21716 THE WORLD BANK ECONOMIC REVIEW Volume 14 September 2000 Number 3 A SYMPOSIUM ON SAVING IN DEvELOPING COUNTRIES Saving in Developing Countries: An Overview Norman Loayza, Klaus Schmidt-Hebbel, and Luis Serven Can Africa's Saving Collapse Be Reversed? Ibrahim A. Elbadawi and Francis M. Mwega The Saving Collapse during the Transition in Eastern Europe Cevdet Denizer and Holger C. Wolf What Drives Consumption Booms? Peter J. Montiel Saving Transitions Dani Rodrik Personal and Corporate Saving in South Africa Janine Aron and John Muellbauer Household Saving in China Aart Kraay Private Saving in India Norman Loayza and Rashmi Shankar A NEW DEVELOPMENT DATABASE A New Database on the Structure and Development of the Financial Sector Thorsten Beck, Aslh Demirgui-Kunt, and Ross Levine ISSN 0258-6770 THE WORLD BANK ECONOMIC REVIEW EDITOR Francois Bourguignon CONSULTING EDITOR Ilyse Zable EI)ITORIAL BOARD Kaushik Basu, Cornell University and Universitr of Delhi Stijn Claessens Carmen Reinhart, University of Maryland David Dollar Mark R. Rosenzweig, University of Pennsylvania Gregory K. Ingram L. Alan Winters, University of Sussex Martin Ravallion The World Bank Economic Review is a professional journal for the dissemination of World Bank- sponsored research that informs policy analyses and choices. It is directed to an international readership among economists and social scientists in government, business, and international agencies, as well as in universities and development research institutions. The Review emphasizes policy relevance and opera- tional aspects of economics, rather than primarily theoretical and methodological issues. It is intended for readers familiar with economic theorv and analysis but not necessarily proficient in advanced mathemati- cal or econometric techniques. Articles will illustrate how professional research can shed light on policy choices. Inconsistency with Bank policy will not be grounds for rejection of an article. Articles will be drawn primarily from work conducted bv World Bank staff and consultants. Before being accepted for publication by the Editorial Board, all articles are reviewed by two referees wvho are not members of the Bank's staff and one World Bank staff member; articles must also be recommended by at least one external member of the Editorial Board. The Review may on occasion publish articles on specified topics by non-Bank contributors. Any reader interested in preparing such an article is invited to submit a proposal of not more than two pages in length to the Editor. The views and interpretations expressed in this journal are those of the authors and do not necessarily represent the views and policies of the World Bank or of its Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any consequences of their use. 'When maps are used, the bound- aries, denominations, and other information do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. Comments or brief notes responding to Review articles are welcome and will be considered for publica- tion to the extent that space permits. Please direct all editorial correspondence to the Editor, The World Bank Economic Review, The World Bank, Washington, D.C. 20433, U.S.A. The World Bank Economic Review is published three times a year (January, May, and September) by the World Bank. Single copies may be purchased at $12.95. Subscription rates are as follows: Individuals Institutions 1-year subscription US$30 US$50 2-year subscription US$55 US$95 3-year subscription US$70 US$130 Orders should be sent to: World Bank Publications, Box 7247-7956, Philadelphia, PA 19170-7956, U.S.A. Subscriptions are available without charge to readers with mailing addresses in developing countries and in socialist economies in transition. Written request is required every two years to renew such subscriptions. (© 2000 The International Bank for Reconstruction and Development / TliE WORLD BANK 1818 H Street, N.W., Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America ISBN 0-8213-4609-1; ISSN 0258-6770 Material in this journal is copyrighted. Requests to reproduce portions of it should be sent to the Office of the Publisher at the address in the copyright notice above. The World Bank encourages dissemination of its xvork and will normally give permission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to make photocopies is granted through the Copyright Clearance Center, Suite 910, 222 Rosewood Drive, Danvers, MA 01923, U.S.A. This journal is indexed regularly in Current Contents/Social & Behavioral Sciences, Index to Interna- tional Statistics, Journal of Economic Literature, Public Affairs Information Service, and Social Sciences Citation Index@. It is available in microform through University Microfilms, Inc., 300 North Zeeb Road, Ann Arbor, Ml 48106, U.S.A. THE WORLD BANK ECONOMIC REVIEW Volume 14 September 2000 Number 3 A SYMPOSIUM ON SAVING IN DEVELOPING COUNTRIES Saving in Developing Countries: An Overview 393 Norman Loayza, Klaus Schmidt-Hebbel, and Luis Serven Can Africa's Saving Collapse Be Reversed? 415 Ibrahim A. Elbadawi and Francis M. Mwega The Saving Collapse during the Transition in Eastern Europe 445 Cevdet Denizer and Holger C. Wolf What Drives Consumption Booms? 457 Peter J. Montiel Saving Transitions 481 Dani Rodrik Personal and Corporate Saving in South Africa 509 Janine Aron and John Muellbauer Household Saving in China 545 Aart Kraay Private Saving in India 571 Norman Loayza and Rashmi Shankar A NEW DEVELOPMENT DATABASE A New Database on the Structure and Development of the Financial Sector 597 Thorsten Beck, Aslt Demirguic-Kunt, and Ross Levine THE WORIL) BANK ECONOMIC REVIEW. VOL. 14, NO0. 3: 393-414 Saving in Developing Countries: An Overview Norman Loayza, Klaus Schmidt-Hebbel, and Luis Serv6n This article reviews the current state of knzowledge on the determinants of saving rates, presenting the main findings and contributions of the recently completed World Bank research project, "Saving Across the World." The article discusses the basic design of the research project and its core database, the World Saving Database. It then summa- rizes the main project resuilts and places them in the context of the literature on saving, identifying the key policy and nonpolicy determinants of private saving rates. Special attention is paid to the relationship between growth and saving and the impact of spe- cific policies on saving rates. The article concludes by introducing the studies included in this special issue. Saving rates around the world vary widely: on average East Asia saves more than 30 percent of gross national disposable income (GNDI), while Sub-Saharan Africa saves less than 15 percent. Regional differences have been rising: over the past three decades saving rates have doubled in East Asia and stagnated in Sub- Saharan Africa and in Latin America and the Caribbean (figure 1). Should these disparities make saving a policy concern? In theory there is little reason why countries facing different income streams, preferences, or demograph- ics, and subject to different types of shocks, should choose similar saving rates. In practice, however, the intertemporal choices that underlie saving are subject to a host of externalities, market failures, and policy-induced distortions that are likely to drive saving away from socially desirable levels. Some market imperfections- such as the unavailability of risk-sharing instruments, overly stringent manda- tory saving schemes, or outright Soviet-style rationing-can lead to socially ex- cessive saving. Others-such as too little government saving or the negative effect on retirement saving of an anticipated public bailout of the poor in old age-can result in too little national saving. Across countries higher saving rates tend to go hand in hand with higher in- come growth-a fact that has been taken as proof of the existence of both virtu- ous cycles of saving and prosperity and poverty traps of insufficient saving and stagnation. If virtuous cycles can be jumpstarted by a hike in aggregate saving, Norman Loayza is with the Central Bank of Chile and the Development Research Group at the World Bank, Klaus Schmidt-Hebbel is with the Central Bank of Chile, and Luis Serven is with the Chief Econo- mist Office of the Latin America and the Caribbean Region at the World Bank. Their e-mail addresses are n1oayza@condor.bcentral.ci, kscbmidt@condor. bcentral.ci, and lserven@worldbank.org. The authors grate- fully acknowledge outstanding research assistance provided by George Monokroussos. They also are grateful to the editor for helpful discussions. C) 2000 The International Bank for Reconstruction and Development/ THE WORLD BANK 393 394 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Figure 1. Median Gross NVational Saving Rates by Region, 1965-94 Percent 40- 35 - 30 -* 1965-73 _ 1974-84 25 -El 1985-94 20- 15- 10 : 5- 0 China East MNiddle Industrial South Sub- Latin Asia East and countries Asia Saharan America and and the North Africa the Caribbean Pacific Africa ,Vote.: Gross national saving rates, including net current transfers, are given as a percentage of gross national disposable income. Source: World Saving Database. then the social value of saving would exceed its private value in many developing countries, particularly poorer countries. The social value of saving could also exceed its private value because of imper- fections in world financial markets: a national saving rate broadly in line with the economy' s investment rate reduces vulnerability to sudden shifts in interna- tional capital flows driven by uncontrollable forces, such as herd behavior or self-fulfilling investor expectations. As the recent turmoil in international finan- cial markets illustrates, low saving and high current account deficits can exacer- bate the likelihood, and the adverse effects, of capital flow reversals. However, the East Asian experience of 1997-98 demonstrates that high saving alone cannot fully insure against the consequences of weak financial systems or unsustainable exchange rate policies. Although a large literature has shed light on some aspects of consumption and saving behavior many empirical puzzles and policy-relevant questions remain. The recently completed World Bank research project, "Saving Across the World," addressed many of them. This article reviews empirical facts on saving behavior in the world, summarizes the main output and results of the World Bank's project, and introduces the articles included in this special issue. Loayza, Scbrzidt-Hebbel, and Serven 395 I. THE WORLD BANK'S SAVING RESEARCH PROJECT The World Bank's research project "Saving Across the World" was motivated largely by behavioral puzzles and policy questions that are at the core of saving experiences and policy discussions in developing, transition, and industrial econo- mies. The project was organized around three broad questions: * Why do saving rates differ so widely across countries and time periods? * What is behind the relationship between saving and growth, and which way does the causal link run? * Which policies have the greatest impact on national saving, and which are unlikely to work? The project addressed these questions by commissioning a set of articles from leading scholars. Most of the studies are empirical, tackling saving issues from the frontiers of consumption theory and econometric methods.1 The studies fall into three broad categories. A first group examines cross-country evidence, focusing on the behavioral and policy determinants of saving. Loayza, Schmidt-Hebbel, and Serven (2000) examine the most important determinants of saving proposed in the literature, and Attanasio, Picci, and Scorcu (2000) focus on the dynamic relationship between national saving, investment, and growth. Deaton and Paxson (2000) also examine the connection between in- come growth and saving, but do so from a microeconomic perspective, making use of household saving data from Indonesia, Taiwan (China), and Thailand. Finally, Deaton and Laroque (1998) reexamine the theoretical relationship be- tween saving and growth, assessing whether the presence of a limited amount of residential land can catalyze a virtuous circle of saving, growth, and rising real estate prices. The second set of articles assesses specific saving-oriented policies, using meth- odologies that range from estimation of parsimonious theoretical models (L6pez, Schmidt-Hebbel, and Serv6n 2000) to reduced-form empirical estimation (Bandiera and others 2000 and Samwick 2000). They assess the impact of saving on do- mestic financial liberalization (Bandiera and others 2000), pension reform (Samwick 2000), tax incentives (Besley and Meghir 1998), and public saving (L6pez, Schmidt-Hebbel, and Serven 2000). The remaining articles focus on specific geographic regions, countries, and country groups selected because many features of their saving experiences are relevant to policy. The articles included in this issue are drawn from this third set.2 1. All are available at the project's website, bttp://wwviw.worldbank.orglresearch/projects/savings/ policies.htm. 2. Other country studies in this third group, but not included in this issue, are Burnside (1998) and Burnside, Schmidr-Hebbel, and Serven (1999) on Mexico and L6pez-Mejia and Ortega (1998) on Colombia. 396 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Empirical studies of consumption and saving in both industrial and develop- ing countries are often hampered by inadequate aggregate data.' Thus in order to address the empirical questions posed in the project, considerable effort was de- voted to constructing a large cross-country and time-series database on saving, consumption, income (at various levels of aggregation), and their major determi- nants, satisfying basic requirements of data quality and consistency. The result- ing World Saving Database, which represents one of the project's major outputs, is described in detail in Loayza and others (1998b). The database and its under- lying documentation are publicly available at the project website. This new database features several improvements over existing, publicly avail- able data sets. First, its broad coverage makes it the largest and most systematic collection of annual time series on country saving rates and saving-related vari- ables, spanning a maximum of 35 years (1960 to 1994) and 112 developing and 22 industrial countries. To illustrate its size, there are, for example, 3,464 country- year observations for the gross national saving rate.4 Second, the database corrects inconsistencies in country time series that are pervasive in existing databases. Apart from checking for accounting consistency in the data, statistical testing for the presence of outliers was also conducted. Third, the World Saving Database unifies definitions regarding the coverage of the public sector, by including separate public saving measures for the consoli- dated central government and the general government or nonfinancial public sector. Fourth, the database contains series on private and public saving both with and without adjustments for capital gains and losses from inflation and real exchange rate devaluation. Fifth, for a limited number of economies the database disaggregates private saving and investment between households and firms. And sixth, the new database includes cross-country and time-series information on determinants of saving, including national and private income measures, proxies for financial depth, interest rates, inflation and other uncertainty-related vari- ables, and demographic variables like urbanization and age dependency.5 II. MAIN FINDINGS ON THE BEHAVIOR OF PRIVATE SAVING Before addressing the main determinants of private saving, we first trace the major trends in national saving rates across regions.6 The world's median gross 3. The criticisms stem partly from conceptual and empirical shortcomings of existing aggregate data and partly from inadequate use of the data in applied research. See Schmidt-Hebbel and Serven (1999b) for a full discussion. 4. Construction of the database involved making consistent use of existing data sources, including the World Bank's World Development Indicators, the International Monetary Fund's International Financial Statistics and Government Financial Statistics, the United Nation's National Income and Product Ac- counts, and data sets from the Organisation for Economic Co-operation and Development (OECD), the Asian Development Bank, and the Inter-American Development Bank. This information was supplemented and made consistent with data gathered from about 1,500 "Recent Economic Development" reports of the International Monetary Fund and about 500 World Bank reports and country government reports. 5. The importance of using appropriate measures of saving can be illustrated by the results of various incremental adjustments applied to Indian saving data (Loayza and Shankar, this issue). 6. A more detailed discussion of these trends and other saving patterns and correlations is presented in Loayza and others (1998a). Loayza, Schmnidt-Hebbel, and Serven 397 national saving rate has declined over the past three decades. It fell from 21 percent in 1965-73 to 20 percent in 1974-84 and further to 19 percent in 1985- 94. The median gross national saving rate in industrial countries increased gradu- ally from 25 percent in the early 1960s to a historical peak of almost 28 percent in 1972-73, just before the first oil shock. Since then, it has declined persistently, reaching 19 percent in 1993-94 (figure 1). The median gross national saving rate in developing countries rose from 17 percent in 1965-73 to 19 percent in 1974-84, falling subsequently to 18 percent in 1985-94. However, this aggregate figure conceals wide divergences in saving patterns within the developing world. Saving rates have risen rapidly in China and most other countries in East Asia. China's already high saving rate (averag- ing 29 percent in 1970-77) rose further during the period of economic reform that began in 1978, reaching 41 percent in 1993-94. The median national saving rate in the East Asia and the Pacific region rose spectacularly from 20 percent in 1966-68 to 33 percent in 1992-94. South Asia's median saving rate also rose substantially over the past three decades. By contrast, saving rates in other developing countries and regions stag- nated or declined. In Latin America and the Caribbean the median national sav- ing rate rose after the first oil shock (1973-80) and then fell after the debt crisis. A similar pattern of rise and fall is observed in the Middle East and North Africa, largely mirroring the path of world oil prices. Sub-Saharan Africa's median sav- ing rate declined from an already low 13 percent in 1965-73 to just over 12 percent in 1974-84, returning to 13 percent in 1985-94. Notwithstanding this recovery, Africa's saving rate continues to be the lowest across all regions. We now turn to the analysis of the main determinants of private saving rates, comparing their expected signs according to consumption theory and their ac- tual signs derived in seven empirical studies of private saving rates in cross- country time-series (panel) samples (table 1).7 The empirical studies cover both industrial and developing countries (Masson, Bayoumi, and Samiei 1995; Edwards 1996; Bailliu and Reisen 1998; and Loayza, Schmidt-Hebbel and Serven 2000), industrial countries alone (Pesaran, Haque, and Sharma 2000), and developing countries alone (Corbo and Schmidt-Hebbel 1991 and Dayal-Ghulati and Thi- mann 1997). The common feature of these articles is that they are based on reduced-form saving equations, not derived from first principles. They differ in that they use different samples, model specifications, and estimation techniques. Still, many of the estimated coefficients are consistently significant across different studies or are consistent with signs predicted by theory. Variables whose signs are consis- tent across studies and are statistically significant include the terms of trade, foreign borrowing constraints, fiscal policy variables, and pension system vari- ables. Regarding the signs of other determinants, on which consumption theories 7. A detailed discussion of the expected signs of saving determinants in table I and how they relate to specific consumption theories is provided in Loavza, Schmidt-Hebbel, and Serven (2000). Further reviews of consumption hypotheses and their relation to empirical findings can be found in Schmidt-Hebbel and Serven (1997, 1999a). 398 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 1. Determinants of the Ratio of Private Saving to Income in Panel Studies Specific Sign predicted Empirical Variable category variable by theory findings Income Income level Actual 0 or + + (1, 2, 3, 4, 7) 0(5, 6) Temporary/permanent + / 0 or + 0 / 0 (7) Terms of trade Actual 0 or + + (2, 4, 6, 7) Temporary/permanent + / 0 or + + / + (7) Growth rate: actual Ambiguous + (2, 3, 7) 0 (4, 5, 6) Rates of return Real interest rate Ambiguous -(7) 0 (1, 3, 5, 6) + (2) Uncertainty Variance of innovations to saving determinants + Inflation or other measures of + macroeconomic instability Measures of political instability + -(4) 0 (1, 2, 3, 6), + (7) Domestic borrowing constraints Private credit flows - + (3) - (7) Broad money flows Income Foreign borrowing constraints Foreign lending Current account deficit - - (1, 2, 3, 7) Financial depth Private or domestic credit stocks Ambiguous - (5) Money stocks Ambiguous + (1, 3, 4) 0 (7) Fiscal policy Public saving - - (1, 3, 7) Public surplus - - (2, 5, 6) 0 (4) Public consumption Ambiguous - (2, 6) Pension system Pay-as-you-go pension transfers 0 or - - (3, 4, 5) Mandatory fully funded pension contributions 0 or + + (4) Fully funded pension assets Ambiguous 0 / + (5) Demographics Old- and/or young-age dependency - - (2, 3, 4, 7) 0 (5, 6) Urbanization Ambiguous - (3, 7) Distribution of income and wealth Income concentration Ambiguous 0 (3) Wealth concentration Ambiguous Capital income share + Note: The qualitative results listed in the last column summarize significant signs of saving regressors in the following studies: 1. Corbo and Schmidt-Hebbel (1991: table 4) 2. Masson, Bayoumi, and Samiei (1995: table 2, "restricted model" column) 3. Edwards (1996: table 2, col. 5) 4. Dayal-Gulati and Thimann (1997: table 4, col. 2) 5. Bailliu and Reisen (1998; table 1, cols. 3 and 4) 6. Pesaran, Haque,and Sharma (2000: table 6, cols. 4 and 5) 7. Loayza, Schmidt-Hebbel, and Serven (2000; table 4, col. 3 and table 7, various columns). Significant coefficient signs are identified by a plus or a minus. Results identified by a zero mean either an insignificant coefficient in the corresponding column of the original study or, when the variable is omitted from the particular specification reported in the column, a significant or insignificant variable in a different column of the same table. A zero in the third column means that theory predicts no effect. Loayza, Schmidt-Hebbel, and Seruen 399 either differ or give ambiguous predictions, such as income growth and the real interest rate, the empirical studies give conflicting results. They also differ in the significance levels of some variables for which theories agree on expected signs: income, inflation, and age-dependency ratios. Keeping these results in mind, we turn to a brief discussion of the literature's findings on saving behavior, relying mainly on the most recent and comprehen- sive of the seven studies in the table, Loayza, Schmidt-Hebbel, and Serven (2000), and the other articles of the World Bank's saving research project. The review starts by identifying nonpolicy saving determinants and subsequently discusses the influence of specific policy variables on private saving. What Drives Private Saving Rates? We begin the review by identifying nonpolicy determinants of saving. These include persistence, income, growth, demographics, and uncertainty. PERSISTENCE. Private saving rates show inertia; that is, they are highly serially correlated even after controlling for other relevant factors. The effects of a change in any determinant of saving thus are fully realized only after a number of years, with long-run responses estimated to be about twice as large as short-run (within a year) effects (Loayza, Schmidt-Hebbel, and Serven 2000).8 INCOME. Several multivariate cross-country studies of saving find that the level of real per capita income positively affects saving rates (see, for example, Collins 1991; Schmidt-Hebbel, Webb, and Corsetti 1992; Carroll and Weil 1994; Edwards 1995; and Schmidt-Hebbel and Serven 2000). Six of the seven panel studies re- ported in table 1 show similar effects for private saving rates. The influence of income typically is greater in developing than in industrial countries, tapering off at medium or high income levels. In developing countries a doubling of income per capita is estimated, other things being equal, to raise the long-run private saving rate by 10 percentage points of disposable income (Loayza, Schmidt-Hebbel, and Serven 2000). Of course, other things are never equal in practice: development also changes demographics and rates of urbaniza- tion, which may reduce saving. Thus the long-term effect of income on saving may be more modest than this figure indicates. Nevertheless, the overall implica- tion is that policies that spur development are an indirect but effective way to raise private saving.9 8. A related but different form of persistence is that which affects consumption levels. One way to explain consumption inertia observed in the data-that is, the finding that future consumption levels are partly predictable-is by introducing consumption habits. They imply that consumer utility in any given period depends on both consumption in that period and a stock of consumption habits. One form is external habits (Abel 1990 and Campbell and Cochrane 1994), in which utility depends positively on the difference between an individual's consumption and (possibly lagged) average per capita consumption levels. An alternative specification is internal habits (Ferson and Constantinides 1991) in which utility depends on the difference between an individual's current and lagged consumption levels. 9. These results are also consistent with the view that the ability to save rises sharply only after income exceeds subsistence consumption levels, as implied by the Stone-Geary specification of consumer prefer- 400 THE WORLD BANK ECONOMIC REVIEW, VOI.. 14, NO. 3 Income inequality is another potentially important determinant of saving. It played a prominent role in post-Keynesian models of saving and growth (Lewis 1954, Kaldor 1957, and Pasinetti 1962), which focus on the functional distribu- tion of income (that is, the distribution of income among classes of consumers, such as workers and capitalists). However, most of the recent theoretical work and the bulk of related empirical studies focus on the personal distribution of income (that is, the distribution based solely on income criteria). Given the links between income inequality and saving, income concentration is expected to have a positive effect on household saving, but a negative effect on corporate and public saving, resulting in an ambiguous effect on aggregate saving (for a discus- sion see Schmidt-Hebbel and Serven 2000). Edwards (1995) and Schmidt-Hebbel and Serven (2000) find that personal income concentration has no significant effect on the private and national saving rates, respectively. Both the permanent-income hypothesis (Friedman 1957) and the life-cycle hypothesis (Modigliani and Brumberg 1954) distinguish between the consump- tion (and saving) effects of changes in permanent and temporary income, whether measured by fluctuations in private disposable income or movements in the terms of trade, in studies using aggregate data. In its simple and extreme form- permanent-income shocks should be entirely consumed, whereas temporary- income shocks should be entirely saved-the permanent-income hypothesis is typically rejected by the evidence. However, the evidence also shows that the positive impact on saving of a temporary increase in real per capita income is greater than that of a permanent rise in income (Loayza, Schmidt-Hebbel, and Serven 2000). GROWTH. The simple permanent-income theory predicts that higher growth (that is, higher future income) reduces current saving. But in the life-cycle model growth has an ambiguous effect on saving, depending on which cohorts benefit the most from income growth, how steep their earning profiles are, and the ex- tent to which borrowing constraints apply (Deaton 1992). Reverse causation from saving to growth also is possible, taking place through capital accumulation. A strong positive association between saving ratios and real per capita growth has been documented amply in cross-country empirical studies (see, for example, Modigliani 1970, Maddison 1992, Bosworth 1993, and Carroll and Weil 1994). Half of the panel studies included in table 1 confirm the positive relationship. How- ever, its structural interpretation is controversial, as it has been viewed both as proof that growth drives saving (for example, Modigliani 1970 and Carroll and Weil 1994) and that saving drives growth through the saving-investment link (for ences, which characterizes utility as a positive function of the difference between current consumption and an exogenously given subsistence level below which no saving takes place. Variants of this model specify the intertemporal elasticity of consumption as an increasing function of wealth (Atkeson and Ogaki 199 1) or of the distance between permanent income and subsistence consumption (Ogaki, Ostry, and Reinhart 1996). These studies provide household and aggregate evidence in support of this view for both industrial and developing countries. Loayza, Schmidt-Hebbel, and Servc)n 401 example, Levine and Renelt 1992 and Mankiw, Romer, and Weil 1992). Recog- nizing the importance of controlling for the joint endogeneity of income growth and saving, Loayza, Schmidt-Hebbel, and Serven (2000) use a panel instrumental- variable approach to estimate the effect of income growth on saving. They find that a 1 percentage-point rise in the growth rate increases the private saving rate by a similar amount, although this effect may be partly transitory. Three other studies in the World Bank's saving project revisit the correlation between saving and growth. Attanasio, Picci, and Scorcu (2000) examine the dynamic relationship between economic growth, the investment rate, and the saving rate using annual time series for a large cross section of countries. Em- ploying a variety of samples and econometric techniques, they consistently find that growth Granger-causes saving, although the effect appears to be quantita- tively weak. They also find that increases in saving rates do not always precede increases in growth. Moreover, there seems to be a negative relationship between lagged saving rates and current income growth (a "saving-for-a-rainy-day" ef- fect) when additional controls (such as dependency rates) are included in the regression specification. Deaton and Paxon (2000) reassess the association be- tween saving and growth using household data and find that the observed corre- lation between both variables can be explained largely as the effect of income growth on saving if individual household members determine their consumption plans on the basis of their respective lifetime income profiles. Finally, Rodrik (this issue) examines both long-lasting and short-lived epi- sodes of saving takeoffs, showing that sustained increases in saving typically are followed by accelerations in growth that persist for several years, but eventually disappear. In contrast, sustained accelerations in growth are associated with per- manent saving hikes. We return to this issue below. DEMOGRAPHICS. The cornerstone of the life-cycle hypothesis is age-related con- sumer heterogeneity and the prediction that saving follows a hump-shaped pat- tern (that is, high at middle age and low at young and old ages). Research has shown that this hypothesis is not problem-free when it comes to interpreting actual saving behavior. Life-cycle saving is not sufficient to account for the high level of aggregate wealth in industrial economies (Kotlikoff and Summers 1981). Changes in growth do not cause the cohort-specific differences in saving levels (Bosworth, Burtless, and Sabelhaus 1991) or in intertemporal consumption pat- terns (Carroll and Summers 1989 and Deaton 1991). Elderly people save or at least do not dissave as much as predicted by the life-cycle hypothesis (Deaton and Paxson 1994 and Poterba 1995), and consumers appear to value bequests (Menchik 1983). Yet microeconomic and macroeconomic evidence, both at the international and single-country level, confirms that a rise in the young-age and old-age depen- dency ratios tends to lower private saving rates-a result in line with the predic- tiorns of the life-cycle theory. Panel evidence indicates that a rise in the young-age dependency ratio by, say, 3.5 percentage points leads to a decline in the private 402 THE WORLD BANK ECONOMIC REVIEW, VOL. 14. NO. 3 saving rate of about 1 percentage point; the negative impact on saving of an increase in the old-age dependency ratio is more than twice as large (Loayza, Schmidt-Hebbel, and Serven 2000). An implication of these results is that devel- oping countries with young populations that want to accelerate their demographic transition-like China-and speed up the decline in young-age dependency may experience a transitory increase in their saving ratios. This increase will continue until the country reaches the next stage of demographic maturity, at which old- age dependency rises swiftly and saving rates level off. Another demographic force that typically affects private saving rates is the degree of urbanization. Its effect on saving has been found to be negative empiri- cally, a result that has been explained along the lines of the precautionary saving motive. UNCERTAINTY. Theory predicts that greater uncertainty should raise saving since risk-averse consumers set resources aside as a precaution against possible adverse changes in income and other factors (Skinner 1988 and Zeldes 1989). Uncertainty helps to explain why consumption follows income so closely (con- tradicting the simple permanent-income hypothesis) in the case of young con- sumers who expect positive but uncertain future income growth: their risk aver- sion is at war with their impatience (Carroll 1991). It also explains why the retired save a positive amount or dissave little, as they face much uncertainty regarding the length of their life and health costs. Direct empirical tests of the precautionary saving motive have been hampered by the difficulty of obtaining estimable closed-form solutions to models with this motive. However, some em- pirical estimates suggest that precautionary saving may account for a substantial fraction of household wealth (Carroll and Samwick 1995a). In the empirical literature on saving and growth the most popular proxy for (macroeconomic) uncertainty is inflation. However, only one of the six panel studies that include inflation among the explanatory variables finds a positive and significant effect on the private saving rate (Loayza, Schmidt-Hebbel, and Serv6n 2000). Another variable related to uncertainty is the rate of urbanization, which is expected to have a negative impact on saving. Rural incomes are more uncertain than urban incomes and, in the absence of financial markets through which risks can be diversified, rural residents would save a greater fraction of their income. Edwards (1996) and Loayza, Schmidt-Hebbel, and Serven (2000) provide supporting evidence for this view. Which Policies Affect Private Saving and Why? In addition to the factors mentioned above, economic policies may also affect saving directly and indirectly. These include fiscal policies, pension reform, fi- nancial liberalization, and external borrowing and foreign aid. FISCAL POLICY. Extending the permanent-income hypothesis, the Ricardian equivalence hypothesis combines consumers' and the government's intertemporal Loayza, Scbnmidt-Hebbel, and Sert4n 403 budget constraints and derives permanent income as net of the discounted value of government spending (Barro 1974). Its implication is that, as long as a number of restrictive conditions hold, a permanent rise in government saving will be fully offset by a corresponding reduction in private saving, leaving national saving unchanged. Most international empirical evidence rejects full Ricardian equivalence, find- ing that the offset is only partial. Six of the seven studies included in table 1 show that public saving or deficits have a negative effect on private saving. However, the estimated contemporaneous offset coefficients are significantly smaller than 1, ranging from 0.23 to 0.65. In the one study that distinguishes between short- and long-term effects, the contemporaneous offset coefficient is only 0.29 but rises to 0.69 in the long term (Loayza, Schmidt-Hebbel, and Serven 2000). L6pez, Schmidt-Hebbel, and Serven (2000) provide evidence that borrowing constraints, rather than finite horizons, lie behind the rejection of full Ricardian equivalence. This conclusion is based on the empirical estimation of a model derived from first principles that aggregates consumption plans of heterogeneous agents. Hence public sector saving seems to be one of the most direct and effective tools available to policymakers targeting national saving. However, its estimated effectiveness varies considerably-not only between the short and long term but also across countries. Offset coefficients vary from less than 30 percent in India (Loayza and Shankar, this issue) to almost 80 percent in Mexico (Burnside 1998). Regarding the composition of public saving, the international evidence shows that cutting expenditures is a more effective way to increase national saving than raising taxes (Corbo and Schmidt-Hebbel 1991; Edwards 1996; and L6pez, Schmidt-Hebbel, and Serven 2000). The evidence on the effectiveness of tax incentives granted to private savers- typically on specific financial instruments-in raising national saving is mixed and, overall, not promising. The elasticity of private saving to net rates of return is ambiguous on theoretical grounds, because of offsetting substitution, income, and human-capital effects. The empirical evidence on interest-rate elasticities of saving reflects the theoretical ambiguity: empirical estimates typically are small and not significantly different from zero. Four of the seven studies reported in table 1 show that the effect of interest rates on private saving is not significantly different from zero; one study (Masson, Bayoumi, and Samiei 1995) reports a positive effect, and one study (Loayza, Schmidt-Hebbel, and Serven 2000) re- ports a negative effect. More direct evidence from industrial countries on the effectiveness of tax incentives for voluntary retirement saving instruments is equally mixed. If the tax incentives have a positive effect on saving, the effect is generally found to be small, particularly when the negative effects of tax incentives on public saving are taken into account (Besley and Meghir 1998). PENSION REFORM. Some countries, especially countries in Latin America and Europe, are replacing pay-as-you-go pension systems with fully funded schemes, a reform often advocated for its favorable impact on saving. However, analytical 404 IHE WORLD BANK EC(ONOMIC REVIEW, VOL. 14, NO. 3 considerations suggest that the impact of pension reform on saving is not a given, but rather hinges on the way the transition deficit is financed and on the reform's efficiency gains. Pension reform should have little short-run impact on private saving if it is financed by issuing public debt, since this entails converting an implicit government liability into an explicit one. If, however, the transition is financed by reducing the nonpension public deficit (by lowering net benefits to current retirees, imposing higher taxes on current generations, or lowering gov- ernment expenditures), saving levels of current generations will decline, while those of future generations will rise, although their saving rates will not necessar- ily change. In the long term pension reform can have additional effects on saving through mandatory saving requirements. A well-known example is Singapore's Central Provident Fund, requiring minimum retirement contributions of 25 percent of salary. These requirements may raise the saving of low-income borrowing- constrained earners well above what they would have saved otherwise, although the welfare implications of such a change are obviously open to question. Pen- sion reform also can have positive, indirect effects on saving if it raises per capita income and growth by reducing labor market distortions and spurring capital market development. Empirical evidence shows that countries that increase the funding of their mandatory retirement programs tend to achieve higher private saving rates. Three of the panel-data studies reported in table 1 show that pay-as-you-go transfers negatively affect private saving rates or that fully funded contributions or assets positively affect private saving rates. Time-series evidence for Chile, the first coun- try that reformed its pension system, suggests that 3.8 percentage points of the 12.2 percentage-point increase in the national saving rate since 1986 can be at- tributed to pension reform (Schmidt-Hebbel 1999). Samwick (2000) provides further time-series evidence for five reforming countries, revealing that national saving increased only in Chile. However, Samwick also reports that pay-as-you- go systems had significant negative effects on saving-the magnitude of which increased with the system's coverage rate-in cross-country regressions of 94 countries. FINANCIAL LIBERALIZATION. Financial liberalization includes interest rate liber- alization, elimination of credit ceilings, easing of entry for foreign financial insti- tutions, development of capital markets, and enhanced prudential regulation and supervision. Until recently, the view that financial liberalization should encour- age aggregate saving was widely held. Analytically, we can separate the effect of financial liberalization on private saving rates into a direct, short-run impact, which is generally negative, and an indirect, long-run impact, which is generally positive. The direct impact is felt through price and quantity channels. The price channel refers to the effect on saving of higher interest rates, which typically result from financial liberalization. Although popularly advocated in the finan- cial press, higher interest rates are seldom found to be effective in raising private Loayza, Schmidt-Hebbel, and Serven 405 saving, suggesting that the negative income effect of higher interest rates tends to neutralize the positive intertemporal substitution effect. The quantity channel works through the expansion of the supply of credit to previously credit-constrained private agents, allowing households and small firms to use collateral more widely, and reducing down payments on loans for housing and consumer durables. Theory predicts that the expansion of credit should re- duce private saving as individuals are able to finance higher consumption at their current income level. This prediction is well supported by the empirical evidence: a 1 percentage-point increase in the ratio of private credit flows to income re- duces the long-term private saving rate by 0.74 percentage point (Loayza, Schmidt- Hebbel, and Serv6n 2000). Deeper analysis of eight episodes of financial liberal- ization fails to find a systematic direct effect on saving rates: it is clearly negative in some cases (the Republic of Korea and Mexico), positive in others (Ghana and Turkey), and negligible in the rest, likely reflecting different emphases on price and quantity channels (Bandiera and others 2000). It should be noted, however, that these studies do not use a measure of saving that includes the future- consumption portion of durable-good purchases. Taking advantage of the exten- sive national account information in India, Loayza and Shankar (this issue) find that financial development has induced private agents to change the composition of their assets to favor durable goods but has not affected the total volume of saving once saving is measured correctly (that is, to include durable purchases). This conclusion should invite a reinterpretation of the negative direct link be- tween financial development and private saving. Moreover, the indirect positive effects of financial liberalization on saving should not be underplayed. Liberalizing domestic financial markets-particularly if done by strengthening the domestic banking sector-improves the efficiency of finan- cial intermediation and hence investment, contributing to higher growth. Thus it is mostly through faster income growth that financial liberalization will increase private saving rates in the long run. EXTERNAL BORROWING AND FOREIGN AID. The relationship between national saving and foreign resource inflows in general, and foreign aid in particular, has attracted considerable attention. Many empirical studies, starting with Chenery and Strout (1966), attempt to establish whether foreign saving crowds national saving in or out, but no consensus has emerged. One problem of this literature is the simultaneity between the two variables, which can be ruled out only if do- mestic savers face binding borrowing constraints in world financial markets. Addressing this issue, Loayza, Schmidt-Hebbel, and Serven (2000) estimate that an increase of 2 percent of GNDI in the exogenous component of foreign lending reduces private (and national) saving by approximately 1 percent of GNDI in the long run.1° 10. Although the authors attempt to control for the endogeneity of foreign saving, this result should be taken with considerable caution in view of the wide disparity in external financial regimes faced by differ- ent countries in different periods. 406 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 The relationship between saving and foreign aid has also been examined in several empirical studies, starting with Griffin (1970). Most conclude that aid crowds out national saving, a finding that is confirmed by recent evidence for Sub-Saharan Africa (Elbadawi and Mwega, this issue). This result also must be taken with caution, however, as aid flows mostly to the poorest countries or to economies in distress, one symptom of which is low saving-so the negative rela- tionship between saving and aid could partly reflect reverse causation. Closer scrutiny of countries experiencing a transition from low to high saving rates re- veals instead that in a number of cases hikes in foreign aid were positively associ- ated with takeoffs in private and national saving (Rodrik, this issue). The conclu- sion is that aid need not invariably crowd out national saving.1 III. ARTICLES IN THE SYMPOSIUM ISSUE This issue includes a substantial part of the output of the World Bank research project on saving. The issue comprises two complementary sets of articles. The first group includes studies that focus on geographic regions or groups of coun- tries sharing a common saving experience: Ibrahim Elbadawi and Francis Mwega on Africa's saving collapse, Cevdet Denizer and Holger Wolf on postsocialist saving adjustment in transition economies, Peter Montiel on temporary consump- tion booms, and Dani Rodrik on saving transitions. The second group comprises case studies of saving in three developing countries: Janine Aron and John Muellbauer on South Africa, Aart Kraay on China, and Norman Loayza and Rashmi Shankar on India. Taken together, the articles in this issue provide a comprehensive assessment of private, national, and, in a few cases, household saving in major developing countries and regions of the world. Regional Studies The poor saving performance of Sub-Saharan Africa over the past three de- cades is a matter of concern given the increasing scarcity of foreign aid. Private saving ratios declined from an already low level of 11 percent in the 1 970s to 8 percent in the 1980s, recovering only partially (to less than 9 percent) in the 1990s. Ibrahim Elbadawi and Francis Mwega examine the pattern of causality between saving and related aggregates, study the determinants of private saving, and identify specific policies that could help to reverse the region's poor saving performance. They find weak evidence that growth precedes saving, but not the reverse. Foreign aid-of major importance in the region-is found to negatively Granger-cause saving. Panel estimation shows that the region's low private sav- ing is not a result of unknown features specific to Sub-Saharan Africa; rather, it reflects differences in saving fundamentals relative to other regions. In explaining the difference in saving performance between Sub-Saharan Africa and the high- 11. This result might reflect the impact of ongoing reforms that invited aid and induced higher invest- ment and growth, so that aid and saving rose together ex post. On this issue see Burnside and Dollar (2000). Loayza, Schmidt-Hebbel, and Serven 407 performing Asian economies, the authors identify low per capita income, the high young-age dependency ratio, and large amounts of foreign aid as the main causal factors. Complementary evidence from country experiences suggests that controls on external trade and capital flows in Zimbabwe and a relatively devel- oped domestic financial system in both Kenya and Zimbabwe contribute to the relatively high private saving rates in these two countries. In turn, prudent public management of a boom in a nonrenewable resource (diamonds) explains Botswana's very high public saving rate, which is reflected-through the imper- fect offset between private and public saving-in a high national saving rate. A striking feature of the economic transition in Central and Eastern Europe has been the dramatic decline in measured ratios of gross domestic saving to gross domestic product (GDP), from about 30 percent in 1989, before the transi- tion, to around 10 percent in 1992-94, with a partial recovery afterward. Cevdet Denizer and Holger Wolf analyze three possible explanations for the saving col- lapse in a sample of 10 Eastern European countries, 3 Baltic countries, and 12 successor countries of the former Soviet Union: elimination of involuntary excess saving during socialism, changes in the responsiveness of saving to its key deter- minants, and changes in the determinants of saving themselves. They assess the extent of involuntary saving by comparing saving rates of market economies with hypothetical saving rates in pretransition economies that are predicted from the same fundamentals underlying saving rates in market economies. On bal- ance, the predicted saving rates fell short of actual saving rates-particularly for the countries of the former Soviet Union and the Baltics and less so for Central European countries-supporting the notion of excessive saving before the transi- tion. Substantial similarities are found between the saving behavior of market and transition economies. The exception is the negative link between growth and saving in transition economies, suggesting that consumption smoothing in the face of a deep recession dominates the other links between growth and saving. Finally, the authors find that economic liberalization reduces saving, a further indication of smoothing in the presence of an output path that follows a J-curve. Transitory surges in consumption have played a major role in boom-bust ag- gregate cycles in many developing countries. Peter Montiel analyzes consump- tion booms across countries, weighing potential macroeconomic and financial explanations. Defining a consumption boom as a large and sustained deviation of the ratio of real private consumption to real GDP from its normal or trend value, Montiel identifies 39 booms that have occurred since the 1960s. Prospec- tive causes include income redistribution through populist policies, changes in intertemporal relative prices (temporarily lower real interest rates and nominal interest rates and appreciated real exchange rates), wealth effects (perceived wealth arising from terms-of-trade windfalls and expectations of higher future growth from successful reforms), and credit booms (rapid expansion of credit to the private sector with implicit or explicit government backing of financial liabilities, resulting from inappropriate domestic financial liberalization and large capital inflows). Empirical results from probit regressions, based on a panel sample of 408 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 52 countries with and without boom episodes, show that booms are more likely to occur because of appreciated real exchange rates and favorable terms of trade. Other factors are not statistically significant in causing consumption booms. However, further country-by-country analysis shows that some of the other po- tential causes-including anticipated higher growth and the expansion of private credit-also have contributed to several episodes. Indeed, in 16 of the 39 booms more than one cause was associated with the temporary consumption surge. Over the past three decades only a handful of countries (the "takeoff" econo- mies) have achieved persistent increases in their growth and saving rates. Dani Rodrik examines the factors that promote such takeoffs and the time precedence between saving and growth, using a formal approach to identify saving transi- tions. For a sample of 20 countries that have experienced saving transitions (de- fined as a sustained increase of 5 percentage points or more in the ratio of na- tional saving to national income) since the 1960s, growth rates tend to return to levels before the saving transition, even though saving rates remain high. This contrasts with the findings for a sample of 18 countries that experienced growth transitions (defined as a sustained increase in the growth rate of 2.5 percentage points or more): growth booms are associated with permanent increases in sav- ing rates. This evidence, complemented by Granger-causality tests, indicates that growth tends to lead saving, not the reverse. Rodrik suggests that behind these results are positive productivity shocks that raise the return on investment, lead- ing to higher aggregate investment and growth. Saving rates follow higher invest- ment and growth because of factors such as consumption habits. Further inquiry into takeoffs of individual countries leads Rodrik to identify favorable policy changes behind the growth spurts. Investment and export incentives and sup- portive public investment were behind the growth (and subsequent saving) take- offs in Korea, Singapore, and Taiwan (China), while financial development and implementation of mandatory fully funded pension systems contributed to the takeoffs in Chile and Singapore. Country Studies South Africa's gross national saving rate declined by half between the 1980s and the 1990s. In net terms it fell from 8 to 1 percent. Janine Aron and John Muellbauer examine the causes, and possible remedies, of this precipitous fall. They show that the trend decline in national saving reflects mainly a deteriora- tion in the government's saving performance. Private saving remained roughly stable until recently, with declining household saving offset by rising corporate saving. Aron and Muellbauer focus on the causes of the disparate evolution of these two components of private saving. For this purpose they construct a com- prehensive database on private sector wealth. Their analysis also includes a num- ber of innovations, such as the explicit inclusion of asset effects and income ex- pectations and the modeling of corporate profits. The empirical findings show that financial liberalization has been a major factor behind the decline in per- Loayza, Schmidt-Hebbel, and Serven 409 sonal saving and the rise in corporate saving, whereas the increase in real interest rates has had a positive impact on both components of saving. Aron and Muellbauer highlight the policy implications of their results in terms of corporate taxation and prudential regulation of financial intermediaries. China, the world's largest and fastest-growing economy, also has national sav- ing rates that are among the highest in the world. Although China's saving rates are similar to those in the East Asian miracle economies, they have been achieved at much lower income levels. Furthermore, China's transition to a market economy has not been associated with declining saving, unlike in most other former social- ist countries. Aart Kraay's study represents a first attempt to sort out these puzzles. Kraay highlights the considerable statistical difficulties that cloud the measure- ment of saving in China, both at the national and household level. For example, depending on the data source, household saving could represent one-fourth or one-half of gross national saving. These unresolved measurement issues make it difficult to interpret the recent trends in China's saving aggregates. With this strong caveat, Kraay explores the main determinants of household saving. He finds that expectations of future income growth, as well as income levels higher than subsistence consumption (proxied by the share of food consumption), play a significant role in the observed evolution of saving, although they account for only a small portion of the observed variation in household saving. In contrast, he finds that demographic factors or income uncertainty have no effect on saving- although the fact that saving rates of rural households are much higher than those of urban households may partly reflect the greater uncertainty of rural incomes. In recent years India's national and private saving performance has surpassed that of countries with comparable per capita incomes-although to a more mod- est extent than in China. Norman Loayza and Rashmi Shankar analyze the fac- tors behind India's high private saving rates. A distinguishing feature of their article is its use of private and public saving measures that are adjusted for infla- tionary capital gains and losses, durable expenditures, and human capital accu- mulation, bringing them close to their theoretically correct counterparts. As Loayza and Shankar show, such adjustments-which are made possible by India's highly detailed data-result in large changes in public and private saving relative to the naive measures based on national accounts data. Using these improved measures of saving, the article identifies the determinants of private saving, after showing that perfect offset between household and corporate saving makes it unnecessary to examine them separately. The ernpirical results show that private saving ra- tios react positively to the real interest rate and are negatively affected by age- dependency ratios. Saving also is positively related to the share of agricultural income in total income. As with cross-country results concerning the degree of urbanization, this likely reflects precautionary saving motives. Loayza and Shankar also show that the naive measure of private saving is adversely affected by finan- cial liberalization, whereas the (theoretically superior) measure of saving inclu- sive of purchases of durable goods is not. 410 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Papers and Publications of the Saving Research Project Aron, Janine, and John Muellbauer. 2000. "Personal and Corporate Saving in South Africa." The World Bank Economic Review. This issue. Attanasio, Orazio, Lucio Picci, and Antonello Scorcu. 2000. "Saving, Growth, and In- vestment." Review of Economics and Statistics 82(2):182-211. Bandiera, Oriana, Gerard Caprio, Patrick Honohan, and Fabio Schiantarei. 2000. "Does Financial Reform Raise or Reduce Private Savings?" Review of Economics and Statis- tics 82(2):239-63. Besley, Timothy, and Costas Meghir. 1998. "Do Tax Incentives Raise Private Saving?" World Bank, Washington, D.C. Processed. Burnside, Craig. 1998. "Private Saving in Mexico." World Bank, Washington, D.C. Processed. Burnside, Craig, Klaus Schmidt-Hebbel, and Luis Serven. 1999. "Saving in Mexico: The National and International Evidence." Economia Mexicana 8(2):181-230. Deaton, Angus. 1999. "Saving and Growth." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge, U.K.: Cambridge University Press. Deaton, Angus, and Guy Laroque. 1998. "Land Prices, Housing, and Household Sav- ing." World Bank, Washington, D.C. Processed. Deaton, Angus, and Christina Paxson. 2000. "Saving and Growth: Another Look at the Cohort Evidence." Review of Economics and Statistics 82(2):212-25. Denizer, Cevdet, and Holger Wolf. 2000. "Aggregate Saving in the Transition: A Cross- Country Study." The World Bank Economic Review. This issue. Elbadawi, Ibrahim, and Francis Mwega. 2000. "Can Africa's Savings Collapse Be Re- versed?" The World Bank Economic Review. This issue. Honohan, Patrick. 1999. "Financial Policies and Saving." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge, U.K.: Cam- bridge University Press. Kraay, Aart. 2000. "Household Saving in China." The World Bank Economic Review. This issue. Loayza, Norman, Humberto L6pez, Klaus Schmidt-Hebbel, and Luis Serven. 1998a. "Sav- ing in the World: Stylized Facts." World Bank, Washington, D.C. Processed. .1998b. "The World Saving Database." World Bank, Washington, D.C. Processed. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serven. 2000. "What Drives Private Saving Across the World?" Review of Economics and Statistics 82(2):165-81. Loayza, Norman, and Rashmi Shankar. 2000. "Private Saving in India." The World Bank Economic Review. This issue. L6pez, Humberto, Klaus Schmidt-Hebbel, and Luis Serv6n. 2000. "How Effective Is Fiscal Policy in Raising National Saving?" Review of Economics and Statistics 82(2): 226-38. Loayza, Schmidt-Hebbel, and Serven 411 L6pez-Mejia, Alejandro, and J. R. Ortega. 1998. "Saving in Colombia." IMF Working Paper WP/98/171. International Monetary Fund, Washington, D.C. Processed. Montiel, Peter J. 2000. "What Drives Consumption Booms?" The World Bank Eco- nomic Review. This issue. Obstfeld, Maurice. 1999. "Foreign Resource Inflows, Saving, and Growth." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cam- bridge, U.K.: Cambridge University Press. Rodrik, Dani. 2000. "Savings Transitions." The World Bank Economic Review. This issue. Samwick, Andrew. 2000. "Is Pension Reform Conducive to Higher Saving?" Review of Economics and Statistics 82(2):264-72. Schmidt-Hebbel, Klaus, and Luis Serven. 1997. Saving Across the World: Puzzles and Policies. World Bank Discussion Paper 354. Washington, D.C.: World Bank. . 1999a. "Aggregate Saving and Income Distribution." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge: Cambridge University Press. , eds. 1999b. The Economics of Saving and Growth. Cambridge, U.K.: Cambridge University Press. .1999c. "Saving in the World: The Stylized Facts." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge, U.K.: Cam- bridge University Press. .1999d. "World Saving: Trends and Theories." Estudios de Economia 25(Decem- ber):191-215. . 2000. "Does Income Inequality Raise Aggregate Saving?" Journal of Develop- ment Economics 61(2):417-46. General References Abel, Andrew B. 1990. "Asset Prices under Habit Formation and Catching up with the Joneses." American Economic Review, Papers and Proceedings 80(1):38-42. Atkeson, Andrew, and Masao Ogaki. 1991. "Wealth-Varying Intertemporal Elasticities of Substitution: Evidence from Panel and Aggregate Data." Working Paper 303. Uni- versity of Rochester, Center for Economic Research, Rochester, N.Y. Processed. Bailliu, Janine, and Helmut Reisen. 1998. "Do Funded Pensions Contribute to Higher Aggregate Savings? A Cross-Country Analysis." Review of World Economics -134(4):692-711. Barro, Robert. 1974. "Are Government Bonds Net Wealth? " Journal of Political Economy 82(6): 1095-117. Bosworth, B. P. 1993. Saving and Investment in a Global Economy. Washington, D.C.: Brookings Institution. Bosworth, B. P., Gary Burtless, and John Sabelhaus. 1991. "The Decline in Saving: Some Microeconomic Evidence." Brookings Papers on Economic Activity 43:183-241. Burnside, Craig, and David Dollar. 2000. "Aid, Policies, and Growth." American Eco- nomic Review 90(4):847-68. Campbell, John Y., and John H. Cochrane. 1994. "By Force of Habit: A Consumption- Based Explanation of Aggregate Stock Market Behavior." Research Working Paper 94-17. Federal Reserve Bank of Philadelphia. Processed. 412 THE WORL D BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Carroll, Christopher. 1991. "Buffer Stock Saving and the Permanent Income Hypoth- esis." Working Paper Series 114. Board of Governors of the Federal Reserve System, Economic Activity Section, Washington, D.C. Processed. Carroll, Christopher, and Andrew Samwick. 1995a. "How Important Is Precautionary Saving?" NBER Working Paper 5194. National Bureau of Economic Research, Cam- bridge, Mass. Processed. 1995b. "The Nature of Precautionary Wealth." NBER Working Paper 5193. Na- tional Bureau of Economic Research, Cambridge, Mass. Processed. Carroll, Christopher, and Lawrence Summers. 1989. "Consumption Growth Parallels Income Growth: Some New Evidence." NBER Working Paper 3090. National Bureau of Economic Research, Cambridge, Mass. Processed. Carroll, Christopher, and David Weil. 1994. "Saving and Growth: A Reinterpretation." Carnegie-Rochester Conference Series on Public Policy 40:133-92. Chenery, Hollis, and Alan Mayne Strout. 1966. "Foreign Assistance and Economic De- velopment." American Economic Review 56(4):679-733. Collins, Susan M. 1991. "Saving Behavior in Ten Developing Countries." In B. Douglas Bernheim and J. B. Shoven, eds., National Saving and Economic Performance. Chi- cago: National Bureau of Economic Research and the University of Chicago Press. Corbo, Vittorio, and Klaus Schmidt-Hebbel. 1991. "Public Policies and Saving in Devel- oping Countries." Journal of Development Economics 36(1):89-115. Dayal-Gulati, Anuradha, and Christian Thimann. 1997. "Saving in Southeast Asia and Latin America Compared: Searching for Policy Lessons." IMF Working Paper 97/110. International Monetary Fund, Washington, D.C. Processed. Deaton, Angus. 1991. "Saving and Liquidity Constraints." Econometrica 59(July):1121- 42. . 1992. Understanding Consumption. Oxford: Clarendon Press. Deaton, Angus, and Christina Paxson. 1994. "Intertemporal Choice and Inequality." Journal of Political Economy 102(3):437-67. Dooley, Michael, Jeffrey Frankel, and D. J. Mathieson. 1987. "International Capital Mobility: What Do Saving-Investment Correlations Tell Us?" IMF Staff Papers 34 (Sep- tember):503-30. Edwards, Sebastian. 1995. "Why Are Saving Rates So Different across Countries? An International Comparative Analysis." NBER Working Paper 5097. National Bureau of Economic Research, Cambridge, Mass. Processed. . 1996. "Why Are Latin America's Saving Rates So Low? An International Com- parative Analysis." Journal of Development Economics 51(1):5-44. Ferson, Wayne E., and George M. Constantinides. 1991. "Habit Persistence and Durabil- ity in Aggregate Consumption: Empirical Tests." Journal of Financial Economics 29(2):199-240. Flavin, Marjorie A. 1981. "The Adjustment of Consumption to Changing Expectations about Future Income." Jotrnal of Political Economy 89(5):974-1009. Friedman, Milton. 1957. "A Theory of the Consumption Function." General Series 63. National Bureau of Economic Research, Cambridge, Mass. Processed. Griffin, Keith. 1970. "Foreign Capital, Domestic Savings, and Economic Development." Bulletin, Oxford University Institute of Economics and Statistics, 32(1):99-112. Loayza, Schmidt-Hebbel, and Serven 413 Hall, Robert. 1978. "Stochastic Implications of the Life Cycle-Permanent Income Hy- pothesis: Theory and Evidence." Journal of Political Economy 86(6):971-87. Kaldor, Nicholas. 1957. "A Model of Economic Growth." Economic Journal 67(268): 591-626. Kotlikoff, Laurence J., and L. H. Summers. 1981. "The Role of Intergenerational Transfers in Aggregate Capital Formation." Journal of Political Economy 89(4):706- 32. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Lewis, Arthur. 1954. "Economic Development with Unlimited Supplies of Labor." Manchester School 22:139-91. Maddison, Angus. 1992. "A Long-Run Perspective on Saving." Scandinavian Journal of Economics 94(2):181-96. Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. "A Contribution to the Empiricis of Economic Growth." Quarterly Journal of Economics 107(2):407-37. Masson, Paul, Tamim Bayoumi, and Hossein Samiei. 1995. "Saving Behavior in Indus- trial and Developing Countries." Staff Studies for the World Economic Outlook. Washington, D.C.: International Monetary Fund. Menchik, M. D. 1983. "Income Distribution, Lifetime Savings, and Bequests." American Economic Review 73(4):672-90. Modigliani, Franco. 1970. "The Life-Cycle Hypothesis of Saving and Intercountry Dif- ferences in the Saving Ratio." In W. A. Eltis, M. F. G. Scott, and J. N. Wolfe, eds., Induction, Trade, and Growth: Essays in Honour of Sir Roy Harrod. Oxford: Clarendon Press. Modigliani, Franco, and Richard Brumberg. 1954. "Utility Analysis and the Consump- tion Function." In K. K. Kurihara, ed., Post-Keynesian Economics. New Brunswick, N.J.: Rutgers University Press. OECD (Organisation for Economic Co-operation and Development). 1994. "Taxation and Household Savings." Paris. Processed. Ogaki, Masao, Jonathan D. Ostry, and Carmen M. Reinhart. 1996. "Saving Behavior in Low- and Middle-Income Developing Countries: A Comparison." IMF Staff Papers 43(1):38-71. Pasinetti, Luigi. 1962. "Rate of Profit and Income Distribution in Relation to the Rate of Economic Growth." Review of Economic Studies 29:267-79. Pesaran, M. Hashem, Nadeem Haque, and Sunil Sharma. 2000. "Neglected Heterogeneity and Dynamics in Cross-Country Savings Regressions." In Jaya Krishnakumar and Elvezio Ronchetti, eds., Panel Data Econometrics-Future Direction: Papers in Honour of Professor Pietro Balestra. Amsterdam: North Holland. Poterba, James. 1995. International Comparisons of Saving. Chicago: Chicago Univer- sity Press for the National Bureau of Economic Research. Schmidt-Hebbel, Klaus. 1999. "Chile. Die Revolution des Alterssicherungssystems tragt Fruchte [Chile's Pension Revolution Coming of Age]." In Deutsches Institut for Altersvorsorge, Gesetzliche Alterssicherung, Reformerfahrungen im Ausland. Ein Systematischer Vergleich aus Sechs Landern [Reforming the Pension System. What Germany Can Learn from Six Countries]. Cologne: Deutsches Institut fur Altersvorsorge. 414 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Schmidt-Hebbel, Klaus, Steven B. Webb, and Giancarlo Corsetti. 1992. "Household Sav- ing in Developing Countries: First Cross-Country Evidence." The World Bank Eco- nomic Review 6(3):529-47. Skinner, Jonathan. 1988. "Risky Income, Life-Cycle Consumption, and Precautionary Savings." Journal of Monetary Economics 22(2):237-55. Zeldes, S. P. 1989. "Consumption and Liquidity Constraints: An Empirical Investiga- tion." Journal of Political Economy 97(2):305-46. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 415-43 Can Africa's Saving Collapse Be Reversed? Ibrahim A. Elbadawi and Francis M. Mwega Private saving in Sub-Saharan Africa declined from more than 11 percent of disposable income in the 1970s to less than 8 percent in the 1980s and only partially recovered (to less than 9 percent) in the 1 990s. This article analyzes the determinants of private saving in Sub-Saharan Africa, seeking to explain the region's dismal performance and identify poli- cies that could help to reverse the region's decline in saving. The analysis shows that in Sub-Saharan Africa causality runs fronm growth to investment (and perhaps to private saving), whereas a rise in the saving rate Granger-causes an increase in investment. For- eign aid Granger-causes a reduction in both savinig and investment, and investment also Granger-causes an increase in foreign aid. The empirical analysis of private saving in Sub- Saharan Africa and other regions over 1970-95 suggests that private saving in Africa can be explained by standard behavioral models. According to these models private saving in Africa lags behind that in other regions (most notably, the high performing Asian econo- mies) because of the region's lower per capita income, high young-age dependency ratio, and high dependence on aid. The combined effects of these factors substantially outweigh Africa's advantage from its lower public saving and higher government consumption. Finally, analysis of the experiences of Kenya, Zimbabwe, and Botswana provides further insight into the saving process in Sub-Saharan Africa. Sub-Saharan Africa's growth and overall economic performance over the past two decades have been described in recent scholarly writings as "tragic" (East- erly and Levine 1997) and of "crisis proportions" (Elbadawi and Ndulu 1996).' Unfortunately, these assessments are not exaggerations. The median annual growth rate of real gross domestic product (GDP) per capita in Africa declined steadily from more than 2.0 percent in the 1970s to -0.3 percent in 1980s and to -1.3 percent in the 1990s. Over that period Africa's growth performance also lagged considerably relative to that of other regions, most notably the high-performing Asian economies. After noting that in 1996 Sub-Saharan Africa registered its highest economic growth in two decades (total income grew at 5 percent), Rodrik (1997:1) cautions, "The sobering reality is that it will take many years of growth 1. See also Collier (1995), Rodrik (1997), Schmidt-Hebbel (1996), and Hadjimichael and Ghura (1995), among others. Ibrahim A. Elbadawi is with the Development Research Group and the Africa Region at the World Bank, and Francis M. Mwega is with the Department of Economics at the University of Nairobi. Their e- mail addresses are ielbadawi@worldbank.org and aercres@form-net.com. This article was prepared for the World Bank's project, "Saving Across the World: Puzzles and Policies." The authors would like to acknowledge the research assistance of Radha Ruparel. They also are grateful to the editor of The World Bank Economic Review and two anonymous referees for helpful comments. © 2000 The International Bank for Reconstruction and Development I THE WORLD BANK 415 416 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 at such levels (or better) to undo the damage that more than two decades of stagnation and decline have inflicted on most countries of the region."2 Growth rates in Sub-Saharan Africa and other regions have been strongly as- sociated with their private saving and investment rates. For example, private saving in Sub-Saharan Africa declined from 11.4 percent of disposable income in the 1970s to only 7.5 percent in the 1980s. It only partially recovered to less than 9 percent in the 1990s. In addition, public saving in Sub-Saharan Africa has remained low-at less than 3 percent of disposable income in the 1990s, declin- ing from 4.3 percent in the 1980s. The slowdown in gross domestic investment (as a fraction of GDP) has been equally dramatic: it declined steadily from 21 percent in the 1970s to about 17 percent in the 1990s. Like growth, saving and investment In Sub-Saharan Africa lagged behind that in all other regions, especially the high-performing Asian economies.3 The Asian econo- mies had private saving (and investment) rates that were almost 10 percent of GDP (or almost 17 percent of disposable income) higher than those of Sub-Saharan Africa. Moreover, saving in Sub-Saharan Africa was not only lower relative to other regions but also less stable. For example, the coefficient of variation of pri- vate (public) saving in Sub-Saharan Africa is more than three (two) times that of the high-performing Asian economies. Yet another worrisome feature of saving and investment in Sub-Saharan Africa is the region's heavy dependence on foreign saving, mostly overseas development assistance, to finance the gap between invest- ment and saving, which averaged slightly less than 11 percent of GDP in 1970-95. This contrasts sharply with the high-performing Asian economies, where saving exceeded investment by 2.6 and 3.4 percent in 1970-79 and 1980-89, respec- tively, and was virtually equal to investment in 1990-95 (tables 1 and 2). The urgent need to reinvigorate and sustain growth in the developing world has inspired a plethora of theoretical frameworks aimed at understanding the observed empirical correlations among growth, saving, and investment, as well as their determinants. Although researchers have made good progress on the empirical modeling of growth and investment in Sub-Saharan Africa and other developing regions, the variation of private saving across countries and over time is less understood (Schmidt-Hebbel, Serven, and Solimano 1994, 1996).4 The presence of significant and negative dummy variables for Africa and Latin America in cross-country regressions of private saving underscores this point (see, for ex- ample, Edwards 1995 and Mwega 1996). Schmidt-Hebbel, Serv6n, and Solimano (1996) review several theories on the direction of causation between saving and growth, ranging from the classical 2. According to Rodrik (1997) roughly a third of the countries in Sub-Saharan Africa (16 countries in all) had higher per capita GDP in the early 1960s than they do three and a half decades later. 3. We include in this category China, Hong Kong, Indonesia, the Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. 4. Easterly and Levine (1997); Schmidt-Hebbel (1996); Elbadawi, Ndulu, and Ndung'u (1997); and Collier (1998) estimate endogenous growth models with a focus on Sub-Saharan Africa. Empirical invest- ment models that emphasize risk and the irreversibility of investment include Serv6n (1997); Elbadawi, Ndulu, and Ndung'u (1997); and Collier (1998). Table 1. Saving and Investment in Sub-Saharan Africa and the High-Performing Asian Economies, 1970-95 (percent) Sub-Saharan Africa' High-performing Asian economiesh Ratio of gross Ratio of gross Ratio of gross domestic Ratio of gross domestic domestic savings investment to domestic savings investment to 4,, to gross domestic gross domestic Investment- to gross domestic gross domestic Investment- Years product product saving gap product product saving gap 1970-79 11.72 21.37 9.65 29.17 26.52 -2.65 1980-89 6.66 18.69 12.03 33.24 29.84 -3.40 1990-95 6.24 17.19 10.95 35.65 36.46 0.81 a. Sub-Saharan Africa includes Angola, Benin, Botswana, Burkina Faso, Butundi, Cameroon, Central African Republic, Chad, Comoros, Congo, Cote d'lvoire, Djibouti, Ethiopia, Gabon, The Gambia, Ghana, Guinea, Guinea Bissau, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger, Nigeria, Rwanda, Sao Tome and Principe, Senegal, Seychelles, Sierra Leone, South Africa, Sudan, Swaziland, Tanzania, Togo, Uganda, Zambia, and Zimbabwe. b. The high-performing Asian economies include China, Hong Kong, Indonesia, the Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. Source: Authors' calculations based on data in the World Bank's World Saving Database. 418 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 2. Aid Dependence in Sub-Saharan Africa and Other Developing Regions, 1970-95 (percent) Sub- High- Latin Saharan performing South America and Indicator Africa, Asian economiesb Asiac the Caribbeand Median percentage of years for 60.87 0.00 0.00 0.00 which the ratio of aid to gross domestic product is greater than 10 percent Mean percentage of years for 47.57 0.00 18.09 6.56 which the ratio of aid to gross domestic product is greater than 10 percent a. Sub-Saharan Africa includes Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Central African Republic, Chad, Comoros, Congo, C6te d'lvoire, Djibouti, Ethiopia, Gabon, The Gambia, Ghana, Guinea, Guinea Bissau, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger, Nigeria, Rwanda, Sao Tome and Principe, Senegal, Seychelles, Sierra Leone, South Africa, Sudan, Swaziland, Tanzania, Togo, Uganda, Zambia, and Zimbabwe. b. The high-performing Asian economies include China, Hong Kong, Indonesia, the Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. c. South Asia includes Bangladesh, Bhutan, Cook Island, India, Maldives, Myanmar, Nepal, Pakistan, and Sri Lanka. d. Latin America and the Caribbean includes Antigua and Barbuda, Argentina, the Bahamas, Barbados, Belize, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominica, Dominican Republic, Ecuador, El Salvador, Grenada, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, St. Kitts and Nevis, St. Lucia, St. Vincent anJ the Grenadines, Suriname, Trinidad and Tobago, Uruguay, and Venezuela. Source: Authors' calculations based cn data in the World Bank's World Saving Database. permanent-income and life-cycle hypotheses to more recent and less conventional models. The newer models-which emphasize slowly changing consumption habits (Carroll, Overland, and Weil 1995), a mixture of strong consumption habits with uncertain incomes (Carroll and Weil 1993), or the presence of consumers who value both consumption and wealth (Cole, Mailath, and Postlewaite 1992; Fershtman and Weiss 1993; and Zou 1993)-broadly suggest that growth drives saving. Moreover, they argue that even when saving is assumed to translate auto- matically into capital accumulation and hence growth, failure to account for reverse causation from growth to saving is likely to overstate the contribution of saving to growth (Carroll and Weil 1993). At the empirical level, however, the evidence on causation from growth to saving is not conclusive (Schmidt-Hebbel, Serv6n, and Solimano 1996). The analy- sis of private saving in Sub-Saharan Africa remains of immense policy impor- tance even though growth in Sub-Saharan Africa has been shown to drive both saving and investment. Admittedly, if we assume that growth drives saving, then policies aimed at directly promoting growth (such as technological innovation or further development of human capital) should be the prime focus of attention. However, policies for directly promoting private saving may still be important for at least three reasons. First, even though capital accumulation may follow Elbadawi and Mwega 419 rather than lead growth, international evidence-most notably the experiences of the high-performing Asian economies-suggests that sustaining high rates of growth requires substantial physical capital accumulation. Second, to the extent that Sub-Saharan Africa faces binding lending constraints in international capital markets or concessional development finance, national saving will drive aggre- gate investment (Summers 1988), and hence saving will influence the sustainability of growth. Third, the recent experiences of Mexico and Chile, for example, have shown that high national saving is a prerequisite for avoiding financial crises and the subsequent collapse of growth (Williamson 1997).5 In this article we analyze the determinants of private saving in Sub-Saharan Africa to explain the region's dismal saving behavior and to identify policies that could help to reverse its decline in saving. Although raising the saving rate is not sufficient for achieving sustained growth, it does appear to be necessary. In fact, identifying policies that promote saving (and the policy distortions that inhibit saving) should be an essential element in any strategy aimed at making Sub- Saharan Africa less dependent on aid, given the extent of its dependence and the anticipated global reduction in foreign aid. I. CAUSALITY AMONG SAVING, INVESTMENT, GROWTH, AND FOREIGN AID IN SUB-SAHARAN AFRICA In this section we establish the causal links among saving, investment, growth, and foreign aid, taking into account their dynamic relationships (see, for ex- ample, Carroll and Weil 1993). Traditional analyses of these relationships have focused on the links between domestic saving and investment, the effects of sav- ing and investment on economic growth, and, more recently, the impact of for- eign aid. The Relationship between Saving and Investment The first issue is whether domestic saving and investment are strongly corre- lated. In an open economy an increase in domestic saving need not translate into an increase in domestic investment. With complete capital mobility, investment and saving could be independent of one another. With full limitations on capital mobility, domestic saving is translated directly into investment. The extent of the correlation between saving and investment is therefore an empirical matter. Empirical studies often find that domestic saving is highly correlated with in- vestment (Feldstein and Horioka 1980, Feldstein and Bacchetta 1991, and Bosworth 1993). Researchers have proposed many reasons for this high correla- tion: productivity and other shocks affect both desired saving and desired invest- ment in the same direction, even if capital is perfectly mobile across countries; an increase in domestic saving induces an increase in investment, particularly in 5. The recent East Asian experience, however, suggests that avoiding financial crises requires more than just attaining high national savings rates. 420 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 large countries; capital controls protect the domestic tax base and the balance of payments; high transaction costs to purchasing securities and investing abroad, fears of expropriation, exchange rate risks, and asymmetric information among countries handicap foreign investors; and access to external credit markets is limited (Gordon and Bovenberg 1994). In Sub-Saharan Africa the gap between saving and investment has been growing since the 1960s. The Relationship between Growth and Saving and Growth and Investment A second question is whether economic growth follows saving and investment or whether saving and investment follow growth. Neoclassical models such as Solow (1956) suggest that an increase in the saving rate generates higher growth only in the short run, during the transition between steady states. The long-run equilibrium rate of growth is exclusively a function of technological progress and growth of the labor force, both assumed to be exogenous. Models based on theo- ries of endogenous growth pioneered by Roemer (1986) and Lucas (1988), how- ever, predict that higher saving and investment rates can permanently raise growth rates. With endogenous technological progress and the accumulation of human capital, saving and investment will increase growth in the steady state. More recent research (for example, Levine and Renelt 1992, De Long and Summers 1993, Easterly and Rebelo 1993, and King and Levine 1994) finds that the in- vestment rate is one of the most important determinants of economic growth. Moreover, other studies using neoclassical production function analysis find that investment, not technological progress, was the main factor behind Southeast Asia's exemplary growth (see, for example, Young 1994). The main argument behind these results is that investment increases the amount of physical capital per worker and therefore raises productivity. The saving literature suggests that economic growth increases the income of workers relative to that of nonworkers (children and retirees). Hence workers' saving could rise, for example, to allow for increased consumption during retire- ment. However, children may borrow against future income, and workers, an- ticipating increased growth, may reduce saving (Tobin 1967 and Bosworth 1993). Growth also may reduce liquidity and borrowing constraints, inducing house- holds to increase consumption. Therefore, economic growth will increase saving if growth is concentrated among households with high saving rates and persis- tent, strong habits and if consumption adjusts with a lag to an increase in in- come, especially if the increase is not anticipated. Also, if income is uncertain, income growth could be positively correlated with saving through a precaution- ary (buffer-stock) motive (Harrigan 1995). Under the permanent-income hypoth- esis unexpected income will largely be saved. Causation is two-way if saving also increases growth. A virtuous cycle results as growth leads to more saving, which in turn enhances growth. It is important to determine which direction of causality is dominant: is rapid growth mainly the result of a higher saving rate, or does saving respond mainly to economic growth? If saving is a major determinant of growth, increasing it would be central to policy, whereas if the converse is true, policy should focus largely on the factors Elbadawi and Mwega 421 driving growth (Schmidt-Hebbel and Serven 1996). The most definitive empiri- cal work on this issue is Carroll and Weil (1993). Using causality tests, they find that growth mainly causes changes in saving rates. Perhaps inspired by this re- sult, the Africa adjustment study (World Bank 1994), which notes that saving rates traditionally have been low in Sub-Saharan Africa and have been affected little by adjustment, postulates that saving largely follows economic growth. Even though the study does not conduct a formal causality analysis, it concludes that policymakers in Sub-Saharan Africa should not attempt to force up the saving rate directly; rather, they should establish an environment that facilitates rapid accumulation, efficient resource use, and high productivity growth. Therefore, a formal test of causality among growth, saving, investment, and foreign aid is warranted. Testing for Causality The variables we use to establish the direction of bivariate causality are gross national saving as a proportion of national disposable income (SY), gross domes- tic investment as a proportion of GDP (IY), real GDP growth (GR), and foreign aid as a proportion of GDP (AIDY). The correlation coefficient between saving and growth (0.247) is slightly higher than that between investment and growth (0.231). However, the strongest link is between saving and investment (0.532). The corre- lation coefficient is lowest between investment and foreign aid (0.166) and is negative with respect to both saving and growth (-0.215 and -0.008). For each relationship we determine whether variable X significantly adds to the explanation of variable Y, controlling for the history of Y (and vice versa) based on the following equations: n m (1) zz" = c° i + -j ,iYt i,i + + Et., n m (2) Xt = 0 +,otXcg}iXt4 l tX i2Yt + t w were i refers to the country, and n and m to the number of lags. We estimate these equations using annual data, which allows us to differentiate between shortrun and long-run causality. Rather than experiment with different lag struc- tures for every equation, we use four lags. One problem with panel data estimation is that differences among countries may reflect country-specific characteristics-such as the efficiency of government, the degree of corruption, the level of violence, or the attitude of the government and population toward individual achievement or enterprise-that jointly influ- ence saving, investment, economic growth, and dependence on foreign aid (Carroll and Weil 1993). Since some of these country-specific effects are difficult to quan- tify and are likely to be correlated with other observed explanatory variables, fixed-effects estimation is commonly used to control for country-specific vari- ables. The success of this framework depends crucially on the extent to which the 422 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 nonparametric proxies (dummy variables) for fixed effects adequately capture the idiosyncratic aspects of country behavior, such as culture and development aspirations, that cannot be easily observed or measured (Harrigan 1995). Fixed effects estimation in effect eliminates differences across countries, leaving only time-series variations to be explained. We present both the pooled and the fixed-effects results for the sample of Sub- Saharan African countries in the World Bank's World Saving Database (tables 3, 4, and 5). The pooled results are less robust, with only causality from economic growth to investment being significant. We therefore focus on the significant fixed-effects results.6 The fixed-effects results show that the saving rate significantly Granger-causes the investment rate (table 3). The effect of the saving rate is persistent, increasing the aggregate coefficient from 0.13 in the short run to 0.37 in the long run. As in the pooled results, the fixed-effects results show significant (at the 5 percent level) and positive causality from economic growth to the investment rate (table 4). The effect of growth on the investment rate is also persistent, increasing the ag- gregate coefficient from 0.14 in the short run to 0.48 in the long run. Both of these results suggest that the data can be viewed either through neoclassical glasses or with an accelerator model in mind. Except for the finding that Granger-causality from economic growth to the saving rate is not significant, these results are largely consistent with those de- rived by Attanasio, Picci, and Scorcu (2000) in their analysis of several groupings of countries. They are also, to some extent, supported by results from South Asia and Latin America and the Caribbean (table 6). In both South Asia and Latin America and the Caribbean the investment rate significantly Granger-causes a reduction in economic growth (at least in the short run). This counterintuitive result may suggest that aggregate investment was driven by public investment, since many countries reduced their growth rates with a public investment boom. Although there are significant effects from economic growth to investment in both South Asia and Latin America and the Caribbean (as in Sub-Saharan Af- rica), the effects from economic growth to saving are significant only in Latin America. The saving rate Granger-causes an increase in the investment rate in Latin America and the Caribbean (as in Sub-Saharan Africa), but not in South Asia. Unlike in Sub-Saharan Africa, there are significant effects from the invest- ment rate to the saving rate in both regions, at least in the short run. 6. Another way to control for fixed effects is by differencing the data. This, however, creates moving- average disturbances that are correlated with the (once) lagged dependent variables. We therefore instru- ment these variables using the generalized methods of moments (GMM) estimator. Instrumenting the (once) lagged dependent variables also accounts for the endogeneity of X and Y in the two estimation equations. We use internal instruments (lagged values of the explanatory variables) as proposed by Holtz- Eakin, Newey, and Rosen (1988) and Allerano and Bond (1991), who suggest that one can use the or- thogonal restrictions implied in the data dynamics to achieve efficiency if the error terms are serially Lncorre]ated. Differencing the data, however, and instrumenting the (once) lagged dependent variables give less precise estimates, with none of the relations between the savings rate, investment rate, economic growth, and foreign aid significant at even the 10 percent level. Elbadawi and Mwega 423 Table 3. Causality among Saving, Investment, and Economic Growth in Sub- Saharan Africa, 1970-95 Saving-p Investment-t Growth-4 Saving-s Model investment saving saving growth Pooled model Sum of coefficients 0.0165 0.0832 0.0635 0.0356 (0.8861) (0.9156) (0.9718) (0.9700) Long-run coefficients 0.2243 0.4674 0.4770 0.0388 (0.9111) (0.9515) (0.9820) (0.9544) Sample size 693 690 629 638 Fixed-effects model Sum of coefficients 0.1300 0.0419 0.0543 0.0050 (0.0010) (0.3800) (0.4810) (0.9060) Long-run coefficients 0.3654 0.1030 0.1191 0.0050 (0.0002) (0.3710) (0.4700) (0.9060) Sample size 693 690 629 638 Note: Sum is the sum of the Granger-causality coefficients in the estimation equations. The long-run coefficient is derived by dividing this sum by I minus the sum of the coefficients of the lagged dependent variables. The numbers in parentheses are p-values. The p-value is the probability that the sum of the estimated Granger-causality coefficients is equal to zero. Source: Authors' calculations based on data in the World Bank's World Saving Database. The fixed-effects results for Sub-Saharan Africa also show that the foreign aid ratio significantly Granger-causes a reduction in the saving rate, as expected from the permanent-income hypothesis, provided that aid is not entirely wasted (table 4). There is no significant bivariate relationship between aid and growth in the long run, as the Solow model predicts. Lastly, the fixed-effects results show that the foreign aid ratio and the invest- ment rate Granger-cause one another, at least at the 10 percent level (table 5). The foreign aid ratio reduces investment, while the investment rate raises the foreign aid ratio. The latter suggests that African countries that increase invest- ment receive more foreign aid. A literal interpretation of the fixed-effects results is that, although the saving rate increases the investment rate, neither variable is relevant for initiating and sustaining growth in Sub-Saharan Africa. However, domestic saving is still likely to be critical not only for increasing investment finance but also for promoting economic growth. Also, avoiding a large gap between investment and saving contributes to macroeconomic stability and hence economic growth. It thus is important to understand the factors that influence the evolution of saving rates in Sub-Saharan Africa over time and the evolution of saving rates across countries. II. DETERMINANTS OF PRIVATE SAVING IN SUB-SAHARAN AFRICA AND IN OTHER REGIONS The empirical analysis of private saving is based on a general functional form, encompassing a wide range of theoretical models derived from first principles. Table 4. Causality between Investment and Economic Growth and between Foreign Aid and Saving in Sub-Saharan Africa, 1979-95 Model Investment --growth Growth -investment Foreign aid-4saving Saving--foreign aid Pooled model Sum of coefficients 0.054982 0.9195 -0.0259 -0.0984 (0.8987) (0.0000) (0.9999) (0.9999) Long-run coefficient 0.0584 0.9892 -0.1660 -0.9406 (0.9545) (0.7637) (0.9941) (0.9999) Sample size 679 674 577 596 Fixed-effects model Sum of coefficients -0.0363 0.1433 -0.0073 0.3883 (0.428) (0.037) (0.014) (0.753) Long-run coefficient 313 0.4842 -0.0139 0.5777 (0.435) (0.031) (0.01 8) (0.753) Sample size 679 674 577 596 Note: Sum is the sum of the Granger-causality coefficients in the estimation equations. The long-run coefficient is derived by dividing this sum by 1 minus the sum of the coefficients of the lagged dependent variables. The numbers in parentheses are p-values. The p-value is the probability that the sum of the estimated Granger-causality coefficients is equal to zero. Source: Authors' calculations based on data in the World Bank's World Saving Database. Table 5. Causality between Foreign Aid and Growth and Foreign Aid and Investment in Sub-Saharan Africa, 1979-95 Foreign aid-- gross Gross domestic investment -k Growths Model domestic investment foreign aid Foreign aid- growtb foreign aid Pooled model Sum of coefficients 0.0067 0.1939 0.0031 -0.1673 (0.9999) (0.9999) (0.9999) (0.9999) Long-run coefficient 0.1510 0.9336 0.0039 -1.1164 (0.9999) (0.9999) (0.9999) (0.9999) Sample size 612 619 608 607 Fixed-effects model Sum of coefficients -0.0263 0.3802 0.0142 -0.3450 (0.087) (0.020) (0.431) (0.266) Long-run coefficient -0.0909 0.6055 0.0132 0.5446 (0.101) (0.0221) (0.435) (0.264) Sample size 612 619 608 607 Note: Sum is the sum of the Granger-causality coefficients in the estimation equations. The long-run coefficient is derived by dividing this sum by 1 minus the sum of the coefficients of the lagged dependent variables. The numbers in parentheses are p-values. The p-value is the probability that the sum of the estimated Granger-causality coefficients is equal to zero. Source: Authors' calculations based on data in the World Bank's World Saving Database. Table 6. Fixed-Effects Granger Causality Results for South Asia and Latin America and the Caribbean, 1970-95 Saving- Investment-* Saving-4 Growthb- Investment --- Growthb- investment saving growth saving growth investment South Asiaa Sum of coefficients -0.0692 0.3260 0.0277 0.1391 0.1521 0.4865 (0.370) (0.000) (0.590) (0.209) (0.000) (0.000) Long-run coeffient -0.3150 0.6547 0.0201 0.6460 0.1560 1.7995 (0.476) (0.000) (0.589) (0.211) (0.146) (0.000) Sample size 406 402 390 379 406 400 a1. Latin America and the Caribbeanb Ch Sum of coefficients 0.1955 0.0774 -0.0405 0.2636 0.1287 0.1 857 (0.000) (0.022) (0.380) (0.002) (0.0191) (0.001) Long-run coefficient 0.398 0.1944 -0.0452 0.7171 0.0116 0.5752 (0.000) (0.250) (0.389) (0.000) (0.0207) (0.001) Sample size 575 568 555 537 568 557 Note: Sum is the sum of the Granger-causality coefficients in the estimation equations. The long-run coefficient is derived by dividing this sum by 1 minus the sum of the coefficients of the lagged dependent variables. The numbers in parentheses are p-values. a. South Asia includes Bangladesh, Bhutan, Cook Island, India, Maldives, Myanmar, Nepal, Pakistan, and Sri Lanka. b. Latin America and the Caribbean includes Antigua and Barbuda, Argentina, the Bahamas, Barbados, Belize, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominica, Dominican Republic, Fcuador, El Salvador, Grenada, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname, Trinidad and Tobago, Uruguay, and Venezuela. Source: Authors' calculations based on data in the World Bank's World Saving Database. Elbadawi and Mwega 427 Our specification is broadly similar to other empirical models in the literature (such as Edwards 1995; Masson, Bayoumi, and Samiei 1995; and Mwega 1996). Private saving is assumed to depend on five types of variables, characterizing the life cycle, fiscal policy, the financial sector, macroeconomic stability, and open- ness of the economy.7 The estimation results are based on pooled ordinary least squares regressions, which include regional dummies for Sub-Saharan Africa and for Latin America and the Caribbean to determine whether the private saving rate in the two regions differs systematically from that in South and East Asia (table 7).8 The shift dummy variable for Africa is not significant at the 10 percent level, while that for Latin America is negative and significant at the 1 percent level. In this database, therefore, the saving rate in Africa is not significantly different from that in Asia after taking into account the various fundamentals. Gross private disposable income (GPDI) per capita has a positive and generally significant coefficient (0.033) among these developing countries. Growth of GPDI per capita also has a positive and highly significant coefficient, suggesting the existence of a virtuous cycle between private saving and growth. A 1 percent increase in economic growth raises the private saving rate by 0.053-0.081 per- centage point. Similarly, growth of the terms of trade has a positive and signifi- cant coefficient, with a 1 percent improvement in the terms of trade increasing the private saving rate by 0.11-0.12 percentage point. An improvement in the terms of trade increases income and hence saving, especially if the improvement is expected to be transitory. This effect is particularly important in Sub-Saharan Africa, where exports are limited to a few primary commodities, which are sold in highly volatile markets. This result is corroborated by other empirical studies (such as Ostry and Reinhart 1992; Bevan, Collier, and Gunning 1992; and Azam 1996). Bevan, Collier, and Gunning (1992) analyze the impact of the 1976-77 coffee boom on rural saving in Kenya. Proceeds from this boom-caused by a frost in Brazil-were fully passed on to farmers, who saved about 60 percent of them. Two demographic variables-the youth dependency ratio (the ratio of the population under 15 years to the population 15-64 years) and urbanization (the 7. A more detailed discussion of the channels through which these determinants affect private saving is contained in Mwega (1996). 8. An expanded version of this article (Elbadawi and Mwega 1999) contains regression results based on fixed-effects and GMM models. Controlling for fixed effects substantially changes the precision of several coefficients because some variables, such as demographics, change slowly over time and because of multicollinearity among variables. Income per capita, the youth dependency ratio, and urbanization, for example, no longer have significant coefficients. Among the financial variables, the coefficient on the ex post real, rate of interest is no longer significant, although the spread between the lending and the deposit interest rates and the degree of financial depth (M2Y) now have positive and significant coefficients. The ratio of private sector credit to gross private disposable income (GPDI) has a larger negative and significant coefficient (changing from -0.22 to -0.27). In addition, inflation is no longer significant. Differencing the variables in the GMM-IV estimation reduces the precision of the pooled point estimates even further. Only three variables remain significant: growth in the terms of trade (the coefficient of 0.06 is in the range estimated earlier), the public savings rate (with a larger offset coefficient of about 1), and the ratio of private sector credit to GPDI (with a reduced coefficient of -0.17). 428 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 7. Private Saving in Developing Countries: Ordinary Least Squares Pooled Regressions with Regional Dummies, 1970-95 Variable (1) (2) Constant -0.3345 -0.3477 (-2.160) (-3.204) Gross private disposable income per capita 0.0327 0.0334 (5.378) (8.131) Growth of gross private disposable income per capita 0.0531 0.0812 (1.715) (2.964) Growth in the terms of trade 0.1129 0.1192 (3.713) (4.285) Real investment rate 0.0447 0.0218 (3.241) (1.900) Ratio of M2 to gross private disposable income 0.0318 (0.583) Rate of public saving -0.6048 -0.6374 (11.202) (13.324) Youth dependency ratio -0.3000 0.2840 (-5.390) (-6.041) Elder dependency ratio 0.7430 (0.0S9) Urbanization index -0.0810 -0.0778 (-2.820) (-3.398) Government consumption 0.6824 0.7659 (7.801) (10.985) Spread between lending and borrowing interest rates 0.0008 (1.265) Ratio of domestic credit to the private sector to gross private disposable income -0.1635 -0.1586 (-4.122) (-5.060) Inflation rate -0.0794 -0.0790 (-4.092) (-4.793) Ratio of foreign aid to gross domestic product -1.0906 -0.8968 (-4.246) (-4.429) Sub-Saharan Africa dummy 0.0461 (1.478) Latin America and the Caribbean dummy -0.0737 -0.0846 (-3.333) (-5.879) Sample size 260 306 Adjusted R2 0.55 0.597 Standard error 0.092 0.090 Note: The numbers in parentheses are p-values. Source: Authors' calculations based on data in the World Bank's World Saving Database. proportion of the urban population in the total population)-have negative and significant coefficients. The life-cycle model predicts a negative relationship be- tween the private saving rate and the dependency ratio, provided that the lifecycle motive for saving to finance retirement is important. Urbanization also may re- duce the saving rate, as precautionary saving associated with the volatility of income in the agricultural sector is reduced. This effect seems to dominate the Elbadawi and Mwega 429 increase in saving arising because urban dwellers may have better access to finan- cial instruments. Of the fiscal policy variables, public saving has a negative and significant ef- fect. An increase in the rate of public saving by 1 percent reduces the private saving rate by about 0.6 percentage point. General government consumption has a positive and significant coefficient of 0.68-0.77. For a given rate of public saving, policies that reduce government consumption expenditures (and reduce tax revenues) have a negative impact on the private saving rate. This result sug- gests that the private sector places positive value on government consumption (Edwards 1995). Turning to the financial variables, the pooled results show that the coefficient on the ex post real rate of interest, which captures financial liberalization (or the absence of financial repression), is positive and significant (0.02-0.04), although the spread between the lending and the deposit interest rates is not significant. Similarly, the degree of financial depth (M2Y) has an insignificant coefficient. The degree to which financial constraints are binding was proxied by domestic credit allocated to the private sector relative to GPDI. The coefficient of this vari- able is negative and significant (-0.16), suggesting that liquidity and borrowing constraints are important among these developing countries. In the category of macroeconomic instability and volatility, the pooled results show. that inflation has a negative and significant coefficient. Inflationary expec- tations lead to a lower real rate of interest and adversely affect private saving through the McKinnon-Shaw mechanism (McKinnon 1973, 1991; Shaw 1973). High inflation also may signal a lack of credibility in government policies, as well as lower expected returns on saving. The estimation suggests that these two ef- fects of inflation more than outweigh its potential effect through the reduction of real wealth (if, because of structural rigidities, it is not fully indexed) and con- sumption, which would lead to an increase in saving. Ending with the open economy variable, the results suggest that the degree of dependence on foreign aid, as measured by the ratio of foreign aid to GPDI, has a negative and significant coefficient, with an almost one-for-one offset relation- ship (that is, a 1 percent rise in foreign aid will lead to a I percent decline in private saving). Although foreign aid traditionally has been regarded as comple- mentary to national saving, a resurgent literature on the macroeconomic effects of foreign aid finds that additional aid is spent mostly on consumption, not in- vestment (Boone 1994 and Obstfeld 1995).9 This means that foreign saving will, on average, act as a substitute for domestic saving by easing liquidity constraints 9. Complementarity can be derived from certain poverty trap models (in which inflows of foreign savings help to unlock these traps) or models in which countries converge to common steady-state income levels (Boone 1994). If incomes converge, then poor countries will save a part of the transfers, since these are temporary. Developing countries also may have a higher marginal productivity of capital, which in- duces them to postpone consumption to the future. Foreign aid may be used to construct infrastructure, leading to a reduction in distortionary taxes and hence promoting economic growth. This assumes that aid is given in a "good" policy environment, in which aid is found to be conducive to growth onlv in such an environment (Burnside and Dollar 1996). But if incomes across countries do not converge, the flow of foreign aid is likely to be perceived as permanent and therefore largely spent on consumption. 430 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 or by inducing Dutch-disease effects. As Africa is the world's main recipient of aid (relative to the size of its economy), the impact of aid on private saving is of utmost importance to both the donor community and African policymakers. III. SAVING PERFORMANCE WITHIN SUB-SAHARAN AFRICA Despite the weak saving performance of Sub-Saharan Africa relative to other regions, it is important to emphasize that Sub-Saharan Africa is not an undiffer- entiated whole. For example, during 1970-95 the median saving rate (as a pro- portion of disposable income) of the high-saving Sub-Saharan African countries was more than 9.8 percent greater than the rate of moderate-saving countries, which in turn was more than 4.7 percent greater than that of the low-saving countries. However, even the average saving rate of the high savers (17.9 per- cent) was much lower than the average rate of the high-performing Asian econo- mies (more than 24.8 percent). Understanding the sources of Asia's superior per- formance as well as the diversity within Sub-Saharan Africa should be an important policy objective. We conduct historical simulations to decompose the differences in saving rates between the high-performing Asian economies and the Sub-Saharan African high and moderate savers (using regression 2 of table 7; see table 8). We carry out the decomposition by computing the contribution to the saving rate of each variable found to be significantly associated with private saving. These shares are derived by multiplying the variable's mean (for each of the four groups) by its estimated coefficient. The differences between the shares of the high-performing Asian econo- mies and the shares of African countries are the sources of the overall difference in saving between the two regions. We also conduct a similar decomposition of the difference between moderate and low savers in Sub-Saharan Africa. These experiments give us some insight into what low-saving Sub-Saharan African coun- tries might do to raise their private saving rates to the level of the moderate savers and what moderate- and high-saving African countries might do to approach the East Asian frontier. Comparing the high-performing Asian economies with the high and moderate African savers, we can make the following points. First, GPDI per capita and the youth dependency ratio emerge as the most important, robust contributors to East Asia's superior saving performance. For example, in 1970-95 the differen- tial GPDI per capita and the youth dependency ratio equaled 18.0 and 6.2 percent in favor of high-performing Asian economies relative to high savers (and 14.4 and 9.2 percent relative to moderate savers). Second, the high aid dependence of moderate savers, and to a lesser degree of high savers, also contributed to East Asia's saving advantage. We estimate that because of this dependence, moderate savers lagged behind the high-performing Asian economies by 3.7 percent in 1970- 95. For high savers this shortfall was much lower, at 1.7 percent. Third, as expected, the first-round effect of East Asia's higher public saving relative to that of Sub-Saharan Africa lessened the saving differential between the Table 8. Savings Performance in Sub-Saharan Africa and the High-Performing Asian Economies, 1970-95 (percentage contribution to differential in savings rates) High-performing Asian economies to High-saving Moderate-saving Moderate-saving Sub- Sub-Saharan Sub-Saharan Sabaran African countries African African to low-saving Sub-Saharan Variable Coefficient, countries countries African countries Gross private disposable income per 0.0334 18.04 14.44 4.16 capita Growth of gross private disposable income per capita 0.0812 0.59 0.44 -0.11 Youth dependency ratio -0.2840 6.24 9.20 -1.06 Urbanization index -0.0778 1.55 -2.20 -0.29 Fiscal policy variables Rate of public saving -0.6374 -2.54 -2.31 0.24 Government consumption 0.7659 -5.85 -3.82 0.50 Financial variables Real interest rate 0.0218 0.06 0.22 -0.05 Ratio of domestic credit to the private sector to gross private disposable income per capita -0.1586 -5.67 -5.34 -1.79 Macroeconomic and external variables Inflation rate -0.0790 0.34 0.57 0.68 Growth in the terms of trade 0.1192 1.05 1.06 0.12 Foreign aid Ratio of foreign aid to gross domestic product -0.8968 1.74 3.68 2.05 Explained difference 12.44 15.93 4.44 Actual difference 7.87 16.34 2.23 Unexplained residual -4.57 0.42 -2.21 Note: The high-performing Asian economies include China, Hong Kong, Indonesia, the Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. The high-Saving Sub-Saharan African countries include Gabon, Seychelles, Liberia, South Africa, Namibia, Mauritius, Lesotho, Kenya, Tanzania, Congo, Togo, and Zimbabwe. The moderate-saving Sub-Saharan African countries include Benin, Burkina Faso, Cameroon, Comoros, Congo, C6te d'lvoire, Ethiopia, Guinea, Malawi, Mali, Mauritania, Niger, Nigeria, Rwanda, Senegal, Sudan, Swaziland, Zambia, and Zimbahwe. The low-saving Sub-Saharan African countries include Botswana, Burundi, Central African Republic, Chad, Gambia, Ghana, Guinea Bissau, Madagascar, Sao Tome and Principe, Sierra Leone, and Uganda. a. The coefficients arc from regression 2 of table 7. Souirce: Authors' calculations based on data in the World Bank's World Saving Database. 432 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 two.10 However, this effect was relatively small (reaching a maximum of -2.5 percent for the high savers) compared with the huge, positive effect of income per capita and the life-cycle variables in favor of saving in the high-performing Asian economies. Higher government consumption in Sub-Saharan Africa lessened the saving differential between Africa and East Asia, compounding the effects of higher public saving in the high-performing Asian economies. Fourth, and perhaps not surprising, the contributions of the financial and macroeconomic variables were small.I' However, the share of private sector credit in GPDI had a large impact (5.7 and 5.4 percent in high and moderate savers, respectively) in favor of Sub-Saharan Africa, whereas the growth of the terms of trade favored private saving in high-performing Asian economies by about 1 percentage point.12 The factors differentiating moderate- and low-saving African countries are broadly similar to those above, with a few differences. Again, the effect of in- come per capita was very important, contributing 4.2 percent to the better saving performance of the moderate savers. This finding underscores the importance of raising income per capita within Sub-Saharan Africa. Unlike the comparisons between Sub-Saharan Africa and the high-performing Asian economies, the higher dependency ratio among the moderate savers re- duces the differential with low savers. Thus these life-cycle variables make a net positive contribution to saving in low savers relative to moderate savers. The net effect of the other life-cycle variables is, however, negligible. This leaves aid de- pendence as the only variable making a relatively appreciable contribution (2.0 percent) to the higher private saving rate in moderate savers. The marginal role of macroeconomic and financial variables in explaining the divergent saving per- formance of the high-performing Asian economies and Sub-Saharan Africa is also corroborated in the comparison between moderate and low savers. Despite the insights into the main reasons that saving diverges across coun- tries, some major gaps remain in the analysis. For example, in comparing the high-performing Asian economies with Sub-Saharan Africa, the cross-country model tends to underestimate saving rates in high- and low-saving African coun- tries, as reflected by fairly large negative unexplained residuals. However, the model tends to slightly overstate saving rates in moderate-saving countries. We may gain further insight into the saving process in Sub-Saharan Africa by analyz- ing country-specific experiences. 10. This analysis is based on partial equilibrium simulations and therefore does not capture the second- round (and positive) effects of public saving on private saving, given public saving's positive influence on growth. Thus this analysis may also overstate the growth effect on private saving. 11. Because macroeconomic variables are part of the economic growth fundamentals, this analysis does not account for their potentially powerful (and positive) second-round effects on private saving. 12. This is despite disregarding the large mean value (6.68) of Hong Kong's ratio of domestic credit to the private sector to GPDI. Elbadawi and Mwega 433 IV. COUNTRY EXPERIENCES IN SUB-SAHARAN AFRICA We analyze three Sub-Saharan African countries, two of which-Kenya and Zimbabwe-are high savers, whereas the third, Botswana, has one of the lowest private saving rates in Sub-Saharan Africa (see table 9). However, Botswana's unusually high public saving rate makes it one of the top savers (in terms of aggregate national saving) in the world. These countries present interesting anoma- lies, which should help to highlight the role of some variables that may not be easily captured by the regression model. Kenya and Zimbabwe The most important reasons for the high private saving rates in Kenya and Zimbabwe are that both countries have relatively dense and diversified financial and banking sectors and have instituted the policy of institutional "forced" sav- Table 9. Savings Performance in Kenya, Zimbabwe, and Botswana Relative to the High-Performing Asian Economies, 1970-95 (percentage contribution to differential in saving rates) Variable Coefficient Kenya' Zimbabwe Botswana Gross private disposable income per capita 0.0334 11.81 3.52 18.32 Growth of gross private disposable income per capita 0.0812 0.63 0.84 -0.01 Youth dependency ratio -0.2840 14.85 11.85 13.42 Urbanization index -0.0778 -2.63 -2.18 -2.64 Fiscal policy variables Rate of public spending -0.6374 -3.65 -3.79 19.47 Government consumption 0.7659 -5.83 -9.09 -26.15 Financial variables Real interest rate 0.0218 0.12 0.12 0.08 Ratio of domestic credit to the private sector to gross private disposable income -0.1586 -5.15 -6.94 -4.59 Macroeconomic and external variables Inflation rate -0.0790 0.53 0.50 0.36 Growth in the terms of trade 0.1192 0.74 1.15 0.03 Foreign aid Ratio of foreign aid to gross domestic product -0.8968 1.64 0.89 2.94 Explained difference 13.06 -3.13 21.23 Actual difference -1.01 3.71 11.26 Unexplained residual -14.07 6.85 -9.98 Note: The high-performing Asian economies include China, Hong Kong, Indonesia, the Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. South Asia includes Bangladesh, Bhutan, Cook Island, India, Maldives, Myanmar, Nepal, Pakistan, and Sri Lanka. a. Kenya savings data are unadjusted for capital gains and losses from inflation. Source: Authors' calculations based on data in the World Bank's World Saving Database. 434 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 ing. The policy of forced saving is based on an elaborate ensemble of regulations for the appropriation of private saving to finance fiscal expenditures.13 Until re- cently, especially in Zimbabwe, the policy also was reinforced by an extensive and effective system of foreign exchange and import controls. The mechanism for mobilizing forced saving in Zimbabwe is succinctly described in Elbadawi and Schmidt-Hebbel (1991: 7-8): To generate a surplus which finances 100 percent or more of the public deficit since 1986/87, the private sector raised significantly its saving rate: since 1984/85 it exceeds 20 percent of GDP and finances more than 100 percent of the economy's gross domestic investment. This private saving rate is very high for a developing economy-a counterpart is a very low private consumption rate, barely exceeding 50 percent of GDP during the second half of the 1980s. High private saving channeled through Zimbabwe's developed financial system to the public sector is probably a result of re- strictions on private consumption (particularly imported consumer durables) and on formal or illegal capital outflows. Perhaps to a lesser extent than Zimbabwe, Kenya also accorded a fair degree of protection to domestic industry until the early 1990s. The history of Zimbabwe's banking and financial system dates back to the 1 890s. The system is composed of nine institutions, including commercial banks, accepting houses, discount houses, finance houses, post office saving funds, build- ing societies, insurance companies, pension and provident funds, and saving clubs. The robustness and diversity of the Zimbabwean financial sector has enabled it to mobilize small saving from a large number of clients from rural communities and small towns, through small institutions, such as post office saving funds, which realized their largest expansion during the first half of the 1980s. The assets of these funds (as a proportion of total assets of the financial sector) rose from 13 percent in 1983 to more than 20 percent in 1986-almost half the assets of commercial banks, which are the most important institutions for mobilizing private saving in Zimbabwe. Commercial banks and other more formal financial institutions also have managed to mobilize large saving from the corporate and commercial sectors (Moyo 1996). The financial sector in Kenya also is considered to be one of the deepest and most diversified in Sub-Saharan Africa. A measure of its success in mobilizing private saving is reflected in its contribution to gross domestic saving, which averaged about 40 percent during the second half of the 1980s. Private nonfinan- cial corporations also are important contributors, accounting for an additional 36 percent of gross domestic saving during the same period (Kwasa 1992). Despite the superior saving performance of Kenya and Zimbabwe relative to that of most other African countries, two important questions arise: what might 13. For example, insurance companies in Zimbabwe are required to invest at least 60 percent of their funds in prescribed government securities (Moyo 1996). Elbadawi and Mwega 435 these countries do to raise their private saving rates to the levels of the high- performing Asian economies, and what is the likely effect of trade and financial liberalization-introduced in both countries since the late 1980s-on private sav- ing? The answer to the first question seems fairly straightforward and is readily obtained from the pooled saving model. The two countries must increase in- come per capita, for which both macroeconomic stability and competitiveness may be critical. Moreover, the two countries need to reduce their age-dependency ratios. Financial liberalization appears to have no significant positive impact on sav- ing, which is consistent with the cross-country evidence of this article and coun- try-specific evidence for a range of African countries. Whereas Kenya's saving rate remained stable between the second half of the 1980s and the first half of the 1990s, Zimbabwe's saving rate fell in the 1990s (Ndung'u 1997). Why financial liberalization may not have yet led to a surge in private saving in these two coun- tries may have to do with their-especially Zimbabwe's-failure to further fi- nancial deepening or to increase the efficiency of the financial sector. Although in Kenya the ratio of M2 to GDP increased from 38 percent in 1980 to around 50 percent in 1994, it declined slightly from more than 50 to 48 percent in Zimba- bwe. More compelling evidence, however, is the perverse rise in the spread be- tween the lending and the deposit rate (an indicator of inefficiency in the finan- cial market) during financial liberalization. In Kenya it increased from more than 5 percent in 1989 to more than 18 percent in 1994, and in Zimbabwe it in- creased from 1.4 percent to more than 8 percent in 1994 (Ndung'u 1997). In the final analysis it seems that the ability of the two countries to maintain relatively high levels of private saving may reflect that both the policy of forced saving (arising because trade liberalization has been only partial so far) and dense and relatively robust financial markets are still operating. Therefore, adjustment to a higher voluntary saving rate-and more efficient investment and therefore higher growth-has yet to begin in the two countries. Botswana Botswana may be one of the few countries that have been able to achieve high and sustained growth despite persistently low private saving rates. Clearly, given its growth record, Botswana's unusually high public saving rate has more than compensated for its low private saving rate. On average, the private saving rate in Botswana during 1970-95 lagged behind the median rate in Sub-Saharan Af- rica by about 3.5 percent.14 In contrast to its record of private saving, Botswana managed to achieve an impressive record of growth, including an average annual per capita growth rate of more than 11 percent during 1969-79 and almost 7 percent during 1980-90. By any standard, Botswana has indeed earned its posi- tion as one of the leading developing countries to make a sustainable transition 14. However, Botswana has closed the gap over time. In 1990-95 its savings rate exceeded the median for Sob-Saharan Africa by 3.6 percent. 436 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 to high growth and low macroeconomic instability (Elbadawi and Schmidt-Hebbel 1997). Public saving constitutes more than 75 percent of Botswana's staggering gross domestic saving rates, which averaged about 42 percent in 1985-89 and more than 33 percent in the first half of 1990s. Government revenue in Botswana was given a tremendous boost by an increase in the production and export of dia- monds. The government collects "between 75 and 82 percent of the profit of the diamond industry through a combination of royalty payments, profit tax, with- holding tax on remitted dividends, and dividends received by virtue of its 50 percent share-holding in the diamond mining company Debswana" (Hope 1997: 107). However, the experiences of other well-endowed resource-based econo- mies suggest that Botswana's unusual record of public saving owes more to pru- dent fiscal management than to the sheer magnitude of its wealth. Naturally, the key test of Botswana's macroeconomic prudence was the chal- lenge of managing the boom in the mining sector (diamonds). In the words of Norberg and Blomstrom (1993: 176-77), The spending effects of the diamond boom have been small, something which is mainly due to government policy. Since the diamond industry is state owned, the government controls the revenues. The revenues have mainly been used to promote national income, rather than to subsidize different sectors (such as import-substituting manufacturing) and support various interest groups. To maximize national income, a large proportion of the revenues from diamonds have been invested, not in Botswana, but in for- eign banks and firms. By sterilizing revenues abroad and executing a non- expansive monetary policy at home, the inflationary pressure has dimin- ished and the negative effects of the diamond boom have been reduced. Despite its enormous success in growth and the management of its nonrenew- able resource-based economy, Botswana must substantially raise its private sav- ing rate without necessarily reducing its current public saving. There are three arguments in support of this view. First, since the mainstay of the economy is a nonrenewable resource, intertemporal welfare maximization suggests that opti- mal saving rates-consistent with maintaining current consumption rates-could be as high as 50 percent of GDP (Elbadawi and Majd 1992). An optimal saving rate (relative to GDP) has been computed at about 37 percent of GDP for countries, such as Egypt, Turkey, and Indonesia, that only partially depend on a nonrenew- able resource, such as oil. For countries that are more heavily dependent on oil, such as Saudi Arabia and the United Arab Emirates, the optimal saving rate could be as high as 60 percent. Botswana clearly belongs to the latter group."5 There- 15. The framework used to derive the optimal rate is a model of a small open economy with two assets (such as oil and an asset other than oil). We can solve for the optimal savings rate in this model as an explicit function of the life span of the reserve of the nonrenewable resource, expected future prices of the resource, the real returns of investment, and the desirable savings rate in the post-resource era. Examples of this literature include Dervis, Martin, and van Wijnbergen (1984); Gelb (1985a, 1985b); Farzin (1990); and Askari (1990). Elbadawi and Mwega 437 fore, a higher private saving rate would be required to help bridge the gap be- tween the prevailing national saving rate and the welfare-maximizing rate for an economy such as Botswana's. Second, despite Botswana's enormous success in sustaining high growth and prudently managing its mineral sector, a new body of evidence suggests that this growth has not been equitable and has coexisted with poverty and unemploy- ment (Hope and Edge 1996). Thus an increasing share of public saving should be spent on the social sectors, with the implication that, to sustain growth, more private saving will be needed to support other vital investments. Further, given the volatility of the mining sector, a high and stable rate of private investment should always be a major policy objective. Even though the partial crowding out effect of the high rate of public saving in Botswana unavoidably reduces private saving, other policy-induced factors also have contributed to this outcome.16 One factor is financial repression-even af- ter rising significantly, real interest rates in Botswana were still negative during the 1990s at -0.3 percent (Ndung'u 1997). More important, however, is the lack of deep and diversified financial and banking sectors. The need for raising private saving is now well understood in Botswana. For example, the private sector is calling for a major drive toward privatizing parastatals-through share offerings-as the most effective method for mobiliz- ing private saving (Hope 1997). V. CONCLUSIONS Sub-Saharan Africa's growth performance over the past two decades has been poor relative to that of other regions, especially the high-performing Asian econo- mies. This poor performance also has been reflected in low saving and invest- ment rates, both of which declined substantially in the 1970s and the 1980s, recovering only partially in the 1990s. A worrisome feature of saving and invest- ment in Sub-Saharan Africa is the region's heavy dependence on foreign saving, mostly overseas development assistance, to finance the gap between investment and saving. This article focused on the determinants of private saving in Sub- Saharan Africa, with the objective of explaining the region's dismal saving be- havior and ultimately identifying policies that could reverse the decline in saving. Although raising saving rates alone is not sufficient for achieving sustained growth, it does appear to be a necessary condition. Results of a fixed-effects estimation suggest that in Sub-Saharan Africa causal- ity runs from growth to investment (and perhaps to private saving), which is consistent with international evidence (see, for example, Carroll and Weil 1993). A rise in the saving rate also seems to Granger-cause an increase in investment. However, foreign aid Granger-causes a reduction in both saving and investment. 16. According to model simulations (not reported) this effect causes private saving in Botswana to be 12 percent less than the median savings rates in Sub-Saharan Africa for the 1990-95 period. 438 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Investment also Granger-causes an increase in foreign aid, so that African coun- tries that increase investment receive more aid. We can make several broad conclusions from the empirical analysis of private saving in Sub-Saharan Africa and other regions over 1970-95. First, including regional dummies for Sub-Saharan Africa and Latin America and the Caribbean in the regressions to test whether the two regions differ systematically from Asia reveals that the Africa shift dummy is not significant at the 10 percent level, while that of Latin America is negative and significant at the 1 percent level. Second, GPDI per capita, the growth of GPDI per capita, and the growth in the terms of trade have positive and significant coefficients. The youth dependency ratio and urbanization ratio have negative and significant coefficients. Third, the public saving coefficient is negative and highly significant, with an offset coefficient of about -0.6 percent (a rise in public saving by 10 percent will reduce private saving by 6 percent). General government consumption, in con- trast, has a positive and significant coefficient, indicating that the private sector places positive value on government consumption (Edwards 1995). The coeffi- cient on the ex post real rate of interest is positive and significant, while the coefficient on the private sector credit ratio is negative and significant. Both in- flation and foreign aid have negative and significant coefficients. We used the regression results to analyze the main factors that drive saving for three subgroupings from Sub-Saharan Africa (high, moderate, and low savers) and three countries (Kenya, Zimbabwe, and Botswana). The analysis is based on historical simulations of the sources of the differences in saving rates between high-performing Asian economies and each of the Sub-Saharan African groups. Income per capita and the youth dependency ratio emerge as the most important contributors to the superior saving performance in high-performing Asian econo- mies relative to that in high and moderate African savers. The high dependence on aid among moderate savers and to a lesser degree among high savers also has given the high-performing Asian economies an advantage in terms of saving per- formance. However, Asia's higher public saving rate relative to that of the two Sub-Saharan African groups reduced the saving differential, although this effect is relatively small compared to the huge effect of the life-cycle variables, which work in the opposite direction. Moreover, higher government consumption in Sub-Saharan Africa (especially among high-saving African countries) lowers the differential, compounding the effect of higher public saving in the high- performing Asian economies. The contributions of financial and macroeconomic variables are small, except for the private sector credit ratio, which favors Sub- Saharan Africa. In comparing moderate savers to low savers in Sub-Saharan Africa, the pat- terns are broadly similar. But there are some differences. First, the effect of in- come per capita is very important, but the higher dependency ratio among the moderate savers reduces the differential with the low savers. Thus life-cycle vari- ables make a net positive contribution to saving in low-saving countries relative to moderate-saving countries. Second, dependence on aid contributes apprecia- bly to the higher private saving rate among moderate savers relative to low sav- Elbadawi and Mvega 439 ers. Third, the marginal role of macroeconomic and financial variables in ex- plaining the divergent saving performance of high-performing Asian economies and Sub-Saharan Africa also is corroborated in the comparison between moder- ate and low savers. Finally, this article provided further insight into the saving process in Sub- Saharan Africa by analyzing the experiences of Kenya, Zimbabwe, and Botswana. DATA APPENDIX: VARIABLE DEFINITIONS AND DATA SOURCES Most of the data used in sections II and III were derived from the World Bank's World Saving Database. * GPDI is measured by gross national disposable income minus gross public disposable income. The latter was estimated by public saving plus public consumption. * GPRS is a residual measure of private saving (gross national saving minus central government saving) in the database adjusted for capital gains or losses due to inflation. This adjustment did not change the results substan- tially. The measure of private saving was scaled by gross private disposable income. The other measures of saving in the World Bank's data that net out saving by other public sector entities and, therefore, provide a better sepa- ration between private and public saving, had too many missing observa- tions, particularly in the Sub-Saharan Africa sample, to permit meaningful analysis. * LRPYC is the log of real per capita gross private disposable income. * ALRYPC is the change in LRYPC (that is, growth in real per capita GPDI). * ALTOT is the change in log terms of trade. * RDRATE is the real deposit rate of interest, expressed as log(1 + p) to reduce the range of variation. * M2Y is M2 as a proportion of GPDI. - GPS is central government saving as a proportion of GPDI. These saving, as in GPRS, are controlled for capital gains or losses due to inflation. - D 15 is the youth dependency ratio, measured by the ratio of the population under 15 years to the population 15-64 years. * D65 is the elder dependency ratio, measured by the ratio of the population over 64 years to the population 15-64 years. * URBAN is the degree of urbanization, measured by the proportion of ur- ban population to the total population. * GCNYis general government consumption expenditures as a proportion of GPDI. * RSPREAD is the spread between lending and borrowing interest rates given in the database. * DCPCY is the proportion of the stock of private sector domestic credit to GPDI. * INFL is inflation, expressed by log (1 + 7t) to reduce the range of variation. 440 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 AIDY is dollar foreign aid as a proportion of dollar GPDI. The data for the dollar foreign aid were derived from a compilation by Chang and others (1998). They define effective development assistance as the sum of grant- equivalents and grants, excluding technical assistance. To derive the dollar GPDI, the exchange rate given in the World Saving Database was used. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Allerano, Manuel, and Stephen Bond. 1991. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equation." Review of Eco- nomic Studies 58(2):277-98. Askari, Hossein. 1990. Saudi Arabia's Economy: Oil and the Search for Economic Devel- opment. Contemporary Studies in Economic and Financial Analysis 67. Greenwich, Conn.: JAI Press. Attanasio, 0. P., Lucio Picci, and Antonello Scorcu. 2000. "Saving Growth and Invest- ment. Review of Economics and Statistics 82(2): 182-211. Azam, Jean-Paul. 1996. "Saving and Interest Rates: The Case of Kenya." Savings and Development. 20(1): 33-44. Bevan, David, Paul Collier, and Jan W. Gunning. 1992. "Anatomy of a Temporary Trade Shock: The Kenya Coffee Boom of 1976-79." Journal of African Economies 1(2): 271-305. Boone, Peter. 1994. "The Impact of Foreign Aid on Savings and Growth." London School of Economics, Centre for Economic Performance, London. Processed. Bosworth, Barry. 1993. Saving and Investment in an Open Economy. Washington D.C.: Brookings Institution. Burnside, Craig, and David Dollar. 1996. Aid, Policies, and Growth. Washington, D.C.: World Bank. Carroll, Chris, Jody Overland, and David Weil. 1994. "Savings and Growth with Habit Formation." Division of Monetary Affairs, Federal Reserve Board, Washington, D.C. Processed. Carroll, Chris, and David Weil. 1993. "Savings and Growth: A Reinterpretation." NBER Working Paper 4470. National Bureau of Economic Research, Cambridge, Mass. Processed. Chang, Charles C., Eduardo Fernandez-Arias, and Luis Serven. 1998. "Measuring Aid Flows." World Bank, Development Economics Division, Washington, D.C. Processed. Cole, H. L., G. J. Mailath, and Andrew Postlewaite. 1992. "Social Norms, Savings Be- havior, and Growth." Journal of Political Economy 100(6): 1092-125. Collier, Paul. 1995. "The Marginalization of Africa." International Labour Review 134(4- 5): 134(4-5): 541-57. .1998. "The Role of the State in Economic Development: Cross-Regional Experi- ences." Journal of African Economies 7(2): 38-76. De Long, J. Bradford, and Lawrence Summers. 1993. "How Strongly Do Developing Economies Benefit from Equipment Investment?" Journal of Monetary Economics 32(3): 395-416. Elbadawi and Mwega 441 Dervis, Kemal, Richard Martin, and Sweder van Wijnbergen. 1984. Policy Analysis of Shadow Pricing, Foreign Borrowing, and Resource Extraction in Egypt. World Bank Staff Working Paper 622. Washington, D.C.: World Bank. Easterly, William, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Division." Quarterly Journal of Economics 112(4): 1203-50. Easterly, William, and Sergio Rebelo. 1993. "Fiscal Policy and Economic Growth: An Empirical Investigation." Journal of Monetary Economics 32(3):417-58. Edwards, Sebastian. 1995. "Why Are Latin America's Saving Rates So Low? An Interna- tional Comparative Analysis." Department of Economics, University of California, Los Angeles. Processed. Elbadawi, Ibrahim, and Nader Majd. 1992. "Managing a Nonrenewable Resource: Sav- ings and Exchange Rate Policies in Bahrain." Policy Research Working Paper 1134. World Bank, Transition and Macro-adustment Division and International Trade Divi- sion, Washington, D.C. Processed. Elbadawi, Ibrahim, and Francis Mwega. 1999. "Can Africa's Saving Collapse Be Re- verted?" World Bank, Development Research Group, Washington, D.C. Processed. Elbadawi, Ibrahim, and Benno Ndulu. 1996. "Long-Term Development and Sustainable Growth in Sub-Saharan Africa." In Mats Lundahl and Benno Ndulu, eds., New Direc- tions in Development Economics: Growth, Environmental Concerns, and Governance in the 1990s. London: Routledge. Elbadawi, Ibrahim, Benno Ndulu, and Njuguna Ndung'u. 1997. "Economic Policy, Risk, and Uncertainty as Determinants of Investments in Sub-Saharan Africa." Paper pre- sented at the tenth anniversary of the Centre for the Studies of the African Economies, University of Oxford, March. Processed. Elbadawi, Ibrahim, and Klaus Schmidt-Hebbel. 1991. "Macroeconomic Adjustment to Oil Shocks and Fiscal Reform: Simulations for Zimbabwe, 1988-95. " Policy Research Working Paper 772. World Bank, Country Economic Department, Washington, D.C. Processed. - 1998. "Macroeconomic Policies, Instability, and Growth across Developing Re- gions." Journal of African Economies (supplementary edition). Volume 7, Supplement 2: December; pp. 116-168. Farzin, Y. H. 1990. "Optimal Saving and Investment Policies for the Petroleum-based Economy of the United Arab Emirates." World Bank, Washington, D.C. Processed. Feldstein, Martin, and Phillipe Bacchetta. 1991. "National Saving and International In- vestment." In B. Douglas Bernheim and John Shoven, eds., National Saving and Eco- nomic Performance. Chicago: University of Chicago Press. Feldstein, Martin, and Charles Horioka. 1979. "Domestic Savings and International Capital Flows." Economic journal 90(3):314-29. Fershtman, Chiam, and Yoram Weiss. 1993. "Social Status, Culture, and Economic Per- formance." Economic Journal 103(July):946-59. Gelb, Alan. 1985a. "Are Oil Windfalls a Blessing or a Curse? Policy Exercise with an Indonesia-Like Model." DRD Discussion Paper 135, World Bank, Development Re- search Department, Washington, D.C. Processed. . 1985b. "The Impact of Oil Windfalls: Comparative Statics with an Indonesia- Like Model." DRD Discussion Paper 133, World Bank, Development Research Depart- ment, Washington, D.C. Processed. 442 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Gordon, Roger, and Lars Bovenberg. 1994. "Why is Capital So Immobile Internation- ally? Possible Explanations and Implications for Capital Income Taxation." Working Paper 4796. NBER 4796:1-30. Hadjimichael, Michael, and Dhaneshwar Ghura. 1995. "Public Policies and Private Sav- ings and Investment in Sub-Sahara Africa: An Empirical Investigation." IMF Working Paper WP/95/19. International Monetary Fund, Washington, D.C. Processed. Harrigan, Frank. 1995. "Saving Transitions in South East Asia." Department of Eco- nomics, University of Oxford, Processed. Holtz-Eakin, Douglas, Whitney Newey, and Harvey S. Rosen. 1988. "Estimating Vector Autoregressions with Panel Data." Econometrica 56(6): 1371-395. Hope, K. R. 1997. "Growth, Savings and Investment in Botswana." Savings and Devel- opment 21(2): 195-210. Hope, K. R., and W. A. Edge. 1996. "Growth with Uneven Development: Urban-Rural Socio-Economic Disparities in Botswana." Geoforum 27(1). King, Robert, and Ross Levine. 1994. "Capital Fundamentalism, Economic Development, and Economic Growth." Carnegie Rochester Series on Public Policy 40(0):259-92. Kwasa, Shadrack 0. 1992. "Kenya." InJ. H. Frimpong Ansah and Barbara Ingham, eds., Saving for Economic Recovery in Africa. London: James Curry. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Lucas, Robert E. Jr. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22(July): 22(1):3-42. Masson, P. R., Tamim Bayoumi, and Hossein Samiei. 1995. "Saving Behavior in Indus- trial and Developing Countries." IMF Working Paper WP/95/51. International Mon- etary Fund, Washington, D.C. Processed. McKinnon, Ronald. 1973. Money and Capital in Economic Development. Washington, D.C.: Brookings Institution. 1. 991. The Order of Economic Liberalization: Financial Control in the Transi- tion to a Market Economy. Baltimore, Md.: Johns Hopkins University Press. Moyo, Sam. 1996. "Trends of Savings in Financial Intermediaries in Zimbabwe." In J. H. Frimpong Ansah and Barbara Ingham, eds., Saving for Economic Recovery in Africa. London: James Curry. Mwega, Francis M. 1996. "Saving in Sub-Sahara Africa: A Comparative Analysis." Pa- per presented at an African Economic Research Consortium (AERC) Plenary, Nairobi, May. Processed. Mwega, Francis M., Njuguna Mwangi, and S. M. Ngola. 1990. "Real Interest Rates and the Mobilization of Private Savings in Africa: A Case Study of Kenya." AERC Research Paper 2. AERC, Nairobi. Processed. Ndung'u, Njuguna. 1997. "The Impact of Financial Sector Liberalization on Savings, Investment, Growth, and Financial Development in Anglophone Africa." African De- velopment Review 9(1):20-51. Norberg, Helene, and Magnus Blomstrom. 1993. "Dutch Disease and Management of Windfall in Botswana." International Economics and Geography. 111:162-78. Obstfeld, Maurice. 1995. "Effects of Foreign Resource Inflows on Saving: A Method- ological Overview." World Bank, Washington, D.C. Processed. Ostry, Jonathan D., and Carmen Reinhart. 1992. "Private Savings and Terms of Trade Shocks: Evidence from Developing Countries." IMF Staff Papers 39: 495-517. Elbadawi and Mwega 443 Rodrik, Dani. 1998. "Why Is Trade Reform So Difficult in Africa." Journal of African Economies 7(1): 43-69. Roemer, Paul. 1986. "Increasing Returns and Long-Run Growth. "Journal of Political Economy 94(5):1002-37. Schmidt-Hebbel, Klaus, 1996. "Fiscal Adjustment and Growth: In and Out of Africa." Journal of African Economies 5(3):7-59. Schmidt-Hebbel, Klaus, and Luis Serven. 1996. "Saving and Investment: Paradigms, Puzzles, Policies." The World Bank Research Observer 11(1):87-117. Schmidt-Hebbel, Klaus, Steven B. Webb, and Giancarlo Corsetti. 1992. "Household Sav- ing in Developing Countries: First Cross-Country Evidence." The World Bank Eco- nomic Review 6(3):529-47. Serven, Luis. 1997. "Irreversibility, Uncertainty, and Private Investment: Analytical Is- sues and Some Lessons for Africa." Journal of African Economies 6(3): 229-68. Shaw, Edward. 1973. Financial Deepening in Economic Development. New York: Ox- ford University Press. Solow, Robert. 1956. "A Contribution to the Theory of Economic Growth." Quarterly Journal of Economics 70(February): Vol. 70(1) pp. 65-94. Summers, Lawrence. 1988. "Tax Policy and International Competitiveness." In Jeffrey Frankel, ed., International Aspects of Fiscal Policies. Chicago: University of Chicago Press. Tobin, James. 1967. "Life-cycle Saving and Balanced Growth. "In William Fellner, ed., Ten Economic Studies in the Tradition of Irving Fisher. New York: John Wily and Sons. Williamson, John. 1994. "The Washington Consensus Revisited." In Louis Emmerij, ed., Economic and Social Development in the Twenty-First Century." Baltimore, Md.: Johns Hopkins University Press. World Bank. 1994. Adjustment in Africa: Reforms, Results and the Road Ahead. New York: Oxford University Press. Young, Alwyn. 1994. "The Tyranny of Numbers: Confronting the Statistical Evidence of East Asian Growth Experience." NBER Working Paper 4680. National Bureau of Eco- nomic Research, Cambridge, Mass. Processed. Zou, Heng-Fu. 1993. "The Spirit of Capitalism, Savings, and Growth." World Bank, Development Research Group, Washington, D.C. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 445-55 The Saving Collapse during the Transition in Eastern Europe Cevdet Denizer and Holger C. Wolf Almost all of the transition economies in Eastern Europe and the former Soviet Union experienced a severe decline in their national saving rates. The saving collapse could be explained by the elimination of involuntary saving, a feature of central planning, or by a change in equilibrium saving reflecting the new economic circumstances following the end of socialism. The predicted saving rates of market economies with the same funda- mentals as the transition economies before the transition are computed to test for the presence of involuntary saving. The results provide some support for the hypothesis of consumption smoothintg. Also considered is whether differences in the extent of liberal- ization affected saving rates in the cross section of transition economies. This is found to be the case: greater liberalization is association with lower saving with a one-year lag. To the extent that liberalizationz is associated with future growth, this finding is consis- tent with smoothing in the face of output evolving along a J-curve. The past decade witnessed wrenching changes in the transition economies of Central and Eastern Europe. One of the puzzling features of the transition has been the precipitous plunge of measured saving rates-from around 30 percent before the transition to less than 20 percent afterward. This article examines alternative explanations of this saving collapse.1 The plunge can be decomposed into three parts: the change due to movements in the determinants of saving, the change due to movements in the equilibrium elasticities of saving with respect to at least some of the determinants, and the elimination of disequilibrium or "involuntary" saving during the transition. We examine the evidence on all three parts for the European transition economies in 1989-95. We first compute hypothetical equilibrium saving rates at the begin- ning of the period, based on the assumption that saving elasticities estimated for market economies at comparable levels of development can be applied to the transition economies. The difference between the actual rates and these hypo- thetical equilibrium rates provides a measure of involuntary saving. 1. See also Conway (1995); Deaton (1995); Masson, Bayoumi, and Samici (1995); and Portes (1986). Cevdet Denizer is with the Europe and Central Asia sector unit at the World Bank, and Holger C. Wolf is with the Department of Economics at George Washington University and the National Bureau for Economic Research. Their e-mail addresses are cdenizer@worldbank.org and holger.wolf@mailexcite.com. This article is part of the World Bank research project "Saving Across the World." The authors thank Eduardo Borenzstein, Norman Loayza, Ratna Sahay, Klaus Schmidt-Hebbel, and Luis Serv6n for many useful suggestions. Nihal Bayraktar and Nikolay Guerguiev provided extremely able research assistance. (D 2000 The International Bank for Reconstruction and Development / THE WORLD BANK 445 446 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 For the later transition years it is reasonable to assume that price liberalization eliminated any disequilibrium saving. Thus additional changes in saving rates must be due to changes in either the determinants of saving or saving elasticities. We compare saving elasticities in transition and market economies and explore the causes behind differences in saving across transition economies. I. DATA ON SAVING DURING THE TRANSITION The sample includes 10 transition economies from Eastern and Central Eu- rope (Albania, Bulgaria, Croatia, Czech Republic, FYR Macedonia, Hungary, Poland, Romania, Slovak Republic, and Slovenia), the 3 Baltic states (Estonia, Latvia, and Lithuania), and the 12 non-Baltic successor states of the Soviet Union (Armenia, Azerbeijan, Belarus, Georgia, Kazakstan, Kyrgyz Republic, Moldova, Russia, Tajikistan, Turkmenistan, Ukraine, and Uzbekistan). Our focus is the ratio of gross domestic saving (GDS) to gross domestic product (GDP). We derive GDS residually from the current account deficit and gross domestic investment using internally consistent World Bank national account statistics for all components. In the period before the transition (defined as 1989 for Central Europe and 1990-91 for the former Soviet Union and Baltic states) saving rates were among the highest in the world, averaging about 30 percent (see table 1).2 They declined sharply to about 10 percent in the early transition years and increased slightly in the later years. The saving data for transition economies suffer from two potential problems. First, it is possible, indeed likely, that the national account statistics capture only a subset of saving activity. If both expenditure and income data are measured incorrectly, the net effect on saving is ambiguous. Second, the nominal data are deflated by price indexes that exclude activity in the informal sector. Depending on the size of the black-market spread, the pretransition price level is under- stated, and thus saving (and the monetary overhang defined with respect to the measured price level) is overstated (Dornbusch and Wolf 2000). II. INVOLUNTARY SAVING The welfare implications of the collapse of measured saving depend crucially on whether pretransition saving was voluntary and stable; voluntary, but driven by expectations of the systemic change-for example, reflecting the expectation of sharp relative price changes and greater availability of goods; or involuntary, reflecting binding limits on consumption spending below desired levels. Of these three explanations, the second commands the least empirical support: saving rates in socialist countries were high throughout the 1980s. 2. The end point of the period before the transition is determined by the beginning of extensive price liberalization. Denizer and Wolf 447 Table 1. Saving Rates in Select Countries of Eastern Europe, 1989-95 Country 1989 1990 1991 1992 1993 1994 1995 Albania 28.9 12.3 -14.2 -75.8 -38.3 -16.8 -7.8 Bulgaria 31.4 22.0 35.8 18.8 10.7 20.9 24.8 Croatia - - 2.2 19.5 13.3 12.0 1.4 Czech Republic 30.6 29.9 36.8 27.4 20.2 20.1 20.2 Hungary 29.9 28.0 18.7 14.9 11.2 15.0 18.9 FYR Macedonia 033.0 23.5 11.0 11.9 11.5 4.1 15.0 Poland 42.7 32.8 18.0 16.7 16.5 16.9 18.3 Romania 29.5 20.8 24.1 23.0 24.0 24.9 22.9 Slovakia 28.5 24.2 28.2 24.1 21.8 28.8 31.6 Slovenia 33.0 32.6 27.4 26.5 20.6 22.6 21.3 Average 32 25 19 11 11 15 17 Average GDP growth 0.6 -6.4 -14.0 -7.4 0.1 3.4 4.3 Armenia 38.2 35.8 20.5 -19.8 -3.7 -19.2 -19.9 Azerbeijan 34.7 31.9 39.4 17.0 -3.2 3.4 3.7 Belarus 35.8 29.3 32.9 33.7 21.4 15.6 22.6 Estonia 25.9 22.3 34.5 29.7 23.5 17.8 25.0 Georgia 25.3 24.9 24.9 1.7 -18.9 -32.7 -9.1 Kazakstan 34.7 31.9 39.4 30.2 21.3 12.3 19.6 Kyrgyz Republic 13.1 4.0 14.0 7.9 4.0 14.0 10.0 Latvia 34.7 38.8 43.5 48.1 25.4 19.5 20.0 Lithuania 25.8 25.2 28.5 17.9 14.5 9.4 14.3 Moldova 34.7 31.9 39.4 15.0 -6.0 0.0 0.0 Russia 34.7 31.9 39.4 38.4 35.0 29.1 25.6 Tajikistan 12.5 13.7 17.3 18.0 -10.9 0.0 18.1 Turkmenistan 34.7 31.9 39.4 21.0 20.0 14.0 12.0 Ukraine - 26.3 28.4 36.4 36.0 19.8 16.8 Uzbekistan 18.2 15.7 23.6 33.8 15.9 24.4 20.0 Average 28.8 26.4 31.0 21.9 11.6 8.5 11.9 Average GDP growth 2.2 -3.2 -8.4 -23.3 -13.2 -13.6 -3.7 - Not available. Source: World Bank data. The possibility of involuntary saving in the period before the transition has been the subject of a lively debate.3 Results have differed sharply, reflecting the high evidentiary hurdles that must be overcome. Involuntary saving can only exist if consumers do not have access to goods or asset markets in which price movements can equate demand and supply (Dornbusch and Wolf 2000).' Conse- quently, the inability to purchase goods at official prices in official stores, as reflected in queues, is a necessary, but not a sufficient, condition to guarantee the presence of involuntary saving. 3. For different viewpoints on these issues see Kornai (1980, 1992), Alexeev (1988), Davis and Charemza (1989), Quant (1989), Ickes (1993), IME (1995), Weitzman (1991), Nove (1991), Easterly and Fischer (1994), and Bennet and Boycko (1995). 4. This definition of involuntary saving has no implications for efficiency; it merely requires con- sumption decisions to be made on the margin without a binding quantity constraint inside the budget constraint. 448 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 We estimate the presence of disequilibrium saving indirectly by comparing the actual saving rates in the period before the transition with the predicted saving rates of market economies with the same fundamentals as the pretransition economies. The difference between the predicted rates and the observed rates measures the extent of involuntary saving. Specifically, suppose that equilib- rium saving for market economies is described by (1) SE = Xi3 + Ui, where i indexes the country, t denotes time, and X denotes a set of relevant saving determinants. Under the dual assumption that, first, equation 1 also de- scribes the pretransition socialist economies had they operated as market econo- mies and, second, that disequilibrium saving in market economies is negligible, we can calculate an indirect estimate of disequilibrium saving, Sit", as: (2) ~~~~~SD' = Sit -SE = Si,t Xitp where Si, denotes observed saving. Estimating the coefficient vector f for a cross section of market economies, and combining it with the observed determinants X of the transition economies, gives an estimate of desired saving in socialist economies. The quality of the prediction depends on the accuracy with which the coefficient vector is mea- sured, the transferability of this vector to transition countries, the explanatory power of the included variables, and the accuracy with which these variables are measured in both market and transition economies. The approach requires that we use the same set of variables for both the mar- ket and the transition subsamples. Given the underlying hypothesis that saving behavior in the socialist economies would not have differed markedly from that of market economies with similar fundamentals if in fact the socialist countries had operated as market economies, we use Edwards' (1996) cross-country re- gression framework for Latin America as the base for our estimation. Since all variables used for the cross-country regression must also be available for the socialist economies, we have to eliminate wealth and effective interest rates, which Edwards uses. The remaining variables include the dependency ratio, the urban- ization ratio, GDP growth, the ratio of broad money (M2) to GDP, inflation, the change in the terms of trade, GDP per capita, and a dummy for military conflict. In the Central European countries, with one exception (the Czech Republic), actual saving rates are either much higher than predicted rates (in Bulgaria, Po- land, and Romania), indicating involuntary saving, or roughly equal to predicted rates (in Hungary and Slovakia; table 2). The findings are thus consistent with the presence of involuntary saving in at least some Central European countries. For the Baltic states and countries of the former Soviet Union the evidence for involuntary saving is strong: in 30 of the 34 cases actual saving exceeds predicted saving. The gap widens substantially in 1990 and 1991, reflecting less a change in actual saving rates than a steep decline in predicted rates. The decline in pre- Table 2. Actual and Predicted Saving Rates in Pretransition Eastern Europe, 1989-91 1989 1990 1991 Country Actual Predicted Difference Actual Predicted Difference Actual Predicted Difference Central Europe 32.1 29.2 2.9 Bulgaria 31.4 22.4 Czech Republic 30.6 39.4 Hungary 29.9 29.4 Poland 42.7 29.6 Romania 29.5 23.9 Slovakia 28.5 30.4 Baltics 28.8 24.9 3.9 28.8 19.2 9.6 35.5 14.0 21.5 Estonia 25.9 24.2 22.3 16.0 34.5 14.7 Latvia 34.7 26.6 38.8 24.9 43.5 15.8 Lithuania 25.8 23.9 25.2 16.6 28.5 11.3 41. Former Soviet Union, excluding the Baltics 28.2 20.4 7.8 25.8 15.4 10.4 29.9 11.0 18.9 Armenia 38.2 25.3 35.8 14.7 20.5 8.3 Azerbeijan 34.7 18.7 31.9 6.4 39.4 16.0 Belarus 35.8 28.0 29.3 24.3 32.9 23.2 Georgia 25.3 25.2 24.9 13.1 24.9 9.6 Kazakstan 34.7 22.3 31.9 19.1 39.4 7.3 Kyrgyz Republic 13.1 14.6 4.0 14.4 14.0 8.7 Moldova - - 31.9 16.2 39.4 4.4 Russia 34.7 29.0 31.9 23.4 39.4 13.7 Tajikistan 12.5 15.8 13.7 14.0 17.3 11.4 Turkmenistan 34.7 13.4 31.9 12.6 39.4 8.4 Ukraine - - 26.3 15.7 28.4 12.2 Uzbekistan 18.2 12.0 15.7 10.7 23.6 8.9 - Not available. Source: World Bank data. 450 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 dicted rates resulted from adverse terms of trade shocks and the beginning of the growth collapse. In interpreting these findings, it must be born in mind that the saving levels before the transition are likely to be biased upward because black- market transaction prices are excluded from the official price indexes. III. SAVING DETERMINANTS IN TRANSITION AND MARKET ECONOMIES The elimination or reduction of price controls after 1989 in Eastern Europe and after 1991 in the former Soviet Union (De Melo, Denizer, and Gelb 1996) opened a margin of substitution, eliminating forced saving (Kornai 1992, Welfe 1989, and Quant 1989), primarily by reducing the real value of saving through inflation. For the 1990-95 period, then, it is best to think of the evolution of saving in terms of changes in desired equilibrium saving. Comparing the results of saving regressions in established market economies and in transition economies reveals both similarities and differences, with a ten- dency toward greater similarity over time (table 3). In the final sample year (1995) the familiar negative link between the dependency and urbanization ratios and Table 3. Saving Rate Regressions for Transition and Market Economies in Eastern Europe, 1989-95 Transition Variable 1989-95 1990-95 1995 Market Constant 0.078 0.152 0.683; 7.758 (0.67) (1.17) (2.05) (1.00) Dependency ratio 0.031 -0.032 -0.518 -0.175* (0.27) (0.25) (1.63) (2.40) Urbanization ratio 0.070 0.010 -0.420 -0.171X (0.74) (0.10) (1.75) (3.70) GDP growth -0.213* -0.263* -0.664** 0.008 (2.41) (2.82) (1.97) (0.05) Ratio of M2 to GDP 0.077* 0.076* 0.283* -0.077* (2.15) (1.98) (2.23) (2.42) CPI inflation -0.001'- _0.001* -0.001 0.001 (2.79) (2.60) (0.23) (0.91) Terms of trade change -0.007 -0.009 0.613* 0.119*' (0.26) (0.31) (3.81) (1.82) GDP per capita 1.6E-5* 1.37E-5** -4.2E-6 3.1E-5* (2.46) (1.89) (0.28) (6.57) War -0.200* -0.214* -0.229* -0.099 (6.03) (6.12) (3.09) (1.42) R2 0.544 0.55 0.78 0.25 Observations 150 131 22 254 Time dummies included? No No No No Significant at the 5 percent level. ** Significant at the 10 percent level. Note: t-statistics are in parentheses. Source: World Bank data. Denizer and Wolf 451 saving rates (Edwards 1996; Bayoumi, Masson, and Samiei 1995; and Loayza, Schmidt-Hebbel, and Serven 1998) holds across the transition economies. The negative effect of wars and the positive effect of terms of trade growth (Ostry and Reinhart 1992 and Bayoumi, Masson, and Samiei 1995) are also common to both the established market economies and the transition economies. The empirical literature has had mixed success in establishing a firm link be- tween inflation and saving rates. Although the entire sample of transition coun- tries displays a significant negative link, the effect is strongest in the early transi- tion years. Indeed, by 1995 the inflation link is both economically and statistically insignificant. Earlier studies on developing economies have found a positive dependence between saving and GDP per capita (Collins 1991; Edwards 1995; Bayoumi, Masson, and Samiei 1995). The positive link also holds for the entire transition sample, but it weakens over time; the coefficient for the 1995 sample is negative, although insignificant. The most striking differences between established market economies and the transition economies emerge for the ratio of M2 to GDP and for economic growth. The expected effect of the M2-GDP ratio on saving depends on whether it is viewed as a proxy for financial deepening-suggesting a positive effect-or as an indica- tor of borrowing constraints-suggesting a negative effect. Empirically, both nega- tive and positive effects have been found (Schmidt-Hebbel, Webb, and Corsetti 1992; Edwards 1996; Loayza, Schmidt-Hebbel, and Serven 1998). For the tran- sition economies the results are unambiguous: for the entire period and for indi- vidual years, a higher ratio of M2 to GDP has a significant positive association with saving. For the sample of established market economies the association is negative. The results also differ with respect to economic growth. Researchers have found that growth is positively associated with saving in developing countries (Collins 1991; Edwards 1996; Bayoumi, Masson, and Samiei 1995).We also find a posi- tive, albeit insignificant, effect for the sample of market economies. In sharp contrast, higher growth is associated with lower saving in the transition econo- mies. Furthermore, the effect becomes stronger over the sample period. This finding may reflect consumption smoothing: the fast-growing countries are located at the bottom of the adjustment J-curve, and thus a substantial part of the growth may reflect a rebound to prior levels. IV. THE ROLE OF TRANSITION STRATEGY The transition from plan to market dramatically changed economic circum- stances in the former socialist economies and may thus have influenced desired saving. From the viewpoint of the precautionary motive for saving, the move from a "cradle-to-grave" social safety net under socialism to a fairly unfettered market economy significantly raised income uncertainty. Viewed from a consumption-smoothing perspective, the transition tilted expected lifetime real income profiles upward for better-skilled and motivated individuals, but down- 452 THF. WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 ward for older, less-skilled individuals, particularly individuals receiving income support from the social security system. We next examine whether differences in the degree of liberalization affected saving in the transition economies. We add to our standard regressions the liberalization index developed by de Melo, Denizer, and Gelb (1996). The index is defined on a range of zero to one and comprises three subindexes measuring the degree of internal liberalization, external liberalization, and price liberalization. We also add the current index value and, to allow for lagged effects, the index value of the previous period. Although the current level of liberalization is not significantly associated with current saving, greater liberalization in the previous year is associated with a significant and large decline in saving in the current year (column 1 of table 4). To the extent that liberalization gradually became associated with future growth, the link is consistent with a smoothing of consumption in the initial phase of the creative destruction. The aggregate index, however, does not rule out the alternative possibility that the negative association reflects the more rapid dissolution of any remaining involuntary saving in the fast-reforming countries. Eliminating common annual effects by including time dummies changes the sign on the current liberalization index from positive to negative, although it still remains insignificant, while hav- ing little effect on the significance of the lagged liberalization index (column 2 of table 4). The last four regressions explore differential effects across the three subin- dexes capturing the restrictions on domestic market activity and foreign trade and the extent of the remaining price controls. We first add all three subindexes (all lagged once) to the regression and then one subindex at a time (columns 3-6 of table 4). Although the three subindexes are highly jointly significant, the dis- aggregation does not noticeably raise the overall explanatory power of the re- gression; the differences between the three indexes are correspondingly muted. In particular, the price liberalization index, arguably a good proxy for any remain- ing involuntary saving, does not enter significantly. V. CONCLUSIONS The transition economies of Eastern Europe almost uniformly experienced a severe decline in saving rates early in the transition, before rebounding slightly. The saving collapse could be explained by the elimination of involuntary saving or by a change in equilibrium saving reflecting the new economic circumstances and long-term prospects. We assessed the extent and presence of involuntary saving by estimating the predicted saving rates of market economies with the same fundamentals as the pretransition economies. On balance, predicted saving rates fell short of actual saving rates, particularly for the countries of the former Soviet Union and the Baltics, providing some support for the notion of excessive saving before the transition. Denizer and Wolf 453 Table 4. Saving Rate Regressions for Transition Economies in Eastern Europe, 1989-95 Variable 1 2 3 4 5 6 Constant 0.258* 0.229** 0.263* 0.274" 0.223** 0.246* (2.09) (1.84) (2.14) (2.24) (1.76) (2.02) Dependency ratio -0.129 -0.137 -0.154 -0.146 -0.082 -0.132 (1.07) (1.12) (1.28) (1.22) (0.66) (1.10) Urbanization ratio 0.048 0.061 0.085 0.060 0.137 0.070 (0.49) (0.62) (0.86) (0.62) (0.13) (0.72) GDP growth -0.063 0.009 -0.094 -0.105 -0.140 -0.084 (0.83) (0.86) (1.00) (1.13) (1.44) (0.89) Ratio of M2 to GDP 0.016 0.002 0.009 0.005 0.036 0.019 (0.41) (0.58) (0.23) (0.12) (0.93) (0.52) cpi inflation -0.001* -0.001* -0.001* -0.001* -0.001 * -0.001* (3.20) (3.18) (3.24) (3.19) (3.04) (3.33) Terms of trade change -0.013 -0.017 -0.011 -0.013 -0.014 -0.013 (0.50) (0.63) (0.43) (0.48) (0.51) (0.48) GDP per capita 1.7E-S* 1.7E-5* 1.3E-5** 1.4E-5* 1.9E-5* 1.6E-5* (2.47) (2.32) (1.85) (2.22) (2.70) (2.40) War -0.201 * -0.195 * -0.202* -0.198* -0.205* -0.205* (6.11) (5.87) (6.16) (6.07) (6.06) (6.33) Current liberalization index 0.020 -0.016 (0.26) (0.21) Lagged liberalization index -0.168* -0.158** (2.14) (1.74) Lagged internal liberalization -0.086 -0.127* (1.33) (4.71) Lagged external liberalization -0.089 -0.127* (1.29) (4.67) Lagged price liberalization 0.073 -0.137* (1.00) (3.33) R2 0.616 0.638 0.627 0.621 0.589 0.620 Observations 131 131 131 131 131 131 Time dummies included? No Yes No No No No r Significant at the 5 percent level. * * Significant at the 10 percent level. Note: t-statistics are in parentheses. Source: World Bank data. We found many similarities between the saving behavior of market econo- mies and transition economies, with the notable exception of a negative link between saving and GDP growth in transition countries. Since the fastest- growing transition economies can be found at the bottom of the adjustment J-curve, this finding is consistent with consumption smoothing. Finally, we explored whether differences in the extent of economic liberaliza- tion affected saving rates in the transition economies. We found that greater liberalization was associated with lower saving with a one-year lag. To the extent that liberalization is perceived as an indicator of future growth, this behavior is consistent with smoothing in the face of output evolving along a J-curve. 454 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Alexeev, Michael. 1988. "Are Soviet Consumers Forced to Save?" Comparative Eco- nomic Studies 30(4):17-23. Bayoumi, Tamim, Paul Masson, and Hossein Samiei. 1995. "International Evidence on the Determinants of Private Savings." IME Working Paper 95/51. International Mon- etary Fund, Washington, D.C. Processed. Bennett, John, and Maxim Boycko. 1995. "Savings and Stabilization Policy in a Pre-Post Socialist Economy." Journal of Money Credit and Banking 27(3). Collins, Susan. 1991. "Saving Behavior in Ten Developing Countries." In Douglas Bernheim and John Shoven, eds., National Savings and Economic Performance. Chi- cago: National Bureau of Economic Research and University of Chicago Press. Conway, Patrick. 1995. "Saving in Transition Economies: The Summary Report." World Bank, Washington, D.C. Processed. Davis, Cahremza, and Wojciech Charemza. 1989. Models of Disequilibrium and Short- age in Centrally Planned Economies. London and New York: Chapman Hall. Deaton, Angus. 1995. "Growth and Saving: What Do We Know, What Do We Need to Know, and What Might We Learn?" Princeton University, Research Program in De- velopment Studies, Princeton, N.J. Processed. De Melo, Martha, Cevdet Denizer, and Alan Gelb. 1996. "From Plan to Market: Patterns of Transition." Working Paper 1564. World Bank, Washington, D.C. Processed. Dornbusch, Rudiger, and Holger C. Wolf. 2000. "Monetary Overhangs." In Maurice Obstfeld, ed., Festscbrift for Robert Mundell. Chicago: University of Chicago Press. Easterly, William, and Stanley Fischer. 1994. "The Soviet Economic Decline: Historical and Republican Data." World Bank, Development Economic Department, Washing- ton, D.C. Processed. Edwards, Sebastian. 1995. "Why Are Savings Rates So Different across Countries?" NBER Working Paper 5097. National Bureau of Economic Research, Cambridge, Mass. Pro- cessed. -. 1996. "Why Are Latin America's Savings Rates So Low? An International Com- parison." Journal of Development Economics 51(1):5-44. Ickes, Barry W. 1993. "Saving in Eastern Europe and the Former Soviet Union." In Blackwell, Heertje, Arnold, eds., World Savings: An International Survey. Cambridge, MA. IMF (International Monetary Fund). 1995. World Economic Outlook. Washington, D.C. Kornai, Janos. 1980. Economics of Shortage. Amsterdam: North-Holland. . 1992. The Socialist System. Princeton, N.J.: Princeton University Press. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serv6n. 1998. "What Drives Saving across the World?" World Bank, Washington, D.C. Processed. Masson, Paul, Tamin Bayoumi, and Hossein Samiei. 1995. "Saving Behavior in Indus- trial and Developing Countries." Staff Studies for the World Economic Outlook. Washington, D.C.: World Bank. Nove, Alec. 1991. The Economics of Feasible Socialism Revisited. New York, Harper Collins Academic Publishers. Denizer and Wolf 455 Ostry, Jonathan, and Carmen Reinhart. 1992. "Private Saving and Terms of Trade Shocks." IMF Staff Papers 39(September):495-517. Portes, Richard. 1986. "The Theory and Measurement of Macroeconomic Disequilib- rium in Centrally Planned Economies." CEPR Discussion Paper 91. Center for Eco- nomic Policy Research, London. Processed. Quant, Richard E. 1989. "Disequilibrium Econometrics for Centrally Planned Econo- mies." In Christopher Davis and Woaciech Charemza, eds., Models of Disequilibrium and Shortage in Centrally Planned Economies. London and New York: Chapman and Hall. Schmidt-Hebbel, Klaus, Steven Webb, and Giancarlo Corsetti. 1992. "Household Sav- ings in Developing Countries." The World Bank Economic Review 6(3):529-47. Tarr, David. 1994. "The Terms of Trade Effects of Moving to World Prices on Countries of the Former Soviet Union." Journal of Comparative Economics 18(February):1-24. Weitzman, Martin L. 1991. "Price Distortion and Shortage Deformation or What Hap- pened to the Soap?" American Economic Review 81(June):401-14. Welfe, Aleksander. 1989. "Savings and Consumption in the Centrally Planned Economy." In Christopher Davis and Woaciech Charemza, eds., Models of Disequilibrium and Shortage in Centrally Planned Economies. London: Chapman and Hall. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 457-80 What Drives Consumption Booms? Peter J. Montiel Consumption booms have been common in both industrial and developing countries, and several explanations have been offered for their occurrence. These include economy- wide wealth effects associated with favorable movements in the terms of trade or eu- phoric expectations triggered by macroeconomic reforms, Ricardian effects associated with fiscal stabilization, lending booms following finanzcial liberalization, and a variety of distortions in intertemporal relative prices. Using a large cross-country sample of booms, this article assesses how widely applicable these explanations are. The key find- ing is that wealth effects linked to favorable movements in the terms of trade and antici- pated improvements in macroeconomic perfornance seem to have been more important empirically than explanations relying primarily on fiscal phenomena or distortions in intertemporal relative prices. Consumption booms, common in both industrial and developing countries, have been associated with a variety of macroeconomic events, including stabilization of high inflation, surges in capital inflows, the implementation of market- oriented structural reforms (especially trade and financial liberalization), and fa- vorable movements in the external terms of trade. They are often perceived as policy problems because of their effects on demand for home goods, the trade balance, and the resources available for investment. The emergence of a con- sumption boom may, for example, undermine the objective of stabilizing infla- tion by putting upward pressure on the prices of home goods. Similarly, a con- sumption boom that arises during an episode of capital inflows may significantly increase the country's current account deficit, which may, in turn, undermine the credibility of the prevailing exchange rate and contribute to capital flow rever- sals. The recent example of Mexico has now become notorious, but similar, if relatively muted, booms have characterized several other capital-importing coun- tries in Latin America. Despite the occasional attention given to specific episodes, researchers have only recently begun to study the causes of consumption booms in a systematic way. By and large, their analysis has been confined to speculation based on ca- sual empiricism in descriptive studies focusing on other issues (such as stabiliza- tion episodes and surges in capital inflows).' In these contexts several competing 1. Exceptions include Rebelo and Vegh (1996) and Reinhart and Vegh (1995). Peter J. Montiel is professor in the Department of Economics at Williams College. His e-mail address is pmontiel@zvilliams.edu. The author gratefully acknowledges comments on earlier drafts from Barry Bosworth, Carlos Vegh, and two anonymous referees. C) 2000 The International Bank for Reconstruction and Development/ THE WORLD BANK 457 4S8 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 hypotheses have been offered to explain the emergence of booms, attributing the phenomenon to changes in various determinants of aggregate consumption expenditures. Because consumption booms have so often been perceived as posing chal- lenges for macroeconomic policy, and because they have been attributed to a variety of different factors, it is useful to examine whether the data reveal system- atic patterns in their emergence. The objective of this article is to establish a set of stylized facts concerning the macroeconomic environments in which consump- tion booms have arisen. Based on a large sample of consumption booms, identified according to a set of objective criteria, I use two approaches to determine the circumstances sur- rounding their emergence. First, for a restricted sample of countries (dictated by the available data), I estimate a probit regression to explain the emergence of booms as a function of the explanatory variables typically emphasized in the literature. This exercise identifies the conditions that favor the emergence of booms in general, without necessarily explaining the emergence of any particu- lar episode. Second, I take up the question of whether consumption booms are all alike by determining the extent to which the behavior of certain key macroeconomic vari- ables during each episode was consistent with the hypotheses that have been offered to explain booms. I take the hypotheses one at a time. This narrows the scope of potential explanations for each episode, without attempting the more ambitious task of measuring the contributions made by each of the remaining hypotheses. I. IDENTIFYING CONSUMPTION BOOMS What exactly do we mean by a consumption boom? Based on considerations discussed in Montiel (1998), in this article I identify a consumption boom as an unusual increase in the ratio of real private consumption expenditures to real national income. Intuitively, a boom should be characterized by a large and sus- tained deviation in the ratio of consumption to gross domestic product (GDP) from its "normal" value that is driven by the behavior of consumption, rather than that of income. I construct the sample of consumption booms used here by applying an operational version of this criterion to the data from a large sample of industrial and developing countries. The sample includes all countries in the World Bank's World Saving Database that have at least 20 annual observations on both real private consumption and real GDP for the period 1960-95. There are 91 such countries, of which 23 are industrial countries and 68 are developing countries (the sample includes no transition economies). To make the intuitive characterization of booms operational, I use the follow- ing procedure to identify consumption booms in the sample. First, for each coun- try I test the ratio of real private consumption to real GDP for trend stationarity. I detrend (if necessary) series that prove to be trend stationary to identify and Montiel 459 separate out secular movements. Series that appear to be nonstationary are ex- amined further for trend breaks and structural shifts that may account for the rejection of stationarity. I introduce shift parameters or trend breaks where ap- propriate to render the series stationary. Second, I calculate transitory compo- nents of each series as deviations of the original series from the deterministic trend component. A boom is said to occur during a period characterized by a "run" of positive deviations of the ratio of private consumption to GDP from its deterministic trend. To minimize the incidence of spurious booms generated by errors in the data, particularly among the large number of low-income countries in the sample, iden- tified booms have to meet three additional conditions: the run must contain at least one observation greater in magnitude than two standard deviations of the transitory consumption-ratio series, the run must consist of at least two consecu- tive annual observations, and the run must contain at least two consecutive ob- servations for which the rate of growth of real private consumption exceeds its sample mean.2 The resulting sample of consumption booms among the 91 sample countries totaled 40 during 1960-95 (table 1). Thirty-five countries experienced at least one boom episode, and 5 of these experienced more than one episode during the period. The sample of consumption booms contains episodes in both industrial (9 episodes) and developing (31 episodes) countries, and in all geographic re- gions. Booms are also widely dispersed across time. Eight booms originated in the 1960s, 19 in the 1970s, 12 in the 1980s, and 1 in the 1990s. The frequency of booms in the sample suggests that they can arise under di- verse circumstances. The fact that consumption booms are fairly common makes it highly unlikely that they arise only as the result of exceptional policy environ- ments. Nonetheless, this conjecture can be tested only by examining the circum- stances surrounding their emergence. II. WHAT CAUSES CONSUMPTION BOOMS? In principle, a consumption boom may be driven by a variety of macroeco- nomic factors. There is no reason to believe that booms share one cause or a set of causes. Nonetheless, it is useful to determine whether booms have arisen more readily in some circumstances than in others-that is, whether any particular set of macroeconomic circumstances has systematically influenced the probability that a boom will emerge. This section reports estimates of a set of probit regres- 2. Private consumption data are often estimated as residuals from categories that are themselves indi- rectly estimated. The compounding of measurement errors in this procedure introduces a substantial amount of noise to the time series for private consumption, a problem that may be particularly severe for small low-income countries. This problem could introduce spurious "boom" observations and generate selectiv- ity bias by overrepresenting poor countries (with presumably worse data) in the boom sample. I have tried to reduce the severity of this problem by relying only on the "cleaned up" consumption and GDP data from the World Saving Database as well as by using a restrictive set of criteria in identifying booms. 460 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 1. Consumption Booms Country Data Stationarity Booms Industrial countries Australia 1960-94 S None Austria 1960-94 D None Belgium 1960-94 S None Canada 1960-94 S 1960-63 Denmark 1960-94 TS 1975-79 Finland 1960-94 S None France 1960-94 D None Germany 1960-94 T2S None Greece 1960-94 T2S None Iceland 1960-94 T2S 1970-74 1987-88 Ireland 1960-94 TS 1978-81 Italy 1960-94 TS None Japan 1960-94 S None Luxembourg 1960-94 T2S None Netherlands 1960-94 S 1975-83 New Zealand 1960-94 D,T 1972-75 Norway - 1960-94 TS 1975-77 Portugal 1960-94 S 1971-78 Spain 1960-94 D None Sweden 1960-94 TS None Switzerland 1960-94 T2S None United Kingdom 1960-94 D None United States 1960-94 T2S None Developing countries Algeria 1960-94 D None Bangladesh 1960-94 D 1983-89 Benin 1960-94 D None Bolivia 1960-94 D None Botswana 1975-94 D None Brazil 1960-94 TS None Burkina Faso 1965-94 T2S None Burundi 1960-94 T2S 1961-63 1979-82 Cape Verde 1973-93 TS 1976-80 Central African Republic 1960-94 TS None Chile 1960-94 TS 1978-81 China 1960-94 D None Colombia 1960-94 D None Congo 1960-94 D None Costa Rica 1960-94 T2S 1964-71 C6te d'lvoire 1965-94 D 1969-70 1976-77 Cyprus 1975-94 S 1991-92 Dominican Republic 1960-94 T2S None Ecuador 1965-94 D None Egypt, Arab Rep. 1974-94 TS None Gabon 1960-93 S None Gambia, The 1960-94 D 1984-85 Montiel 461 Country Data Stationarity Booms Ghana 1960-94 S None Guinea-Bissau 1970-94 D 1971-75 Guyana 1960-88 TS None Honduras 1960-94 D 1976-77 Hong Kong i965-94 TS None India 1960-94 TS None Indonesia 1960-94 S 1981-87 Iran 1974-94 TS None Israel 1960-94 T2S None Jamaica 1960-93 T2S 1987-90 Kenya 1960-94 T2S None Korea, Rep. of 1960-94 T2S None Lesotho 1960-94 T2S 1975-77 Madagascar 1960-94 S None Malawi 1960-94 T2S None Malaysia 1960-94 T2S 1978-85 Mali 1967-94 T2S 1983-85 Mauritania 1960-94 TS 1973-77 1982-84 Mexico 1960-94 D,T 1989-94 Morocco 1960-94 T2S None Nicaragua 1960-94 D None Niger 1960-94 D None Nigeria 1960-94 D 1978-86 Pakistan 1960-94 T2S None Panama 1960-94 T2S None Papua New Guinea 1961-94 T2S None Paraguay 1960-94 S 1962-69 Peru 1960-94 S None Philippines 1960-94 D None Rwanda 1960-94 S(1960-90) None Senegal 1960-94 S None Singapore 1960-94 S None South Africa 1960-94 TS None Sri Lanka 1960-94 D,T 1980-83 Sudan 1960-91 T2S 1977-85 Taiwan (China) 1965-94 T2S None Thailand 1960-94 TS None Togo 1960-94 T2S 1966-70 Trinidad and Tobago 1960-94 TS 1982-84 Tunisia 1961-94 D None Turkey 1960-94 D,T 1982-86 Uruguay 1960-94 T2S 1969-75 Venezuela 1974-94 T2S None Zaire 1970-89 S None Zambia 1960-94 S 1966-70 1987-90 Zimbabwe 1965-93 S 1978-83 Note: S = stationary, D = stationary after introduction of a dummy, TS = trend stationary, T2S = stationary after removal of quadratic trend. Source: Author's calculations based on information in the World Bank's World Saving Database. 462 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 sions that seek to explain the probability of booms as a function of macroeco- nomic variables derived from existing hypotheses of their origins.3 Specification The first step in the analysis is to express each hypothesis in terms of links be- tween observable macroeconomic variables. I take them up one at a time. INCOME REDISTRIBUTION. Consumption booms may arise as the result of a sig- nificant redistribution of income in favor of lower-income groups, itself the out- come of populist policies. Since lower-income groups are more likely to be liquidity- constrained than higher-income groups, such a redistribution would tend to shift income from unconstrained to constrained households, thereby causing aggre- gate consumption to rise. The question, then, is whether the booms identified above were indeed associated with large redistributions of income in favor of lower-income households. Unfortunately, to answer this question, we would need annual data on house- hold income distribution for large groups of countries. The most comprehensive data set of this type is the Deininger and Squire (1996) database. But even these data barely provide enough information to draw tentative inferences about what may have been happening to the distribution of household income for a subset of the consumption booms in the sample. Ideally, we want data on the distribution of household income for a number of years immediately before, during, and after the boom in order to assess whether the boom was associated with a systematic change in that distribution. But the best we can do in most cases is to derive the Gini coefficient or the household distribution of income by quintile for one or more noncontiguous years shortly before, during, and after the boom. For this reason I exclude the income-distribution hypothesis from the probit estimation. CHANGES IN INTERTEMPORAL RELATIVE PRICES. A second category of explana- tions for the emergence of booms focuses on changes in intertemporal relative prices. These hypotheses come in several forms. The first operates in the context of exchange rate-based stabilization with inflation inertia. Rodriguez (1982) ar- gues that in economies that are financially open, domestic nominal interest rates would tend to fall when the exchange rate is used as a nominal anchor to stabi- lize inflation. If expectations of inflation remain high, however (say, because of inflation inertia), the domestic real interest rate would tend to fall. And if con- sumption responds to the real interest rate, this fall would trigger an increase in consumption spending. The empirical question, then, is whether booms are asso- ciated with a contemporaneous reduction in real interest rates. The second mechanism involves "incredible" exchange rate-based stabiliza- tion. Dornbusch (1985) suggests that exchange rate-based stabilizations that are not expected to last tend to create an intertemporal distortion in the form of an 3. The approach here is analogous to that which has recently been emploved to study the determinants of currency crises. See, for example, Eichengreen, Rose, and Wyplosz (1995). Montiel 463 expected future increase in the real price of importables (a real exchange rate depreciation). To the extent that consumption durables fall into this category, the anticipation that their price will rise in the future will drive consumers to shift the purchase of durables to the present. The implication is that such stabilizations, if they are not expected to be sustainable, may be associated with private con- sumption booms driven by temporary real exchange rate appreciation. But even if the real exchange rate is not expected to change, intertemporal relative prices may change so as to favor increased consumption in the present. The mechanism involves a cash-in-advance technology for making transactions. If money reduces the cost of economic transactions, lower nominal interest rates will lower the opportunity cost of holding the money balances required to effect current transactions, and thus current consumption will be cheap relative to fu- ture consumption. Assuming sufficient intertemporal substitutability in consump- tion, households will shift consumption to the present. Thus to assess the poten- tial role of this mechanism-due to Calvo (1989) and dubbed the "temporariness" hypothesis-we must determine whether booms are associated with nominal in- terest rates that are temporarily lower. Each of these hypotheses focuses on a specific macroeconomic variable: the real interest rate, the real exchange rate, or the nominal interest rate. Each im- plies a specific path for the relevant variable before, during, and after the boom in order to account for the boom's inception and end: the Rodriguez hypothesis suggests that the real interest rate should fall during the boom and rise subse- quently; the Dornbusch hypothesis implies that the real exchange rate should appreciate during the boom and be expected to depreciate subsequently; and the Calvo hypothesis holds that the nominal exchange rate should fall during the boom and be expected to rise subsequently. In the probit regression I attempt to capture these transitory movements by measuring each variable in the form of deviations from the country-specific sample mean. According to the three hypotheses, then, the probability that a particular observation will be associated with a consumption boom will be higher if the real and nominal interest rates are below their sample means and if the real exchange rate is above its sample mean.4 WEALTH EFFECTS. A boom may arise because the private sector perceives an increase in its wealth and adjusts its consumption path accordingly. Private wealth may increase with an increase in national wealth, which is shared by the private and public sectors, arising from a perceived permanent improvement in the country's terms of trade or a change in the policy regime that is expected to accelerate economic growth. Alternatively, private wealth may increase because 4. 1 take data on nominal interest rates from the World Saving Database, using market-determined rates (money market rates or T-bill rates) whenever possible and deposit rates otherwise. I compute the real interest rate as an exact ex post real interest rate based on the consumer price index. The real exchange rate data are the trade-weighted real effective exchange rate series computed by the International Mon- etary Fund, which defines an appreciation as an increase in the real effective exchange rate. 464 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 of a change in fiscal policy that is perceived to reduce the present value of the private sector's future tax liabilities. To evaluate the potential role of these variables, the probit regression exam- ines whether the booms in the sample were associated with contemporaneous improvements in the country's terms of trade, increases in the rate of economic growth, and reductions in the ratio of real public consumption to GDP. CREDIT EXPANSION. Finally, consumption booms also have been attributed to a rapid expansion of credit to the private sector. The expansion may arise from a variety of sources, involving the relaxation of credit constraints triggered by ex- plicit or implicit backing of financial liabilities by the public sector. Two versions are common. First, booms have been said to result from the relaxation of credit constraints at the level of the household or firm following inappropriate domestic financial liberalization. Specifically, the rapid liberalization of a previously repressed fi- nancial system without appropriate regulatory and supervisory mechanisms for financial institutions, but with explicit or implicit deposit insurance and poor capitalization of banks, creates a structure of incentives for bank managers that fosters excessive risk taking in their portfolio decisions. This risky lending may take the form of an excessive expansion of consumption loans or loans used to acquire (or secure) stocks or real estate. Lending booms for the latter may, in turn, trigger asset price bubbles in real estate or in the stock market. Whether consumption loans expand disproportionately (thus alleviating liquidity con- straints on households) or asset price bubbles arise (thus exerting wealth effects on household spending), a boom in private consumption may be the result. In the second version of this credit story, the credit expansion originates abroad, and liquidity constraints are no longer binding for the country as a whole. Just as in the case of a domestically driven credit expansion, a large capital inflow that is at least partially intermediated by the domestic financial system can set off the rapid expansion of domestic credit and the development of asset price bubbles. What these phenomena have in common is that the resulting consumption boom is associated with a rapid expansion of credit to the private sector from domestic financial intermediaries. This expansion may itself be the source of the boom, or it may in part reflect growth in the demand for credit triggered by the wealth effects associated with asset price bubbles. In either case, a rapid increase in the rate of expansion of credit to the private sector will be a symptom of such finance-driven booms. Accordingly, the final variable to be included in the probit regression is the ratio of the flow of credit to the private sector to GDP. If the credit expansion view is correct, an increase in this variable should be associated with a higher probability of observing a boom in progress. Probit Results The data set I use in this article consists of annual observations of the variables listed above for 91 countries during the period 1960-95-a total of 3,276 poten- Montiel 465 tial panel observations. However, complete data for all of the variables (except for the income distribution variable) were available for only 640 observations, including 40 boom observations. I thus base the probit estimates on this reduced sample, consisting of an unbalanced panel of 51 countries. To cope with potential endogeneity problems, I estimate four different probit regressions. Equation 1 includes the contemporaneous values of all the variables in the form suggested by the hypotheses described above. One problem with estimating the regression in this form, however, is that many of the explanatory variables are likely to respond endogenously to the emergence of a boom. This obviously hampers our ability to draw inferences about the causal roles that the explanatory variables may have played. As a simple procedure to reduce the potential effects of feedback from booms to the explanatory variables, equation 2 replaces all of the potentially endog- enous variables with their lagged values.5 However, this procedure may bias the results against hypotheses that rely on the response of consumption expenditures to contemporaneous changes in domestic interest rates. Accordingly, equation 3 adopts the alternative approach of instrumenting for the two domestic interest rate variables, using as instruments the contemporaneous U.S. nominal and real interest rates. It also includes the lagged values of the other endogenous variables and the two exogenous variables. I run two variants of this regression, with and without the real interest rate (equations 3a and 3b). Overall, the estimates suggest that changes in domestic interest rates have not played an important role in generating the booms observed in this sample (table 2). The coefficients of the nominal and real interest rate variables do not ap- proach statistical significance with the theoretically predicted (negative) signs in any of the regressions. Similarly, the results provide little support for the credit hypothesis or for the hypothesis of a wealth mechanism operating through higher growth of real GDP. With the exception of equation 1, in which endogeneity is a strong suspect, the coefficients of the credit and growth variables have negative signs, the opposite of what would be predicted by the "lending boom" or "growth euphoria" hypotheses. Booms in this sample instead tend to be associated with appreciated real ex- change rates and favorable terms of trade. The coefficients on both of these vari- ables are statistically significant and have theoretically appropriate signs in equa- tions 1 and 2, and the terms of trade variable continues to perform well in equation 3a. Although the p-value of the lagged real exchange rate rises somewhat (to 0.19) in equation 3a, the sign of this variable proves to be robust across equa- tions. The loss of statistical precision in the estimate of its coefficient in equation 3a may be due to covariation with the instrumented value of the real interest rate. Indeed, the correlation between the lagged real exchange rate and the real interest rate rises from 0.13 in equation 2 to 0.43 in equation 3a. Dropping the 5. 1 assume that the terms of trade and ratio of public consumption to GDP are exogenous in this exercise. The latter is obviously problematic, but the exogeneity of public consumption within a particular year seems warranted by well-known problems of fiscal rigidity. 466 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 2. Determinants of Consumption Booms: Probit Results Equation Variable (1) (2)a (3a)b (3b), Constant -1.702 -1.626 -1.580 -1.634 (0.12) (0.109) (0.142) (0.113) Real exchange rate 0.007* 0.002* 0.001 0.002* (0.003) (0.001) (0.001) (0.001) Real interest rate 0.016 0.003 -0.037 (0.009) (0.007) (0.059) Nominal interest rate -0.002 0.000 0.005 0.006 (0.004) (0.000) (0.023) (0.023) Real GDP growth 0.030 -0.019 -0.022 -0.017 (0.022) (0.019) (0.021) (0.019) Terms of trade 1.470'* 2.024** 1.995** 2.021** (0.352) (0.316) (0.317) (0.315) Government consumption 1.063 -3.016 -3.041 -3.179 (3.286) (4.012) (3.971) (3.922) Credit to the private sector 1.063 -2.311 -2.182 -2.232 (3.286) (1.194) (1.173) (1.178) Likelihood ratio -105.97 -106.81 -106.83 -107.02 Significant at the 1 percent level. *x Significant at the 5 percent level. Note: Standard errors are in parentheses. a. Lagged values are used for all variables except the terms of trade and the ratio of government consumption to GDP. b. The contemporaneous U.S. real and nominal interest rates are used to instrument for the real and nominal interest rate. c. The contemporaneous U.S. nominal interest rate is used to instrument for the nominal interest rate. Source: Author's calculations based on information in the World Bank's World Saving Database. real interest rate variable from equation 3a, as in equation 3b, lowers the p-value for the real exchange rate from 0.19 to 0.036. The implication is that both the intertemporal reallocation of consumption expenditures in response to changes in intertemporal relative prices and wealth effects were operative in generating the booms observed in this sample. The real- location of consumption expenditures, however, is more likely to have worked through anticipated future increases in the price of consumer durables with a large imported component (as in the Dornbusch hypothesis) rather than through interest rate effects, either real or nominal. Similarly, wealth effects have gener- ated booms more often when such effects were produced by improvements in the economy's terms of trade rather than by increases in the growth rate or by sectoral effects operating through the share of the economy's resources absorbed by the public sector. These results, however, speak to the factors that have been associated most often with the emergence of booms and thus to the conditions under which booms are most likely to emerge. They cannot be interpreted as ruling out other mecha- nisms in driving specific episodes. In other words, we cannot, with these results, presume that all booms are alike. To examine the extent of uniformity in the Montiel 467 factors driving booms, in the next section I consider the potential role that each of the mechanisms may have played in the individual booms included in the sample. III. EXPLAINING INDIVIDUAL BOOMS Explaining each boom-that is, measuring the empirical contribution of each potential causal variable to movements in the ratio of private consumption to GDP during each episode-is obviously beyond the scope of this article. Instead, I take up the more limited and logically prior question of whether the time-series behavior of the relevant macroeconomic variables during each boom is consis- tent with the hypotheses described above. In particular, I ask: * Did domestic real interest rates fall at the inception of the boom? * Was the emergence of the boom associated with a reduction and subse- quent increase in the nominal interest rate? * Did the boom accompany a major program of structural reform or a large change in the terms of trade, or was it followed by an acceleration of eco- nomic growth or by a fiscal consolidation that could be related to a revision in households' expectations of future income? * Were there changes in the domestic financial system or in households' ac- cess to external financial markets that were consistent with a finance- induced boom? Such changes would involve not just financial liberalization but also, as indicated above, an approach to liberalization that may have encouraged imprudent behavior on the part of lenders. Detecting the appropriate change in the macroeconomic variable emphasized by each hypothesis is a necessary, though not sufficient, condition to conclude that the hypothesis provides a valid explanation for the emergence of the boom. Further investigation would be required to establish its empirical relevance. If the postulated change did not occur, however, then the hypothesis cannot be sus- tained for the boom in question. In other words, this approach may permit us to rule out some explanations in individual cases, but it cannot identify the relative importance of surviving explanations. The objective is to expand the set of styl- ized facts about consumption booms by examining individual cases, while keep- ing the task to manageable proportions. I consider the alternative hypotheses one by one. Booms and Intertemporal Relative Prices The previous section linked the three boom hypotheses that rely on changes in intertemporal relative prices to the behavior of a specific macroeconomic vari- able. In order to account for the inception and end of the boom, each hypothesis implies a specific path for the relevant variable before, during, and after the boom. The Rodriguez hypothesis suggests that the real interest rate should fall during the boom and rise subsequently. The Dornbusch hypothesis implies that the real 468 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 exchange rate should appreciate during the boom and be expected to depreciate subsequently. And the Calvo "temporariness" hypothesis suggests that the nomi- nal exchange rate should fall during the boom and be expected to rise subsequently. Since we cannot observe the expectations that operated during each of these episodes, I proxy them by the actual ex post behavior of the variable in question. Accordingly, in the discussion that follows I describe the behavior of the relevant variable during the boom episode as consistent with the hypothesis if its realized values follow the temporal pattern of the actual and expected future values that the hypothesis suggests. Consider first the Rodriguez real interest rate mechanism. I examine the be- havior of real interest rates around consumption booms for all of the booms in the sample for which I could extract market-determined interest rate data from the World Saving Database (table 3). Consistency in this case requires that the real interest rate be lower during the boom than both before and after the boom. Overall, the real interest rate hypothesis does not fare well. For only two of the sample booms does the behavior of the real interest rate match that suggested by the Rodriguez hypothesis: Cyprus in 1991-92 and Zambia in 1987-90. In four Table 3. Consumption Booms and Real Interest Rates Country Boom Before During After Consistent? Bangladesh 1983-89 -0.72 1.60 5.03 N Burundi 1979-82 -8.58 -7.96 -1.61 N Canada 1960-63 - 0.00 3.50 ? Chile 1978-81 -51.25 10.08 11.30 N C6te d'lvoire 1976-77 - -10.21 -6.60 ? Cyprus 1991-92 1.67 0.16 1.15 Y Denmark 1975-79 -1.35 1.42 4.76 N Gambia, The 1984-85 0.27 -9.40 -16.24 N Iceland 1970-74 - -100.00 0.50 ? 1987-88 -6.46 -1.28 0.03 N Indonesia 1981-87 -8.16 2.75 10.73 N Ireland 1978-81 -8.46 -4.03 -1.14 N Jamaica 1987-90 -2.92 6.93 -7.38 N Malaysia 1978-85 -2.11 0.95 2.91 N Mexico 1989-94 -14.45 3.93 - N Netherlands 1975-83 -0.01 0.90 4.87 N Nigeria 1978-86 -9.51 -8.03 -11.50 N Norway 1975-77 -1.26 -0.99 1.15 N 1985-88 2.63 6.00 7.87 N Portugal 1971-78 - -5.44 0.51 ? Sri Lanka 1980-83 -1.11 2.80 6.53 N Turkey 1982-86 -38.58 4.84 -6.37 N Zambia 1987-90 -21.81 -37.61 45.79 Y Zimbabwe 1978-83 -7.03 -5.14 -3.03 N - Not available. Note: Y indicates a lower real interest rate during the boom than either before or after the boom, corresponds to a lower real interest rate during the boom than before, or a higher real interest rate after the boom than during it, and no data are available for the other period. All other cases are labeled N. Source: Author's calculations based on information in the World Bank's World Saving Database. Montiel 469 other cases-Canada, C6te d'Ivoire, Iceland, and Portugal-no data were avail- able to confirm a reduction in the real interest rate during the boom, but the real interest rate did increase after the boom. The temporal pattern in the data did not match that predicted by the hypothesis in 18 of 24 cases. The temporariness hypothesis requires that the nominal interest rate fall dur- ing the boom and rise subsequently. None of the boom episodes in the table clearly fits this pattern (table 4). Post-boom developments are often consistent with the postulated time path of the nominal interest rate (which increased after the boom in 16 of the 24 cases), but the nominal interest rate does not drop in any of the cases for which data are available, making it difficult to account for the timing of the boom's emergence using this mechanism. In three cases, how- ever, the hypothesis cannot be ruled out, because of the absence of data. Overall, the predicted temporal pattern is rejected in 21 of 24 cases. For the most part, then, the evidence does not suggest that a large number of booms have emerged as a consequence of movements in nominal interest rates. The mechanism may be important in the context for which the hypothesis was intended (exchange rate- Table 4. Consumption Booms and Nominal Interest Rates Country Boom Before During After Consistent? Bangladesh 1983-89 8.57 12.00 9.00 N Burundi 1979-82 3.00 4.00 5.25 N Chile 1978-81 95.00 47.00 34.25 N C6te d'Ivoire 1976-77 - 7.00 7.50 ? Cyprus 1991-92 6.00 6.00 6.00 N Denmark 1975-79 9.00 11.60 14.80 N Gambia, The 1984-85 9.00 9.50 16.00 N Iceland 1970-74 - 14.50 26.40 ? 1987-88 18.50 21.00 18.00 N Indonesia 1981-87 9.00 12.00 19.40 N Ireland 1978-81 8.00 10.00 9.25 N Jamaica 1987-90 16.25 20.5 33 N Malaysia 1978-85 3.50 5.63 5.75 N Mauritania 1982-84 6.00 6.00 6.67 N Mexico 1989-94 68.00 21.33 16.00 N Netherlands 1975-83 6.11 7.11 6.67 N Nigeria 1978-86 3.00 6.67 16.25 N Norway 1975-77 6.67 8.33 9.33 N 1985-88 12.75 13.50 11.75 N Portugal 1971-78 - 13.67 21.75 ? Sri Lanka 1980-83 10.50 20.25 15.50 N Turkey 1982-86 10.80 46.20 49.60 N Zambia 1987-90 13.25 19.25 124.00 N Zimbabwe 1978-83 4.00 6.33 9.00 N - Not available. Note: Y indicates that the nominal interest rate falls during the boom and rises after, ? indicates that the nominal interest rate rises after the boom, and there are no data before the boom, while N indicates all other cases. Source: Author's calculations based on information in the World Bank's World Saving Database. 470 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 based stabilization in high-inflation countries), but it seems to have limited appli- cability outside of that setting. Finally, to determine whether an anticipated future increase in the real price of durables may have played a role in triggering some of the consumption booms in the data, I examine the extent to which booms have been associated with a tran- sitory appreciation of the real exchange rate. Unfortunately, data for real effec- tive exchange rates-taken from the International Monetary Fund's trade-weighted series-are only available for a sufficiently large group of countries after 1979, limiting the sample of booms that could be examined for consistency with the Dornbusch hypothesis. The Dornbusch hypothesis also fails to find strong support in the data, but it fares somewhat better than the two preceding hypotheses (table 5). The available data are consistent with the predicted temporal pattern in 9 of the 22 cases. In the remaining 13 episodes the real exchange rate either depreciated during the boom or appreciated after the boom. Both events are inconsistent with the emergence of a boom as the result of a temporary real appreciation. However, using a uni- form standard for all three hypotheses, the incidence of inconsistency is much lower for this hypothesis than for the preceding ones. Table 5. Consumption Booms and Real Appreciation Country Boom Before During After Consistent? Bangladesh 1983-89 109.77 107.41 94.31 N Burundi 1979-82 - 148.16 169.36 N Cape Verde 1976-80 - 87.50 98.64 N Chile 1978-81 - 201.99 173.13 ? Cyprus 1991-92 101.06 101.29 104.35 N Denmark 1975-79 - 100.89 89.43 ? Gambia, The 1984-85 128.96 126.91 97.06 N Iceland 1987-88 98.10 108.28 100.54 Y Indonesia 1981-87 198.75 183.72 100.04 N Ireland 1978-81 - 87.51 94.66 N Jamaica 1987-90 117.28 104.83 85.54 N Malaysia 1978-85 - 142.80 107.38 ? Mauritania 1982-84 130.54 148.06 122.42 Y Mexico 1989-94 93.83 112.68 81.94 Y Netherlands 1975-83 - 106.04 100.11 ? Nigeria 1978-86 - 456.83 104.84 ? Norway 1985-88 100.40 99.22 98.58 N Sri Lanka 1980-83 98.79 122.53 124.55 N Sudan 1977-85 - 43.62 61.87 N Turkey 1982-86 122.56 101.26 90.13 N Zambia 1987-90 108.65 94.55 95.36 N Zimbabwe 1978-83 - 160.25 132.05 ? - Not available. Note: Y indicates real appreciation during the boom, while ? indicates real depreciation after the boom, and no data before the boom. All other cases are labeled N. Source: Author's calculations based on information in the World Bank's World Saving Database. Montiel 471 Overall, none of the mechanisms that rely on changes in intertemporal relative prices appears to offer promising explanations for a large number of consump- tion booms. Consistent with the results of the probit estimation, neither of the interest rate-based hypotheses fares well in the examination of individual booms. They may be important in particular circumstances, as in the case of incredible stabilizations, but the triggers for the majority of the consumption booms uncov- ered in the data appear to lie elsewhere. The results for the Dornbusch hypothesis are more favorable, again consistent with the probit estimates, but its applicabil- ity is clearly less than general. The "Euphoria" Factor in Consumption Booms Unfortunately, the relationship between the wealth variables and consump- tion booms is not straightforward. Theory suggests that a permanent improve- ment in the terms of trade, for example, should increase the ratio of private consumption to GDP, but it should do so permanently and not result in a transi- tory consumption boom. On the other hand, a transitory improvement in the terms of trade should have a minimal impact on the ratio of consumption to GDP. The same reasoning applies to the other variables that may increase private con- sumption through wealth effects-although they may account for a one-time change in the consumption-income ratio, they would not in general produce tran- sitory consumption swings of the type associated with booms in this sample. A possible explanation concerns the role of consumer durables. Since what is mea- sured in the data is consumption expenditures, rather than consumption itself, an adjustment in the stock of durables in response to a perceived increase in household wealth could, in principle, trigger an overshooting in consumer expenditures. Accordingly, the standards of consistency are somewhat looser in this section than in the previous one. Here I judge the data to be consistent with each wealth hypothesis if the hypothesis can account for the increase in the ratio of consump- tion to GDP during the boom, without requiring it also to explain why the boom ended. In the case of a change in the terms of trade, for example, the emergence of a consumption boom is taken to be consistent with an explanation relying on a perceived permanent improvement in the terms of trade if the terms of trade improved during the boom and were, on average, more favorable during and after the boom than before the boom. I employ similar criteria for the other two variables considered in this section, the growth rate of real GDP and the share of public consumption in GDP. According to this criterion, a terms of trade explana- tion is consistent with the evidence for 22 of the 3 8 cases for which complete data are available (table 6). In 3 of the 16 cases that are rejected (Portugal, Sudan, and Togo), the terms of trade improved during the boom, but deteriorated subse- quently by an amount sufficient to make the average terms of trade less favorable during and immediately after the boom than before the boom. Overall, it is clear that booms have routinely been accompanied by improvements in the terms of trade. Since there is no question of reverse causality for the countries in this 472 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 6. Consumption Booms and Terms of Trade Shocks Country Boom Before During After Consistent? Bangladesh 1983-89 0.85 1.00 0.96 Y Burundi 1961-63 1.38 1.38 1.30 N 1979-82 1.97 1.25 1.18 N Cape Verde 1976-80 0.37 0.71 1.00 Y Chile 1978-81 1.35 1.14 0.97 N Costa Rica 1964-71 1.19 1.12 0.98 N Cote d'Ivoire 1969-70 1.13 1.51 1.15 Y 1976-77 1.22 1.74 1.73 Y Cyprus 1991-92 0.96 0.98 0.99 Y Denmark 1975-79 1.15 1.07 0.95 N Gambia, The 1984-85 0.82 0.93 1.02 Y Guinea-Bissau 1971-75 0.14 0.61 0.44 Y Honduras 1976-77 0.97 1.29 1.22 Y Iceland 1970-74 0.87 1.01 0.99 Y 1987-88 0.93 1.00 0.94 Y Indonesia 1981-87 0.75 1.17 0.92 Y Ireland 1978-81 1.01 0.96 0.96 N Jamaica 1987-90 0.77 1.02 0.72 Y Lesotho 1975-77 1.06 1.54 1.22 Y Malaysia 1978-85 0.94 1.08 1.01 Y Mauritania 1973-77 0.97 1.24 0.97 Y 1982-84 0.93 0.98 1.07 Y Mexico 1989-94 1.11 1.07 0.99 N Netherlands 1975-83 1.07 1.02 1.01 N New Zealand 1972-75 1.00 1.08 0.93 Y Nigeria 1978-86 0.95 1.99 0.86 Y Norway 1975-77 1.26 1.15 1.17 N 1985-88 1.31 1.08 0.99 N Paraguay 1962-69 0.70 0.91 1.02 Y Portugal 1971-78 0.96 1.02 0.89 N Sri Lanka 1980-83 1.41 0.99 1.00 N Sudan 1977-85 0.98 1.21 0.76 N Togo 1966-70 1.58 1.78 0.96 N Turkey 1982-86 0.86 1.02 0.96 Y Uruguay 1969-75 1.28 1.29 0.88 N Zambia 1966-70 1.41 2.24 1.59 Y 1987-90 0.69 0.92 0.57 N Zimbabwe 1978-83 0.35 1.06 1.05 Y Note: Y indicates that the terms of trade improve during the boom and are more favorable on average during and after the boom than before, and N refers to all other cases. Source: Author's calculations based on information in the World Bank's World Saving Database. sample, the suggestion is strong that improvements in the terms of trade often have contributed to the emergence of consumption booms. The criterion for consistency of the growth rate of real output is that the growth rate accelerated during the boom, and the average growth rate during and after the boom exceeded the average growth rate before the boom (table 7). The first point to notice is that booms are not always accompanied by an acceleration of economic growth. Indeed, the growth rate fell during the boom in 17 of the 35 episodes for which it is possible to make the comparison. In 9 of the 35 booms Montiel 473 Table 7. Consumption Booms and Economic Growth Country Boom Before During After Consistent? Bangladesh 1983-89 4.02 3.97 4.30 N Burundi 1979-82 4.92 1.52 3.94 N Cape Verde 1976-80 -3.08 7.14 13.64 Y Chile 1978-81 -3.63 7.31 -2.00 Y Costa Rica 1964-71 0.76 6.73 6.07 Y Cote d'lvoire 1969-70 9.51 10.49 4.76 N 1976-77 9.14 3.85 6.68 N Cyprus 1991-92 8.39 3.88 5.45 N Denmark 1975-79 3.90 1.39 1.13 N Gambia, The 1984-85 11.27 -3.34 0.29 N Honduras 1976-77 4.60 4.81 10.46 Y Iceland 1970-74 3.75 7.20 5.22 Y 1987-88 3.68 8.27 -0.56 N Indonesia 1981-87 7.22 6.01 7.74 N Ireland 1978-81 3.45 4.60 0.77 N Jamaica 1987-90 -2.52 6.56 4.41 Y Lesotho 1975-77 15.79 11.79 8.68 N Malaysia 1978-85 7.12 6.78 6.38 N Mauritania 1973-77 4.24 1.86 2.83 N 1982-84 3.43 2.31 -0.21 N Mexico 1989-94 0.24 3.91 - ? Netherlands 1975-83 4.88 1.65 2.90 N New Zealand 1972-75 1.10 5.61 -1.20 N Nigeria 1978-86 8.35 0.32 4.01 N Norway 1975-77 4.50 5.11 3.50 N 1985-88 2.69 4.86 0.98 N Portugal 1971-78 6.28 5.26 1.59 N Sri Lanka 1980-83 5.14 6.08 4.50 N Sudan 1977-85 2.70 3.07 2.33 N Togo 1966-70 9.45 8.37 4.40 N Turkey 1982-86 2.22 4.74 5.68 Y Uruguay 1969-75 2.24 0.25 4.84 N Zambia 1966-70 3.11 6.51 3.80 Y 1987-90 -3.27 2.57 1.40 Y Zimbabwe 1978-83 3.96 3.14 3.04 N - Not available. Note: Y indicates that growth increases during the boom and is greater on average during and after the boom than before; ? indicates that growth increases during the boom and no data are available for the period after the boom. All other cases are denoted N. Source: Author's calculations based on information in the World Bank's World Saving Database. for which data are available, the evidence is consistent with wealth effects from current and anticipated future economic growth driving the booms. Reverse cau- sation could play a role here, since the emergence of a private consumption boom could stimulate growth during the boom through expansionary effects on aggre- gate demand. Nonetheless, this is unlikely to be all of the story, since the growth rate after the boom exceeded that before the boom in the nine cases judged to be consistent with the wealth hypothesis. In short, for many countries the data are consistent with the hypothesis that booms are driven by wealth effects operating at the national level, in the form of 474 THE WORLD BANK ECONOMIC REVIEW, VOL. 14. NO. 3 favorable movements in the terms of trade. An anticipated acceleration in the rate of economic growth also may have played a role, but in many fewer episodes. The final wealth mechanism explored here operates at the sectoral level, re- flecting the potentially favorable effects on private wealth of an anticipated fu- ture fiscal consolidation, which takes the form of a reduction in the share of public consumption in GDP. The criterion used to judge consistency in this case is the same as that for the other applications-to be judged consistent with the hypothesis, the ratio of public consumption to GDP must fall during the boom and remain, on average, below its pre-boom level during and after the boom. In only a few cases-4 of the 37 booms with the requisite data-are the data consis- tent with an anticipated fiscal consolidation (table 8). Overall, then, the data from individual booms suggest that wealth effects are a more promising avenue for understanding the origins of consumption booms than effects operating through intertemporal relative prices, although the latter may be important in specific circumstances. Current and anticipated increases in the terms of trade may have generated wealth effects that triggered consumption booms in nearly two-thirds of the episodes in the sample. The evidence is sub- stantially weaker for anticipated acceleration in economic growth or anticipated fiscal consolidation. Financial Policies: Booms and Rapid Credit Expansion To consider the potential empirical role of credit expansion in the generation of consumption booms, I examine the extent to which the booms in the sample were associated with an increase in the ratio of the flow of domestic credit to the private sector to GDP. I compute average ratios for periods of equal length before, during, and after each of the consumption booms identified previously (table 9). I judge the data to be consistent with the hypothesis that booms are driven by rapid credit expansion if credit expansion accelerated during the boom and de- celerated after the boom. Of the 31 booms in this sample, in 22 the ratio of the flow of credit to the private sector to GDP increased from the period before the boom to the period during the boom, sometimes dramatically. More impressive is the fact that in 12 of these 22 cases (marked as "consistent" in table 9) the acceleration of credit expansion during the boom was followed by deceleration after the boom, sug- gesting a role for credit in both starting and ending the boom. Still, although lending booms may have been important in a substantial minority of the cases, the predicted temporal association with consumption booms failed to emerge in two-thirds of all cases.6 6. This result may arise in part from the particular credit variable used in the exercise. I use credit to the private sector to proxy for the more appropriate (but unavailable) variable, credit to households. To the extent that lending booms reflect a redirection of credit from firms to households without a corresponding expansion of credit to the private sector as a whole, their role in triggering consumption booms would not be revealed through the use of this proxy. Montiel 475 Table 8. Consumption Booms and Public Sector Consumption Country Boom Before During After Consistent? Burundi 1961-63 0.03 0.04 0.08 N 1979-82 0.13 0.11 0.10 Y Cape Verde 1976-80 0.08 0.11 0.22 N Chile 1978-81 0.17 0.14 0.13 Y Costa Rica 1964-71 0.15 0.16 0.16 N Cote d'Ivoire 1969-70 0.11 0.13 0.16 N 1976-77 0.17 0.19 0.20 N Cyprus 1991-92 0.17 0.18 0.15 N Denmark 1975-79 0.23 0.25 0.27 N Gambia, The 1984-85 0.20 0.15 0.28 N Guinea-Bissau 1971-75 0.10 0.11 0.10 N Honduras 1976-77 0.13 0.14 0.12 N Iceland 1970-74 0.14 0.16 0.17 N 1987-88 0.19 0.19 0.20 N Indonesia 1981-87 0.08 0.10 0.09 N Ireland 1978-81 0.19 0.19 0.19 N Jamaica 1987-90 0.10 0.15 0.12 N Lesotho 1975-77 0.11 0.14 0.18 N Malaysia 1978-85 0.15 0.17 0.15 N Mauritania 1973-77 0.14 0.29 0.26 N 1982-84 0.25 0.18 0.14 Y Mexico 1989-94 0.08 0.08 - N Netherlands 1975-83 0.15 0.15 0.15 N New Zealand 1972-75 0.16 0.16 0.17 N Nigeria 1978-86 0.12 0.17 0.13 N Norway 1975-77 0.19 0.19 0.19 N 1985-88 0.21 0.20 0.21 N Paraguay 1962-69 0.02 0.08 0.11 N Portugal 1971-78 0.08 0.10 0.13 N Sri Lanka 1980-83 0.12 0.11 0.13 N Sudan 1977-85 0.18 0.12 0.12 Y Togo 1966-70 0.14 0.13 0.16 N Turkey 1982-86 0.07 0.07 0.08 N Uruguay 1969-75 0.10 0.10 0.11 N Zambia 1966-70 0.16 0.22 0.30 N 1987-90 0.26 0.21 0.29 N Zimbabwe 1978-83 0.10 0.19 0.26 N - Not available. Note: Y indicates that the share of public consumption falls during the boom and is lower on average during and after the boom than before, and N denotes all other cases. Source: Author's calculations based on information in the World Bank's World Saving Database. The conclusion that financial sector developments, either in the form of domes- tic policies or external financial shocks, may have played a role in triggering a significant number of booms must be tempered by the strong likelihood of reverse causation in this case. Booms emerging for other reasons obviously could result in greater private demand for credit, which would appear as an increase in the ratio of the flow of credit to the private sector to GDP. This possibility seems more likely in this case than in that of the other hypotheses examined. It also is consistent with 476 THE WORLD BANK ECONOMIC REVIEW, VOl. 14, NO. 3 Table 9. Consumption Booms and Credit Expansion Country Boom Before During After Consistent? Bangladesh 1983-89 0.017 0.034 0.021 Y Burundi 1979-82 0.003 0.027 -0.007 Y Cape Verde 1976-80 0.084 0.073 0.094 N Chile 1978-81 0.077 0.162 0.183 N Costa Rica 1964-71 0.084 0.073 0.094 N C6te d'Ivoire 1969-70 0.019 0.036 0.038 N 1976-77 0.084 0.073 0.094 N Cyprus 1991-92 0.084 0.095 0.095 N Denmark 1975-79 0.054 0.047 0.031 N Gambia, The 1984-85 0.006 0.044 0.018 Y Honduras 1976-77 0.032 0.042 0.031 Y Iceland 1970-74 0.042 0.052 0.079 N 1987-88 0.123 0.087 0.113 N Indonesia 1981-87 0.050 0.026 0.086 N Ireland 1978-81 0.042 0.054 0.027 Y Jamaica 1987-90 0.047 0.063 0.058 Y Malaysia 1978-85 0.036 0.076 0.078 N Mauritania 1973-77 0.025 0.053 0.033 Y 1982-84 0.023 0.032 0.023 Y Mexico 1989-94 0.041 0.076 0.119 N Netherlands 1975-83 0.049 0.067 0.052 Y New Zealand 1972-75 0.010 0.024 0.025 N Nigeria 1978-86 0.012 0.021 0.026 N Norway 1975-77 0.042 0.034 0.035 N 1985-88 0.040 0.104 0.024 Y Portugal 1971-78 0.062 0.102 0.110 N Sudan 1977-85 0.015 0.030 0.025 Y Togo 1966-70 0.011 0.005 0.021 N Turkey 1982-86 0.043 0.059 0.063 N Uruguay 1969-75 0.065 0.061 0.128 N Zambia 1987-90 0.031 0.058 0.044 Y Note: Y indicates that credit expansion was targer during the boom than both before and after, and N indicates all other cases. Source: Author's calculations based on information in the World Bank's World Saving Database. the results of the previous section, in which the availability of credit had no inde- pendent effect on the probability of a boom after accounting for other potential causal factors and in which the domestic credit variable reversed signs after feed- back effects were ameliorated through the lagged specification. An alternative, albeit crude, way to assess the potential role of domestic finan- cial sector policies in generating booms is to examine how frequently financial crises follow consumption booms. Booms arising from excessive lending because of inappropriate financial sector policies are more likely to result eventually in a financial crisis than those arising from other sources. Using a five-year post-boom window and taking the dating of crises from Caprio and Klingebiel (1997), I find that eight of the booms accompanied by rapid credit expansion were followed by financial crises-Bangladesh, Chile, Malaysia, Montiel 477 Mauritania (1982-84), Mexico, Nigeria, Norway (1975-77), and Turkey. None of the countries with booms that were classified as inconsistent with the credit expansion hypothesis experienced a crisis within the five-year window. This pro- cedure is obviously highly imperfect, since crises can have many other causes, and the incidence and timing of a financial crisis after an episode of excessive lending can depend on many factors. But it at least suggests that the coincidence of consumption booms and credit expansion in table 9 may not reflect reverse causation in every case. IV. SUMMARY AND CONCLUSIONS The evidence compiled in the last section examined the potential driving forces of consumption booms one at a time. Tables 10 and 11 bring this information together for all of the consumption booms in the sample. The following con- clusions follow from these tables, as well as from the material in sections II and III. Consumption booms are not all driven by the same factors. No single expla- nation appears to dominate for all, or even for a significant majority, of the booms in the sample. As expected, booms arise in a wide variety of macroeco- nomic circumstances. Nonetheless, some hypotheses seem more applicable than others. Booms appear to be associated most commonly with long-lasting im- provements in the terms of trade. The terms of trade variable is statistically sig- nificant in each of the probit regressions, and a large proportion of individual booms are associated with improvements in the terms of trade. This result is particularly significant because, given that the terms of trade are exogenous for the small countries involved, reverse causation is not an issue. The role of the Table 10. Factors Driving Consumption Booms in Industrial Countries Intertemporal relative prices Wealth Real Nominal Real Terms of Public interest interest exchange trade GDP sector Credit Country Booms rate rate rate shocks growth consumption expansion Canada 1960-63 ? Denmark 1975-79 N N ? N N N N Iceland 1970-74 ? ? Y Y N N 1987-88 N Y Y N N N Ireland 1978-81 N N N N N N Y Netherlands 1975-83 N N ? N N N Y New Zealand 1972-75 Y N N N Norway 1975-77 N N N N N N N 1985-88 N N N N N Y Portugal 1971-78 ? ? ? N N N Note: Y indicates that growth increases during the boom and is greater on average during and after the boom than before; ? indicates that growth increases during the boom and no data are available for the period after the boom. All other cases are denoted N. Source: Author's calculations based on information in the World Bank's World Saving Database. 478 'HF WORLD BANK ECONOMIC REVIEW, VOI. 14, NO. 3 Table 11. Factors Driving Consumption Booms in Developing Countries Intertemporal relative prices Wealth Real Nominal Real Terms of Public interest interest exchange trade GDP sector Credit Country Booms rate rate rate shocks growtlh consumption expansion Bangladesh 1983-89 N N N Y N Y Burundi 1961-63 N N 1979-82 N N N N N Y Y Cape Verde 1976-80 N Y Y N N Chile 1978-81 N N Y N Y N N Costa Rica 1964-71 N Y N N C6te d'lvoire 1969-70 Y N N N 1976-77 ? ? Y N N N Cyprus 1991-92 Y N N Y N N N Gambia, The 1984-85 N N N Y N N Y Guinea-Bissau 1971-75 Y N Honduras 1976-77 Y Y N y Indonesia 1981-87 N N N Y N N N Jamaica 1987-90 N N N Y Y N Y Lesotho 1975-77 Y N N Malaysia 1978-85 N N ? Y N N N Mali 1983-85 N Mauritania 1973-77 Y N N Y 1982-84 N Y Y N Y Y Mexico 1989-94 N N Y N ? N N Nigeria 1978-86 N N ? Y N N N Paraguay 1962-69 Y N SriLanka N N N N N N Sudan 1977-85 N ? N N Y Togo 1966-70 ? N N N Trinidad and Tobago 1982-84 Turkey 1982-86 N N N Y Y N N Uruguay 1969-75 ? N N N Zambia 1966-70 Y Y N 1987-90 Y N N Y Y Y Y Zimbabwe 1978-83 N N ? Y N N Note: Y indicates that growth increases during the boom and is greater on average during and after the boom than before; ? indicates that growth increases during the boom and no data are available for the period after the boom. All other cases are denoted N. Source: Author's calculations based on information in the World Bank's World Saving Database. terms of trade suggests that wealth (euphoria) effects most often account for the emergence of consumption booms. Other factors may also be important. The probit results suggest that booms occurred during periods in which the real exchange rate was relatively appreci- ated. The results for individual booms suggest that booms often have been fol- lowed by real depreciation, consistent with the Dornbusch hypothesis. Booms may also have been driven by an anticipated acceleration of economic growth or by financial sector expansion caused by domestic financial policies or capital inflows. The evidence is weaker for both of these mechanisms, because of Montiel 479 the conflicting results in sections II and III and because of a strong likelihood of reverse causation. The increases in the growth rate of real output and in the share of credit to the private sector that were often associated with the emergence of individual booms are consistent with hypotheses emphasizing these phenomena as driving forces. But neither variable proved to have an independent effect in the probit regressions after other potential causal forces and potential endogeneity were taken into account. Although reverse causation is an obvious interpreta- tion, it may not be a complete explanation; the acceleration in economic growth was not restricted to the boom period, but often followed it, and when lending booms were associated with consumption booms, consumption booms tended to be followed by financial crises, suggesting a potential causal role for financial market policies. Factors relying oil changes in intertemporal relative prices arising from inter- est rate movements appear to have a more restricted scope in triggering booms. The evolution of real and nominal interest rates did not often accord with such hypotheses, and where it did, the magnitudes involved were not generally large enough to suggest that intertemporal reallocation of consumption may have cata- lyzed the booms. Of course, these hypotheses were designed to apply only in the specific circumstances of exchange rate-based stabilizations. The upshot is that they are not likely to provide an understanding of the emergence of consumption booms outside of these specific circumstances. A similar conclusion applies in the case of mechanisms relying on Ricardian effects on consumption. Relatively few of the consumption booms in the sample were followed by fiscal consolidation in the form of reduced public consump- tion. Again, this factor may have been important in specific cases. But booms do not seem to arise from the (correct) anticipation of reduced future tax liabilities due to fiscal consolidation. Finally, disentangling the forces behind specific consumption booms usually will involve much more than determining whether or not the data are consistent with the specific macroeconomic mechanism underlying any single explanation. In many cases the data will prove to be consistent with more than one hypothesis. This is true of 15 of the 40 booms in tables 10 and 11. The scope for competing explana- tions may be narrowed by digging just a little deeper in some of these cases. But, more generally, understanding the factors underlying any single consumption boom, and thus understanding the policy implications of the boom, is likely to require quantifying the separate contributions made by more than one factor. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Calvo, Guillermo. 1989. "Incredible Reforms." In Guillermo Calvo, Ronald Findlay, Penti Kouri, and Jorge Braga de Macedo, eds., Debt, Stabilization, and Development. Ox- ford: Basil Blackwell. 480 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Caprio, Gerard, and Daniela Klingebiel. 1997. "Bank Insolvency: Bad Luck, Bad Policy, or Bad Banking?" In Michael Bruno and Boris Pleskovic, eds., Annual World Bank Conference on Development Economics 1996. Washington, D.C.: World Bank. Deininger, Klaus, and Lyn Squire. 1996. "A New Data Set Measuring Income Inequal- ity." The World Bank Economic Review 10(September):565-91. Dornbusch, Rudiger. 1985. "External Debt, Budget Deficits, and Disequilibrium Exchange Rate." In Gordon W. Smith and John T. Cuddington, eds., International Debt and the Developing Countries, pp. 213-35. Washington, D.C.: World Bank. Eichengreen, Barry, Andrew K. Rose, and Charles Wyplosz. 1995. "Exchange Market Mayhem: The Antecedents and Aftermath of Speculative Attacks." Economic Policy 21(October):249-312. Montiel, Peter J. 1998. "Consumption Booms." Williams College, Department of Eco- nomics, Williamstown, Mass. Processed. Rebelo, Sergio, and Carlos A. Vegh. 1996. "Real Effects of Exchange Rate-Based Stabilizations: An Analysis of Competing Theories." In Ben Bernanke and Julio Rotemberg, NBER Macroeconomics Annual, pp. 125-73. Cambridge, Mass.: National Bureau of Economic Research. Reinhart, Carmen M., and Carlos A. Vegh. 1995. "Nominal Interest Rates, Consump- tion Booms, and Lack of Credibility: A Quantitative Examination." Journal of Devel- opment Economics 46(April):357-78. Rodriguez, C. A. 1982. "The Argentine Stabilization Plan of December 20th." World Development 10(September):801-11. THE WORL.D BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 481-51)7 Saving Transitions Dani Rodrik This article takes a systematic cross-national approach to identifying saving transitions- defined as sustained increases in the saving rate of S percentage points or more-to study their determinants and to reexamine the question of causality between growth and saving. Countries that undergo saving transitions do not necessarily experience suis- tained increases in their growth rates. In fact, growth rates typically return to their levels before the transition within a decade. By conitrast, couintries that undergo growth tran- sitions-arising from improved terns of trade, increased domestic investment, or other souirces-do end up with permanently higher saving rates. Hence saving transitions do not appear to be caisal with respect to superior economic performance. Capital accumulation is the proximate source of economic growth. Physical in- vestment is generally the most robust correlate of long-run growth, even though the relationship between investment and growth tends to be weak in the short run. ' As a matter of accounting, investment has to be financed by saving, from either domestic or foreign sources. In only a few high-investment countries has foreign saving accounted for more than 20 percent of investment over long stretches of time. In an economy investing, say, 30 percent of its gross domestic product (GDP), relying on foreign saving beyond this limit would imply running a persis- tent current account deficit in excess of 6 percent of GDP, which would be court- ing disaster. Hence the critical importance of domestic saving in economic growth follows from a few straightforward facts of economic life. Indeed, differences in saving rates clearly distinguish thriving from stagnant economies. During 1984-94, 31 countries had average annual per capita GDP growth rates of 2.5 percent or higher. In these successful countries the median saving rate was 24 percent.2 By contrast, the median saving rate stood at 16 percent in the 59 countries in which per capita income grew at less than 1 percent 1. See Easterly (1997) on the short-run relationship, and Levine and Renelt (1992) on the long-run relationship. 2. Unless otherwise mentioned, all saving rates in this article refer to the ratio of gross national saving to gross national disposable income, as defined in the World Bank's World Saving Database. Dani Rodrik is a professor at the John F. Kennedy School of Government at Harvard University. His e- mail address is dani rodrik@harvard.edu. This article was prepared as part of the World Bank research project "Saving Across the World" and was funded by the World Bank. The author is grateful to Luis Serven for detailed suggestions, Chad Steinberg for excellent research assistance, and Joanna Veltri for editorial and research help. He is also indebted to Francois Bourguignon and two of the three anonvmous referees for helpful suggestions that improved the article. O 2000 The International Bank for Reconstruction and Development/THE WORLD BANK 481 482 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 a year. Assuming that all domestic saving translates into domestic investment and that the long-run incremental capital-output ratio is around 5, virtually all of the gap in growth between these two groups of countries can be attributed to the difference in their saving performance. Such comparisons, however, tell us little about the underlying economics of high growth and the policies that enable it. High-growth countries share many characteristics other than high saving and investment: they tend to have lower inflation rates, smaller budget deficits, better human resources, lower current account deficits, and higher shares of trade in GDP. Which of these factors, if any, are the real determinants of growth? Must a country have them all, or are some the consequence of growth? And even if we accept the causal role played by investment, are increases in saving sufficient and necessary for investment and growth? How likely are saving transitions to result in higher growth? Finally, to the extent that saving is responsible for investment and growth, which policies and institutional arrangements generate increased saving? The empirical literature on saving has three strands. One line of research fo- cuses on the cross-national determinants of saving, applying econometric tech- niques to large cross-national or panel data sets (Giovannini 1985, Edwards 1996, Harrigan 1996, and Loayza, Schmidt-Hebbel, and Serven 1998). This research emphasizes factors that can be quantified, in particular demographic conditions, fiscal policy, financial depth, and economic growth itself. Its initial focus was on the role of deposit interest rates in mobilizing saving. Partly because of negative findings, attention has recently turned to a broader set of structural and institu- tional determinants. A second strand of the literature focuses on the question of causality between saving and growth (Carroll and Weil 1993 and Attanasio, Picci, and Scorcu 1997). There are strong hints in this research that growth drives saving rather than the reverse, especially over short horizons. This result has led some analysts to sug- gest that saving should not receive high priority in designing growth strategies: it is thought that once the obstacles to growth are removed, the response of saving could be nearly automatic (see, for example, Gavin, Hausmann, and Talvi 1996). Finally, several analytical case studies focus on high-saving countries or those that have undergone transitions to become high-saving countries-such as Japan, the Republic of Korea, Taiwan (China), and Chile-to uncover the determinants of saving and growth transitions in specific settings (Marfan and Bosworth 1994, Hayashi 1986, and Rodrik 1995). This strand of research reinforces some of the findings of the cross-national regressions-the importance of demography, for ex- ample--and points out idiosyncratic conditions, such as investment subsidies, as in Korea and Taiwan (China), or pension-system reforms, as in Chile. This article relates to all three strands. I focus on countries that have under- gone sustained saving transitions, which I define more precisely below. The ob- jective is to understand the causes and consequences of saving transitions. Thus the article has a natural link to the case study literature. However, I take a sys- tematic cross-national approach to identifying saving transitions and discover Rodrik 483 many cases that received scant attention in the past. This approach allows me to reexamine the question of causality in the relationship between growth and sav- ing using a different approach and a longer time horizon than earlier studies. The central message of this article is summarized in figure 1. The top panel of the figure shows the median saving and growth rates in the group of 20 countries identified as having experienced saving transitions. I define a saving transition as a sustained increase in the saving rate of 5 percentage points or more (subject to certain other restrictions discussed later). Saving transitions are associated with only temporary increases in economic growth. After a decade or so growth rates tend to return to their levels before the saving transition, even though saving rates remain high. The analogous picture for growth transitions-a sustained increase in the growth rate of 2.5 percentage points or more-is a striking con- trast. Growth booms are associated with permanent increases in saving rates. Taken together, the two pictures underscore the insignificance of saving as a causal factor of long-term growth. High saving rates tend to be the outcome of high growth-regardless of the channel through which high growth is attained- and not a determinant of it. The second half of the article summarizes the evidence from Korea, Taiwan (China), Singapore, Mauritius, and Chile. These cases suggest that idiosyncratic factors often drive sustained transitions in growth and saving. Changes in poli- cies and institutions that enhance the productivity of domestic output and raise the return to domestic investment are frequently the crux of the matter. I. DEFINING A SAVING TRANSITION The central problem in the theory of economic development, wrote W. Arthur Lewis (1954: 155), "is to understand the process by which a community which was previously saving and investing 4 or 5 percent of its national income con- verts itself into an economy where voluntary saving is running about 12 to 15 percent of the national income or more."3 Saving transitions, Lewis thought, are key to economic development. Countries with the most successful records of growth in the postwar period have indeed gone through spectacular saving transitions. Consider the examples of Korea and Botswana. In Korea the saving rate was barely more than 10 percent in the early 1960s. By the mid-1970s it had risen to more than 20 percent, and by the late 1980s it was more than 30 percent. Botswana's saving rate has been more erratic, although rising from 11 percent in 1971 (the earliest year for which the World Bank's World Saving Database pro- vides a figure) to more than 30 percent in the mid-1980s and reaching 53 percent in 1989 before declining thereafter. Lewis would have been astonished to see saving rates rise so high, but not surprised to learn that these two countries were at the top of the economic growth league in the past three decades. 3. Lewis's answer is based on the classical model, emphasizing the functional distribution of income: as the profit share of national income rises, the rate of aggregate saving rises alongside it. 484 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Figure 1. Saving and Growth Booms Saving rate Savinig Booms Growth rate (percent) (percent) 30 4.5 25 ~~~~~~~~~~~~~~~~~~~~~~4.0 2.___ngaving rate 35 20 3.0 2.5 15 2.0 10 1.5 Growth (relative to world) 1.0 5 0.5 0 0.0 [T-5, T-1] T T+Y4 e [Tr5,7:-9[ [7T+10,7+-141 Year Saving rate Growth Booms Growth rate (percent) (percent) 8 4 Saving rate (relative4 7 ~~~~~~~~~~~~~~~to world). 6-70 5- Growth -2 4 = 4 3 2 0- -10 IT-5ST-11 [TT+41 [T+5,TT+9] [T+10,T+14] Year ,Note, Trefers to transition vear. Source: Author's calculations. The term "saving transition" immediately conjures up thoughts of a handful of East Asian countries as well as a few others, such as Botswana and Chile since the second half of the 1980s. In this article I look at saving transitions more systematically by applying a common definition to the cross-national data. By doing so I look beyond the usual suspects and avoid the optical illusions pro- duced by focusing on a narrow set of countries. My definition of a saving transition is inspired by Lewis. As noted, I define a transition as a sustained increase in the saving rate by more than 5 percentage points of national income. To make this definition operational, I apply the fol- Rodrik 485 lowing filter to the saving-rate time series for each country. A country is said to undergo an investment transition in year T if the three-year moving average of its investment rate over a nine-year period starting at T exceeds by more than 5 percentage points the five-year average of its investment rate prior to T. I exclude from the analysis countries that received large resource windfalls, such as the major oil-exporting countries. I also exclude cases in which the saving rate after transition remained less than 10 percent. More precisely, I define ST as the three-year moving average of the saving rate with year T as the first year of the average and 5T as the five-year moving average with year T as the terminal year. For example, S1965 corresponds to the average for the years 1965, 1966, and 1967, while S1965 iS the average for the years 1961- 65. Applying the filter amounts to searching through the data for occurrences of any T such that the following are true: (1) 5T+i > T-l + x for all i = 0,1 n (2) ST+i > 0.10 for all i = 0,1,...,n where the parameter x stands for the threshold increase in the saving rate (set to 0.05), and n captures the length of the horizon over which the transition is ex- pected to be sustained. With a nine-year horizon starting at year 0, n = 6. The first of these conditions checks that the (moving average of the) saving rate after year Texceeds the average prior to T by more than 5 percentage points. The second condition ensures that the average saving rate after the candidate transition year exceeds 10 percent. If these conditions are satisfied for more than a single year in any country, I check to see whether 10 years or more separate the dates. If not, I assume that there is a single transition and designate the earliest year in the sequence as the transition year.4 This kind of a definition does a much better job of capturing cases in which the saving rate increases sharply in a relatively short time span than cases in which it rises steadily, but gradually. The advantage of this approach is that I am able to identify instances in which saving behavior exhibits a sharp break with the recent past. Methodologically, it allows me to identify the date of transitions with better precision, getting a better handle on their antecedents and consequences. By setting x sufficiently low, I am able to capture more gradual transitions (see the sensitivity analysis), but at the cost of blurring the meaning of a transition. The World Bank's World Saving Database covers the years 1960-95.5 How- ever, given the leads and lags involved in the definition of a transition, the earliest possible year for a transition is 1965, and the latest year is 1987. I find a total of 4. If I did not do this, countries would be listed with multiple transition years. However, with the values of n and x selected for the central case (x = 0.05; n = 6), I find no multiple transitions. 5. Of course, not all countries have coverage throughout the entire period, and most countries do not have data for 1995. Choosing a definition of saving that is less appropriate theoretically, such as gross domestic saving as a ratio of GDP, would have increased the available number of observations. Preliminary work indicates, however, that this would not affect the qualitative conclusions. 486 THE WORLI) BANK ECONOMIC REVIEW, VOL. 14, NO. 3 20 transitions (table 1). There are two cases in the 1960s (Portugal 1965, Panama 1968), thirteen in the 1970s (China 1970, Egypt 1974, Jordan 1972, Lesotho 1977, Malta 1975, Pakistan 1976, Paraguay 1972, Philippines 1972, Singapore 1971, Sri Lanka 1976, Suriname 1972, Syria 1973, and Taiwan [China] 1970), and five in the 1980s (Belize 1985, Chile 1985, Costa Rica 1983, Korea 1984, and Mauritius 1984). The list includes many well-known cases, such as Korea, Singapore, Taiwan (China), China, Chile, and Mauritius, as well as several sur- prises (to me at least).6 The transition years generally accord with conventional wisdom regarding the better-known cases. However, Korea's transition date, 1984, is rather late. This is because prior to 1984 Korea's saving rate was increasing at a steady but slow pace, and a filter that requires a jump of 5 percentage points does not pick up such a transition. When the threshold is lowered to 4 percentage points, Korea is listed as having two transitions, one in 1965 and another in 1975. In general, however, the selection and dating of transitions are not very sensi- tive to the thresholds used in operationalizing my definition. The remaining col- umns in table 1 show alternative transition dates for different values of x and n. The second column reduces by two years the horizon over which the increase in saving must be maintained. This change results in a second transition for Mauritius (1971) and Suriname (1972) as well as 12 transitions in previously unlisted coun- tries. Among the additions, the case of Uganda (1988) is particularly intriguing, in view of the significant reforms that this country has undertaken since 1987. The third column raises the required increase in the saving rate to 7.5 percent- age points (while keeping the horizon the same as in the second column). This change reduces the total number of transitions to 16, knocking out some clear- cut cases like Taiwan (China) and Pakistan. Finally, the fourth column restores the original horizon but lowers the threshold to 4 percentage points. This change includes a few new countries such as Hong Kong (1971), Malaysia (1973), and Turkey (1984) and also dates Korea's transitions differently. A certain arbitrariness in the definition of a saving transition is unavoidable. Each version of the filter I use reveals its own anomalies. If the threshold value for the rise in saving is set high, we overlook cases of gradual, but sustained, increases in saving; if it is set low, we include too many instances of simple vola- tility. If the horizon is kept long, we lose countries with transitions in the 1980s; if it is too short, we pick up many increases in saving that are temporary. In view of these tradeoffs the list of transitions in the first column of table 1 strikes me as a good starting point. II. THE CONTOURS OF SAVING TRANSITIONS The usual pattern of a saving transition is represented in figure 2. The typical jump in the saving rate around year 0 (the transition year) is much larger than 5 6. Botswana is not included in the table because the discovery of diamonds classifies it as a resource- boom countrv. Rodrik 487 Table 1. Saving Transitions Using Different Filters Transition year x S percent, x = 5 percent, x = 7.5 percent, x = 4 percent, Economy n=6 n=4 n=4 n=6 Belize 1985 1985 1985 1985 Chile 1985 1985 1986 1985 China 1970 1970 1970 Costa Rica 1983 1983 1983 Egypt, Arab Rep. of 1974 1974 1974 Jordan 1972 1972 1973 1972 Korea, Rep. of 1984 1984 1985 1965, 1975 Lesotho 1977 1977 1977 1977 Malta 1975 1975 1975 Mauritius 1984 1971, 1984 1984 1984 Pakistan 1976 1976 1976 Panama 1968 1968 1966 Paraguay 1972 1972 1972 Philippines 1972 1972 1972 Portugal 1965 1965 1965 Singapore 1971 1971 1972 1970 Sri Lanka 1976 1976 1976 Suriname 1972 1972, 1986 1972, 1986 1972 Syrian Arab Rep. 1973 1973 1973 1973 Taiwan (China) 1970 1970 1970 Barbados 1979 1978 Dominican Republic 1973 1973 Gambia, The 1987 1987 Kiribati 1984 1984 Malawi 1972 1972 Mali 1986 Mauritania 1987 Mozambique, Rep. of 1986 1986 Seychelles 1984 St. Lucia 1985 1985 Thailand 1987 1965, 1986 Uganda 1988 Burkina Faso 1984 Cameroon 1976 Hong Kong 1971 Malaysia 1973 Tunisia 1970 Turkey 1984 Morocco 1969, 1984 Number of transitions 20 34 16 33 Note: x is the threshold increase in the saving rate, and n is the length of the horizon over which the transition is expected to be sustained. Source: Author's calculations based on data from the World Saving Database. 488 THE WORLD BANK ECONOMIC REVIEW, VOL. 14. NO. 3 Figure 2. Saving Transitions Percent 28 26 24 _ 22 20 18 10 -14 -12 -10 -8 -6 4 -2 0 2 4 6 8 10 12 14 16 18 20 Year Note: Figure shows mean and median saving rates for the 20-country sample. The data are shown in transition time, so that year 0 corresponds to the transition year in each country, year 1 is the first year thereafter, and so on. Source: Author's calculations. percentage points. The median saving rate in our sample goes from 14 percent in the five years before the transition, to 23 percent in the next five years, and to 25 percent in the five years thereafter. The most spectacular cases are those of Belize (an increase from 12 to 24 percent), Lesotho (from 9 to 22 percent), and Suriname (from 20 to 41 percent; table 2). Suriname, however, eventually shows an equally spectacular reversal. Egypt, the Philippines, Portugal, and Syria display similar reversals. In each case the saving rates in years [T + 10, T + 14] fall back to their levels before the transition. Since the reversal takes place after a long lag, however, it makes sense to keep these countries in the sample. In the remaining countries saving rates are substantially higher 10 to 15 years down the line-in some instances by more than 20 percentage points (such as Lesotho and Singapore). One respect in which this sample seems distinctive is the significant role of workers' remittances in many of the countries. Seven countries in the sample received remittances in excess of 1 percent of gross national product (GNP) over sustained periods: Belize, Egypt, Jordan, Malta, Pakistan, Portugal, and Sri Lanka. Remittances were particularly large in Jordan, Egypt, Pakistan, and Portugal. These countries have supplied significant amounts of labor to booming econo- mies nearby-the oil-rich Gulf states in the case of Egypt, Jordan, and Pakistan; Germany (former West Germany) in the case of Portugal. If balance-of-payments and national income data are to be believed, since the early 1970s, remittances Table 2. Contours of Saving Transitions (percent) Saving' Investment5 Growthb Transition IT-5, [T, [T+5, [T+ 10, /T-5, fT, [T+5, [T+l 0, [T-5, fT. [T+5, [T+10, Economy year T-1] T+4] T+9] T+14] T-1] 1+4] T+9] T+14] T-1] r+41 T+9] T+14] Belize 1985 11.5 23.5 24.8 -2.0 1.0 6.7 0.4 4.5 2.9 -2.6 Chile 1985 7.8 18.0 25.4 25.8 -5.3 2.0 5.9 5.8 -3.5 4.7 4.0 3.7 China 1970 22.6 29.1 32.6 34.6 0.7 6.3 8.0 10.0 2.6 4.1 1.7 6.8 Costa Rica 1983 13.6 20.9 22.2 23.1 3.2 4.7 4.8 5.9 -4.0 2.3 2.2 1.7 Egypt 1974 11.7 18.6 17.8 12.2 -8.9 3.1 4.7 2.6 -0.7 4.4 5.1 2.6 Jordan 1972 10.2 18.1 24.7 17.0 -7.8 -6.3 0.7 0.2 0.3 Korea, Rep. of 1984 24.7 33.2 35.8 35.0 7.6 9.1 14.6 14.5 3.0 6.9 4.4 5.7 Lesotho 1977 8.5 22.3 22.2 32.8 -11.6 -8.6 -3.2 5.1 8.7 1.8 -0.8 0.7 Malta 1975 19.5 26.6 27.8 25.6 0.0 -2.7 0.8 3.6 2.8 8.0 1.2 1.5 Mauritius 1984 14.5 24.6 26.8 24.9 -1.4 3.5 7.5 9.8 -2.0 4.4 3.0 0.8 Pakistan 1976 10.4 17.7 22.6 21.6 -8.6 -8.5 -7.2 -4.9 -1.0 3.3 4.7 1.4 Panama 1968 15.5 22.8 22.8 21.5 -1.8 6.7 7.8 5.3 3.1 0.1 -2.0 4.6 Paraguay 1972 11.7 18.0 21.4 18.8 -5.4 -2.3 3.9 1.4 -0.6 2.3 7.3 -2.8 Philippines 1972 20.0 26.0 26.9 20.0 -0.3 3.0 5.1 0.3 -0.7 1.6 1.3 -3.9 Portugal 1965 20.3 25.6 28.6 20.9 2.4 1.9 4.2 1.5 1.5 0.9 2.0 -2.0 Singapore 1971 17.4 25.3 33.6 42.3 6.0 19.2 17.3 24.6 6.2 4.6 2.9 5.4 Sri l anka 1976 11.9 17.0 19.6 18.2 -6.8 -3.4 2.4 -0.1 -1.1 1.0 3 1 0.2 Suriname 1972 20.2 40.6 29.3 11.0 4.8 13.7 3.3 -4.1 5.5 1.9 -0.6 5.3 Syria 1973 12.2 23.2 22.7 14.1 -6.1 -0.9 -2.8 -0.1 4.6 3.9 3.6 -3.4 Taiwan (China) 1970 22.4 30.4 31.9 32.0 2.3 6.9 5.8 3.3 4.5 4.8 5.2 5.1 Median 14.1 23.4 25.1 21.6 -1.6 2.5 4.8 3.3 1.5 3.9 2.9 1.5 Mean 15.3 24.1 26.0 23.8 -2.0 2.4 4.5 4.5 1.5 3.4 2.7 1.6 Note: T is the first year of transition for each country. a. Gross national saving relative to gross national disposable income. b. Relative to the world average. Source: Author's calculations based on data from the World Saving Database. 490 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 have averaged 18 percent of GNP in Jordan, 8 percent in Egypt, 5 percent in Pakistan, and 8 percent in Portugal.7 In most of these countries the saving transitions and the initial spurt in remit- tances were closely synchronized. This was certainly true of Egypt, Jordan, Paki- stan, and Sri Lanka, all of which benefited from the boom in the oil-producing states during the 1970s. The case of Portugal, which benefited from the German boom of the 1960s, is analogous. Hence remittances appear to have been an important determinant of saving transitions in a subsample of our countries. Since it is a relatively exogenous factor, the availability of remittances provides a convenient test for determining whether a rapid increase in domestic saving trans- lates into growth. III. GROWTH, SAVING, AND INVESTMENT In this section I analyze the three-way relationship among saving, investment, and growth for the whole sample of transition countries. The investment rate refers to gross investment as a ratio of gross national disposable income, while the growth rate is the growth of GNP. Saving transitions are associated with noticeable increases in both investment and growth rates (figures 3 and 4). The correlation with investment is particu- larly strong. The median investment rate in the sample is 1.6 percentage points below the world average prior to the saving transition. In the first five years following the transition it increases to 2.5 percentage points above the world average, and in the next five years it rises to 4.8 percentage points above the world average (see table 2). In other words the median investment rate rises by about 6.4 percentage points relative to the world average in countries undergo- ing saving transitions. The growth rate also displays a significant spike around the time of the transi- tion year (figure 4), with its median value rising from 1.5 percent (relative to the world average during the five years preceding the transition) to 3.9 percent (dur- ing the five years thereafter). So, saving transitions are clearly associated with sharp increases in growth rates. But the striking message delivered by figure 4 is that the increase in growth tends to be temporary. Following the initial spike, the growth rate starts to de- cline, and 10 years or so into the transition it is back to the level prevailing in the years prior to the transition. The median growth rate in years [T + 10, T + 14] is 1.5 percent, the same as that in years [T- 5, T- 1] (see table 2). The conclusion is that, on average, saving transitions do not seem to produce lasting increases in growth, even when the rise in saving itself is permanent. Excluding the five coun- tries that eventually experience a reduction in saving rates-Egypt, the Philip- pines, Portugal, Suriname, and Syria-does not affect this conclusion. 7. The source of these data is the World Bank's World Development Indicators 1998. Data are not available for Chile, Costa Rica, Lesotho, Mauritius, Singapore, Syria, and Taiwan (China). Some of these economies also received significant remittances as well. Rodrik 491 Figure 3. Saving Transitions and Investment Saving rate Jnvestment rate (percent) (percent) 28 8 26 ~~~~~~~~~~~~~~~~~~~6 24~~~~~~~~~~~~~~~~~~~~~ 22~~~~~~~~~~~~~~~~~~~~~~~~ 1 ,;e~~~~~~~~~~Iv n 18 ~~~~~~~~~~~~~~~~~~~~0 -2 14 12 10 - -14-12-10 -8 -6 -4 -2 0 2 4 6 8 1) 12 14 16 18 20 Year Note: Figure shows median saving and investment rates for the 20-country sample. The data are shown in transition time, so that year 0 corresponds to the transition year in each country, year 1 is the first year thereafter, and so on. Investment refers to gross investment as a ratio of gross national disposable income. Investment is shown relative to the world average for the relevant year, so as to take out the effect of cycles common to most countries. Son rce: Author's calculations. Figure 4. Saving Transitions and Growth Saving rate Growth rate (percent) (percent) 32 6 5 4 2 2 ~~~~~~~~~~~~~~~~~~~~3 2 h ~~~0 -1 7 i -2 -14-12-10-8 -6 4 -2 0 2 4 6 8 10 12 14 16 18 20 Year Note: Figure shows median saving and growth rates for the 20-country sample. The data are shown in transition time. so that year 0 corresponds to the transition year in each country, year 1 is the first year thereafter, and so on. Growth refers to growth of GNP. It is shown relative to the world average for the relevant year. so as to take out the effect of cycles common to most countries. Source: Author's calculations. 492 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Pakistan may represent the paradigmatic case (figure 5). Pakistan's saving rate went through a sustained increase after 1976, rising from close to 10 percent to more than 20 percent. Until about 1982 this rise was accompanied by a signifi- cant increase in Pakistan's relative growth performance-more than 6 percent- age points relative to the world average. Throughout the 1980s and early 1990s, however, Pakistan's relative performance steadily slipped, to the point where the gap in its favor was eliminated entirely by 1994. The country's actual growth rates (without the benchmarking) show a similar, if less marked, cycle. The figures say nothing about the direction of causality between saving and growth, and the language I use ("accompanied by," "correlated with," and so on) reflects that fact. A plausible hypothesis is that causality runs from growth to saving. What we observe around the transition dates could be an increase in saving resulting from an increase in growth, where growth is the product of de- terminants other than saving. IV. GROWTH TRANSITIONS AND SAVING BEHAVIOR One way of gaining more insight into this issue is to reverse the direction of the exercise and look for saving patterns in countries that undergo sustained growth transitions. If saving remains high well into the transition, this would strengthen our suspicions that growth is the driving force behind saving. Figure 5. Saving and Growth in Pakistan, 1969-93 Saving rate Growth rate (percent) (percent) 30 7 6 25~~~~~~~~~~~~~~~~~~~~~~~~ 3 15 2 15 V O 10~~~~~~~~~~~~~~~~~~~~ -2 O -3 ANote: Growith refers to growth of GNP. It is shown relative to the world (3-year) average. Source. Author's catCLclations. Rodrik 493 I define a growth transition in a manner analogous to a saving transition. A growth transition is a sustained increase in the growth rate of real GNP by more than 2.5 percentage points. In particular, a country is said to undergo a growth transition at year T if the three-year moving average of its growth rate over a nine-year period starting at T exceeds by more than 2.5 percentage points the five-year average of its growth rate prior to T. I exclude from the analysis coun- tries whose post-transition growth rates average less than 4 percent. I also ex- clude, as before, resource-boom countries.8 The resulting list includes 18 countries (table 3). Many of these countries have also had saving transitions, although the dates for the two kinds of transitions do not always coincide There are also several new countries, such as Bangladesh (1974), Brazil (1966), Ghana (1984), and Thailand (1986). The median growth rate of income in these 18 countries rises from 1.1 percent prior to the transition to 7.0 percent in the five years following the transition and to 7.2 percent in the five years following that period (these are actual growth rates, not relative to the world growth rate). The growth rate eventually falls off somewhat (10 years or more after the transition date), even though it remains higher than the rate before the transition on average. My main interest lies in what happens to the saving rate in the countries expe- riencing increased growth. Benchmarked against the world norm, the 18 coun- tries are significant underperformers on the saving front before the transition begins: their median saving rate is 7.5 percentage points below the world aver- age. After the growth transition, however, their performance steadily improves. The median saving rate rises to 3.7 points below the world average to 1.2 and 2.5 percentage points above the world average in the three five-year periods follow- ing the transition year. Saving performance continues to improve even in years [T + 10, T + 14], when growth slows. The cumulative improvement in median saving (relative to the world average) amounts to a striking 10 percent of na- tional income. The results without benchmarking the saving rate are virtually identical. The conclusion is clear: growth transitions tend to be followed by sig- nificant, and sustained, improvements in saving performance.9 What the data show, therefore, is an interesting asymmetry between saving and growth transitions. A significant increase in either saving or growth is gener- ally accompanied by a contemporaneous increase in the other. But while growth transitions lead to sustained increases in saving rates, saving transitions generate only temporary increases in growth. These findings are in line with the hypoth- esis that it is mainly growth that drives the time-series relationship between the two variables. 8. I also exclude a few very small island economies: Cape Verde, Dominica, Solomon Islands, St. Vincent, and the Grenadines. 9. 1 calculate the median (and mean) values reported in table 4 using the whole sample, including countries for which observations are not available in certain time periods. Excluding such countries and using a consistent set of countries throughout does not change the conclusions in this and the previous paragraph. Table 3. Contours of Growth Transitions (percent) Growtha Savingb Investment' Transition [T-S, [T, [T+S, [T+1 0, [T-5, [T, [T+S, [T+1 0, [T-S, [T, [T+5, [T+10, Economy year T-1] T+4] T+9] T+14] T-1] T+4] T+91 T+14] T-1] T+4] T+9] T+14] Bangladesh 1974 -1.9 4.6 4.8 3.9 -14.4 -17.4 -9.8 -9.0 -13.7 -15.4 -11.1 -9.5 Brazil 1966 4.2 8.0 9.8 6.1 1.0 -1.1 -1.5 -2.0 -2.2 -0.9 1.1 -1.3 Cameroon 1976 3.6 7.2 9.3 -3.2 -12.4 -6.0 1.2 -1.8 -2.9 2.6 4.8 1.0 Chile 1984 -1.1 6.8 8.0 5.7 -8.4 -5.1 6.9 6.8 -4.5 -0.3 5.8 5.8 China 1977 3.0 8.1 10.6 7.8 7.4 13.7 17.1 17.7 5.5 9.4 13.2 14.0 Costa Rica 1983 -1.1 5.0 5.2 4.6 -6.0 2.8 3.5 4.4 3.2 4.7 4.8 5.9 Dominican Republic 1969 2.0 12.3 4.7 3.4 -12.4 -5.9 -3.7 -1.4 -5.0 -2.4 -1.6 -1.2 Ghana 1984 -3.1 5.2 4.3 3.9 -14.3 -11.3 -9.4 -7.6 -19.4 -12.0 -8.3 -6.8 Lesotho 1969 6.4 12.0 11.3 3.0 -13.1 -8.8 4.7 -11.3 -11.8 -6.2 a~. Mali 1985 -2.6 5.8 1.2 7.1 -11.9 -6.7 -1.2 -8.6 -2.9 -1.3 Malta 1966 0.8 8.7 9.3 10.5 4.9 7.5 -1.5 5.9 -1.5 3.3 -1.7 -2.5 Mauritius 1983 0.9 6.3 6.1 4.0 -4.5 4.1 8.3 8.0 0.7 0.9 7.6 9.1 Pakistan 1976 3.4 7.7 6.7 4.6 -11.0 -3.6 4.9 2.7 -8.6 -8.5 -7.2 4.9 Paraguay 1972 4.6 6.9 11.0 -0.6 -7.5 -3.8 0.8 1.1 -5.4 -2.3 3.9 1.4 Philippines 1986 -1.8 5.2 3.1 3.4 0.8 1.2 1.6 -2.0 -0.1 Seychelles 1985 -0.7 5.6 5.3 3.2 12.8 2.9 4.4 1.9 0.0 Syrian Arab Rep. 1969 1.3 7.8 11.3 5.3 -8.0 -5.7 1.8 2.4 -7.7 -6.8 1.1 -2.6 Thailand 1986 5.0 10.0 7.8 6.0 11.0 16.6 6.2 11.5 20.4 Median 1.1 7.0 7.2 4.6 -7.5 -3.7 1.2 2.5 -2.9 -1.4 0.5 -1.2 Mean 1.3 7.4 7.2 4.4 -5.0 -1.5 1.6 2.3 -3.4 -1.7 1.1 0.2 Note: T is the first year of transition for each economy. a. Growth of gross national product. b. Relative to the world economy. Source: Author's calculations based on data from the World Saving Database. Rodrik 495 V. GRANGER CAUSALITY TESTS With the preceding analysis I have examined the dynamic relationship be- tween growth and saving over varying time horizons. The standard procedure for examining issues of time precedence is to use some type of Granger causality test. Such tests are better at picking up short-run leads and lags than long-term relationships. Nonetheless, it is instructive to know whether the conclusions reached are borne out by the evidence from more formal tests of this type. In the working paper version of this article I report results of Granger causality tests on saving, investment, and growth using the sample of countries with saving transi- tions (Rodrik 1998). I run these tests with annual data as well as with five-year averages. The outcomes are revealing. I find strong evidence that growth pre- cedes saving in the pooled annual data for the countries with saving transitions. The evidence using five-year averages is somewhat weaker. As for the reverse relationship, the results indicate, if anything, a negative, perverse effect from saving to growth. Growth Granger-causes saving, while saving (negatively) Granger-causes growth. Putting all the pieces together, the emerging story emphasizes economic growth as the driving force behind the saving transitions observed. Economic growth tends to have a clear positive effect on the saving rate, both in the short run and in the long run. Increases in saving per se do not seem to produce a sustained rise in growth. The typical pattern for countries that undergo saving transitions is that their growth rates eventually return to their levels before the transition. What explains these patterns, particularly the finding that saving transitions produce, at best, temporary growth spikes? One possibility is that we are observ- ing the implications of the Solow growth model. According to Solow's model, a permanent rise in saving would increase the steady-state capital stock, but raise the economy's growth rate only temporarily until a new, balanced growth path (equal to the previous rate of growth) is reached. However, the Solow model tends to adjust too slowly to explain the rapid declines in growth observed from our data. For example, under a typical calibration (carried out in Romer 1996: 22), the half-life of convergence is around 18 years. Romer assumes a capital share of one-third in his calibration. If I use a larger capital share, in accordance with conditions in developing countries, the convergence rate would be slower, rendering the gap between Solow's model and my findings even larger. More- over, the Solow model, with its constant saving rate, cannot explain the sus- tained rise in saving subsequent to growth booms. Assume instead that the saving rate is determined endogenously through intertemporal optimization on the part of households, as in the Ramsey- Cass-Koopmans model. In this model adjustment to a new balanced growth path could be much more rapid than in the Solow model because the saving rate tends to overshoot on the way to the steady state, producing more rapid capital accu- mulation (see Romer 1996: 58-59). A rise in saving brought about by, say, a 496 THE WORLD BANK ECONOMIC REVIEW, VOL.. 14, NO. 3 decline in the discount rate could produce the initial spurt in growth followed by the rapid decline in growth associated with the saving transitions. Could such a model also explain the saving pattern that follows growth booms? Assume that the growth spurt is produced by an increase in the productivity of capital. The higher productivity of capital would not be associated with a rise in the growth rate or the saving rate in the new steady state, because the marginal product of capital must eventually fall. Another explanation is needed for the seemingly permanent increase in saving that follows growth booms. We might look for an endogenous growth model, which can yield higher saving and growth rates in the long run following a positive productivity shock. Or we might appeal to hysteresis in saving behavior. Consider, for example, the consequences of persistence in consumption habits or the consequences of a more sophisticated financial system when higher in- come levels are reached. Under either scenario temporary growth may generate higher saving rates, in the former case because consumption levels do not adjust rapidly enough and in the latter case because high-yield saving instruments are more available. Hence, the observed pattern could be the joint product of hyster- esis and of saving being driven by positive productivity shocks. VI. CASE STUDIES Not all saving transitions lead to high growth in the long run. Only a small number of our sample countries have managed to sustain increased saving and increased investment and growth. How do these virtuous saving-investment- growth cycles get started? What are the respective roles of external factors, gov- ernment policies, and institutional determinants?10 To help answer these ques- tions, I focus in greater detail on a few economies in which such cycles seem to have taken hold. Korea The saving rate in Korea increased steadily from the early 1960s, rising from around 10 percent in 1960 to more than 35 percent by the late 1980s (figure 6). But saving lagged behind investment until the second half of the 1980s. My filter does not pick up a saving transition in Korea until 1984, which is fairly late in view of the sharp pickup in growth during the 1960s. Applying the same filter to the investment rate, I find an investment transition date of 1965-two decades prior to the saving transition. Korea is a prime example of a country in which high saving has been largely the product of high growth-itself the result of an investment boom that began in the early 1960s. 10. Fixed-effects panel regressions reveal that the standard determinants of saving apply equally well (or badly) to the sample of saving transition countries. In particular, I find that national saving is affected positively by lagged income growth, public saving, the terms of trade, urbanization, and foreign aid (the results are available on request). These regularities do not go far in explaining the onset of a saving transition in any of the sample countries, which is why case studies are helpful. Rodrik 497 Figure 6. Savitng and lInvestmenzt in Korea, 1960-94 Percent 45 40 35 Investment 25 20 15 10 o~~~~~~~ 4 Note: Investment is gross domestic investment relative to gross national disposable income. Saving is gross national saving relative to gross national disposable income. Soiirce Author's calculations. What generated the investment boom? In Rodrik (1995) I argue that the boom was largely the result of government policies that substantially increased the private profitability of investment from the early 1960s onward. With the inauguration of President Park, who took power in a military coup in 1961, the investment climate in Korea improved sharply. In addition to eliminating ob- stacles to investment, the government heavily subsidized investment. The chief form of subsidy was the extension of credit to large business groups at negative real interest rates. Korean banks were nationalized after the military coup of 1961, giving the government exclusive control over the allocation of funds in the economy. Investment was also subsidized through the socialization of investment risk in selected sectors. This came about because the government-most notably Presi- dent Park himself-provided an implicit guarantee that the state would bail out entrepreneurs investing in "desirable" activities if circumstances later threatened the profitability of those investments. The government played a direct, hands-on role by organizing private entrepreneurs to make investments that they otherwise may not have made. In the words of Amsden (1989: 80-81), "The initiative to enter new manufacturing branches has come primarily from the public sphere. Ignoring the 1950s, . . . every major shift in industrial diversification in the de- cades of the 1960s and 1970s was instigated by the state." Finally, public enterprises played a very important role in enhancing the prof- itability of private investment. They did so by ensuring that key inputs were 498 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 available locally for private producers downstream. The government established many new public enterprises in the 1960s and 1970s, particularly in basic indus- tries characterized by high linkages and scale economies (Jones and Sakong 1980). Not only did public enterprises account for a large share of manufacturing out- put and investment in Korea, but their importance actually increased during the critical takeoff years of the 1960s. It is true that other countries applied many of the same policies, with much less favorable results. Korea differed in that there was much greater consistency, predictability, and coherence in the application of the investment incentives; the government bureaucracy was less corrupt and more competent; and the educa- tional attainment of the labor force was very high for a country at Korea's level of development. These factors reduced the administrative and rent-seeking costs of the interventions, while enhancing their efficiency. Taiwan (China) and Singapore The saving transitions in Taiwan (China) and Singapore (1970 and 1971, re- spectively) are dated earlier than in Korea, but otherwise the stories are similar. In both economies investment booms that began in the 1960s led growth. Gov- ernment policies that encouraged and subsidized private investment played a criti- cal role in these booms. And in both countries the saving rate eventually over- took the investment rate, but not until the 1980s. In Taiwan (China) an important turning point was the Nineteen-Point Reform Program instituted in 1960. This program contained a wide range of subsidies for investment and signaled a major shift in government attitudes toward invest- ment. The most important direct subsidies came in the form of tax incentives. The Statute for Encouragement of Investment (enacted in 1960 in conjunction with the Nineteen-Point Reform Program) significantly expanded the prevailing tax credit system for investment. The government further expanded these incen- tives in 1965, at which time specified manufacturing sectors (in basic metals, electrical machinery and electronics, machinery, transportation equipment, chemi- cal fertilizers, petrochemicals, and natural gas pipe) were completely exempted from import duties on plant equipment. As in Korea, it was common for the state to establish new plants in upstream industries. These were then either handed over to private entrepreneurs (as in the case of glass, plastics, steel, and cement) or run as public enterprises (Wade 1990: 78). In Singapore investment was also heavily subsidized. According to Young (1992: 21), in 1968 the Singaporean government dramatically expanded its involvement in investment activities, with the Development Bank of Singapore increasing its financial commitments eightfold over a two-and-a-half year pe- riod. The government came to own, directly or indirectly, a substantial share of the economy. The year 1968 also marked the beginning of an investment boom, preceding the saving transition in 1971 (figure 7). The government funded these large investments in part by running surpluses on the current account of its budget and in part by borrowing from the Central Provident Fund. Foreign Rodrik 499 Figure 7. Saving and In vestment in Singapore, 1965-93 Percent 60 40 20 l o ,& pl;\ , C) NI) , \ 7 ^ 9 9 1. 26. Recent evidence for just such an effect in the United States comes from Gordon and Slemrod (1998). 27. We measure this tax ratio as log(CORPTAXR) = log[1+ (tax paid/net profits)]. By using this measure of corporate taxation, we include the influences of variations in depreciation allowances and the differing tax regimes applied to mining companies, absent from the statutory corporate tax rate. For institutional details on tax regimes and depreciation allowances in the manufacturing sector, see Tsikata (I1998). Aron and Muiellbauer 535 from accounting conventions. We include it as Alog( GNDI) on the right side in order to preserve a coefficient of 1 on log(GNDI) in the long-run solution for log(CPgross). Given that crises such as the 1985 debt crisis saw sharp rises in nominal interest rates, which can affect gross profits by, for example, having a negative short-term impact on sales before production responds, we also include changes in the prime rate charged by banks. We estimate an error-correction model of the form in equation 10 for Alog(CPgross/GNDI). The x variables classified as 1(1)-see table 1-are log(WPI/ ULC), log(TOTRGOLD), RPRIME, TAXDIF, RTARIF, and log(CORPTAXR); those classified as 1(0) are log(CAPUT), APRIME, and Alog(GNDI). A general- to-specific testing procedure on annual data for 1971-97 gives the parsimonious equation shown in column 1 of table 3.28 We could accept the hypothesis that the coefficient on log(CP9russ/GNDI)1 was -1, thus simplifying the model to one with a levels-dependent variable. The level of the real prime rate was insignificant. The contribution of the 1(1) regressors weighted by their coefficients is shown in figure 10. Both the gold terms of trade and the ratio of wholesale prices to unit labor costs fell in the 1990s, the latter because of stronger international competi- tion and stronger trade unions. Despite these negative effects, the profit share in national income recovered after 1994, with improved capacity utilization and lower import tariffs, a temporary fall in the ratio of company taxes to profits, and a rise in TAXDIF (which reflects a compositional change with little real significance for the economy as a whole). Turning to the 1(0) regressors, the interest rate terms simplify into a negative effect from the acceleration of the prime rate. In column 2 of table 3, we replace the acceleration in the prime rate by the change in the real rate and achieve simi- lar results.29 Capacity utilization enters as a two-year moving average. We could also accept (but have not imposed) the hypothesis of a coefficient of -0.5 on Alog(GNDI). This would be equivalent to omitting this term and redefining the dependent variable as log(CP9rrss/GNDIma), that is, scaling by the two-year moving average of national income. The coefficient on Alog (WPIULC), while positive, is insignificant; omitting this variable means the output price to average cost effect enters as a one-year lag. These results are consistent with lags in re- porting profits or in production. The tariff and tax effects are current effects, possibly because tax and tariff rates typically are known at the beginning of the production year. The model easily passes various specification tests for lack of residual autocorrelation, heteroskedasticity, a Chow test for stability over a mid-sample split, and Ramsay's RESET specification test. Columns 3 and 4 report estimates for 1971-93 and 1971-89 samples, which also indicate parameter stability. With six 28. We were restricted to estimates from 1971, as ULC data begin in this year. 29. At the suggestion of a referee, we tested to see if these interest rate effects were not largely a proxy for particular events such as the 1983-85 period of exchange rate unification and subsequent debt crisis, when interest rates were particularly volatile. Both sets of results hold up when the 1983-85 data are excluded from the sample. 536 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 3. Corporate Profits Equation Estimates Dependent variable (1) (2) (3) (4) log(CP r..SIGNDI) 1971-97 1971-97 1971-93 1971-89 Regressors Intercept -6.54 -8.78 -7.23 -5.82 (4.34) (3.85) (5.02) (4.16) log(CAPUTMA) 1.26 1.75 1.40 1.32 (3.72) (3.37) (4.37) (3.64) log(CORPTAXR) -0.38 -0.36 -0.41 -0.49 (3.49) (2.47) (4.18) (3.03) RTARIF -2.77 -1.88 -2.48 -2.67 (4.99) (2.53) (4.98) (4.51) log(TOTRGOLD) 0.56 0.58 0.57 0.56 (13.98) (11.26) (15.73) (11.72) log(WPIULC)_1 1.18 1.07 1.02 0.83 (10.21) (7.45) ( 6.59) ( 3.87) TAXDIF 0.72 0.83 0.82 0.99 (5.39) (4.91) (6.37) (4.71) A log(GNDI) -0.63 -0.88 -0.61 -0.64 (3.12) (2.91) (3.41) (3.08) A2 (PRIME) -0.88 -0.81 -0.81) (4.75) (4.81) (4.22 A RPRIME -0.80 (2.58) Diagnostics Standard error 0.0309 0.0396 0.0269 0.0296 R 2 0.982 0.971 0.989 0.986 Adjusted R2 0.975 0.958 0.982 0.975 Durbin Watson 2.49 2.11 2.77 2.66 LM1 (Lagrange multiplier test for 1st order serial correlation) 2.86 0.18 2.86 2.63 LM2 (Lagrange multiplier test for 2nd order serial correlation) 2.62 0.21 3.24 6.52 Ramsey reset test 0.00 0.00 0.01 0.19 [0.96] [0.97] [0.91] [0.68] CHOW test 0.27 0.87 1.20 0.78 [0.97] [0.58] [0.44] [0.71] Note: Absolute values of asymptotic t-ratios are in parentheses. P-values for Chow and Ramsey tests are in square brackets. Definitions of the variables and statistics are given in table 1. Equation 2 tests robustness by replacing Al (PRIME) in equation 1 by ARPRIME. Equations 3 and 4 use a shorter sample, defined by particular regime breaks, to test for the parameter stability of equation 1. Source: Authors' calculations I(1) variables and at most 27 observations, the Johansen procedure (for example, Johansen and Juselius 1990) lacks power. However, a linear combination of I(1) variables from table 3, column 1, easily passes a Dickey-Fuller test for stationarity. IV. CONCLUDING REMARKS AND POLICY IMPLICATIONS In this article we explored the determinants of private saving in South Africa, separately examining personal (or household) and corporate saving behavior from the late 1960s to 1997 and emphasizing links between them. Aron and Muellbauer 537 Figure 10. Decomnposition of the Log Profit Share into the Main (Nonstationary) Determiniants Multiplied by Their Regression Coefficients, 1966-96 log (profit share) (level-adjusted) o.6 0.4 0.2 0.0 -0.2 --0.4' -0.6 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 19886 1988 1990 1992 1994 1996 - Log (gross profits/GNDO) - Gold terms of trade - Tax differential 7 Log (WPI,unit labor cost)(-1) (Tariffs+surcharges)/import-s Source. Table 3, equation 1. We examined personal saving behavior using a quarterly solved-out consump- tion function for households. This allowed a fuller treatment of a range of exten- sions and approximations to theoretical behavior than is usual with simple sav- ing functions. Particular innovations were the inclusion of asset effects, financial liberalization, and income expectations, in addition to the more usual determi- nants of consumption. The main conclusions from this model are fourfold. First, much of the rise in the ratio of consumption to income has been the result of financial liberalization, in part because of reduced down payments for housing purchases. However, the general equilibrium reductions in personal saving due to financial liberalization are substantially smaller than the partial equilibrium effects. This is because fi- nancial liberalization also raised real interest rates, caused household debt to rise relative to liquid assets, and raised unincorporated business income relative to labor income, all of which tend to raise personal saving rates. Second, real inter- est rates have significant direct negative effects on consumption, presumably be- cause of the mix of substitution, income, and user-cost-of-durables effects pre- dicted by economic theory. The estimates throw important light on the monetary 538 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 transmission mechanism, showing that there are multiple channels through which interest rates affect consumption expenditures. In addition to the direct effect, a rise in the real interest rate appears to have even larger indirect effects through asset prices, income, and income expectations. Without official data on the stock of wealth, these apparently large asset effects have not been measured previously. Third, the inclusion of expected income growth clarifies the channel through which fiscal policy is transmitted to personal saving, generating links between personal and government saving rates. This approach is missing in previous work. Fourth, the model suggests that the effect of a permanently higher growth rate on the personal saving ratio in South Africa is likely to be small, given real interest rates and asset-to-income ratios. There is a striking paucity of literature on corporate saving behavior, both in general and specific to South Africa. We examined the determinants of annual net corporate saving relative to national income (GNDI) by disaggregating this ratio into three components and investigating separately the determinants of net corporate saving relative to net corporate income (or profits) and gross corporate profits relative to GNDI. The third component, the ratio of net to gross profits, is taken as given, being largely outside the control of firms. The profit share equation has two groups of determinants. One consists of the components of the price to average cost markup: the share of pretax profits in national income rises with the ratio of wholesale prices to unit labor costs, the gold terms of trade, and capacity utilization (capturing a lower share of fixed costs), but falls with import tariffs, a cost component (see figure 10). The other consists of two tax effects: a rise in the difference between the highest tax rate on individuals and the corpo- rate tax rate raises the profit share, while a rise in the ratio of company taxes to profits has the opposite effect. The first of these tax effects largely reflects the incentives of businesses to incorporate when tax rates on corporations are lower than those on private individuals. The second suggests that higher tax rates on corporations tend to stifle profit-earning activities. The main results from the corporate saving equation are that the rise in the inflation rate between the late 1960s and late 1980s was an important factor in explaining the rise in the share of corporate saving in net profits; whereas with the decline of inflation in the 1990s, the corporate saving rate was bolstered by high real interest rates and liberalization of consumer credit markets. Changes in personal tax rates on dividends play no role in explaining the rise in the corpo- rate saving rate out of net profits, but they help to account for some decline in the 1990s (see figure 9). From these two models, and the log ratio of net to gross profits, we derive a model for the log ratio of corporate saving out of national income, consistent with the corporate saving rate shown in figure 3. For example, we can now un- derstand the 1980 peak to be the result of the peak in the share of profits associ- ated with the gold price boom and not the result of any significant rise in the propensity to save out of profits. There are important links between household and corporate saving. A key question is whether the well-known stability of private saving can be attributed Aron and Muellbauer 539 fully to households piercing the corporate veil. We tested this hypothesis for both the personal and corporate saving models and found evidence of piercing the veil in both cases: household expenditures appear to respond to the value of equities rather than to after-tax dividend payments, whereas the corporate tax rate is influenced by the personal dividend tax rate and by inflation. However, the significant role of inflation in explaining the secular rise in the corporate saving rate out of net profits is explained only partly by corporations serving the tax needs of their shareholders. Other reasons include poor returns on alternative assets in an inflationary environment. Moreover, the role of changing dividend taxation should not be exaggerated quite apart from the fact that, as we have seen, its variations cannot explain the secular rise in the corporate saving rate. From the mid-1980s, most equities held on behalf of individuals were held by pension funds and thus were tax-exempt.30 The above tests do not indicate the extent of sectoral offset. However, even if households saw through the corporate budget constraint perfectly, piercing the veil turns out to be only one factor in explaining these sectoral shifts. Our personal and corporate sector saving models both show clearly the im- portance of many other economic factors besides piercing the veil that change sectoral saving behavior. These fall into two types. First, there are factors that move household and corporate saving rates out of national income in opposite directions, with differing degrees of offset. Perhaps most obviously, the shares of corporate profits and of personal disposable income in national income are nega- tively correlated (see figures 4 and 10). Factors behind this negative correlation are gold booms and other causes of the procyclical pattern of profit shares and the strengthening of union power, which raises labor's share and lowers the profit share. To the extent that there is some stability in the sectoral saving ratios out of sectoral income, the two sectoral saving ratios out of national income will be negatively correlated. There are also factors that can move household and corporate saving rates out of their respective sectoral incomes in opposite directions. The most important of these is financial liberalization. Although there are general equilibrium effects that make it complicated to quantify the degree of offset, our results suggest that financial liberalization reduced personal saving far more than it raised corporate saving. Inflation surprises lower corporate saving and increase measured house- hold saving (through wealth effects and measured disposable income, since mea- sured interest income rises with higher nominal interest rates). Second, there are other factors that cause a positive correlation between the two saving rates out of sectoral income: both increasing or both decreasing. Higher real interest rates positively affect both corporate and personal saving rates, given income and growth. Other factors can be ambiguous. A rise in government saving reduces the personal saving rate out of personal income by 30. Pensions paid out to individuals are subject to income tax. However, pensions are paid in part as a tax-free sum, while income is received at a time when other income is low or nonexistent, so that low marginal tax rates will tend to apply. After 1996 a change in legislation provided for direct taxation of pension fund income (Katz Commission 1996). 540 THE WORLD BANK ECONO1MIC REVIEW, VOI. 14, NO. 3 raising income expectations (which we modeled separately), given current in- come, and also to the extent that it lowers real interest rates. The income ex- pectations effect of a rise in government saving on the corporate saving rate is in the opposite direction, although the real interest rate effect is in the same direction as for households. The net effect of all these factors suggested by our models is thus more com- plex than the simple story of piercing the veil. This view suggests that there are good reasons to be concerned not only with reducing government dissaving but also about the compositional change in private saving and the decline in both private saving ratios from 1995. Our models represent a significant advance on earlier partial equilibrium models of saving. However, to examine the general equilibrium effects for monetary transmission implications, these richer partial models should be included in larger models, in particular to trace through the real interest rate effects working through asset price changes for bonds, housing, and equities. Such work should also ad- dress risk premiums more explicitly than we have been able to do. There have been extensive debates on the effects of financial liberalization on private saving. Our results concur with findings in the empirical survey in Fry (1995: ch. 8): if financial liberalization increases the availability of consumer credit, private saving tends to decline. Policy changes in the new millennium that bear on saving behavior in South Africa include the introduction of capital gains taxation on securities directly held by households. The regime announced is stringent, with minimal allow- ances and no inflation indexing. Since we cannot reject the hypothesis that the corporate veil is pierced, our models predict little impact on total private saving; they suggest some increase in personal saving, with a reduction in corporate sav- ing (since corporate retentions resulting in taxable capital gains for households will now be less attractive to households). An important change is the transition toward inflation targeting and the decision to issue inflation-indexed bonds. These moves aim to improve the transparency and stability of policy, dampen inflation- ary expectations, and reduce risk premiums, thus lowering real interest rates. The fall in the corporate saving ratio since 1995, despite rises in the real inter- est rate, appears consistent with our model (see figure 9 and, for the history of the real interest rate, figure 2). However, if real interest rates are to fall, our partial equilibrium models suggest that both personal and corporate saving rates out of (higher) sectoral incomes will fall still further. Taking further system feed- backs into account, the asset price effects of lower interest rates will reinforce the fall in personal saving, whereas the effect of a higher growth rate on personal saving is probably fairly small (for given ratios of assets to income). For corpora- tions, however, the growth effects on saving from lower interest rates are likely to be larger, and computing their size has great policy relevance. Encouraging corporate saving would seem to be particularly important if the sustainability of foreign saving remains uncertain because of the volatility of capital inflows, if government saving improves only slowly, and if household saving remains low. Aron and Muellbauer 541 Even without embedding our models in a full macroeconometric model, we can draw some policy conclusions-supplementing these models by simple hy- potheses about, for example, the effect on asset prices of higher interest rates. One possible measure related to our model concerns prudential regulation limit- ing the degree to which companies can take on short-term foreign debt. Another concerns corporate taxation and depreciation allowances, where more generous treatment encourages investment and enables companies to earn more profit. Since corporations have much higher saving rates than households, this should raise the private saving rate out of national income. Returning to personal saving, it is clear that despite the direct and indirect effects of high real interest rates in the 1990s, net household saving has fallen to very low levels, mainly because of financial liberalization. The South African financial system encourages personal borrowing to an excessive degree-in the sense that the high interest rates considered necessary to restrain consumer credit and spending also restrained investment and economic growth in the 1990s. At- tention should be given to tightening prudential controls, to allow for the possi- bility of macroeconomic risk or a shock to the financial system, for instance, through a major fall in asset values. Prudential controls not only should stabilize the financial system but also should keep individual default rates reasonably low, particularly for new consumers, many of whom may have little experience in risk management. One possible move would be to impose higher risk weights for mortgage loans with high loan-to-value ratios. Another area for tightening regu- lations concerns the use of pensions for housing collateral. Although current regu- lations may have implications for reducing the housing deficit, forcing income risk, pension risk, and housing risk to be correlated violates the general notion of spreading risk. Housing policies that encourage the rental sector would help to reduce overborrowing by young, middle-income households. Apart from encouraging prudent lending by the financial sector, there may be a role for tax incentives to encourage household saving. Although we have not investigated a separate role for the rate of return on assets, as opposed to the borrowing rate, our evidence is at least consistent with the idea that extending the generous tax treatment of pensions to some other saving products would raise the personal saving rate. Such schemes may help to install a long-term cul- ture of saving, although it is fair to say that empirical evidence for the United States has not found large aggregate saving effects (see, for example, Hubbard and Skinner 1996 and other articles in this volume). REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Aron, Janine, and John N. J. Muellbauer. 2000a. "Financial Liberalisation, Consump- tion, and Debt in South Africa." Working Paper Series 2000.22. Oxford University, Centre for the Study of African Economies, Oxford. Processed. 542 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 2000b. "Inflation and Output Forecasts for South Africa: Monetary Transmis- sion Implications." Working Paper Series 2000.23. Oxford University, Centre for the Study of African Economics, Oxford. Processed. Bandiera, Oriana, Gerard Caprio, Patrick Honohan, and Fabio Schianterelli. 2000. "Does Financial Reform Raise or Reduce Saving?" Review of Economics and Statistics 82(2):239-63. Banerjee, Anindya, Juan Dolado, John W. Galbraith, and David F. Hendry. 1993. Co- Integration, Error Correction, and the Econometric Analysis of Non-Stationary Data. Oxford: Oxford University Press. Barr, Graham, and Brian Kantor. 1994. "The Changing Pattern of Savings in South Af- rica 1970-91." Journal of Studies in Economics and Econometrics 18(3):59-76. Bayoumi, Tamim. 1993a. "Financial Deregulation and Consumption in the United King- dom." Review of Economics and Statistics 75(3):536-39. . 1993b. "Financial Deregulation and Household Saving." Economic Journal 103(421):1432-43. Bond, Stephen, Lucy Chennells, and Michael Devereux. 1995. "Company Dividends and Taxes in the U.K." Fiscal Studies 16(1):1-18. Caggese, Andrea. 1999. "Financial Constraints, Precautionary Saving and Firm Dynam- ics." Financial Markets Group, London School of Economics. Processed. Campbell, J. Y., and Gregory Mankiw. 1989. "Consumption, Income, and Interest Rates: Reinterpreting the Time Series Evidence." In 0. J. Blanchard and Stanley Fischer, eds., NBER Macroeconomics Annual 1989. Cambridge, Mass.: MIT Press. . 1991. "The Response of Consumption to Income: A Cross-Country Investiga- tion." European Economic Review 35(4):723-56. David, P. A., and J. L. Scadding. 1974. "Private Savings: Ultrarationality, Aggregation, and Denison's Law." Journal of Political Economy 82(2):225-49. Davidson, James, D. F. Hendry, Frank Srba, and Stephen Yeo. 1978. "Econometric Mod- elling of the Aggregate Time-Series Relationship Between Consumers' Expenditure and Income in the United Kingdom." Economic Journal 88(352):661-92. Deaton, Angus. 1999. "Saving and Growth." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge: Cambridge University Press. De Kock, Gerhard. 1978. "Commission of Inquiry into the Monetary System and Mon- etary Policy in South Africa, Interim Report." Government Printer, Pretoria. Processed. 1. 985. "Commission of Inquiry into the Monetary System and Monetary Policy in South Africa, Final Report." Government Printer, Pretoria. Processed. Dickey, David A., and Wayne A. Fuller. 1979. "Distribution of the Estimators for Autoregressive Time Series with a Unit Root." Journal of the American Statistical Association 74(366):427-31. Engle, Robert F., and Clive W. J. Granger. 1987. "Cointegration and Error Correction: Representation, Estimation, and Testing." Econometrica 55(2):251-76. Federal Reserve Bank of Kansas. 1999. New Challenges for Monetary Policy. Jackson Hole, Wyoming. Feldstein, M. S. 1970. "Corporate Taxation and Dividend Behaviour." Review of Eco- nomic Studies 37(1):57-72. .1973. "Tax Incentives, Corporate Saving, and Capital Accumulation in the United States." Journal of Public Economics 2(1):159-71. Aron and Muellbauer 543 Fry, Maxwell. 1995. Money, Interest, and Banking in Economic Development, 2d ed. Baltimore and London: Johns Hopkins University Press. Gordon, R. H., and Joel Slemrod. 1998. "Are Real Responses to Taxes Simply Income Shifting between Corporate and Personal Tax Bases?" NBER Working Paper 6576. National Bureau for Economic Research, Cambridge, Mass. Processed. Hall, Robert E. 1978. "Stochastic Implications of the Life Cycle-Permanent Income Hy- pothesis: Theory and Evidence." Journal of Political Economy 86(6):971-87. Hall, Robert E., and Frederick S. Mishkin. 1982. "The Sensitivity of Consumption to Transitory Income: Estimates from Panel Data on Households." Econometrica 50(2):461-81. Hendry, D. F., and Thomas von Ungern Sternberg 1981. "Liquidity and Inflation Effects on Consumer's Expenditure." In Angus Deaton, ed., Essays in Theory and Measure- ment of Consumers' Behaviour. Cambridge, U.K.: Cambridge University Press. Honohan, Patrick. 1999. "Financial Policies and Saving." In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge: Cambridge Uni- versity Press. Hubbard, R. Glenn, and Jonathan S. Skinner. 1996. "Assessing the Effectiveness of Sav- ing Incentives." Journal of Economic Perspectives 10(4):73-90. Jappelli, Tullio, and Marco Pagano. 1994. "Saving, Growth, and Liquidity Constraints." Quarterly Journal of Economics 109(1):83-109. Johansen, Soren, and Katarina Juselius. 1990. "Maximum Likelihood Estimation and Inference on Cointegration-with Applications to the Demand for Money." Oxford Bulletin of Economics and Statistics 52(2):169-210. Kaldor, Nicholas. 1957. "A Model of Economic Growth." Economic Journal 57(268): 591-624. Katz Commission. 1996. Third Interim Report of the Commission of Inquiry into Cer- tain Aspects of the Tax Structure of South Africa. Pretoria: Government Printer. King, M. A. 1977. Public Policy and the Corporation. London: Chapman and Hall. Kingdon, Geeta G., and John B. Knight. 2000. "Are Searching and Non-searching Unem- ployment Distinct States When Unemployment Is High? The Case of South Africa." Working Paper Series 2000.2. Oxford University, Centre for the Study of African Econo- mies, Oxford. Processed. Klasen, Stephan. 1997. "Poverty and Inequality in South Africa." Cambridge University, King's College, Centre for History and Economics, Common Security Forum. Processed. Koopman, S. J., A. C. Harvey, J. A. Doornik, and Neil Shephard. 1995. Structural Time Series Analyser, Modeller and Predictor (STAMP). London: Chapman and Hall. Lucas, Robert E. 1976. "Econometric Policy Evaluation: a Critique." Carnegie-Rochester Conference Series Supplement to the Journal of Monetary Economics 1:19-46. . 1981. Studies in Business Cycle Theory. Cambridge, Mass: MIT Press. MacKinnon, James G. 1991. "Critical Values for Co-Integration Tests." In Robert Engle and Clive Granger, eds., Long-run Economic Relationships: Readings in Co-Integra- tion. Oxford: Oxford University Press. Miles, David. 1994. Housing, Financial Markets, and the Wider Economy. New York: Wiley. Moll, Peter. 1998. "Discrimination Is Declining in South Africa, But Inequality Is Not." Working Paper 5. South African Network for Economic Analysis, School of Econom- 544 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 ics, University of Cape Town, Cape Town. Processed. Mouton Report. 1992. "Report of the Committee of Investigation into a Retirement Provision System for South Africa." Government Printer, Pretoria. Processed. Muellbauer, John N. J., and Janine Aron. 1999. "Estimates of Personal Sector Wealth for South Africa." Working Paper Series 99.17. Oxford University, Centre for the Study of African Economies, Oxford. Processed. Muellbauer, John N. J., and Ralph Lattimore. 1995. "The Consumption Function: A Theoretical and Empirical Overview." In Hashem Pesaran and Michael Wickens, eds., Handbook of Applied Econometrics. Oxford: Blackwell. Muellbauer, John N. J., and A. J. Murphy. 1995. "Explaining Regional Consumption in the U.K." Nuffield College, Oxford. Processed. Patterson, K. D. 1984. "Net Liquid Assets and Net Illiquid Assets in the U.K. Consump- tion Function: Some Evidence for the U.K." Economic Letters 14(4):389-95. Poterba, Jim M. 1987. "Tax Policy and Corporate Saving." Brookings Papers on Eco- nomic Activity 2:455-503. .1991. "Dividends, Capital Gains, and the Corporate Veil: Evidence from Britain, Canada, and the United States." In B. D. Douglas and John B. Shoven, eds., National Saving and Economic Performance. Chicago: University of Chicago Press. Pretorius, C. J., and S. Knox. 1995. "Private Consumption Expenditure in the Macro- Model of the Reserve Bank." Quarterly Bulletin, South African Reserve Bank, 196(June):27-40. Prinsloo, Johan W. 1994. "Private Sector Saving in South Africa." Quarterly Bulletin, South African Reserve Bank no. 193 (September):25-35. =. 1997. "Corporate and Household Saving in South Africa: A Macroeconomic Perspective." South African Reserve Bank, Pretoria. Processed Quattek, Kristina, and Thea Fourie. 2000. "Economic Impact of AIDS on South Africa." ING Barings Global Research. Johannesburg (April). RDP Office. 1995. "Key Indicators of Poverty in South Africa." Pretoria. Processed. Sarno, Lucio, and Mark Taylor. 1998. "Real Interest Rates, Liquidity Constraints, and Financial Deregulation: Private Consumption Behaviour in the U.K." Journal of Mac- roeconomics 20(2):221-42. Schmidt-Hebbel, Klaus, and Luis Serven. 1997. Saving Across the World: Puzzles and Policies. World Bank Discussion Paper 354. Washington, D.C.: World Bank. .1999. The Economics of Saving and Growth. Cambridge, U.K.: Cambridge Uni- versity Press. Tsikata, Yvonne. 1998. "Savings in South Africa: An Empirical Investigation." World Bank, Southern Africa Department, Washington, D.C. Processed. Zellner, Arnold, D. S. Huang, and L. C. Chau. 1965. "Further Analysis of the Short-Run Consumption Function with Emphasis on the Role of Liquid Assets." Econometrica 33(3):571-81. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 545-70 Household Saving in China Aart Kraay China, in recent years the world's largest and fastest growing economy, also has na- tional saviing rates that are among the highest in the world. This article considers a variety of statistical issuies that clouid the measurement of aggregate and household sav- ing in China. It also provides new empirical evidence on the importance of intertemporal considerations in explaining the variation in household saviing across China's provinces. China, which has been the world's largest and fastest growing economy for the past two decades, has saving rates that are among the highest in the world. Since it took the first steps toward economic reform in 1978, China's gross national saving rate has averaged 37 percent of gross national product (GNP), and its economy has expanded at a remarkable 9 percent a year in per capita terms, lifting 200 million Chinese out of absolute poverty. Rapid growth has been accompanied by an equally rapid structural transition, as China has progressed from a primarily rural, agrar- ian, and state-run economy to a more urban and industrial society in which most economic interactions are governed by market forces. These changes have had profound consequences for saving. Consider China's transition from plan to market. Before 1978 China's high saving rates, averag- ing 27 percent of GNP, were engineered by state fiat. Distorted relative prices favored industry, concentrating profits in state-owned enterprises, which then could be directed toward the state's investment priorities. Household incomes were very low, and households accounted for only a small proportion of total saving. Economic reforms since 1978 have transformed public and private sav- ing. Price reform and vigorous competition from collectively owned and pri- vate enterprises have eroded the operating surpluses of state industry and, with them, the importance of public saving. In contrast, rising household incomes and rising household saving rates have made household saving newly promi- nent, with households contributing between a quarter and half of total savings. The transition to a market economy has not only given new importance to households' voluntary consumption and saving decisions, but has also shaped the economic environment in which these decisions are made. In rural areas the collapse of the agricultural commune system and the emergence of more secure Aart Kraay is with the Development Research Group at the World Bank. His e-mail address is akraay@worldbank.org. This article was prepared as part of the World Bank research project, "Saving Across the World: Puzzles and Policies." The author is grateful to the project's organizers, Norman Loayza, Klaus Schmidt-Hebbel, and Luis Serven, for their encouragement and to Nick Lardy, Lihong Wang, the editor of this journal, and three anonymous referees for their helpful discussions. (C 2000 The International Bank for Reconstruction and Development/ THE WORLD BANK 545 546 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 property rights to land and buildings sparked a boom in rural household saving in the form of investment in land and housing. In addition, the proliferation of bank branches even in remote areas and the rapid growth of rural industry have given rural households access to a broader range of assets. Urban households, once covered by generous cradle-to-grave benefits through employment in state- owned enterprises, are finding their futures increasingly uncertain as the finan- cial performance of their employers has weakened. At the same time, growth in the nonstate sector has meant that a small but increasing share of urban workers no longer enjoy the same generous benefits afforded their counterparts in state industries. These factors likely have given urban households strong new motiva- tions to save. Furthermore, as incomes have risen from low levels, especially in rural areas, people are devoting a smaller share to meeting subsistence consump- tion requirements, leaving more income available to save. Against this backdrop, this article makes two contributions. The first, although mundane, is important, since it concerns the measurement of saving. In China, as in many developing countries, substantial statistical difficulties arise when mea- suring saving. In China these difficulties are particularly acute. I discuss discrep- ancies between aggregate saving and its components and between alternative measures of household saving. I examine the implications of these discrepancies for views of saving in China and the relevance for China of standard forward- looking models of consumption and saving. In particular, I provide new evidence on how expectations of future income growth and future income uncertainty, as well as demographic variables and proxies for the importance of subsistence consumption, explain interprovincial differences in household saving rates. Given China's size and its high saving rates, it is not surprising that in recent years considerable theoretical and empirical effort has been devoted to under- standing its saving behavior. This research can be divided into two broad strands. The first emphasizes the relevance for China of traditional equilibrium theories of saving, ranging from simple Keynesian consumption/saving functions to vari- ants of the life-cycle and permanent-income hypotheses.' The second argues that equilibrium theories of saving are unlikely to be relevant in an economy in tran- sition from plan to market. Instead, disequilibrium factors, especially shortages and rationing in goods and credit markets, explain China's saving experience.2 1. Examples of these include estimates of simple Keynesian consumption or saving functions (Wong 1993, Qian 1988, and World Bank 1988), tests of various implications of the permanent-income hypoth- esis (Chow 1985, Qian 1988, Wong 1993, and Wang 1995), and tests of the life-cycle model (Jefferson 1990, Pudney 1991, Dessi 1991, and Modigliani and Cao 1996). Bai, Zhu, and Wang (1993), Yusuf (1994), Zhang (1994), and Arora (1995) offer interesting descriptive analyses of saving in China, but they do not formally test alternative theories of saving. 2. See Feltenstein, Lebow, and van Wijnbergen (1990) and Ma (1993), as well as indirect evidence given by estimates of the effects of rationing in demand systems provided in Wang and Chern (1992); Fleisher, Liu, and Li (1994); and Wang and Kinsey (1994). Direct survey evidence on the availability of consumer goods can be found in Hussain and others (1990). Naughton (1987) offers a critical review of this evidence and suggests that involuntary saving is unlikely to be empirically important other than in the short period immediately following reform. Kraay 547 These two hypotheses have been subject to empirical scrutiny using both aggre- gate and household-level data, with varying success. In addition to its treatment of data issues, this article differs from much of the existing literature on China in that it directly tests the importance of expectations of future income growth and future income uncertainty, following the approach suggested by Carroll and Weil (1994) and Carroll (1994). I. MEASURING SAVING IN CHINA National accounts data show that saving in China has been extraordinarily high over the past 20 years. In this section I briefly review this performance and present rough estimates of the composition of national and household saving. This exercise highlights a number of data problems, frequently ignored in empiri- cal work, that have important implications for understanding saving in China. I then describe the panel of provincial saving data taken from household surveys, which I use in section II. Aggregate Saving Official statistics for China reveal high and rising saving rates and rapid growth over the past 30 years. Between 1965 and 1977 gross national saving rates aver- aged 27 percent of GNP; they climbed to an average of 37 percent between 1978 and 1995. Growth of GNP per capita averaged 6 and 9 percent, respectively, during these periods, although the latter growth rate is probably overstated by 1 to 2 percentage points. China's saving and growth performance is similar to that of other rapidly growing economies in East Asia, although China reached its high saving rates at much lower income levels.3 More striking is the difference between China's saving experience and that of the transition economies of Eastern Europe and the former Soviet Union. Whereas China weathered the early stages of transition with only a small drop in its na- tional saving rate, many of the other transition economies saw sharp declines in national saving rates, mirroring the well-documented collapse of their output.4 The resilience of China's aggregate saving rate throughout the transition process reflects the fact that household saving increased rapidly during the early years of reform, offsetting the decline in public saving directed through the planning and state-enterprise apparatus. In contrast, the collapse in public saving in many Eu- ropean transition economies was paralleled by sharp declines in household saving. More broadly, China's saving rate remains unusually high relative to interna- tional experience, even after controlling for some of the determinants of saving. To illustrate this point, I estimate a cross-section regression of gross national saving rates in a large sample of countries on a set of variables that Loayza, Schmidt-Hebbel, and Serv6n (1998) identify as the "core" determinants of sav- 3. A more descriptive discussion of China's growth and saving experience can be found in World Bank (1997a). 4. See, for example, Denizer and Wolf (1998). 548 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 1. Cross-Country Determinants of Saving Variable Coefficient Standard error Constant -0.050 0.071 Real interest rate 0.012 0.102 Urbanization ratio -0.041 -0.038 Log (real GNDI per capita) 0.039* 0.009 Growth in real GNDI per capita 0.991 * 0.260 M2 to GNDI 0.098* 0.031 Old-age dependency ratio -0.600* 0.159 Young-age dependency ratio -0.028 0.048 Terms of trade -0.035 0.029 Inflation rate -0.032 0.134 Domestic credit flow to GNDI 0.076 0.292 *Significant at the 1 percent level. Note: The dependent variable is gross national saving/gross national disposable income (GNDI). For variable definitions, see Loayza, Schmidt-Hebbel, and Serven (1998). Source: Loayza, Schmidt-Hebbel, and Serven (1998). ing (table 1).5 I then express each of the explanatory variables for China as a deviation from the average across all countries and multiply these deviations by the estimated coefficients. This yields a measure of the extent to which differ- ences in China's saving rate from that of a "typical" country can be attributed to differences in the known determinants of saving. China's saving rate is unusually high given its income level, as indicated by the negative contribution to the predicted saving differential of the natural log of real income per capita (figure 1). In contrast, China's high growth rate and financial depth, as well as its relatively low old-age dependency ratio (people 65 years and older as a proportion of the total population) contribute positively to China's saving differential. Most important, even after controlling for these determinants of saving, China remains a large outlier in the regression, which underestimates China's saving rate by nearly 10 percentage points.6 What underlies these high saving rates? I decompose gross national saving in China since 1978 into public saving, corporate saving, and household saving (figure 2). The data for this decomposition are given in table A-1. This decompo- sition highlights the changing role of government and household saving over the course of China's economic reform, and it draws attention to a number of anoma- lies in the data that cloud the measurement of saving in China. Public saving is defined as the current balance on the budgetary and extra- budgetary accounts of the central and local governments. Since extrabudgetary resources consist primarily of the operating surpluses of state-owned enterprises and revenues from a range of informal levies, they provide a rough measure of total public sector saving.7 Because I do not have data on the current balances of 5. 1 am grateful to these authors for sharing their data. 6. This residual is statistically different from zero at the 5 percent level and is the third largest in the sample (only the residuals of Togo and Bahrain are larger). 7. A more detailed decomposition of general government and public enterprise saving for 1987-94 can be found in World Bank (1996: annex 3). Fiscal decline during the reform period is documented at length Kraay 549 Figure 1. Accounting for China's Saving Rate Residual Domestic credit Inflation rate Terms of trade Young-age dependency ratio Old-age dependency ratio M2/GNP Real per capita income growth ln(real per capita income) Urbanization ratio Real interest rate -0.1 -0.05 0 0.05 0.1 Contribution to predicted saving rate differential Note: The figure plots China's deviation from the mean of the entire sample of countries for each of the indicated variables, multiplied by the coefficient estimated from a cross-country regression of gross na- tional saving on these variables. Source: Author's calculations using data from Loayza, Schmidt-Hebbel, and Serven (1998). the nonstate sector, I estimate corporate saving as the total investment in fixed assets of the nonstate sector, excluding investment by individuals and by foreign- owned enterprises.8 This requires the assumption that investment of the nonstate sector is financed primarily out of retained earnings. Although there is some evidence in favor of this assumption, anecdotal evidence also suggests that a significant portion of investment, particularly that made by collectively owned enterprises, is financed by direct and unrecorded contributions from workers. To the extent that these contributions are important, this proxy double counts a portion of saving that is attributed more appropriately to households. The mea- in World Bank (1995). Barandiaran (1997) provides a more detailed flow-of-funds analysis of the compo- sition of saving. 8. I include individual investment in household saving. I measure foreign investment as investment in fixed assets by foreign-funded and overseas Chinese-funded enterprises. These data are available in the China Statistical Yearbook only beginning in 1993. 550 THF WORLD BANK ECONOMIC RFVIlW, VOL. 14, NO. 3 Figure 2. 7he Composition of National Saving in China, 19 78-95 Percentagc of GNP 415 40 35 30 20 15 10 Household r- r- xc x 0 c x cc o c cc xc c,. C Cv ,' ' Source: World Bank (1997a, 1997b) and China, State Statistical Bureau (1996). Data are given in table A-1. surement of household saving also presents a number of difficulties (discussed more fully below). For the purposes of the decomposition in figure 2, I employ my preferred measure, which is the difference between income and expenditures from China's household surveys plus the investment of individuals in fixed assets. Figure 2 has two striking features. First, the decline in public saving is largely due to the sharp decline in the surpluses of the state-owned enterprise sector that has accompanied China's transition to a more market-oriented economy. Before reform, distortions in the pricing system and discrimination against nonstate in- dustry concentrated surpluses in the state sector, from where the state could readily allocate them to its own investment priorities. Since the initiation of reform, however, price reforms and competition from a vigorous nonstate sector have combined to shrink dramatically the surpluses of state enterprises and, with them, the share of the public sector in total saving. The second feature of figure 2 is the substantial and widening residual that remains after direct measures of public and private saving have been subtracted from total saving. Although large residuals in saving decompositions such as these are typical in developing (and often in industrial) countries, the size of China's discrepancy is disconcertingly large, calling into question the magnitude Kraay 551 Figure 3. The Composition of Household Saving in China, 1978-95 Pr rcCnta5e of GNP Survey-based measure Percentage of GNP Asset-based measure Pecntg o N 20/ 18 18- -/ 14 < 14 12 L \~~>~\- 12 /\/ Insdstment-11 80 / Income - expenditures - 6 ~ 6 2~~~~~~~~~~~~~~~~~~~ F~i . . &d (vki~~~~~~~~~ it 0 4 0' /O 00>. Source: World Bank (1997a, 1997b) and China, State Statistical Bureau (1997). Data are given in table A-2. (although probably not the direction) of the increase in saving since the begin- ning of reform in 1978. One important factor contributing to this residual is the unusually large posi- tive contribution of inventory accumulation to gross national saving, which is reflected in the total but is not included in the estimate of private saving reported above. In most countries changes in stocks are small and generally average to zero over time. But in China they have been positive and large, averaging 6.5 percent of GNP between 1978 and 1995. It is unclear why changes in stocks have been so large in China. Anecdotal evidence suggests that it reflects in part the accumulation of output produced by state-owned enterprises that cannot be sold. Although this constitutes saving inasmuch as it represents forgone current con- sumption, it is misleading to think of this output as saving in the usual sense because it cannot be translated into future consumption. Household Saving Two measures of household saving have been used in academic research and policy discourse in China (figure 3). The first is based on the difference between household income and expenditures, as reported by China's household survey (described in more detail in the appendix). Since this survey does not distinguish between households' current and capital expenditures (counting both as con- sumption), I arrive at household saving by augmenting the difference between income and expenditures with a national accounts measure of investment made by individuals.9 I also construct another measure of household saving, derived from changes in aggregate stocks of assets held by households (right panel of figure 3). These changes consist of the change in household saving deposits, net subscriptions to 9. Investment made by individuals is not reported separately prior to 1983. For the purposes of figure 3, I assume that the share of individual investment in total saving is constant at its 1983 value for earlier years. 552 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 government bonds, the change in currency in circulation held by households, and investment by individuals.10 The measure based on household surveys indicates that household saving rose sharply from 7 to 15 percent of GNP during the first six years of the postreform period, but then declined and stabilized at around 10 percent of GNP (roughly one-fourth of total saving).1" In contrast, the asset-based measure of saving indi- cates that household saving has increased steadily from 5 to 20 percent of GNP, or from roughly one-seventh to one-half of gross national saving. The main source of the discrepancy between these two measures is the rapid growth of household deposits in the banking system, which accounts for the bulk of the increase in the second measure of saving. Ordinarily, one would expect that household income less expenditures would exceed the change in deposits, as households distribute their saving among deposits and other assets. Yet since 1986 the converse has been true, with the change in household saving deposits exceeding household saving by a large and rising margin. Several factors may have contributed to the divergence between the change in household saving deposits and household income less expenditures. First, the rapid development of China's financial sector since the initiation of economic reform in 1978 has improved households' access to banking institutions, espe- cially in rural areas.12 This is likely to have contributed to the growth of deposits simply by encouraging a shift in the composition of household saving from physical commodities, such as grain, to deposits. Second, in the late 1980s and early 1990s inflation-indexed saving deposits offering very attractive real returns were made available to households, and there is some evidence that significant volumes of corporate saving found their way into these instruments illicitly. These two rea- sons suggest that using the deposit-based measure of household saving might overstate actual household saving. However, to the extent that the household survey underestimates income and to the extent that the propensity to save out of unrecorded income is positive, the survey-based measure will underestimate the level of saving. If this omitted sav- ing is held in the form of deposits, it can help to account for the discrepancy between the two measures. This explanation is plausible, since growth in house- 10. It is difficult to determine household cash holdings. Following Qian (1988), I assume that 85 percent of currency in circulation is held by households. Mehran and others (1996: 38) report a lower share of 77 percent. In both measures I also am assuming implicitly that, in the aggregate, individual investment (a substantial portion of which is investment in housing in rural areas) is financed solely by household saving. 11. Although variants of these two measures have been used in many studies of household saving in China (see section I), the discrepancy between them has mostly escaped attention (see Xie 1995 for an oblique reference to this issue). 12. For example, the number of branches of the Agricultural Bank of China, the smallest of China's four large state commercial banks, more than doubled between 1981 and 1995, from 29,000 to 67,000. See also World Bank (1995: annex 3.2) for a description of the proliferation of urban credit cooperatives in Shanghai over the past decade; see Kumar and others (1997) for a description of the burgeoning net- work of nonbank financial institutions. Honohan (1995) provides an overview of international evidence on the role of institutional factors, such as access to banks, in mobilizing household saving. Kraay 553 hold income per capita recorded in the household survey lagged behind growth in GNP per capita by about 2.5 percent a year in nominal terms between 1985 and 1995. However, even if the growth rates taken from the national accounts are correct, and even if saving out of unrecorded income occurs at the same rate as saving out of recorded income, only 2 percentage points of GNP would be added to the household survey-based estimate of saving in 1995-still too little to ac- count fully for the difference between the two measures.13 How much does this discrepancy matter? On the one hand, it is not surprising that household survey measures of saving do not correspond closely to aggregate measures. Even in industrial economies with strong statistical systems, there are often large gaps between survey-based measures of household deposits and those reported by the banking system.'4 However, the magnitude of the discrepancy between these two measures and their diverging trends make this issue of some concern for the interpretation of saving behavior in China. At a basic level the gap between these two measures calls into question the overall importance of household saving in understanding aggregate saving in China. The divergent trends also have implications for different explanations of household saving behavior. For example, one might find support for the hypothesis of involuntary saving due to scarcity or rationing of consumer goods using the household survey-based measure of saving (which first rises and then declines), but not using the deposit- based measure of saving (which rises continuously), given that shortages of con- sumer goods have declined considerably since the mid-1980s. For the remainder of this article I employ the measure of saving based on the household survey, since it probably gives a more accurate picture of household saving rates. China's household survey has been scrutinized closely by a number of outside users, who, while noting several difficulties, generally conclude that its quality is fairly high.'5 Two drawbacks of this choice are that the individual investment data required to construct these series are available for a shorter time period than the income and expenditure data from the household survey (1983- 95 compared with 1978-95) and that investment data disaggregated by province are available only after 1985. In order not to lose the data from before 1985, I take as my basic measure of the household saving rate income less expenditures from the household survey, for which I have a reasonably complete panel for 30 provinces over the period 1978-95. 13. Cumulating 2.5 percent annual growth over 10 years would raise the income-less-expenditures component of the household survey estimate by 28 percent, from around 6 percentage points of GNP to 8 percentage points of GNP in 1995. 14. See, for example, the discussion in Brandolini and Cannari (1994). They note that although esti- mates of income and expenditures based on the household survey generally correspond fairly well with national income accounts, there are much larger discrepancies in estimates of wealth. They note that Canada's Survey of Consumer Finances understates bank deposits by as much as 60 percent, whereas the corresponding U.S. survey understates deposits by 44 percent. 15. See, for example, Chen and Ravallion (1996) and World Bank (1997b), as well as the more de- tailed description of the survey in the appendix. 554 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 A shortcoming of this measure is, of course, that it does not include household capital expenditures, which, particularly in rural areas, are an important compo- nent of household saving. For the shorter period for which data are available, I construct two other measures of saving to remedy this deficiency. The first sim- ply adds individual investment to the basic measure of saving. As noted earlier, however, it is difficult to distinguish between household investment and total private investment, given the close ties between many collectively owned enter- prises and their employees. I therefore also construct a broad measure of house- hold saving that includes investment by collectively owned enterprises, recogniz- ing that this measure is close to the measure of private, rather than household, saving. I then use these two measures to check the robustness of the results ob- tained using the basic measure for the periods in which the alternative measures are available.'6 Given that there is likely to be considerable measurement error in all three measures of saving, I focus on long time-series averages of saving rates, wherever possible, in order to mitigate the effects of the time-varying component of this measurement error. Household Saving Rates, Income, and Growth: Stylized Facts I document the stylized facts on average household saving rates, household income, and growth in a panel spanning China's 30 provinces between 1978 and 1995 in order to establish a set of facts with which explanations of household saving behavior should be consistent. The evolution over time of national aver- ages of household income per capita, growth of household income per capita, and household saving rates shows three interesting regularities (figure 4). First, there is a very close high-frequency time-series correlation between growth of income per capita and saving rates in both rural and urban households, averag- ing 0.49 and 0.52, respectively. Second, the relationship between saving and lev- els of income per capita differs markedly in the urban and rural data. In urban areas saving rates and income per capita generally rose together, with a time- series correlation of 0.82. But in rural areas between 1984 and 1992, incomes remained stagnant, while saving first declined sharply and then rose again, mak- ing an S-shaped pattern over time. The time-series correlation between the two variables is essentially zero. Third, household saving rates in rural areas are substantially higher than those in urban areas. Between 1978 and 1995 rural saving averaged 16 percent of household income, whereas urban saving rates averaged 5 percent. Although conceptual differences between the urban and rural household survey measures of income undoubtedly play a role (see the appendix), this difference is likely to reflect the distinct institutional and social environments in urban and rural areas. In particular, urban households enjoy access to highly subsidized housing, edu- cation, and health care, and most urban households are covered by generous (although unfunded) pension schemes through their employers. Few rural house- 16. These results are available on request. Kraay 555 Figure 4. Household Saving Rates, Income, and Growth in China, 1978-95 Urban households I A) Nruan P--t Rural households 25 1 2,S00> 'S_ ; 2\> 11) 21) Pir rPfr, -Ini,,re 2, ,101 20 2 I I \ing r It c ~~~~~S,-ig rem I. I z- nn _ 5J '' 5 en e1 enl en n e enen x 1 55 r r . D r x 3 ,c: ,§C ,rc v4G -5 I o i; Source World Bank (1997a, 1997h) and Chima, State Statistical Bureau (1997). Data are given in tahle A-2. holds enjoy these benefits, and most rely primarily on their own saving and their children for support in old age. Given these differences in the environments in which rural and urban households make their saving and consumption decisions, I consider the rural and urban data separately wherever possible. Saving rates, income levels, and income growth vary widely across China's 30 provinces, a fact that I exploit in the remainder of the article. Rural household saving rates range from a minimum of 9 percent (Hunan) to a maximum of 28 percent (Tianjin), whereas urban household saving rates range from 2 percent (Guizhou) to 14 percent (Tibet). In contrast with the time-series evidence in fig- ure 4, cross-sectional correlations between saving rates and growth, and between saving and income per capita, are modest (table 2). The correlation between saving rates and income per capita is only 0.26 in urban and rural areas, and the correlation between saving rates and growth is 0.05 in urban areas and 0.14 in rural areas. A second interesting feature of the saving-growth nexus is that there is weak evidence of a negative correlation between saving rates and future growth of Table 2. Correlations of Household Saving Rates with Income and Growth across Provinces in China Income Growth of income per capita Correlation per capita Contemporaneous Future Lagged Cross-sectional Urban 0.26 0.05 -0.13 0.09 Rural 0.26 0.14 -0.14 0.16 Time-series Urban 0.81 0.68 0.05 -0.33 Rural 0.14 0.21 -0.20 0.08 Note: Data are from a panel of three six-year averages for 30 provinces. Cross-sectional (time-series) correlations are correlations of variables in deviations from period (province) means. Past (future) growth refers to average growth during the previous (next) six-year period. Source: Author's calculations. 556 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 income per capita, both in the cross-sectional and time-series dimensions for ru- ral households and in the cross-sectional dimension for urban households. This is also apparent from figure 4, which shows a sharp rise in rural household sav- ing during the first six years of the reform period, preceding six years of nearly stagnant incomes. Thus a first glance at the data on household saving, income per capita, and growth reveals a number of interesting regularities that can help to identify promising explanations for household saving in China and caution against oth- ers. First, saving rates and levels of income per capita exhibit a modest positive correlation in a cross section of provinces and a considerably stronger time- series correlation within provinces for urban households. This lends credence to the importance of subsistence consumption in household saving decisions, with average saving rates rising as household income progresses beyond the bare minimum required for survival. Second, despite the strong time-series cor- relation between saving and growth documented in figure 4, saving and con- temporaneous growth rates are much less strongly correlated in the cross sec- tion of provinces. This cautions against recourse to the prediction of the life-cycle hypothesis that high growth leads to high saving as the saving of the young outweighs the dissaving of the old. Finally, there is weak, but suggestive, evi- dence that saving rates are negatively correlated with future income growth. This points to the prediction of standard forward-looking behavior in which households facing high expected future income growth will save less than house- holds facing low growth, as the former consume more in anticipation of higher future income. II. EVIDENCE ON SAVING AND FUTURE INCOME EXPECTATIONS In this section I present new empirical evidence on the relevance of standard forward-looking models of consumption and saving behavior for China, using the panel of rural and urban provincial average saving rates. The methodology I employ follows closely the work of Carroll and Weil (1994) and Carroll (1994). These authors note two fundamental predictions of standard forward-looking models of consumption and saving. First, expectations of higher future income (or future income growth) will lead to higher (lower) present consumption (sav- ing), as households smooth their consumption in the face of a rising income pro- file. Second, higher future income uncertainty will lower (raise) current consump- tion (saving) to the extent that households have precautionary saving motivations. These predictions can be tested empirically by regressing saving rates on suitable proxies for expected future income and the variability of future income. Consis- tent with the theory, I find that expected future income growth is negatively associated with saving for rural households, but not urban households. How- ever, proxies for future income uncertainty are not significantly associated with higher current saving and, surprisingly, often enter negatively. Kraay 557 Empirical Framework To assess the impact of expected future income growth on current saving be- havior, I estimate the following specification: (1) sa = Po + PE,[gi,, ,] + 2 x, + -it where s, is the average household saving rate in province i at time t; Et[gi,+,] is the expected growth rate of household income per capita in province i between periods t and t + 1 based on information available at time t; xit is a vector of other potential determinants of saving rates; cit is a disturbance term; and t = 1,2,3 indexes the three six-year periods for which data are available. The coefficient P3 captures the effect of expected future income growth on current saving and is predicted to be negative under the null hypothesis of forward-looking saving behavior. As Carroll and Weil (1994) note, the assumption of rational expecta- tions suggests that a good proxy for expected future income growth is actual future income growth, since rational expectations imply that expected income growth is equal to actual income growth plus an error term, that is Ejgj,+,] = gl t+l + vi. However, the presence of measurement error in this proxy variable, and the possibility that this measurement error is correlated with saving, implies that I have to instrument for future income growth in order to identify the pa- rameter t1. Given a vector of instrumental variables z,t that satisfy E[zi,'(g1t+1 + vit)] X 0 and E[z,,'Ei,] = 0, and the assumption that the variables in x,t are exog- enous, the model is overidentified and can be estimated using two-stage least squares. Since the first-stage regression in this procedure is simply a regression of future growth on current variables, it is natural to turn to variables that are standard in growth regressions as candidate instruments. As instruments for future growth, I use current income, the share of state-owned enterprise em- ployment in total employment, and a dummy variable taking on the value 1 if the province is located on China's coast and 0 otherwise. Current income can be thought of as capturing convergence effects, which have been found to be significant in a number of studies of growth in China's provinces.7 The coastal dummy is highly significant in the first-stage regressions, picking up the faster growth in coastal provinces, which is due in part to the preferential policies granted to those regions and their favorable geographical location. Finally, the share of state-owned enterprise employment is intended to proxy for growth- inimical distortions and generally enters negatively in the first-stage regressions. Standard tests of overidentifying restrictions permit me to determine the valid- 17. See for example, Jian, Sachs, and Warner (1996) and Chen and Fleisher (1996). Both of these studies use provincial gross domestic product per capita rather than the household survey measures of income per capita that I use. 558 THE WORLD BANK ECONOMIC RFVIEW, VOL. 14, NO. 3 ity of the identifying assumption that these variables affect saving only through their effects on growth."8 This procedure has the natural interpretation that agents form their expec- tations of future income growth using a simple cross-sectional growth regres- sion. This feature is somewhat more satisfying than the implicit assumption of backward-looking expectations used to justify proxying permanent income with a moving average of past income (as Qian 1988 does for China). It is also more appealing than using the predicted values of cross-sectional regressions of in- come on household-specific characteristics, such as age and education (as Wang 1995 does for China), since this methodology implicitly assumes that households do not expect any changes in their economic environment that will affect, for example, returns to their education. Given the rapidly changing economic envi- ronment in China during the period under consideration, this assumption is not very satisfying. I augment the regressions with two additional variables that are likely deter- minants of saving. I include the share of food consumption in total consumption expenditures as a proxy for the effects of subsistence consumption. Other au- thors have noted that people near subsistence levels of consumption will have lower average saving rates than richer people, as the share of their income avail- able for smoothing consumption intertemporally is smaller.'9 Accordingly, this variable is expected to enter with a negative sign in the saving equation. I also include the ratio of population to employment as a proxy for the dependency ratio. Although this is a rather crude proxy for a range of demographic determi- nants of aggregate saving rates (since it conflates demographic variables with labor force participation decisions), to the extent that low values of this ratio reflect a high ratio of prime-age earners to the total population saving for their retirement, one might expect this variable to enter negatively into the saving equation.20 To assess the effects of future income uncertainty on current saving, I use the same methodology described above, simply augmenting the regressions with a proxy for future income uncertainty. Theories of precautionary saving suggest that variability in the component of income that cannot be forecast is the rel- evant measure of income uncertainty. I construct this component of (the log of) income as the one-year-ahead prediction errors from two specifications of in- 18. This is particularly important in the case of income, which may be correlated with saving directly, for two reasons: it appears in the denominator of the left-side variable and is measured with error, and it may have a direct influence on saving because of, for example, subsistence consumption. However, these concerns are not as serious as they might seem. First, I use long (six-year) averages of all variables, which should help to alleviate problems of measurement error. Second, I include the share of food in total con- sumption as one of the explanatory variables. In any case, in almost all specifications the overidentifying restrictions are not rejected at conventional levels of significance. 19. See, for example, Gersovitz (1983) for a theoretical exposition and Ogaki, Ostry, and Reinhart (1995) and Atkeson and Ogaki (I996) for recent empirical contributions. 20. Unfortunately, further disaggregations of this variable into its demographic and labor force partici- pation components, and by old and young age, are not available by province. Kraay 559 come: a random walk with drift (in which case the prediction errors are simply demeaned income growth itself) and an autoregressive process of order 1 [AR(1)] around a deterministic trend.21 I then compute the standard deviation of both of these measures over the three six-year periods to obtain two measures of income uncertainty. Under the same assumptions of rational expectations, I can use appropriately instrumented actual future income uncertainty as a proxy for expected future income uncertainty. However, the choice of instruments for future income un- certainty is less obvious than the choice of instruments for future income growth. For simplicity, I use the same set of instruments, while recognizing that they are probably weaker for income uncertainty than for growth. Indeed, the first-stage regressions typically perform worse than the first-stage growth regressions. How- ever, the coastal dummy typically enters negatively in the first-stage regressions (significantly in the case of rural income uncertainty), as does the share of state- owned enterprise employment, suggesting that these instruments do have some predictive power for future income uncertainty as well as future income growth. Estimation Results I carry out the regressions of saving on expected future income growth, for rural and urban households, using a panel of three six-year averages for China's 30 provinces. Since expected future income growth is an explanatory variable for current saving, the model can be estimated only for the first two subperiods, 1978-83 and 1984-89. In addition to the explanatory variables noted above, the regressions also include period intercepts, so that the regressions exploit only the cross-sectional variation in the data.22 I also report results for the full sample of provinces, as well as for a subsample of higher-income provinces. To the extent that residents of poor provinces are more likely to face borrowing constraints or are near subsistence levels of consumption, it is reasonable to expect a priori that future income growth will have a stronger effect on current saving in richer prov- inces. Splitting the sample provides a crude control for these effects. The most interesting feature of the results is that expected future income growth enters negatively and significantly for rural households, as predicted by standard forward-looking models of consumption and saving (table 3). The magnitude of the coefficients suggests that the effects are economically significant as well. A 1 percentage point decline in expected future income growth results in slightly more than a 1 percentage point increase in the saving rate, as households reduce con- sumption and increase saving in anticipation of slower future income growth. 21. In principle, one can test to determine which specification is more appropriate. However, given the well-known low power of tests that might discriminate between these two alternatives, it seems prudent to consider both measures. 22. It is possible that the results are misspecified because of the absence of controls for unobserved province-specific effects that are correlated with the explanatory variables of interest. However, simple diagnostic tests do not lead to a rejection of the null hypothesis that there is no first-order serial correlation in the residuals of almost every specification reported, casting doubt on the importance of province- specific effects. 560 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 3. Saving and Expected Future Income Growth in China Full sample High-income samplea Variable Urban Rural Urban Rural Future income growth 1.857 -1.042** 0.128 -1.090** (1.521) (0.504) (0.493) (0.506) Share of food in total consumption -0.401 -0.374* -0.094 -0.776* (0.263) (0.145) (0.150) (0.222) Dependency ratio 0.115 0.046 0.033 0.040 (0.094) (0.037) (0.084) (0.038) R2 0.004 0.137 0.203 0.199 P-value for test of overidentifying restrictions 0.141 0.105 0.015 0.477 Number of observations 55 55 30 32 *Significant at the 1 percent level. **Significant at the 5 percent level. Note: The dependent variable is the household saving rate. Instruments for future income growth consist of current income, a coastal province dummy, and the share of state enterprise employment in total employment. Results are based on a panel of three six-year averages. All regressions include period intercepts. Standard errors are in parentheses. a. Provinces with average household income greater than 500 and 1,000 constant 1990 yuan in rural and urban areas, respectively. Source: Author's calculations. These results also stand in sharp contrast to those of Carroll and Weil (1994), who use cross-national data and U.S. household-level data. They find that higher expected future income growth is associated with higher, rather than lower, sav- ing rates. For urban households in China, however, expected future income growth has a positive, but statistically insignificant, effect on saving. The results for rural households may not be that surprising, to the extent that they reflect high saving in the first six years of reform, followed by lower saving in the next six. Many observers have noted that the large gains in rural incomes in the early 1980s reflect primarily the one-time benefits of dismantling the com- mune system. It is quite plausible that rural households knew that the growth benefits of these reforms were transient and increased their saving accordingly. As for the other control variables, the share of food consumption in total consumption enters with the expected negative sign, consistent with the idea that households nearer to subsistence levels of consumption have lower saving rates. Curiously, the dependency ratio enters with a positive sign, although the esti- mated coefficients are small and not significantly different from zero. This con- trasts with the large cross-country literature, beginning with Modigliani (1970), that finds that high dependency ratios are typically associated with lower saving rates. However, we can in principle account for this result to the extent that high values of the dependency ratio reflect a large young-age population and to the extent that households have strong bequest motives.23 Finally, it is worth noting 23. Collins (1991) notes this as a possible reason why dependency rates have an ambiguous effect on saving. She also notes that the standard implication that saving rates are negatively associated with dependency rates requires the assumption that the economy is growing. On the basis of these observations, she argues that dependency rates should enter alone and interacted with growth in saving rate regressions. However, including this interacted variable in the regressions reported above does not significantly affect the results. Kraay 561 that for both urban and rural households in the full sample, and for rural house- holds in the high-income sample, the null hypothesis that the overidentifying restrictions are valid is not rejected. This result is useful because it indicates that the instruments (initial income, a coastal province dummy, and the share of state enterprise employment) affect saving only indirectly through their effect on ex- pected future income growth. In the regressions of future income uncertainty on current saving rates, neither measure of future income uncertainty enters significantly, and the signs are mixed, suggesting little evidence of a precautionary saving motivation in this sample (table 4). This contrasts with Jalan and Ravallion (1998), who find a small, but statistically significant, effect of income uncertainty on wealth in a panel of rural households over the period 1985_90.24 Expected future income growth, in con- trast, continues to enter with the expected negative sign for rural households. The results for the other control variables are similar to those for the previous regressions of saving rates on expected future income alone. The share of food consumption in total consumption enters negatively (and significantly in the ru- ral sample), while the dependency ratio again enters positively. As before, in most cases tests of overidentifying restrictions do not reject the null hypothesis that the model is specified correctly. Thus the empirical evidence suggests that expectations of future income growth affect current saving rates in a manner consistent with the predictions of stan- dard intertemporal models of consumption, at least for rural households. In ad- dition, the declining importance of subsistence consumption offers a promising explanation for China's rising saving rates, as the share of food consumption in total consumption (a proxy for the importance of subsistence effects) is a robust predictor of saving rates in a panel of provincial saving rates. However, these modest empirical successes are tempered by at least two factors. First, the much poorer performance of the model for urban households and the modest fit of the regressions warn that there is much more to be understood regarding the deter- minants of household saving across provinces in China. The second is the ab- sence of significant results on future income uncertainty. This may simply reflect the drawbacks of working with aggregate data. If shocks to households are large and idiosyncratic, they may affect saving at the household level (as found by Jalan and Ravallion 1998), but not at the aggregate level. It may also reflect the fact that the simple aggregate measures of macroeconomic volatility used here do not adequately proxy for the shocks households actually experience. III. CONCLUSIONS In light of the data problems discussed and the nature of the econometric results, which are suggestive, at best, firm conclusions are less in order than ques- 24. One reason for this difference may be that Jalan and Ravallion consider wealth, rather than saving, as the dependent variable. As Carroll and Samwick (1997) note, precautionary motivations are more likely to be manifested in wealth than in saving rates in "buffer-stock" models of saving. 562 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 4. Saving and Expected Future Income Uncertainty in China Full sample High-income sample' Independent variable Urban Rural Urban Rural Future income growth 1.926 -1.276* -0.339 -1.153** (2.071) (0.549) (0.765) (0.586) Future income uncertaintyb -0.075 -1.194 0.713 -0.210 (1.493) (0.808) (0.644) (0.560) Share of food in total consumption -0.408 -0.585* -0.076 -0.810* (0.299) (-0.210) (0.147) (0.261) Dependency ratio 0.118 0.061*** 0.017 0.040 (0.112) (0.037) (0.084) (0.040) R2 0.003 0.093 0.128 0.178 P-value for test of overidentifying restrictions 0.045 0.472 0.012 0.250 Number of observations 55 55 30 32 Future income growth 1.235 0.906**t 2.309 -0.101** (0.935) (0.485) (2.318) (0.512) Future income uncertaintyc -0.899 -0.619 -2.109 0.115 (-0.705) (-0.518) (-1.860) (-0.453) Share of food in total consumption -0.456*** -0.542* -0.433 0.722** (-0.260) (-0.192) (-0.391) (-0.238) Dependency ratio 0.055 0.077 0.241 0.067 (-0.063) (-0.050) (-0.256) (-0.054) R2 0.005 0.178 0.001 0.236 P-value for test of overidentifying restrictions 0.463 0.533 0.248 0.326 Number of observations 52 53 28 31 Significant at the 1 percent level. * * Significant at the 5 percent level. * Significant at the 10 percent level. Note: The dependent variable is the household saving rate. Instruments for future income growth and future income uncertainty consist of current income, a coastal province dummy, and the share of state enterprise employment in total employment. Results are based on a panel of three six-year averages. All regressions include period intercepts. Standard errors are in parentheses. a. Provinces with average household income greater than 500 and 1,000 constant 1990 yuan in rural and urban areas, respectively. b. Future income is calculated as the standard deviation of the one-year-ahead forecast errors assuming that income is a random walk with drift. c. Future income is calculated as the standard deviation of the one-year-ahead forecast errors assuming that income is an AR(1) around a deterministic trend. Source: Author's calculations. tions for further research. Although the empirical evidence on the determinants of household saving presented here reflects favorably on two complementary explanations (expectations of future income growth and the role of subsistence consumption), these factors capture only a small fraction of the variation in house- hold saving rates across provinces. Moreover, important and unresolved mea- surement issues complicate the interpretation of trends in aggregate and house- hold saving rates. At least five sets of issues deserve further investigation, as they are likely to contribute to a better understanding of this variation and have im- portant consequences for policy. Consider first measurement issues. The large discrepancies between house- hold survey measures of saving and growth of deposits in the banking system Kraay 563 raise wider concerns. If this discrepancy reflects inadequacies of the household survey, it has important implications for a range of policies (such as targeting poverty alleviation expenditures) that rely on these survey data. If, as is more plausible, this discrepancy reflects an exaggeration of the growth of deposits in the banking system, it has important implications for the stability of China's financial system, which is already under pressure on the asset side from poorly performing loans to the state enterprise system. A second set of issues concerns the role of credit constraints. The limited re- sults suggest that intertemporal considerations are important in saving decisions among rural households. This is despite the fact that few formal consumer credit mechanisms are available in China to enable households to shift consumption from the future to the present. In surveys of saving motivations, Chinese house- holds consistently rank saving for anticipated purchases of consumer durables and life-cycle events, such as wedding celebrations, as important factors in their saving decisions. As financial markets develop, it is likely that these credit con- straints will become less important, raising the possibility that saving will fall. Although simulations of theoretical models suggest that the presence of liquidity constraints does not lead to very large buffer stocks of wealth, this factor may be important in understanding saving in China, where households have responded to unprecedented affluence by making large adjustments to their stocks of durables over a relatively short period of time. Studying the transitional dynamics of theo- retical models with these features, as well as more careful analysis of household- level data, is required to shed light on this issue. A third set of issues relates to the ongoing process of transition to a market economy and the greater uncertainties this has created as households are exposed to the vicissitudes of the market. In this article I used crude measures of aggregate income uncertainty to investigate whether this has resulted in higher precaution- ary saving. There is little evidence in favor of precautionary saving motivations. More work using household-level data, along the lines of Jalan and Ravallion (1998), is likely to be informative in this regard. Not only would this permit construction of individual measures of income uncertainty, but it would also allow investigation of more qualitative contributions to household income inse- curity, such as employment status in the private or state sector. This may in turn shed light on the consequences of further enterprise reforms for household and aggregate saving rates. The fourth set of issues concerns differences in rural and urban saving rates and the large differences in saving rates across provinces with different income levels. It seems intuitively plausible that the higher saving rates observed in rural areas can be attributed in part to the lower coverage of pension and other social benefits in those areas. Similarly, the data suggest that income levels in excess of subsistence requirements are important in understanding interprovincial saving differentials. However, carefully quantifying the magnitude of these effects is important for policy and can probably be done better using household-level data rather than aggregate data. For example, exploiting regional variations in pen- 564 THE WORLD BANK ECONOMIC REVIEW, VOL 14, NO. 3 sion reform policies could shed light on the consequences of further pension re- form for aggregate saving. Finally, the analysis in this article does not shed any light on the higher- frequency determinants of saving, since it has relied on long averages of saving rates to mitigate the effects of measurement error. Progress on this front will first require improvements in data quality. The results of such research would have implications for the effectiveness of Keynesian demand management policies that in recent years have ranked highly among the concerns of Chinese policymakers. APPENDIX: HOUSEHOLD SURVEY-BASED MEASURES OF SAVING IN CHINA The rural and urban household survey teams of China's State Statistical Bu- reau conduct annual household surveys. The surveys collect detailed data on income and expenditures as well as stocks of consumer durables and agricultural producer goods.25 I use data published in China, State Statistical Bureau (1997) on provincial averages of household income and expenditures for China's 30 provinces between 1978 and 1995 to construct flow measures of household saving.26 In 1995 the survey covered 35,000 urban households and 67,000 rural house- holds, representing 0.04 and 0.03 percent of the urban and rural populations, respectively.27 The sampling frame for both surveys is based on the administra- tive classification of household registrations (hukou). All rural households are classified as "agricultural" regardless of their primary source of income, and the population of agricultural households in rural areas constitutes the sampling frame of the rural survey. Thus the rural survey can be considered to be generally repre- sentative of the rural population. The frame of the urban survey is the urban population with "nonagricultural" household registrations.28 Since roughly 20 percent of the urban population is classified as agricultural, this constitutes a sizable gap in the coverage of the urban household survey. A second gap is due to the fact that the urban survey largely excludes migrant workers, who seldom obtain hukou registrations. Since the long-term migrant population is estimated at about 50 million, or about one-sixth of the urban population, this too consti- tutes a substantial omission. However, aggregate measures of saving rates will be biased only to the extent that saving propensities differ between omitted and included groups of the population. Additional concerns relate to the measurement of income and expenditures. The principal issue is that the concepts of income employed in the rural and 25. For a more detailed description of the rural household survey, see Chen and Ravallion (1996). World Bank (1997b: box 1.1) provides a summary of the main deficiencies of the rural and urban surveys. 26. Both surveys also collect data on household deposits and cash balances. However, summary statis- tics of these measures are not reported systematically in Chinese publications, so it is not possible to consider measures of stocks of financial assets based on the household survey. 27. China, State Statistical Bureau (1995: tables 9-5 and 9-15). Figures are for 1995. 28. However, within this broad definition there have been changes in the sample frame. For example, pensioners were included in the urban survey only after 1985. Kraay 565 urban surveys are different. In rural areas the household income measure I use, net income, is calculated net of production costs associated with household pro- duction. In urban areas the household survey is geared primarily toward record- ing labor income and is likely to exclude other forms of income (such as interest or self-employment income). Again, measures of household saving rates will be biased only to the extent that saving propensities out of the omitted component of income differ from those out of included income. A final set of concerns regarding the household survey relates to the valuation of in-kind income. In rural areas the prices at which nonmarketed household production is valued may not adequately reflect market prices. To the extent that this undervalues this source of income and form of consumption, household sav- ing rates will be biased downward. In urban areas a range of in-kind income (such as medical, education, and housing benefits) is not included in income. This too will bias urban saving rates downward. The specific measures of household income employed are as follows. In urban households, I use income available for living, which consists of the total income from all sources covered by the survey, less transfers to family members not re- siding in the household, gifts, payments received from boarders in the household, and subsidies received for responding to the survey. The expenditure measure is living expenditures, which nets out from total expenditures transfers to family members not residing in the household, gifts, and payments received from board- ers in the household. In rural areas I use net income as discussed above, less living expenditures. Table A-1. Composition of National Saving, 1978-95 Ildicator 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 Percentage of GNc at current prices Gross national saving 37.75 36.O)8 34.91 32.67 35.07 34.86 34.71 33.92 35.28 36.26 35.S8 35.03 37.72 37.86 37.41 41.36 41.28 41.25 Changes in stocks 8.39 8.00 6.02 6.75 5.03 4.97 4.76 8.29 7.33 4.85 5.84 10.38 9.21 7.28 4.95 5.84 3.73 5.29 Gross national saving net of changes in stocks 29.36 28.09 28.89 25.92 30.04 29.89 29.95 25.63 27.95 31.41 30.04 24.65 28.52 30.5S 32.46 35.52 37.55 35.95 Households (survey measure)' 7.70 9.57 10.29 9.89 13.06 13.70 16.54 13.77 12.62 11.94 11.13 10.2S 11.72 10.34 10.36 10.05 10.23 10.56 lHouseholds (assets measure), 4.83 6.49 8.10 8.25 8.34 9.67 13.55 12.77 14.70 15.67 16.13 16.10 16.94 17.55 17.65 19.23 22.30 21.10 Corporate sector 2.19 1.87 4.63 2.84 4.65 3.65 3.31 3.64 3.84 4.58 4.77 3.37 2.85 3.22 5.10 7.35 7.24 7.56 Public secror 20.11 15.99 13.86 13.16 13.47 13.37 13.56 14.78 12.81 12.00 9.16 8.29 7.57 735 6.89 2.68 2.06 0.53 Budgetarygovernment' 14.92 9.92 7.19 6.47 5.29 5.54 6.30 6.70 5.60 4.23 2.86 2.26 2.59 1.79 1.08 1.49 0.43 0.53 l xtrabudgetary government 5.18 6.07 6.67 6.69 8.18 7.83 7.26 8.08 7.21 7.77 6.30 6.03 4.99 5.56 5.81 1.18 1.63 - Residual (using household survey neasurc) 7.76 8.66 6.13 6.77 3.89 4.13 1.30 1.73 6.01 7.75 10.82 13.10 15.58 16.95 15.06 21.29 21.76 22.60 Residual (using household assets measure) 10.62 11.74 8.31 8.41 8.62 8.18 4.29 2.73 3.93 4.02 5.82 7.28 10.37 9.74 7.76 12.11 9.69 12.06 l'ercentage of gross national saving Changes in stocks 22.22 22.17 17.24 20.66 14.35 14.25 13.71 24.44 20.78 13.38 16.27 29.63 24.40 19.23 13.23 14.12 9.04 12.84 Public sector 53.16 44.31 39.71 411.29 38.41 38.34 39.07 43.57 36.31 33.08 25.53 23.65 20.08 19.42 18.42 6.47 4.98 1.27 Corporate sector 5.81 5.17 13.27 8.69 13.25 10.46 9.54 10.74 10.89 12.62 13.29 9.62 7.55 8.51 13.63 17.77 17.53 18.33 Households (survey measure)' 20.39 26.53 29.46 30.29 37.25 39.30 47.65 40.60 35.77 32.93 31.02 29,33 31.07 27.31 27.69 24.29 24.78 25.61 Households (assets measure)a 12.80 17.99 23.21 25.27 23.78 27.73 39.03 37.64 41.68 43.21 44.97 45.96 44.90 46.35 47.19 46.48 54.03 51.16 Residual (using household survey measure) 20.55 23.99 17.S6 20.73 11.10 11.90 3.73 5,09 17.04 21.37 30.16 37.40 41.31 44.77 40.26 51.48 52.72 54.79 Residual (using household assets measure) 28.14 32.53 23.81 25.75 24.57 23.47 12.36 8.05 11.13 11.09 16.21 20.78 27.48 25.72 20.75 29,28 23.46 29.24 GNP (billions (if current yuan) 362.41 403.80 451.78 486.00 530.613 595.70 720.48 898.91 1,020.14 1,195.45 1,492.23 1,691.78 1,859.84 2,166.25 2,665.19 3,456.05 4,653.29 5,727.73 Current account (percentage of GNP) -0.27 -0.42 -0.28 0.14 1.90 1.21 0.45 -3.75 -2.42 0.11 -0.95 -0.99 3.08 3.16 1.25 -2.03 1.33 0.07 - Not available. a. See table A-2. b. Current budgetary surpltis of central and local governiments (World Bank 1997a: statistical annex tables 19 and 22). c. Current surpluls on extrabudgetary accounts of central an d local governments (China, State Statistical Bureau 1996: table 7-16, revenue less non-fixed-asset investment). Source: Table A-2 and World Bank (World Development Indicators). Table A-2. Alternative Measures of Household Saving in China, 1978-94 Measure 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 Household surrey measure National per capita income, (yuan) 162.43 196.27 233.56 265.26 310.58 349.23 403.94 455.98 503.53 552.10 661.80 738,20 832.76 888.86 1,008.98 1,236.13 1,672.54 Expenditures, 146.93 172.23 204.80 238.23 265.46 295.32 328.92 389.61 444.20 494.30 604.28 675.55 729,68 797.60 877.66 1,067.76 1,439.86 Saving = income less expenditurcs 15.50 24.04 28.76 27.03 45.13 53.92 75.01 66.37 59.33 57.80 57.52 62.65 10.3.08 91.26 131.32 168.37 232.68 Population (millions) 962.59 975.42 987.05 1,000.72 1,016.54 1,030.08 1,043.57 1,058.51 1,075.07 1,093.00 1,110.26 1,127.04 1,143.33 1,158.23 1,171.71 1,185.17 1,198.50 Total income less cxpcnditures (billionis of currcntyuaii) 14.92 23.45 28.39 27.05 45.87 55.54 78.28 70.25 63.78 63.18 63.86 70.61 117.85 105.70 153.87 199.54 278.87 Individual investment (billionsofcurrentyuan)b 12.97 15.21 18.08 21.04 23.44 26.08 40.90 53.52 64.94 79.59 102.21 103.23 100.12 118.29 122.20 147.62 197.06 Household saving (billions of current yuan)' 27.89 38.66 46.47 48.08 69.31 81.61 119.18 123.78 128.72 142.76 166.07 173.83 217.97 223.99 276.07 347.16 475.93 Household assets measure (billions of current yuan) Currcncy in circulanton - 26.77 34.62 39.63 43.91 52.98 79.20 98.80 121.80 145.40 213.30 234.20 264.10 317.40 432.90 585.00 729.00 85 percent of change in currency in circulation 1.63 3.96 6.67 4.26 3.64 7.71 22.29 16.66 19.55 20.06 57.72 17.77 2S.42 45.31 98.18 129.29 122.40 Household deposits" 21.06 28.10 39.95 52.37 67.54 89.25 121.47 162.26 223.76 307.33 380.15 514.69 703.42 911.03 1,154.54 1,520.35 2,151,88 Change in deposits 2.90 7.04 11.85 12.42 15.17 21.71 32.22 40.79 61.50 83.57 72.82 134.54 188.73 207.61 243.51 365.81 631.53 Net subscriptions to government bonds' 0.0) 0.00 0.00 2.40 2.00 2.10 2.20 3.80 4.00 4.11 7.99 16.85 0.75 9.00 6.64 21.76 86.80 Subscriptions 0.00 0.00 0.00 2.40 2.00 2.10 2.20 3.80 4.00 4.50 8.80 18.10 9.30 19.94 29.60 31.48 106.74 Rtedemptions 0.00 0.00 0.00 0,00 0.00 0.00 0.00 0.00 0.00 0.39 0.81 1.25 8.55 10.94 22.96 9.72 19.94 Inidividual investment' 12.97 15.21 18.0S 21.04 23.44 26.08 40.90 53.52 64.94 79.59 102.21 103.23 100.12 118.29 122.20 147.62 197.06 Household savinig (assets measurc)( 17.51 26.21 36.61 40.12 44.25 57.59 97.61 114.77 149.99 187.33 240.74 272.38 315.01 380.20 470.52 664.48 1,037.79 Retail price index 48.15 49.11 52.05 53.30 54.31 55.13 56.67 61.68 65.38 70.15 83.15 97.93 100.00 102.89 108.43 122.73 149.35 - Not available. Note: Shaded areas indicate estimates. a. Population-weighted average of urban and rural values. Urban measures are income available for living and living expenditures (China, State Statistical Bureau, Chinia Statistical Yearbook, various issues: table 9-5). Rural measures are net income anid living expenditures. Population weights are nioniagricultural and agricultural shares of the population. b. China, State Statistical Bureau, C6ina Statistical Yearbook (various issues: table 5-1). c. Income less expenditures plus individual investment. d. World Bank (1997a: statistical annex table 14). e. China, State Statistical Bureau, Chinta Statistical Yearbook (various issues: table 9-3). f. World Bank (1997a: statistical annex table 12). g. Change in currency in circulation plus change in deposits plus tiet stubscriptions to government bonds plus itldividual investment. Souirce: China, State Statistical Bureau, Chinia Statistical Yearbook (various issues) and World Bank (1997a). 568 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Arora, Vivek. 1995. "National Saving in China: Trends, Measurement, and Determi- nants." International Monetary Fund, East Asia Department, Washington, D.C. Processed. Atkeson, Andrew, and Masao Ogaki. 1996. "Wealth-Varying Intertemporal Elasticities of Substitution: Evidence from Panel and Aggregate Data." Journal of Monetary Eco- nomics 39(3):507-34. Bai, Baier, Yan Zhu, and Ling Wang. 1993. "A Quantitative Analysis of the Consump- tion Structure of China's Urban Households." In Tien-Tung Hsueh, Yun-Wing Sun, and Jingyuan Yu, eds., Studies on Economic Reforms and Development in the People's Republic of China. Hong Kong: Chinese University Press. Barandiaran, Edgardo. 1997. "China: Flow of Funds Analysis." World Bank, East Asia Department, Washington, D.C. Processed. Brandolini, Andrea, and Luigi Cannari. 1994. "Methodological Appendix: The Bank of Italy's Survey of Household Income and Wealth." In Albert Ando, Luigi Guiso, and Ignazio Visco, eds., Saving and the Accumulation of Wealth: Essays on Italian Household and Government Saving Behaviour. Cambridge, U.K.: Cambridge Uni- versity Press. Carroll, Christopher D. 1994. "How Does Future Income Affect Current Consumption?" Quarterly Journal of Economics 109(1): 111-48. Carroll, Christopher, and Andrew Samwick. 1997. "The Nature of Precautionary Wealth." NBER Working Paper 5193. National Bureau of Economic Research, Cambridge, Mass. Processed. Carroll, Christopher D., and David N. Weil. 1994. "Saving and Growth: A Reinterpreta- tion." Carnegie-Rochester Conference Series on Public Policy 40:133-92. Chen, Jian, and Belton M. Fleisher. 1996. "Regional Income Inequality and Economic Growth in China." Journal of Comparative Economics 22(1):141-64. Chen, Shaohua, and Martin Ravallion. 1996. "Data in Transition: Assessing Rural Liv- ing Standards in Southern China." China Economic Review 7(1):23-56. China, State Statistical Bureau. 1992. China Statistical Yearbook. Beijing: China Statisti- cal Publishing House. 1995. China Statistical Yearbook. Beijing: China Statistical Publishing House. 1996. Cbina Statistical Yearbook. Beijing: China Statistical Publishing House. 1997. China Regional Economy: A Profile of 17 Years of Reform and Opening Up. Beijing: China Statistical Publishing House. Chow, Gregory. 1985. "A Model of Chinese National Income Determination." Journal of Political Economy 93(4):782-92. Collins, Susan. 1991. "Saving Behaviour in Ten Developing Countries." In B. Douglas Bernheim and John B. Shoven, eds., National Saving and Economic Performance. Chicago: University of Chicago Press. Denizer, Cevdet, and Holger Wolf. 1998. "Aggregate Savings in the Transition: A Cross- Country Study." World Bank, Europe and Central Asia Department, Washington, D.C. Processed. Kraav 569 Dessi, Roberta. 1991. "Household Saving and Wealth in China: Some Evidence from Survey Data." Working Paper 9112. University of Cambridge, Department of Applied Economics, Cambridge, U.K. Processed. Feltenstein, Andrew, David Lebow, and Sweder van Wijnbergen. 1990. "Saving, Com- modity Market Rationing, and the Real Rate of Interest in China." Journal of Money, Credit, and Banking 22(2):235-52. Fleisher, Belton M., Yunhua Liu, and Hongyi Li. 1994. "Financial Intermediation, Infla- tion, and Capital Formation in Rural China." China Economic Review 4(1):101-15. Gersovitz, Mark. 1983. "Saving and Nutrition at Low Incomes." Journal of Political Economy 91(5):841-55. Honohan, Patrick. 1995. "The Impact of Financial and Fiscal Policies on Saving." Eco- nomic and Social Research Institute, Dublin. Processed. Hussain, Athar, Steve Ludlow, Limin Wang, and Qian Wei. 1990. "The Chinese Acad- emy of Social Sciences Data Set." London School of Economics, London. Processed. Jalan, Jyotsna, and Martin Ravallion. 1998. "Precautionary Wealth in Rural China." World Bank, Development Research Group, Washington, D.C. Processed. Jefferson, Gary. 1990. "The Impact of Economic Structure on Fertility, Saving, and Re- tirement Behaviour of Chinese Households." Journal of Asian Economics 1(2):205- 23. Jian, Tianlun, Jeffrey Sachs, and Andrew Warner. 1996. "Trends in Regional Inequality in China." NBER Working Paper 5412. National Bureau of Economic Research, Cam- bridge, Mass. Processed. Kumar, Anjali, Nicholas Lardy, William Albrecht, Terry Chuppe, Susan Selwyn, Panda Perttunen, and Tao Zhang. 1997. China's Non-Bank Financial Institutions. Washing- ton, D.C.: World Bank. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serven. 1998. "What Drives Saving across the World?" Policy Research Working Paper 2309. World Bank, Development Research Group, Washington, D.C. Processed. Ma, Guonan. 1993. "Macroeconomic Disequilibrium, Structural Changes, and House- hold Saving and Money Demand in China." Journal of Development Economics 41(1):115-36. Mehran, Hassanali, Marc Quintyn, Tom Nordman, and Bernhard Laurens. 1996. Mon- etary and Exchange System Reforms in China: An Experiment in Gradualism. IMF Occasional Paper 141. Washington, D.C.: International Monetary Fund. Modigliani, Franco. 1970. "The Life Cycle Hypothesis of Saving and Intercountry Differ- ences in the Saving Ratio." In W. A. Eltis, M. F. Scott, and J. N. Wolfe, eds., Induc- tion, Growth, and Trade. Oxford: Clarendon Press. Modigliani, Franco, and Larry Shi Cao. 1996. "The Chinese Saving Puzzle and the Life Cycle Hypothesis." Massachusetts Institute of Technology, Department of Econom- ics, and Harvard University, Department of Economics, Cambridge, Mass. Processed. Naughton, Barry. 1987. "Macroeconomic Policy and Response in the Chinese Economy: The Impact of the Reform Process." Journal of Comparative Economics 11(3):334- 53. Ogaki, Masao, Jonathan Ostry, and Carmen Reinhart. 1995. "Saving Behaviour in Low- and Middle-Income Developing Countries: A Comparison." IMF Working Paper 95/3. International Monetary Fund, Washington, D.C. Processed. 570 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Pudney, Stephen. 1991. "Income, Wealth, and the Life Cycle: A Non-Parametric Analysis for China." Working Paper 9106. University of Cambridge, Department of Applied Economics, Cambridge, U.K. Processed. Qian, Yingyi. 1988. "Urban and Rural Household Saving in China." IMF Staff Papers 35(4):592-627. Wang, Yan. 1995. "Permanent Income and Wealth Accumulation: A Cross-Sectional Study of Chinese Urban and Rural Households." Economic Development and Cul- tural Change 43(3):523-50. Wang, Zhi, and Wen S. Chern. 1992. "Effects of Rationing on the Consumption Behaviour of Chinese Urban Households during 1981-87." Journal of Comparative Economics 16(1):1-26. Wang, Zhi, and Jean Kinsey. 1994. "Consumption and Saving Behaviour under Strict and Partial Rationing." China Economic Review 4(1):83-100. Wong, Richard Yue-Chim. 1993. "Estimating and Interpreting Chinese Consumption Functions." In Tien-Tung Hsueh, Yun-Wing Sun, and Jingyuan Yu, eds., Studies on Economic Reforms and Development in the People's Republic of China. Hong Kong: Chinese University Press. World Bank. 1988. China: Finance and Investment. Washington, D.C. . 1995. China: Macroeconomic Stability in a Decentralized Economy. Washing- ton, D.C. 1996. China: Fighting Inflation, Deepening Reforms. Washington, D.C. 1997a. China 2020: Developmentt Challenges in the New Century. Washington, D.C. 1997b. China 2020: Sharing Rising Incomes. Washington, D.C. Various issues. World Development Indicators. Washington, D.C. Xie, Ping. 1995. "Individual Saving Behaviour in China's Structural Transitional Period." Paper presented at the International Seminar on Income Distribution Systems, Beijing, June 27-29. Processed. Yusuf, Shahid. 1994. "China's Macroeconomic Performance and Management during Transition." Journal of Economic Perspectives 8(2):71-92. Zhang, Meiling. 1994. "Household Saving Since the Late 1970s in China." International Monetary Fund, Washington, D.C. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 571-94 Private Saving in India Norman Loayza and Rashmi Shankar This article studies the evolution of the private saving rate in India during 1960-9S. Its distinctive feature is that it proposes three new measures of private saving, which are incremental improvements to the (naive) national accounts measure. The improvements consist of accounting for capital losses to private net worth due to inflation, including expenditures on durable goods as a form of saving, and expanding the definition of saving to include human capital expenditures. After examining descriptive trends and reviewing the related literature, the article tests the hypothesis that households that save in India "pierce the corporate veil. " The evidence shows that, in fact, changes in corpo- rate saving are offset by changes in household saving, indicating that the unit of analysis should be aggregate private saving. The core analytical section of the article studies how the behavior of the private saving rate is related to the real interest rate, per capita income, the dependency ratio, financial depth, the government saving rate, and the share of agriculture in gross domestic product. The empirical analysis is done by estimating error-correction models on aggregate annual data, although most of the discussion cen- ters on long-ruin effects. There are many reasons for studying private saving in India. India's saving per- formance has surpassed that of other developing countries with comparable per capita income. However, given the Indian government's ambitious growth tar- gets, and the need in the present global environment to generate investable re- sources by and large internally, the design of policies aimed at enhancing saving acquires great significance. With this in mind we assess the extent to which an increase in public saving is offset by a reduction in private saving (effectively testing for Ricardian equivalence), whether an increase in rates of return in finan- cial markets leads to a rise in private saving rates (which would have implications for the effectiveness of tax incentives on saving instruments), and the extent to which an increase in average private disposable income brings about a rise in private saving rates (which has implications for the wide range of policies that raise private income). In addition to studying the policy determinants of private saving, it is also important to take into account the impact that India's remarkable demographic Norman Loayza is with the Central Bank of Chile and the Economic Research Group at the World Bank, and Rashmi Shankar is a doctoral candidate at the University of California, Santa Cruz. Their e- mail addresses are nloayza@condor.bcentral.c1 and rshankar@worldbank.org. The authors are grateful to Luis Serven, an anonymous referee, and Fran,ois Bourguignon for valuable comments. C) 2000 The International Bank for Reconstruction and Development/ THE WORLD BANK 571 572 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 transition in the past 30 years has had on current and future saving rates. The age structure of the population in India has changed dramatically. The number of people more than 65 years of age as a proportion of total adults (that is, people between 16 and 65 years of age) rose from 6 to 8 percent. The corresponding figure for children less than the age of 16 has declined from 72 percent in 1965 to 58 percent in 1994. We are interested in determining the impact of this change on saving because in the next few years "demographic inertia" is likely to result in a continuing rise in the share of working adults in the population. The major difference between this study and other macroeconomic studies of saving in India is that here we adjust private saving and income figures to derive measures that are conceptually more correct. The goal of these adjustments is to approach saving figures that correspond to changes in net worth. The first adjustment concerns asset price changes. Although in theory we should correct for capital gains and losses for all types of assets, data availability con- strains us to focusing only on inflation-related capital changes. Since we do not have long enough time series for stock, private bond, and real estate prices, the first adjustment considers only the effect that inflation has on private and public income and saving. Although limited, this correction is important. When infla- tion occurs, the value of government debt (including money balances) held by the private sector decreases, which implies a loss of wealth to the private sector and a concomitant gain to the government. We obtain the inflation-adjusted saving and income figures from the World Saving Database (see Loayza and others 1998 for a complete explanation of the adjustment procedure, input sources, and the corresponding quality controls). In fact, the provision of these adjusted saving figures, as well as the consistent definition of the public and private sector for a large number of countries since the 1960s, is one of the main contributions of this new database. Clearly, correcting for capital gains and losses due to inflation becomes more important as the level of government debt held by the private sector increases or inflation rises. In India the inflation-adjusted figures are significantly different from the unadjusted figures, particularly those in the 1980s. We find significant differences between our results and those emphasized in the received historiogra- phy, particularly regarding Ricardian offset coefficients, and believe this to be largely due to our use of inflation-adjusted figures. We also correct private and public saving figures to include expenditures on consumer durables and investment in human capital. We first present a measure of household saving, and hence private and national saving, that includes private expenditures on consumer durables, including personal transportation.' This is not only because households perceive expenditures on durables as deferred con- sumption but also because in India most consumer durables and motor vehicles have an extremely high resale value. In the absence of data on gold imports over 1960-94, we also hope to use this definition of saving to correct partly for the 1. We follow the methodology in Jorgenson and Fraumeni (1989). Loayza and Shankar 573 understatement of household physical saving that excludes jewelry, a tradition- ally important form of saving in Indian families. The third measure of saving involves both private and public saving. To ad- justed private saving, we add not only expenditures on consumer durables but also expenditures on education and health. Correspondingly, we augment public adjusted saving with the government's final consumption expenditures on health and education. These adjustments are made to proxy for the accumulation of hu- man wealth. I. MAJOR TRENDS IN INDIAN SAVING RATES The trend of the national saving rate (defined as gross national saving, includ- ing net current transfers, divided by gross national disposable income) in India compares favorably to that of low-income countries and countries of the Organisation for Economic Co-operation and Development (OECD) but falls short of that of China and other East Asian countries (table 1). In 1965 India's na- tional saving rate was similar to the average of all developing countries. It has since followed a rising trend, surpassing the average national saving rate of the OECD countries by the late 1980s and gaining about 4 percentage points on the average of developing countries by 1995. Although strong, the rising trend of India's national saving rate has fallen short of that of East Asian countries, pro- ducing a gap between the two of more than 10 percentage points by 1995. The inflation-adjusted median private saving rate appears to be consistently lower in India than in the East Asian or OECD countries, although it has been growing faster than both (table 2). For India and other countries, except the OECD countries, the definition of the public sector we use includes not only the central government but also state and local governments and public enterprises. In the OECD countries the public sector does not include public enterprises, which are ascribed to the private sector. This overstates the OECD'S private saving rate relative to that of India and other countries. We next look at the relationship among saving, investment, and growth across countries. Each country relies to a varying extent on national saving to finance Table 1. Median National Saving Rates, 1960-94 (percentage of gross national disposable income) Period India China Low-income South Asia, East Asia OECD 1960-73 15 26 11 12 20 25 1974-82 21 33 13 15 29 23 1983-91 21 36 11 17 33 21 1992-94 22 41 12 18 34 19 Note: For country groups, the median is of individual country averages, while for India and China the averages across time are reported. The saving rate is gross national saving, including net current transfers, divided by gross national disposable income. a. Excluding India. , Source: Authors' calculations based on data in the World Bank's World Saving Database. 574 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 2. Inflation-Adjusted Median Private Saving Rates, 1960-94 (percentage of inflation-adjusted private disposable income) Period India Low-income South Asiaa East Asia OECD 1960-73 12 9 n.a. 20 23 1974-82 16 10 10 27 23 1983-91 17 9 10 29 22 1992-94 18 10 n.a. 29 23 Note: See table 1 for the composition of country groupings. a. Excluding India. Source: Authors' calculations based on data in the World Bank's World Saving Database. the investment necessary for growth. Comparing performances on these fronts allows us to make some inferences about the best use of resources garnered through saving. The saving rate in China, for example, was, on average, 70 percent higher than that in India, investment was approximately 50 percent higher, and growth was 100 percent higher (table 3). East Asia achieved a growth rate twice as high as India's with an investment rate that was 34 percent higher and a saving rate that was about 45 percent higher. India's long-run averages of saving, invest- ment, and growth are similar to those of the OECD countries. Looking at the saving-investment gap, India had to rely on foreign saving more than the East Asian countries, but notably less than other low-income countries. Saving Rates and Other Macroeconomic Aggregates in India We divide the period 1960-95 into four subperiods (table 4). The first two, 1960-1973 and 1974-82, are separated by an oil crisis, as are the second and third. The third and fourth subperiods, 1983-91 and 1992-95, are distinguished by the Indian government's 1991 structural adjustment program. Inflation increased fairly smoothly across these broad subperiods, as did real GDP growth and the rate of investment (given by the ratio of gross domestic in- vestment to gross national disposable income), although the investment rate dipped slightly in the last period. Table 3. Median Saving, Investment, and Gross Domestic Product, 1960-94 Low- South East Indicator India Cbina income Asia' Asia OECD National saving rate (percent of gross national disposable income) 20 34 11 16 29 22 Gross domestic investment (percent of national disposable income) 21 32 17 17 27 21 GDP growth (percent) 4 8 3 4 8 4 a. Excluding India. Source: Authors' calculations based on data in the World Bank's World Saving Database. Loayza and Shankar 575 Table 4. Basic Macroeconomic Indicators in India, 1960-98 Gross domestic National saving rate Inflation investment (gross national saving (GDP GDP (percentage of as a percentage deflator; growth gross national of gross national Period percent) (percent) disposable income) disposable income) 1960-73 7.0 3.3 17 15 1974-82 7.7 3.9 21 21 1983-91 8.5 5.2 23 21 1992-95 9.0 4.9 23 22 1996-98 7.0 5.9 23 21 Source: Authors' calculations based on data in the World Bank's World Saving Database. In 1960-82, India achieved its peak saving rate in 1978, when the national saving rate rose to 22 percent. Private saving increased sharply in the second period, 1974-82, by about 5 percentage points. Several factors operated to in- crease the saving rate in the late 1970s, notably, the end of a decade of vigorous bank expansion and foreign remittances from Indians working in the Gulf area. These changes are also indicated by the fall in the average current account deficit (by definition, the difference between investment and saving) during this period. Real GDP growth did not exhibit any marked trend relative to the previous period. The national saving rate remained at about 21 percent in the last three peri- ods, despite the sharp increase in the GDP growth rate between the second and third periods, its slight decline in the fourth, and the post-1996 increase. The investment rate remained closely linked to the saving rate, resulting in a small current account deficit throughout 1960-95. This points to an important styl- ized fact: saving in India depends on national, as opposed to foreign, savings. The slight increase in the investment-saving gap in the last period still leaves the deficit at a low 2 percent of gross national disposable income. Changing Composition of National Saving in India We study the composition of national saving by, first, describing the behavior of the components of private saving and, second, describing the trends in public and private saving, paying special attention to the comparison between adjusted and unadjusted figures. (Recall that the definition of the public sector includes the central government, state and local governments, and public enterprises.) The entire 1960-94 period was marked by an increase in the share of house- hold saving as a proportion of unadjusted private disposable income, whereas the analogous measure for corporate saving remained mostly stagnant, picking up only in the mid-1990s (table 5). It can be argued that the sharp rise in national saving following the nationalization of banks and vigorous branch expansion beginning in 1969 was spurred by a rapid growth in financial saving in the 1970s. The 1970s were also characterized by a jump in remittances from abroad-mostly from the Middle East-which could have contributed some of the increase in household saving between the first and second periods. 576 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 5. The Composition of Private Saving in India, 1960-95 (percentage of unadjusted private disposable income) Household Household Total Corporate Period physical saving financial saving household saving saving 1960-73 7.61 3.62 11.23 2.46 1974-82 9.83 6.40 16.23 3.03 1983-91 9.82 8.81 18.63 2.64 1992-95 8.67 11.16 19.83 3.46 Note: Figures are not adjusted for capital losses due to inflation, since there is no sound basis on which to apportion the adjustment to household and corporate sectors. Source: Authors' calculations based on data in the World Bank's World Saving Database. The rate of household physical saving also increased in the first three periods; however, in the last period, 1992-95, it fell somewhat, reflecting in part a portfo- lio shift from physical to financial saving. The slight decline in household physi- cal saving generated some debate in India, especially by 1994, when the decline appeared to be more substantial. One view holds that this decline is a statistical illusion (Athukorala and Sen 1995). The Central Statistical Organization (cso) sets household physical saving exactly equal to household investment, which in turn is determined residually. In national accounts domestic investment is ad- justed to equal the sum of domestic and foreign saving. Then only investment is adjusted for errors and omissions. This asymmetric adjustment is advocated on the argument that public and corporate saving data are more reliable than invest- ment data. This might be true in the case of public saving; however, the estimate of corporate saving (and investment) is based on small, not necessarily represen- tative, samples and relies on voluntary responses from enterprises. At any rate, although there are reasons to believe that the figures for private saving, particu- larly those of household physical saving, are erroneous, it is not at all obvious that measurement error is behind the recent trends.2 Private income and saving figures must be adjusted to take into account the redistribution of wealth from the public to the private sector due to inflation's erosion of the value of public debt held by the private sector. This correction is sizable. Adjusted public saving as a ratio of gross national disposable income increased by 3 percentage points over 1960-94, while the corresponding unad- justed figure rose by about half as much. There was a steady increase in adjusted private saving over the entire period, except between 1974-82 and 1983-91, when it remained stable at about 14 percent (table 6). The unadjusted figures for private saving show a sharper in- crease, from 12 to 21 percent over 1960-94, as opposed to the adjusted figures, which show an increase from 11 to 16 percent. The unadjusted public saving rate declined from 3 to 1 percent over the full period, reaching a low of 0.94 percent 2. Some physical assets, traditionally preferred as saving tools in India, such as jewelry and gold, are not covered in the cso estimates. This further increases the likelihood of measurement error in this cat- egory of saving. Loayza and Shankar 577 Table 6. Private and Public Saving Rate in India, 1960-94 Adjusted Adjusted Unadjusted Unadjusted private public private public Period saving rate saving rate saving rate saving rate 1960-73 Unaugmented 0.11 0.04 0.12 0.03 Includes consumer durables 0.13 0.15 Includes expenditures on health and education 0.16 0.06 0.18 0.04 1974-82 Unaugmented 0.14 0.07 0.17 0.04 Includes consumer durables 0.18 0.21 Includes expenditures on health and education 0.22 0.10 0.24 0.07 1983-91 Unaugmented 0.14 0.07 0.19 0.02 Includes consumer durables 0.19 0.24 Includes expenditures on health and education 0.23 0.11 0.27 0.06 1992-94 Unaugmented 0.16 0.07 0.21 0.01 Includes consumer durables 0.20 0.25 Includes expenditures on health and education 0.23 0.11 0.28 0.05 Note: Adjusted means adjusted for capital gains or losses due to inflation. Source: Authors' calculations based on data in the World Bank's World Saving Database. in 1984. However, adjusted public saving increased from 4 to 7 percent between 1960 and 1994. What is striking about these data is that the difference between the adjusted and unadjusted figures becomes larger as time goes on. In the first period, 1960- 73, the differences are marginal. However, in the next subperiod, there is a 3 percent difference between the adjusted and unadjusted figures. This difference increased to 5 percent in later years. These increases were due to the gradual rise of inflation and, most important, the increase in public debt held by the private sector. Whereas the unadjusted public saving rate shows a declining trend since the early 1970s, the adjusted public saving rate remained basically flat. Table 6 also presents private saving rates augmented to include expenditures on consumer durables and expenditures on health and education. Saving by any definition rose sharply in the early and mid-1970s and then again in the early 1980s. The flattening observed in common measures of saving since the early 1990s was less pronounced for the augmented measures, which continued to rise, albeit by less than in the 1986-88 period. This is in part due to the continu- ous increase in spending on health and education. 578 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 II. SAVINGS ISSUES IN INDIA Much of the literature on saving in India distinguishes between private and public saving for analytical purposes and treats public saving as exogenous. Pub- lic saving has been conspicuous in the lack of attention it has received. According to Pandit (1991) public saving in India has been largely residual and mostly driven by expansionary fiscal policy and public sector pricing policy. A notable excep- tion to this view is that of Cashin, Olekalns, and Sahay (1998). They find evi- dence that the central government in India exhibits both tax-smoothing and tax- tilting behavior. Thus the government relies on seigniorage and financial repression as revenue sources, whereas taxes rise only in response to permanent changes in expenditures. According to their study, state tax revenues are relatively volatile in the face of even temporary changes in expenditure. Most of the literature, how- ever, focuses, as we do, on aggregate private saving. In this context several issues have been raised. Dependence on Public Saving The issue here is the extent to which the private sector internalizes the government's budget constraint and hence the extent to which an increase in public saving is offset by an increase in private saving. Muhleisen (1996) presents evidence that Ricardian equivalence is of minor importance in India. He finds that long-run aggregate private saving decreases by 0.25 percent in response to a 1 percent increase in public saving. Estimates for other developing countries range from 0 percent (Haque and Montiel 1989) to 50 percent (Corbo and Schmidt- Hebbel 1991). Income Growth Lahiri (1989) finds that the rate of growth of personal disposable income is a significant determinant of private saving in all the countries in his sample of Asian countries, including India. Lahiri bases his empirical findings on individual time-series analyses for each country in his sample. He claims that such an ap- proach has an advantage over a panel-based analysis in that the marginal re- sponse of the saving rate to various factors need not be assumed uniform across countries. Muhleisen (1996) uses a vector autoregression (VAR) process in loga- rithms to jointly model the relationship between the private saving rate and growth and between private and public saving (that is, Ricardian equivalence). Using Granger-causality tests, Muhleisen shows that growth leads to both higher pub- lic and private saving rates. Demography Lahiri (1989) finds that the age-dependency ratio (the fraction of the popula- tion under the age of 16 and over the age of 64) is a significant determinant of private saving. Under Lahiri's specification a 1 percentage point increase in the Loayza and Shankar 579 dependency ratio lowers the long-run average propensity to save by 1.6 percent- age points in India, the Republic of Korea, Malaysia, Singapore, and Sri Lanka. Muhleisen (1996) finds that the age-dependency ratio is the most significant de- terminant of private saving, with the usual negative relationship between the two variables. Level of Financial Development According to Muhleisen (1996) public policy could play a role in providing credible and stable rules for financial intermediation and thus, by encouraging saving in financial assets, also further the development of financial markets. He highlights the need for reform in insurance markets, pension schemes, and mutual funds. He calls for policies that would mobilize greater financial saving and favor long-term saving instruments, such as reducing the government's recourse to cap- tive saving by allowing greater flexibility in portfolio allocation to pension funds and the Life Insurance Corporation. Muhleisen bases his call for greater financial development on his empirical results, in particular those that find that the ratio of M2 to GDP-a proxy for financial depth-has a positive long-run relationship with private saving. His methodological tool is cointegration analysis of time-series data, following the maximum-likelihood method of Johansen (1991). Urban versus Rural Propensities to Save Pandit (1991) tests single-equation cross-sectional models of aggregate house- hold saving with the purpose of, among others, contrasting the propensities to save in rural and urban areas. He compares cross-sectional data for 1967-68 and 1975-76. He does not find convincing or consistent evidence supporting the view that both the average propensity to save and the marginal propensity to save are higher in urban areas and thus that a worsening of agricultural terms of trade could lead to higher saving. Pandit does not attempt a more disaggregated analy- sis, that is, one that studies the propensities to save at the level of the physical and financial components of household saving. This distinction might be important given that urban and rural households have different access to financial saving instruments. Rate of Interest The presence of imperfect, segmented capital markets with administered rates of interest in India may make estimating the interest elasticity of private saving problematic. The available empirical evidence (see, for example, Muhleisen 1996) does not support the hypothesis that aggregate private saving is increasing in the real rate of interest. Muhleisen argues that, to the extent that the measured inter- est rate reflects only the rate of return on controlled financial instruments and not the rate of return on investment, changes in measured interest rates mostly generate a substitution effect between physical and financial assets and between different kinds of financial assets. 580 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Decomposition of Private Saving Pandit (1991) further decomposes private saving into household and corpo- rate saving, mostly in an attempt to explain the household's portfolio decision. * Household physical and household financial saving. Pandit finds that house- hold investment (equal to household physical saving) declines when house- hold financial saving increases, suggesting substitutability between the two components of household saving and, therefore, the need to treat them jointly. Further decomposing financial saving into the real demand for money, sav- ing in time deposits, saving in insurance premiums, and saving in provident funds, Pandit finds that the composition of household financial saving is, as expected, driven by the rates of return on each type of financial saving and, to some extent, by bank expansion. * Corporate saving. Corporate saving is defined as the excess of profits over dividends. Pandit estimates a two-equation model using ordinary least squares, one equation for after-tax corporate profits and one equation for corporate dividends. He finds that after-tax profits are positively related to sales and to the nonagricultural terms of trade and negatively correlated with per unit wage costs. Dividends are explained by after-tax profits and the firm's access to external financing. [II. Do HOUSEHOLDS IN INDIA "PIERCE THE CORPORATE VEIL"? The household sector, which includes unincorporated enterprises, is the ulti- mate owner of incorporated business. The question we address in this section is the extent to which the household sector takes into account the saving decisions of corporations in formulating its own saving decisions. In order for households to treat corporate saving as their own, they must both understand corporate actions and have a marginal propensity to save out of wealth equal to the mar- ginal propensity to save out of disposable income. As Poterba's (1987) seminal paper explains, if households "pierce the corporate veil," aggregate private sav- ing becomes the variable of interest, and little information is gained from break- ing it down further into household and corporate saving. However, if household and corporate saving decisions are made independently, then they must be stud- ied separately, for it is likely that they have different determinants. We argue here that household saving reacts sufficiently to corporate saving so as to allow a focus on aggregate private saving. We study the relationship between household and corporate saving in the frame- work of time-series cointegration. Augmented Dickey-Fuller tests provide evi- dence that all series involved are integrated of order 1 and, therefore, that cointegration analysis can be relevant (table A-1). However, if household saving is perceived as equivalent to corporate saving, the two variables should be cointegrated, possibly including other variables in the long-run relationship, with the cointegrating vector (1, 1). Loayza and Shankar 581 We first test for cointegration using the Engle and Granger approach and then follow a maximum-likelihood method to estimate the cointegrating vector, al- lowing for short-run dynamics (see table 7). In addition to the ratios of (unad- justed) household and corporate saving to private disposable income, we include in the model unadjusted private disposable income, the dependency ratio, the real interest rate (an ex post composite of borrowing and lending rates), and the rate of inflation (to serve as an additional measure of the rate of return on non- financial assets). The saving rates and income figures are not adjusted for infla- tionary capital losses, given our inability to apportion the adjustment on private sector saving between the household and corporate sectors. However, we do account for consumer durables as a form of household saving. Thus we present estimation results for two representations of the dependent variable: the house- hold saving ratios obtained excluding and including durable consumption. We reject the null of no cointegration when we include in the model the vari- ables mentioned above (other variables could be excluded from the long-run re- lationship of household saving). Both excluding and including consumer durables in household saving, the coefficients on the corporate saving ratio are signifi- cantly negative and large in magnitude (table 7). This result is consistent with the hypothesis that households allocate their portfolio between corporate and house- Table 7. Empirical Results: Do Indian Households Pierce the "Corporate Veil"? Household saving Household saving ratio not including ratio including Variable consumption durables consumption durables Dependency ratio 1.2 -0.18 (4.4)* (9.5) Inflation in GDP deflator -0.35 0.09 (2.7)* (1.94) Real interest rate -0.16 0.24 (1.7) (5.9)# Ratio of corporate saving to unadjusted -0.72 -0.99 private disposable income (5.07)- (11.3)P Log of real unadjusted private 0.49 0.05 disposable income per capita (6.12)' (18.5)- Engel-Granger cointegration test (Ho: no cointegration): Augmented Dickey-Fuller test -5.5/14.7` -8.01/-4.7* statistic/5 percent critical value (Phillips-Ouliaris) Test on piercing the corporate veil (Ho: coefficient on corporate saving rate =-1): t-test statistic/critical value 1.97/1.96"' 0.114/1.96* * Significant at 5 percent. Note: t-statistics are in parentheses. Source: Authors' calculations based on data in the World Bank's World Saving Database. 582 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 hold saving and that there is a strong substitution between the two. Moreover, in the case of household saving including consumer durables, the estimated coeffi- cient is close to and not statistically different from -1, indicating that (correctly measured) household saving offsets a change in corporate saving. This result is consistent with the hypothesis that households pierce the corporate veil and thus indicates that we should focus on the behavior of aggregate private saving. We follow this strategy in the following section, where we study the economic and demographic determinants of private saving rates. IV. DETERMINANTS OF PRIVATE SAVING In this section we examine how the evolution of the private saving rate over 1960-94 has been related to the behavior of other economic and demographic variables. Instead of adhering to a specific consumption/saving model, we esti- mate a reduced-form model with the private saving rate as the dependent vari- able and its most important proposed determinants as explanatory variables. We define the public sector as including not only the central government but also state and local governments and public enterprises. We consider three new mea- sures of the private saving rate, each of which represents an incremental adjust- ment to the naive (unadjusted) measure. Thus we consider the unadjusted saving rate, the saving rate adjusted for inflationary capital gains and losses, the inflation- adjusted saving rate that includes expenditures on consumer durables, and the adjusted saving rate that includes expenditures on education and health in addi- tion to consumer durables. Using the conceptually correct definition of the public sector, adjusting for capital gains, and applying the theoretically correct concept of net worth (which should include consumer durables and human capital assets) lead us to different results. In selecting the explanatory variables, we take into account the literature on Indian saving, the current cross-country studies on private saving (see Loayza, Schmidt-Hebbel, and Serven 2000), and the availability of data for the whole period. The explanatory variables we consider are the ratio of public saving to private income (to evaluate the extent of Ricardian equivalence), the ratio of private domestic credit to GDP (to examine the importance of financial depth), the dependency ratio (to account for life-cycle effects), the share of agriculture in GDP (to gauge the effect of occupational structure and income uncertainty), the real interest rate (to serve as the relative price of current consumption with re- spect to future consumption), and the log of per capita disposable income (to examine income elasticities and the importance of subsistence consumption). We use aggregate annual data and therefore work with a sample of 35 obser- vations. We are interested mostly in the long-run relationship between the pri- vate saving rate and various economic variables. However, working with annual data forces us to consider a model that also takes short-run effects into account. In addition, unit-root tests indicate that the variables in our model are integrated of order 1 (see table A-1). Both the need to account for long-run and short-run Loayza and Shankar 583 effects and the importance of avoiding spurious correlations among integrated variables lead us to estimate the relationship between saving and its determinants using an error-correction model (see Johansen 1991 and Pesaran and Shin 1997, 1999). We performed an Engle-Granger test for cointegration (see table 8), and we could not reject the hypothesis that the variables in the model are cointegrated; that is, at least one linear combination of the variables is stationary. We estimate the error-correction model under the assumption that there is a single long-run relationship between private saving rates and their proposed determinants. Al- though Johansen tests indicated the presence of more than one cointegrating re- lationship, we chose to work under the restriction of a single cointegration vec- tor, given that we did not have appropriate restrictions to differentiate between various long-run saving relationships (see Pesaran 1997). Given that we estimate a dynamic model with only 35 observations, degrees-of-freedom considerations prevent us from either working with a larger set of explanatory variables or ex- amining the importance of nonlinear and interactive effects. Correlation Results Before presenting the estimation results, we consider simple time-series corre- lations between saving rates and potential saving determinants (table A-2). The first point to notice is that the four measures of the private saving rate are highly correlated. The correlation between the unadjusted and inflation-adjusted pri- vate saving rates is 0.87, which means that the two series share basic trends and cyclical fluctuations. Introducing consumer durables produces a larger change in the saving figures (the correlation of the series with and without durables is 0.75), whereas the inclusion of health and education expenditures produces a series that is highly correlated with the series that ignores them (correlation 0.99). Regarding public saving ratios, it appears that the inflation adjustment does produce a major change in the series (correlation 0.39), while the inclusion of health and education expenditures does not (correlation 0.99). The private sav- ing rate that includes all adjustments is negatively correlated with the depen- dency ratio (-0.74) and the share of agriculture in GDP (-0.72) and is positively correlated with per capita private income (0.72), financial depth (0.65), the real interest rate (0.50), and the government saving rate (0.43). We do not dwell on the meaning of these simple correlations because, as we will see, the signs of the corresponding coefficients change when estimated in a multivariate setting. We present the correlation coefficients to serve as a reference when analyzing the results of the error-correction model. Estimation Results Although we present the estimated long-run and short-term coefficients, in- cluding the adjustment term, of the error-correction model (tables 8 and 9), we emphasize the long-run coefficients for two reasons. First, we are mainly inter- ested in the long-term evolution of the private saving rate, and, second, most of the estimated short-term parameters are not statistically significant. Among the 584 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 8. Empirical Results: Determinants of the Private Saving Rate, Long- Run Effects Unadjusted Adjusted private Adjusted private private saving! Adjusted saving including saving including unadjusted private saving! consumer consumer durables private adjusted private durables/adjusted and human capitall disposable disposable private disposable adjusted private Variable income income income disposable income Public saving/private -0.166 -0.411* -0.31 * -0.454* disposable income' (1.07) (4.76) (2.79) (3.19) Domestic -0.002 -0.002* -0.00003 0.0002 credit/GDP (7.13's) (4.97) (0.07) (0.456) Dependency ratio -1.26"' -2.04 * -0.657* -1.25* (4.83) (8.05) (2.8) (4.9) Real interest rate -0.1087'" 0.08 0.28* 0.26* (4.23) (1.56) (5.47) (4.31) Log of real private 0.05 -0.09* -0.16* -0.21 * disposable income (1.96) (2.7) (4.44) (5.21) per capitab Share of agriculture 0.69* 1.14* 0.06 0.47* in GDP (4.17) (6.57) (0.39) (2.8) Engel-Granger cointegration test (Ho: no cointegration): Augmented -5.961-4.74 -5.18/-4.74 -5.241-4.74 -5.17/-4.74 Dickey-Fuller test statistic/5 percent critical value (Phillips-Ouliaris) *Significant at 5 percent. Note: t-statistics are in parentheses. a. Adjusted for inflation if dependent variable is adjusted private saving/private disposable income; adjusted for inflation and inclusive of human capital expenditure if dependent variable is adjusted private saving including expenditure on human capital/adjusted private disposable income. If dependent variable is private saving including consumer durables/adjusted private disposable income, then we use adjusted public saving unaugmented/adjusted private disposable income; if dependent variable is unadjusted private saving/unadjusted private disposable income, we use unadjusted public saving/unadjusted private disposable income. b. If dependent variable is adjusted private saving/adjusted private disposable income, augmented or unaugmented, we use log of real private disposable income per capita; if dependent variable is unadjusted private saving/unadjusted private disposable income, we use log of real unadjusted private disposable income per capita. Source: Authors' calculations based on data in the World Bank's World Saving Database. measures of the private saving rate, we give more importance to the measure that we believe is conceptually most correct: the private saving rate that adjusts for inflationary capital gains and losses and includes expenditures on consumer du- rable goods and expenditures on health and education. The first explanatory variable is the public saving ratio. We can reject full Ricardian equivalence for all measures of the private saving rate. That is, the coefficient on the public saving ratio is statistically less than -1. However, we do find a negative relationship between private and public saving rates. This rela- Loayza and Shankar 585 tionship is significant for all measures except for the unadjusted private saving rate. The "offset" coefficient is lowest for the unadjusted measure of private saving (-0.166) and largest for the most conceptually correct measure (-0.454). The latter result indicates that, beyond internalizing the public budget constraint, the private sector considers some degree of substitution between public and pri- vate expenditure on health and education. It is noteworthy that we find an offset between private and public saving only when we account for inflationary capital gains and losses. Assuming that we have made this adjustment correctly, we can interpret the offset as an indication that the private sector internalizes the government budget constraint and that the private sector recognizes that inflation entails a transfer of income to the government. Table 9. Empirical Results: Determinants of the Private Saving Rate, Short- Run Effects Inflation-adjusted Inflation-adjusted private saving private saving Unadjusted Inflation- including including consumer private savizg! adjusted private durablesl durables and unadjusted saving/adjusted adjusted human capital private private private expenditure/adjusted disposable disposable disposable private disposable Variablea income income income inconme CR -0.39 -0.30 -0.07 -0.015 (1.46) (0.77) (0.17) (0.04) Self(-1) -0.30 -0.09 -0.04 -0.06 (1.24) (0.28) (0.09) (0.04) D[DOMCRGDP(-1)] 0.0004 -0.12 0.0002 0.0002 (0.69) (1.05) (0.16) (0.19) D[RR70OEX(-l)] 1.30* 0.07 -0.14 -0.14 (2.08) (0.68) (1.26) (1.31) D[SHAGRI(-1)] -0.14 0.65 0.29 0.29 (0.49) (1.37) (0.55) (0.55) D[GSR(-1)]y 0.39 -0.36 -0.20 -0.22 (1.31) (1.44) (0.67) (0.74) D[LRPDI(-1)]' 0.04 -0.25 -0.19 -0.20 (0.32) (1.40) (0.88) (0.96) D[DEP(-1)] -2.53 -3.14 -0.47 -0.01 (1.25) (0.90) (0.15) (0.003) Note: t-statistics are in parentheses. a. See table A-2. b. Adjusted for inflation if dependent variable is adjusted private saving/private disposable income; adjusted for inflation and inclusive of human capital expenditure if dependent variable is adjusted private saving including expenditure on human capital/adjusted private disposable income. If dependent variable is private saving including consumer durables/adjusted private disposable income, then we use adjusted public saving unaugmented/adjusted private disposable income; if dependent variable is unadjusted private saving/unadjusted private disposable income, we use unadjusted public saving/unadjusted private disposable income. c. If dependent variable is adjusted private saving/adjusted private disposable income, augmented or unaugmented, we use log of real private disposable income per capita; if dependent variable is unadjusted private saving/unadjusted private disposable income, we use log of real unadjusted private disposable income per capita. Source: Authors' calculations based on data in the World Bank's World Saving Database. 586 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 However, another, less amenable, interpretation is also possible. If the inflation- related adjustment introduces measurement error in private and public saving, the negative relationship between the two would be spurious (given that, by construc- tion, the measurement error would have the opposite sign for private and public saving). Unfortunately, our methodology does not control for measurement error, and the interpretation of the negative coefficient on public saving has to rely on the quality of the adjustment. We believe that, on balance, the adjustment for inflation is correct and brings the saving figures closer to analytically correct measures. Fur- thermore, the size of the offset coefficient is consistent with that estimated for other countries, in both panel and single-country studies (see Loayza, Schmidt-Hebbel, and Serven 2000 and Corbo and Schmidt-Hebbel 1991). The second explanatory variable, the ratio of domestic credit to the private sector to GDP, is an indicator of financial depth. It has a negative and significant relationship with private saving rates, both unadjusted and adjusted for infla- tionary capital gains and losses. This negative relationship is consistent with the notion that financial development allows households and small firms to use col- lateral more widely to reduce down payments on loans for housing and con- sumer durables. This should reduce private saving as individuals are able to fi- nance higher consumption at their current income level. When the measure of private saving includes consumer durables, the negative relationship drops nota- bly (in absolute value) and becomes insignificant. This indicates that financial development induces private agents to change the composition of their assets to favor durable goods but does not affect the total volume of saving once this is correctly measured. This conclusion invites a reinterpretation of the finding that financial development reduces private saving, which is common in several indi- vidual and cross-country studies that ignore consumer durables (see Loayza, Schmidt-Hebbel, and Serven 2000 and Bandiera and others 2000). The third explanatory variable is the dependency ratio, which captures life-cycle effects. As expected, its estimated coefficient is negative and significant for all four measures of the private saving rate. In other words, the private saving rate moves in the same direction as the share of working-age people in the total population. Thus India's demographic transition in the past 30 years must have contributed to expand the aggregate private saving rate. We should expect that as India moves to the next phase of the demographic transition, in which the share of old people in total population expands, private saving rates will decrease accordingly. The effect of the real interest rate on saving differs drastically depending on the measure of saving used. The effect of the real interest rate on unadjusted private saving appears to be negative, meaning that the (negative) income effect of an increase in the real interest rate dominates the (positive) substitution effect. Once we adjust for inflation-related capital losses, however, the effect of interest rate changes sign, although it is not statistically significant. Given that our mea- sure of real interest rates is derived from nominal rates that may not respond fully to price changes, a decrease in real interest rates reflects a rise in inflation. Thus the more positive impact of real interest rates on saving rates once they are Loayza and Shankar 587 inflation-adjusted can be explained by the fact that a change in inflation affects both real interest rates and adjusted saving rates in the same direction. A corol- lary of this explanation is that if we had a measure of real interest rates that were fully independent of inflation, we would not see a different effect of the former on adjusted or unadjusted private saving rates. When we include consumer durables in the measure of private saving, the positive effect of the real interest rate becomes stronger and statistically significant. Taken at face value, this indi- cates that the substitution effect of interest rate changes is larger (or the income effect is smaller) when applied to expenditures on durables with respect to other forms of saving. The effect of per capita private disposable income on saving is also highly dependent on the measure of private saving used. This effect is positive for unad- justed saving, but becomes significantly negative for inflation-adjusted saving and even more negative when saving is augmented to include consumer durables and expenditures on education and health. Most of the literature finds that per- manent rises in income have a positive effect on the private saving rate. This has been regarded as a puzzle, given that a permanent rise in income should increase consumption by the same magnitude, and thus have no effect on the level of saving, but should reduce the saving rate. In India the puzzling positive effect of income does not apply once we correct for inflationary wealth losses and include durables and human capital investment in the measure of private saving. That is, a permanent increase in income is consumed, and this consumption mostly takes the form of nondurables. Given that in a country as poor as India most people have yet to satisfy their vital consumption needs, it is only natural that an in- crease in income translates into an increase in basic consumption, and not into financial saving, durable consumption, or expenditures on human capital. Finally, we consider the long-run effect on saving of an occupation structural variable, the share of agriculture in GDP. This variable is important given that India remains a largely agricultural economy. The dominance of agriculture im- plies that a large share of the population faces an uncertain income. The share of agriculture in GDP appears to have a positive effect on private saving rates (table 8). This finding is consistent with a precautionary saving motive. Taking into account monsoon risk, Indian farmers have a higher propensity to save than people dedicated to other activities. For the sake of completeness, we present the short-term coefficients of the error-correction model for each measure of the private saving rate. These coeffi- cients are, however, uninteresting as they are not statistically significant. V. CONCLUSION: WHAT IS SPECIAL ABOUT SAVING IN INDIA? India's saving rate has been consistently higher than that of most other coun- tries with comparable per capita income. Regarding aggregate national saving as a ratio of gross national disposable income, India has performed as well as the OECD countries. India also traditionally has relied largely on national saving to 588 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 fuel investment needs, relying relatively less on foreign saving compared with other developing countries. Even the 1991 liberalization that led to an increase in foreign capital inflows has not significantly changed this scenario. Given the ambitious growth targets of the government and the current global environment, it is likely that policies oriented toward raising aggregate national saving will play a key role in Indian economic development. Our results on the determinants of private saving rates in India highlight six issues. First, Indian households that save appear to pierce the corporate veil, in that they internalize the saving actions of the corporate sector. This finding supports our strategy of focusing on aggregate private saving, rather than on its components. We find equivalence between household and corporate saving only when we augment the measure of household saving to include expendi- tures on durable goods. As with many other conclusions of the article, the cor- rections we make to measures of private saving are key to uncovering impor- tant relationships. Second, the private saving rate rises with the share of agriculture in GDP. A large share of India's GDP comes from agriculture, and a large proportion of the population lives in rural areas. This suggests that the income of many households is characterized by the uncertainty associated with agriculture in India. Uncer- tainty should introduce a precautionary motive to save that should manifest itself as a positive relationship between aggregate private saving rates and the share of agricultural income in GDP. Our empirical results support this claim. Third, the real interest rate is positively associated with private saving rates once these are adjusted for inflation-related capital losses and augmented to in- clude consumer durables. Taken at face value, this result indicates that in India the substitution effect of interest rate changes is larger (or the income effect is smaller) when applied to consumption of durables with respect to other forms of saving. A controversial implication of this result is that tax incentives that in- crease the net rate of return on saving instruments could increase private saving rates. We are, however, hesitant to recommend tax incentives given their distor- tionary effect on resource allocation and the associated fiscal impact. Fourth, in India financial development has induced private agents to change the composition of their assets in favor of consumer durable goods, but this does not affect the total volume of saving once it is correctly measured. Extending this result to other countries, this conclusion invites a reinterpretation of the finding that financial development reduces private saving (see Loayza, Schmidt-Hebbel, and Serven 2000 and Bandiera and others 2000). Fifth, in India the puzzling positive effect of income on private saving rates found in other studies does not apply once we correct for inflationary wealth losses and include durables and human capital investment in the measure of pri- vate saving. That is, a permanent increase in income is consumed, and this con- sumption mostly takes the form of nondurables. Whether this reflects the behav- ior of forward-looking agents who are not financially restricted or the behavior of people living below subsistence consumption, we cannot tell for sure. Natu- Loayza and Shankar 589 rally, given the extent of poverty in India, we favor the subsistence-consumption interpretation. Last, the dependency ratio has a negative effect on private saving rates, as expected from life-cycle considerations. Thus India's demographic transition in the past 30 years must have contributed to an increase in the aggregate private saving rate. As India moves to the next phase of the demographic transition, in which the share of elderly people in the total population expands, private saving rates should decrease accordingly. Although important, demographic variables such as the dependency ratio do not appear to be the sole determinants of private saving rates, as previous studies had indicated. We contend that the differences between our study's findings and previous findings are at least partially due to our use of inflation-corrected figures for income and saving and our inclusion in saving of private expenditures on consumer durables and accumulation in hu- man capital. Table A-1. Unit-Root Test Results Adjusted Dickey- Fuller test statistic! Order of Variable critical value' integration DEP (dependency ratio) 0.38/-3.6 I(1) ' 1 lag; intercept GSR (inflation-adjusted public saving including human -1.99/-3.6 I(1)-- capital expenditure/adjusted private disposable income) 1 lag; intercept GSRJPU (unadjusted public saving/unadjusted private -0.99/-3.6 I(1)*# disposable income) 1 lag; intercept GSRPA (inflation-adjusted public saving/adjusted private -1.3/-3.6 I(1) disposable income) 1 lag; intercept PSR (inflation-adjusted private saving including human -1.7/-3.6 I(1) # capital expenditure/adjusted private disposable income) 1 lag; intercept PSRJPA (inflation-adjusted private saving/adjusted private -2.0/-3.6 I(1) disposable income) 1 lag; intercept PSRCON (inflation-adjusted private saving including -1.8/-3.6 I(1)** durables/adjusted private disposable income) 1 lag; intercept PSR_PU (unadjusted private saving/unadjusted private -1.2/-3.6 I(1)* disposable income) 1 lag; intercept RR70-EX (real ex post rate of interest) -3.2/-3.6 1(1)## 2 lags; intercept SHAGRI (share of agricultural income in GDP) -0.81/-3.6 1(1)#* 2 lags; intercept LRPDI_PU (natural log of unadjusted private disposable 1.2/-3.6 1) * income per capita) 1 lag; intercept LRPDI_PA (natural log of adjusted private disposable 1.39/-3.6 I(1)## income per capita) 1 lag; intercept CORPSR (corporate saving/unadjusted private -0.87/-3.6 I(1)## disposable income) 1 lag; intercept HHCON (household saving including durables/ -1.28/-3.6 1(1)*# unadjusted private disposable income) 1 lag; intercept DOMCRGDP (domestic credit to the private sector/GDP) -1.1/-3.6 I(1) * 1 lag; no inter- cept, no trend Significant at 5 percent. * * Significant at 1 percent; the null hypothesis is that there is a unit root. Source: Authors' calculations based on data in the World Bank's World Saving Database. 590 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table A-2. Correlations between Saving Rates and Potential Saving Determinants Inflation-adjusted public saving Corporate including human saving! Domestic capital unadjusted credit to the expenditure! private private adjusted private disposable Dependency sector! disposable income ratio GDP income Corporate saving/unadjusted private disposable income 1.00 -0.17 0.16 0.02 Dependency ratio -0.17 1.00 -0.88 -0.61 Domestic credit to the private sector/GDP 0.16 -0.88 1.00 0.62 Inflation-adjusted public saving including human capital expenditure/adjusted private disposable income 0.02 -0.61 0.62 1.00 Inflation-adjusted public saving/adjusted private disposable income -0.01 -0.49 0.53 0.99 Unadjusted public saving/ unadjusted private disposable income 0.03 0.39 -0.25 0.29 Household saving rate -0.13 -0.87 0.74 0.76 Inflation rate -0.05 -0.19 0.16 0.49 Natural log of adjusted private disposable income per capita 0.16 -0.95 0.79 0.47 Natural log of unadjusted private disposable income per capita 0.14 -0.97 0.82 0.54 Inflation-adjusted private saving including human capital expenditure/adjusted private disposable income 0.06 -0.74 0.65 0.51 Inflation-adjusted private saving/adjusted private disposable income 0.42 -0.72 0.57 0.35 Unadjusted private saving/ unadjusted private disposable income 0.29 -0.88 0.74 0.69 Inflation-adjusted private saving including durables/adjusted private disposable income 0.04 -0.73 0.62 0.47 Real ex post interest rate 0.23 -0.36 0.22 -0.07 Share of agricultural income in GDP -0.17 0.98 -0.86 -0.67 Loayza and Shankar 591 Inflation adjusted public Unadjusted savingl public saving! adjusted unadjusted Natural log of private private adjusted private disposable disposable Household Inflation disposable income income saving rate rate income per capita -0.01 0.03 -0.13 -0.05 0.16 -0.49 0.39 -0.87 -0.19 -0.95 0.53 -0.25 0.74 0.16 0.79 0.99 0.29 0.76 0.49 0.47 1.00 0.39 0.68 0.51 0.35 0.39 1.00 -0.14 0.00 -0.47 0.68 -0.14 1.00 0.20 0.85 0.51 0.00 0.20 1.00 0.03 0.35 -0.47 0.85 0.03 1.00 0.43 -0.45 0.88 0.10 0.99 0.43 0.07 0.75 0.03 0.72 0.27 -0.03 0.68 -0.22 0.77 0.61 -0.12 0.90 0.14 0.88 0.39 0.02 0.73 0.04 0.72 -0.14 -0.11 0.34 -0.72 0.54 -0.57 0.33 -0.87 -0.27 -0.90 (Table continues on the following page.) 592 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table A-2. (continued) Inflation-adjusted private savings Inflation-adjusted Natural log of including human private unadjusted private capital expenditure/ saving/adjusted disposable income adjusted private private disposable per capita disposable income income Corporate saving/unadjusted private disposable income 0.14 0.06 0.42 Dependency ratio -0.97 -0.74 -0.72 Domestic credit to the private sector/GDP 0.82 0.65 0.57 Inflation-adjusted public saving including human capital expenditure/adjusted private disposable income 0.54 0.51 0.35 Inflation-adjusted public saving/adjusted private disposable income 0.43 0.43 0.27 Unadjusted public saving/ unadjusted private disposable income -0.45 0.07 -0.03 Household saving rate 0.88 0.75 0.68 Inflation rate 0.10 0.03 -0.22 Natural log of adjusted private disposable income per capita 0.99 0.72 0.77 Natural log of unadjusted private disposable income per capita 1.00 0.71 0.74 Inflation-adjusted private saving including human capital expenditure/adjusted private disposable income 0.71 1.00 0.77 Inflation-adjusted private saving/adjusted private disposable income 0.74 0.77 1.00 Unadjusted private saving/ unadjusted private disposable income 0.89 0.74 0.87 Inflation-adjusted private saving including durables/adjusted private disposable income 0.71 0.99 0.75 Real ex post interest rate 0.47 0.50 0.72 Share of agricultural income in GDP -0.93 -0.72 -0.69 Source: Authors' calculations based on data in the World Bank's World Saving Database. Loayza and Shankar 593 Inflation-adjusted private saving Unadjusted private including saving/unadjusted durablesladjusted Share of private disposable private disposable Real ex post agricultural income income interest rate income in GDP 0.29 0.04 0.23 -0.17 -0.88 -0.73 -0.36 0.98 0.74 0.62 0.22 -0.86 0.69 0.47 -0.07 -0.67 0.61 0.39 -0.14 -0.57 -0.12 0.02 -0.11 0.33 0.90 0.73 0.34 -0.87 0.14 0.04 -0.72 -0.27 0.88 0.72 0.54 -0.90 0.89 0.71 0.47 -0.93 0.74 0.99 0.50 -0.72 0.87 0.75 0.72 -0.69 1.00 0.72 0.47 -0.88 0.72 1.00 0.50 -0.71 0.47 0.50 1.00 -0.28 -0.88 -0.71 -0.28 1.00 594 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Athukorala, Premachandra, and Kunal Sen. 1995. "Economic Reforms and Rate of Sav- ing in India." Economic and Political Weekly 30:2184-90. Bandiera, Oriana, Gerard Caprio, Patrick Honohan, and Fabio Schiantarelli. 2000. "Does Financial Reform Raise or Reduce Private Savings?" Review of Economics and Statis- tics 82(2):239-63 Cashin, Paul, Nilss Olekalns, and Ratna Sahay. 1998. "Tax Smoothing in a Financially Repressed Economy: Evidence from India." IMF Working Paper WP/98/122. Interna- tional Monetary Fund, Washington, D.C. Processed. Corbo, Vittorio, and Klaus Schmidt-Hebbel. 1991. "Public Policies and Saving in Devel- oping Countries." Journal of Development Economics 36(1):89-115. Haque, Nadeem U., and Peter Montiel. 1989. "Consumption in Developing Countries: Tests for Liquidity Constraints and Finite Horizons." Review of Economics and Sta- tistics 71(August):408-15. Johansen, Sorenoren. 1991. "Estimation and Hypothesis Testing of Cointegrating Vec- tors in Gaussian Vector Autoregressive Models." Econometrica 59:1551-80. Jorgenson, Dale W., and Barbara M. Fraumeni. 1989. "The Accumulation of Human and Nonhuman Capital, 1948-84." In Robert E. Lipsey and Helen Stone Tice, eds., The Measurement of Saving, Investment, and Wealth. Cambridge, Mass.: National Bureau of Economic Research. Lahiri, Ashok. 1989. "Dynamics of Asian Savings: The Role of Growth and Age Struc- ture." IMF Staff Papers 36:228-61. Loayza, Norman, Humberto L6pez, Klaus Schmidt-Hebbel, and Luis Serven. 1998. "The World Saving Data Base." World Bank, Washington, D.C. Processed. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serven. 2000. "What Drives Private Saving across the World?" Review of Economics and Statistics 82(2):165-81. Muhleisen, Michael. 1996. "India-Policies to Increase Domestic Saving." International Monetary Fund, Washington, D.C. Processed. Pandit, B. L. 1991. The Growth and Structure of Savings in India. Bombay: Oxford University Press. Pesaran, Hashem. 1997. "The Role of Economic Theory in Modelling the Long-Run." Economic Journal 107:178-91. Pesaran, Hashem, and Yongcheol Shin. 1997. "An Autoregressive Distributed Lag Mod- elling Approach to Cointegration Analysis." University of Cambridge, Cambridge, U.K. Processed. . 1999. "Long-Run Structural Modelling." University of Cambridge, Cambridge, U.K. Processed. Poterba, James M. 1987. "Finite Lifetimes and the Effects of Budget Deficits on National Saving." Journal of Monetary Economics 20(September):369-91. A NEW DEVELOPMENT DATABASE The following article is one in an occasional series introducing new databases. The series intends to make new development databases more widely available and to contribute to discussion and further research on economic development issues. The databases included in the series are selected for their potential useful- ness for research and policy analysis on critical issues in developing and transi- tion economies. Some are drawn from micro-level firm or household surveys; others contain country-level data. The authors describe the data contents, crite- ria for inclusion or exclusion of values, sources, strengths and weaknesses, and any plans for maintenance or updating. Each database is available on the Internet, at the address provided in the article. THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 597-605 A New Database on the Structure and Development of the Financial Sector Thorsten Beck, Ash Demirgui-Kunt, and Ross Levine This article introduces a new database of indicators of financial structure and financial de- velopment across countries and over time. The database is unique in that it combines a wide variety of indicators that measure the size, activity, and efficiency of financial intermediaries and markets. It improves on previous efforts by presenting data on the public share of commercial banks, introducing indicators of the size and activity of nonbank financial insti- tutions, and constructing measures of the size of bond and primary equity markets. An expanding literature establishes the importance of financial development for economic growth (Levine 1997). Measures of the size of the banking sector and the size and liquidity of the stock market are shown to be highly correlated with subsequent growth of gross domestic product (GDP) per capita. Moreover, emerging evidence suggests that both the level of banking and the development of the stock market have a causal impact on economic growth.1 The recent financial crises in Southeast Asia and Latin America further underline the importance of a well- functioning financial sector for the whole economy. This article introduces a new database, the first to provide comprehensive measures of the development, structure, and performance of the financial sector. The database includes statistics on the size, activity, and efficiency of banks, nonbank institutions, equity markets, and bond markets (including both primary and secondary markets) across a broad spectrum of countries and through time. It thus enables financial analysts and researchers to investigate a wide array of issues.2 1. See King and Levine (1993a, 1993b) and Levine and Zervos (1998) for the correlation between these variables, and Levine, Loayza, and Beck (2000), Beck, Levine, and Loayza (2000), Neusser and Kugler (1998), and Rousseau and Wachtel (1998) for evidence on causality. Also Demirgii-Kunt and Maksimovic (1998) show that firms in countries with an active stock market and large banking sector grow faster than predicted by the characteristics of individual firms. Rajan and Zingales (1998) show that industries that rely more heavily on external finance grow faster in countries with better-developed financial systems. 2. The database and further details on its construction are available at the following address: http:ll www.worldbank.org/research/projects/finstructure/database.htm. Thorsten Beck is with the Financial Sector Policy and Strategy Group at the World Bank, Ash Demirgiuc- Kunt is lead economist with the Development Research Group at the World Bank, and Ross Levine is with the Carlson School of Management at the University of Minnesota. Their e-mail addresses are tbeck@worldbank.org. ademirguckunt@worldbank.org, and rlevine@csom.umn.edu. The authors are grateful to Joe Attia and Ian Webb for technical assistance and to Gerard Caprio and two anonymous referees for comments. © 2000 The International Bank for Reconstruction and Development/THE WORLD BANK 597 598 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Thus far the profession has relied on a few indicators of the banking sector and the stock market using data from the International Monetary Fund's (IMF) International Financial Statistics and the International Finance Corporation's Emerging Market Database. To build the database introduced in this article, we draw on a wider array of sources and develop indicators of the size, activity, and efficiency of a much broader set of financial institutions and markets. We use bank-specific data to construct indicators of the market structure and efficiency of commercial banks. Furthermore, we compile data on the split between public and private ownership in the banking sector. This database is the first to define and construct indicators of the size and activity of nonbank financial intermediaries, such as insurance companies, pen- sion funds, and nondeposit money banks. It is also the first to include indicators of the size of primary equity markets and primary and secondary bond markets. In constructing the database, we carefully deflate measures and match stock and flow variables.3 This effort produces a unique set of indicators that capture the development and structure of the financial sector across countries and over time along many different dimensions (table 1). I. THE SIZE AND ACTIVITY OF FINANCIAL INTERMEDIARIES The first group of measures compares the size and activity of central banks, deposit money banks, and other financial institutions relative to each other and relative to GDP. The IMF'S International Financial Statistics covers the period from 1960 to 1997 for up to 175 countries. Groups of Financial Institutions The indicators in this section distinguish between three groups of financial institutions.4 The first group-central banks-comprises the central bank and other institutions that perform the functions of the monetary authorities.5 The second group-deposit money banks-comprises all financial institutions that have "liabilities in the form of deposits transferable by check or otherwise usable in making payments" (IMF 1984: 29). The third group-other financial institu- tions-comprises other bank-like institutions and nonbank financial institutions. 3. We deflate the end-of-period stock variables by end-of-period deflators, and we deflate the flow variables by deflators measured over the period. Then we use the average of the real stock variable in years t and t - 1 and the real value of the flow variable in year t. We use the consumer price index from the International Financial Statistics as the deflator. Since we detail the raw data at the website, researchers can easily reconstruct our indicators using alternative deflators. 4. For a detailed description see IMF (1984). The three groups correspond to lines 12, 22, and 42 of the International Financial Statistics. 5. Exchange stabilization funds are the most typical case of monetary authority functions that are performed separately from the central banks' balance sheets. Furthermore, the central bank might perform commercial banking tasks. Where possible, we exclude these from the central bank balance sheets when they are reported in the International Financial Statistics. Beck, Demirgui,-Kunt, and Levine 599 Table 1. Coverage of the Variables Included in the Database Time Number of Number of Variable span countries observations Ratio of central bank assets to total financial assets 1960-97 79 2,177 Ratio of deposit money bank assets to total financial assets 1960-97 79 2,177 Ratio of other financial institutions' assets to total financial assets 1960-97 79 2,177 Ratio of deposit money bank assets to central bank and deposit money bank assets 1960-97 169 4,651 Ratio of liquid liabilities to GDP 1960-97 159 3,873 Ratio of central bank assets to GDP 1960-97 153 3,671 Ratio of deposit money bank assets to GDP 1960-97 160 3,912 Ratio of other financial institutions' assets to GDP 1960-97 80 2,008 Ratio of private credit by deposit money banks to GDP 1960-97 160 3,901 Ratio of private credit by deposit money banks and other financial institutions to GDP 1960-97 161 3,923 Net interest margin 1990-97 129 721 Overhead costs 1990-97 129 719 Concentration 1990-97 137 822 Foreign bank share (assets) 1990-97 111 673 Foreign bank share (number) 1990-97 111 673 Public share 1980-97 41 213 Ratio of total assets of bank-like institutions to GDP 1980-97 54 766 Ratio of total assets of life insurance companies to GDP 1980-97 24 333 Ratio of total assets of insurance companies to GDP 1980-97 40 547 Ratio of total assets of private pension and provident funds to GDP 1980-97 16 185 Ratio of total assets of pooled investment schemes to GDP 1980-97 27 295 Ratio of total assets of development banks to GDP 1980-97 46 634 Ratio of private credit by bank-like institutions to GDP 1980-97 43 652 Ratio of private credit by life insurance companies to GDP 1980-97 17 258 Ratio of private credit by insurance companies to GDP 1980-97 19 275 Ratio of private credit by private pension and provident funds to GDP 1980-97 11 126 Ratio of private credit by pooled investment schemes to GDP 1980-97 10 106 Ratio of private credit by development banks to GDP 1980-97 38 555 Life insurance penetration 1987-96 85 682 Life insurance density 1987-96 85 682 Ratio of stock market capitalization to GDP 1976-97 93 1,171 Ratio of stock market total value traded to GDP 1975-97 93 1,264 Stock market turnover ratio 1976-97 93 1,154 Ratio of private bond market capitalization to GDP 1990-97 37 287 Ratio of public bond market capitalization to GDP 1990-97 37 287 Ratio of equity issues to GDP 1980-95 42 586 Ratio of long-term private debt issues to GDP 1980-95 40 508 Source: Database was constructed by the authors from data in IMF (various years), International Finance Corporation (various years), and other sources. 600 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 These are institutions that serve as financial intermediaries, while not incurring liabilities usable as means of payment. We distinguish between two different balance sheet items: total claims on do- mestic nonfinancial sectors (lines a through d) and claims on the private sector (line d).6 In this article we denote total claims on domestic nonfinancial sectors as "assets" and claims on the private sector as "private credit." Assets refer to total domestic financial intermediation that the respective intermediary performs, and private credit captures financial intermediation with the private nonfinancial sec- tor. For both measures we exclude claims on central banks, deposit money banks, and other financial institutions (lines e through g)-and therefore any cross-claim that one financial sector has on another. Measures of the Size of Financial Intermediaries We construct relative size indicators that measure the importance of the three financial sectors relative to each other, and we construct absolute size indicators that measure size relative to GDP. The relative size measures are the ratio of central bank assets to total financial assets, the ratio of deposit money bank assets to total financial assets, and the ratio of other financial institutions' assets to total financial assets, where total financial assets are the sum of the assets of the central bank, deposit money banks, and other financial institutions. Since we calculate these measures only if data are available for all three categories, we construct an alternative indicator that measures the relative importance of deposit money banks relative to central banks: the ratio of deposit money bank assets to central bank and deposit money bank assets. Three indicators measure the size of the three financial sectors relative to GDP: the ratio of central bank assets to GDP, the ratio of deposit money bank assets to GDP, and the ratio of other financial institutions' assets to GDP. The assets include claims on the whole nonfinancial real sector, including government, public en- terprises, and the private sector. The sum of the three measures equals the total claims that financial intermediaries have on nonfinancial domestic sectors, relative to GDP, and thus constitutes a comprehensive measure of financial intermediation. Since many researchers have focused on the liability side of the balance sheet, we include a measure of absolute size based on liabilities: the ratio of liquid lia- bilities to GDP. Liquid liabilities equal currency plus demand and interest-bearing liabilities of banks and other financial intermediaries. This is the broadest avail- able indicator of financial intermediation, since it includes all three financial sec- tors. Liquid liabilities are a typical measure of financial depth, and thus of the overall size of the financial sector, that does not distinguish between the financial sectors or between the use of liabilities. 6. In the case of other financial institutions we also include line h-claims on real estate-in total claims on domestic nonfinancial sectors and in private credit. Beck, Demirgiiu-Kunt, and Levine 601 Measures of the Activity of Financial Intermediaries Two indicators focus on intermediary claims on the private sector: the ratio of private credit by deposit money banks to GDP and the ratio of private credit by deposit money banks and other financial institutions to GDP. Both measures iso- late credit issued to the private sector as opposed to credit issued to governments and public enterprises. Furthermore, they concentrate on credit issued by inter- mediaries other than the central bank. They measure one of the main activities of financial intermediaries: channeling savings to investors. II. EFFICIENCY AND MARKET STRUCTURE OF COMMERCIAL BANKS We collected data on the efficiency and market structure of commercial banks from individual banks' balance sheets, provided by Fitch IBCA's Bankscope data- base, and from individual country sources, such as central bank and supervisory body publications.7 We first present two efficiency measures of commercial banks. Then we define several indicators of market structure, in terms of concentration, foreign bank penetration, and public versus private ownership. Measures of Efficiency One of the main functions of financial intermediaries is to channel funds from savers to investors. We construct two measures of the efficiency with which com- mercial banks perform this function. The net interest margin equals the account- ing value of a bank's net interest revenue as a share of its total assets. The net interest margin also can be used as an indicator of the financial sector's competi- tive structure, although many factors may influence interest margins. Overhead costs equals the accounting value of a bank's overhead costs as a share of its total assets. We construct both measures as unweighted averages across all banks in a country for a given year. Measures of Market Structure The concentration of commercial banks equals the ratio of the three largest banks' assets to total banking sector assets. A highly concentrated commercial banking sector might reflect a lack of competition. We use two measures of for- eign bank penetration: foreign bank share (number) equals the number of for- eign banks in total banks, and foreign bank share (assets) equals the share of foreign bank assets in total banking sector assets.8 A bank is defined as foreign if at least 50 percent of the equity is owned by foreigners. 7. The classifications "commercial" and "deposit money banks" are close, but not exactly the same. Whereas the IMF database defines deposit money banks consistently across countries, Bankscope uses country-specific definitions of commercial banks. Unfortunately, Bankscope's coverage is less than 100 percent for most countries' banking sectors. This poses relatively few problems for the efficiency measures, but poses more for the measures of market structure, as discussed below. 8. There are important measurement problems with these shares. The Bankscope coverage is less than 100 percent. To the extent that foreign and large banks are more likely to be included in the Bankscope 602 THF WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Public versus private ownership has become an increasingly important issue for both researchers and policymakers, not only in the banking sector, but also in the whole economy (Demirgiiu-Kunt and Levine 1996). Our database compiles panel data on the public ownership of commercial banks. The measure public share equals the share of publicly owned commercial bank assets in total com- mercial bank assets. A bank is defined as public if at least 50 percent of the equity is held by the government or a public institution. Separately, La Porta, Lopez-de- Silanes, and Shleifer (2000) put together data on public ownership of commercial banks using detailed assessments from each country.9 III. OTHER FINANCIAL INSTITUTIONS This section presents measures of bank-like institutions, insurance companies, private pension and provident funds, pooled investment schemes, and develop- ment banks. We collected data from the IMF (various years), individual country sources (central banks, bank and insurance supervisory publications, statistical yearbooks), and SIGMA, a monthly publication from the reinsurance company Swiss Re. Categories of Other Financial Institutions Bank-like institutions comprise intermediaries that accept deposits without providing transferable deposit facilities and intermediaries that raise funds on the financial market mainly in the form of negotiable bonds. Examples of the first group are savings banks, cooperative banks, mortgage banks, and building societies. An example of the second group is finance companies. These institu- tions are often specialized in very specific activities, for historic, legal, or tax reasons.10 Within the category of insurance companies we can distinguish between life insurance companies and other insurance companies. We do not include insur- ance funds that are part of a government social security system. Like life insurance companies, pension and provident funds serve for pooling risk and accumulating wealth. We do not include pension funds that are part of a government social security system. Pooled investment schemes include financial institutions that invest on behalf of their shareholders in a certain type of asset, such as real estate investment schemes or mutual funds. database, then both foreign bank indicators and the concentration measure will be biased upward. This might be particularly relevant in developing countries. Also, a bank is defined as foreign if it was foreign in 1998. Thus takeovers of domestic banks by foreign banks are not taken into account. Given these problems, these indicators have to be used with caution in cross-country comparisons. 9. Their country coverage is broader, but their measure considers the largest 10 banks of a country and is only available for two points in time. 10. This definition is more restricted than the definition of other bank-like institutions given in the International Financial Statistics. Beck, Demirgiiu-Kunt, and Levine 603 And, lastly, development banks are financial institutions that derive their funds mainly from the government, other financial institutions, and supranational or- ganizations. On the asset side they are often concentrated on specific groups of borrowers. Most development banks were set up after World War II or after independence, in the case of developing countries, in an effort to foster economic development. Measures of the Size and Activity of Other Financial Institutions For all five groups constituting other financial institutions we construct mea- sures of their size relative to GDP by calculating the ratio of total assets to GDP. Unlike in section I, here total assets refer to total assets of balance sheets."1 We also construct activity indicators by measuring claims on the private sector rela- tive to GDP. For the insurance sector we include additional measures: assets and private credit of the life insurance sector, where disaggregated data are available; life insurance penetration, measured by the ratio of premiums to GDP; and life insurance density, measured by the ratio of premiums to population size. Life insurance penetration provides evidence on the importance of the life insurance sector relative to the total economy, and life insurance density provides evidence on the expenditure per capita on life insurance provision.'2 IV. STOCK AND BOND MARKET DEVELOPMENT The database includes measures of the primary and secondary stock and bond markets. By including bond markets and primary equity markets, this database significantly improves on previous work. Most of the data on the secondary stock market come from the International Finance Corporation's Emerging Market Database. Data on the secondary bond market are from the Bank for Interna- tional Settlements' Quarterly Review on International Banking and Financial Market Development. Data on the primary equity and debt markets come from country-specific sources, which were collected by Aylward and Glen (1998), and from the Organisation for Economic Co-operation and Development's (OECD) Financial Statistics Monthly. Indicators of Stock Market Size, Activity, and Efficiency The ratio of stock market capitalization to GDP indicates the size of the stock market relative to the size of the economy. Stock market capitalization equals the value of listed shares. 11. Using the total assets of balance sheets is problematic because they might include cross-claims within a category of other financial institutions and claims on other groups of financial intermediaries. A size measure, such as the measure in section I that includes only claims on the nonfinancial sector, is therefore preferable, but not available for most countries. 12. Life insurance density is constructed as premiums in local currency divided by the purchasing power parity conversion factor (obtained from World Bank 1997) and the population. To obtain the real density, we adjust these numbers by the annual consumer price index of the United States. 604 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 The ratio of stock market total value traded to GDP measures the trading vol- ume of the stock market as a share of national output and should reflect the degree of liquidity that stock markets provide to the economy. Total value traded equals the value of total shares traded on the stock market exchange. The stock market turnover ratio is the ratio of the value of total shares traded to market capitalization. It measures the activity or liquidity of a stock market rela- tive to its size. A small, but active, stock market will have a high turnover ratio, whereas a larger, but less liquid, stock market will have a low turnover ratio. Indicators of Bond Market Size As indicators of the size of the domestic bond market we use the ratios of private and public bond market capitalization to GDP. Bond market capitaliza- tion equals the total amount of outstanding domestic debt securities issued by private or public domestic entities. These two indicators thus measure the size of the market for public and private bonds relative to the real economy. Indicators of the Size of Primary Stock and Bond Markets For the indicator of size we use the ratio of equity issues to GDP for primary stock markets and the ratio of long-term private debt issues to GDP for bond markets. While the indicators of bond and stock market capitalization measure the outstanding publicly traded equity and debt, equity and long-term debt issues indicate the extent to which enterprises use these vehicles to raise external finance. V. CONCLUDING REMARKS This article introduced new indicators of the size, activity, and efficiency of financial intermediaries and markets across countries and over time. The data- base is part of a broader research project that tries to understand the determi- nants of financial development and its importance for economic development. These indicators can be used to investigate a wide range of financial issues (see Beck, Demirgu-Kunt, and Levine 2000; Beck and Levine 2000; Demirgiu-Kunt and Levine 1999; Demirguc,-Kunt and Maksimovic 2000; and Levine 2000). REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Aylward, Anthony, and Jack Glen. 1998. "Emerging Primary Markets." International Finance Corporation, Washington, D.C. Processed. Bank for International Settlements. Various years. Quarterly Review on International Banking and Financial Market Development. On-line at www.bis.org. Beck, Thorsten, AslI Demirguc-Kunt, and Ross Levine. 2000. "Politics, Law, and Fi- nance. " World Bank, Development Research Group, Washington, D.C. Processed. Beck, Thorsten, and Ross Levine. 2000. "New Firm Formation and Industry Growth: Does Having a Market- or Bank-Based System Matter?" Policy Research Working Paper 2383. World Bank, Development Research Group Washington, D.C. Processed. Beck, Demirgii-Kunt, and Levine 605 Beck, Thorsten, Ross Levine, and Norman Loayza. 2000. "Finance and the Sources of Growth." Journal of Financial Economics (forthcoming). Demirgii-Kunt, Aslh, and Ross Levine. 1996. "The Financial System and Public Enter- prise Reform: Concepts and Cases." In Niels Hermes and Robert Lensink, eds., Finan- cial Development and Economic Growth. London and New York: Routledge. 1999. "Bank-based and Market-based Financial Systems: Cross-Country Com- parisons." Policy Research Working Paper 2143. World Bank, Development Research Group, Washington, D.C. Processed. Demirgiiu-Kunt, Asli, and Vojislav Maksimovic. 1998. "Law, Finance, and Firm Growth." Journal of Finance 53(6):2107-37. . 2000. "Funding Growth in Bank-based and Market-based Financial Systems: Evidence from Firm-level Data." World Bank, Development Research Group, Wash- ington, D.C. Processed. IMF (International Monetary Fund). 1984. " A Guide to Money and Banking Statistics in International Financial Statistics." Washington, D.C. . Various years. International Financial Statistics. Washington, D.C. International Finance Corporation. Various years. Emerging Market Database. Wash- ington, D.C. King, Robert G., and Ross Levine. 1993a. "Finance and Growth: Schumpeter Might Be Right." Quarterly Journal of Economics 108(3):717-38. . 1993b. "Finance, Entrepreneurship, and Growth: Theory and Evidence." Jour- nal of Monetary Economics 32(3):513-42. La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer. 2000. "Government Ownership of Commercial Banks." NBER Working Paper 7620. National Bureau of Economic Research, Cambridge, Mass. Processed. Levine, Ross. 1997. "Financial Development and Economic Growth: Views and Agenda." Journal of Economic Literature 35(2):688-726. . 2000. "Bank-based or Market-based Financial Systems: Which Is Better?" Uni- versity of Minnesota, Department of Economics, Minneapolis. Processed. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. "Financial Intermediation and Growth: Causality and Causes." Journal of Monetary Economics 46(1):31-77. Levine, Ross, and Sara Zervos. 1998. "Stock Markets, Banks, and Economic Growth." American Economic Review 88(June):537-58. Neusser, Klaus, and Maurice Kugler. 1998. "Manufacturing Growth and Financial De- velopment: Evidence from OECD Countries." Review of Economics and Statistics 80(No- vember):636-46. OECD (Organisation for Economic Co-operation and Development). Various years. Fi- nancial Statistics Monthly. Paris. Rajan, Raghuram G., and Luigi Zingales. 1998. "Financial Dependence and Growth." American Economic Review 88(June):559-86. Rousseau, Peter L., and Paul Wachtel. 1998. "Financial Intermediation and Economic Performance: Historical Evidence from Five Industrial Countries." Journal of Money, Credit, and Banking 30(4):657-78. Swiss Re. 1998. SIGMA. Monthly. Zurich, Switzerland. World Bank. 1997. World Development Indicators 1997. Washington, D.C. Coming in the next issue of THE WORLD BANK ECONOMIC REVIEW Januiiary 2001 Voluime 15, Numniber 1 * The Impact of Early Childhood Nutrition Status on Cognitive Development: Does the Timing of Malnutrition Matter? Pautl Glewwe and Elizabeth M. King * Multi-tier Targeting of Social Assistance: The Role of Intergovernmental Transfers Harold Alderman * Does Ignoring Heterogeneity in Impacts Distort Project Appraisals? An Experiment for Irrigation in Vietnam Dominiquie van de Walle and Dileni Gunewardena * Circumstance and Choice: The Role of Initial Conditions and Policies in Transition Economies Martha de Melo, Cevdet Denizer, Alan Gelb, and Stoyan Tenev * Flight Capital as a Portfolio Choice Paul Collier, Anke Hoeffler, and Catherine Pattillo * The Mystery of the Vanishing Benefits: An Introduction to Impact Evaluation Martin Ravallion * New Tools in Comparative Political Economy: The Database of Political Institutions Thorsten Beck, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh The World Bank 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. Telephone: (202) 477-1234 Facsimile: (202) 477-6391 Telex: MCI 64145 WORLDBANK or MCI 248423 WORLDBANK World Wide Web: www.worldbank.org E-mail: feedback@worldbank.org ISBN 0-8213-4609-1