of change in consumption. The simplest specifica- of their consumption through borrowing. (See, for instance, tions identify pre- and post-liberalization periods with a 5 Campbell and Mankiw (1989, 1991) and Zeldes (1989).) dummy variable (an early example is de Melo and Tybout The developing-country literature here generally confirms (1986) for Uruguay); an alternative is to specify a linear the importance of such dependence, with some indication trend reflecting gradual liberalization (Muellbauer and Mur- that it has been higher for developing countries. (See, for phy (1993) for the UK). instance, Rossi (1988), Haque and Montiel (1989), and Others have employed such proxy variables as the volume Corbo and Schmidt-Hebbel (1991).) of consumer credit (such as Jappelli and Pagano (1989, 1994)). Ostry and Levy (1995) used this variable both on its B. Transitional Effects of Liberalization own and in interaction with an interest rate, and concluded The impact effect of financial liberalization on saving that liberalization had not only lowered saving in France but could be larger than the sustained long-term effect. This is had transformed a negative association between saving and because households will be able to revise target precaution- interest rates into a positive one (cf. Bayoumi (1993) for the ary balances, allowing, for example, some middle-aged UK). An easing of credit market conditions facing house- households that had hitherto been constrained from lifecycle holds was also detected for the 1980s in Scandinavian borrowing to consume at a higher rate than they would have countries by Koskela, Loikkanen, and Vire´n (1992), and over a full lifetime of unconstrained access to borrowing. Lehmussaari (1990). Here the effect on saving came indi- These transitional effects suggest that aggregate household rectly through the impact of increased housing finance on saving could dip below its steady-state level, and that a surge house prices. in consumption may be observed (Muellbauer (1994)). In their thirty-country study, Jappelli and Pagano (1994) Moreover, as noted above, financial liberalization has been also found another type of credit availability variable to be accompanied by real-estate booms in some countries; the highly significant, namely the normal loan-to-value ratio resulting increase in real wealth also may have a temporarily obtainable from mortgage finance institutions: an increase of negative impact on saving.6 fifteen percentage points in the loan-to-value ratio reducing the national saving rate by 2.6 percentage points. This substantial effect may not be entirely housing related, as the 5 The household’s inability to borrow at wholesale-market interest rates may be a rationing phenomenon, or it may reflect a large wedge between variable may be capturing movements in wider credit retail deposit and borrowing rates (for example, money-lender rates). A availability. lower wedge would reduce saving, as King (1985) found for the United Other proxy measures of the prevalence of credit con- Kingdom. See also Alessie, Devereux, and Weber (1993) for an analysis of the effects of abolition of credit controls on the demand for cars in the straints that have been used include the percentage of United Kingdom. homeowners in certain age groups, the interest rate wedge 6 Financial liberalization could affect the value of human and nonhuman on consumer and mortgage loans (Jappelli and Pagano wealth in a variety of ways. An increase in the value of nonhuman wealth will normally, ceteris paribus, reduce saving as consumption out of income (1989)), and the rate of consumer credit delinquencies can now be permanently higher. However, it is hard to isolate such wealth (Carroll (1992)). Confirming the evidence for industrial effects on saving of financial liberalization, not only because of the countries, Vaidyanathan (1993) shows that international difficulty of measuring human and nonhuman wealth, but also because other reforms affecting wealth are usually being undertaken at the same variations in the sensitivity of consumption to income are time. positively related to financial depth (measured by the ratio of

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 242 THE REVIEW OF ECONOMICS AND STATISTICS

M2 to GDP), suggesting again the importance of liquidity responsibility was left to the banks.9 In such cases, with most constraints. of their balance sheet effectively out of their own control, More directly, Miles (1992) estimated that 80% of the banks invested little in credit assessment, monitoring, or total amount of home equity withdrawn by U.K. households asset-liability skills, and in the extreme cases—formerly each year in the 1980s was consumed (rather than involving socialist economies—the result was a low skill base in just a portfolio shift), accounting for essentially all of the finance and little of the infrastructure that supports market- collapse in the U.K. personal saving ratio from 12% to less based financial intermediaries.10 than 5%. Beginning in many countries in the 1970s and accelerat- The existence of well-functioning stock markets could ing subsequently, governments began to reconsider more- also be a factor influencing saving by offering an improved direct interventions, and financial reform programs have risk-return frontier while retaining liquidity. Again, as included attempts to reduce or redirect the government’s mentioned, the predicted impact on aggregate saving is role, most noticeably in the area of pricing and directing theoretically ambiguous, and recent empirical evidence credit. The path of reforms often was influenced by govern- suggests that funds attracted to liquid stock markets in ment views, initial conditions, and political pressures for developing countries come mainly as a switch from other reform. For example, in Chile, real interest rates had been assets.7 negative for decades prior to the removal of controls in 1974, and this decontrol was quite sudden. In contrast, following mild repression in the 1960s, Malaysian authori- III. Financial Reform: Measurement and Effects ties in early 1973—like their Japanese counterparts much later—began deregulating some longer-term interest deposit A. Financial Repression and the Process of Reform rates but let several years pass before all controls were The multifaceted nature of financial reform—involving removed. A very gradual process also characterized the deregulation, liberalization, globalization, and privatization— Korean experience. At times, the process was rather bumpy complicates the measurement of its effects. In addition, the with reimposition of controls after an initial bout of liberal- reforms undertaken in each country have reflected the ization, as in Chile and Malaysia. Often the reimposition of perceived problems of the pre-reform environment. Prior to controls was a consequence of a severe banking crisis that reform, most countries experienced a period of mild or developed in an unstable macro context, characterized by severe financial repression: intervention by governments in large capital inflows, and excessive risk-taking in the allocating and pricing credit, controlling what banks and absence of effective prudential regulation, as in Chile in the other intermediaries could do, using intermediaries as tax- early 1980s. collection devices, and often limiting competition, in particu- Reforms in general include two parts: outright deregula- lar from foreign institutions. These interventions varied by tion, limiting the government’s direct intervention, and country, and in some countries included government owner- putting in its place a system of prudential regulation aimed at ship of banks as a very direct way of influencing how they ensuring the safety and soundness of banking. In addition, did business.8 there is an institution-building component. The latter likely In developing countries, intervention in the financial is a key component of the reform process: During periods of sector went considerably further than the regulation of substantial intervention, especially where most risk is born interest rates and of credit expansion that characterized by government, the demand for financial infrastructure— industrial-country policy. In some countries, banks were accounting, auditing, legal systems, and other finance- required to hold as much as one-half or more of their related skills—is quite limited. When this intervention is liabilities in the form of reserve or liquid assets (often lessened, and if the incentive structure is right, intermediar- deposits at the central bank), and another large part of the ies start devoting more resources to risk and credit analysis, portfolio was dominated by directed credit. Although the for example, and spend more to upgrade the quality of their latter might have been structured so as to leave significant staff. discretion to the banks for credit assessment and monitoring (as in Japan), in practice, in many cases little power or B. Measuring Financial Reform The ideal index of financial reform would attempt to 7 Levine and Zervos (1996); see also Bonser-Neal and Dewenter (1996). measure both the various aspects of the deregulatory and the This conclusion was drawn from the insignificance of indicators of stock market development in cross-sectional regressions in which the dependent variable was the ratio of private saving to GDP. 9 In some cases, the small size of the economy meant that a government 8 In addition to concerns about an inherent instability of finance, these requirement to provide financial support for a sector such as steel-making interventions were often rationalized by a view that finance was not meant in practice lending to a single steel company, with the result that the decisive for growth unless harnessed by a benevolent planner. Levine banks viewed the risk as belonging to the authorities. (1997) discusses some of the historical context and developments of 10 See Caprio and Claessens (1997) for a discussion of initial conditions attitudes about the financial system, and Caprio, Atiyas, and Hanson in reforming financial systems. They argue that long periods of financial (1994) describe financial reforms in Chile, Malaysia, Indonesia, Korea, repression greatly weakened the skills, incentives, and infrastructure in Turkey, and New Zealand. See also Caprio and Klingebiel (1996). finance and therefore complicated the reform process.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 243

institution-building process. Unfortunately, short of using short lived in Malaysia (see figure 5a), in part because banks outcome measures such as the development of markets as a were observed to be slow to reduce rates as their cost of proxy—an approach leading to ambiguity in interpretation funds declined, but also because a moderately severe and endogeneity problems—it seems impossible at present banking crisis led Malaysian authorities to reimpose interest- to find useful measures of institutional development. For rate controls for several years in the mid-1980s.14 these reasons, we have chosen to build our index of reform Significant but different reforms were introduced in all of from explicit policy changes that, though not wholly indepen- the countries under review. As seen in the data for Chile and dent of wider economic conditions, should be less subject to Malaysia, reform can see significant reversals, and more endogeneity problems. generally is not a linear process but one that proceeds in fits Our index thus summarizes exogenous changes in interest- and starts. rate regulation, reserve requirements, directed credit, bank ownership (moves toward privatization), procompetition C. Visual Evidence on Saving and Reform measures, liberalization of securities markets, prudential regulation, and international financial liberalization. Based The figures provide no visual evidence of a clear positive on an analysis of the historical evolution in each case, we association between either index (or real interest rates) and have identified the timing of major moves towards, and private saving for most countries. This is also confirmed by sometimes away from, a more liberalized system under each the bivariate correlation coefficient between saving and the of these eight headings. (Appendix A gives the details.) For index, contemporaneous or lagged (not reported for reasons the particular dimension of deregulation (or institution of space). Only for Turkey and Korea is the association building) affected by each move, we have created a 0–1 significant and positive. For some periods and in some of the dummy variable, with 1 characterizing the more liberalized countries, there appears to be a negative relationship be- regime. This yields a matrix of 0–1 variables for each tween saving and the index. For instance, saving plummets country. Rather than attempting to use all of these dummy in Chile (figure 1a and 1d) with the onset of reform— variables in the econometrics (leading to degrees of freedom perhaps reflecting the easing of credit constraints—then and collinearity problems), we have constructed for each recovers gradually until a more significant increase starts in country the principal components of the matrix. We use the 1985, associated in part with the introduction of a fully first principal component (or a weighted average of the most funded pension system. In Mexico (figures 6a and 6d), we observe a protracted decline in the saving ratio since reforms important principal components) as our liberalization index 15 in the regressions of section IV below.11 In all cases, a higher began. A lack of correlation between the index of financial value of the index in a given year captures a regime that is reform and saving is evident in the cases of Ghana (figures overall more market oriented.12 For all countries, the first 2a and 2d) and Zimbabwe (figures 8a and 8d), where saving first rose then fell, while the index was registering continu- principal component accounts for a high percentage of the 16 total variation: more than 62% for all countries except ing gains. In Malaysia, saving did rise in the 1970s as Malaysia, for which the percentage is also high at 44% (cf. reforms began, but then subsequently leveled off and fell table 2, part I). This gives us confidence that the index based back to its original level. In contrast, there is a clearer on the first principal component is capturing much of the positive association between the index and saving in Korea total information in the matrix of dummies. (figures 4a and 4d), particularly until the late 1980s, Turkey The resulting index is shown in figures 1 through 8, with (figures 7a and 7d), and, to a lesser extent, Indonesia, where, data on financial depth (M2 or total credit to the private however, part of the increase in saving occurred before sector as a percentage of gross national disposable income domestic financial reforms began (figures 3a and 3d). (GNDI)), real interest rates, and the private saving rate It is noteworthy that the figures (and bivariate correla- (measured as a share of GNDI). We have used a definition of tions) suggest a closer association between the behavior of the private saving rate, both unadjusted and adjusted for the index and measures of financial depth for a majority of capital losses due to inflation on domestic assets denomi- the countries. The exceptions are Turkey, Ghana, and nated in local currency. For example, in figure 1a, the index Zimbabwe. There also appears to be a generally positive captures the partial reversal of reforms in Chile resulting 13 raising of ceilings on foreign borrowing. After the reversals of 1982, from the twin banking and debt crises of 1982. Likewise, liberalization resumed in 1986. figure 5a clearly charts the fact that decontrol was initially 14 Caprio, Atiyas, and Hanson (1994) describe financial reforms in Chile, Malaysia, Indonesia, Korea, Turkey, and New Zealand. See also Caprio 11 As an alternative, we also experiment below with a weighted average and Klingebiel (1996, 1996a) for a further discussion of the Malaysian of the more important principal components, using as weights the fraction experience. of the total standard deviation explained by each component. The results 15 The sharp drop in the adjusted series in 1988 is due to a large increase are very similar and are not reported for reasons of space. in the measured stock of debt to which the adjustment applies in that year. 12 See also Demetriades and Luintel (1997) for an application to India of 16 Albeit with continued negative interest rates. The persistence of the principal-components approach to aggregating the information con- negative real interest rates, notably in the cases of Ghana and Zimbabwe, tained in a combination of policy changes and outcome variables. after the onset of reform measures calls into question how real reforms 13 The Chilean reforms had begun in 1974 with the freeing of interest have been. Even though interest rates were deregulated, in some countries rates and the beginning of the easing of reserve requirements, and they continue to be controlled by a cartel of banks, often at the informal continued in the mid- and late-1970s with bank privatization and the behest of the authorities.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 244 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 1.—CHILE

association between our index and real interest rates (which the substitution effect dominates the income effect), is statistically significant in half of the countries). but more predictably reducing disintermediation. 2. Reduction in reserve requirements: This increases the D. How Do the Main Components of Liberalization resources available for lending by the formal financial Affect Saving? sector, and also reduces the break-even intermediation While our main focus has so far been on overall summary spread, thereby likely increasing deposit rates for a indicators of reform, reflecting the view that the overall given level of lending rates. The net effect will also be reform process—though it takes some time and may be influenced by the response of other government policy reversed—ultimately comes as a package, it is worth instruments: For example, the monetary policy impact pausing to consider how each of the eight main components of the reduction in reserve requirements may be of our index is likely to affect saving. There is no unambigu- neutralized through an open market sale. ous theoretical prediction because each change has several channels of effect; thus, the listing is chiefly of heuristic 3. Reduction in directed credit: The response of the value: banks may well be to increase lending to households (never favored in directed-credit regimes), thereby 1. Interest liberalization: On its own, this will tend to tending to lower net household saving. Inasmuch as increase interest rates, perhaps encouraging saving (if the rest of credit is reallocated within the business

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 245

FIGURE 2.—GHANA

sector, there is no obvious basis for predicting the into the formal financial sector. The overall effect on impact on overall business-sector saving. saving is, however, ambiguous. 4. Privatization: This too may be associated with an 8. International financial liberalization: Even if foreign increase in lending to households, a line of business funds flow in following liberalization, the net effect is from which state-owned banks have often been discour- likely to include an increase in rates of return, as aged. barriers to capital outflow are removed. The net effect 5. Procompetition measures: A more competitive environ- on saving could be positive as in 1 above, but this is ment may lead to a move to increased risk taking in less clear, as banks can now use funds borrowed from lending, as the return on low-risk lending narrows. abroad to support lending to households or firms. Households offer an inexhaustible supply of high-risk lending opportunities, so this could be associated with The contradictory likely effects from different compo- lower net household saving. nents of liberalization will be clear from this brief review. 6. Increased prudential regulation: This will tend to Grouping the components into two subaggregates may help. offset risk taking discussed under 5, above, hence For example, one natural dividing line is between macroeco- perhaps increasing net household saving; furthermore, nomic liberalization components 1, 2, and 8, and the a new perception that the banks are sound may microeconomic measures 3 through 7. Even here, there is no encourage a growth in intermediation. However, regu- clear prediction, although some presumption from the above lation may also serve to reduce upward pressure on discussion that the macro components would be somewhat deposit interest rates. more likely to increase saving than the micro. An alternative 7. Securities market development: By providing a wider way of grouping the components into two subaggregates is range of savings instruments, this could channel funds to separate out the measures that are most concerned with

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 246 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 3.—INDONESIA

reducing the role of the state in the allocation of credit 2, 3, IV. Econometric Evidence: Saving Functions and 4 from the rest that act mainly through their effect on the A. Benchmark Model: Country-by-Country Estimates characteristics of asset returns. To the extent that the reduced state involvement helps to reduce liquidity constraints for We begin by estimating the long-run relationship between households, there is a presumption that 2, 3, and 4 could saving and its determinants separately for each country over reduce net household saving, with perhaps the opposite the period 1970–1994.18 In the benchmark specification, the likely on balance for the rest. Based on these two alternative (unadjusted) private saving rate st/yt is modeled as a function aggregations of the various aspects of financial reform, we of the natural log of real per capita GNDI ln yt, the real have extracted the first principal component for each group- interest rate rt, our overall index of financial liberalization ing of the components of liberalization. Each pair will be indext, the inflation rate ␲t, and the government saving rate, 19 used in our econometric work as an alternative to the govst. The choice of variables included in the equation is aggregate index. We use the abbreviations i-mac and i-mic, limited partly by series availability and partly by the length and i-liq and i-rsk,17 for the two pairs of subindices thus computed. 18 Except for slightly shorter samples for Indonesia (1971–1994), Korea (1970–1993), and Zimbabwe (1974–1993). 19 Gross national disposable income is used as a proxy for income. The 17 This term is used because the components involved, 1 and 5 through 8 real interest rate is defined as a short-term rate (continuously compounded) can manifest themselves to the households as an altered risk-return profile minus the inflation rate (calculated as the forward log difference). The rather than reducing liquidity constraints. conclusions reached below are not sensitive to the definition of the real

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 247

FIGURE 4.—KOREA

of the sample period. In particular we would like to have the saving rate, we also present a specification in which included a satisfactory proxy for nonhuman wealth, but income is excluded from the set of regressors. available ones, such as the stock of high-powered money or We have tested the order of integration of the variables government debt, are more likely to be misleading than both country by country, using the ADF test, and by panel, helpful. using the Im, Pesaran, and Shin (1996) test. The results of Inclusion of income in the specification might raise the tests suggest that we cannot reject the hypothesis that questions, given that it would not be expected to influence st/yt,lnyt,lnyt, indext, and govst are integrated of order one. the saving ratio at the individual level in a standard (See table 1, part I, for the panel tests; the country-by- formulation of the lifecycle model. However, any of several country ADF tests are not reported for reasons of space.) considerations can suggest income as an explanatory vari- However, for some countries, there is evidence against the able, including minimum subsistence consumption, nonho- unit root hypothesis for rt and ␲t. For instance, when a trend motheticity of the utility function, precautionary consider- is included, the hypothesis that r has a unit root is rejected at ations, and liquidity constraints.20 In case of any residual t the 5%-significance level in Malaysia, Korea, and Indonesia. doubt about the direction of causality between income and Also, a unit root in ␲t is rejected for Malaysia, Indonesia, interest rate and of the inflation rate. See appendix B for further details on and Turkey. The panel test suggests the rejection of the unit variable constructions and on the data sources. root hypothesis for both variables. 20 On the last two issues, see Deaton (1992, ch. 6). To the extent that Using the Dickey-Fuller (DF) or the adjusted Dickey- borrowing constraints vary with financial liberalization, the coefficient on income might also vary over time. We explore this issue empirically in Fuller (ADF) tests on the residuals of the cointegrating section V, in the context of Euler equations for consumption. regressions, country by country, and the critical values

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 248 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 5.—MALAYSIA

calculated as suggested by MacKinnon (1990) to adjust for dynamic GLS estimates obtained by including the contempo- sample size, we cannot reject the hypothesis of no cointegra- raneous differences of the right-side variables as additional tion between the vector of variables mentioned above regressors and allowing for an AR(1) error term.21 The main (whether or not income, the real interest rate, and inflation drawback of the dynamic GLS estimates is the small number are included). These cointegration tests must be treated with of degrees of freedom available, so that it is probably wise to a healthy dose of caution both because of the low power of consider both sets of results. Although estimates of the such tests against reasonable alternatives and because of the coefficients of rt and ␲t would be problematic if those small number of observations available relative to the variables were truly stationary, their inclusion does not number of variables. As shown in table 1, part II, moreover, invalidate the consistency for the coefficients (and associ- the panel cointegration test proposed by Pedroni (1997a, ated inference) for the other nonstationary variables. 1997b), and the Im, Pesaran, and Shin test applied to the Despite the fact that we have here corrected the omission residuals of the cointegrating vector are consistent with the of other dimensions of financial liberalization, there is existence of a cointegrating relationship between st/yt, (except for the OLS estimate for Mexico) no evidence from indext, govst—or between these three and ln yt if it is the country-by-country estimates of a significant distinct included (or between st/yt, indext, govst,lnyt, rt, and ␲t,ifthe positive effect of the real interest rate on saving. In most troublesome unit root tests on the last two variables are cases, the long-run point relationship is negative (and disregarded). The same propositions hold if we replace index significantly so in the case of Ghana and Indonesia). The with either of the pairs of split indices. evidence based on the time series for individual developing In table 2, we present two estimates of the cointegrating vector, when ln y , r , and ␲ are included. The estimates of 21 Ideally, one would have wanted to include additional leads and lags of t t t the differences; however, the length of our sample precludes us from doing part I are OLS. Because the conventional OLS standard that. Our procedure can be seen as an approximation to the DGLS errors are not valid in this context, part II shows approximate procedure in Stock and Watson (1993).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 249

FIGURE 6.—MEXICO

countries confirms, therefore, the general conclusion derived hand, we estimate that liberalization has raised the saving from previous studies using pooled time-series cross- rate in Turkey by 13% and in Ghana by 6%. country data that it is not possible to pin down a systematic So far as the other variables are concerned, the income positive effect on saving of increases in the interest rate. variable is significant in most cases (both in the long and So far as the effect of financial liberalization itself is short run). The sign of the coefficient of the inflation rate concerned, the picture is mixed. For Korea and Mexico, the differs across countries: significantly negative in Ghana, coefficient of the index of financial liberalization is negative Indonesia, and Malaysia; significantly positive in Mexico.23 and significant in the long run (using the dynamic GLS Finally, there is evidence that an increase in government estimates). On the other hand, for Turkey and Ghana, there is saving leads to a decrease in private saving. Actually, for evidence of a positive and significant long-run effect.22 Korea, Malaysia, and Mexico (and, depending upon the The point estimates imply sizable impacts of the index on specification, Chile and Zimbabwe), the estimates are consis- private saving. For example, the results of table 2, part II, tent with Ricardian equivalence, in that the coefficient of 24 imply that liberalization in Korea and Mexico has perma- govst is not significantly different from Ϫ1. nently lowered saving by 12% and 6% of GNDI, respec- tively (using the realized change in the index). On the other 23 A variety of effects may be associated with inflation, including the fact that it is positively correlated with the private-sector’s capital loss on monetary assets, the relative-price confusion effect of Deaton (1977), 22 Note that these are also the two countries where indext is uncorrelated substitution of consumer durables for financial assets as an inflation hedge, with private credit, suggesting that borrowing constraints may not have or various forms of uncertainty. been much eased. 24 See also Burnside (1996) on Mexico.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 250 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 7.—TURKEY

B. Disaggregating the Effect of Different Elements components most associated with reduced saving, probably of Liberalization through relaxation of borrowing constraints. In the lower panels of table 2, we also report the results obtained when the overall index of liberalization is replaced C. Robustness by the pairs of subindices (i-mac and i-mic, i-rsk and i-liq) previously discussed.25 We report only the coefficients of the The general conclusions we have reached concerning the subindices because the rest of the estimates are little altered effects of financial liberalization in individual countries are when we substitute the subindices for the overall index. The broadly robust to several changes in the specification, some most striking feature is that the estimated effect of i-liq is of which are reported in table 3. For instance, table 3, part I, negative for every country, and significantly so for three shows the dynamic GLS results when we exclude ln yt from countries (Chile, Malaysia, and Mexico). The estimated the cointegrating vector; part II of the same table excludes rt, ␲ effect of i-rsk is positive but not significant. A quite similar and t, while part III uses adjusted private and public saving pattern is observed for the alternative pair of subindices: rates and income to allow for capital gains and losses Here, it is i-mic that tends to be negative, and is always induced by inflation on assets denominated in local currency. negative when significant (for Korea, Malaysia, and Mexico). Excluding the interest rate and inflation and adjusting for This suggests that i-liq and i-mic include the liberalization capital gains and losses leaves the results virtually un- changed. When we exclude the income variable, the overall index is more often positive and significant, probably 25 Note that i-liq cannot be computed for Korea and Turkey, because no significant changes occurred along the dimensions it is meant to summa- because it captures income’s upward trend. The negative rize. impacts of i-mic and i-liq are less unambiguous in this

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 251

FIGURE 8.—ZIMBABWE

specification. Similarly, when we exclude the interest rate, adding the dependency ratio to the cointegrating vector; we i-rsk becomes more significant as it proxies for interest-rate included an interaction term between the interest rate and the changes. financial liberalization index to allow for the interest-rate The general conclusions derived from the benchmark effect to differ depending upon the degree of liberalization; model are also robust to other changes in the specification, we used both linear and quadratic interest-rate terms to which have not been reported for reasons of space. For capture the idea that the effect on saving may depend upon instance, we have used a weighted average of the first few the value of the interest rate itself. These additional variables principal components (with the ratio of the standard devia- did not have significant coefficients. tion relative to the total standard deviation as weights) to Finally, in order to assess also the short-run effects of calculate the aggregate index of financial liberalization. This liberalization, we estimated an error-correction model for is equivalent to including the principal components sepa- saving (not reported here for the sake of brevity). Only for rately and imposing the restriction that their coefficient is one country is there a significant coefficient on the first proportional to the fraction of the total variance explained by difference of the liberalization index (Korea, where the each one of them. We also reestimated the model by using a impact effect is negative). The coefficient on the change in ‘‘backward’’ real interest rate, defined as the nominal the interest rate is positive and significant only for Mexico. interest rate minus the inflation rate over the preceding Thus, there is, for the short-run also, no general evidence of period. In each case, our conclusions do not change. We tried an increase in saving induced by financial liberalization.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 252 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 1.—PANEL INTEGRATION AND COINTEGRATION TESTS significant at the 5% level in the benchmark specification Part I: Univariate Tests (constrained dynamic GLS specification), although only at the 10% level in the specification without income. The (s/y)t ln yt rt Indext ␲␶ govst negative and significant effect of the i-liq subindices is without trend P ϭ 0 Ϫ0.617 2.623 Ϫ4.152 7.165 Ϫ3.495 Ϫ0.141 confirmed by the panel estimates, the latter strongly suggest- P ϭ 1 Ϫ0.458 1.300 Ϫ3.393 5.090 Ϫ3.300 Ϫ0.859 ing that relaxation of liquidity constraints is a key factor with trend underlying the fall in saving. I-mic also remains negative in P ϭ 0 0.027 1.213 Ϫ3.707 4.676 Ϫ3.517 Ϫ0.589 26 P ϭ 1 Ϫ1.645 Ϫ0.302 Ϫ4.076 3.491 Ϫ3.892 Ϫ2.014 all specifications but one. However, the panel estimates are not unproblematic: A likelihood-ratio test does imply that Part II: Panel Tests the assumed equality of coefficients can be rejected at Vector Vector Vector Vector Vector Vector conventional significance levels, casting doubt on the impo- 1a 1b 1c 2a 2b 2c sition of equality across countries. ADF t-test (2) Ϫ2.27 Ϫ3.89 Ϫ2.00 Ϫ2.74 Ϫ3.38 Ϫ2.07 Taking the panel results at face value, we would conclude Panel ADF (3) Ϫ4.68 Ϫ5.85 Ϫ5.16 Ϫ4.95 Ϫ4.76 Ϫ4.35 that financial liberalization is likely to lower saving rates. 1. P denotes the number of lags in the country-specific ADF test. For instance, in a benchmark model with the aggregate 2. The panel ADF t-test is based on Pedroni (1997a, 1997b). The test is distributed N(0, 1). 3. The panel integration test is based on Im, Pesaran, Shin (1995). The test is distributed N(0, 1). index, a typical combination of interest-rate change and 4. Cointegrating vectors: 1a: [(s/y)t,lnyt, rt, indext, ␲t, govst)] other liberalizations is estimated to lower saving by 1.2% of ␲ 1b: [(s/y)t,lnyt, rt, imact,imict, t, govst)] GNDI, using the estimates in the first column, and by 5.5% 1c: [(s/y)t,lnyt, rt, irskt, iliqt, ␲t, govst)] 2a: [(s/y)t,lnyt, indext, govst)] using the estimates in the fourth column. When the i-rsk/i-liq 2b: [(s/y)t,lnyt, imact,imict, govst)] 2c: [(s/y)t,lnyt, irskt, iliqt, govst)] pair is used, the decrease equals 1.75% of GNDI, using the estimates in the third column.27

D. Panel Results V. Econometric Evidence: Augmented Euler Equations It is worth investigating whether we can sharpen our The negative impact of financial liberalization on saving conclusions by assuming that coefficients are equal across found for some countries above suggests that liberalization countries, thereby making fuller use of the panel nature of may have weakened credit or liquidity constraints. Curi- the data. If the coefficients are truly the same across ously, despite dramatic changes in financial structure world- countries, then there should be a gain in efficiency by wide, the Euler equation literature on liquidity-constrained imposing such restrictions. The drawback may be that one consumption has not focused on time-varying constraints. may be imposing invalid restrictions, because of differences Here, we start with the Campbell-Mankiw (1989, 1991) in preferences, institutional settings, and nature of the approach of estimating an Euler equation augmented by the liberalization. Moreover, the construction of the index does presence of liquidity-constrained consumers, and attempt to not guarantee comparability of scale across countries. estimate variation in the proportion of constrained consum- Table 4 shows the results of this approach. Part I contains ers as liberalization proceeds. Thus, let ␾ be the proportion the benchmark model (using the aggregate index and its of unconstrained consumers and assume that the remainder subcomponents), while, in part II, the income variable is consume all their income. If ␾ is constant, then two standard excluded. In each part, the first three columns include only Euler equations are28 the levels of the explanatory variables and present estimates For constant interest and quadratic utility: of the cointegrating vector using SURE. The last three columns include first difference of the regressors, and an ⌬ ϭ Ϫ␾ ⌬ c ϩ␾⑀ ct (1 ) yt t (1) AR(1) error term is allowed for (as in the approximate dynamic GLS specification). The estimated coefficients on For time-varying interest and CRRA utility: the change terms (as well as the first-order autocorrelation coefficient) are allowed to differ across countries, and the ⌬ ln c ϭ µ␾ϩ⌬ln y c ϩ␾␴r ϩ␾⑀ (2) model is estimated by nonlinear methods. In all cases, we t t t t report the full set of results (including the coefficients on all where µ ϭϪ␴ln (1 ϩ␦), the level regressors) for both the specifications with the ␦ is the subjective discount rate, aggregate index and for those with its subcomponents. ␴ is the intertemporal elasticity of substitution, and In this constrained estimates, the coefficient on govern- ment saving is negative, but significantly different from 1, 26 The constrained dynamic GLS specification failed to converge for the implying a rejection of Ricardian equivalence. The coeffi- i-mic/i-mac pair. Note that Loayza, Schmidt-Hebbel, and Serve´n (2000) find a negative effect of credit flows on savings using panel data. cient of the inflation rate is positive and mostly significant. 27 We assumed an interest rate increase from Ϫ10% per annum to ϩ5% More importantly, from our point of view, the panel (typical of the sample, apart from the inflationary episodes in Chile and estimates imply that the real interest rate has a significant Ghana). We set the change in index to 7, in i-rsk to 6, and in i-liq to 4, equal to their respective mean change between the initial and final year of the positive effect and that the aggregate index of financial sample. liberalization has a negative effect on saving, which is 28 Cf. Attanasio and Browning (1995).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 253

TABLE 2.—COUNTRY-BY-COUNTRY ESTIMATES OF THE COINTEGRATING VECTOR:BENCHMARK MODEL Part I: OLS Estimates Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

ln yt 0.17 0.10 0.09 0.19 0.18 0.13 Ϫ0.01 0.28 (3.38) (2.55) (2.52) (8.32) (2.71) (3.52) (Ϫ0.19) (1.01) rt Ϫ0.04 Ϫ0.17 Ϫ0.61 Ϫ0.20 Ϫ0.65 0.11 Ϫ0.06 0.12 (Ϫ0.74) (Ϫ0.64) (Ϫ4.03) (Ϫ1.23) (Ϫ1.99) (2.66) (Ϫ1.18) (0.28) ␲t Ϫ0.03 Ϫ0.14 Ϫ0.68 Ϫ0.06 Ϫ0.86 0.13 Ϫ0.09 0.04 (Ϫ0.52) (Ϫ0.53) (Ϫ3.81) (Ϫ0.48) (Ϫ2.44) (2.42) (Ϫ1.27) (0.08) govst 0.26 Ϫ0.64 Ϫ1.42 Ϫ0.01 Ϫ1.32 Ϫ0.51 Ϫ0.32 Ϫ0.14 (0.89) (Ϫ3.19) (Ϫ2.67) (Ϫ2.61) (Ϫ7.88) (Ϫ2.35) (Ϫ1.75) (Ϫ0.39) Indext 0.001 0.005 0.002 Ϫ1.08 Ϫ0.003 Ϫ0.01 0.02 0.005 (0.32) (1.31) (0.63) (Ϫ2.59) (Ϫ0.47) (Ϫ7.10) (4.16) (0.18) R2 0.58 0.75 0.47 0.88 0.82 0.95 0.79 0.21

I-mact Ϫ0.03 Ϫ0.001 Ϫ0.015 Ϫ0.001 0.002 Ϫ0.003 0.012 Ϫ0.011 (Ϫ3.60) (Ϫ0.33) (Ϫ0.76) (Ϫ0.06) (0.36) (Ϫ1.04) (2.16) (Ϫ0.41) I-mict 0.03 0.04 0.005 Ϫ0.015 Ϫ0.001 Ϫ0.012 0.012 Ϫ0.05 (3.88) (1.67) (1.14) (Ϫ1.40) (0.25) (Ϫ3.89) (1.27) (Ϫ1.20)

I-rskt 0.008 0.008 0.002 0.002 0.010 Ϫ0.06 (0.62) (0.19) (0.19) (0.29) (1.18) (Ϫ1.23) I-liqt Ϫ0.004 0.004 0.004 0.001 Ϫ0.02 0.50 (Ϫ0.44) (0.18) (0.18) (0.29) (Ϫ3.17) (0.47) %var 0.62 0.80 0.73 0.71 0.44 0.73 0.75 0.64 NOBS 25 25 24 24 25 25 25 20 Part II: DGLS Estimates Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

ln yt 0.32 0.17 0.21 0.47 0.05 0.24 Ϫ0.05 0.34 (3.13) (3.44) (4.74) (4.35) (1.14) (4.64) (Ϫ0.25) (0.68) rt 0.13 Ϫ1.49 Ϫ0.81 Ϫ0.35 Ϫ0.69 Ϫ0.06 Ϫ0.06 Ϫ1.36 (1.45) (Ϫ2.22) (Ϫ4.89) (Ϫ1.6) (Ϫ1.71) (Ϫ1.00) (Ϫ0.71) (Ϫ1.33) ␲t 0.15 Ϫ1.2 Ϫ0.68 Ϫ0.38 Ϫ0.79 0.00 Ϫ0.08 Ϫ2.1 (1.58) (Ϫ2.31) (Ϫ3.64) (Ϫ1.64) (Ϫ2.17) (1.64) (Ϫ0.68) (Ϫ1.45) govst Ϫ1.08 1.02 Ϫ2.16 Ϫ0.96 Ϫ0.97 Ϫ0.86 Ϫ0.17 0.18 (Ϫ2.01) (1.21) (Ϫ4.2) (Ϫ2.38) (Ϫ4.9) (Ϫ3.66) (Ϫ0.38) (0.31) Indext Ϫ0.001 0.02 Ϫ0.004 Ϫ0.02 0.01 Ϫ0.01 0.02 0.06 (Ϫ0.10) (2.02) (Ϫ1.03) (Ϫ2.36) (1.34) (Ϫ3.82) (2.1) (0.73) R2 0.84 0.81 0.80 0.94 0.92 0.98 0.78 0.12

I-mact Ϫ0.007 0.007 Ϫ0.020 0.02 0.002 0.001 0.011 0.07 (Ϫ0.47) (0.68) (Ϫ0.97) (2.07) (0.61) (0.18) (0.95) (0.81) I-mict 0.02 0.04 Ϫ0.004 Ϫ0.03 Ϫ0.012 Ϫ0.010 0.015 0.05 (0.9) (1.47) (0.89) (Ϫ3.66) (Ϫ3.7) (Ϫ3.62) (0.91) (0.7)

I-rskt 0.003 0.06 0.010 0.004 0.007 0.20 (0.22) (1.95) (0.68) (0.83) (0.94) (1.07) I-liqt Ϫ0.04 Ϫ0.04 Ϫ0.03 Ϫ0.012 Ϫ0.010 Ϫ0.04 (Ϫ2.71) (1.35) (Ϫ1.01) (Ϫ3.57) (Ϫ2.23) (0.29) NOBS 25 25 24 24 25 25 25 20

Notes: 1) Dependent variable is saving ratio. 2) In addition to the estimated coefficients (t-statistics in parentheses) and R2 of the basic regression, the table shows the estimated coefficients of the split indices in the two alternative versions where these replace the basic liberalization index. 3) %var denotes the proportion of total liberalization matrix variance accounted for first principal component (Index). 4) The dynamic GLS estimates are obtained by including the contemporaneous changes of all the RHS variables as additional regressors and allowing for AR(1) errors (not shown).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 254 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 3.—COUNTRY-BY-COUNTRY ESTIMATES OF THE SAVING RATIO COINTEGRATING VECTOR-ALTERNATIVE SPECIFICATIONS—DGLS ESTIMATES Part I: Excluding Income Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

rt Ϫ0.1 0.17 Ϫ0.69 Ϫ0.17 Ϫ1.41 Ϫ0.01 Ϫ0.06 Ϫ1.1 (Ϫ1.27) (Ϫ0.24) (Ϫ2.65) (Ϫ0.49) (Ϫ3.71) (Ϫ0.12) (Ϫ0.64) (Ϫ2.93) ␲␶ Ϫ0.01 Ϫ0.05 Ϫ0.83 Ϫ0.07 Ϫ1.44 0.11 Ϫ0.08 Ϫ1.08 (Ϫ0.09) (Ϫ0.08) (Ϫ2.66) (0.22) (Ϫ3.61) (0.94) (Ϫ0.79) (Ϫ2.78) Govst 0.69 Ϫ1.42 Ϫ0.48 Ϫ0.11 Ϫ0.82 Ϫ0.7 Ϫ0.17 Ϫ0.08 (1.39) (Ϫ1.91) (Ϫ1.02) (Ϫ0.18) (Ϫ6.99) (Ϫ1.7) (Ϫ0.4) (Ϫ.18) Indext 0.03 Ϫ0.002 0.01 Ϫ0.001 0.01 Ϫ0.01 0.01 0.04 (1.45) (Ϫ0.32) (3.32) (Ϫ0.12) (2.8) (Ϫ2.06) (2.45) (2.78) R2 0.63 0.71 0.37 0.85 0.82 0.95 0.80 0.34

I-mact 0.0006 Ϫ0.0009 0.04 0.03 0.003 Ϫ0.002 0.01 0.05 (0.03) (Ϫ1.02) (3.15) (2.37) (1.43) (Ϫ0.36) (0.98) (1.73) I-mict 0.08 0.06 Ϫ0.001 Ϫ0.02 Ϫ0.01 Ϫ0.01 Ϫ0.01 0.0004 (2.73) (2.04) (Ϫ0.11) (Ϫ1.41) (Ϫ4.19) (Ϫ1.97) (0.99) (0.02)

I-rskt 0.03 0.04 Ϫ0.01 0.003 Ϫ0.003 Ϫ0.03 (1.95) (1.24) (Ϫ0.51) (1.18) (0.18) (Ϫ0.31) I-liqt Ϫ0.11 Ϫ0.05 0.05 Ϫ0.01 Ϫ0.01 0.09 (Ϫ1.75) (Ϫ1.35) (1.14) (Ϫ3.99) (Ϫ1.21) (1.64) NOBS 25 25 24 24 25 25 25 20 Part II: Excluding Inflation and Interest Rate Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

ln yt 0.14 0.16 0.04 0.41 0.06 0.26 Ϫ0.08 0.07 (2.38) (4.57) (0.56) (5.88) (0.65) (4.56) (Ϫ0.82) (0.21) Govst Ϫ0.2 Ϫ0.46 Ϫ0.72 Ϫ1.0 Ϫ1.01 Ϫ1.29 Ϫ0.18 0.15 (Ϫ0.48) (Ϫ2.33) (Ϫ0.74) (Ϫ2.9) (Ϫ4.63) (Ϫ8.8) (Ϫ0.55) (0.28) Indext 0.003 0.0003 0.002 Ϫ0.02 0.004 Ϫ0.010 0.016 Ϫ0.016 (0.27) (0.13) (0.32) (Ϫ2.26) (0.56) (Ϫ3.54) (3.02) (Ϫ0.32) R2 0.73 0.79 0.42 0.94 0.86 0.96 0.80 0.08

I-mact Ϫ0.03 Ϫ0.003 Ϫ0.03 Ϫ0.004 Ϫ0.001 Ϫ0.002 0.012 Ϫ0.009 (Ϫ3.23) (Ϫ0.59) (Ϫ1.32) (0.49) (Ϫ0.17) (Ϫ0.42) (1.26) (Ϫ0.28) I-mict 0.05 0.02 0.005 Ϫ0.02 Ϫ0.012 Ϫ0.011 0.008 0.02 (3.57) (0.71) (0.79) (Ϫ2.64) (Ϫ3.3) (Ϫ2.9) (0.75) (Ϫ0.86)

I-rskt 0.02 0.03 0.042 0.001 0.017 Ϫ0.02 (1.86) (1.24) (3.85) (0.21) (2.1) (Ϫ0.38) I-liqt Ϫ0.06 Ϫ0.04 Ϫ0.08 Ϫ0.013 Ϫ0.02 0.003 (Ϫ2.05) (Ϫ1.2) (Ϫ3.81) (Ϫ3.4) (Ϫ3.72) (0.03) NOBS 25 25 24 24 25 25 25 20 Part III: Income, Saving Adjusted for Domestic Capital Gains and Losses Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

ln yt 0.17 0.13 0.19 0.47 Ϫ0.06 0.11 Ϫ0.16 0.68 (1.97) (3.17) (5.05) (3.93) (Ϫ0.83) (3.94) (Ϫ0.93) (1.47) rt 0.04 Ϫ0.53 Ϫ0.83 Ϫ0.34 Ϫ1.14 0.004 Ϫ0.05 Ϫ3.72 (0.66) (Ϫ1.50) (Ϫ5.01) (Ϫ1.51) (Ϫ3.13) (0.5) (Ϫ0.75) (Ϫ3.38) ␲␶ 0.04 Ϫ0.60 Ϫ0.003 Ϫ0.38 Ϫ1.25 0.24 Ϫ0.18 Ϫ4.98 (0.58) (Ϫ1.84) (Ϫ0.80) (Ϫ1.56) (Ϫ3.50) (4.64) (Ϫ1.74) (Ϫ3.45) govst Ϫ0.24 0.17 Ϫ2.0 Ϫ0.88 Ϫ0.69 Ϫ0.99 Ϫ0.14 Ϫ1.77 (Ϫ0.70) (0.32) (Ϫ4.51) (Ϫ2.18) (Ϫ3.46) (Ϫ6.24) (Ϫ0.33) (Ϫ2.24) Indext 0.007 0.005 Ϫ0.003 Ϫ0.02 0.016 Ϫ0.004 0.02 0.26 (0.43) (1.20) (Ϫ0.80) (Ϫ2.26) (2.11) (Ϫ1.82) (3.22) (3.19) R2 0.87 0.68 0.73 0.94 0.87 0.95 0.80 0.48

I-mact 0.001 Ϫ0.004 Ϫ0.017 0.02 0.002 Ϫ0.002 0.015 0.19 (0.09) (Ϫ0.39) (Ϫ0.86) (2.46) (0.70) (Ϫ0.76) (1.73) (1.56) I-mict 0.04 0.04 Ϫ0.002 Ϫ0.03 Ϫ0.015 Ϫ0.005 0.03 0.08 (1.44) (1.1) (Ϫ0.49) (Ϫ4.20) (Ϫ4.44) (Ϫ1.58) (1.96) (1.10)

I-rskt 0.010 0.10 0.009 0.005 Ϫ0.009 0.36 (0.86) (3.10) (0.66) (1.28) (Ϫ1.32) (1.67) I-liqt Ϫ0.03 Ϫ0.12 Ϫ0.02 Ϫ0.016 0.001 0.10 (Ϫ2.23) (Ϫ2.89) (Ϫ0.93) (Ϫ4.99) (0.17) (0.49) NOBS 25 23 24 24 25 24 25 17

Notes: See footnotes to table 2.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 255

TABLE 4.—SAVING RATIO:PANEL RESULTS Part I: Benchmark Specification Method SURE Constrained DGLS

ln yt 0.11 0.13 0.11 0.207 NA 0.17 (16.93) (19.34) (14.52) (14.13) (31.67) rt 0.06 .07 .05 0.101 0.02 (3.42) (4.74) (3.13) (3.62) (1.82) ␲␶ 0.06 .07 .04 0.168 0.005 (3.49) (4.33) (2.35) (10.696) (0.43) govst Ϫ0.74 Ϫ0.73 Ϫ.69 Ϫ0.660 Ϫ1.22 (Ϫ14.42) (Ϫ14.4) (Ϫ13.37) (Ϫ16.446) (Ϫ34.36) indext Ϫ0.003 Ϫ0.010 (Ϫ3.12) (Ϫ8.440)

I-mact Ϫ0.0002 (Ϫ0.14) I-mict Ϫ0.008 (Ϫ5.06)

I-rskt 0.001 0.006 (1.21) (10.49) I-liqt Ϫ0.01 Ϫ0.016 (Ϫ6.96) (Ϫ21.15) Part II: Excluding Income Method SURE Constrained DGLS

rt 0.004 0.009 Ϫ0.01 0.17 0.17 0.18 (0.23) (0.48) (Ϫ0.91) (5.90) (7.04) (9.56) ␲␶ 0.06 0.07 0.006 0.11 0.16 0.07 (3.28) (3.30) (0.30) (3.00) (4.96) (2.74) govst Ϫ0.55 Ϫ0.54 Ϫ0.57 Ϫ0.61 Ϫ0.6 Ϫ0.61 (Ϫ12.24) (Ϫ11.71) (Ϫ11.18) (Ϫ5.58) (Ϫ6.81) (Ϫ6.85) indext 0.006 Ϫ0.004 (7.17) (Ϫ1.78)

I-mact 0.004 0.001 0.002 (2.90) (0.47) (1.25) I-mict 0.002 Ϫ0.01 Ϫ0.02 (0.97) (Ϫ5.21) (Ϫ6.75)

I-rskt 0.01 (13.01) I-liqt Ϫ0.02 (Ϫ9.38)

Notes: 1) Number of observations is 160. 2) ‘‘NA’’ denotes not available because of noncovergence.

yc is the per capita income of the constrained consumers, ⌬␾ ⌬ ϭ Ϫ␾ ⌬ ϩ␾␴ Ϫ t assumed to be a constant29 fraction ␩ of per capita income in ln ct (1 t) ln yt t rt ␾ Ϫ (4) the economy. t 1 ϫ Ϫ ϩ ␩ ϩ␾ ϩ␰ If we allow ␾ to change through time, then equation (1) (ln ytϪ1 ln ctϪ1 ln ) tµ t and (2) become30 where ␰t ϭ␾t⑀t. ⌬␾ Equation (3) and (4) emphasize that the sensitivity of ⌬ ϭ Ϫ␾ ⌬ Ϫ t ␩ Ϫ ϩ␰ ct (1 t) yt ␾ ( ytϪ1 ctϪ1) t (3) consumption to income varies over time, as the share of tϪ1 liquidity-constrained consumers varies. Indeed, the sensitiv- ity of aggregate consumption to current income is due to the 29 Assuming ␩ constant is necessary to have equation (3) and (4) below in a tractable form. fact that some consumers consume their income, and as such 30 ϭ Ϫ␾ To derive equation (3), define per capita consumption ct (1 t) is proportional to the relative size of the credit-constrained u ϩ␾ c ct tct , where the superscripts u and c denote unconstrained and group. The sensitivity of consumption to the interest rate constrained consumers, respectively. Then, take first differences to obtain ⌬ Ϫ␾ u ϩ⌬␾ c ϭ Ϫ␾ ⌬ u ϩ u ⌬ Ϫ␾ ϩ␾ ⌬ c ϩ also changes over time in equation (4). But the main novelty (1 t) ct t ct (1 t) ct ctϪ1 (1 t) t ct c ⌬␾ c ϭ␩ ␾ ctϪ1 t. Substituting the definition of ctϪ1 and using ct yt gives is that the time variation of introduces additional regres- equation (3). The derivation of equation (4) proceeds along similar lines, sors in equation (3) and (4). In particular, there is a new term using a geometric mean with population weights for average per capita consumption (whereas, in the empirical implementation, we substitute a of error-correction type involving lagged consumption and simple mean consumption). income, whose coefficient is equal to the rate of change in

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 256 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 5.—EXCESS SENSITIVITY TESTS AND THE AUGMENTED EULER EQUATION FOR CONSUMPTION (GMM ESTIMATES) Panel Chile Ghana Indonesia Korea Malaysia Mexico Turkey Zimbabwe

Part I: fixed ␾ ⌬ln yt 0.68 0.550 1.705 1.606 0.359 1.185 0.687 0.575 1.088 (23.031) (3.222) (1.056) (1.113) (2.089) (2.593) (3.127) (1.445) (3.392) rt Ϫ0.002 0.076 0.349 0.490 0.275 0.767 0.181 0.028 0.280 (Ϫ0.109) (2.464) (Ϫ0.828) (0.955) (1.412) (1.056) (2.688) (0.233) (Ϫ0.837) BG 0.653 0.041 0.870 0.424 0.042 0.041 0.634 0.491 NOBS 152 24 24 23 23 24 24 24 19 Part II: ␾ affected by liberalization ␣0 NA Ϫ0.81 Ϫ7.566 Ϫ2.673 6.683 Ϫ0.136 Ϫ4.811 (Ϫ2.914) (Ϫ1.454) (Ϫ0.462) (0.30) (Ϫ0.016) (Ϫ3.411) ␣1 Ϫ0.109 0.272 Ϫ1.813 Ϫ1.42 0.701 0.731 (Ϫ0.735) (0.90) (Ϫ0.733) (Ϫ0.591) (0.49) (2.36) ␩ 0.771 0.841 0.540 0.570 0.560 1.471 (2.668) (2.241) (91.844) (18.943) (2.791) (34.736) ␴Ϫ0.001 1.066 6.278 1.055 0.449 0.477 (Ϫ0.013) (1.111) (0.163) (1.255) (1.728) (0.472) OIR 0.057 0.880 0.931 0.079 0.655 0.916 NOBS 24 23 23 24 24 24

Notes: 1) See equation (4) and preceding text for a definition of ␩ and ␴. ␣1 captures the dependence of the proportion of unconstrained consumers upon the financial liberalization index. 2) BG denotes the marginal significance level for the Breusch-Godfrey test for serial correlation up to the second order. 3) OIR denotes the marginal significance level of the test of overidentifying restrictions. 4) ‘‘NA’’ denotes not available because of noncovergence.

the proportion of unconstrained consumers. This conse- babwe), we have not succeeded in obtaining convergence. quence of time-varying liquidity constraints seems to have For the remaining six countries, the results, on the whole, been overlooked in the literature. show lack of a significant relationship between ␲t and the 31 The error term in equation (3) and (4) also depends on index. In the only case in which ␣1 is significant at ␾t giving rise to a need to seek consistent estimates by conventional levels (Turkey), it is indeed positive. However, IV or GMM techniques. For instance, assume that the Turkey was the country in which the saving-function results set of instruments used, ztϪ1, belongs to the information suggested a positive direct effect of liberalization on saving. set available at time t Ϫ 1. If ␾t is also a function of Bearing in mind the indications from the saving function variables known at time t Ϫ 1, then E(ztϪ1␰t) ϭ 0, because (section IV)—that liquidity constraints might be more the forecast error ⑀t is by definition uncorrelated with associated with subindiced i-mic and i-liq—we experi- variables at t Ϫ 1. More precisely, E(ztϪ1␰t) ϭ E(ztϪ1␾t⑀t) ϭ mented with substituting these for the overall index in the Ez,␲[E(ztϪ1␾t⑀t 0ztϪ1, ␾t)] ϭ E[ztϪ1␾tE(⑀t 0ztϪ1, ␾t)] ϭ 0if Euler equation. However, the results do not improve, and not E(⑀t 0ztϪ1, ␾t) ϭ 0. much is learned from this additional exercise. (The results The last assumption is plausible if financial liberalization are not reported for brevity’s sake.) measures are actually effective one period after being The Euler equation results suggest that financial liberaliza- implemented, so that ␾t depends upon the lagged value of tion has had little impact on the amount of credit available to the liberalization index. consumers through the formal financial sector. Alternatively, The final step is to relate ␾t to financial liberalization. We the inconclusive results may stem from the econometric will assume that ␾t is an increasing function of the index (or problem of pinning down what is essentially the coefficient subindices) of financial liberalization lagged one period, of the product of a nonstationary variable (index,orits indextϪ1. In table 5, we summarize the empirical results for subcomponents), with a stationary one (⌬ ln yt). More the specification that includes the interest rate, estimated by generally, one might question the adequacy of the instru- GMM (past values of the included variables are used as ments used in estimating the augmented Euler equations. instruments). In the first part of the table, we present the A further reason for us to expect to find (as we do) a estimates of the model under the assumption of a constant stronger influence of liberalization in our saving equation by ␾t. (See equation (2).) For the majority of countries, the comparison with the Euler equation is that the dependent coefficient of income is significant at conventional levels. variable of the former relates to total private saving (busi- The coefficient is positive and significant when it is re- ness as well as household sectors), while the latter relates in stricted to be equal across countries. (See the first column.) principle only to household-sector behavior. Just as it is All this evidence is consistent with the presence of liquidity more sensitive to exogenous shocks (Honohan and Atiyas, constraints. In the second part of the table, we have adopted ␾ ␾ ϭ ϩ 31 ␣ a logistic specification for t, so that t 1/(1 exp The basic sense of the results does not change if we allow 0 to be different when the growth rate of income is negative, or if we choose a (Ϫ␣0 Ϫ␣1 indextϪ1)), and we have estimated model (4). ␣ different functional form (such as the Gumbel) for ␾. Finally, note that we Iffinancial liberalization relaxes financial constraints, 1 have failed to achieve convergence when we impose equality of coeffi- would be positive. For two countries (Ghana and Zim- cients across countries.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 257

1993), business saving in developing countries may be more policy from financial liberalization would require data on a influenced by liberalization than household behavior, espe- larger set of countries. cially as the latter may be more conditioned by informal Our estimates of an augmented Euler equation for con- finance than by reforms that affect mainly the formal sector. sumption confirm previous evidence of excess sensitivity of consumption to income. However, with the exception of VI. Conclusions Turkey, there is not much evidence that such sensitivity has Attention has long focused on the interest-elasticity of decreased with financial liberalization, although this may be saving as a key parameter in gauging the impact of financial due to the econometric difficulty of obtaining precise liberalization. Our econometric results confirm the visual estimates of the parameters. impression from the figures, as well as much previous For the present, our results suggest that, while financial literature, that there is no strong, reliable, interest-rate effect liberalization may sometimes increase private saving, the on saving. Only when the data is pooled and one assumes opposite is more likely to be the case. Considering that 32 that the long-run coefficients are equal across countries (a government saving can also be adversely affected, it is at restriction that the data rejects) can we find evidence of a best unwise to rely on an increase in saving as the channel significant, positive, interest-rate effect on saving, and even through which financial liberalization can be expected to then the effect is small. increase growth. Our aggregate index of financial liberalization captures Even if financial liberalization does not increase private several aspects of reform that are not fully represented by saving, it does not follow that the process contracts the changes in the interest rate, such as the increased availability volume of funds applied to productive investment. For one of a variety of saving media with better risk-return character- thing, liberalization can increase the inflow of capital, istics or the relaxation of borrowing constraints. including the return of flight capital (Bartolini and Drazen But here too, the econometric evidence on the impact of (1997)). For another, by strengthening market discipline and reform on saving is mixed. When saving functions are increasing the autonomy of banks and other financial estimated for each of the countries separately, the overall institutions, the various elements of the reform process can long-run effect is found to be significantly negative for two have the effect of eliminating less-productive uses of (Korea and Mexico), positive for two (Ghana and Turkey), loanable funds. These two potentially important aspects are with no clear effect discernible in the others. When the not considered in the present paper. long-run responses are constrained to be equal, the effect of the financial liberalization index is significantly negative REFERENCES and large enough to offset (in these constrained estimates) Alessie, R., M. P. Devereux, and G. Weber, ‘‘Intertemporal Consumption, the estimated positive effect of the interest-rate increases Durables and Liquidity Constraints: A Cohort Analysis,’’ European that have accompanied the reforms. Economic Review, 41:1 (1993), 37–60. Attanasio, O., P. and M. Browning, ‘‘Consumption over the Life Cycle and Delving behind the aggregate index to explore the differ- Over the Business Cycle,’’ American Economic Review 85:5 ential impact of different elements of the overall process of (1995), 1187–1207. financial liberalization, we found contrasts between the Balassa, B., ‘‘The Effects of Interest Rates on Savings in Developing Countries,’’ Banca Nazionale del Lavoro Quarterly Review 172 estimated impact of micro and macro reforms, and between (1990), 101–118. the liberalization of access to credit and remaining reforms. Bartolini, L., and A. Drazen, ‘‘Capital Account Liberalization as a Signal,’’ Interestingly, a negative impact on saving is more reliably American Economic Review 87 (1997), 138–155. Bayoumi, T., ‘‘Financial Deregulation and Household Saving,’’ Economic associated with micro than with macro, and with reforms Journal 103 (1993), 1432–1443. liberalizing the allocation of credit (thereby reducing liquid- Blanchard, O. J., ‘‘Debts, Deficits and Finite Horizons,’’ Journal of ity constraints). These negative effects are statistically Political Economy 93 (1985), 1045–1076. Bonser-Neal, C., and K. Dewenter, ‘‘Does Financial Market Development significant for three countries (Malaysia, Mexico, and either Stimulate Savings? Evidence From Emerging Market Stock Mar- Chile or Korea), and in the panel. The panel results based on kets,’’ University of Indiana mimeograph (1996). the decomposed subindices confirm a negative and signifi- Browning, M., and A. M. Lusardi, ‘‘Household Savings: Micro Theories and Micro Facts,’’ Journal of Economic Literature 34 (Dec. 1996), cant association between financial liberalization and saving 1797–1855. that dominates the positive interest-rate effect. These results Burnside, C., ‘‘Private Consumption and Public Sector Deficits in Mexico, suggest that the dominant factor is the negative impact on 1985–1995,’’ World Bank mimeograph (1996). Campbell, J., and N. G. Mankiw, ‘‘Consumption, Income and Interest saving of a relaxation, varying from country to country, of Rates: Reinterpreting the Time Series Evidence,’’in O. J. Blanchard borrowing constraints for households and firms. and S. Fischer (Eds.), NBER Macroeconomic Annual 1989 (Cam- These estimates, although obtained from dynamic mod- bridge, MA: MIT Press, 1989), 185–216. els, refer to long-run effects. Pinning down the short-run effects seems elusive. The impact effect of liberalization is 32 As noted earlier, financial liberalization phases out below-market often associated with a transitory boom in credit and a surge financing for government, and may also increase expenditures by requiring in property prices, but whether this occurs depends strongly that subsidies be explicit. However, in addition to longer-term gains in revenue due to more-rapid growth, in the near-term bank privatization, at on the conduct of monetary policy and macro-management least in the right regulatory environment, can save substantial sums, as in generally. Sorting out the short-run influence of macro Argentina (Clarke and Cull, 1998).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 258 THE REVIEW OF ECONOMICS AND STATISTICS

——— ‘‘The Response of Consumption to Income: A Cross-Country Miles, D., ‘‘Housing Markets, Consumption and Financial Liberalization Investigation,’’ European Economic Review 35 (1991), 715–721. in the Major Economies,’’ European Economic Review 36 (1992), Caprio, G. Jr., I. Atiyas, and J. Hanson, Financial Reform: Theory and 1093–1135 (with discussion). Evidence (Cambridge University Press, 1994). Muellbauer, J., ‘‘The Assessment: Consumer Expenditure,’’Oxford Review Caprio, G., Jr., and D. Klingebiel, ‘‘Bank Insolvency: Bad Luck, Bad of Economic Policy 10 (1994), 1–39. Policy or Bad Banking?’’ Presented to the World Bank Annual Muellbauer, J., and A. Murphy, ‘‘Is the U.K. Balance of Payments Conference on Development Economics (1996). Sustainable?’’ Economic Policy 11 (1990), 345–383. ——— ‘‘Bank Insolvency, Cross Country Experience,’’ World Bank ——— ‘‘Income Expectations, Wealth and Demography in the Aggregate UK Consumption Function,’’ Paper presented to the HM Treasury Policy Research Paper No. 1620 (July 1996a). Academic Panel (1993). Carroll, C. D., ‘‘The Buffer-Stock Theory of Saving: Some Macroeco- Ostry, J. D., and J. Levy, ‘‘Household Savings in France: Stochastic nomic Evidence,’’ Brookings Papers on Economic Activity 1992–2 Income and Financial Deregulation,’’ International Monetary Fund (1992), 61–135. Staff Papers 4 (1995), 375–397. Clarke, G., and R. Cull. ‘‘Why Privatize? The Case of Argentine Public Pedroni, P., ‘‘Panel Cointegration: Asymptotic and Finite Sample Proper- Provincial Banks,’’ World Bank Policy Research Working Papers n. ties of Pooled Time Series Tests with an Application to the PPP 1972 (1998). Hypothesis,’’ University of Indiana mimeograph (1997a). Corbo, V., and K. Schmidt-Hebbel, ‘‘Public Policies and Saving in ——— ‘‘Approximate Critical Values for Cointegration Tests in Heteroge- Developing Countries,’’ Journal of Development Economics 36 neous Panels with Multiple Regressors,’’ University of Indiana (1991), 89–115. mimeograph (1997b). Deaton, A. S., ‘‘Involuntary Saving through Unanticipated Inflation,’’ Reynoso, A., Essays on the Macroeconomic Effects of Monetary Reform, American Economic Review 67 (1977), 899–910. Price Controls and Financial Repression, MIT Dissertation (1989). ——— ‘‘Savings and Liquidity Constraints,’’ Econometrica 59 (1991), Rossi, N., ‘‘Government Spending, the Real Interest Rate, and the 1221–1248. Behavior of Liquidity Constrained Consumers in Developing ——— Understanding Consumption (Oxford: Clarendon Press, 1992). Countries,’’ IMF Staff Papers 35 (1988), 104–140. de Melo, J., and J. Tybout. ‘‘The Effects of Financial Liberalization on Schmidt-Hebbel, K., S. B. Webb, and G. Corsetti, ‘‘Household Saving in Savings and Investment in Uruguay,’’ Economic Development and Developing Countries: First Cross-Country Evidence,’’ World Bank Cultural Change 34 (1986), 561–588. Economic Review 6 (1992), 529–547. Shaw, E. S., Financial Deepending in Economic Development (New York: Demetriades, P., and K. Luintel. ‘‘The Direct Cost of Financial Repression: Oxford University Press, 1973). Evidence from India,’’ this REVIEW 79:2 (1997), 311–320. Srinivasan, T. N., ‘‘Saving in the Development Process,’’ in J. H. Gapinski Dornbusch, R., and A. Reynoso. ‘‘Financial Factors in Economic Develop- (Ed.), The Economics of Saving (Dordrecht: Kluwer, 1993). ment,’’ American Economic Review, Papers and Proceedings, 79 Stock, J. H., and M. W. Watson, ‘‘A Simple Estimator of Cointegrating (1989), 204–209. Vectors in Higher Order Integrated Systems,’’ Econometrica 61 Fry, M. J., Money, Interest, and Banking in Economic Development, 2nd (1993), 783–820. Ed. (Baltimore: Johns Hopkins, 1995). Taylor, L., Structuralist Macroeconomics: Applicable Models for the Third Hall, R. E., ‘‘Intertemporal Substitution in Consumption,’’ Journal of World (New York: Basic Books, 1983). Political Economy 96 (1978), 339–357. Vaidyanathan, G., ‘‘Consumption, Liquidity Constraints and Economic Haque, N. U., and P. Montiel. ‘‘Consumption in Developing Countries: Development,’’ Journal of Macroeconomics 15 (1993), 591–610. Tests for Liquidity Constraints and Finite Horizons,’’ this REVIEW Van Wijnbergen, S., ‘‘Stagflationary Effects of Monetary Stabilization 34 (1989), 408–415. Policies: A Quantitative Analysis of South Korea,’’ Journal of Honohan, P., and I. Atiyas, ‘‘Intersectoral Financial Flows in Developing Development Economics 10 (1982), 133–169. Countries,’’ Economic Journal 103:418 (1993), 666–679. Zeldes, S. P., ‘‘Consumption and Liquidity Constraints: An Empirical Im, K. S., M. H. Pesaran, and Y. Shin, ‘‘Testing for Unit Roots in Investigation,’’ Journal of Political Economy 97 (1989), 305–346. Heterogeneous Panels,’’ University of Cambridge mimeograph (1996). APPENDIX A Jappelli, T., and M. Pagano, ‘‘Consumption and Capital Market Imperfec- tions: An International Comparison,’’ American Economic Review Building an Index for Financial Liberalization 79 (1989), 1088–1105. ——— ‘‘Saving, Growth and Liquidity Constraints,’’ Quarterly Journal of Financial liberalization packages generally consist of a wide range of Economics 109 (1994), 83–109. different measures. We identify eight main dimensions, six of them King, Mervin A., ‘‘The Economics of Saving,’’ in Arrow, K., and S. grouped under the heading ‘‘domestic financial liberalization,’’ one on Honkanpohja (Eds.), Frontiers of Economics (Oxford: Basil Black- securities markets, and the last on international financial liberalization. well, 1985). Dummy variables are constructed to represent liberalization: Each takes Koskela, E., H. A. Loikkanen, and M. Vire´n, ‘‘House Prices, Household the value 1 when the system was liberalized in the relevant dimension. Savings and Financial Market Liberalization in Finland,’’ European The result is a matrix X of dummies for each country. Each column Economic Review 36 (1992), 549–558. represents a single dummy (the rows represent years). Although there are Koskela, E., and M. Vire´n, ‘‘Taxation and Household Saving in Open eight main dimensions of policy change, the number of columns differs Economies—Evidence from the Nordic Countries,’’ Scandinavian from country to country because the complexity of the policy changes Journal of Economics 96 (1994), 425–441. differed, often requiring more than one dummy for a given dimension. Our Lehmussaari, O. P., ‘‘Deregulation and Consumption Saving Dynamics in approach to summarizing liberalization is to reduce the dimensionality of the Nordic Countries,’’ IMF Staff Papers 37 (1990), 71–93. this matrix using principal components. Levine, R., ‘‘Financial Development and Economic Growth,’’ Journal of In addition to using the first principal component of the matrix X,we also computed a weighted average of the more relevant components that Economic Literature 35:2 (1997), 688–727. explain, cumulatively, 95% of the total variation of X. The weights used are Levine, R., and S. Zervos, ‘‘Stock Markets, Banks and Economic the fraction of the total standard deviation explained by each component, Growth,’’ World Bank Policy Research Working Paper 1690 so that the first principal component is more heavily weighted than the (1996). second, and so on. Loayza, N., K. Schmidt-Hebbel, and L. Serve´n, ‘‘What Drives Private Finally, as described in the text, we also computed two pairs of Saving across the World,’’ this review 82:2 (2000), 165–181. subindices by constructing two submatrices from X (and extracting the first McKinnon, R., Money and Capital in Economic Development (Washing- principal components) as follows: ton, D.C.: Brookings Institution, 1973). MacKinnon, J. G., ‘‘Critical Values for Cointegration Tests,’’ Economics, University of California at San Diego, Discussion Paper 90-4 Pair 1 i-mac: 1.a, 1.b and 3; i-mic the others (January 1990). Pair 2 i-liq: 1.b, 1.c, 1.d; i-rsk the others

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 259

The columns of X representing the timing of the most important (f) 1982 (Dec.) Central bank ‘‘suggests’’ deposit rates. liberalization measures and are ordered according to the following scheme: (g) 1986 (Jan.) Controls (i.e., suggestions) on interest rates are definitely abolished. 1. Domestic Financial Liberalization We ignore events (a) to (d) because they cancel out within a year, and our data is annual. This leaves us with two dummies. The first one is ϭ 1 after 1.a Interest rates 1975 to capture event (e). The second one is ϭ 1 between 1976 and 1982 and then after 1985 to capture events (f) and (g). Dummies for the timing of liberalization of interest rates (Dri ϭ 1 when interest rates are freed.) 1.b Procompetition measures

1.b Procompetition measures 1975 Entry barriers are lowered. Dummy variable ϭ 1 after 1974. Includes lowering of entry barriers, permissions to offer new services, and other measures intended to foster competition in the financial markets. (Dco ϭ 1 when measures are taken.) 1.c Reserve requirements

( h ) 1974 (Oct.) Reserve requirements on short-term (1–12 m) 1.c Reserve requirements time deposits are reduced (from 40% to 8%). ( i ) 1975 (Jul.) Reserve requirements on demand deposits are Most financial liberalization packages include a reduction in reserve reduced (before base rate ϭ 100% marginal requirements, which increases the funds available for lending. (Dres ϭ 1 rate ϭ 80%, after uniform rate ϭ 80%). when reserve requirements are reduced.) ( j ) 1975 (Aug.) Technical reserve requirements on short-term time deposits are increased (to 80%, to be fulfilled by a mandatory investment in T-bills). 1.d Directed credit ( k ) 1976 (May) CB pays interest rate on reserves. ( l ) 1977 (May–Dec.) Reserve requirements are reduced on demand This set of variables includes all the measures aimed to reduce the deposits (to 59%), on 1–3 m time deposits (to amount of preferential loans (or loans at a preferential rate) that banks are 20%), and on 3–12 m time deposits (to 8%). forced to make. (Dpr ϭ 1 when directed credit is reduced.) (m) 1978 (Jan.–Jul.) Reserve requirements are reduced on demand deposits (to 42%). ( n ) 1979 (Apr.–Dec.) Reserve requirements are reduced on demand 1.e Banks’ ownership deposits (to 21%), on 1–3 m time deposits (to ϭ 8%). Dpriv 1 when banks are privatized or government controls are ( o ) 1979 (Sep.) CB stops paying interest on reserves. reduced. ( p ) 1980 Reserve requirements are reduced on demand deposits (to 10%), on 1–3 m time deposits (to 4%), and on 3–12 m time 1.f Prudential regulation deposits (to 4%).

Typically, financial liberalization programs include a strengthening of We ignore events (h) and ( j) because they cancel out within a year and our prudential regulation and supervisory powers of the CB. This is relevant in data is annual. We have five dummies to summarize changes in reserve which it can increase the trust in the financial system and hence attract requirements. The first is ϭ 1 after 1974 (event (i)), the second is ϭ 1 more deposits. (Dreg ϭ 1 when prudential regulation measures are in between 1976 and 1979 (events (k) and (o)), the third is ϭ 1 after 1976 force.) (event (l)), the fourth is ϭ 1 after 1978 (event (m)), and the fifth is ϭ 1 after 1980 (events (n) and (p)). 2. Securities Markets 1.d Directed credit and credit ceilings These variables capture the measures aimed at deregulating and ϭ developing the securities and stock markets. (Dst 1 when markets are (q) 1974 (Jan.–Sep.) New, more-relaxed ceilings are introduced. deregulated.) (r) 1974 (Oct.–Dec.) Ceilings are completely abolished. (s) 1975 (Jan.–Jul.) Ceilings are reestablished, but banks are al- 3. International Financial Liberalization lowed to increase their loans by the increment in time deposits over the outstanding amount as Domestic financial liberalization is generally paired with international of September 1974. liberalization both in the capital and in the current account. Here we use the (t) 1976 (Mar.) The ceiling is set at the amount of outstanding information relative to the capital account and the exchange rate. (Df ϭ 1 loans as of July 1975. when capital movements and/or the exchange rate are liberalized.) (u) 1976 (Apr.) Ceilings are definitely abandoned. We list below for each country the policy events that are used to construct the dummies, and the dates on which the changes occured. We ignore event (r) because it cancels out within a year and our data is annual. We have three dummies to summarize changes in credit ceilings. CHILE The first is ϭ 1 after 1973 (event (q)), the second is ϭ 1 after 1974 (event 1. Domestic Financial Liberalization (s)), and the third is ϭ 1 after 1975 (event (u)).

1.a Interest rates. 1.e Banks’ ownership

(a) 1974 (May) Interest rates for institutions other than commercial 1975 Banks are privatized. banks, the state bank, and saving and loans are freed. 1982 Banks are under special government administration. (b) 1974 (Jun.) Controls on deposit rates are abolished. 1986 Banks are reprivatized. (c) 1975 (May) Controls on lending rates are abolished. (d) 1975 (Oct.) Controls are reimposed on both rates. The dummy for bank ownership is ϭ 1 between 1975 and 1982 and after (e) 1976 (Jan.) Controls are removed. 1986.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 260 THE REVIEW OF ECONOMICS AND STATISTICS

1.f Prudential regulation 1.e Banks’ ownership

1986 (Nov.) New banking law, includes prudential measures and The sector is dominated by State-owned institutions. There have not strengthens the supervisory system. been changes in this dimension. 1987 Deposit insurance scheme is introduced. 1.f Prudential regulation The dummy for prudential regulation is ϭ 1 after 1986. 1989 (Aug.) Enacted a banking law providing for minimum capital 3. International Financial Liberalization and prudential lending guidelines.

( v ) 1975 (Jan.) Ceilings on foreign borrowing are reduced (from The dummy for prudential regulation is ϭ 1 after 1989. 200% to 100% of capital and reserves). ( w ) 1976 (Jun.) Ceilings on foreign borrowing are increased (to 2. Securities Markets 150%). ( x ) 1979 (Jan.) Foreign borrowing is authorized for every purpose (before it was allowed only for financing loans 1986 Introduced weekly foreign exchange auction. related to foreign trade). 1990 (Nov.) Stock Exchange operations begin. ( y ) 1978 (Mar.) Ceilings on foreign borrowing are increased (to 160%). The first dummy is ϭ 1 after 1986; the second is ϭ 1 after 1990. ( z ) 1978 (Apr.) Short-term foreign borrowing by banks is forbid- den. Reserves are imposed on long-term borrowing 3. International Financial Liberalization (rates are differentiated according to maturity). Limits on foreign-currency loans are imposed (both No reforms in this area. Capital movements are still subject to controls. stock and flow). (aa) 1978 (Dec.) Ceilings on foreign borrowing are increased (to 180%). INDONESIA (bb) 1979 (Jun.) Ceilings on foreign borrowing are abolished. 1. Domestic Financial Liberalization (cc) 1980 (Jan.) Reserve requirements on foreign currency deposits are reduced. (dd) 1980 (Apr.) Limits on foreign currency loans are eliminated. 1.a Interest rates (After this, the only restrictions left on capital movements are the prohibition of short-term (2 yrs) 1983 Interest rates on loans and deposits are freed (except rates on foreign loans and reserve requirements on loans loans refinanced by CB). with maturities between 2 yrs and 5.5 yrs.) (ee) 1984 Capital movements are restricted. The dummy for interest rates is ϭ 1 after 1982. ( ff ) 1991 Restrictions are reduced again.

We ignore event (v) because it was shortly neutralized by events (w), (y), 1.b Procompetition measures and (aa). We have three dummies to summarize international financial liberalization. The first is ϭ 1 after 1978 (event (bb)). The second is ϭ 1 1988 (Oct.–Dec.) Entry of new banks is allowed. Banks that satisfy the between 1979 and 1983, and then after 1990 (events (bb), (ee), and (ff)). criteria for financial soundness are allowed to open The third is ϭ 1 between 1980 and 1983, and then after 1990 (events (cc), new branches. All banks can issue CDs and offer (dd), (ee), and (ff)). new services.

ϭ GHANA The dummy is 1 after 1988. 1. Domestic Financial Liberalization 1.c Reserve requirements 1.a Interest rates 1988 Reserve requirements are reduced from 15% to 2%. 1987 (Sep.) Decontrolled all interest rates. 1987 (Oct.) Introduce weekly auctions of T-bills. The dummy ϭ 1 after 1987.

ϭ The dummy for interest rates is 1 after 1987. 1.d Directed credit

1.b Procompetition measures 1983 The role of CB in allocating credit is reduced. The number of categories of credit for which banks would be refinanced by CB is reduced. 1993 (May) Enacted new law to foster competition among commercial 1990 Most of the liquidity credit arrangements for priority loans are banks. Also enacted Home Mortgage Finance Law to eliminated. support development of housing finance. The first dummy is ϭ 1 after 1982; the second is ϭ 1 after 1989. The dummy for competition is ϭ 1 after 1993.

1.f Prudential regulation 1.d Directed credit 1989 Prudential measures such as capital adequacy ratio are intro- 1988 (Feb.) Removed almost all credit controls (except agriculture). duced. 1990 (Nov.) Removed lending targets for the agricultural sector. 1991 Prudential measures are reinforced. 1992 New prudential measures are approved, and the supervisory The first dummy is ϭ 1 after 1987; the second is ϭ 1 after 1990. power of CB is reinforced.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 261

The first dummy is ϭ 1 after 1988; the second is ϭ 1 after 1990; the third MALAYSIA is ϭ 1 after 1991. 1. Domestic Financial Liberalization

2. Securities Markets

1977 The stock exchange opens in its present form but remains 1.a Interest rates virtually inactive until 1989. 1988 New measures to strengthen and deregulate the stock market. (a) 1971 Interest rates on long term (four or more years) deposits are liberalized. ϭ The dummy 1 after 1988. (b) 1972 Rates on deposits with maturity greater than one year are freed. 3. International Financial Liberalization (c) 1973 Rates on deposits placed with finance companies are freed. (d) 1973 Discount rates on T-bills are determined by open tender in No controls on capital movements and foreign exchange. the money market. (e) 1978 All interest rates of commercial banks are freed. KOREA (f) 1984 New controls are set on the lending rates. Specifically, the 1. Domestic Financial Liberalization base lending rate (BLR) is introduced. Lending rates offered by every bank and finance company are then anchored to 1.a Interest rates their declared BLR, determined on the basis of the cost of funds (taking into account the cost of statutory reserves, 1984 Financial intermediaries (nonbank) are given discretionary power liquid assets requirements, and overheads). in determining lending rates. (g) 1985 (Oct.) Controls on deposits rates are reintroduced. 1988 Most banks’ lending and long-term deposits rates are deregu- (h) 1987 (Jan.) Controls on deposits rates are eliminated. lated. (i) 1987 (Sep.) The CB imposes new, and more restrictive, guide- lines for the determination of the BLR. The first dummy is ϭ 1 after 1983; the second is ϭ 1 after 1987. (j) 1991 The BLR is freed from CB’s control.

1.b Procompetition measures We ignore event (a) because it is at the beginning of the sample. Seven dummies summarize interest rates liberalization. Dummies 1 to 4 are for the deposit rates, dummies 5 to 7 for the lending rates. The first is ϭ 1 after 1983 Entry barriers are lowered, and banks are allowed to introduce 1971 (b), the second is ϭ 1 after 1972 (c,d), the third is ϭ 1 after 1977 (e), new services. the fourth is ϭ 1 between 1978 and 1985 and then after 1986 (events (g) 1989 Entry barriers are lowered again. The establishment of new and (h)), the fifth is ϭ 1 between 1978 and 1983 (events (e) and (f), the financial institutions is approved. sixth is ϭ 1 between 1978 and 1983 and then after 1990 (events (e), (f), ϭ ϭ ϭ and ( j)), the seventh is 1 between 1978 and 1987 and then after 1990 The first dummy is 1 after 1982; the second is 1 after 1988. (events (e), (f), (i), and ( j)).

1.d Directed credit

The share of policy loans is quite high, after peaking at the end of the 1.d Directed credit 1970s. No significant measures taken within sample period to reduce it. 1975 Priority lending is introduced. CB controls both the quantity and 1.e Banks’ ownership the interest charged on primary borrowers. 1979 CB issues annual priority lending guidelines, still leaving Although privatized in 1981–1983, banks remained heavily controlled considerable discretion to the banks and without seriously by the State. distorting the interest rates. 1991 The number of priority sectors and the required loan amount is 1.f Prudential regulation reduced.

1991 General Banking Act introduces new prudential measures and The first dummy is ϭ 1 before 1975; the second is ϭ 1 before 1975 and imposes supervisory regulations. after 1978; the third is ϭ 1 before 1975 and after 1990.

The dummy ϭ 1 after 1990.

2. Securities Markets 1.f Prudential regulation

1992 The stock market opens for direct purchase by foreigners. 1988 The Banking and Financial Institutions Act extends and strength- ens CB’s supervisory powers. The dummy ϭ 1 after 1991. Dummy ϭ 1 after 1988. 3. International Financial Liberalization

Capital movements and the exchange rate are still heavily regulated. Significant dates are 3. International Financial Liberalization

1981 Capital movements are less controlled. 1973 Exchange rate regulations are relaxed. 1989 Foreign exchange market is established. 1987 New measures to provide investors with greater access to credit.

The first dummy is ϭ 1 after 1980; the second is ϭ 1 after 1988. The first dummy ϭ 1 after 1972; the second ϭ 1 after 1986.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 262 THE REVIEW OF ECONOMICS AND STATISTICS

MEXICO The first dummy ϭ 1 in 1981–1982 and after 1987; the second dummy ϭ 1 1. Domestic Financial Liberalization after 1984.

1.a Interest rates 1.b Procompetition measures 1988–89 Interest rates are liberalized. 1980 CDs are introduced. 1981 Barriers to entry are lowered. Dummy ϭ 1 after 1988. The first dummy ϭ 1 after 1979; the second ϭ 1 after 1980. 1.b Procompetition measures

1990 New legal framework for banks and nonbanks financial interme- 1.f Prudential regulation diaries. The new law promotes competition, allows the introduc- tion of new services, and establishes prudential measures. Also, 1986 A new banking law becomes effective. The law provides favors the development of nonbank financial institutions. supervisory and prudential measures. A Bank Supervision Unit is created within the Central Bank. Dummy ϭ 1 after 1989. Dummy ϭ 1 after 1985. 1.c Reserve requirements 2. Securities Markets 1989 Reserve requirements are reduced. 1983 The Capital Market Board is established. CMB promotes and Dummy ϭ 1 after 1988. monitors developments in the securities markets. 1986 The Istanbul Stock Exchange becomes operative. 1.d Directed credit The first dummy ϭ 1 after 1982; the second ϭ 1 after 1985. 1988 Elimination of forced lending. 1991 Elimination of the ‘‘liquidity coefficient,’’ requiring that 30% of 3. International Financial Liberalization deposits be invested in T-bills. 1992 Elimination of regulations requiring that banks hold long-term 1984 Foreign exchange deregulation: Residents are allowed to hold government bonds until maturity. foreign-currency denominated deposits (FCCDs). Banks are allowed to keep foreign currency abroad and are given some The first dummy ϭ 1 after 1987; the second ϭ 1 after 1990; the third ϭ 1 discretionary power in determining the exchange rate. after 1991. 1985 New restrictions on foreign exchange are introduced. 1988 Foreign exchange is liberalized. 1989 Capital movements are liberalized. 1.e Banks’ ownership 1990 Exchange rate is liberalized.

1982 Banks are nationalized. Credit to private sector falls sharply. The first dummy is ϭ 1 after 1983; the second is ϭ 1 in 1983 and after 1992 Banks are privatized. 1987; the third is ϭ 1 after 1987; the fourth is ϭ 1 after 1988; the fifth is ϭ 1 after 1989. The dummy is ϭ 1 before 1982 and after 1991.

2. Securities Markets ZIMBABWE 1. Domestic Financial Liberalization 1988–1992 During this period, measures have been taken to deregu- late the securities market and promote its development. 1.a Interest rates Despite recent growth, the securities market is still under- developed. 1990 Restrictions on all interest rates are eliminated. 3. International Financial Liberalization Dummy ϭ 1 after 1990. 1989 Restrictions on foreign direct investment are removed. 1.b Procompetition measures Dummy ϭ 1 after 1988. No special measures have been taken in this field and banks have only TURKEY very recently started to offer new services. Although the number of 1. Domestic Financial Liberalization financial institutions and the range of services offered are impressive by African standards, competition is scarce. Some new institutions have entered the financial market, but this has not changed the status quo. 1.a Interest rates

1981 Interest-rate ceilings are abolished (except on sight deposits and 1.c Reserve requirements on preferential lending). 1983 Ceilings are reintroduced. 1985 Government securities are auctioned. Their yields are market 1991 Reserve requirements are reduced. determined. 1988 Ceilings are eliminated. Dummy ϭ 1 after 1990.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021 DOES FINANCIAL REFORM RAISE OR REDUCE SAVING? 263

2. Securities Markets (3) For Ghana—savings of the consolidated central govern- ment computed as budget surplus plus public investment— 1973 Stock exchange is created. Source: Ghana-Quarterly Digest of Statistics. 1993 Stock exchange is opened to foreign investors. (4) For Zimbabwe—savings of the consolidated central gov- ernment computed as budget surplus plus public invest- ϩ The first dummy ϭ 1 after 1972; the second ϭ 1 after 1992. ment—Source: World Bank National Accounts Easterly database. ϩ 3. International Liberalization GNDI: GNP External Transfers—Source: WB ‘‘World Sav- ings Database’’ Rev. 3.00. defl)t ء yt ϭ log of real per capita income ϭ ln (GNDI/population 1994 The current account and the foreign exchange are liberalized. population: Source WB BESD database. ␲ ϭ inflation rate ϭ⌬ln (defl ϩ ). ϭ t t 1 Dummy 1 after 1993. defl: implicit consumption price deflator—year average—Source: WB ‘‘World Savings Database’’ Rev. 3.00. rt ϭ real interest rate ϭ APPENDIX B ϭ ϩ a Ϫ⌬ (1) For IDN, KOR, MYS rt ln (1 it ln (defltϩ1). ϭ ϩ d ϩ (2) For CHL, GHA, MEX, TUR, ZWE rt 0.5(ln (1 it ) ϩ d Ϫ⌬ Variables Definitions and Data Sources 0.5(ln (1 itϪ1)) ln (defltϩ1)). ia ϭ nominal interest rate ϭ short-term deposit rate, year average— ϭ ϭ (s/y)t private saving rate (private savings/GNDI)t Source: Central Banks Bulletins. private savings ϭ gross national savings—public sector sav- id ϭ nominal interest rate ϭ short-term deposit rate, December value— ings Source: Central Banks Bulletins. gross national savings, Source: WB ‘‘World Savings Data- index ϭ index of financial liberalization—Source: our calculations base’’ Rev. 3.00. t ϭ i-mic, i-mac; i-rsk, i-liq; subindices—Source: our calculations public sector savings govs ϭ public-sector saving rate (relative to GNDI)—Source: WB ‘‘World (1) For CHL, KOR, MEX, MYS, TUR—savings of the t ϭ Savings Database’’ Rev. 3.00. nonfinancial public sector ( consolidated central a ϭ government ϩ state and local governments ϩ nonfinancial (s/y)t private savings ratio adjusted for domestic capital gains—Source: public enterprises) computed as revenues minus consump- WB ‘‘World Savings Database’’ Rev. 3.00. a ϭ tion—Source: WB ‘‘World Savings Database’’ Rev. 3.00. yt GNDI adjusted for domestic capital gains—Source: WB ‘‘World (2) For IDN savings of the consolidated central government Savings Database’’ Rev. 3.00 ϩ our calculations a ϭ computed as revenues minus consumption—Source: WB govst public-sector saving rate adjusted for domestic capital gains— ‘‘World Savings Database’’ Rev. 3.00. Source: WB ‘‘World Savings Database’’ Rev. 3.00.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/003465300558768 by guest on 28 September 2021