The finding of Feldstein and Horioka (1980) that countries’ investment rates are highly correlated with their national saving rates has by now been confirmed by many subsequent studies. With few exceptions, industrial and developing countries with relatively high levels of or changes in average saving ratios over the past 25 years have also had relatively high levels or changes in investment ratios. There remains little agreement, however, on the lessons that can be drawn from this empirical regularity. This paper focuses on the inferences that can be drawn from data for 64 industrial and developing countries over the 1960-84 period. In Section I of the paper, international capital mobility is defined as the condition under which expected differential yields on physical capital in different countries are eliminated by net saving flows, as conventionally measured by current account imbalances.

Abstract

The finding of Feldstein and Horioka (1980) that countries’ investment rates are highly correlated with their national saving rates has by now been confirmed by many subsequent studies. With few exceptions, industrial and developing countries with relatively high levels of or changes in average saving ratios over the past 25 years have also had relatively high levels or changes in investment ratios. There remains little agreement, however, on the lessons that can be drawn from this empirical regularity. This paper focuses on the inferences that can be drawn from data for 64 industrial and developing countries over the 1960-84 period. In Section I of the paper, international capital mobility is defined as the condition under which expected differential yields on physical capital in different countries are eliminated by net saving flows, as conventionally measured by current account imbalances.

The finding of Feldstein and Horioka (1980) that countries’ investment rates are highly correlated with their national saving rates has by now been confirmed by many subsequent studies. With few exceptions, industrial and developing countries with relatively high levels of or changes in average saving ratios over the past 25 years have also had relatively high levels or changes in investment ratios. There remains little agreement, however, on the lessons that can be drawn from this empirical regularity. This paper focuses on the inferences that can be drawn from data for 64 industrial and developing countries over the 1960-84 period. In Section I of the paper, international capital mobility is defined as the condition under which expected differential yields on physical capital in different countries are eliminated by net saving flows, as conventionally measured by current account imbalances.

If capital were mobile in this sense, there would be no reason to predict that countries with relatively high saving ratios over a given period would also have relatively high investment ratios, since savings would be redistributed to countries that offered relatively high rates of return on physical capital. The finding that saving and investment ratios are highly correlated, however, is also consistent with several plausible alternatives to the view that capital is immobile. To organize systematic review of these alternatives, a very general framework is proposed in Section I that allows the statistical covariance of saving ratios and investment ratios to be decomposed into economically meaningful components. In general, the alternative hypotheses do not provide an adequate explanation of the high saving-investment correlations.

This leaves outstanding the apparent contradiction between the saving-investment correlations and the widely discussed integration of international financial capital. To evaluate this problem, an analytical framework is developed in Section II that shows that the failure of net saving flows to equalize international rates of return on physical capital is possible even with highly developed financial markets. It is widely accepted that integration of international goods markets cannot be evaluated by noting that the common currency prices of a subset of goods are equalized. Although there are many “traded” goods for which the “law of one price” holds strictly, the more interesting measure of goods-market integration—purchasing power parity among broad price indices—has in general failed to hold in recent years. This failure, of course, reflects substantial changes in the relative prices of traded and nontraded goods. In contrast, the evidence that expected yields are equalized for a subset of financial assets is widely interpreted as supporting the view that “capital” markets are highly integrated. The implicit assumption is that the observed yields for a subset of “traded” securities are fixed relative to the unobserved yields for claims on physical capital. Because there is little evidence that such an assumption is warranted, the role of current account imbalances in equalizing international rates of return on physical capital remains an interesting empirical issue, and directions for further research are suggested in Section III.

I. Endogeneity and the Interpretation of Saving-Investment Correlations

Many of the difficulties in interpreting saving-investment correlations arise because national saving and investment ratios are endogenous variables. Thus, the finding in cross-section studies that relatively high saving ratios are associated with relatively high investment ratios may not mean that residents accept lower yields on domestic investments, but that factors that generate relatively high saving ratios in a given country also generate relatively high investment ratios. The capital mobility issue can be conceptually disentangled from such problems by considering three conditions that must hold in a very simple framework before no correlation between saving and investment would be expected.

Condition 1. A country’s investment rate must depend on a “representative” national real rate of return r but not on other variables that are correlated with domestic saving.

Thus, while investment is endogenous it depends only on r, a variable that because of capital mobility is determined entirely outside the country in question. For convenience, we assume the relation to be linear:

I/Y=ahr+,(1)

where I is domestic fixed investment, Y is gross national or domestic product (GNP or GDP), h denotes a linear coefficient, and ∈ is an error term. The statistical support for the relationship between investment and the real interest rate has in fact always been weak. We know that, at a minimum, the error term ∈ must be large. But if investment is to be uncorrected with saving, it is crucial that this error term be purely random—that it be uncorrelated not only with the national rate of return, but also with national saving. The first subsection below explores this aspect of the endogeneity of saving and investment in some detail and concludes that this problem is so serious that little confidence can be placed in estimates that do not attempt to control for common determinants of national saving and investment.1

Condition 2. The foreign expected rate of return relevant for saving and investment, r*, must be determined exogenously.

In other words, the country in question must not be large enough in world financial markets to affect the world interest rate. This problem is also taken up in the following subsections, where it is argued that a more careful evaluation of existing empirical evidence suggests that exogenous determination of r* is not an important obstacle in interpreting the correlations.

Condition 3, The domestic expected real rate of return relevant for real investment and saving decisions must equal the foreign expected rate of return: r = r*.

If we think of the capital account balance, KA, as a function of the differential in returns,

KA=k(rr*),(2)

then the hypothesis is that the coefficient k is infinite. Of the three conditions, only this one can properly be associated with the phrase “international capital mobility” as traditionally understood. But, as we argue in Section II, financial capital mobility by itself is not sufficient to ensure that even this condition holds.

By using this framework, the covariance between investment and national saving can be decomposed into three parts:

cov(I/Y,NS/Y)=cov(,NS/Y)hcov(r*,NS/Y)hcov(rr*,NS/Y).(3)

The assumption that investment depends only on domestic interest rates would suggest that the first term on the right-hand side of equation (3) is zero. The exogeneity of the world interest rates would suggest that the second covariance is also zero. Finally, the assumption of perfect capital mobility would suggest that the third covariance is zero. If one of the three covariances fails to hold—if any one of the links is broken—then there is no reason to expect the investment rate to be uncorrelated with the saving rate.

Each of the three conditions in fact often fails to hold, and so we would not expect the covariance of national saving and investment to be zero, no matter what the degree of international capital mobility. To gauge the empirical importance of these failures, the paper examines the statistical relationship between national saving and investment in a sample that includes not only 14 industrial countries, but also 50 developing countries.2

Endogeneity of National Saving and Investment

One obvious version of the endogeneity problem that arises in time-series studies is the strongly procyclical nature of both saving and investment, even when expressed as shares of GNP. If an exogenous boom causes both to rise, we do not want to attribute the correlation to low capital mobility. For this reason, Feldstein and Horioka (1980) restricted their analysis to cross-section data, as did most who followed in their footsteps.3 But even in time-series studies, one can cyclically adjust the saving and investment data.4

An alternative version of the problem that is relevant even for cross-section studies is that the saving and investment rates both depend on the rate of growth of national income—as determined, for example, by population growth or productivity growth. This problem is particularly relevant if the sample includes both industrial countries and developing countries. One solution that has been applied is to add the rate of growth as a second explanatory variable. But the finding has been that holding the growth rate constant, like holding the business cycle constant, does not reduce the coefficient in the saving-investment regressions.5

The most popular version of the endogeneity critique is that governments react systematically to current account imbalances so as to offset these imbalances. For example, if the government reacts to a trade deficit induced by an increase in investment by cutting government expenditure or raising taxes, then national saving and investment will be correlated for reasons having nothing to do with capital mobility.6

It is important to realize how general the endogeneity argument is. Any economic variable, in addition to the cost of capital that influences the investment rate, will probably be correlated with the national saving rate. This is true not only for the level of income, population growth, and productivity growth, but also for energy shocks, real wages, strikes, and so forth.7 If factors, other than the cost of capital, that determine investment happen to be uncorrected with national saving, then there will be no econometric problem. (This is true even if government policy reacts systematically to current account imbalances.) But such a lack of correlation, with or without perfect capital mobility, is an absurdly strong condition. Because the difference between national saving and investment is identically equal to the current account, the lack of correlation would imply that the factor in question has the identical effect on the current account as on investment.

If national saving depended on only the national rate of return and other factors thought to be random (uncorrected not only with the rate of return but also with investment), one could invert the equation and regress national saving against investment. The hypothesis that foreign capital is in infinitely elastic supply at a given rate of return would imply a zero coefficient for national saving. This test would be equivalent (given the identity for national saving) to regressing the current account against investment. In that case the null hypothesis would be a unitary coefficient, implying that any exogenous changes in investment are fully financed by borrowing from abroad. This in fact is the equation run by Sachs (1981, 1983). But the idea that the rate of return is the only systematic determinant of both private and public saving has received even less support than the analogous assumption for investment. Clearly the right answer is that national saving, investment, and the current account are all endogenous, and that some covariance among these variables is to be expected.

The question of why there is so little independent variation among these variables, however, remains. Surely some of the factors that influence domestic saving do not have an equal and positive effect on investment. Yet it is shown in the next subsection that a positive covariance “near” to unity is evident for very different country groups and over very different time periods. Moreover, the country group with the lowest correlation is also the group that relies the least on market-determined capital flows. It follows that factors that generate high correlations between national saving and investment in the presence of capital mobility must be remarkably similar both across country groups and over long time periods.

It seems important, therefore, to confront the “endogeneity” problem directly through an instrumental variables approach. Indeed, Feldstein and Horioka (1980) used instrumental variables.8 Two candidates for instruments for national saving are the ratio of military expenditure to GNP and the dependency ratio (the population 15 years old or younger or 65 years or older, divided by the population of working age in between). The former is most immediately a determinant of the government budget deficit (government dissaving) and the latter of private saving. It is possible to think of ways that either instrumental variable could be endogenous; it is conceivable that military expenditure could be cut back in response to trade deficits,9 and that the age composition of the population could respond to the growth rate. But these two variables seem to meet the criteria for instruments better than most instrumental variables in macroeconomics. Regressions based on them are reported in the next subsection.

Empirical Analysis of the Endogeneity Problem

In this subsection we examine the relationship between national saving, defined to include both private and public sector saving and investment, for a sample comprising 14 industrial countries and 50 developing countries (see the Appendix). Our first objective is to consider whether the empirical results obtained for industrial countries (for example, in Feldstein and Horioka (1980) and Penati and Dooley (1984)) also extend to the developing countries, and whether the developing countries constitute a homogeneous group or differ according to the nature of the resource transfers they receive from foreign sources.

Table 1 summarizes some of the empirical results based on ordinary least-squares regressions that have been obtained in past studies for the saving and investment relationships for the industrial countries. Three preliminary conclusions are evident from these results. First, industrial countries that had relatively high rates of gross fixed investment also had relatively high rates of gross domestic saving. Second, examined over time, industrial countries that accumulated capital more rapidly in the most recent periods also experienced increases in saving as a share of GNP. Conflicting evidence was found, however, for the relationship between investment ratios and current account ratios. For example, Sachs (1981, 1983) showed a negative and significant slope coefficient in cross-section regressions for either the levels or changes in current account balances and investment ratios. Such a relationship would be consistent with a high degree of capital mobility. In contrast, Feldstein and Horioka (1980) and Penati and Dooley (1984) did not find a significant negative coefficient; as a result, this evidence is consistent with the view that changes in the propensity of residents of an industrial country to save or to invest were reflected in changes in the share of investment or saving in that country.

Table 1.

Saving, Investment, and Current Account Balances in Industrial Countries

article image
article image
Sources: The sources for regressions 1-4 and 6-8 are as follows: 1, Feldstein and Horioka (1980. p. 321); 2, Feldstein (1983. p. 135); 3 and 4, Sachs (1983, p. 105); 6, Feldstein and Horioka (1980. p. 327); 7, Sachs (1981, p. 250); and 8, Feldstein (1983, p. 144). The remaining regressions are from Penati and Dooley (1984).Note: Based on ordinary least-squares regressions. Parentheses enclosing a period of years indicate average values of the variables during that time. The delta (Δ) indicates the change from the average of the first period indicated in parentheses to the average of the second; R2 is the coefficient of determination; three dots indicate that the statistic of the parameter is not reported by the authors; / denotes gross domestic fixed investment, Ygross national or domestic product. CA the current account balance including official transfers, 5 gross national saving, and OIL the net imports of oil at constant prices; r-statistics are shown in parentheses below the coefficients; * indicates that the coefficient is significant at the 5 percent level.

To examine whether such relationships exist for developing as well as developed countries, our analysis first examines saving and investment behavior in these two groups of countries in two periods: 1960-73 and 1974-84. The first of these time periods represents the Bretton Woods era of fixed exchange rates. It might be expected that the degree of integration of markets for real saving and investment would be much higher in the second period, when the largest industrial countries removed their capital controls, the capital surpluses of the Organization of Petroleum Exporting Countries were recycled, and the Eurocurrency markets experienced rapid growth.

Table 2 presents ordinary least-squares regressions that relate saving and investment ratios (for both the levels and changes in the ratios) for the groups of industrial countries and developing countries and for all countries. In the regressions relating the levels of the saving and investment ratios, all coefficients are highly significant statistically. Moreover, the coefficients are higher for the industrial countries than for the developing countries, although we find that the difference in these coefficients is not statistically significant when the entire sample is pooled. This is the same finding as in Summers (1985): there is no sign of the greater ease that one would have expected the more open industrial countries to have in financing shortfalls in saving. Nor is there any sign of the expected increase in capital mobility after 1973. The coefficient for all countries rises from 0.46 to 0.61. This failure of the coefficient to decline over time is the same result found by Feldstein (1983) and Penati and Dooley (1984, pp. 9-10).10

Table 2.

Ordinary Least-Squares Regressions of the Investment Ratio (I/Y) Against the National Savings Ratio (NS/Y)

article image
Note: Parentheses enclosing a period of years indicate average values for the variables during that time. Standard errors are given in parentheses below coefficients; Δ denotes the change in a given variable, R¯2 is the adjusted coefficient of determination, and DV is a dummy variable with the value of unity (otherwise zero) when the country is an industrial country; * indicates significance at the 5 percent level, ** significance at the 1 percent level. See the Appendix for definitions and data sources.

These results are also supported by examining the relationship between the changes in the average saving and investment ratios over the periods 1960-73 and 1974-84. As indicated in Table 2, the slope coefficient relating the change in the average investment to income ratios to the change in the average saving to income ratios is positive and significant (at the 99 percent level) for both industrial and developing countries, although the coefficient for the industrial countries is considerably higher.

To cope with the government policy-reaction argument and the other forms of possible econometric endogeneity of national saving, we turn to instrumental variables regression. The two instrumental variables used in Table 3 are the ratio of military expenditure to GNP and the ratio of dependents to working-age population. The choice of instruments was dictated largely by the requirement that time series were available for the large sample of countries considered in this paper. Other equally plausible instruments may, of course, lead to different results.

Table 3.

Instrumental Variables Regressions: Military Expenditure/GNP and Dependency Ratio, Developing Versus Industrial Countries

article image
Note: Parentheses enclosing a period of years indicate average values for the variables during that time. Standard errors are given in parentheses below coefficients; Δ denotes the change in a given variable, and DV is a dummy variable with the value of unity (otherwise zero) when the country is an industrial country; * indicates significance at the 5 percent level, ** significance at the 1 percent level. See the Appendix for definitions and data sources.

The results are striking. In the case of the developing countries, the coefficients lose all statistical significance. Indeed, for the 1960-73 period the sign is not even positive. This finding would appear to suggest that the high coefficients reported in ordinary least-squares regressions by many authors are entirely attributable to problems of econometric endogeneity rather than to imperfect capital mobility, and that these problems are easily solved by the use of instrumental variables. The results for industrial countries, however, suggest the opposite conclusion. For these countries, the coefficient increases somewhat. These results only heighten the puzzling conflict between the regressions and the presumption of higher capital mobility among the industrial countries than among the developing countries.

A somewhat similar result emerges when changes in the average saving and investment ratios are used. For the developing countries, the coefficients associated with the change in the saving ratio are approximately the same size and positive in both the ordinary least-squares and instrumental variables equations, but the coefficient is not significantly different from zero in the instrumental variables case. For the industrial countries, however, the comparable saving coefficient rises quite sharply in value but again becomes insignificant.

One possible explanation for these different results for industrial and developing countries is that the 48 developing countries in the sample are too diverse to treat as having the same degree of capital mobility. In Tables 4 and 5, we focus on the distinction between a group of 21 market borrowers and 14 countries that depend primarily on official financing. (Fifteen countries are not classified by the Fund as either sort of borrower.) In the ordinary least-squares regressions (Table 4) the positive and significant coefficients for the market borrowers, the official borrowers, and the combined set of market and official borrowers that were evident in Table 2 are still evident. Moreover, the coefficients are higher after 1973 than before, and they are also higher for the market borrowers than the official borrowers.

The use of the instrumental variables technique (Table 5) has a much less important effect on the size and significance of the coefficients than it did when all the industrial countries were included together, especially for the post-1973 period.11 For the market borrowers in particular, the point estimate of the coefficient now even exceeds unity, as it did with the industrial countries in Table 3.

Table 4.

Ordinary Least-Squares Regressions of I/Y Against NS/Y: Market Borrowers Versus Official Borrowers

article image
Note: Parentheses enclosing a period of years indicate average values for the variables during that time. Standard errors are given in parentheses below coefficients; Δ denotes the change in a given variable, R¯2 is the adjusted coefficient of determination, and DO is a dummy variable with the value of unity (otherwise zero) when the country is an official borrower; * indicates significance at the 5 percent level, ** significance at the 1 percent level. See the Appendix for definitions and data sources.
Table 5.

Comparison of Regressions of I/Y Against NS/Y: Market Borrowers Versus Official Borrowers

article image
article image
Note: Parentheses enclosing a period of years indicate average values for the variables during that time. Standard errors are given in parentheses below coefficients; Δ denotes the change in a given variable, R¯2 is the adjusted coefficient of determination, and DO is a dummy variable with the value of unity (otherwise zero) when the country is an official borrower; * indicates significance at the 5 percent level, ** significance at the 1 percent level. See the Appendix for definitions and data sources.

To sum up the results, the instrumental variables estimation does little to clear up the mysteries of the saving-investment correlations; why they are so high in general, and why they are even higher for the industrial countries than for the developing countries, for the market borrowers than for the official borrowers, and for the period after 1973.

Country Size and the Saving-Investment Correlation

We noted at the outset three conditions, all of which taken together were sufficient if investment were to be uncorrelated with national saving: (1) investment depends only on the national real rate of return (and purely random factors); (2) the world real rate of return is exogenous; and (3) the domestic real rate of return equals the world real rate of return. It was argued that, even if condition 3 holds, condition 1 is very unlikely to hold because the gap between the domestic and world rates of return may be affected by endogenous domestic factors. The instrumental variables regressions are a way of attempting to deal with the econometric problems created by such endogeneity.

The failure of condition 2 is another possible econometric problem that has been pointed out by several authors. For a country large enough to affect world financial market conditions, a fall in national saving might drive up interest rates and crowd out investment everywhere in the world. It would be erroneous to conclude from the fact that domestic investment fell when domestic saving fell that there was low capital mobility if the rise in interest rates and fall in investment were as large abroad as in the domestic country. This argument is of particular interest here because it might seem to explain our findings in Section II, that the saving-investment coefficients appear to be higher for industrial countries (which of course tend to be larger in world financial markets) than for developing countries.

Among the Feldstein-Horioka (1980) critiques, Murphy (1984) focuses explicitly on the failure of the world interest rate to be exogenous, cov (r*, NS/Y) ≠ 0 in our equation (3).12 To address the effect of country size, Murphy divided his sample of 17 Organization for Economic Cooperation and Development countries in two and found that the 7 large countries had a higher coefficient on saving (0.98) than did the 10 small countries (0.59). He interpreted the results as supporting the claim that the high Feldstein-Horioka coefficients could be due to country size rather than to imperfect capital mobility.

The argument is particularly relevant in the case of time-series studies. Obstfeld (1986), for example, attributed the high correlation coefficient he got for a U.S. time series to the size of the United States in world markets. As a correction, Tobin (1983) proposed including, in the equation for any given country’s investment, the saving rates for all the other countries in the sample; he argued that, under the hypothesis of perfect capital mobility, a country’s saving will have no more effect on its own investment than on other countries’ investment.13 This technique would use up too many degrees of freedom. But a natural solution is to include the aggregated saving rates of the rest of the world. Frankel (1985) converted the U.S. saving rate to deviations from the world saving rate, converted the U.S. investment rate to deviations from the world investment rate, and found that the two variables remained highly correlated. The argument is that if capital were perfectly mobile, and crowding out of investment were due only to the large size of the United States in world capital markets, then there should be no effect of saving on the U.S. investment rate beyond the effect on the rest of the world’s investment rate. The rejection of this hypothesis suggests that the saving-investment correlation cannot be attributed to the large-country effect.

An insufficiently appreciated point is that, in cross-section studies, the endogenous foreign interest rate r* is not a problem to begin with. The statement that “a country’s saving rate affects r*, “ is a statement about alternative states of the world, as in time series. It is not possible that cross-section effects on the saving rate can be attributed to r*, for the simple reason that all countries share the same r*. In equation (3), cov(r*, NS/Y) is necessarily zero when r* is the same for every observation.

If one expressed national rates of saving and investment as deviations from world levels as in the time-series regression, only the constant term in the cross-section regression would change. To take the concrete example of the fall in the U.S. national saving rate in the 1980s, the large-country effect alone could in theory explain high real interest rates and a depressing effect on U.S. investment. But if real interest rates were equalized, the large-country effect alone could not explain a combination of low saving and investment rates in the United States together with simultaneous high saving and investment rates in Japan.

Harberger (1980) offers an argument to explain why the Feldstein-Horioka (1980) results are spurious that relies on the factor of country size, but in a way unrelated to the endogeneity of the world interest rate. The argument is that large countries tend to be more diverse than small countries (more like U.S. states and less like city blocks, in Harberger’s words). For this reason, multiple saving and investment shocks tend to cancel out, and there is proportionately less need in the aggregate for the country to borrow or lend to the rest of the world. When there is a drought in a small country, for example, it affects the whole economy; the country has to borrow from abroad. But when there is a drought in Kansas, there may be a good harvest in California, or a “high-tech” boom in Massachusetts, or an oil discovery in Alaska, so that there is proportionately less need for the United States to borrow from abroad. The argument is analogous to that in the literature of the early 1960s on optimum currency areas, which relied on the regularity that the larger and more economically diverse a region was, the less need it would have (proportionately) for trade in labor or goods with other regions.14

Harberger (1980) specifically addressed the issue of industrial versus developing countries:

The point to be borne in mind here is that the evidence of the Feldstein-Horioka paper was assembled from the OECD countries only. Casting the net wider would have surely thrown up indications of much greater divergence between saving and investment rates (p. 334).

Harberger’s point is a convincing explanation of why ratios of the current account to GNP (CA/Y)—-such as calculated by him, Feldstein and Horioka (1980), Feldstein (1983), Sachs (1981, pp. 233-37), Penati and Dooley (1984), Caprio and Howard (1984), and Summers (1985)— might be closer to zero for large countries than for small countries. But it does not seem in itself to explain why the saving-investment regression coefficient would be higher for large countries than for smalt countries. If the variance of the saving rate S/Y (or of the investment rate disturbances) is smaller for large countries because shocks cancel out, but changes in net capital inflow per unit change in the current account are the same, then the variance of the current account will also be smaller, but the regression coefficient need not be.

Let

I/Y=α+β(NS/Y)+.(4)

Here, β depends on parameters, such as the degree of capital mobility and the sensitivity of investment to the interest rate, which we will assume are the same for different countries. Then,

CA/Y=NS/YI/Y=α+(1β)NS/Y,(5)

and var(CA/Y) = (1 - β)2var(NS/Y) + var (∈) + 2(1 - β)cov(NS/Y, ∈). The last term on the right-hand side is zero in large samples, in the event (extremely unlikely without the use of instrumental variables, as we saw earlier) that the error term is well behaved. If var (S/Y) and var(∈) are smaller for large countries, then var (CA/Y) will be smaller as well, as Harberger claims. Nevertheless, a regression coefficient will be estimated as

βcov(I/Y,NS/Y)var(NS/Y)=β+cov(NS/Y,)var(NS/Y),(6)

where again the last term is zero in large samples in the event that the error term is well behaved. The important point is that we should expect B to be no higher among large countries than among small countries (but the standard error of B in small samples could be affected one way or the other).

Murphy (1984) pursued Harberger’s suggestion that one should find more apparent capital mobility for smaller than for large units. He regressed the investment rate against the saving rate for a cross section of 143 U.S. corporations and found a significant coefficient. He argued that, since the corporate capital market is “by almost all standards, a well-integrated capital market with a high degree of capital mobility” (p. 335), the correlation test must not be correct. But there is an element of circularity to this argument. On the one hand, the proposed explanation for the spurious Feldstein-Horioka results—an endogenous market-wide interest rate—is not relevant for the cross section of corporations. On the other hand, there are perfectly good theoretical and empirical reasons for thinking that the liabilities of corporations are not perfect substitutes for each other in investors’ portfolios and that corporations cannot borrow unlimited amounts at the going interest rate. We are left doubting the perfect integration of the corporate capital market more than the reliability of the test.

II. Financial and Real Capital Mobility

In our view the evidence that levels and changes in national saving and investment ratios move together stands up to the empirical issues raised in the previous section. In terms of the three conditions laid out at the beginning of this paper, it is the failure of the third, r = r*, that remains the most likely explanation for the positive covariance of national saving and investment. Many observers continue to resist the implications of this evidence because it seems so clearly at odds with the apparent integration of financial capital markets, especially those of developed countries.

If by capital mobility one means the tendency of investors to equalize expected rates of return on a subset of liquid, short-term, default-free, financial assets denominated in different currencies or issued by residents of different countries, then there can be little doubt that capital is mobile among the major industrial countries.15 This definition of capital mobility, however, is of limited value. It is analogous to measuring the degree of integration of international goods markets by noting that prices (measured in a common currency) are equalized for a subset of goods. There certainly are many agricultural and mineral commodities for which this condition holds quite strictly. It is clear, however, that this condition tells us nothing about the tendency of prices of goods in general to be equalized across countries. In fact, this more interesting purchasing-power-parity measure of “goods mobility” has failed to hold in recent years. The key to this more general condition is that within countries the relative prices of goods change by substantial amounts and with no apparent tendency to return to their original levels.

An analogous argument can be made in the context of international markets for securities. That some financial instruments issued by residents of different countries appear to be perfect substitutes does not mean that the financial markets are fully integrated. As is also true for goods markets, a useful measure of capital mobility would require that “traded” and “nontraded” financial assets be perfect substitutes in residents’ portfolios so that the relative yields on these assets within countries are largely independent of international influences.

The existence of “nontraded” securities is, of course, not obvious. It can be argued, however, that the possibility of ex post taxation of some types of international positions acts as a barrier to trade in such assets. Moreover, it is possible that the net indebtedness of a country is an important determinant in its incentive to impose such taxes.

The idea that net claims on a given country might be important to wealth holders is difficult to model. There is no aggregate asset called “net claims” that we can identify, nor can we directly measure the yield necessary to induce wealth holders to hold this net position. In most cases net claims on a given country are a small difference between two large numbers, gross claims and gross liabilities. Thus, although it may be clear that a risk exists when a country’s net debt to the rest of the world grows, it is not clear who holds that risk—that is, how a loss would be allocated among holders of various financial claims on that country.

The government, for example, may perceive different costs in tarnishing the reputation of various forms of investments. Its own securities might be an example of an asset unlikely to be taxed, particularly if these are held by foreign governments. equity claims, however, might be relatively easy to identify as being held by nonresidents; because they impart control of domestic firms, equity claims may constitute a relatively attractive tax base for political reasons.

If assets such as treasury securities and equity claims are imperfect substitutes within countries, the structure of yields between countries and within countries will reflect a complex set of expectations concerning the likely incidence of an ex post tax on nonresident claims as well as “arbitrage” conditions that will determine how and by whom such risks are borne.

It follows that arbitrage among a select set of internationally traded “safe” financial assets need not imply that yields on more vulnerable nontraded assets are also equalized. What is necessary is that differentials that might arise within a country are not closely arbitraged. That is, the yields on treasury securities and other tradable securities would have to diverge from yields on equities and other nontradable assets. The analogy in goods markets would be that prices of traded goods are equalized across countries, but that the relative prices of traded and nontraded goods within countries will change in response to economic conditions.

This framework might be clarified by illustrating the financial and real links between and within two very simple economies. Financial assets consist of claims on physical capital, E, interest-bearing claims on the government, B, and fiat money, M. The yield on equity claims is the marginal revenue product of the capital stock, rp. Interest-bearing government bonds yield a nominal rate of return, rg, and fiat money, which is also a liability of the governments, yields no nominal return:

Md=Md(r¯p,r¯g)(7)
Bd=Bd(r¯p,r+g)(8)
Ed=Ed(r+p,rg).(9)

For the links between the two countries’ financial systems, it is assumed that government bonds are traded securities and that equity claims are nontraded securities. As with goods and services, the terms “traded” and “nontraded” securities are an analytical convenience that represents a continuum of assets. For simplicity it is assumed that government-issued securities are perfect substitutes across countries, so that

rg=rg*+E(e),(10)

where E(e) is the expected depreciation of the domestic currency. Because equity claims are not traded, however, the links between rp and rp* are only indirect. The two currencies are tradable but are assumed to be dominated by government securities in the portfolios of nonresidents.16 These financial market assumptions are illustrated as follows:

Note that the arbitrage between the two capital stocks is limited both across, because of country risk, and up and down, because of our assumptions about substitution parameters between equities and government bonds.

The links between the financial sector and real sector within each country are as follows. Real investment is defined as transforming current output into a capital good that has an “own” rate of return in terms of future output. Investment (/) will be positive when the present value of the expected future output exceeds the cost of the capital. As discussed above, the discount rate (rp) need not be equal to the government security rate because domestic savers might require a variable differential in order to hold willingly existing government debt and claims on the capital stock:

I=I(rp).(11)

Saving (5) in each country is a function of two independent rates of return, the two government bond yields (which are assumed to be equal because of arbitrage across countries) and the yield on the domestic capital stock:

S=S(rp,rg).(12)

The real sector links between countries are derived from the above conditions and the balance of payments constraint. That is, a net excess demand for the traded security must be identically equal to the differences between output and absorption (or public plus private saving and investment) in each country. The mechanism that ensures this equality is left in the background. It may or may not involve income, prices, and real exchange rate changes, which in this discussion are assumed to remain unchanged.

The essential features of this framework are illustrated in Figure 1. Because these relations are drawn with respect to returns on traded securities, the yield on nontraded securities, rp is a parameter in the saving and investment functions. Because government bonds are perfect substitutes across countries, rg and rg* are always equal. In the initial position, domestic saving for both countries is set equal to domestic investment plus the domestic fiscal deficit (G) at rg0 assumption.

Figure 1.
Figure 1.

Shift in Home-Country Fiscal Deficit Schedule

Citation: IMF Staff Papers 1987, 002; 10.5089/9781451972931.024.A004

It is evident that shifts in the I, I* or S, S* functions, or a change in either country’s fiscal deficit, will result in some net trade of government bonds and a current account imbalance in this framework. But there is no presumption that this trade in “safe” securities will be sufficient to equalize rp and rp* In general, the disturbance in this model will be dissipated by both a change in the structure of yields, saving, and investment behavior within countries, and in the net trade of goods and securities between countries. The relative importance of these equilibrating mechanisms depends on the degree of segmentation of financial markets, the response of saving and investment to the two financial yields, and the nature of the shock to the system.

Suppose in this system the domestic fiscal deficit increases (because of a bond-financed increase in government expenditures), shifting the I + G curve rightward to I + G (Figure 1). In autarky, rg would be bid up and, if rp were bid up only slightly so that shifts in I and S were negligible, rg0 would rise to rg1 If we now open the economy, and assume that rp* is about unchanged, rg* will also be bid up, and the new equilibrium will be at rg2=rg*2. At this point the yields of traded securities will have risen sufficiently to increase S + S*. so that the domestic fiscal deficit is covered. This increase in yields will generate a current account deficit in the home country, ab, equal to the difference between domestic saving and I + G, and a current account surplus, cd, in the foreign country. There is an equivalent net trade of government securities.

If we assume that the financial markets are highly integrated within each country, a rise in rg = rg* will tend also to raise rp and rp*. In this case the S and S* curves in Figure 1 would shift to the right, and the I + G and I* + G* curves would shift to the left, as private investment is crowded out. If the two countries were identical, the rise in rg = rg* would be less, but we would observe the same current account deficit following a fiscal deficit as described above.

A shift in the investment schedule or the saving schedule would differ in at least one important respect from the fiscal deficit shock discussed above. Assume that the private investment function shifts to the right by the same amount as the fiscal deficit (as discussed above), so that I + G shifts to I + G (Figure 2). The effects of such a shift appear to be similar as described above in that rg0=rg*0 would rise until the level of saving in both countries covered the increase in investment at rg1,rg*1. In this case, however, the shift in the domestic investment function would tend to raise rp directly, and this rise would shift I + G to I′′ + G. Moreover, the rise in rp would cause SS in Figure 2 to shift to the right to SS For these reasons an ex ante shift in the domestic investment schedule would have a smaller effect on the traded-securities market, as shown by the broken line in Figure 2. The initial disequilibrium in the domestic traded-securities market would be mitigated by a rise in rp. It follows that the size of the shock transmitted to the other country is smaller compared with a domestic fiscal deficit if the domestic financial maket is poorly integrated. For this reason the current account imbalance would also be smaller.

Figure 2.
Figure 2.

Shift in Home-Country Investment Schedule

Citation: IMF Staff Papers 1987, 002; 10.5089/9781451972931.024.A004

In general, it appears that a more complete description of the links among financial markets within and between countries might help to explain the correlations between domestic saving and domestic investment. The very simple model proposed above suggests that these correlations should break down in cases where the dominant shocks to the system are unusually large fiscal deficits that are uncorrelated across countries. In the absence of such conditions, differences or changes in domestic saving and investment functions might have limited impact on current account imbalances.

II. Concluding Remarks

The evidence brought together in this paper suggests that a close association between national saving and national investment is a robust empirical regularity. This finding casts considerable doubt on the view that national markets for physical capital are highly integrated. The positive correlations between levels and changes in national saving rates and investment rates—-which are apparent both for industrial countries and developing countries, and which have been higher in recent years compared with earlier periods—stand up to a variety of econometric objections. The only data for which the empirical regularity is not apparent includes developing countries that depend primarily on aid to finance their current account imbalances.

That national markets for some types of financial capital are integrated may be irrelevant in evaluating the degree of integration of national markets for physical capital. The tendency for expected returns on liquid, default-free financial assets to be equalized does not imply that expected returns on physical capital are also equalized.

We do not know why the apparent isolation of national markets for physical capital has persisted in the face of substantial expansion of trade in goods and services and in financial capital. Further research into these matters might focus on the impediments to net transfers of real savings to “foreign” political jurisdictions. A better understanding of such factors might suggest policy measures that would encourage more productive use of world saving.

APPENDIX

Data Sources

GDP at market prices (Y), gross domestic investment (I), and gross domestic saving (NS) are conventional concepts of national income drawn from EPDNA data files of the World Bank, Economic Analysis and Projections Department. The dependency ratio is the ratio of dependent population (persons under 16 years old and over 64 years old) to working-age population (ages 15 to 64 years), drawn from the same source. For the ratio of military expenditure to GNP, military expenditures were taken from data files of the U.S. Arms Control and Disarmament Agency. The definitions for official borrowers, market borrowers, and combined borrowers—as well as lists of countries in each category—can be found on pages 173-74 of the International Monetary Fund’s World Economic Outlook (Washington, April 1986).

REFERENCES

  • Caprio, Gerard, Jr., and David H. Howard, “Domestic Saving, Current Accounts, and International Capital Mobility,” International Finance Discussion Paper 244 (Washington: Board of Governors of the Federal Reserve System, June 1984).

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cumby, Robert E., and Frederick S. Mishkin, “The International Linkage of Real Interest Rates: The European-U.S. Connection,” NBER Working Paper 1423 (Cambridge, Massachusetts: National Bureau of Economic Research, August 1984).

    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin, “Domestic Saving and International Capital Movements in the Long Run and the Short Run,” European Economic Review (Amsterdam), Vol. 21 (March-April 1983), pp. 12951.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin, and Charles Horioka, “Domestic Saving and International Capital Flows,” Economic Journal (London), Vol. 90 (June 1980), pp. 31429.

    • Search Google Scholar
    • Export Citation
  • Fieleke, Norman, “National Saving and International Investment,” in Saving and Government Policy, Federal Reserve Bank of Boston Conference Series, No. 25 (Boston, 1982).

    • Search Google Scholar
    • Export Citation
  • Frankel, Jeffrey A., “International Capital Mobility and Crowding Out in the U.S. Economy: Imperfect Integration of Financial Markets or of Goods Markets?,” NBER Working Paper 1773 (Cambridge, Massachusetts: National Bureau of Economic Research, December 1985).

    • Search Google Scholar
    • Export Citation
  • Harberger, Arnold C, “Vignettes on the World Capital Market,” American Economic Review (Nashville, Tennessee), Vol. 70 (May 1980), pp. 33137.

    • Search Google Scholar
    • Export Citation
  • McKinnon, Ronald I., “Optimum Currency Areas,” American Economic Review (Nashville, Tennessee), Vol. 53 (September 1963), pp. 71725.

    • Search Google Scholar
    • Export Citation
  • Mundell, Robert A., “A Theory of Optimum Currency Areas,” American Economic Review (Nashville, Tennessee), Vol. 51 (September 1961), pp. 65765.

    • Search Google Scholar
    • Export Citation
  • Murphy, Robert G., “Capital Mobility and the Relationship Between Saving and Invstment in OECD Countries,” Journal of International Money and Finance (Guildford, England), Vol. 3 (1984), pp. 32742.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Obstfeld, Maurice, “Capital Mobility in the World Economy: Theory and Measurement,” Carnegie-Rochester Conference Series on Public Policy (Amsterdam), Vol. 24 (Spring 1986), pp. 10514.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Penati, Alessandro, and Michael Dooley, “Current Account Imbalances and Capital Formation in Industrial Countries, 1949-81,” Staff Papers, International Monetary Fund (Washington), Vol. 31 (March 1984), pp. 124.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sachs, Jeffrey D., “The Current Account and Macroeconomic Adjustment in the 1970s,” Brookings Papers on Economic Activity: 1 (1981), The Brookings Institution (Washington), pp. 20168.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sachs, Jeffrey D., “Aspects of the Current Account Behavior of OECD Economies,” in Recent Issues in the Theory of Flexible Exchange Rates, ed. by E. Claasen and P. Salin (Amsterdam and New York: North-Holland, 1983).

    • Search Google Scholar
    • Export Citation
  • Summers, Lawrence H., “Tax Policy and International Competitiveness,” paper presented at the Conference on International Aspects of Fiscal Policies, Cambridge, Massachusetts, December 13-14, 1985 (Cambridge, Massachusetts: National Bureau of Economic Research, December 1985).

    • Search Google Scholar
    • Export Citation
  • Tobin, James, “‘Domestic Saving and International Capital Movements in the Long Run and the Short Run’ by M. Feldstein: Comment,” European Economic Review (Amsterdam), Vol. 21 (March-April 1983), pp. 15356.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Westphal, Uwe, “ ‘Domestic Saving and International Capital Movements in the Long Run and the Short Run’ by M. Feldstein: Comment,” European Economic Review (Amsterdam), Vol. 21 (March-April 1983), pp. 15759.

    • Crossref
    • Search Google Scholar
    • Export Citation
*

Mr. Dooley, Chief of the External Adjustment Division in the Research Department of the Fund, is a graduate of Duquesne University, the University of Delaware, and the Pennsylvania State University.

Mr. Frankel is Professor of Economics at the University of California, Berkeley, and is a graduate of Swarthmore College and the Massachusetts Institute of Technology. This paper was written while he was a consultant in the Research Department.

Mr. Mathieson is Chief of the Financial Studies Division in the Research Department. Educated at the University of Illinois and Stanford University, he was on the staff of Columbia University before coming to the Fund.

1

A similar problem arises if measurement errors for investment are incorporated in measures of national saving. In that case the covariance of ∈ and national saving would clearly not be zero. The instrumental variables approach is used in Section II to minimize these difficulties.

2

Summers (1985) and Fieleke (1982) also include developing countries in their samples.

3

Other cross-section studies include Fieleke (1982), Feldstein (1983), Penati and Dooley (1984), Murphy (1984), Caprio and Howard (1984), and Summers (1985).

4

Sachs (1981) included a GNP gap variable in his regressions. Frankel (1985) tried two approaches: decade averages on a ten-year time sample of U.S. data, and cyclically adjusted annual saving and investment rates on shorter postwar time samples. A third time-series study is Obstfeld (1986).

5

Summers (1985) argued, for developing countries in particular, that the influence of the growth rate on the other two variables explains the saving-investment correlation. Obstfeld (1986) makes the argument carefully, in the context of OECD countries. But Summers (1985, p. 22) added the rates of population growth and GNP growth to his regressions and found no effect on the saving coefficient.

6

The ‘“policy-reaction” argument has been made by Fieleke (1982), Tobin (1983). Westphal (1983), Caprio and Howard (1984), and Summers (1985). Summers called it the “maintained external balance” hypothesis.

7

Many of these factors would probably bias the correlation upward. But some would go in the direction of a negative correlation. If a country discovers oil, investment should go up, but saving should go down. Similarly, an investment tax credit should raise investment but lower the budget surplus and, therefore, national saving.

8

Feldstein and Horioka’s four instrumental variables were the ratio of retirees over the age of 65 to the population aged 20-65, the ratio of younger dependents to the working-age population, the labor force participation rate of older men, and the benefit-earnings “replacement ratio” under social security.

9

Total government expenditure may not be a good enough instrument, as Summers (1985) found, because under the policy-reaction argument it is endogenous.

10

Obstfeld (1986) and Frankel (1985) obtained the same result for the United States.

11

The ordinary least-squares regressions in Tables 4 and 5 differ for official borrowers and the combined set of market and official borrowers. One official borrower was excluded in Table 5 because of lack of data for some of the instrumental variables; since the samples are identical, the results for the market borrowers in Table 4 are not repeated in Table 5.

12

Harberger (1980), Tobin (1983), and Obstfeld (1986) have also criticized Feldstein and Horioka on the grounds of the “large-country” problem.

13

Tobin (1983) acknowledged that the solution is relevant only for time-series, not cross-section, studies, but he seems to believe that the problem itself is relevant and serious even in cross-section studies.

14

It followed from this argument that, if two regions joined together, the aggregate unit would be less open than either region taken individually. Mundell (1961) put the argument in terms of labor mobility, and McKinnon (1963) couched it in terms of openness to trade.

15

But note that, even if capital is sufficiently mobile to equalize internationally expected rates of return on two assets, the equalization will take place in terms of any common currency or other numeraire, not in terms of the countries’ respective goods. Thus, real interest parity, which is the condition relevant for saving and investment, need not hold unless investors anticipate no changes in real exchange rates. See Frankel (1985) for an elaboration of this point.

16

In terms of the literature, we are thus assuming “perfect capital mobility” but no currency substitution. Note, however, that the addition of nontraded securities to the model will alter the “standard” interpretation of capital mobility as defined by equation (10).

IMF Staff papers: Volume 34 No. 3
Author: International Monetary Fund. Research Dept.