I. Alternative Approaches to Measuring International Capital Mobility

Tests of the level of international capital mobility generally fall into two categories. The first type uses the behavior of nominal interest rates to measure capital mobility. The cleanest of these tests involves a comparison of movements of nominal onshore and offshore interest rates for the same instrument and currency. The logic behind these tests is that if there are no barriers to capital mobility then these rates should move very closely together, in order to eliminate arbitrage opportunities. If, on the other hand, there are constraints to moving capital between the host country and the rest of the world, then the two interest rates will be able to diverge significantly. The results from these tests almost uniformly show that international capital markets are currently very integrated.³ In addition, they indicate that switches in regime such as the elimination of capital controls by the Thatcher Government in the United Kingdom in 1979 led, almost immediately, to a dramatic increase in capital mobility.

A second type of test, originally developed by Feldstein and Horioka (1980), measures capital mobility by using the correlation between saving and investment across countries. The logic behind this test is that if capital is fully mobile then there is no reason to expect higher saving in any particular country to be reflected in higher investment in that country. Rather, the total pool of world saving will be augmented, and this saving will then be used to finance the most efficient investment available across all countries. By contrast, if there are barriers to capital mobility then saving will tend to be used domestically, creating a correlation between saving and investment. Numerous studies have found a strong positive correlation between saving and investment across countries, which, using the Feldstein and Horioka approach, implies low international capital mobility.

Attempts to reconcile the Feldstein-Horioka results with those using nominal interest rates have generally focused on potential problems with the Feldstein and Horioka test. Some authors have criticized the implicit assumption that saving and investment are uncorrelated in a world of high international capital mobility by developing models in which saving and investment are highly correlated even in such a regime.⁴ Other authors have suggested that governments may have targeted the current account. As the current account is equal to the difference between saving and investment, this could create an artificial correlation between saving and investment.⁵ Still others have suggested that the onshore-offshore interest rate tests look at capital mobility between the large financial centers (New York, Tokyo, London, etc), while the saving investment correlations look at the capital mobility between countries (the United States, Japan, the United Kingdom, etc.).⁶

The implication from this last line of reasoning is that the two tests may be measuring different concepts of capital mobility: that based on interest differentials is a test of short-term capital mobility, while that based on saving-investment correlations is concerned with more long-term capital movements. One difference is clear. While onshore-offshore interest rate comparisons focus on the behavior of nominal interest rates, saving and investment respond to real interest rates. Hence, the work on saving-investment correlations is probably more closely related to work on comparisons of real interest rates across countries (for example, Gagnon and Unferth (1993) and Goodwin and Grennes (1994); see MacDonald and Taylor (1990) for a survey). Although this work has generally rejected the equalization of real interest rates, one significant problem with this line of inquiry has been that it requires the auxiliary assumption that ex ante purchasing power parity is expected to hold. Recent empirical work (for example, Abuaf and Jorion (1990) and MacDonald (1993)) has cast doubt on the accuracy of this assumption, and hence the value of these tests.

The conflicting results from the different tests have revived interest in alternative approaches to measuring international capital mobility. In particular, Obstfeld (1994) suggested that consumption paths could be used as a measure of capital mobility.⁷ The idea behind this test is that in a fully integrated world, consumption paths across countries should move together, reflecting the desire of individuals to smooth consumption and hence maximize welfare. This approach, which is more fundamental than those discussed earlier, implies a definition of capital market integration in which real interest rates are equalized across countries.

Focusing on consumption has several attractive features. The underlying theory is stronger than that for saving-investment correlations. In addition, since consumption is the ultimate goal of economic activity, it is a more basic test of the effects of financial integration on economic welfare than either the saving-investment correlations or interest rate comparisons. Also, it appears unlikely that macroeconomic policy is directed at private nondurable consumption patterns in the way it may be at the current account. Finally, as discussed below, it also incorporates a test of real interest rate parity that does not require the assumption that ex ante purchasing parity is expected to hold.

II. Consumption Across Countries in a Financially Integrated Area

Within a financially integrated area, consumers can use capital markets to smooth consumption in response to anticipated movements in income.⁸ As a result, consumption patterns across different countries should move in the same manner, except for the impact of unforeseen shocks to income. To see this, consider the Euler equation characterization of optimal consumption behavior. As formulated by Hall (1978), this assumes that rational, forward-looking consumers maximize the expected value of lifetime utility, subject to an intertemporal budget constraint. This yields the equation

where C_t is consumption, E_t-1 is the mathematical expectation conditional on the information available at t-1, β is a subjective discount factor, σ is the intertemporal elasticity of substitution, and R_t-1 is the real interest rate between t – 1 and t. Using lowercase letters to represent logarithms, this can be rewritten:

where the error ϵ_t is uncorrelated with information available in period t – 1 or earlier.

In order to see how this can be used as a test of capital mobility, consider the behavior of consumption across different countries. The path of consumption across countries a, b, …, z is

where equations (2) assume that each country has the same utility function. In these circumstances, the only reason for consumption paths to differ across countries (other than possible differences in discount rates) is if the expected real interest rates faced by consumers in different countries diverge.

Expected real interest rates can differ across countries with limited financial linkages. However, if capital markets are integrated in the sense used in this paper, then expected real interest rates will be equal. In this case, except for differences in the discount rate, consumption growth should follow the same pattern across countries. Mathematically,

for any two countries i and j, where ϵ_ijt is equal to ϵ_it – ϵ_jt. Equation (3) implies that consumption across countries in a financially integrated area should follow a random walk, and can be seen as an inter-country version of Hall’s original consumption model.⁹ It can also be seen as a test of the level of insurance that capital markets provide for consumption. If such markets are complete, then consumption should vary with aggregate values but be unrelated to other factors. Cochrane (1991) and Mace (1991) both use versions of equation (3) to test the complete markets model of consumption using data on individuals.

Equation (3) assumes that the random walk model of consumption is an accurate representation of consumption behavior in countries i and j. However, this model has not performed well in formal tests. In particular, models that assume that some proportion of the change in consumption is associated with anticipated changes in income, such as that estimated in Campbell and Mankiw (1989 and 1990) consistently reject the forward-looking model. In these models consumption is divided into a proportion (1 – λ) that is associated with forward-looking consumers and a proportion (λ) that is associated with “rule of thumb” consumers who vary consumption in line with the change in their income. Aggregate consumption therefore equals

where y_t, is disposable income.

Equation (4) can be extended to the multicountry case. To see this rewrite equation (4) as an equation in the real interest rate:

If capital markets are integrated across countries, so that in ex ante terms R_it = R_jt, this implies that for two countries i and j

where ϵ_ijt is a function of the error terms in ϵ_i and ϵ_j. This is the basic estimating equation used in this paper. It states that, in a world of high capital mobility and rule of thumb consumers, growth of consumption in any one country is related to growth in home income and growth in foreign consumption and income. It is also a test of real interest rate parity, which does not rely upon the assumption that purchasing power parity is expected to hold.

As written, equation (5) is a test of capital mobility between two countries, and could be used in that way. However, it is often more informative to derive a test of capital mobility between a particular country and the rest of the world. This is done in the empirical work by making “country” j represent the rest of the world.¹⁰

Equation (5) provides a tractable method of investigating the importance of both rule of thumb consumers and real interest rate equalization on consumption across countries. Consider an unrestricted version of the model written with the difference of the growth in consumption on the left-hand side:

The coefficients on income can be used to test the importance of rule of thumb consumers. Real interest equalization across countries can be tested by looking at the coefficient on foreign consumption. A negative coefficient on this term would tend to indicate a lack of real interest rate equalization with the rest of the world as it would imply a low level of correlation between domestic and foreign consumption, even when liquidity constraints have been taken into account. A constant term is included in this version of the model to take account of possible differences in subjective discount factors across countries.

It is, however, possible to push our tests of equation (5) further. There is an important parameter constraint on the coefficients in equation (5). This is best seen by reparameterizing the equation as follows:

where t_i is equal to (λ_j – λ_i)/(1 – λ_j) and α_i = (λ_i – λ_j)σlnβ. The model implies that the sum of the coefficients on home income, external consumption, and external income should equal zero. The importance of this constraint can be seen by considering the case in which capital mobility is not high, and hence the ex ante real interest rate in country i, R_it, differs from that in the rest of the world. The error-in-variable bias resulting from this inequality will cause the coefficients on foreign consumption and foreign income to be biased downward. This, in turn, implies that the sum of the estimated coefficients on home income, foreign consumption, and foreign income will be less than zero.¹¹ Hence, the restriction that the sum of the coefficients on income and consumption is equal to zero is a formal test of the integration of real interest rates across countries, even taking account of liquidity constraints.

The assumption that the utility functions are the same across countries is important in the derivation of (5″). In particular, if the rate of intertemporal substitution differs across countries, then the restriction that the sum of the coefficients is zero will not hold. Hence, this restriction involves a joint test of the assumptions that capital markets are integrated and that rates of intertemporal substitution are equal. The same assumption is made by most other consumption-based tests of international capital integration, such as those in Obstfeld (1994).

One disadvantage of equation (5′) is that it includes several series that may be highly collinear. In particular, income growth at home and abroad are often highly correlated, which lowers the level of precision of the individual parameter estimates. Hence, results from a simplified version of equation (5′), in which it is assumed that λ_i = λ_j, are also reported. This implies the following estimation equation:

In this model, the sensitivity of consumption with respect to income is given by parameter β₁, while the level of correlation of domestic and foreign consumption, and hence real interest rate equalization, is measured by the amount by which parameter β₂ differs from zero.

It is important to differentiate between excess sensitivity of consumption to income and lack of real interest rate equalization as reasons for the deviation of consumption from its optimal path, because the two sources of failure potentially imply very different underlying problems. If the failure is caused by excess sensitivity of consumption to income, this generally implies that consumers are not taking full advantage of asset markets. Hence, excess sensitivity to consumption can be seen as a test of the degree to which local consumers have access to capital markets. By contrast, if home consumption growth has a low correlation with external consumption growth even when taking into account the impact of excess sensitivity of consumption with respect to income, this implies a failure of real interest rate equalization between national capital markets, A lack of correlation with external consumption is thus a test of the integration of capital markets, where integration is defined as equalization of ex ante real interest rates. Hence, in testing international capital mobility, we differentiate between imperfect access of consumers to capital markets and imperfect integration between capital markets.

This approach can be compared with that taken in Obstfeld (1994), who uses an approach suggested by Mace (1991) to examine the degree of capital mobility using consumption across countries. He estimates the following regression:

where c_jt is consumption in the rest of the world and y_it is a measure of domestic income that is used to control for the effect of domestic income on consumption. The restriction β = 1 and Ψ = 0 is then used as a test of capital mobility, with β = 0 and Ψ = 1 being the result expected under autarky.

Comparing equations (5′) and (6), the most important point is that equation (6) does not include the change in aggregate income, Δy_jt. To the extent that Δy_jt is positively correlated with Δy_it, and Δc_jt (as it generally will be), the coefficients on domestic income and foreign consumption will be biased downward. In addition, equation (5′) implies that, in the face of rule of thumb consumers, the coefficient on the growth in consumption in the rest of the world need not equal one, as assumed in the Mace/Obstfeld formulation. Both considerations imply that the hypothesis β = 1 and Ψ = 0 is not a very powerful test of optimum consumption patterns.¹²

III. Data and Estimation Issues

Annual data on real nondurable consumption and real household disposable income were collected from the OECD’s National Accounts for 15 countries: Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Luxembourg, the Netherlands, the United Kingdom, and the United States. All seven of the major industrial countries are included in our data set, as are almost half of the other OECD countries. Hence, we believe that the data set provides a reasonable proxy for industrial countries as a whole.¹³

The data cover the period 1971 to 1992, which is the period of floating exchange rates after the breakup of the Bretton Woods fixed exchange rate system.¹⁴ Annual data were chosen because they generally convey more information than quarterly data (Shiller and Perron (1985)) and because they avoid the contamination introduced into forward consumption models by seasonal elements. Nondurable consumption was used because the theory is concerned with the marginal utility derived from consumption. Durable goods provide utility over several years, making measured consumption a bad proxy for the marginal utility derived from their ownership.

It is clearly important to make the definition of consumption as consistent as possible across countries. Since official measures of nondurable consumption are not available in most of our countries, we have constructed our own measure of nondurable consumption by collecting data on the individual components of consumption that most closely approximate nondurable consumption and, for any one country, used the sum of these items as our measure of nondurable consumption. Our measure of real nondurable consumption (at 1980 prices) includes consumer spending on clothing, education, food, medical care, and miscellaneous goods and services (which are largely personal services).

Real personal disposable income (at 1980 prices) is defined as real income minus taxation for all countries except Denmark, Greece, Ireland, Italy, and Luxembourg.¹⁵ For this latter group of countries, reliable income and tax series are not available for the full sample period and we therefore used real GDP. Both our real consumption and income measures have been divided by population to produce corresponding per capita measures. Consumption and income in the rest of the region were calculated by weighting indices together using OECD estimates of purchasing power parity GDPs in 1980.

Equations (5′) and (5‴) were estimated using the Generalized Method of Moments (GMM). GMM is particularly appropriate in this case for several reasons. Liquidity constraint models of this type should be estimated using instrumental variable estimation (such as GMM) since disturbances to domestic income contain information about permanent income and are therefore correlated with consumption. In addition, because the data on consumption are time averages, the model implies a first-order moving average process in the error (Working (1960)); GMM techniques provide an adjustment for this effect, as well as being robust to heteroscedasticity.¹⁶ Finally, GMM provides a direct test of the overidentifying restriction that the instruments are uncorrelated with the errors.¹⁷ Hence, GMM can be used to test simultaneously the importance of the parameters and the orthogonality of the errors to the instruments.

The choice of instruments is important for this type of model.¹⁸ Past changes in income should be useful in helping to predict future changes in consumption and income. In addition, since the consumption model underlying our approach is the permanent income model, it follows that current consumption will summarize agents’ information about the future path of income (Campbell (1987)). Thus, lagged values of the change in consumption should also be a useful predictor of changes in income. Finally, the permanent income model also implies that the ratio of consumption to income should be useful in predicting future income (Campbell (1987)). An important limitation on the instrument set is that first lags are inadmissable as instruments because the time averaging of the consumption data induces a correlation between the change in consumption and its first lag. To preserve degrees of freedom in the estimation, only second lags of the instruments were used. Accordingly, the instrument set contained the second lag of the growth in real consumption, real disposable income, and ratio of nondurable consumption to disposable income for both the home country and the rest of the world.

IV. International Results

Table 1 reports the results of testing the importance of income and financial market integration for the 15 industrial countries in the data set. The first column reports a Wald test for the joint significance of the six instruments in explaining the behavior of consumption growth for the country relative to the “rest of the world.”¹⁹ Since the random walk consumption model in equation (3) implies that no prior information should be useful in predicting the relative growth of consumption across countries, this is a test of the optimality of consumption growth across countries. The results indicate that the random walk model is rejected for the majority of countries in the sample (9 of the 15). Generally, these rejections are decisive, in the sense that they occur at the 1 percent level of significance.

The second column of Table 1 reports on a test of the overidentifying restriction that the errors in equation (5′) are orthogonal to the instruments. The results indicate that the generalized model, which includes home income, external consumption, and external income, is an adequate description of the data, in the sense that the errors from this equation are no longer significantly correlated with the instruments used in the estimation.

The next three columns report the coefficient estimates for the independent variables in the regressions.²⁰ Although there is some variation across countries, all of the coefficients on home income (β₁) have the expected positive sign, while those on external income (β₃) generally have the expected negative sign. The coefficients on home income are significant in 12 of the 15 regressions, indicating that consumption is excessively sensitive to local income in all countries but Austria, Greece, and Japan.

The coefficients on external consumption (β₂) are often both large and negative, indicating that consumption at home and abroad are not highly correlated and hence a lack of real interst rate equalization, particularly for countries such as Belgium, Denmark, Greece, and Italy. The sum of the coefficients on home income, foreign income, and foreign consumption (β₁ + β₂ + β₃), together with inferred standard errors, is shown immediately after the results for each coefficient individually. The sum is negative in the majority of cases, again suggestive of a lack of real interest rate equalization. The significance of these tests is, however, rather weak, and only two countries fail the test at conventional levels. One reason for the relatively low power of these tests may be attributable to the collinearity among the right-hand side variables in the estimated version of equation (5), which is why we also propose estimating equation (5‴).

Table 1.

International Data: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_j + ϵ_ijt

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income.

Table 1.

International Data: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_j + ϵ_ijt

	Significance of instruments χ2(6)		Orthogonality of errors χ2(3)	Home income β1		External consumption β2		External income β3	Sum of coefficients β1+β2+β3		Equality of income parameters λi = λj χ2(1)
Austria	4.0		0.3	0.27		-0.74		-0.14	-0.61	*	0.8
				(0.17)		(0.39)		(0.33)	(0.26)
Belgium	24.4	**	1.5	0.34	*	-1.54	**	0.62	-0.57		1.6
				(0.15)		(0.40)		(0.64)	(0.59)
Canada	12.6		0.6	0.64	**	0.09		-0.88	-0.14		10.1	**
				(0.19)		(0.56)		(0.49)	(0.41)
Denmark	73.9	**	2.7	1.02	**	-2.07	**	0.29	-0.76	**	27.5	**
				(0.18)		(0.53)		(0.28)	(0.31)
Finland	15.8	*	0.2	1.15	**	0.26		-0.97	0.44		10.3	**
				(0.28)		(0.71)		(0.64)	(0.41)
France	16.6	*	1.0	0.41	*	-0.20		-0.27	-0.06		3.5
				(0.16)		(0.26)		(0.26)	(0.21)
Germany	10.2		1.7	0.90	**	-0.06		0.49	0.34		17.8	**
				(0.21)		(0.49)		(0.37)	(0.39)
Greece	9.7		1.3	0.33		-2.65	**	0.52	-1.80		2.4
				(0.31)		(1.03)		(0.88)	(1.04)
Ireland	33.6	**	0.8	1.19	**	0.48		-0.74	0.93		5.3	*
				(0.38)		(1.63)		(1.47)	(0.67)
Italy	21.4	**	0.2	0.82	*	-1.62	*	-0.01	-0.81		0.9
				(0.33)		(0.72)		(0.66)	(0.94)
Japan	5.6		3.0	0.11		-0.41		0.29	-0.01		0.0
				(0.24)		(0.56)		(0.90)	(0.42)
Luxembourga	33.9	**	0.8	1.19	**	0.68		-0.90	0.98		6.3	*
				(0.40)		(1.98)		(1.58)	(0.92)
Netherlands	4.7		1.3	0.64	**	-0.60		0.27	0.30		0.0
				(0.09)		(0.40)		(0.28)	(0.43)
United Kingdom	12.6	*	3.7	0.47	*	-0.11		-0.18	0.18		0.6
				(0.20)		(1.03)		(0.90)	(0.39)
United States	22.5	**	1.1	0.27	*	0.38		-0.80	-0.15		17.0	**
				(0.14)		(0.21)		(0.15)	(0.12)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income.

View Table

Table 1.

International Data: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_j + ϵ_ijt

	Significance of instruments χ2(6)		Orthogonality of errors χ2(3)	Home income β1		External consumption β2		External income β3	Sum of coefficients β1+β2+β3		Equality of income parameters λi = λj χ2(1)
Austria	4.0		0.3	0.27		-0.74		-0.14	-0.61	*	0.8
				(0.17)		(0.39)		(0.33)	(0.26)
Belgium	24.4	**	1.5	0.34	*	-1.54	**	0.62	-0.57		1.6
				(0.15)		(0.40)		(0.64)	(0.59)
Canada	12.6		0.6	0.64	**	0.09		-0.88	-0.14		10.1	**
				(0.19)		(0.56)		(0.49)	(0.41)
Denmark	73.9	**	2.7	1.02	**	-2.07	**	0.29	-0.76	**	27.5	**
				(0.18)		(0.53)		(0.28)	(0.31)
Finland	15.8	*	0.2	1.15	**	0.26		-0.97	0.44		10.3	**
				(0.28)		(0.71)		(0.64)	(0.41)
France	16.6	*	1.0	0.41	*	-0.20		-0.27	-0.06		3.5
				(0.16)		(0.26)		(0.26)	(0.21)
Germany	10.2		1.7	0.90	**	-0.06		0.49	0.34		17.8	**
				(0.21)		(0.49)		(0.37)	(0.39)
Greece	9.7		1.3	0.33		-2.65	**	0.52	-1.80		2.4
				(0.31)		(1.03)		(0.88)	(1.04)
Ireland	33.6	**	0.8	1.19	**	0.48		-0.74	0.93		5.3	*
				(0.38)		(1.63)		(1.47)	(0.67)
Italy	21.4	**	0.2	0.82	*	-1.62	*	-0.01	-0.81		0.9
				(0.33)		(0.72)		(0.66)	(0.94)
Japan	5.6		3.0	0.11		-0.41		0.29	-0.01		0.0
				(0.24)		(0.56)		(0.90)	(0.42)
Luxembourga	33.9	**	0.8	1.19	**	0.68		-0.90	0.98		6.3	*
				(0.40)		(1.98)		(1.58)	(0.92)
Netherlands	4.7		1.3	0.64	**	-0.60		0.27	0.30		0.0
				(0.09)		(0.40)		(0.28)	(0.43)
United Kingdom	12.6	*	3.7	0.47	*	-0.11		-0.18	0.18		0.6
				(0.20)		(1.03)		(0.90)	(0.39)
United States	22.5	**	1.1	0.27	*	0.38		-0.80	-0.15		17.0	**
				(0.14)		(0.21)		(0.15)	(0.12)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income.

The last column in Table 1 reports a Wald test of the hypothesis that the effect of home income on home consumption is the same as the effect of external income on external consumption. This restriction is accepted in 8 of the 15 countries being studied. Accordingly, Table 2 reports the results from estimating equation (5‴), the simplified version of the model in which this restriction is imposed. For completeness, we estimated the specification across all countries in our sample.²¹ The coefficient on income growth, which measures the importance of income in consumption, is now significant in 13 of the 15 countries, the exceptions being Austria (where the coefficient is significant at the 10 percent level) and Japan. The coefficient on external consumption, which measures the level of capital market integration, is significant and negative for seven countries: Austria, Belgium, Denmark, Greece, Italy, Luxembourg, and the Netherlands. The rise in the number of countries rejecting capital market integration confirms our earlier hypothesis as to the low degree of power of the tests using the general model.

Table 2.

International Data: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵijt

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

Table 2.

International Data: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵijt

	Income β1		External consumption β2		Adjusted R2
Austria	0.24		-0.78	*	0.32
	(0.14)		(0.36)
Belgium	0.31	*	-1.32	**	0.28
	(0.14)		(0.46)
Canada	0.67	**	0.25		0.49
	(0.18)		(0.42)
Denmark	0.82	**	-1.20	**	0.65
	(0.11)		(0.24)
Finland	1.18	**	-0.29		0.60
	(0.26)		(0.58)
France	0.42	**	-0.20		0.67
	(0.16)		(0.27)
Germany	0.73	**	-0.49		0.67
	(0.17)		(0.26)
Greece	0.28	*	-2.37	**	0.02
	(0.14)		(0.85)
Ireland	1.14	**	-0.66		0.03
	(0.37)		(0.84)
Italy	0.70	**	-1.35	**	0.81
	(0.07)		(0.22)
Japan	0.15		-0.15		0.00
	(0.25)		(0.30)
Luxembourga	0.26	*	-4.82	**	0.34
	(0.13)		(1.02)
Netherlands	0.68	**	-0.93	**	0.94
	(0.06)		(0.06)
United Kingdom	0.48	*	0.34		0.35
	(0.20)		(0.70)
United States	0.43	**	-0.20		0.57
	(0.06)		(0.20)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

View Table

Table 2.

International Data: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵijt

	Income β1		External consumption β2		Adjusted R2
Austria	0.24		-0.78	*	0.32
	(0.14)		(0.36)
Belgium	0.31	*	-1.32	**	0.28
	(0.14)		(0.46)
Canada	0.67	**	0.25		0.49
	(0.18)		(0.42)
Denmark	0.82	**	-1.20	**	0.65
	(0.11)		(0.24)
Finland	1.18	**	-0.29		0.60
	(0.26)		(0.58)
France	0.42	**	-0.20		0.67
	(0.16)		(0.27)
Germany	0.73	**	-0.49		0.67
	(0.17)		(0.26)
Greece	0.28	*	-2.37	**	0.02
	(0.14)		(0.85)
Ireland	1.14	**	-0.66		0.03
	(0.37)		(0.84)
Italy	0.70	**	-1.35	**	0.81
	(0.07)		(0.22)
Japan	0.15		-0.15		0.00
	(0.25)		(0.30)
Luxembourga	0.26	*	-4.82	**	0.34
	(0.13)		(1.02)
Netherlands	0.68	**	-0.93	**	0.94
	(0.06)		(0.06)
United Kingdom	0.48	*	0.34		0.35
	(0.20)		(0.70)
United States	0.43	**	-0.20		0.57
	(0.06)		(0.20)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

The results contained in Table 2 clearly indicate the pervasiveness of liquidity constraints across our sample of countries; it is only in the cases of Austria and Japan that there is no evidence of this pattern. In addition, note that all of the countries that produce a significant coefficient on external consumption, and hence evidence of a lack of real interest rate equalization, are European and, indeed, all but Austria were continental members of the European Community.²² Adding Germany (which fails at the 10 percent significance level) to the list, we have a clear majority of the EC members in the sample. Hence, the world appears to be divided into two zones. The United Kingdom and countries outside the European Community appear to have national capital markets that are quite well integrated with the rest of the world, but, except for Japan, their consumption is excessively sensitive to income. Within the European Community, excess sensitivity to income is also important, but (maintaining the assumption that intertemporal rates of substitution do not vary across countries) these countries also appear to have capital markets that are not well integrated with the rest of the world. The only country for which consumption appears to be close to its optimal path is Japan, the sole Asian country in our sample.

V. European Community

The results from the previous section indicate that, measured in terms of consumption behavior, most countries in the European Community fail our test of capital market integration. One possible explanation for this result is that there is an integrated capital market within the European Community, but that this market is imperfectly connected to the rest of the world. To investigate this possibility, the model for EC countries was re-estimated with the “rest of the world” being limited to current members of the European Community.²³

There are a number of other reasons for being particularly interested in results that focus on the European Community. The European Community contains a number of relatively homogeneous industrialized countries that have close economic links, and therefore a priori would be expected to have a high degree of capital mobility. On the other hand, many of its members also maintained capital controls over most of the estimation period.²⁴ Finally, moves toward regional integration, such as the single market program and plans for a single currency, make it particularly interesting to focus upon the existing level of economic integration in the European Community.

Tables 3 and 4, which have the same format as Tables 1 and 2, respectively, show the results from estimating the model for the ten countries in our sample that were members of the European Community. The rejection of the restrictions tests for the optimality of relative consumption growth, reported in the first column of Table 3, is even more decisive here than for the international data set: only two countries, the Netherlands and Belgium, passed this test. Both versions of the model provide clear evidence that consumption is excessively sensitive to home income (this is shown in eight of the ten instances reported in Table 4). Also, one or other of the tests of the equality of real interest rates fails in six out of ten countries. Thus even when behavior is compared to the rest of the European Community, consumption across these countries still appears to be impeded by the lack of integration of capital markets. These results imply that recent moves to increase capital market integration within the European Community, through such initiatives as the EC-wide single market program and moves toward a single currency, may well provide significant benefits in terms of intraregional consumption patterns.

Table 3.

European Community: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_i + ϵijt

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

Table 3.

European Community: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_i + ϵijt

	Significance of instruments χ2(6)		Orthogonality of errors χ2(3)	Home income β1		External consumption β2		External income β3		Sum of coefficients β1+β2+β3		Equality of income parameters λi = λj χ2(1)
Belgium	5.8		2.2	0.26	*	-1.61	*	0.64		-0.72		4.9*
				(0.13)		(0.81)		(0.59)		(0.32)
Denmark	29.8	***	3.5	0.67	**	-0.13		-0.56		-0.02		43.8	**
				(0.13)		(1.13)		(0.79)		(0.29)
France	14.5*		4.2	0.34	**	-1.04		0.24		-0.47	*	0.0
				(0.08)		(043)		(0.30)		(0.21)
Germany	20.8	**	3.0	1.00	**	-0.67		-0.27		0.06		14.1	**
				(0.16)		(0.52)		(0.41)		(0.27)
Greece	18.4	**	1.7	0.17		-2.19	*	1.46		-0.56	*	13.1	**
				(0.10)		(0.99)		(0.79)		(0.27)
Ireland	15.8*		2.9	0.50		0.14		0.32		0.95		0.0
				(0.44)		(3.00)		(2.26)		(1.15)
Italy	134.1	**	0.3	0.94	**	-2.27	**	0.74	*	-0.59	**	163.9	**
				(0.14)		(0.56)		(0.30)		(0.22)
Luxembourga	48.1	**	3.5	-0.52		1.63		-1.32		-0.20		0.0
				(0.35)		(1.94)		(0.97)		(0.74)
Netherlands	3.0		2.1	0.65	**	-0.28		-0.34		0.03		5.8*
				(0.10)		(0.44)		(0.40)		(0.23)
United Kingdom	22.2*		6.2	0.48		-1.28		0.47		-0.33		0.1
				(0.33)		(2.45)		(1.53)		(0.80)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

View Table

Table 3.

European Community: General Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt + β₃Δy_i + ϵijt

	Significance of instruments χ2(6)		Orthogonality of errors χ2(3)	Home income β1		External consumption β2		External income β3		Sum of coefficients β1+β2+β3		Equality of income parameters λi = λj χ2(1)
Belgium	5.8		2.2	0.26	*	-1.61	*	0.64		-0.72		4.9*
				(0.13)		(0.81)		(0.59)		(0.32)
Denmark	29.8	***	3.5	0.67	**	-0.13		-0.56		-0.02		43.8	**
				(0.13)		(1.13)		(0.79)		(0.29)
France	14.5*		4.2	0.34	**	-1.04		0.24		-0.47	*	0.0
				(0.08)		(043)		(0.30)		(0.21)
Germany	20.8	**	3.0	1.00	**	-0.67		-0.27		0.06		14.1	**
				(0.16)		(0.52)		(0.41)		(0.27)
Greece	18.4	**	1.7	0.17		-2.19	*	1.46		-0.56	*	13.1	**
				(0.10)		(0.99)		(0.79)		(0.27)
Ireland	15.8*		2.9	0.50		0.14		0.32		0.95		0.0
				(0.44)		(3.00)		(2.26)		(1.15)
Italy	134.1	**	0.3	0.94	**	-2.27	**	0.74	*	-0.59	**	163.9	**
				(0.14)		(0.56)		(0.30)		(0.22)
Luxembourga	48.1	**	3.5	-0.52		1.63		-1.32		-0.20		0.0
				(0.35)		(1.94)		(0.97)		(0.74)
Netherlands	3.0		2.1	0.65	**	-0.28		-0.34		0.03		5.8*
				(0.10)		(0.44)		(0.40)		(0.23)
United Kingdom	22.2*		6.2	0.48		-1.28		0.47		-0.33		0.1
				(0.33)		(2.45)		(1.53)		(0.80)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

Table 4.

European Community: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵ_ijt

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

Table 4.

European Community: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵ_ijt

	Income β1		External consumption β2		Adjusted R2
Belgium	0.36	**	-1.08	**	0.21
	(0.13)		(0.56)
Denmark	0.65	**	0.01		0.66
	(0.09)		(0.74)
France	0.37	**	-0.50		0.77
	(0.10)		(0.27)
Germany	1.05	**	-0.78	**	0.69
	(0.12)		(0.30)
Greece	0.24	*	-0.60		0.44
	(0.11)		(0.42)
Ireland	0.34		2.06		0.05
	(0.46)		(13.26)
Italy	0.80	**	-1.49	**	0.78
	(0.06)		(0.25)
Luxembourga	-0.19		-0.80	*	0.31
	(0.21)		(0.38)
Netherlands	0.67	**	-0.27		0.94
	(0.06)		(0.41)
United Kingdom	0.71	**	-2.95		0.37
	(0.09)		(1.84)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

View Table

Table 4.

European Community: Simplified Model

Δc_it − Δc_jt = α_i + β₁Δy_it + β₂Δc_jt − β₁(1 + β₂)Δy_j + ϵ_ijt

	Income β1		External consumption β2		Adjusted R2
Belgium	0.36	**	-1.08	**	0.21
	(0.13)		(0.56)
Denmark	0.65	**	0.01		0.66
	(0.09)		(0.74)
France	0.37	**	-0.50		0.77
	(0.10)		(0.27)
Germany	1.05	**	-0.78	**	0.69
	(0.12)		(0.30)
Greece	0.24	*	-0.60		0.44
	(0.11)		(0.42)
Ireland	0.34		2.06		0.05
	(0.46)		(13.26)
Italy	0.80	**	-1.49	**	0.78
	(0.06)		(0.25)
Luxembourga	-0.19		-0.80	*	0.31
	(0.21)		(0.38)
Netherlands	0.67	**	-0.27		0.94
	(0.06)		(0.41)
United Kingdom	0.71	**	-2.95		0.37
	(0.09)		(1.84)

^aThe estimation period ends in 1991.

Notes: The estimation period is 1973–92. Standard errors (adjusted for heteroscedasticity and the implied moving average process) are reported in parentheses. One asterisk indicates that the coefficient is significantly different from zero at the 5 percent significance level, and two asterisks, at the 1 percent level, respectively. The instruments were the second lags of home and external consumption growth, income growth, and the ratio of consumption to income (saving).

It is worth noting that since all of the above tests assume that the rate of intertemporal substitution is equal across countries, an alternative interpretation of these results is that it is a variation in this characteristic that is causing the results. While acknowledging that this could be the cause, we find this explanation unlikely for several reasons. First, intertemporal rates of substitution are a feature of individual utility functions, not of national policy. There appears to be no reason to assume that the preferences of, say, German consumers vary in this dimension from those of, say, British ones. Second, the coefficients on external consumption generally have the negative signs, which would occur if capital markets were not integrated. Finally, many EC countries relied heavily on capital controls over the estimation period. Hence, the results appear to accord with casual observation.

VI. Conclusions

This paper has explored the issues of income constraints and capital mobility using data on consumption across countries. From a model that assumes that intertemporal rates of substitution are equal across countries, a procedure was proposed and executed that tests the optimality of the underlying consumption path across different countries, and identifies whether rejections were caused by excess sensitivity of consumption to income or by lack of correlation with consumption in the rest of the region. It is important to differentiate between these two underlying causes because they identify very different reasons for the failure. If the failure is caused by excess sensitivity of consumption to income, this implies that the problem may well be associated with access of individuals to capital markets. By contrast, if home consumption growth has a low correlation with external consumption growth, this implies that the problem has to do with the lack of real interest rate equalization between capital markets and not with access within them.

The results indicate that Japan is the only industrialized country in the sample for which national consumption appears to be fully integrated with the rest of the world. For the remaining countries, the main source of failure is the excess sensitivity of consumption to home income, a finding that parallels the work of Jappelli and Pagano (1989) and Bayoumi and Koujianou (1990). There are, however, important instances in which the rejections also reflect a failure of real interest rates to be equalized, most notably for the EC countries.²⁵ We conclude that greater integration of national capital markets caused by moves toward a single market and, possibly, a single currency could therefore provide potentially significant gains in economic welfare by improving consumption patterns across the European Community.

These results using consumption to measure capital market integration can be compared with the wider literature on international capital market integration. Overall, the results from this approach are closer in spirit to those using saving-investment correlations than those using nominal interest rate differentials, although our approach does not reject full capital market integration as consistently as do saving-investment correlations. We reject full capital market integration for about half the countries. In addition, even in those countries where capital markets may be well integrated, consumption still appears almost always to deviate from its optimal path because of limited access to asset markets.

One lesson from this work appears to be that even if capital is mobile between money center banks, this is not enough to ensure full integration of consumption patterns, or even to ensure capital market integration between national capital markets. It is worth stressing that the tests implemented in this paper probably capture a different type of capital mobility than that contained in a comparison of the nominal interest rates set in financial centers. This paper uses a theory that is based on the behavior of consumers, real interest rates, and annual data, and can therefore be seen as measuring the economic integration of real activity, while tests involving nominal differentials can be seen as tests of the integration of short-term capital. Given that the latter concept of capital mobility is likely more closely related to speculative behavior, the former concept is probably of more interest to those who are interested in economic “fundamentals.” It also helps to explain why Japan, which is the only country to have experienced a significant debt cycle over the period in the sense of having imported large amounts of capital to industrialize and then having repaid that capital, is the only country in our sample to exhibit a high degree of capital mobility.

Finally, a note on directions for future research. One obvious extension would be to obtain data for the pre-1973 period, in order to investigate the impact of the breakup of the Bretton Woods exchange rate system on capital mobility. Another would be to use higher frequency data, which would facilitate investigation of interesting subperiods within this more general sample. This might include the impact of the exchange rate mechanism and single market program within Europe, and the more general trend toward financial liberalization in the industrial countries since the late 1970s.