77334 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3: 509-44 Personal and Corporate Saving in South Africa Janine Aron and John Muellbauer Low domestic saving rates in South Africa may perpetuate a low-growth trap. The de- cline in government satfing, a major reason for the overall decline in saving, is now being reversed. However, personal saving rates have fallen since 1993, and corporate rates since 199S, and both may decline further with lower real interest rates. It is important to understand both personal and corporate saving behavior in order to formulate policies to raise the domestic saving rate in line with the needs of economic growth. This article summarizes previous work on the household sector, emphasizing the role of financial liberalization, assets, and income expectations, and explains sectoral links and policy implications. Further, it analyzes South Africa's corporate saving rate in detail. Models are developed both for the share of profits in national income, including the roles of the terms of trade, tax effects, and the price to unit labor cost ratio, and for the share of corporate saving in profits, which is found to depend on inflation, the real interest rate, dividend taxation, and financial liberalization. Corporate saving is remarkably underresearched, given its importance in many economies. This research thus puts the saving and growth concerns ofKaldor into a modern empirical context. Low domestic saving rates in South Africa may hinder investment-driven growth in the medium term. The ratio of gross domestic saving to gross national dispos- able income (GNDI) fell sharply from the early 1980s to the late 1990s, from an average of 24 percent in 1982-89 to 14 percent in 1998, and the net saving ratio (after depreciation) fell from 8 to 1 percent (see figure 1). These low gross saving rates rank poorly relative to rates in comparable emerging-market economies such as Chile. Low and falling saving rates imply increasing reliance on foreign capital inflows. Yet the sustainability of higher medium-term capital inflows is uncertain, given recent currency crises and increased risk premiums for emerging markets more generally. Real interest rates were very high in the 1990s, particu- Janine Aron is with the Centre for the Study of African Economies, Department of Economics, Univer- sity of Oxford, and John Muellbauer is with Nuffield College, University of Oxford. Their e-mail ad- dresses are janine.aTon%economici.oxMC.uk. and john.mueUbauer@economict.ox.ac.uk. Earlier versions of this article were given at the Econometric Society European Meeting, Berlin, August 29-September 2, 1998, and at the conference "Saving Across the World: Puzzles and Policies," September 16-18,1998, at the World Bank. The authors are grateful for comments and advice from Christopher Adam, Gavin Cameron, Alan Gelb, Brian Kantor, Jonathan Leapt, Cathy Pattillo, Jim Poterba, Klaus Schmidt-HebbeL, Luis Serven, Francis Teal, Lynne Thomas, Peter Worthington, and two anonymous referees. They acknowledge fund- ing support from the South Africa Country Team of the World Bank and the Economic and Social Re- search Council, United Kingdom (award number HJ2427003594 and grant number R000237500). Assis- tance from the South African Reserve Bank is gratefully acknowledged, particularly from Danie Meyer, Johan Prinsloo, Herman Smith, and Johan Van den Heever. O 2000 The International Bank for Reconstruction and Development /THE WORLD BANK 509 510 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Figure 1. Gross and Net Domestic Sewing Rates as a Percentage of Gross National Disposable Income, 1966-98 Percentage of GNDI 40 -r 30 - - -10 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 -e— Gross domestic saving —•— Net domestic saving -*— Net private saving - » - Net Government saving Source: Authors' calculations from the Quarterly Bulletin, South African Reserve Bank. larly after the currency crises of 1996-98, when real growth fell sharply (see figure 2). The implications of low saving are even more far-reaching considering the spillover effects from growth in South Africa to growth in its impoverished neighbors. Moreover, the AIDS crisis is Likely to exacerbate the pressure on saving rates in the next decade (see ING Barings 2000). The falling government saving rate since the early 1980s is largely responsible for the decline in South Africa's domestic saving rate (figure 1). The private sav- ing rate has been more stable, except for in the late 1970s, when gold prices rose, and in 1979-80, a period of very high gold prices from which the mining indus- try retained large windfall profits. Increased corporate saving sustained the level of private saving in the face of a decline in household saving, particularly in the 1980s (figure 3). Personal saving exceeded corporate saving until the mid-1970s, when corporate saving rose strongly, almost doubling as a ratio of national in- come by 1980-85. It remained high even after the 1979-80 gold price boom. Although government dissaving peaked in 1993, the government's improved saving performance, particularly creditable given weak growth rates, has been offset by a fall in private saving. This fall was driven by a decline in personal (or house- hold) saving, associated with a rapid rise in credit-financed consumer spending (figure 3). Corporate saving has also weakened since 1995. Aron and Muellbauer 511 Figure 2. Real Annual Growth and the Real Prime Interest Rate 0.05 -0.05 - - -0.1 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1969 1991 1993 1995 1997 - • - real annual GDP growth — e — real prime rate (4 q moving average) Source: Authors' calculations from the Quarterfy Bulletin, South African Reserve Bank. Figure 3. The Composition of Net Private Saving as a Percentage of Gross National Disposable Income, 1966-98 Percentage of GNM 20 T 15 - - 10 - • 1996 1998 —•— Net private saving —•*— Net corporate saving Net personal saving Source: Quarterfy Bulletin, South African Reserve Bank. S12 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 There is little published literature on saving behavior in South Africa, which is surprising given the importance of household consumption and corporate invest- ment in gross domestic product (GDP) and their role in the monetary policy trans- mission mechanism that runs from interest rate changes to output. Prinsloo (1994, 1997), Barr and Kantor (1994), and Tsikata (1998) have analyzed private sav- ing. These authors hypothesize that the offsetting patterns of personal and cor- porate saving referred to above had much to do with households "piercing the corporate veil." Thus they suggest that households rationally reduced their sav- ing to reflect the fact that when corporate saving increased in response to changed inflation and tax rates (with correspondingly reduced dividend payments), this raised the value of equities held directly by households or by pension funds on their behalf.1 If this rise in corporate saving were completely offset by the fall in household saving, due only to piercing the veil, the compositional change in pri- vate saving would be of little policy consequence. Indeed, these authors suggest that policies aimed at improving national saving should be focused firmly on raising the government's saving rate, rather than the private saving rate. But there are several reasons why the compositional change in private saving may not be innocuous and why the stability of private saving may be in doubt. First, several studies provide evidence suggesting that the piercing of the veil is incomplete. If households see through corporate budget constraints perfectly, dividend income should haye no direct effect in a consumption model after incor- porating the equity wealth effect. Poterba (1991) finds that the effect of dividend income is not zero in the Canada, United Kingdom, and United States. Second, even if households did pierce the veil, there is no guarantee that corporations would then behave in the way households desire. For example, a number of ar- ticles explain the large issue of dividends in the United Kingdom and the United States, despite unfavorable individual tax treatment, by corporate signaling or reputational devices (Bond, Chennells, and Devereux 1995 and Poterba 1987). Third, several other factors could change relative sectoral saving ratios, with differing degrees of offset and hence differing implications for the overall private saving rate. One long-term factor implying some offset between saving rates in the two sectors is financial liberalization, which has allowed South African house- holds to strongly raise their ratios of debt to income since the early 1980s. House- hold debt has its counterpart in corporate assets, particularly assets of financial corporations, which might help to account for the relatively high corporate sav- ing rate sustained into the 1990s. Similarly, incomplete offset could result from the response of the two sectors to changing terms of trade or to a shift in govern- ment saving behavior. Still other factors, such as lower real interest rates, might move corporate and household saving in the same direction. Far from being irrelevant to the question of raising national saving, therefore, the changing composition of private saving could have multiple causes and merit 1. See David and Scadding (1974) for an early discussion of piercing the veil or, in the authors' terms, the "ultrararionality* of households. Aron and Muellbauer S13 specific policy responses. Thus a rigorous empirical examination of the factors driving household saving and the factors driving corporate saving is warranted. In this article we explore the determinants of private saving in South Africa, separately examining personal and corporate saving behavior from the late 1960s to 1997. In each case we subject the corporate veil hypothesis to more rigorous empirical testing than has been conducted thus far. We examine the observed decline in personal saving in South Africa, drawing on Aron and Muellbauer (2000a), which estimates a quarterly consumption func- tion for South African households. Prior to this study, the most comprehensive consumption function came from the model of the South African Reserve Bank, which separately models four components of consumption: durables, nondurables, semidurables, and services (Pretorius and Knox 1995). However, this model ex- cludes relative prices, assets, debt, proxies for expectations, and measures of fi- nancial liberalization. These are important omissions. Because of fluctuations in asset prices and changes in financial liberalization, omitting these variables can lead to huge forecasting errors (as in the United Kingdom in the 1980s; see Muellbauer and Murphy 1995). It is not surprising that the South African Re- serve Bank omitted asset stock variables, since it does not construct these mea- sures. However, the omission makes it impossible to test rigorously whether house- holds pierce the corporate veil. Aron and Muellbauer (2000a) remedy some of these problems. Income fore- casts are generated from a separate income-forecasting model; an innovative in- dicator for financial liberalization is derived, which is closely linked to institu- tional changes; and wealth estimates are constructed, given the absence of official data (details are given in Muellbauer and Aron 1999). The general issue of corporate saving is highly underresearched, and not only in developing countries. It receives less than one page in Deaton's (1999) com- prehensive survey. There is no generally agreed theory of corporate finance to parallel the intertemporal models of households. Two of the few empirical analy- ses of aggregate corporate retention rates are by Poterba (1987,1991). Our analysis extends his treatment by articulating several of the underlying economic mecha- nisms. Moreover, because the share of national income that is corporate profits can be quite variable, unlike Poterba, we model the profit share separately in order to derive corporate saving as a share of national income. We therefore incorporate one of the neglected key insights of Kaldor's (1957) growth model. Kaldor argues that since the propensity to save out of corporate profits is higher than the propensity to save out of labor income, the private sector saving rate out of national income depends on the share of profits in national income. However, unlike Kaldor, we endogenize the sectoral saving propensities and address intersectoral links. There is a large literature on the effects of financial deregulation on saving, investment, and growth (see the survey by Fry 1995: ch. 8). Our analysis of the overall effects of financial liberalization on private saving in South Africa con- tributes to this literature, although it should be noted that South Africa has long 514 THE •WOULD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 had a sophisticated financial system and has never experienced financial repres- sion on the scale seen in many developing countries. Finally, our research throws light on the monetary transmission mechanism in South Africa, highlighting some of the policy dilemmas faced by the South Afri- can Reserve Bank. I. DATA ISSUES AND REGIME SHIFTS South Africa is a dualistic economy, with a highly unequal income distribu- tion. Its income-based Gini coefficient is 0.61, calculated using 1993 household survey data, making it second in inequality only to Brazil among middle-income countries (Klasen 1997). Gini measures of the inequality of earnings among ur- ban employed men are about 0.5 for both 1980 and 1993, suggesting little de- cline in inequality, although racial discrimination lessened (Moll 1998). With such unequal income and expenditure distributions, the bulk of aggre- gate saving will be accounted for by the most affluent households.2 In this article we thus model aggregate behavior focusing on the formal part of the economy, which has the dominant economic weight. Doing so neglects welfare issues con- nected with the saving of poorer households, important in its own right, but which the national aggregates cannot address.3 Soudi Africa has a large urban unemployed class, in contrast with many coun- tries in Sub-Saharan Africa. Recent research based on different measures of un- employment in South Africa suggests, using a broad definition of unemployment, a range from 30 to 40 percent (Kingdon and Knight 2000). Clearly, it is difficult to measure the unemployment rate in a meaningful sense. High unemployment, frequently used as a measure of income uncertainty in countries with lower un- employment, is unlikely to have much of an influence on the saving behavior in the formal part of the economy and hence on aggregate saving behavior in South Africa (controlling for income, interest rates, and other variables). In South Africa's highly segmented labor market, it makes sense instead to proxy income uncer- tainty with formal sector indicators of labor market tightness, such as capacity utilization. The South African economy has suffered many political shocks and discontinuities, as well as terms of trade shocks. International trade and financial sanctions, foreign disinvestment, capital flight and emigration during the apart- heid era, and their subsequent reversal (to some extent) during the 1990s have greatly influenced macroeconomic outcomes and policy. There also have been important changes in domestic regimes, notably in the exchange rate and mon- etary policy regimes, and in financial regulation. 2. Bated on a comprehensive socioeconomic survey conducted in 1993, the upper 30 percent of house- holds in South Africa in the distribution of household expenditures, making up 21 percent of the population, accounted for 75 percent of the country's expenditures (World Bank research quoted in RDP Office 1995). 3. With data revisions in the early 1990s, the South African Reserve Bank attempted to incorporate more of die informal sector and service economy into GDP data. Revisions since June 1999 (after the period modeled in this article) have continued this process. Such shifts have required creative modeling of regime breaks. For instance, we employ stochastic trends in income expectations models, which help to capture changes in income growth trends linked to political events. Further, using detailed institutional changes, we construct indicators for crucial shifts in the monetary policy regime in the 1980s and for financial liberalization in the 1980s and 1990s. Shifts in monetary policy play an important role in our income-forecasting models in capturing shifts in the influence of interest rates. Financial liberalization has important effects in our household and corporate saving models. The main impact of financial liberalization in South Africa's already sophisticated financial sector, through more competitive markets, was on saving and spending in the formal sec- tor, rather than on bringing financial services to poor black households. The real interest rate was largely positive during 1960-98 (see figure 2), and the main con- sequence of financial liberalization has been more easily obtained credit and more easily collateralized, and thus more spendable, illiquid assets. However, under the apartheid system in South Africa, rural and urban black households had highly constrained access to formal saving and credit opportunities. With incomplete or less than comprehensive data, we construct data, where possible; but in some instances the lack of data constrains the possible analysis. For example, in the absence of published data on stocks of personal assets (mostly available only on a flow basis and often at book rather than market value), we construct the first series for a range of personal assets at market value (see Muellbauer and Aron 1999). These wealth effects prove very important in personal saving. We are, however, unable to construct asset series for unincorporated busi- nesses and directly held foreign assets. Our efforts to capture the effects of in- come distribution using ratios of black formal sector wages to white formal sec- tor wages failed, perhaps because of die weaknesses of these data, especially in the 1990s. Time-series data on the age distribution are not available, although it would have been useful to test for demographic effects by age of formal sector savers. Similarly, the figures on international migration are poor. But to the ex- tent that recent emigrants have been among the higher savers, their leaving may have reduced the saving ratio and possibly also wealth holdings in South Africa. n. MODELING PERSONAL SAVING The approach we adopt is to split the ratio of personal saving to national income into two ratios: the ratio of personal saving (S) to personal disposable income (PDI) and the ratio oiPDI to gross national disposable income (GND7).4 Thus (1) S/GNDI = 5 • (PDI/GNDI) where s is the personal saving ratio S/PDI. 4. Personal saving in the national accounts is conventionally defined as personal disposable income minus consumer expenditures. Consumer expenditures include expenditures on consumer durables, but not on the acquisition of owner-occupied housing, which is treated as part of saving. 516 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Figure 4. Afe/ Personal Saving Relative to Personal Disposable and Gross National Disposable Income, 1966-98 PDI/GNDI Saving ratio 0.72 -i r0.12 • 0.10 -0.08 -0.06 --0.04 • -0.02 0.40 \ 1,1,1 I I. M I I I I I I I. I. I I I I I J I I I I I I I J I J.J'0.00 - B — N « personal saving/GNDl —•— Net personal saving/PDl PDI/GNDI Source: Authors' calculations from the Quarterly Bulletin, South African Reserve Bank. The share of PDI in GNDI is inversely correlated with me share of profits, given that labor income is a large component of PDI (figure 4). We do not model this ratio formally, but it also depends on personal tax rates, income transferred to individuals from the government, and self-employment and personal property income. We can now link the saving ratio, 5, to the consumption function. Since 5 is a small number, it can be approximated as (2) s » - l o g ( l - s ) = -log(aPD7) where C is consumer expenditures in current prices. The models we discuss for the log of consumption can therefore be translated easily into models of the ratio of personal saving to gross national disposable income. Theoretical Issues The canonical rational expectations permanent-income hypothesis, as derived by Hall (1978), has both a Euler and a "solved-out" representation. Under a number of simplifying assumptions,5 Hall derived a martingale property for the intertemporal efficiency condition on consumption, or the Eulcr equation: (3) ct = c,_i + e, 5. These include the absence of credit restrictions, quadratic utility and certainty equivalence, rational expectations, intertemporally additive preferences implying the absence of habits, a constant real interest rate equal to the rate of subjective time preference, and behavior equal to that of an infinitery lived house- hold. Much research since Hall (1978) has been concerned with relaxing such assumptions. Aron and Muellbauer 517 where c, is consumption measured in constant prices and e, is a stochastic vari- able, not predictable from information available in f - 1, which captures news about permanent income. Equation 3 embodies the extreme consumption-smooth- ing implication of the rational expectations permanent-income hypothesis, since at t - 1, the consumer plans future consumption levels to be the same as the current level. Solving this efficiency condition and its equivalents for all future periods gives the solved-out form of the consumption function: (4) ct = rA^ + Etft where -ft is permanent nonproperty income (consisting of employment income after taxes, and transfers from the government), r is the real rate of return, and A M is the stock of real assets at the end of the previous period.6 To make the solved-out form empirically useful requires at least one more equation, namely an income-forecasting model, which relates permanent nonproperty income to observable variables. The solved-out model has a num- ber of advantages relative to the more fashionable Euler equation. First, it is directly relevant for policy analysis—for instance, to analyze the effects of a tax reform (using the income forecasting model), which could alter the profile of future household income. Second, the solved-out consumption function does not throw away important long-run information from the consumption, income, and asset data (see the cointegration literature, such as Davidson and others 1978, Engle and Granger 1987, and Banerjee and others 1993). The Euler approach does not use asset data, while differencing makes consumption and income stationary, where they are typically nonstationary. Third, although the solved-out approach is clearly more demanding in terms of the information and modeling effort required, the approximations needed to obtain policy-relevant consumption functions (described in the next section) are no more extreme than those popularly made in the Euler equation approach—by, for example, Hall and Mishkin (1982) and Campbell and Mankiw (1989, 1991)—to incorporate credit constraints or myopia. Derivation of a Basic Solved-out Consumption Function We set out an empirical methodology for estimating the consumption func- tion, drawing on the theoretical literature, but, which is important, introducing income forecasts, estimates of wealth—distinguishing between liquid and illiquid assets—and an indicator of financial liberalization (for more details see Aron and Muellbauer 2000a). At the micro-level, a solved-out consumption function is the solution to an intertemporal utility maximization problem, the classic case being the canonical rational expectations permanent-income hypothesis in equation 4. A log-linearization of equation 4 makes log(c/y) a linear function of the asset-to- 6. Nonproperty income is the theoretically relevant measure of income and excludes interest income, rents, and returns on equities and stocks. 518 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 income ratio (AJy) and of log {"/ly), where we can proxy for yf/y with forecasts of the income growth rate. Introducing habits or adjustment costs implies a par- tial adjustment model. Extending the model from point to probabilistic income expectations introduces a measure of income uncertainty, 0fi as well as a measure of expected income growth, E, Alog yTfi, which is a weighted moving average of forward-looking growth rates. Further, if real interest rates are variable, theory suggests that the real interest rate rt will enter the model. Incorporating these three additional variables, a simple linearization gives the following generalization of the canonical rational expecta- tions permanent-income hypothesis model of equation 4: (5) Alogc,« P[Oo - as, - a 2 9 r + logy, + c c ^ ^ o g y 1 " 1 + yA,Jyt - logc^] + e, where |$ measures the speed of adjustment. Making the same assumptions as Campbell and Mankiw (1989,1991) to incorporate credit-constrained behavior by households can be shown to result in an equation for aggregate consumption that is a generalization of equation 5, containing an additional income growth term, Alogjv Although consumption theory puts great weight on income expectations, ex- plicit empirical models of the income-forecasting process are rare in consump- tion modeling. The Lucas critique (Lucas 1976, 1981) argues that conventional reduced-form econometric models break down when policy rules change since agents' expectations-generating functions change. He concludes that the models are of little use for policy analysis. One solution is to address the critique directly, by building an income-forecasting model that recognizes the importance of policy feedback rules and is sensitive to possible shifts in these rules. The aggregation of assets into a single quantity, A, in equation 5 is an oversim- plification. We argue that wealth effects differ according to the liquidity charac- teristics of different types of wealth. Households usually hold a balance of as- sets—liquid assets, which can be converted easily into expenditures when needed, and illiquid assets, which typically yield higher rates of return. This suggests that we should associate different weights reflecting different propensities to spend with different types of assets and debt.7 Housing, pension funds, and life insurance funds are at the illiquid end of the spectrum.8 Pension wealth is likely to have a delayed impact on consumption. Contractual saving contribution rates often respond with considerable lags to changes in the asset values of such pension funds, suggesting that we should test for longer lags on consumption. 7. Several studies, such as Patterson (1984), allow different weights on liquid and illiquid assets, whereas others, such as Zellner, Huang, and Chau (1965) and Hendry and von Ungern Steinberg (1981), include the effects of liquid assets alone. 8. Housing wealth is a special case because housing has consumption as well as wealth value (housing services also appear in the utility function). Thus an increase in the real price of housing has both an income and a substitution effect on consumption, partly offsetting the wealth effect. See Miles (1994) and, for a simple derivation, Muellbauer and Lattimore (1995). Aron and MueUbauer S19 Another factor, financial liberalization, theoretically has consequences for the liquidity characteristics of different assets and might be expected to affect con- sumption directly and indirectly. Although the implications of financial liberal- ization have aroused interest, controversy, and a growing literature (such as Bayoumi 1993a, 1993b; Schmidt-Hebbel and Serven 1997; Bandiera and others 2000; and Honohan 1999), the applied analysis of these implications in the con- sumption literature has not been entirely satisfactory. The difficulty has been to find an indicator of credit market deregulation with which to model the direct and interaction effects of financial liberalization. These effects are as follows. First, financial deregulation is likely to raise the propensity to spend directly by reducing the required deposit needed to buy a house (see Jappelli and Pagano 1994 and Deaton 1999: 53-54). Second, finan- cial deregulation is likely to increase the "spendability" weights on illiquid as- sets, since both housing collateral and, to a lesser extent, pension collateral could be used for borrowing and hence for spending. Third, financial liberalization should reduce the proportion of credit-constrained households. With the increase in households that are not credit-constrained, the real interest rate and expected real income growth will have a greater effect on aggregate consumption. How- ever, the effect of uncertainty on consumption may weaken if consumers antici- pate that they can borrow more easily in the event of a negative income shock. Finally, financial liberalization and income distribution may interact as access to credit is extended at lower income levels. Income Expectations In Aron and Muellbauer (2000a) we generate forecasts of personal disposable nonproperty income from an income-forecasting model that includes interest rates, capacity utilization, the government budget deficit, and the Johannesburg Stock Exchange (JSE) share price index expressed in real terms. The model accounts for important regime shifts: it is estimated using a stochastic trend based on Koopman and others (1995), thereby capturing the decline in the growth rate in the 1980s as trade and financial sanctions and political difficulties increased. Using dummy indicators constructed by altering statutory liquidity ratios for commercial banks in the 1980s, the model also captures the changing sensitivity of income growth to interest rates as the monetary policy regime changes. By incorporating impor- tant regime shifts in the model, the resulting consumption function should be fairly immune to the Lucas critique. The model suggests that government deficits have persistent negative effects on subsequent income growth. These effects could reflect typical concerns for budget deficits followed by higher taxes or lower government expenditures, but these deficits may also signal political shocks. In the past political unrest was often followed by higher social or military expenditures, which thus may serve as a proxy for a direct negative effect on growth through falling investment. This generates a link between personal and government saving rates, which was miss- ing in previous work. 520 THE WORLD BANK ECONOMIC REVIEW, VQL. 14, NO. 3 Nominal rises in interest rates and levels of real rates have strong negative effects on subsequent growth. The shift toward more market-oriented monetary policy in the 1980s appears to have weakened the influence of changes in nomi- nal rates. Before the shift, high liquidity ratios and other quantitative methods of controlling credit growth were correlated with changes in nominal rates, exag- gerating the apparent influence of interest rates on growth. After the shift, firms and households could refinance more easily, so that higher interest rates had a weaker effect on expenditures. However, although the coefficients on interest rates are lower, the greater volatility of interest rates in the market regime means that the proportion of the variance of growth explained by interest rates remains high. Changes in capacity utilization, proxying changes in labor market tight- ness, have the expected positive effects on nonproperty income. Finally, the JSE index, sensitive to changes in the price of gold and other minerals, captures the positive effect that improving terms of trade have on income. Wealth Effects The estimates of illiquid and liquid personal wealth used in the consumption model were constructed in Muellbauer and Aron (1999); see figure 5. The house- Figure 5. Debt and Liquid and Illiquid Assets Relative to Personal Disposable Nonproperty Income, 1970-97 Assewo- income ratio 2-5 T 0.0 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 Housing weahh/PDNi Directly held securitiea/PDNi Debc/PDNI Penskms/PDNi Liquid assets/PDNl Source: Muellbauer and Aron (1999). Aron and Muellbauer 521 hold liquid assets ratio seems to have been relatively stable in the 1970s. In the 1980s, however, there was a large fall in households' holdings of liquid assets relative to nonproperty income. This coincided with both a drop in the personal saving ratio, as implied by the income and expenditure accounts, and a switch to saving in pension and retirement funds offering superior returns to those on liq- uid assets. Pension wealth has grown relative to income since the 1980s, despite the fall in the personal saving ratio, and has greatly exceeded the growth of debt.' Yet although pension wealth is now the single biggest asset, its growth has been off- set to a considerable degree by the decline of housing wealth relative to income. The wealth measures in our model will be understated because we omit foreign assets accumulated illegally during the era of exchange controls on domestic resi- dents. Similarly, our measures of household assets do not include assets of unin- corporated businesses, except for their inclusion in private housing stocks. For further discussion, see Muellbauer and Aron (1999). Financial Liberalization The government initiated financial liberalization following the de Kock Com- mission reports (1978, 1985), which advocated a more market-oriented mon- etary policy. Interest and credit controls were removed in 1980, and banks' li- quidity ratios were reduced substantially between 1983 and 1985. Competition rose in the mortgage market following the 1986 Building Societies Act and amend- ments to it in 1987-88. Demutualization and takeovers in 1989-90 consolidated the stronger competition in the credit market. In the 1990s pensions were in- creasingly used to provide additional collateral for housing loans, while begin- ning in 1995 special mortgage accounts (access bond accounts) allowed house- holds to borrow and pay back flexibly from these accounts up to an agreed limit set by the value of their housing collateral. After the 1994 elections more black South Africans obtained formal employment, particularly in the public sector, gaining access to credit that they previously would have been denied.10 Exchange controls on nonresidents were eliminated in early 1995: large nonresident capital inflows from mid-1994 induced a temporary endogenous financial liberaliza- tion. Finally, exchange controls on domestic residents, in existence since before the 1960s, were partially relaxed after 1997. Financial liberalization had a major impact on the ratio of consumption to income, but it had an even larger effect on the ratio of debt to income.11 An 9. Three of the factors behind this growth are die relaxation of restrictions on official pension funds (that is, for government employees), which had prevented their holding of equities (Mouton Report 1992), the improvement in the returns on government and parastatal bonds with the deregulation of interest rates after 1980 and the decline of inflation in the 1990s, and relaxation of prescribed holdings of government bonds for all pension funds. 10. However, total formal employment continued to decline. 11. Bayoumi (1993a, 1993b) and Sarao and Taylor (1998) use the ratio of debt to income as an indicator of finanrijil liberalization, although the ratio also depends on factors such as income, asset levels, and interest rates. 522 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Figure 6. Household Debt to Personal Disposable Nonproperty Income and Financial Liberalization, 1970-97 Financial liberalization (PUB) 0.7 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 m Log (household debt/PDNi) • Financial liberalization (PUB) Source: Aron and Muellbauer (2000a). innovation in Aron and Muellbauer (2000a) is to treat financial liberalization as an unobservable indicator entering both household debt and consumption equa- tions. The indicator, FLIB, is proxied by a linear spline function, and the param- eters of this function are estimated jointly with the consumption and debt equa- tions (subject to cross-equation restrictions on the coefficients in the spline function). The estimated parameters for FLIB in the model reflect the key institu- tional changes in credit markets. Our estimated indicator shows strong rises in 1984,1988, and 1995, with more moderate increases in 1989, 1990, and 1996 (figure 6). It is noteworthy that both the consumption function and debt equa- tion are subject to major structural breaks (failing Chow tests) when allowance is not made for financial liberalization. In the consumption equation the most important role of FLIB is its direct effect on the average propensity to consume, in part related to the lower mort- gage deposits that banks require of home buyers.12 Further, a proxy for in- come uncertainty has a weaker effect on consumption when interacted with FLIB (that is, the precautionary role of income uncertainty declines with financial liberalization). 12. We also interact FLIB with housing assets, expecting to find that housing wealth is effectively more spendable with liberalization. However, the coefficient is hard to pin down accurately, probably because the ratio of housing assets to income has been so trend-like since 1983 (see figure 5). Aron and Muellbauer S23 In the debt equation the direct effect oiFLIB on the ratio of debt to income is significant. But this effect is dominated by the interaction effects of FLIB with asset-to-income ratios: debt increases as the collateral weight of housing assets rises with liberalization.13 However, there is a small offset: the size of liquid as- sets has less effect on the volume of debt with financial liberalization, probably because mortgage lenders are less constrained by the personal sector's liquid deposits. Personal Satnng As shown in equation 2, the personal saving ratio is close to log(C/PDZ). Since our consumption function is formulated in terms of log( C/PDNJ), where PDNI is personal disposable nonproperty income, we use the following conversion to get from our consumption function to the personal saving rate: (6) s = -log(C/PDNI)-log(PDNJ/PDJ). Our quarterly consumption model applies the extended version of equation 5 for 1970-97, using our own income forecasts, our own liquid and illiquid asset variables, and our own indicator of financial liberalization (Aron and Muellbauer 2000a). We summarize first the long-run results from this model and then dis- cuss three further implications for saving from the full model. The long-run solu- tion—in terms of nonstationary variables integrated of order 1 [1(1)]—is: (7) log(aPDNJ) = constant + 0.33 FLIB + 0.073 log(PD//PDN7) + 0.147 RLADB + 0.068 (RHA + RJFA + RPAMA) - 0.25 RPRIMEMA. Although self-employment income is part of the theoretical definition of nonproperty income, these data are not separately available in the national ac- counts. We correct for this by including log(PDJ/PDNI) as a proxy for self- employment income.14 The asset variables RLADB, RHA, and RIFA are, respec- tively, the ratio of liquid assets minus debt to current PDNI, the ratio of gross housing assets to current PDNI, and the ratio of directly held, illiquid financial assets (at the end of the previous quarter) to current PDNI. RPAMA measures pension and insurance company assets owned by the personal sector, expressed as the moving average of the previous four quarters, again relative to current PDNI. The longer lag reflects the delays by which pension and insurance funds adjust contribution rates to changing asset values. RPRIMEMA is the four- 13. In most other countries housing booms have followed significant financial liberalization, thereby compounding the rise in debt. In South Africa, however, the declining personal saving ratio and rising debt-to-income ratio due to finam-ial liberalization were accompanied by a falling ratio of house prices to income after the gold boom of the early 1980s. Possible reasons include poor growth, high real interest rates, and emigration. 14. Note that log(PDI/PDNJ) = log(l + property income/PDNT) - property income/PDN/. We assume that self-employment income (a major component of property income in other countries) is highly corre- lated with property income in South Africa. 524 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 quarter moving average of the real interest rate using the prime rate charged by banks. 15 Following equation 6 the long-run solution for the personal saving ratio is then derived as (8) s = constant - 0.33 FLIB + 0.927 log(PDI/PDNI) - 0.147 RLADB -0.068 (RHA + RIFA + RPAMA) + 0.25 RPRIMEMA. The long-run saving ratio emphasizes the role of wealth effects and financial liberalization, factors typically neglected in conventional analyses. There are two wealth effects. Liquid assets minus debt appear to be about twice as spendable as illiquid assets, the components of which have similar degrees of spendability.1* We illustrate equation 8 by plotting 5 against all five explanatory components weighted by their parameter estimates (figure 7). This clarifies longer-term trends and shows how short-run movements in s are dominated by the volatility of property income (reflected in PDI/PDNI). Our model emphasizes the important and complex role of financial liberaliza- tion. Financial liberalization has a huge effect on the personal saving ratio (figure 7). Our model suggests that between 1983 and 1997, financial liberalization di- rectly reduced the personal saving ratio by around 21 percent (for given PDll PDNI, ratios of assets to income, and real interest rates). Since the average ratio of PDI to gross national disposable income between 1983 and 1997 was 0.62 (see figure 4), this would imply a fall of 12.5 percentage points in the ratio of personal saving to gross national disposable income. However, important offsetting factors cushioned the fall in the saving ratios (figure 7). Some of these factors were influenced by financial liberalization itself: implying that our partial equilibrium calculation overstates the general equilib- rium effects. The full general equilibrium picture of the effect of financial liberal- ization on 5 and on S/GDNI includes complex, indirect effects working through interest rates, income ratios, and portfolios of assets and debt. More comprehen- sive models are needed to quantify these effects. The largest offsetting factor in figure 7 is the rise in PDI/PDNI, where increased property income and self- employment income in the unincorporated business sector, relative to formal labor income, increased saving. One of the factors likely to have contributed to the relative rise of PDI is increased access to credit, because of the stimulus to unincorporated enterprise. The fall in liquid assets minus debt is the second Largest offsetting factor (fig- ure 7). This trend was strongly influenced by financial liberalization, which al- lowed households to hold more debt and fewer liquid assets relative to income (by the late 1980s household debt exceeded liquid assets). Such a deterioration in households' net liquid asset position tends to increase saving. 15. The real interest rate is a borderline 1(1) variable on annual data, although the quarterly moving average appears to be 1(0). We therefore prefer to include it in the long-run solution. 16. Illiquid assets comprise housing assets, directly held illiquid finjinrial assets, and pension assets. Aron and Muellbauer 52S Figure 7. Decomposition of the Personal Saving Ratio into Asset-to-income and Relative Income Effects, Financial Liberalization, and Real Interest Rate Effects, 1970-97 Personal saving ratio 0.4-r 0.3-" -0.1-- -0.2-- -0.3 2 1984 1986 1988 1990 1992 1 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 " Personal saving/PDl - a - (liquid assets - debO/PDNi Composite illiquid assets/PDNl -"— Log (PDl/PDNi) —»— Financial liberalization Real interest rate Source: Equation 8 and Aron and Muellbauer (2000a). Financial liberalization also indirectly influenced the rise in real interest rates. With the reduction of credit rationing from the early 1980s, high interest rates were required to control credit demand. We find that a 1 percent rise in the real prime interest rate directly increases the personal saving ratio by 0.25 percent (see figure 7). Further, increased real interest rates undoubtedly helped to prevent a financial liberalization-induced boom in asset prices, which would have led to a deterioration in the saving ratio. The long-run solution also lends insight into how monetary policy is transmit- ted to the personal sector. Our research emphasizes multiple transmission chan- nels: probably even larger than the direct effect shown in equation 7 is the indi- rect effect working through asset prices. Theoretical considerations as well as empirical results for a range of other countries emphasize the relative importance of such indirect effects.17 An important future avenue for research is to quantify these asset price effects for South Africa. 17. For a recent symposium on monetary policy emphasizing the asset price channel, see Federal Re- serve Bank of Kansas (1999). 526 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Three key implications for saving from the full model in Aron and Muellbauer (2000a) concern fiscal policy effects, saving and growth, and piercing the veil. The inclusion of expected income growth in the consumption model clarifies the channel through which fiscal policy is transmitted to personal saving. Our income- forecasting model suggests that a 1 percentage point reduction in the ratio of the government deficit to GDP raises personal disposable nonproperty income by around 1 percent in the long run. Since income expectations and current con- sumption adjust before actual current income, this would result in a short-term decline in the ratio of personal saving to gross national disposable income. In the longer term, the effect of the deficit on saving depends on how it influences the ratio of personal income to national income. The estimated model also suggests that, at given asset-to-income ratios and real interest rates, an increase in the steady-state growth rate of income has little effect on the personal saving rate. Habit formation in consumption tends to generate a positive effect of growth on saving, while if households have high growth expectations, this reduces sav- ing (see Deaton 1999: 61 for a summary of the issues on the effect of growth on steady-state personal saving). In our model the two effects tend to cancel one another. Finally, we followed Poterba (1991) in testing whether dividend income has no direct effect on consumption after incorporating the equity wealth effect, as is implied if the corporate veil is pierced. Dividend income is insignificant, suggest- ing that households perceived the tradeoff between increased dividends and lower equity prices to be almost complete and thus did not increase consumption. This result, however, is subject to possible measurement error in the proxy for after- tax dividends.18 HI. CORPORATE SAVING AND PROFITS Since corporations now dominate private saving in South Africa it is impor- tant to understand the determinants of the corporate saving rate. In this article we examine the determinants of annual net corporate saving (relative to GNDI) from 1966 to 1997, including the role of dividend and other tax changes. The net corporate saving ratio can increase through both a rise in corporate saving rela- tive to profits and a rise in the profit share in national income: (9) CS/GNDI = (CS/CP"") (CP^/GNDI) where CS is net corporate saving, and CP*" is corporate income (profits) net of tax and interest payments. There are multiple influences on each of the ratios on the right side of equation 9. In order not to conflate these influences, we model 18. Unlike in the United Kingdom and the United States, the national accounts in South Africa do not provide a separate dividend income measure. We use the JSE dividend-to-price ratio, adjusted for taxes and multiplied by estimated equity holdings. Further, the self-employment income proxy in equation 7 unfor- tunately also includes a small dividend component. In the consumption equation this term has a low but significant coefficient, and hence dividend income may exert some effect on consumption. Aron and MueUbauer S27 Figure 8. The Corporate Saving Rate, Shares of Net and Gross Profits in National Income, and the Ratio of Net to Gross Profits, 1966-98 Log scale 0.0 x -0.2-• -0.4-- -0.6 -0.8 -1.0-• -1.2 - - -1.4 M l II 1966 1968 1970 197:2 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 —•— CS (neO/CP (net) -*— CP (net)/GNDi - s - CP (neO/CP (gross) —»— CP (gross)/GNE« Source: Authors' calculations from the Quarterly Bulletin, South African Reserve Bank. the two ratios separately. Further, we break the second ratio, CP^/GNDI, into (CPH"/CP*wuHCP'ro"/GNDr), and model CP^°"/GNDI, thereby separating out tax effects from the share of gross profits in income (see the section on the corpo- rate profits equation). This type of disaggregation extends the work of Poterba (1987), who examines only saving or dividend payments out of net corporate profits. The log of the ratio of corporate saving to net profits rises until the late 1970s (figure 8). It then fluctuates around the higher level until the early 1990s, when a declining trend appears. The ratio of net to gross profits, which largely reflects the ratio of company taxes to profits, shows little trend, but again a tendency to decline since the early 1990s. The upturn of the share of gross profits in national income in the 1990s has been more than offset in recent years by higher taxation; thus net profits to GNDI have fallen since 1996. Corporate Saving Equation The literature on corporate saving is largely policy-oriented and empirical, but it generally lacks an agreed theoretical model.19 Poterba (1987) examines the mirror image of corporate saving, namely corporate dividend or payout equa- tions (see also earlier work by Feldstein 1970, 1973 and King 1977), and sum- 19. The only one is an econometric study for South Africa by Tsikata (1998). 52 8 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 marizes some of the alternative theories and the empirical literature for the United States and the United Kingdom.20 In the absence of an accepted theoretical model on dividend behavior, Poterba argues for an error-correction model of aggregate real dividends, in which the target dividend is a function of equity earnings (using several different measures of earnings) and a tax discrimination variable, mea- sured as a weighted average across shareholders of the after-tax household in- come from a dividend payout, divided by the after-tax income associated with undistributed profits (with associated capital gains). Such models focus on the link between net profits and the demand for payouts by shareholders (reflecting their tax position). But, as Poterba acknowledges, another influence on dividend behavior could come from the demand for reten- tions induced by corporate investment. If retentions offer a source of funding expected to be cheaper than equity or debt issues or bank borrowing, corpora- tions may reduce dividend payments if they anticipate a higher cost of outside funds or greater profit-earning opportunities (see also Caggese 1999 for a recent formal model of this type of saving motive). Poterba (1987) attempts to control for this influence by including in his dividend equation the effective tax rate on corporate investment and Tobin's q, but he finds neither to be significant. In his results for the United States the tax discrimination variable hovers around the 5 percent significance level over different samples and specifications. In the United States, therefore, the relative tax burden on dividends and capital gains does appear to affect, but only weakly, the extent to which corporate earnings are distributed to shareholders. Our measure of net corporate saving covers saving by South African financial and nonfinancial corporations as well as foreign affiliates of South African firms and is defined as net corporate income less net tax and net interest paid (a rela- tively small component) and less net dividends, after adjusting for depreciation and inventory valuation. Poterba (1987) discusses adjusting a related measure of net saving, defined as undistributed profits less capital consumption, for infla- tionary gains on corporate debt. We make no adjustment for inflation, but we do include inflation as a regressor in the corporate saving equation. We analyze two tax effects, both associated with piercing the veil. The first is the impact of dividend taxation, which underwent several changes in the pe- riod. 21 If corporations pursued a dividend policy to mfiyimizp after-tax returns of shareholders, then Poterba's tax discrimination variable—which for South Af- rica, without a capital gains tax in the period estimated, is simply measured as 1 minus the dividend tax rate—will negatively affect corporate saving (see Poterba 1987). Second, without a capital gains tax, inflation is expected to affect corpo- 20. See also Bond, Qiennells, and Devereux (1995) for a cross-sectional study on company dividends. 21. Between 1960 and 1990 dividends on directly held equine* were effectively taxed twice: once through the corporation tax on company profits (out of which dividends are paid) and again through a dividend tax on individuals, in which one-third of dividends were untaxed, and the remaining two-thirds were taxed at die individual's marginal income tax rate. The dividend tax was temporarily abolished in 1990-93 and then replaced by a secondary dividend tax on companies. Aron and Muellbauer 529 rate saving positively. Inflation raises both nominal dividend payments and nomi- nal share prices. This increases the benefits of retained earnings to shareholders, since only dividends are taxed.22 However, an inflation surprise, in principle, will diminish debt and hence reduce the need to save for debt repayments. Since corporate saving takes place largely to finance investment, when outside funds are expensive because real interest rates are high, or are difficult to obtain because access to foreign finance is denied, corporate saving should be higher. Furthermore, when expected income growth rates are high, a higher rate of sav- ing is expected since investment will be particularly profitable (we compare re- sults using forecast rates of income growth, from one-year-ahead real PDNI per capita, and real GDP per capita forecasting equations, derived, respectively, in Aron and Muellbauer 2000a, 2000b). There may also be short-run smoothing effects. In cyclical upturns and in gold price booms, when profits are temporarily high, one might expect companies to save a larger percentage of their income. Finally, we include our financial liberalization indicator, FLIB, to reflect the fact that household debt has its counterpart in corporate assets, particularly assets of financial corporations, and this might help to account for the relatively high corporate saving rate sustained into the 1990s. We estimate an error-correction model of the form: (10) Ay, = Oo + X, P, Axl4 + po( ^ a, */, M - y»-i) + e, where y is the log of the ratio of corporate saving to profits, log(OS/OP"*). The x variables include RLTFLOW, the ratio of long-term capital inflows to GDP; TAXDISCR, defined as 1 minus the dividend tax rate; RPRIME, the real prime rate of interest on borrowing; Alog(PC), the rate of change of consumer prices; and FLIB, the estimated indicator of financial liberalization. These variables are all 1(1), except for the real interest rate, which is borderline 1(0); see table 1. We also examine other cyclical indicators, including capacity utilization in manufac- turing, the gold terms of trade, and the log of real corporate profits, as well as the forecast rates of income growth. A general-to-specific testing procedure on annual data for 1966-97 gives the parsimonious equation shown in the first column of table 2. The lag structure implies a speed of adjustment of around 0.5, meaning that half of the deviation of the lagged dependent variable from the long-run solution is corrected within one year. Figure 9 shows the contribution to the long-run solution of the 1(1) regressors weighted by their coefficients.23 We find that the tax effects on corporate saving are indeed twofold. First, a rise in personal tax rates on dividends (a fall in tax discrimination) raises corporate saving. Second, inflation has a positive effect; that is, higher inflation encourages corporations to retain earnings, given the 22. Corporate saving may also have been encouraged by low or negative real after-tax returns on other assets, such as bonds or liquid deposits, in an inflationary environment. 23. It achieves this by plotting (Ay( + P o y w ) inequation 10 against (P ; Ax^+ P o O(jX iM ) for each 1(1) variable. S30 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 1. Statistics and Variable Definitions Augmented Dickey-Fuller Definition Standard statistics Variable of variable Mean deviation KD K2) Corporate saving equation (1966-97) A logfCS/CP") Growth rate of ratio of corporate saving to GNDI 0.0287 0.1507 -7.78* LogfCS/CP-) Log of ratio of corporate saving to GNDI -0.6074 0.4238 -2.12 -7.784* ARLTFLOW Change in ratio of long-term capital flows to GDP 0.0003 0.0051 -6.76* A RTOTFLOW Change in ratio of total capital flows to GDP -0.00004 0.0073 -5.67* TAXDISCRMA Two-year moving average of tax 0.7086 0.1126 4J5* discrimination, defined as 1 minus effective rate of dividend tax FLIB Estimated indicator of finanrial liberalization 0.1574 0.2210 RPRIME Real prime interest rate/100 0.0370 0.0425 -3.72** ^.44* Alog(PC) Inflation rate (consumption deflator) 0.1035 0.0379 -2.17 -4.92* A2 log(PC) Acceleration of prices (consumption deflator) 0.0013 0.0179 -4.91* AloglRYNf+l)]1"1" Forecast real income growth rate 0.0042 0.0312 -3.67* * Alog[RGDP(+l)]forat Forecast real GDP growth rate 0.0243 0.0246 ^».34* Corporate profits equation (1971-97) Log(CP»n>"/GNDI) Log of gross profits/GNDi -1.9178 0.1937 -1.5272 -4.37* Log(CAPUTMA) Log of manufacturing capacity utilization index (four-quarter 4.4098 0.0253 -7.03* moving average) Log(CORPTAXR) Log of the ratio of corporate tax payments to pretax and interest 0.4981 0.0886 -2.91 -5.0569* corporate profits RTARD? (Tariffs + import surcharges Rvalue of imports 0.0720 0.0189 -1.8910 -4.73* LogfTOTRGOLD) Log of the real rand price of gold -0.0353 0.2467 -3.24" -429* (deflated by import price index) Log(WPIULQ Log ratio of wholesale price index -0.4303 0.1017 -3.25 -4.17* (domestically produced output) and unit labor cost in non- agriculture sectors TAXDIF (Maximum personal rate of income tax minus corporate 0.0844 0.0946 -1.90 -2.41 tax)/100 A log(GNDI) Growth rate of national disposable income 0.1448 0.0412 -4.29" -6.84* A2 (PRIME) Acceleration in prime interest rate/100 0.0001 0.0364 -5.82* A RPRIME Change in the real prime interest rate/100 0.0033 0.0354 -426* Aron and Muellbauer 531 Table 1. (continued) * Rejection at the 1 percent critical value. * * Rejection at the 5 percent critical value. Note For a variable X, die augmented Dickey-Fuller (1981) statistic is the f-ratio on it from the regression: AX, • = it X M + £*,,, 8; AX^,. + v 0 + v , f+ e p where k is the number of lags on the dependent variable, Vo is an intercept term, and t is a trend. The Jkth-order augmented Dickey-Fuller statistic is reported, where k is the last significant lag of the 2 lags employed. The trend is included only if significant For null order I (2), AX replaces X in the equation above. Critical values with constant, and with and without trend, are obtained from MacKinnon (1991). Source: Authors' calculations. lack of a capital gains tax. Thus there is supporting evidence in favor of the sector saving on behalf of households, although the inflation effect may also reflect another motive. However, it is clear from figure 9 that variations in tax discrimi- nation cannot explain the rise in corporate saving, although they may have helped to sustain the level of corporate saving in the past 35 years. The rise in inflation in the 1970s accounts for a large part of the rise in the corporate saving rate. The fall in inflation in the 1990s was offset by higher real interest rates and the liberalization of consumer credit markets, which helped to sustain corporate saving (figure 9). The fall in the dividend tax rate (rise in tax discrimination) between 1979 and 1990 would have led to a decline in corporate saving, but for the upward push from higher real interest rates and from financial liberalization. The change from 1983 to 1997 in our measure of financial liberal- ization for consumer credit markets raised the ratio of corporate saving to gross national disposable income by around 4 percentage points, given real interest rates, the ratio of net profits to national income, and other variables. Three variables help to explain the short-run dynamics of the log of the corpo- rate saving ratio (table 2). One is the forecast growth rate of real PDNI per capita. This variable is significant, whereas the forecast rate of real GDP per capita is not (see column 2), possibly reflecting the fact that PDNI is dominated by wages and salaries. A rise in PDNI thus signals both greater final demand and higher labor costs, capturing both an expansion and a capital-deepening motive for investment, which current corporate saving serves. We find the expected posi- tive effect from the change in the ratio of international long-term capital inflows to GDP especially important from the 1980s to 1994 under financial sanctions.24 There is also an indication of a negative effect on saving from an inflation surprise. The equation has satisfactory diagnostics: there is no sign of residual auto- correlation or heteroskedasticity; it passes a Chow test for parameter stability splitting the sample halfway and Ramsey's RESET specification test. Estimates for the 1966-93 and 1966-89 samples also indicate parameter stability (columns 3 and 4). The first potential break (1966-93) captures the change of regime after democratic elections in April 1994. The second potential break (1966-89) cap- 24. Since June 1999 the South African Reserve Bank has published capital flow figures by functional type of investment rather than categorizing them as long-term or short-term. An alternative specification using total capital inflows fits less well but otherwise gives similar results (column 5 in table 2). S32 THE WORLD BANK ECONOMIC REVIEW, VOL 14, NO. 3 Table 2. Corporate Saving Equation Estimates Dependent variable (1) (3) W (S) A logiCSICP^) 1966-97 1966-97 1966-93 1966-89 1966-97 Regressors Intercept -0.38 -0.11 -0.33 -0.40 -0.42 (1.22) (0.31) (0.67) (0.46) (1.23) Log(CS/CP«) (-1) -0.53 -0.43 -0.50 -0.53 -0.55 (3.58) (2.53) (3.09) (2.35) (3.34) ARLTFLOW -3.47 -3.41 -4.34 ^».25 (3.30) (2.71) (3.33) (2.91) A RTOTFLOW -6.99 (2.17) TAXDISCRMA -1.29 -1.05 -1.23 -1.25 -1.30 (2.71) (2.06) (1.78) (1.17) (-2.47) FLIB(-l) 0.57 0.52 0.55 0.23 0.60 (2.11) (1.75) (1.11) (0.25) (2.00) RPRIME 3.80 1.59 3.53 3.87 3.63 (2.56) (1.13) (2.19) (1.87) (2.18) A log(PC) 7.30 4.68 6.75 7.39 7.66 (3.24) (1.89) (2.73) (2.00) (3.08) A2 log(PC) -2.97 -1.59 -3.06 -3.04 -2.93 (2.05) (1.10) (-1.78) (-1.56) (-1.83) Alog[RYN(+l)]feOT 3.64 3.24 3.51 3.84 (2.29) (1.93) (1.65) (2.18) Alog[RGDP(+l)]foraI 0.98 (0.60) Diagnostics Standard error 0.1138 0.1250 0.1180 0.1291 0.1259 R1 0.577 0.489 0.612 0.594 0.482 Adjusted R2 0.430 0.311 0.448 0.377 0.302 Durbin Watson 2.20 2.27 2.16 2.17 2.26 LM1 (Lagrange multiplier test for 1st order serial correlation) 0.42 0.72 0.20 0.34 0.72 LM2 (Lagrange multiplier test for 2nd order serial correlation) 0.34 2.89 0.21 0.43 0.55 Ramsey reset test 0.02 0.40 0.03 0.21 0.10 [0.89] [0.53] [0.86] [0.86] [0.76} CHOW test 1.297 0.998 0.668 1.198 1.158 [0.320] [0.483] [0.715] [0.428] [0.389] 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 1 uses the ann11ali7.fi four-quarter ahead forecast real income growth rate estimated using quarterly data and a stochastic trend in STAM P 5.0 (Koopman and others 1995), while equation 2 replaces this with the forecast real GD P per capita growth rate, estimated in the same manner (see Aron and Muellbauer 2000b). Equa- tions 3 and 4 use shorter samples, defined by particular regime breaks, to test for the parameter stability of equation 1. Equation 5 replaces the long-term flows to G DP variable in equation 1 by total flows. Source: Authors' calculations. Aron and MueUbauer 533 Figure 9. Decomposition of the Log Ratio of Corporate Saving into (Non- stationary) Determinants Multiplied by Their Regression Coefficients, 1966-96 log.Ccorporate saving ratio) 1.5 T 1.0 - - -0.5 -- -1.0 - - Dlog (CSY) + 0.53 x log (CSYX-D Real prime rate —a— Consumer price Inflation Financial liberalization —•— Tax discrimination Source: Table 2, equation 1. tures the change of monetary policy regime with the appointment of a new cen- tral bank governor, Dr. Stals, as well as the onset of political reforms after the release of Nelson Mandela in 1990. A linear combination of 1(1) variables from the first column of table 2 easily passes a Dickey-Fuller stationarity test. Corporate Profits Equation The profits ratio in equation 9, CP"*/GNDI, is broken into two components, (CPnet/CP*roa)-(CPtrot'/GNI>I); both are plotted in figure 8. Net profits are de- fined as corporate income, net of tax and interest payments. However, if tax and interest payments are regarded as items over which corporations have little con- trol in the short run, the ratio of after-tax and after-interest profits to before-tax and bcfore-interest profits can be regarded as given. Thus our second model con- centrates on identifying the determinants of the ratio of gross corporate profits to gross national disposable income. To derive an empirical equation for the share of profits in national disposable income, we begin with the standard analysis of a monopolistically competitive firm. This model assumes constant average and marginal costs, AC = MC. The share of gross profits in total revenue is 534 THE WORLD BANX ECONOMIC REVIEW, VOL. 14, NO. 3 (11) C P ^ / P Y = (P - AQY/PY where Y is real GNDI. Defining the markup, m, as the ratio of price, P, to aver- age or marginal cost, that is, m = PIAC, the profit share becomes (12) CPP°"/PY=l-frt-\ We approximate log(CPgro"/PY) by a linear function of log(m). Although m is unknown, we have price data for the main elements of P and AC, relative to a base year. Average costs are proxied by a weighted average of labor and other costs, assuming the cost share of labor to be a and the cost share of other inputs to be (1 - a). The markup, m, is approximated by (13) m = m0WPI/(ULQa (WPI) *-« = m^WPI/ULC)a where ULC is the index of unit labor costs, WPI is the wholesale price index for domestically produced output, and mo is the markup in the base year. For ex- ample, with plausible values of a = 0.6 and m0 = 1.5, equations 12 and 13 imply a coefficient around 1.2 in the regression of log(CPgroa/GNDI) on log(WPI/ ULC).25 The approximation in equation 13 can be improved by including other vari- ables. First, the gold terms of trade (TOTRGOLD, defined as the ratio of the rand gold price divided by an index of import prices) improves on WPI as an output price measure: gold has a small weight in the WPI but is a major source of revenue. Second, the ratio of tariffs and surcharges to imports (RTARIF) reflects another cost component. Third, the level of real interest rates, RPRIME, cap- tures an aspect of costs for corporations holding floating debt, although since financial corporations benefit from raised interest rates, the overall effect on sav- ing may be small. Fourth, to capture the role of fixed costs in reducing average costs as output increases, we include capacity utilization, CAPUT. Finally, we note that the size of the corporate sector can vary with tax policy. We control for two effects: a measurement effect and a real effect. If personal income tax rates are high rela- tive to corporate tax rates, individuals with a business income have an incentive to become incorporated. We define the difference between the highest personal income tax rate and the corporate tax rate as TAXDIF, which can be expected to raise the share of profits in national income.26 Second, a high ratio of taxes on companies relative to profits can act as a disincentive to earn or declare profits.27 There are two dynamic effects. The model includes lagged GNDI among the explanatory variables to reflect lags in production and lags in recording profits 25. Note that dlog(CPgross/PY)/dlog[m) = l/(m - 1) > 1. 26. Recent evidence for just tuch an effect in the United States comes from Gordon and Slemrod (1998). 27. We measure this tax ratio as log(CORPTAXR) • = log(l+ (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 (1998). Aron and Muellbauer S3S 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{CPgro"). 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(CPs^/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(CAPlTT), APRIME, and Alog(GNDJ). A general- to-specific testing procedure on annual data for 1971-97 gives the parsimonious equation shown in column 1 of table 3.u We could accept the hypothesis that the coefficient on log( CP*ro"/GNDJ)_i 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(GNDT). This would be equivalent to omitting this term and redefining the dependent variable as log(CPgro"/GNDIma)i 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 tee 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. 53 6 THE WORLD BANK ECONOMIC REVIEW, VOL. 14, NO. 3 Table 3. Corporate Profits Equation Estimates Dependent variable logfCPv-ZGNDI) (1) 1971-97 a) 1971-97 (3) 1971-93 (4) 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) logfTOTRGOLD) 0.56 0.58 0.57 0.56 (13.98) (11.26) (15.73) (11.72) logfwTIULC)_, 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 . ARPRLME -0.80 (2.58) Diagnostics Standard error 0.0309 0.0396 0.0269 0.0296 R1 0.982 0.971 0.989 0.986 Adjusted R1 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 (-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 A2 (PRIME) in equation 1 by ARPRLME. Equations 3 and 4 use a shorter sample, defined by particular regime breaks, to test for die parameter stability of equation 1. Source: Authors' calculations 1(1) variables and at most 27 observations, the Johansen procedure (for example, Johansen and Juselius 1990) lacks power. However, a linear combination of 1(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. Decomposition of the Log Profit Share into the Main (Nonstationary) Determinants Multiplied by Their Regression Coefficients, 1966-96 log (profit share) Oevd-adjusted) 0.6 -r -0.2 -0.4 - - -0.6 - - 1 I I I I I I I I I 1 I I II I I I I I I I I I I 1 I I I I I LI 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 —— Log (gross profits/GNDi) Gold terms of trade Tax differential —"— Log (WPI/unit labor costX-D CTarilrs+surchargesyimports 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 at 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). S40 THE WORLD BANX ECONOMIC REVIEW, VOL. 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 die 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 BANX 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. Bandicra, Oriana, Gerard Caprio, Patrick Honohan, and Fabio Schianterelli. 2000. "Does Financial Reform Raise or Reduce Saving?" Review of Economics and Statistics 82(2)^39-63. 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 O. 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. . 1985. "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(l):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 S43 Fry, Maxwell. 1995. Money, Interest, and Banking m 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 Serv6n, 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(l):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: M TT Press. MacKinnon, James G. 1991. "Critical Values for Co-Integration Tests." In Robert Engle and Clive Granger, eds., Long-run Economic Relationships: Readings m 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 BANX 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 TheoreticaJ 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. Samo, 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 Serv6n. 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.