Government Spending, the Real Interest Rate, and the Behavior of Liquidity-Constrained Consumers in Developing Countries

The importanceof the mobilization of domestic savings in developing countries can hardly be understated, given the present state of inter-national capital markets. Unfortunately, however, the understanding of the actual determinants of domestic savings in developing countries is still scanty. A number of empirical investigations (mainly reproducing the literature relating to developed economies) have been carried out in recent years, but the paucity of reliable data has made it difficult to test the underlying hypotheses and obtain results that warrant a reasonable degree of confidence.

Abstract

The importanceof the mobilization of domestic savings in developing countries can hardly be understated, given the present state of inter-national capital markets. Unfortunately, however, the understanding of the actual determinants of domestic savings in developing countries is still scanty. A number of empirical investigations (mainly reproducing the literature relating to developed economies) have been carried out in recent years, but the paucity of reliable data has made it difficult to test the underlying hypotheses and obtain results that warrant a reasonable degree of confidence.

The importanceof the mobilization of domestic savings in developing countries can hardly be understated, given the present state of inter-national capital markets. Unfortunately, however, the understanding of the actual determinants of domestic savings in developing countries is still scanty. A number of empirical investigations (mainly reproducing the literature relating to developed economies) have been carried out in recent years, but the paucity of reliable data has made it difficult to test the underlying hypotheses and obtain results that warrant a reasonable degree of confidence.

The purpose of this paper is to provide additional and more conclusive empirical evidence on this topic. In order to permit comparison with the most recent work on the subject, Euler equations for the representative consumer’s stochastic dynamic optimization problem are estimated. In contrast with much of the previous literature, however, the theoretical framework used here is based on the consideration that some basic assumptions on which savings functions for developing countries have been estimated may not be entirely realistic. In particular, a significant fraction of the population in developing countries can be expected to be affected by liquidity constraints that substantially diminish consumers’ ability to substitute consumption intertemporally, as is assumed by the well-known life-cycle theory. This liquidity problem is attributable to a number of factors, including capital market imperfections. The gravity of capital market imperfections continues to be a matter of debate even in countries with apparently sophisticated financial institutions and well-developed capital markets (see Hayashi (1985), for a review, and Hubbard and Judd (1986)), but this phenomenon has never been ex-plicitly accounted for in developing countries, although there are several reasons why such imperfections are likely to be exacerbated in those countries (Blejer and Cheasty (1986)). Allowance is therefore made for departing from optimal behavioral rules by the representative consuming unit described by the theory.

In the context of a theoretically plausible model of consumer behavior that allows for borrowing constraints, two major issues are addressed.

First, the long-debated issue of the real interest rate elasticity of savings is re-examined. As is well known, the effects of the rate of return on the level of savings and the rate of capital formation are important to both economists and policymakers, since they bear on a number of central macroeconomic questions. The relevance of the interest rate elasticity of savings is further enhanced in development economics, where competing views on the role of financial conditions in the eco-nomic growth process rely crucially on the degree of responsiveness of aggregate savings to changes in the rate of return.

Notwithstanding considerable research to determine the interest rate responsiveness of savings behavior in developed economies, the tradi-tional view that changes in the rate of return are likely to have only a minor effect on the savings rate holds (Modigliani (1986), p. 304)), but controversy still exists (Summers (1984)). In the case of developing countries, the lack of empirical work on the responsiveness of savings in the 1960s and early 1970s is emphasized in the surveys by Mtkesell and Zinser (1973), and by Snyder (1974), which describe the evidence as sketchy at best.1 More recent attempts (Fry (1978, 1980), Giovannini (1983, 1985), McDonald (1983), and Leite and Makonnen (1984))can be questioned on the basis of their limited geographical coverage, the un-reliability of available data, and, in some cases, the underlying meth-odology. Consequently, their results cannot be used with confidence for policy analysis.

Fry estimates a (national) savings function for seven Asian countries2 for the period 1962-72 and finds strong support for the hypothesis of a negative real interest rate elasticity of domestic consumption. He esti-mates this elasticity to be about —0.2. Similar conclusions are reached by McDonald and Leite and Makonnen. McDonald focuses on factors de-termining savings behavior in 12 Latin American countries3 and provides evidence of a negative relationship between the real interest rate and private consumption in most of the countries examined of a magnitude roughly comparable with that found by Fry. Leite and Makonnen’s study concentrates on six African countries4 and provides evidence of a small but positive relationship between private savings and the real interest rate.

The hypothesis of a positive and significant relationship between real interest rates and savings in developing countries is questioned by Giovannini (1985, p. 199), who shows that “the apparent empirical success of the high interest elasticity hypothesis depends in a crucial way on the presence in the sample of a few observations that have a disproportionately large influence on the estimated” response of savings to the real interest rate. In this work, Giovannini extends the analysis to 18 developing countries5 and, bypassing many of the econometric problems associated with aggregate savings equations, estimates the elasticity of intertemporal substitution in (private) consumption. Using annual data, Giovannini finds that in only 5 out of 18 countries is the intertemporal substitutability in consumption not likely to be smalt,6 therefore im-plying, other things being equal, that the interest rate elasticity of savings is positive.

Giovannini’s work represents a considerable improvement in the un-derstanding of savings behavior in developing countries. However, it can hardly be considered conclusive. First, his main result proceeds from parameters for which precise estimates were difficult to obtain. In 11 out of 18 cases, the coefficient of intertemporal substitution is positive, but with standard errors so large as to make any conclusion questionable. Since Giovannini’s sample period in most cases covers the 1960s, the inconclusiveness of his results is not unexpected, given the low variability of real rates in that period. Second, as far as geographical coverage is concerned, Giovannini’s work neglects those regions for which evidence is most lacking—that is, Africa and the Middle East.7 Third, as Giovan-nini points out, some of the assumptions under which the elasticity of substitution is estimated in his 1985 study may not be realistic in develop-ing countries. In particular, some proportion of aggregate consumption is likely to be accounted for by the consumption of liquidity-constrained individuals, for which the first-order condition on which the estimation is based does not hold.8 While the existence of liquidity constraints implies a relatively small elasticity of savings9 and can therefore explain Giovannini’s results, it also implies misspecification of the estimated unrestricted first-order condition and calls for further investigation.

Section I provides a framework for the estimation of the degree of intertemporal substitution in consumption, in which liquidity constraints are explicitly allowed for. Section I also addresses the second major issue dealt with in this paper—extending the representative consumer’s utility function to include government spending and assessing the role of public expenditure in private consumption decisions. The importance of the response of private spending to changes in government spending stems from the observation that, if government spending is a substitute for private spending, then government expenditure restraint policies are likely to induce higher private consumption. Most adjustment programs in developing countries, however, attempt to ensure that government deficits do not absorb an unduly high share of private savings. Indeed, the existence of public surpluses (presumably to be achieved through tight expenditure policies) is often seen as a way to provide a pool of loanable fund savings to private sector investors, thereby avoiding the problems that characterize more traditional policies aimed at mobilizing private savings. This argument, however, disregards the fact that direct crowding out can partly or fully counteract government efforts.10 Indeed, for some Latin American countries, McDonald (1983) provides evidence of a sizable degree of substitution between private and public consump-tion. But his results rest on an inappropriate definition of disposable income and require further investigation.

As Section II makes clear, an effort is made to construct as accurate and extensive a data set as existing sources allow. In particular, the empirical analysis in Sections III and IV focuses on private savings behavior over the period 1973-83 in 49 developing countries, grouped in 6 sets of pooled time-series cross-section observations, each one refer-ring to a single geographical region.

Finally, Section V presents the main conclusions of the analysis and its policy implications.

I. Theory

Research on consumption in the early 1980s (reviewed in Deaton (1986), and King (1986)) has been influenced by the important works of Hall (1978), Grossman and Shiller (1981), and Hansen and Singleton (1982), which open the possibility of a direct estimation of the parame-ters of the intertemporal utility function that characterize the behavior of a representative individual without requiring explicit solutions of the consumers’ dynamic optimization problem. In addition. Hansen and Singleton have shown how to test the overidentifying restrictions implied by the hypothesis of continuous optimization of a stable, additively separable objective function.

The present study follows this line of research by suggesting that aggregate consumption can be modeled as the outcome of optimizing decisions of a representative consumption unit (household).11 The household faces an economic environment in which future opportunities are uncertain, and has a stationary utility function that is additively separable through time, and is defined over a composite consumption good as follows:

Vt=Et[ΣτtTρτt(Uτγ/γ)],γ<1,(1)

with

Uτ=(Cτ1αGτα),0α1.(2)

In equations (1) and (2), Vt is expected utility att, Et is the expectations operator conditional on information available at t,ρ is a constant dis-count factor, and U is a function increasing and concave in a Cobb-Douglas aggregate of per capita private and public consumption. Cτ is private consumption of goods at τ, and Gτ is government expenditure in period τ. The parameter γ in equation (1) controls intertemporal substi-tution: large and negative values of γ characterize consumers who are willing to smooth consumption over time and who respond only to substantial changes in incentives.

The consumer (household) maximizes equation (1), subject to the following period-to-period budget constraint:

Aτ=Aτ1Rτ+YτCτ,(3)

where Aτ defines real assets at the end of period τ, Rτ is the real rate of return between periods τ-1 and τ, and Yτ is real nonproperty income (net of taxes) in period τ. As long as the optimum path lies in the interior of the budget set, simple perturbation arguments can be used to establish certain characteristics of the optimal path. At any point along an optimal path the representative consumption unit cannot make itself better off by forgoing one unit of consumption at time r and using the proceeds to purchase any other good at any other point in time. Formally, at time t the marginal condition will be given by

Et[Rt+1(Vt/Ct+1)/(Vt/Ct)1]=Et(Ft+11),(4)

which, apart from implicitly defining Ft+1, is satisfied for any free-traded risky asset (even if other assets such as human capital cannot be traded freely) and holds for consumers who expect with certainty to be alive in the next period regardless of the length of horizon of their maximization problem. Notice also that equation (4) does not depend on any assump-tion about expectations regarding future labor income, government spending, or rates of return.

Estimation of the first-order condition for utility maximization (for example, equation (4)) represents an alternative approach to estimating standard consumption functions. The difficulties associated with the latter are well known and mostly concern the Lucas critique: the relation between consumption, income, and interest rates depends on the wider macroeconomic context and may not be stable over time, even though preferences remain unchanged. The research done so far, however, has provided only limited support for the econometric restrictions implied by the Euler equation approach. Furthermore, the assumptions usually underlying the application of this approach are not generally accepted (see Ando and Kennickell (1986), Blinder and Deaton (1985), Deaton (1986), and King (1986)).

Under rational expectations and market clearing, the first-order con-dition (equation (4)) holds ex post except for an error term uncorrelated with information available to the consumption unit at time t. In other words,

Ft+1=1+t+1,(5)

where Єt+1 is the forecast error with a zero mean and a constant variance.

In the case of the time separable, constant relative-risk-aversion set-ting given by equations (1) and (2), with lowercase letters denoting natural logarithms and Δ acting as the difference operator, equations (4) and (5) imply12

Δct+1=ψc+ψrEtrt+1+ψgEtΔgt+1+ut+1,(6)

where ψc=ψc(ρ,σ2,γ,α),ψr=1/[γ(α1)+1],andψg=γαψr,, since ψr>0,ψg is greater, equal, or less than zero depending on whether γ is positive, zero, or negative. In equation (6), the error term ut+1 reflects the impact of “news” (or “surprises”) about current levels of income, interest rates, and government spending. It is therefore orthogonal to all past information.13

As it stands, equation (6) still disregards the possibility that some consumers may face quantity constraints on the amount of borrowing or that loan rates available to them may be higher than the corresponding lending rates. These possibilities may arise for a number of reasons, including imperfections in capital markets and tax policy. For example, the tax system can generate divergences between after-tax rates on bor-rowing and lending. Alternatively, large transaction costs and the possi-bility of bankruptcy or asymmetric information about creditworthiness between lenders and borrowers or both can result in lenders denying loans to potential borrowers with particular characteristics.

Suppose, then, that the liquidity constraint takes the form of a re-striction on the total net stock of traded assets, as follows:

AtΦt+Φyt,forallt,(7)

where a negative value of At indicates net indebtedness in period t. The additional condition AT≤0 provides the necessary endpoint constraint. Equation (7) is expressed in terms of the net position in order to allow the use of illiquid assets as collateral. According to the equation, potential lenders make the size of the loan conditional upon nonproperty income. Notice that the lending rule is time dependent, since the intercept Φt, is allowed to respond to changes in government legislation and macroeconomic conditions in general.

Expressing liquidity constraints in the form of equation (7)—that is, exogenous stock constraints, is important because such borrowing restrictions are exploitable by stabilization policy (Hubbard and Judd (1986)). Of course, other alternatives are conceivable. For example, Hayashi (1985) discusses the case of imperfect information in the loan market and shows that it is not necessarily exploitable for stabilization purposes (see also King (1986)).

Under the additional constraints given by equation (7). it can be shown (see Muellbauer (1983, 1986a) and Zeldes (1985)) that equation (6) has to be augmented by adding the term (ψrt), where μt, is an increasing function of the shadow price associated at t with being credit-rationed, or, in other words, it is the marginal increase in expected lifetime utility derived from a unit relaxation of the credit constraint in period t. Since agents are constrained from borrowing more, but not from saving more, μt, is zero when the constraint is not binding and positive when it is binding.

In principle, μt, could be derived by solving the whole intertemporal programming problem. However, a sufficiently general solution is hardly likely to be operational. As an alternative, Muellbauer (1986a), p. 10) suggests that, if “consumers are most likely to want to borrow and hence, other things being equal, to encounter credit restrictions when future income prospects look bright compared with current circumstances,” then, in the aggregate, μt, is likely to depend positively on terms like Et(Zt+1-ct), where Zt=At-1(Rt-1)+Yt—that is, real disposable in-come.14 In other words, consumers who are liquidity constrained at t may not expect to be constrained at t+1 and may therefore be forced to let their consumption path follow more closely their income path. Equation (6) would therefore be rewritten as

Δct+1=ψ0+ψrEtrt+1+ψgEtΔgt+1+ψμEt(zt+1ct)+ut+1,(8)

which can be interpreted as an approximation to the Euler equation for consumption incorporating credit constraints. Abstracting from real interest rate and government spending effects, as ψμ moves closer to 1 in equation (8), consumption increasingly mimics income developments. Notice that, for ψg = ψμ = 0, equation (8) reduces instead to Hall’s (1981) original formulation estimated by Giovannini (1985).15

The Data Set

A thorough empirical analysis of private savings behavior in developing countries raises several difficult statistical problems, mostly stemming from inadequacies in the data and their lack of comparability. A reasonable number of observations on aggregate time-series data are available, on a consistent basis, for only a few developing countries. In the great majority of cases, less than 20 annual observations are available.16 In such a situation, pooling cross-section and time-series data for a number of countries seems to be the most sensible procedure, provided that sufficient allowance is made for obvious institutional and cultural differences among countries.

Following this line of research, the empirical analysis presented here is based on six sets of pooled time-series cross-section data, each one referring to what is intended to be a homogeneous geographical region.17

The first set includes 12 countries in sub-Saharan Africa. To give a different order of magnitude, this sample covers 40 percent of the 1975 gross domestic product (GDP) of the whole region as defined in the World Bank’s World Tables (1983). The second set includes five coun-tries in North Africa and the Middle East, accounting for 61 percent of GDP of the whole region for 1975. The third set covers nine countries in East and South Asia and the Pacific, or 46 percent of the region’s GDP for 1975. The fourth and fifth sets cover eight countries in Central America (including the Caribbean) and nine in South America, respectively, with coverage in terms of 1975 regional GDP of 76.2 percent and 83.1 percent, respectively.18 Finally, the sixth set of data includes six Southern European countries, totaling 77 percent of the regional GDP for 1975. The sample as a whole contains 11 low-income, and 38 middle-income countries. Low-income countries are, therefore, somewhat underrepresented.19 A description of the data set is provided in the Appendix.

It is important to recall that appropriate measurement is particularly difficult in the case of real interest rates, where the problem of choosing a particular interest rate series from those series that may be available is coupled with the question of appropriately deflating nominal interest rates (Khatkhate (1985)). In this respect, the approach described in the previous section turns out to be particularly useful, because the relation-ship represented by equation (6) should hold for all real rates of return corresponding to freely traded assets.

To provide an indication of the robustness of the results, two alternative measures of the nominal rate are used. On the one hand, domestic interest rates on term deposits of commercial banks, which constitute a relatively large segment of the financial system in developing countries, are considered.20 On the other, the nominal interest rate is derived, with implicit reference to the small open economy model, as the relevant foreign interest rate adjusted for expected changes in the exchange rate. The latter alternative implies that the relevant real interest rate depends on the rate of change of the real price of home goods (Dornbusch (1983)). Of course, it may be argued that the small open economy stereotype is inappropriate for most developing countries that are characterized by pervasive foreign exchange and trade controls. It has been suggested, however (Tanzi and Blejer (1982)) that, even in countries with severe restrictions on capital movements and other exchange controls, it is unlikely that economic agents will be prevented from illicitly substituting foreign currency and foreign financial assets for domestic currency and domestic financial assets if incentives are sufficiently strong.21

III. Estimation

For estimation purposes, the theoretical model described in Section I is rewritten as follows:

Δct+1i=Ψci+ΨrEtrt+1i+ΨgEtΔgt+1i+Ψμ,lEt(zt+1jctj)+ψμ,mEt(zt+1kctk)+ζz[zt+1iEt(zt+1i)]+ζg[gt+1iEt(gt+1i)]+ζr[rt+1iEt(rt+1)]+ν¯t+1+νt+1i,foralljlandkm,(9)

where the superscript i identifies the ith country in each of the geographical areas referred to in the previous section. In other words, the constant term in equation (9), being a function of the variance of the forecast error, is allowed to differ among countries because countries with a higher share of the product originating from agriculture, for example, are likely to face higher uncertainty. In addition, the coefficient of the proxy for borrowing constraints,ψμ, is allowed to take different values in low-income countries (that is, those countries identified by the superscript j and belonging to the subset identified by the subscript l) and in middle-income countries (that is, identified by the superscript k and belonging to the subset m> Low-income countries are taken to be the countries currently eligible for use of International Development Association (IDA) resources. Of course, under the interpretation given to ψμ in this paper, ψμ, l>ψμ, m is expected.22

Finally, the original error term in equation (6)—that is. ut+l, is now linearly decomposed into three innovation terms referring to z, g, and r, respectively, as well as two random components that have mean zero but are not necessarily homoscedastic, because the variance of different countries’ forecast errors may differ, and this difference could be only partially incorporated into the innovation terms. The first component is country-specific and is uncorrelated across countries,(νt+1i), and the second is an area-wide component, which equally affects all countries in a particular geographical area (ν¯t+1).23 The obvious example of the latter component is given by the recent drought in sub-Saharan Africa, as long as its effects are not already in the income “news,”

Notice that the variable rti is alternatively defined as (qtiΔpti), where qt is the domestic nominal interest rate, and pi is the (logarithm of the) consumer price level; or as (qt*+ΔetiΔpti),where,apartfrompti,qt* is the representative nominal interest rate paid on foreign currency assets, and et is the (logarithm of the) exchange rate defined as domestic currency per unit of foreign currency.

If the expected (or unexpected) nature of the variables on the right-hand side of equation (9) is disregarded for the time being, the appropriate estimator for the setting described by equation (9) is given by what is known as the “between-within groups fixed-effects estimator,’’ if we can regard each country as a group. As Mundlak (1978) shows, this estimator amounts to applying ordinary least squares to equation (9), expressed in terms of “transformed” variables—that is:

Δc¯t+1i=ψrEtr¯t+1i+ψgEtΔg¯t+1i+ψμ,lEt(z¯t+1jc¯tj)+ψμ,mEt(z¯t+1kc¯tk)+ζz[z¯t+1iE(z¯t+1i)]+ζg[g¯t+1iEt(g¯t+1i)]+ζr[r¯t+1iEt(r¯t+1i)]+ν¯t+1i,foralljlandkm,(10)
x¯ti=xti(1/T)Σjxji(1/N)Σkxtk+(1/NT)ΣjΣkΣjk,j=1,,T;K=1,,N,(11)

for a generic variable xt, and where T and N denote the number of time periods and the number of countries, respectively—that is, the transformed variable is the original variable minus the country and time means plus the total mean. Notice that the transformation eliminates the constant term and the area-wide error term. In general, the transformation would eliminate all variables not simultaneously indexed on i and t. Therefore, if the nominal interest rate is given by the adjusted foreign interest rate, the term Etr¯t+1ireducestoEt[Δ(et+1ipt+1i)].

To incorporate expected and unexpected (or “surprise”) variables in equation (10) the well-known two-step procedure is used. This procedure involves the estimation of an auxiliary set of equations describing the variables about which expectations are formed and then substituting the estimated residuals and predicted values as appropriate in the structural equation of interest.24 A vector autoregression (VAR) is estimated for the transformed variables z¯,g¯,andr¯, the right-hand-side variables of the VAR including lagged consumption, lagged disposable income, lagged government spending, lagged nominal interest rate (or lagged devaluation), two lags of the price level, and a time trend.25 In general, the VAR equations for z¯andg¯ fit the transformed data quite well, but, as would be expected (Hall (1981)), the real interest rate r¯ appears to be more difficult to predict (see the Appendix). Disposable income and government spending are strongly autoregressive, and, in addition, they help to predict each other; increased inflation signals a future slowdown in the rate of growth.

Once anticipated and unanticipated series are available, equation (10) can be estimated by ordinary least squares. As shown by Pagan (1984), however, the two-step procedure does not yield correct estimates of all the standard errors. In particular, although the standard errors of the coefficients of the “surprise’ variables are correct, standard errors for the remaining coefficients have to be obtained from a two-stage least squares regression that omits the surprise terms and uses the VAR as the first stage.

IV. Empirical Results

Tables 1-6 report the estimates of the coefficients of equation (10) for the six geographical regions described in Section II Before the tables are examined in detail, it is worth emphasizing their implications. First, the omission of liquidity constraints appears consistently and seriously to bias downward the estimates of the intertemporal elasticity of substitution. Second, where liquidity constraints are substantial (as in regions where the use of IDA resources is common), intertemporal substitution is weak, and very large changes in incentives are necessary to induce postponement of consumption. Third, as expected, low-income countries suffer most from liquidity constraints and therefore react strongly to expected income changes, although there is no clear-cut pattern in the way different countries react to unexpected income shocks. In short, the picture that emerges from the evidence is a highly coherent one in which differences in behavioral responses appear to be linked more to the stage of development of different areas or countries than to unexplained shifts in preferences.

Table 1.

Parameter Estimates and Test Statistics: Sub-Saharan Africa

article image
Note: The figures in parentheses are heteroscedasticity-consistent standard errors. The regressions also include a dummy variable taking a value of 1 in 1973 and of —1 in 1974 for Swaziland. This accounts for two large outliers but does not affect the remaining coefficients. Its coefficient takes a value of -0.51 (0.03) in the equation of columns (1) and (3), and -0.47 (0.02) in the equations of columns (2) and (4). R2 is the coefficient of determination; DW denotes the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(20,82).

Distributed as f(20,76).

Distributed as F(‘20,77).

In Tables 1-6, the columns labeled (2) and (4) report the coefficient estimates for the two measures of the real interest rate and the six subsamples, respectively, their heteroscedasticity-consistent (White (1980)) standard errors (derived as above), as well as some diagnostic statistics such as the Durbin-Watson statistic for fixed-effects models given in Bhargava, Franzini, and Narendranathan (1982), and a Chow stability test across the 1981-83 period.26 This period coincides with the downward trend of oil prices (in U.S. dollars) and, therefore, also with substantial (and, in recent times, unprecedented) shifts of real income from oil exporting to oil importing countries. In addition, the same period witnessed the emergence of the debt crisis. Stability tests are expected to detect possible structural breaks related to these events. To allow for comparison with previous work, columns (I) and (3) report the results derived from following Giovannini (1985) and estimating Hall’s (1981) original formulation (corresponding to equation (10), with ψg = ψμ,l = ψμ,m = ζz = ζg = ζr = 0).

Table 2.

Parameter Estimates and Test Statistics: Middle East and North Afica

article image
Note: The figures in parentheses are heterosce elasticity-con si stent standard errors. R2 is the coefficient of determination; DW denotes the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(10, 33).

Distributed as F(10, 29).

Tables 16 also report the estimates of the implied behavioral parameters, as well as some interesting functions of the same parameters, along with their standard errors derived by linearizing the underlying non-linear functions.27 In particular, the tables show estimates of the parameters γ and α In the restricted model, 7 controls the intertemporal elasticity of substitution in consumption, which is given by γ(1 - α) in the full model (equation (10)). The parameter a defines the weight of government spending in the Cobb-Douglas consumption index (equation (2)) and, if nondistortionary taxes are available and perfect transformation in production is assumed, it also defines the optimal provision of public goods as a percentage of private ones (that is, α/(l-α)).

Table 3.

Parameter Estimates and Test Statistics: East and South Asia and the Pacific

article image
Note: The figures in parentheses are heteroscedasticity-consistent standard errors.R2 is the coefficient of determination; DW denotes the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(21, 62).

Distributed as F(21, 56).

Distributed as F(21, 57).

In general, equation (10) constitutes a substantial improvement over its restricted version. The available diagnostic does not suggest mis-specification, and, in particular, the hypothesis of parameter constancy across the 1981-83 period cannot be rejected except in South America.

Contrary to Giovannini’s (1985) findings, there is clear-cut evidence of a positive relationship between the rate of growth of per capita consump-tion and the expected real interest rate. Furthermore, in three regions out of six (Middle East and North Africa, Southern Europe, and Central America), the coefficient ψr also turns out to be positive and significantly different from zero, although this result depends on the definition of the real interest rate. It may be argued that assets denominated in foreign currency are unlikely to be a significant item in private portfolios in sub-Saharan Africa, in contrast to those in the Middle East and North Africa and in Southern Europe.28 In general, however, the restricted model estimated by Giovannini tends to bias downward the estimate ofψr

Table 4.

Parameter Estimates and Test Statistics; Southern Europe

article image
Note: The figures in parentheses are heteroscedasticity-consistent standard errors. The regressions also include two dummy variables taking a value of 1 in 1974 for both Cyprus and Portugal. They do not affect the remaining coefficients, and their coefficients take the following values: -0.19 (0.01), -0.10 (0.02), -0.20 (0.01), -0.08 (0.02) for the first dummy variable in the equations of columns 1 to 4, respectively; 0.09 (0.01), 0.07 (0.01), 0.09 (0.01), 0.05 (0.01) for the second dummy variable in the equations of columns 1 to 4, respectively. R2 is the coefficient of determination; DW is the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(14,39).

Distributed as f(14,36).

Distributed as F(14,35).

Table 5.

Parameter Estimates and Test Statistics: Central America and the Caribbean

article image
Note: The figures in parentheses are heteroscedasticity-consistent standard errors. R2 is the coefficient of determination; DW denotes the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(18, 55).

Distributed as F(18, 50).

Distributed as F(18, 51).

Notwithstanding these results, ψr still remains quite small, and the intertemporal elasticity of substitution therefore tends to take on negative values that are larger, in absolute terms, than those observed in developed economies.29 In addition, with the exception of the Middle East and North Africa and Central America and the Caribbean, the estimates of γ do not tend to differ widely across regions, and they indicate a reduced response by consumers to changes in incentives. It is important to stress that if the sample excluded 1982 and 1983, South America too would show a positive ψr coefficient that was significantly different from zero.30 As is apparent from the Chow test, however, the relationship weakens considerably in the early 1980s. Therefore, al-though the extent of the misspecification of the restricted model is apparent and substantial, the main thrust of Giovannini’s work remains largely unaffected.

Table 6.

Parameter Estimates and Test Statistics: South America

article image
Note: The figures in parentheses are he teroscedasti city-con si stent standard errors. R2 is the coefficient of determination; DW denotes the Durbin-Watson statistic; σ^ denotes the standard error of the regression.

Incorrectly signed coefficient (-0.40 with standard error equal to 0.10) set to zero.

(Incorrectly signed and insignificant) coefficient set to zero.

Distributed as F(25, 62).

Distributed as F(25, 58).

Distributed as F(25, 59).

Government spending never appears to play a substantial role in the regressions. No definite pattern of substitution emerges from the estimates. On the contrary, private consumption is mostly insensitive to the expected path of government spending, with the exception of sub-Saharan Africa where the implied estimate of the optimal provision of public goods (as a percentage of private consumption) exceeds the average government spending to private consumption ratio over the 1973-83 period (that is, 0.27).

Most of the improvement with respect to the restricted formulation shown in columns (1) and (3) of Tables 1-6 is therefore clearly attributable to the liquidity constraint proxies and to the impact of “news” on disposable income. The “surprise” variables explain a substantial amount (from 10 percent to nearly 35 percent)31 of the variance of the error in the regressions reported in columns (2) and (4), as the rational expectations approach to the consumption function suggests. Statisti-cally, the innovation in income is the most important such variable.

It cannot be safely said, however, that only unexpected changes in income cause consumption to change, as modern versions of the permanent income hypothesis suggest. The coefficients of the liquidity constraint proxies, ψμ,l and ψμ,m, are always positive, of substantial magnitude, and significantly different from zero.32 In addition, ψμ,l turns out to be always greater and significantly different from ψμ,m and both coefficients are roughly of the same order of magnitude across regions.33 As is to be expected, the relationship between the rate of growth of consumption and the expected real interest rate shows up more clearly and strongly where the proxy for liquidity constraints plays a minor role.

To clarify these issues further. Table 7 focuses on the effects of changes in interest rates on consumption. To characterize fully the consumer’s response to random shocks, a closed-form solution to the stochastic control problem described in Section I would be needed. Since such solutions remain intractable, the approach used by Mankiw, Rotemberg, and Summers (1985) is followed, which concentrates on the effects of interest rate changes in a deterministic environment. The elasticities reported in Table 7 illustrate the changes in consumption at t in response to temporary changes in the real interest rate, from t to t+1. These are short-run elasticities in the sense that the effect of such changes after t + 1 is ignored.34

Table 7.

Real Interest Rate Elasticities of Consumption

article image
Note: Elasticities are computed assuming ρ = 1 + r = 1.03, and assuming, for simplicity, ρ = 0. Changes in these assumptions imply only marginal changes in the elasticities.

Figures in Table 7 describe the percentage change in consumption following a 1 percent change in the variable 1 + r. Hence, if the real rate of interest jumps, say, from 3 percent to 4 percent in sub-Saharan Africa, the corresponding reduction in consumption as implied by equation (10) is about ¼ of 1 percent.

Table 7 conveys, although in a different form, the same message as in Tables 1-6. In particular, the relationship between the degree of responsiveness of consumption to changes in the real interest rate and the magnitude of liquidity constraints (as described by the coefficients ψμ,l and ψμ,m) is, if anything, emphasized.

Assuming that ψμ = 0 yields the Euler equation satisfied under market clearing, or, in other words, assuming that {ψμ[Et(zt+1)-ct]} can provide an estimate of the Lagrange multiplier associated with transferring re-sources between tomorrow and today,35 the pervasiveness of liquidity constraint can be easily seen by computing the (lower) rate of growth of consumption that would have taken place in the absence of such con-straints.36 It turns out that sub-Saharan Africa, North Africa and the Middle East, and South America, which witnessed an average rate of growth of per capita consumption of about 0.6 percent, 4.8 percent, and 1,7 percent, respectively, in 1973-83, would actually have experienced much lower rates of growth (around -0,4 percent, 3.2 percent, and 0.9 percent, respectively). Instead, East and South Asia and the Pacific, and Southern Europe, whose per capita consumption grew by 3.2 per-cent and 2.4 percent, respectively, over the same period, would have had annual growth rates approximately 0.3 percent lower. The only region that actually experienced, on average, negative values of the [Et(zt+1)-ct] variable over that period—that is. Central America and the Caribbean—is actually the only region to show an estimate of Ψμ not significantly different from zero.

V. Policy Implications

Analyses based on cross-country data are subject to several well-known caveats, which warrant appropriate caution in the interpretation of the results, particularly when, as in the present case, data problems are known to be substantial. Nonetheless, the results of the present study appear to be robust in most respects. They provide a coherent picture of private savings behavior in developing countries, offer reasons for the existing behavioral differences among geographical regions, and suggest a number of important policy implications.

With respect to the issue of the real interest rate elasticity of savings, the available evidence indicates that in all regions considered the expected growth of consumption does change with changes in the real interest rate. In addition, in regions such as the Middle East and North Africa, Southern Europe, and Central America and the Caribbean, the response of consumption growth to the expected real interest rate is also significantly different from zero. However, if the magnitude of the estimated parameters is to be taken seriously, the effective mobilization of domestic savings through changes in savings incentives is likely to require changes in the real interest rates, which, given the existing constraints, may prove unfeasible, especially in low-income developing countries. In such a case, a viable alternative is the one considered by Blejer and Cheasty (1986)—that is, the generation of budgetary surpluses. As long as these are derived by expenditure restraints, they are not likely to crowd in additional private expenditure and thereby be counteracted by the behavior of private agents.

More far-reaching, however, are the implications of the existence of pervasive liquidity constraints for fiscal policy design and implementation. The fact that current resources are low relative to lifetime resources but consumers are, to some extent, not permitted to borrow against future income clearly affects the way we look at issues such as the efficacy of temporary tax cuts and the effects of government budget deficits on aggregate demand. As Tobin (1980, p. 57) says, liquidity-constrained consumers are not “indifferent to the opportunity to defer tax payments. Even if they themselves must pay the taxes later, they will increase their consumption now. In effect the government lends to them at its borrowing rate of interest, an option not otherwise available in the credit market.” Arguments about fiscal policy ineffectiveness are therefore affected if a substantial number of consumers are liquidity con-strained,37 although, in an assessment of the real world effects of debt-financed tax cuts, the actual distribution of tax changes is likely to be of significance. Of course, the reduced responsiveness of savings to changes in the real interest rate further emphasizes the role of traditional stabilization policies.

The same arguments that can lead to countercyclical policies on efficiency grounds in the presence of borrowing constraints impinge also on a number of issues in tax policy evaluation and tax reform. If liquidity constraints are present, traditional arguments in favor of wage and con-sumption taxation or proportional taxation and against capital taxation and progressive income taxation lose some of their appeal. For example, Hubbard and Judd (1986, p. 27) show that “[a] switch from progressive to proportional income taxation would speed up tax collection, raising tax rates on low-income consumers and reducing their consumption substantially when liquidity constraints are important.” In other words, tax exemptions, as well as other forms of social insurance, would not only obey considerations of equity, but would also be grounded on efficiency.

Similarly, the usual conclusion suggesting that substantial efficiency costs are likely to characterize capital income taxation as opposed to labor income taxation is likely to be reversed to some extent when liquidity constraints are introduced. Again, the reason is that capital income taxation effectively delays the collection of tax payments over an individual’s life cycle.

It should be stressed that tax policies designed to lessen the burden of borrowing constraints may induce substantial welfare gain if the public does not substitute easily between present and future consumption. If, as appears to be the case in developing countries, people prefer an even consumption path and show low elasticity of intertemporal substitution, the welfare cost of borrowing constraints is likely to be exacerbated. In this respect, the results of this paper underline the role of financial reforms in developing countries.

In recent years, a substantial amount of work has been carried out in developed economies on the effects of liquidity constraints on consumer behavior. Given the importance of considering capital market imperfections as pre-existing distortions in normative and positive economic analyses, it is surprising that liquidity constraints have received so little attention in the analysis of savings behavior in developing countries, where they seem to be a simple matter of commonsense observation.

APPENDIX Data Sources and VAR Estimation

This Appendix lists the sources used for the data set in this study and provides information on the coverage and main characteristics of the data.

The Appendix also provides a detailed presentation of the VAR equations as discussed in the text. In Tables 8-20, symbols are as in the main text.

Data Sources and Definitions

The data set for the present study has been constructed from all available international sources: National Accounts Statistics (NAS) (United Nations (1983)); International Financial Statistics Yearbook (IFS) (International Mone-tary Fund (1985)); Government Finance Statistics Yearbook (GFS) (International Monetary Fund (1985)); World Tables (WT) (World Bank (1983)); and Social Indicators of Development 1986 (SID) (World Bank (1986)); as well as national sources as needed.

As is well known, because of the unreliability and internal inconsistency of data and the varying methodology in different countries, data in the present sample may be subject to a wide margin of error. In addition, attention should be paid to conceptual differences in the various sources and their implications. These remarks apply, in particular, to the construction of the variable Z, (that is, per capita private disposable income in constant prices), to the estimation of the real interest rate, and to the definition of Gt. In particular:

Ct is per capita private final consumption expenditure, in constant prices (index: 1980 = 1). Sources were NAS, Tables 1.1 and 1.2, for consumption data; and SID for population data.

Gt is per capita government final consumption expenditure, in constant prices (index: 1980= 1). Sources were the same as for Ct above. According to the definitions in the NAS this item comprises compensation of employees and other purchases of goods and services. Capital expenditure is therefore disregarded here and, what is more important, as well as the long-debated question of the correct definition of current as opposed to capital expenditure.

Zt is per capita private disposable income, in constant prices (index: 1980 = 1), defined as gross national product (GNP), less consumption of fixed capital plus net transfers from abroad (when available), less tax revenue, plus subsidies and current transfers (when available), deflated by private final consumption implicit price index. Sources were NAS, Table 1.12, for GNP, consumption of fixed capital, and net current transfers from abroad; NAS, Table 1.4, and GFS, Summary Table and Table C, for tax revenue and subsidies and current transfers; and SID for population data. Notice that, although national disposable income (that is, GNP less consumption of fixed capital plus net current transfers from the rest of the world) is often reported in NAS, general government current receipts and disbursements, and, in particular, current tax revenue and subsidies and current transfers seldom are. In such cases, GFS data were used, thereby combining transactions recorded on a payments basis and flows measured and classified by their characteristics at the time of transaction (as in GFS), with transactions recorded on an accrual basis and flows measured and classified by future use or purpose (as in NAS).

The symbol rt denotes the real interest rate, defined as (1 + ri) = (1+qi)/(1 + Δpi), or as (1 + ri) = (1 +q*)(1 + Δei)/(1 + Δpi). Sources (apart from na-tional sources) were NAS, Tables 1.1 and 1.2, for the private final consumption deflator; and IFS for interest rates, exchange rates, and the consumer price index.

For each country in the six subsamples. Table 8 reports the time period considered, the average and the standard deviation of the ratio to GNP of gross private savings as derived by subtracting private final consumption from the measure of disposable income mentioned above, as well as the average and the standard deviation of the ratio to GNP of gross private savings as derived by adding the current account surplus to gross capital formation and subtracting government gross savings. The comparison of the two average ratios is a useful check of the quality of the approximation embodied in the definition used here of private disposable income. As is apparent from the table, in most cases the two averages match quite closely. However, substantial discrepancies arise in a few cases such as South Africa, the Islamic Republic of Iran, Jamaica, Greece, and Israel. Tracing the reasons for these discrepancies is, of course, far from easy. However, the discrepancies are likely to be due partly to the fact that the approximation to the concept of disposable income used here disregards interest payments on the public debt and, therefore, in some cases, substantially underestimates income.38 Unfortunately, there are very few countries for which statis-tics are available that allow for isolating the volume of interest payments on domestic public debt paid to the private sector. In addition, in cases such as South Africa, the difference partially derives also from a sizable statistical dis-crepancy that allows the reconciliation of the national accounting aggregates.

Table 8.

Coverage and Main Characteristics of the Data

article image
article image

Standard error.

Indicates an IDA (International Development Association)-designated country.

As Table 8 shows, the sample is characterized by a substantial variability across time and across countries, and, for the latter case, both between and within regional subsets.

VAR Estimation

The VAR equations estimated for the six geographical regions and the two alternative definitions of the real rate of return are reported in Tables 9-20. In all tables, symbols are as in the main text.

Table 9.

VAR Estimation: Sub-Saharan Africa

article image
Note: The variable d takes values of 1 in 1974 and of - 1 in 1975 for Swaziland. White’s (1980) standard errors are shown in parentheses. DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 10.

VAR Estimation: Sub-Saharan Africa

article image
Note: The variable d takes values of 1 in 1974 and of - 1 in 1975 for Swaziland. White’s (1980) standard errors are shown in parentheses. DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 11.

VAR Estimation; Middle East and North Africa

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 12.

VAR Estimation: Middle East and North Africa

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 13.

VAR Estimation: East and South Asia and the Pacific

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 14.

VAR Estimation: East and South Asia and the Pacific

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 15.

VAR Estimation: Southern Europe

article image
Note: The variables d1 and d2 take values of 1 in 1974 for both Cyprus and Portugal. White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 16.

VAR Estimation: Southern Europe

article image
Note: The variables d1 and d2 take values of 1 in 1974 for both Cyprus and Portugal. White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 17

VAR Estimation: centeral America and the Caribbean

article image
Note:White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 18.

VAR Estimation: Central America and the Caribbean

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 19.

VAR Estimation: South America

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.
Table 20.

VAR Estimation: South America

article image
Note: White’s (1980) standard errors are shown in parentheses; DW denotes the Durbin-Watson statistic; R2 is the coefficient of determination; σ^ denotes the standard error of the regression; x and x^ denote, respectively, actual and fixed values of the dependent variable.