Article

Current Account Imbalances and Capital Formation in Industrial Countries, 1949–81

Author(s):
International Monetary Fund. Research Dept.
Published Date:
January 1984
Share
  • ShareShare
Show Summary Details

The integration of financial markets among industrial countries in the 1970s and 1980s is often cited as indirect evidence that changes in net foreign asset positions of countries (which are the sum of net private and official capital flows over a given period) have become more sensitive to differential rates of return on investment in physical capital across countries. An increase in the sensitivity of net capital flows would imply that national savings have been increasingly allocated among countries through current account imbalances in a manner that more nearly equalizes rates of return. Increased sensitivity would also imply that foreign savings have played an important role in the capital accumulation process of the industrial world.1

Assumptions about the nature of net capital flows are crucial in evaluating economic policy in an open economy. Because such flows can substantially modify the relationship between domestic savings and investment, they are central to the predictions of a wide range of models of open economies. For example, if changes in net foreign assets are extremely responsive to cross-country differences in rates of return, a small country can experience a prolonged budget deficit without reducing the rate of domestic private capital accumulation or increasing domestic private savings. The crowding out of domestic investment suggested by many closed-economy models would not occur, because any tendency for domestic interest rates to rise would attract an inflow of foreign savings through a deficit in the current account of the balance of payments. If the country were large, this would imply a noticeably lesser amount of savings available for investments in foreign countries and, ultimately, a rise in world real interest rates and a slowdown in the growth rate of world output. Furthermore, policies that stimulated domestic savings would not necessarily increase the rate of accumulation of the domestic capital stock.

Despite its importance, the behavior of net capital flows among industrial countries has not been the subject of extensive empirical testing. This may be due to the difficulty in measuring expected real rates of return on capital in various countries, but it is also possible that the assumption has been considered obviously appropriate and not in need of formal testing. However, two recent empirical studies by Feldstein and Horioka (1980) and Feldstein (1983) have concluded that the assumption that net capital flows tend to equalize rates of return on physical capital is at least debatable. This paper re-examines and extends some of the studies on this topic. The results presented suggest that there is little support for the assumption that changes in the net foreign assets of industrial countries are sensitive to cross-country differences in rates of return, at least for periods longer than five years. There are two important implications of these findings. First, disturbances to domestic propensities to save and invest are not systematically accommodated by international transfers of re-sources. Second, the integration of financial markets among industrial countries and the associated rapid growth in two-way trade in financial assets do not necessarily imply that net trade in financial assets has played a measurable role in equalizing rates of return on physical capital among countries.

Section I reviews selected contributions in this area. From this review, it appears that an important unresolved issue is whether changes in net foreign assets of industrial countries have become substantially more sensitive to differential rates of return in recent years. Section II presents evidence suggesting that these changes were no more sensitive in 1971–81 than they had been in the 1950s. Section III focuses on experience from 1970 to the present.

In contrast to recent empirical studies—Sachs (1981, 1983)—this study shows that no systematic relationship between current account imbalances and investment rates is apparent during this period. Section IV presents some concluding remarks.

I. A Short Survey of the Literature

If changes in net foreign assets were sensitive to cross-country differences in real rates of return, the location of investment in physical capital would be unrelated to the location of savings. In such a world, differences among investment rates in various countries would depend on the expected real rates of return on their capital stocks, whereas differences among saving rates would depend upon demographic and cultural elements and on the distribution of income. New investment opportunities in a country would not need to be financed by increases in domestic savings, because every country would face an elastic supply of funds from abroad.

These ideas were utilized by Feldstein and Horioka (1980) to test whether changes in net foreign assets of industrial countries, which are the counterpart of current account imbalances, respond to changes in expected real rates of return. The hypothesis tested was that the share of income saved by individual countries is unrelated to the share of output devoted to gross investment—that is, to increasing and maintaining the capital stock. Feldstein and Horioka took a sample of 15 industrial countries and calculated the average gross domestic saving and gross fixed investment rates for each country during 1960–74. Then they regressed the cross-section of the average investment rates on a constant and the cross-section of the average saving rates. They found that the slope coefficient was 0.88 and that the R2 was 0.91. The regression is reported in row 1 of Table 1. For a slightly larger sample of countries, Feldstein (1983) presented similar results for various sample periods, including 1975–79 (see row 2 of Table 1). These data confirmed Feldstein and Horioka’s earlier findings: industrial countries that had relatively high rates of gross fixed investment also had relatively high gross domestic saving rates.

Table 1.Industrial Countries: Savings, Investments, and Current Account Balances1
Number of Industrial

Countries
Sample

Period
Regression

Equation
R2
1.152(1960–74)(I/Y)=0.035(1.94)*+0.88(12.6)*(S/Y)0.91
2.172(1975–79)(I/Y)=0.046(1.09)+0.86(1.89)*(S/Y)0.57
3.142(1971–79)(CA/Y)=0.039(1.49)0.20(1.89)*(I/Y)0.21
4.142(1971–79)(CA/Y)=0.03(1.27)0.20(1.9)*(I/Y)+0.28(1.0)OIL0.28
5.19(1960–74)(CA/Y)=0.002(0.08)0.02(0.27)(I/Y)0.01
6.152(1960–69)–(1970–74)Δ(I/Y)=0.002(0.50)+0.72Δ(4.50)*(S/Y)0.52
7.152(1968–73)–(1974–79)Δ(CA/Y)=0.64Δ(6.2)*(I/Y)0.72
8.172(1968–73)–(1974–80)Δ(I/Y)=+1.04()Δ(S/Y)(…)
9.19(1968–73)–(1974–79)Δ(CA/Y)=0.018(4.53)*0.55Δ(3.6)*(I/Y)0.43
10.19(1968–73)–(1974–80)Δ(CA/Y)=0.018(4.51)*0.39Δ(2.48)*(I/Y)0.27
11.19(1968–73)–(1974–79)Δ(I/Y)=0.014(1.95)*+0.83Δ(3.58)*(S/Y)0.44
12.19(1968–73)–(1974–80)Δ(I/Y)=0.013(2.15)*+0.81Δ(4.36)*(S/Y)0.53

Parentheses enclosing a period of years indicates the average value of the variables during that period. The delta (A) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate that the statistic of the parameter is not reported by the authors. I denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

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.

Parentheses enclosing a period of years indicates the average value of the variables during that period. The delta (A) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate that the statistic of the parameter is not reported by the authors. I denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

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.

Feldstein and Horioka also calculated the average saving and investment rates for the 15 countries in two subperiods and tested whether the cross section of the changes in the average saving rates from one subperiod to the other was correlated with the cross section of the changes in investment rates between the subperiods. This regression, which is reported in row 6 of Table 1, suggests that the industrial countries that accumulated capital stock more rapidly in the second subperiod also experienced increases in savings as a share of gross national product (GNP). This evidence led Feldstein and Horioka (1980, p. 317) to conclude that “the statistical estimates indicate that nearly all of the incremental saving remains in the country of origin. These results are quite incompatible with the assumption of complete arbitrage in a perfect world capital market.”2

In two recent papers, Sachs (1981, 1983) has analyzed the relationship between investment and the current account balance. The aim of these papers was to show that changes in domestic investment opportunities, rather than changes in the price of oil, were the predominant cause of current account imbalances among industrial countries. Sachs computed the average current account-to-GNP ratios for 14 industrial countries and their average gross fixed investment rates for 1971–79. He then regressed the cross section of current account ratios on the cross section of investment ratios and found that the slope coefficient was negative and significantly different from zero. In addition, he found a significant inverse relationship between cross sections of changes in average current account balances and changes in investment rates over various sample periods from 1968 through 1979. Based on this evidence, Sachs (1983, p. 106) concluded that “since the coefficient on investment is -0.65, a 1 percent rise in the investment rate [between 1961–70 and 1971–79] was financed on average 0.65 percent by foreign capital inflows, and only 0.35 percent by gross national savings.”3

The contradiction between Sachs’s and Feldstein and Horioka’s conclusions is obvious and is well illustrated by rows 7 and 8 in Table 1, which are taken from Sachs (1981) and Feldstein (1983). However, it is possible that the two regressions are not strictly comparable because they differ both in the cross section of countries included and in the sample period. Therefore, this study re-estimated them using data for 19 industrial countries and the same time period. As rows 9 through 12 of Table 1 show, the estimated parameters changed very little compared with the results reported in the studies by Sachs, Feldstein and Horioka, and Feldstein. Because it is clear that Feldstein and Horioka’s and Sachs’s interpretations of their results cannot both be correct, one of their equations cannot be viewed as truly structural—that is, one of the two sets of correlations found in these empirical studies cannot be used to draw inferences about the behavior of economic agents.

II. Have Current Account Imbalances Become More Sensitive to Investment Changes in the 1970s?

A plausible interpretation of the data that would reconcile the conclusions of Sachs and Feldstein and Horioka is that national capital markets were not integrated during the 1950s and 1960s but were liberalized in the 1970s, so that in a world of imperfect but increasing capital mobility, there would still be a significant relationship between domestic savings and investment ratios. This correlation, however, should have declined over time as increasing integration of capital markets tended to break down the relationship between domestic savings and domestic investment. As a corollary, the inverse correlation between current account balances and investment rates should have become more apparent over time.

This possibility was advanced by Harberger (1980) and Feldstein (1983). Feldstein (1983) argued that capital market integration was enhanced in the 1970s by various measures, an example being the elimination of the interest rate equalization tax in the United States. As a result, the industrial countries may have moved toward a world of equal rates of return even though domestic savings patterns still largely determined the accumulation of domestic capital stock. Feldstein supported this hypothesis by pointing out that the R2 in the regression of the cross section of investment rates on that of saving rates sharply declines if the sample period that is used to calculate the average rates includes the years after 1974 (see rows 1 and 2 of Table 1). He also pointed out that the negative correlation between current account balances and investment rates that Sachs found for the 1970s had not been apparent for the 1960s. This can be seen by comparing rows 3 and 5 in Table 1.

The conjecture that the development of international credit markets has tended to break the linkage between domestic savings and investment is certainly appealing. The growth of Eurocurrency and Eurobond markets, as well as the expansion of international banking and other types of financial intermediation among residents of different countries, has been a prominent institutional change during the 1970s. However, it is not necessarily true that a large volume of two-way trade in financial assets is associated with net trade in financial assets. But it is the net trade, together with the associated net trade in goods and services, that allows domestic investment to diverge from domestic savings.

This section tests the hypothesis that the sensitivity of changes in net foreign assets of industrial countries to differential rates of return increased from the 1950s to the 1970s. Because the 1950s are widely viewed as a decade in which the industrial countries behaved as insular economies, the test provides a good benchmark for evaluating the degree of integration of the national markets for physical capital in the 1970s.4 In practice, the period 1949–59, which was characterized by extensive trade barriers and exchange controls on current account transactions, was selected to represent the 1950s. As for the 1970s, two periods were selected. The first period covered 1971–81, so that its initial year coincided with the first devaluation of the dollar and the gradual move toward the managed float. The second period covered 1974–81, the years that followed the first oil shock.

Table 2 shows that the industrial world as a whole increased its investment rate by nearly 2 percentage points between the 1950s and the 1970s. The increase in the median investment rate was even larger, moving from 20.3 percent to nearly 23 percent. Thus, there is enough variation in the data to make the comparison between the two decades significant.

Table 2.Nineteen Industrial Countries: Comparison of Variances, 1950s and 1970s
1949–591974–811971–81
Fixed investment/GNP
Overall ratio for industrial countries0.1990.2160.216
Median0.2030.2200.228
Variance0.001142 = S10.001644 = S20.001572 = S3
Savings/GNP
Overall ratio for industrial countries0.2150.2200.223
Median0.2270.2230.225
Variance0.001673 = S40.001345 = S50.001394 = S6
Current account/GNP
Overall ratio for industrial countries0.005–0.003–0.002
Median0.003–0.019–0.011
Variance0.000338 = S70.000563 = S80.000430 = S9
Current account/fixed investment
Variance0.009325 = S100.011347 = S110.008821 = S12

Under the alternative hypothesis, the value of the F-statistics should exceed the critical value of the F-distribution, which IS F. 95 (18,18) = 2.22.

Test of Equality of Variances
Null hypothesisAlternative hypothesisValue of F-statistics1
S1 =S2S1 <S21.44
S1 =S3S1 <S31.38
S4 =S5S4>S51.24
S4 =S6S4>S61.20
S7 =S8S7 <S81.66
S7 =S9S7 <S91.27
S10 =S11S10 <S111.22
S10 =S12S10>S121.06

Under the alternative hypothesis, the value of the F-statistics should exceed the critical value of the F-distribution, which IS F. 95 (18,18) = 2.22.

Under the alternative hypothesis, the value of the F-statistics should exceed the critical value of the F-distribution, which IS F. 95 (18,18) = 2.22.

If net foreign assets were more sensitive to differential rates of return in the 1970s than in the 1950s, the following testable implications would arise: the correlation between the cross section of the average saving and investment rates would be lower in the 1970s than in the 1950s; the correlation between the cross section of average current accounts and investment rates would be negligible in the 1950s but significantly negative in the 1970s; and the cross section of the changes in the average investment rates between the two decades would be unrelated to the cross section of the changes in the average saving rates but negatively related to the cross section of the changes in the average current account rates. Table 3 shows that the data clearly lead one to reject the hypothesis that changes in net foreign assets have become more sensitive to yield differentials, regardless of the period used to represent the 1970s.5

Table 3.Nineteen Industrial Countries: Capital Mobility, 1950s and1970s 1
Sample PeriodRegression EquationR2
1.(1949-59)(I/Y)=0.053(2.13)*+0.69(6.25)*(S/Y)0.69
2.(1949-59)(CA/Y)=0.009(0.32)0.04(0.30)(I/Y)0.005
3.(1971–81)(I/Y)=0.030(0.88)+0.88(6.12)*(S/Y)0.71
4.(1971–81)(CA/Y)=0.033(1.17)0.19(1.66)(I/Y)0.11
5.(1974–81)(I/Y)=0.034(0.91)+0.88(5.47)*(S/Y)0.68
6.(1974–81)(CA/Y)=0.04(0.76)0.24(0.93)(I/Y)0.12
7.(1949–59)–(1971–81)Δ(I/Y)=0.021(4.87)*+0.78(6.11)*Δ(S/Y)0.70
8.(1949—59)—(1971—81)Δ(CA/Y)=0.015(3.15)*+0.05(0.46)Δ(I/Y)0.01
9.(1949–59)–(1974–81)Δ(I/Y)=0.024(5.67)*+0.77Δ(5.81)*(S/Y)0.70
10.(1949–59)–(1974–81)Δ(CA/Y)=0.019(3.58)*0.02(0.19)Δ(I/Y)0.01

Parentheses enclosing a period of years indicate the average value of the variables during that period. The delta (Δ) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate the statistic of the parameter is not reported by the authors. 7 denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

Parentheses enclosing a period of years indicate the average value of the variables during that period. The delta (Δ) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate the statistic of the parameter is not reported by the authors. 7 denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

Additional information about the hypothesis that changes in net foreign assets were more sensitive to differential rates of return in the 1970s can be obtained by comparing the variances of the average investment rates, of the average domestic saving rates, and of the average current account-to-GNP ratios. Table 2 shows these variances calculated for the cross section of 19 industrial countries for 1949–59, 1974–81, and 1971–81. The null hypothesis that the variances of the saving rate and investment rate did not change between the 1950s and the 1970s could not be rejected at the usual significance level.6 Thus, the differences that existed among the saving and investment rates of the industrial countries in the 1950s continued to exist in the 1970s—that is, the proportion of industrial countries that invested and saved more than the average remained constant. The same thing is true for the proportion of industrial countries that invested and saved less than average. If national markets for real capital had become more integrated in the 1970s, and if savings and investment had responded to real interest rate movements, the dispersion of the savings and/or investment rates around their means would have increased or decreased but would not have remained constant. In addition, if national markets for real capital had become more integrated, the differences among the current account-to-investment ratios of the industrial countries would have substantially declined in the 1970s, compared with the 1950s.7 When the variance of this ratio was computed for the three periods, it was found to be almost the same in 1974–81 and 1971–81 as in 1949–59. The comparison among these variances thus provides another bit of evidence against the hypothesis that capital became more mobile in the 1970s.

III. The Evidence Re-examined

This study’s failure to find evidence of increased integration among national markets for physical capital in the 1970s and 1980s leaves unresolved the apparent contradiction between Feldstein and Horioka’s and Sachs’s results. In this section, their work is re-examined and extended. First, the focus is on the stability of the estimated slope coefficients in Sachs’s and Feldstein and Horioka’s equations.

stability of coefficients

The first set of regressions used by Feldstein and Horioka (1980) and Sachs (1983) relates a cross section of average saving rates (I/Y) to cross sections of average investment rates (I/Y) and of the current account-to-GNP ratios (CA/Y). Feldstein and Horioka (1980) and Feldstein (1983) found that (S/Y) was highly correlated with (I/Y) in various sample periods. By contrast, rows 3 and 5 of Table 1 show that the negative correlation between (CA/Y) and (I/Y) is not apparent before the 1970s. As was discussed in Section II, a plausible explanation of these results is that since the beginning of the 1970s the industrial world has slowly moved toward a regime under which real rates of return tend to be equalized across countries. The second set of regressions used by Feldstein and Horioka (1980) and Sachs (1981, 1983) relates a cross section of changes in average investment rates to both a cross section of changes in average saving rates and a cross section of changes in current account rates.

In order to check the stability of the regression coefficients through time, the years after World War II were divided into five subperiods, and Feldstein and Horioka’s and Sachs’s equations were refitted for each pair of consecutive subperiods. The sub-periods were selected so that all of them included roughly the same number of business cycles; the last two subperiods—1968–73 and 1974–79—coincided with the sample period used in Sachs (1981). The results are reported in Table 4. The correlation be-tween the cross sections of changes in saving and investment rates appears to be high throughout the years, and it increases substantially after 1973.8 By contrast, the negative correlation between the investment and current account rates takes a large value only between 1968 and 1979, which is the sample period used by Sachs (1981). However, this value declines substantially, from –0.55 to –0.39, if 1980 is added to the sample period. Before the 1970s, the estimated slope coefficient in Sachs’s equation is significantly different from zero only between 1956 and 1967.

Table 4.Nineteen Industrial Countries: Capital Stability of Coefficients1
Sample PeriodRegression EquationR2
(1950–55)–(1956–61)Δ(I/Y)=0.018(2.65)*+0.55Δ(4.74)*(S/Y)0.57
Δ(CA/Y)=0.005(0.446)0.12(0.59)Δ(I/Y)0.02
(1956–61)–(1962–67)Δ(I/Y)=0.009(1.83)*+0.77Δ(2.73)*(S/Y)0.30
Δ(CA/Y)=0.000(0.10)0.32(2.39)*Δ(I/Y)0.25
(1962–67)–(1968–73)Δ(I/Y)=0.002(0.79)+0.55Δ(3.18)*(S/Y)0.37
Δ(CA/Y)=0.007(2.45)*0.28Δ(1.25)(I/Y)0.08
(1968–73)–(1974–79)2Δ(I/Y)=0.014(1.95)*+0.83Δ(3.58)*(S/Y)0.43
Δ(CA/Y)=0.017(4.53)*0.55Δ(3.61)*(I/Y)0.43
(1968–73)–(1974–80)2Δ(I/Y)=0.013(2.15)*+0.81Δ(4.36)*(S/Y)0.53
Δ(CA/Y)=0.018(4.51)*0.39Δ(2.48)*(I/Y)0.27

Parentheses enclosing a period of years indicate the average value of the variables during that period. The delta (A) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate the statistic of the parameter is not reported by the authors. I denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

These equations also appear in Table 1.

Parentheses enclosing a period of years indicate the average value of the variables during that period. The delta (A) indicates the change from the average of the first period indicated in parentheses to the average of the second. Dots indicate the statistic of the parameter is not reported by the authors. I denotes gross domestic fixed investment, Y gross national or domestic product, CA the current account balance including official transfers, S gross national savings, and OIL the net imports of oil at constant prices. See the Appendix for the sources of the data and the definitions adopted. The t-statistics are shown in parentheses below the coefficients, and an asterisk (*) indicates that the coefficient is significant at the 5 percent level.

These equations also appear in Table 1.

Because the cross section consists of only 19 countries and because the negative correlation between the cross section of changes in investment rates and the cross section of changes in current account balances emerges in only two periods, a natural question to ask is whether the correlation depends on a few out-liers.9Charts 1 and 2 show the scatter plots of the average changes in investment rates and the average changes in current account rates between 1956–61 and 1962–67 and between 1968–73 and 1974–79—that is, for the only two subperiods in which the negative correlation between the two variables is significant.10 The plots clearly indicate that the negative slope of the regression line depends on one outlier in the first subperiod, Canada, and on three outliers in the second subperiod, Switzerland, Norway, and New Zealand. When Sachs’s equation was re-estimated without these outliers, an estimate of the slope coefficient not significantly different from zero was obtained.11

Chart 1.Nineteen Industrial Countries: Changes in Fixed Investment and Current Account Balances, 1956–61 to 1962–671

(Percent of GNP)

1 The line depicts the regression Δ (CA/Y) = –0.000 –0.32Δ(I/Y).

Chart 2.Nineteen Industrial Countries: Changes in Fixed Investment and Current Account Balances, 1968–73 to 1974–791

(Percent of GNP)

1 The line depicts the regression Δ (CA/Y) = –0.017 –0.55Δ(I/Y).

Not only does the investment-current account relation depend on a few outliers but—for Switzerland—this relation may not stem from differentials in rates of return on domestic capital. The sharp increase in the Swiss current account surplus after 1973 reflected a marked improvement in investment income that was probably related to the “safe haven” status of Swiss capital markets. As a consequence, it is difficult to interpret the data as showing that changes in the current account position were due to changes in the expected real rate of return on the capital stocks in the country. Chart 2 further illustrates this point. From 1968–73 to 1974–79, industrial countries differed markedly in their capital accumulations. For example, Ireland and Canada increased the share of income allocated to fixed investment by nearly 4 and 2 percentage points, respectively. By contrast, the Federal Republic of Germany and the Netherlands decreased this share by over 3 percentage points. However, notwithstanding the differences in investment rates, the vast majority of industrial countries experienced similar changes in the current account rates.

It seems reasonable to conclude that a few isolated episodes appear to determine the statistical negative correlation between current account balances and investment rates in the 1970s. This correlation is thus insufficient to support the hypothesis that changes in net foreign assets were responsive to changes in investment opportunities in the industrial world.

joint estimation

The contrasting results that Feldstein and Horioka and Sachs obtained with the same set of data from the 1970s may be caused by some specification error owing to omitted variables.12 The best way to test this form of misspecification is to include variables in the regression that might be important. The problem with this approach is that it is difficult to find a few variables that capture the effect of the numerous shocks that might have affected both endogenous and exogenous variables during the sample period.13 An alternative strategy for testing the “omitted” variable hypothesis is to use an instrumental variable approach. It was assumed that Feldstein and Horioka’s equation was truly structural—that is, that changes in saving rates determined the systematic part of changes in investment rates.14 Under this assumption, an omitted variable would affect investment rates through their random component. If the negative correlation between investments and current accounts was due to any omitted variable, then changes in investment rates would not have an explanatory power in Sachs’s equation once the random component had been eliminated from the changes. Therefore, Sachs’s equation was re-estimated using the predicted values from Feldstein and Horioka’s equation as the explanatory variable. As Table 5 shows, the correlation between the cross sections of changes in investment rates and of changes in current account balances disappears. Although one can always argue that the results stem from the choice of an unsatisfactory instrumental variable, the authors of this study think that these regressions are sufficient to cast some doubt on the specification of Sachs’s equation.

Table 5.Nineteen Industrial Countries: Instrumental Variable Estimation1
Sample PeriodRegression EquationR2
(1968–73–1974–80)Δ(CA/Y)=0.014(0.005)+0.02Δ(0.30)(I/^Y)0.001
(1968–73)–(1974–79)Δ(CA/Y)=0.014(0.004)+0.09Δ(0.24)(I/^Y)0.007

A circumflex (ˆ) indicates the predicted value of the variable it appears above. Standard errors of estimate appear in parentheses below the coefficients.

A circumflex (ˆ) indicates the predicted value of the variable it appears above. Standard errors of estimate appear in parentheses below the coefficients.

adding-up constraint

Given the constraint resulting from the national income identity, an increase in the investment rate cannot be associated with an increase in the saving rate of the same amount unless the current account balance remains unchanged.15 Because this adding-up constraint is used by both Sachs and Feldstein and Horioka to interpret their results, this study jointly estimated their equations by constraining the sum of the slope coefficients to be equal to one.16 To implement the estimation, the observations were stacked as follows:

where i denotes the country index, which ranges from 1 to 19. The system can be rewritten in a more compact way as Y = XB + U. The assumption that the disturbances in both Sachs’s and Feldstein and Horioka’s equations are homoskedastic implies that

and

The variance-covariance matrix of the disturbances can then be expressed as

where

The constraint on the coefficients can be expressed in matrix form as

where

The constrained estimates of the parameters B^ can then be obtained as (Theil (1971))

where

The constrained estimates are shown in Table 6 for the two sample periods.

Table 6.Nineteen Industrial Countries: Constrained Estimates1
Sample PeriodRegression Equation
(1968–73)–(1974–80)Δ(CA/Y)=0.016(0.004)0.15Δ(0.11)(I/Y)
Δ(I/Y)=0.021(0.004)+1.15Δ(0.11)(S/Y)
(1968–73)–(1974–79)Δ(CA/Y)=0.016(0.004)0.33Δ(0.12)(I/Y)
Δ(I/Y)=0.026(0.004)+1.33Δ(0.12)(S/Y)

The standard errors are in parentheses. They were obtained from the variance-covariance matrix

The standard errors are in parentheses. They were obtained from the variance-covariance matrix

If the sample period includes 1980, the estimate of the slope coefficient in Sachs’s equation is smaller than two standard deviations, while the estimate in Feldstein and Horioka’s equation is not significantly different from 1 at the usual significance levels. By contrast, if 1980 is left out, the slope estimate in Sachs’s equation is equal to -0.34 and is now significantly different from zero even though it is substantially smaller than the estimate obtained in the unconstrained regression. For the Feldstein and Horioka equation, the constraint has the effect of increasing the slope coefficient to 1.33, a value that has no immediate economic interpretation. The point that emerges from this joint estimation is that one minus the estimated coefficient in the original Sachs

equation grossly understates the fraction of gross capital formation that is financed by domestic savings.

analysis of time series

An alternative procedure used to test for misspecification, though not a rigorous one, is to analyze the time-series properties of the variables used in the regressions. The two basic equations can be rewritten as

where Xti and Zti denote two variables that might have some explanatory power in the regressions but are omitted; uti and eti denote serially uncorrelated disturbance terms; and i denotes the country index. Because the most conflicting results are obtained by Feldstein and Horioka and Sachs with regressions that use changes in the variables over two periods, a t subscript is attached to the variables to make explicit the time dependency of the regressions. If changes in net foreign assets were insensitive to differential rates of return and if the equation adopted by Feldstein and Horioka was correctly specified—that is, if no relevant variables were left out—then a0 and a2 would be equal to zero and a1 equal to one. The first equation could then be rewritten as

so that savings would differ from investments only by serially uncorrelated error terms, and current account balances would simply offset unanticipated shocks to the investment and the saving functions. Thus, uti could be interpreted as a forecasting error since, by assumption, people would not plan to accumulate net claims on nonresidents. A test that uti is white noise can then be interpreted as a test that national markets for physical capital are not integrated. Such a test would be meaningful only if a proper unit of time were chosen—the equality between ex ante savings and investments would not be expected to hold on, say, a monthly basis. In what follows, annual data are used. Although a year is probably too short a unit of time, the use of longer units would have resulted in insufficient numbers of observations.

In the first column of Table 7, it is shown that there is a high degree of serial correlation of investment rates or, in other words, that shocks to investment tend to persist, so that movements in investment rates are characterized by long cycles. If these movements were driven by movements in the saving rates, then saving rates would also be characterized by the same cycles, and the saving-investment gaps would be white noise. This is true for 12 out of 19 countries. For these countries, the null hypothesis of no serial correlation could not be rejected at the 0.05 significance level (see the second column of Table 7).

Table 7.Nineteen Industrial Countries: Autocorrelation Functions1
Significance Level

of the Box-Pierce

Q(n) Statistics for

n Lags
(I/Y)ti(I–S)Yti(I–CA)Yti
Q(4)Q(6)Q(4)Q(6)Q(4)Q(6)
United States0.0040.0150.5700.7100.0340.090
Canada0.0000.0000.0630.1400.0560.090
Australia0.0000.0000.1600.2900.9370.990
Japan0.0000.0000.0620.1070.0000.000
New Zealand0.0000.0000.5000.6700.0690.070
Austria0.0000.0000.0000.0000.0000.000
Belgium0.0000.0000.0150.0400.0000.000
Denmark0.0000.0000.0000.0000.0000.000
Finland0.0000.0000.0520.1180.0100.022
Germany,
Fed. Rep. of0.0000.0000.0030.0130.0000.000
Iceland0.0000.0000.3360.4030.0560.127
Ireland0.0000.0000.0650.1330.0000.000
Italy0.0010.0030.6500.4890.0010.003
Netherlands0.0000.0000.1890.2100.0080.023
Norway0.0380.0640.0090.0310.0060.014
Spain0.0000.0000.0560.1110.0050.017
Sweden0.0000.0000.0850.1700.0000.000
Switzerland0.0000.0000.0010.0000.0000.000
United Kingdom0.0000.0000.0100.0280.0000.000

The sample period was 1948-81 for the majority of the 19 countries; 1949-81 for Australia, the Federal Republic of Germany, and New Zealand; 1950-81 for Italy and Belgium; 1952-81 for Japan; and 1954-81 for Spain.

The Q(n) statistics are equal to TΣi=1nri where ri denotes the ith estimated autocorrelation and T denotes the sample size. A number in the table exceeding 0.05 indicates that the null hypothesis that the first n autocorrelation coefficients are equal to zero cannot be rejected at the 5 percent significance level.

The sample period was 1948-81 for the majority of the 19 countries; 1949-81 for Australia, the Federal Republic of Germany, and New Zealand; 1950-81 for Italy and Belgium; 1952-81 for Japan; and 1954-81 for Spain.

The Q(n) statistics are equal to TΣi=1nri where ri denotes the ith estimated autocorrelation and T denotes the sample size. A number in the table exceeding 0.05 indicates that the null hypothesis that the first n autocorrelation coefficients are equal to zero cannot be rejected at the 5 percent significance level.

A similar test can be performed for the difference between investment rates and current-account ratios. If the long cycles of investment rates were accounted for by movements in net foreign assets, their difference would be white noise, and this result could be interpreted as supporting the capital mobility hypothesis. In the third column of Table 7, it is shown that the null hypothesis that (ICA)/Yti is white noise can be rejected at the 0.05 significance level for all countries except Australia, Canada, Iceland, and New Zealand. Although this test is only suggestive, it adds to the evidence that national markets for real capital were not highly integrated among industrial countries during the sample period.

IV. Conclusion

In this paper, it has been shown that the hypothesis that national markets for real capital are highly integrated receives little support from the data for 19 industrial countries. By re-examining and extending existing studies, this study has found that cross-country differences in investment rates have mirrored the differences in saving rates during the years after World War II. More-over, this relation was as strong in the 1970s as in the 1950s, when international trade in goods and factors of production was hindered by extensive restrictions and controls. Another finding has been that the link between current account balances and investment opportunities that seems to have emerged in the 1970s is accounted for by data from only a few countries and perhaps by shocks to those two variables that were common to all countries.

In view of the evidence presented here, it is concluded that changes in the propensity to save or to invest on the part of residents of an industrial country result in changes in that country’s investment share or saving share, while current account balances act as temporary shock absorbers. The authors are unable to explain why the industrial countries were still behaving like “insular” economies in the late 1970s, even though a substantial part of the barriers to the international mobility of goods and factors of production, which existed in the 1950s, had been phased out.

It must be acknowledged that the evidence presented might be consistent with the hypothesis that no differentials in real rates of return existed among industrial countries over the period studied. This could have occurred if fiscal policies in industrial countries were constantly aiming at balancing current accounts. Although this possibility cannot be ruled out, it is unlikely that governments have been able to systematically influence aggregate savings rates.

The apparent lack of integration among national markets for real capital can be reconciled with the existence of current account imbalances among industrial countries and with the expansion of the international capital markets in the 1970s. In view of the evidence presented in the paper, current account imbalances seem unrelated to differences in rates of return. Instead, they seem to reflect a variety of unanticipated shocks to incomes and terms of trade, or shocks that are believed to be temporary. Over time, these unintended changes in net foreign assets might sum to zero as portfolios are adjusted to desired levels. There have clearly been isolated instances of industrial countries—for example, Norway, Canada, and New Zealand—importing a substantial amount of foreign savings. The possibility remains, however, that these are unusual events that should not be viewed as reflecting an increased integration of national capital markets among industrial countries.

The virtual elimination of effective controls over financial capital movements among industrial countries and the existence of extraterritorial credit markets provide ample opportunity for savings from one country to find their way into investment or consumption loans in another country. However, it is also clear that the very large volume of two-way trade in financial assets that has developed in recent years could accommodate portfolio preferences associated with diversification, tax avoidance, or avoidance of controls on domestic financial intermediation, without any net capital flow associated with redistribution of world savings. But it is the net flow that plays an important role in equalizing rates of return on real capital among countries.

APPENDIX: Data Sources and Definitions

Current account of balance of payments

Data were obtained for the current accounts of the balance of payments of the industrial countries by adding lines 77a.d, 77and, 77aed, and 77afd on the appropriate country pages of the Fund’s monthly publication, International Financial Statistics (IFS). Because the current accounts were expressed in dollars, they were converted to local currencies using the average market exchange rate (line ah). Only the time series for Canada and Ireland were available in IFS for the entire period beginning in 1949. The initial dates for the other countries ranged from 1951 for Italy and the Netherlands to 1967 for Sweden. For most countries, these time series were extended back to 1949 using the data published in the Fund’s Balance of Payments Yearbook, volumes 3-19. However, there were no current account data for Switzerland in 1949 and for Spain from 1949 until 1954. In addition, there were no data on the transactions between metropolitan France and the franc area before 1967. As a result, it was decided to leave France out of the sample.

GDP/GNP17

The GDP and GNP data were obtained from lines 99a and 99b, respectively, of IFS. The GDP/GNP series for Canada, Iceland, Ireland, Switzerland, the United Kingdom, and the United States are available from 1949. However, the GDP/GNP series for Austria, Belgium, Denmark, the Federal Republic of Germany, Italy, the Netherlands, Norway, and Sweden were available in IFS only from 1950. For this group of countries, the time series was extended back to 1949 by using the volume and price indices of GDP/GNP published in the Organization for Economic Cooperation’s Statistics of National Product and Expenditure (No. 2,1957). The indices of GDP/GNP were taken at constant prices for Japan and Finland from United Nations, Statistics of National Income and Expenditure, Statistical Papers, Series H, No. 9 (May 1956). This publication was also the source of the GNP at current prices in Japan for 1949–51. However, because these series systematically differed from the series taken from IFS for those years in which there was overlapping, the GNP at current prices in 1949–51 was multiplied by 1.23, which was equal to the ratio between the series in IFS and in Statistics of National Income and Expenditure during 1952–54.

Fixed investment

Gross national fixed investment was taken from IFS (line 93e). For the years during which it was not available in IFS, it was obtained from the same sources used to obtain the missing GDP/GNP data.

Savings

Gross national savings were obtained by adding gross domestic fixed investments, changes in inventories, and current accounts of the balance of payments. This measure of gross national savings is slightly biased because the current account of the balance of payments differs from the equivalent concept calculated on a national account basis. Sachs (1981) used the same procedure that is adopted here. By contrast, Feldstein (1983) used the conventional national income accounts measure. However, he showed that his results did not depend on the way in which savings were calculated.

REFERENCES

    FeldsteinMartin“Domestic Saving and International Capital Movements in the Long Run and the Short Run,”European Economic Review (Amsterdam) Vol. 21 (March/April1983) pp. 12951.

    FeldsteinMartin and C.Horioka“Domestic Savings and International Capital Flows,”Economic Journal (London) Vol. 90 (June1980) pp. 31429.

    HarbergerArnold C.“Vignettes on the World Capital Market,”American Economic Review Papers and Proceedings of the Ninety-Second Annual Meeting of the American Economic Association (Nashville) Vol. 70 (May1980) pp. 33137.

    McKinnonRonald I.“The Exchange Rate and Macroeconomic Policy: Changing Postwar Perceptions,”Journal of Economic Literature (Nashville) Vol. 19 (June1981) pp. 53157.

    SachsJeffrey D.“The Current Account and Macroeconomic Adjustment in the 1970s,”Brookings Papers on Economic Activity: I (1981) The Brookings Institution (Washington) pp. 20168.

    SachsJeffrey D.“Aspects of the Current Account Behavior of OECD Economies,”in Recent Issues in the Theory of Flexible Exchange Ratesed. byE.Claassen and P.Salin (Amsterdam: North-Holland1983) pp. 10128.

    TheilHenriPrinciples of Econometrics (New York: John Wiley & Sons1971).

    TobinJames“Comments on ‘Domestic Saving and International Capital Movements in the Long Run and the Short Run’ by M. Feldstein,”European Economic Review (Amsterdam) Vol. 21 (March/April1983) pp. 15356.

Mr. Penati, economist in the Research Department, is a graduate of Bocconi University, Milan, and the University of Chicago.

Mr. Dooley, Senior Economist in the Research Department, is a graduate of Duquesne University, the University of Delaware, and the Pennsylvania State University.

An example of this view appears in McKinnon (1981, p. 533): “The development of the eurocurrency market now enables both firms and governments to borrow (or lend) internationally, on a large scale, for long periods in a variety of convertible currencies. Clearly, the international integration of capital markets in the 1980s parallels that prevailing in world trade in goods and services, whereas in the late 1940s national capital markets were segmented by exchange controls and eurocurrency transacting did not yet exist.”

Harberger (1980) criticized Feldstein and Horioka’s conclusions by pointing out that the correlation they found between saving and investment shares was biased upward by the fact that Feldstein and Horioka used gross investment and gross savings. However, Feldstein (1983) showed that Feldstein and Horioka’s results hold even though net data are used instead of gross data. Harberger also showed that the sizes of the current account-to-GNP ratios were inversely related to the sizes of the countries. This might support the view that for small countries, in which the location of investment opportunities is more likely to differ from that of savings, net capital flows systematically tend to offset any shortage or abundance of domestic savings. Tobin (1983) pointed out that the correlation between domestic saving and fixed investment rates might be biased upward because corporate profits account for a large fraction of gross domestic savings in industrial countries, but, at the same time, they are a major determinant of gross fixed investment.

Table 1 did not report the regression Sachs refers to because it was almost identical to the regression that is shown in row 7. The regression of the quotation is

Δ(CA/Y)=0.01(4.2)0.65(4.6)Δ(I/Y),R2=0.65 where the sample is the change from the average for 1961–70 to the average for 1971–79; the cross section consists of 13 industrial countries and excludes Japan; and t-statistics are shown in parentheses below the coefficients.

For example, see McKinnon (1981).

The Appendix contains the sources and the definitions of the data used in this study.

Because the distribution used to test the hypothesis of the equality of the variances is based on the assumption that the random variable is distributed normally, a X2 goodness-of-fit test was used to test the normality of the data. In every case, one could not reject the hypothesis of normality.

It is assumed here that the two oil shocks did not systematically affect the relationship between current account balances and investment rates among industrial countries in the 1970s. This assumption is supported by the data. For example, see the regression shown in row 4 of Table 1.

This increase, however, may be overstated on account of several exogenous factors. During the sample period, total savings in many countries were depressed by a redistribution of income in favor of labor, which reduced corporate profits, and by a decline in public sector savings caused by an expansion in social welfare programs. At the same time, the increase in energy prices and the larger role played by governments in the economies caused a decline in the investment rate by reducing the expected return on physical capital.

No precise statistical meaning of the word “outlier” is provided herein.

The scatter plot for 1968–80 looks almost identical to that for 1968–79.

The fitted regressions were the following:

(195661)(196267)Δ(CA/Y)=0.004(1.34)0.19(1.60)Δ(I/Y)R2=0.14(196873)(197479)Δ(CA/Y)=0.011(5.14)*0.11(1.01)Δ(I/Y)R2=0.07 The notation used here is explained in footnote 1 of Table 1.

A second misspecification, involving the functional forms of the equations chosen, may have occurred. It has been noted that the constant terms in the regressions reported were large and precisely estimated. This implies, for example, that even if the average investment rates of industrial countries had not changed between the 1960s and 1970s, their average current account balances would have deteriorated by nearly 2 percentage points in proportion to GNP, a conclusion that contrasts with the tone of Sachs’s papers. Thus, a large constant may be an indication that the functional forms of the equation are misspecified. However, quadratic functions fitted the data very poorly.

Given that the sample period covers the first oil shock, changes in the terms of trade were included, but without much success. Sachs (1983) reports a similar regression with oil imports as an independent variable; see rows 3 and 4 in Table 1

The assumption that Feldstein and Horioka’s equation is truly structural was justified by the robustness of their results, which was made clear in the previous section.

Here the changes in the inventory investment that are very small for the two periods considered are disregarded.

For example, see the quotation from Sachs in the penultimate paragraph of Section I.

GDP denotes gross domestic product, and GNP denotes gross national product.

Other Resources Citing This Publication