Limiting Volatile Short-Term Capital Flows

Foreign exchange markets are very liquid and react very quickly to shocks. Because of factors such as price and wage rigidities and investment irreversibility, the real economy has a slower speed of adjustment. Authors such as Tobin (1978) and Dornbusch (1986) argue that this differential speed of adjustment, together with exogenous excess volatility in financial markets, may induce excess exchange rate volatility (overshooting, bubbles, etc.), with negative effects on real economic activity. Tobin proposed “throwing sand in the wheels” of short-run capital flows by imposing a uniform tax on all foreign exchange transactions, thereby discouraging very short-term capital flows, but with negligible effects on long-run ones.⁶ Dornbusch (1986) suggests the adoption of measures such as dual exchange rate systems, which are able to shield, at least partially, the real economy from the vagaries of short-term financial market behavior. Tornell (1990) presents a model in which Tobin taxes can help increase real investment by reducing the volatility of returns on financial investment, resulting from rumors unrelated to economic fundamentals. Since real investment is assumed to be irreversible, it is discouraged by high volatility of returns.

With pegged exchange rates, unrestricted short-term capital flows may cause large variations in foreign exchange reserves, the collapse of the peg, or high interest rate variability. The turbulence experienced in late 1992 and in 1993 in countries belonging to the European Monetary System and in countries that unilaterally pegged their rate to the ECU or the deutsche mark, and the more recent developments in Mexico prove this point very effectively. According to their proponents, effective capital controls can at least mitigate these undesirable effects in the short run. Obviously crises can occur because fundamentals are out of line, as is the case when two macroeconomic policy objectives (say, domestic credit expansion and fixed exchange rates) are mutually inconsistent, as shown in the literature on speculative attacks and balance of payments crises.⁷ In the absence of capital controls, sustainability of an adjustable-peg mechanism requires large interest rate changes before realignments, to compensate asset holders for capital losses. This interest rate variability is particularly damaging in countries where the government has a large short-term public debt, or when longer-term debt instruments are indexed to short-term interest rates.⁸ In order to justify the imposition of controls, one would need to find arguments in support of the adoption of policy measures that are inconsistent with the exchange rate peg in the long run.

However, the possibility of self-fulfilling speculative attacks against a fixed exchange rate, not motivated by market fundamentals, would provide an additional justification for the imposition of capital controls: the exchange rate peg can collapse even when current fundamentals are consistent with the peg (Obstfeld (1986 and 1988)). This line of argument has been adopted recently by Eichengreen and Wyplosz (1993). Overall, the analysis suggests that governments with stronger credibility problems would be more likely targets of speculative attacks and would therefore be more likely to impose capital controls.⁹

Retention of Domestic Savings

If the private return from holding domestic instruments is below the social return, for example because of the existence of positive externalities from domestically invested capital, there is a rationale for limiting capital outflows and/or encouraging capital inflows. As observed by Mathieson and Rojas-Suarez (1993), however, restricting domestic residents’ ownership of foreign assets implies reduced portfolio diversification and increased vulnerability to domestic macroeconomic shocks.

A related argument is that a government may be willing to adopt measures that stimulate savings if savings are prevented from flowing abroad by low capital mobility or by capital controls. For example, a panel study of OECD countries by Jappelli and Pagano (1994) finds that savings are higher in countries with restrictions on household borrowing. To be effective in raising domestic savings, these restrictions also require the presence of capital controls (Pagano (1994)).

Finally, capital inflows, especially in the form of foreign direct investment, may be discouraged by countries that wish to limit foreign ownership of domestic factors of production, for political or ideological reasons.

Help for Stabilization and Structural Reform Programs

There is an extensive literature on the optimal sequencing of external sector liberalization. Authors such as Frenkel (1982), Edwards (1984, 1989), and van Wijnbergen (1990) have stressed the effects of liberalizing the capital account on the real exchange rate. In the context of an inflation stabilization plan accompanied by trade liberalization, an early opening of the capital account can cause a real appreciation, because of the high interest rates typically associated with a stabilization plan, and increased real exchange rate volatility. Both these effects would make trade liberalization more problematic.

The credibility of the stabilization plan plays a key role in determining the consequences of free capital mobility. Lack of credibility of the stabilization plan may cause capital flight and a balance of payments crisis, making the plan failure more likely. If the plan is credible, the high real interest rates typically associated with a stabilization program may cause temporary large capital inflows. If these inflows are sterilized, domestic interest rates would remain high, thereby encouraging further inflows, and the central bank would incur a quasi-fiscal cost, because the return on foreign exchange reserves would be below the return on assets denominated in domestic currency.¹⁰ If no sterilization occurs, the increase in the money supply could jeopardize the control of inflation. Finally, letting the nominal exchange rate appreciate may hamper a trade reform aiming at lower barriers to imports.¹¹

The appropriate response to a surge in capital inflows cannot be determined without a close examination of the causes of the inflow, and may differ depending on the composition of inflows. Clearly, portfolio investment is more reversible than foreign direct investment. Indeed, proponents of foreign exchange restrictions tend to propose measures that hamper short-term flows, rather than long-term ones (see previous subsection). From a political economy point of view, one should consider the relation between political stability, government preferences, and credibility. Again, governments with lower initial credibility may be those with stronger incentives to introduce capital controls (see above). Overall, this motivation for the introduction of capital controls may be more applicable to developing countries.

Maintenance of the Domestic Tax Base and Distributional Issues

Restrictions on foreign currency holdings reduce the ability of domestic agents to avoid the inflation tax. The government can impose measures such as high reserve requirements that raise the demand for money and, therefore, the inflation tax base. As stressed by Drazen (1989), these measures can have negative long-run effects, because they may discourage capital accumulation by raising the interest rates that banks charge on loans.¹² In order to maintain seigniorage revenue after the dismantling of barriers to trade and capital flows, Brock (1984) argues, the central bank can impose a reserve requirement on foreign capital inflows and a prior import deposit.

More generally, foreign exchange restrictions are often accompanied by various types of financial market restrictions, such as controls on interest rates, constraints on banks’ portfolios, and credit controls. These measures can be used by the government to reduce the cost of domestic borrowing. Giovannini and de Melo (1993) compare the domestic and foreign costs of borrowing for a sample of developing countries, and show that this source of revenue can be substantial. Even in the absence of financial repression, effective controls on capital outflows may allow the government to reduce the cost of financing its debt by lowering real interest rates. Aizenman and Guidotti (1994) present a second-best argument in favor of this policy choice when tax distortions are high and domestic debt is large.¹³

The links between financial development and economic performance, highlighted in the work of Goldsmith (1969) and McKinnon (1973), among others, have been recently re-examined by Roubini and Sala-i-Martin (1992 and 1995) and King and Levine (1993). These authors underline that an underdeveloped and repressed financial system allows the government to finance public expenditure more easily when the tax system is inefficient, but it may be an obstacle to growth.

Giovannini (1988) and Razin and Sadka (1991) argue that taxation of domestic capital can induce capital flight when it is difficult to tax foreign-source income, thus making a case for controls on capital outflows. According to Giovannini, the distortions introduced by capital controls may be less than those implied by the inability to tax foreign-source income. Razin and Sadka show that when taxing foreign-source income is impossible, it may be optimal to impose a restriction on capital exports in order to generate overinvestment domestically.

Alesina and Tabellini (1989) examine the distributional aspects of tax policy and capital controls when economic agents are heterogeneous. The authors view capital controls as a form of limiting holdings of foreign assets that are nontaxable. Individuals would accumulate foreign assets to avoid the risk of future domestic taxation. In their model, there are two social groups—“workers” and “capitalists”—and two parties that represent them. The workers’ source of income is labor (they cannot own domestic capital), while the capitalists’ income comes from capital holdings. Under reasonable assumptions about initial endowments and distribution it is shown that fear of a future workers’ government may induce capitalists to export capital. Among other things, the paper shows that once homogeneity between private agents is removed, distributional factors become important in the evaluation of foreign exchange restrictions.¹⁴ Epstein and Schor (1992) use a Keynesian framework to argue that capital controls enhance monetary policy autonomy, and that an expansionary monetary policy in the presence of controls can raise employment and capacity utilization by reducing interest rates, thus favoring workers and damaging financial sector interests.

Capital Controls and Inflation

The discussion in Section I and the theoretical literature provide ambiguous predictions of the impact of capital controls on the rate of inflation. According to the optimal taxation literature, inflation is a tax that is chosen optimally together with other tax instruments: tax rates are set at values that equalize the marginal distortions of the different tax instruments, weighted by the size of the respective tax bases. By limiting the scope for portfolio diversification and facilitating the imposition of financial repression measures, capital controls can enlarge the inflation tax base. As a result, the optimal rate of inflation may rise or fall, while seigniorage revenue would unambiguously rise (for given distortions).¹⁸

In other work, inflation is viewed as a residual form of taxation, given other taxes and bond financing. In the latter view, financial liberalization may reduce the tax base for the inflation tax, implying that the same financing gap would have to be covered by a higher rate of inflation for a given amount of bond financing. This effect would be enhanced by the necessity to raise revenue to finance the higher interest payments on government debt.¹⁹

In this section we present results of regressions in which the dependent variable is the rate of inflation. The results are qualitatively similar to those that obtain when the share of seigniorage over total GDP or over total tax revenue is used (results are available from the authors).

In order to proxy, albeit roughly, for the intensity of capital controls, we included restrictions to current account transactions among the regressors, with the idea that current account restrictions may make it more difficult to evade capital controls through leads and lags in import and export billing. We also included among the regressors two measures of central bank independence, taken from Cukierman (1992). The first variable, LEGAL, measures the legal independence of the central bank: higher numbers imply a more independent central bank. According to Cukierman (1992), this is a good measure of central bank independence for industrial countries. The second variable, TURNOVER, measures the turnover in central bankers, which is considered to be negatively correlated with the degree of independence of the central bank. According to Cukierman (1992), this is a good measure of central bank independence for developing countries. Both measures change every ten years for most countries, and are constant across time for the rest. We expect inflation to be lower in countries with a more independent central bank, because central bank independence is associated with the choice of central bankers that are more inflation averse than is the government.

We introduce three time-varying political variables. The first variable, LEFT, captures the political orientation of the government: it takes the value of one when a democratic left-wing government is in power, and zero otherwise. “Partisan” models of monetary policy (Alesina (1987)) predict that inflation should be higher under left-wing governments, which are more concerned about output and employment performance than about inflation. The second variable, COAL, takes the value of one when a coalition government is in power and zero otherwise. Coalition governments may find it more difficult to reach agreement on tax increases and may therefore rely more heavily on seigniorage. The third political dummy, NODEM, takes the value of one when the country is not a democracy, and zero otherwise. There is no a priori presumption about its effect on inflation.

We also introduce two political variables that are country specific but time invariant. The first, TCHANGE, equals the total number of government changes in the period 1950–82.²⁰ The second, COUP, gives the total number of successful coups in the same period. Both variables are proxies for the degree of political instability. As several studies have shown, inflation and political instability are strongly correlated, although the issue of causality is not firmly resolved (see Cukierman, Edwards, and Tabellini (1992); and Roubini and Ozler (1994)).

Finally, we include among the regressors three macroeconomic variables. The first is the (log of) initial level of income, GDP1966. The expected sign on this variable is a priori ambiguous: on the one hand, countries with lower income have a less efficient tax system and may therefore rely more heavily on the inflation tax; on the other hand, the degree of monetization of the economy is lower in poorer countries. The second variable is the lagged share of the budget balance to GDP (negative numbers indicate a deficit). We expect inflation to be higher in countries with large budget deficits. The third variable is the degree of openness of the economy, OPEN, measured as the ratio of exports plus imports to GDP. According to Romer (1993), countries that are more open to international trade have lower incentives to use unexpected inflation to stimulate economic activity, because of the negative effects of real depreciation.²¹ We therefore expect to find a negative correlation between inflation and openness. The last explanatory variable is a dummy variable identifying the type of exchange rate regime, EXR. It takes the value of one when the exchange rate is fixed or managed, and zero when it is floating. We expect inflation to be lower in countries that peg or manage their exchange rate, because of the constraints that such a regime imposes on monetary policy.

The inflation equations were estimated using three different specifications. The regressions were first run on the pooled cross-sectional time-series data, using ordinary least squares (OLS) (column (1) in Table 2). The residuals from this regression were then used to calculate appropriate weights, and the subsequent regressions were run using weighted least squares (WLS). This procedure was used given the difference in the variance of inflation across the countries in our sample. The second column in Table 2 reports results from the WLS regressions using a dummy variable for each year in the sample. Time dummies help control for the effects of omitted variables that are time varying but have the same effect across countries. Finally, the third column reports results from WLS regressions that include country-specific dummies as well as time dummies. These country dummies are introduced in order to control for omitted variables that are country specific but constant through time. It should be noted that introducing country-specific fixed effects implies that it is impossible to identify the coefficient on variables that are country specific and time invariant, such as the initial level of income and the TCHANGE and COUP variables.²²

Table 2.

Determinants of Inflation, 1966–89

Note: t-statislics are in parentheses. Regression (1) is ordinary least squares. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

Table 2.

Determinants of Inflation, 1966–89

	Whole Sample						Induslrial Countries						Developing Countries
	(1)		(2)		(3)		(1)		(2)		(3)		(1)		(2)		(3)
CAPCON	2.51	(3.21)	2.89	(8.96)	2.02	(4.12)	3.67	(7.37)	2.56	(8.19)	1.94	(3.92)	1.49	(0.83)	0.90	(0.53)	3.06	(3.19)
CURRCON	1.64	(1.39)	2.14	(4.28)	1.17	(2.41)	2.52	(3.21)	2.16	(3.98)	1.19	(2.19)	0.53	(0.28)	1.58	(0.85)	-1.40	(-1.26)
MULTER	5.38	(4.55)	3.22	(6.77)	3.04	(5.02)	-0.03	(-0.04)	0.57	(1.18)	0.55	(1.37)	6.98	(3.73)	5.35	(2.94)	4.49	(4.02)
LEGAL	-8.83	(-4.11)	-3.66	(-4.06)	-8.13	(-1.03)	-1.61	(-1.16)	-4.11	(-4.77)	6.65	(-1.10)	…	…	…	…	…	…
TURNOVER	13.63	(4.84)	10.20	(6.47)	3.88	(1.82)	…	…	…	…	…	…	6.42	(219)	7.38	(2.54)	2.33	(1.00)
LEFT	2.82	(3.18)	0.93	(2.84)	0.42	(1.35)	1.30	(3.15)	0.70	(2.56)	0.59	(2.20)	3.27	(1.62)	1.88	(0.93)	-0.14	(0.11)
NODEM	6.95	(5.72)	3.72	(5.38)	0.67	(0.73)	…	…	…	…	…	…	7.52	(5.18)	6.47	(4.73)	1.73	(1.32)
COAL	-0.01	(-0.02)	1.09	(2.85)	0.18	(0.35)	0.71	(1.22)	1.22	(3.46)	0.99	(2.10)	-4.86	(-3.01)	-4.58	(-2.96)	-2.46	(-2.16)
TCHANGE	0.05	(1.17)	0.01	(0.23)	…	…	0.03	(0.89)	0.03	(1.13)	…	…	0.11	(0.88)	0.15	(1.22)	…	…
TCOUP	0.91	(3.44)	-0.01	(-0.39)	…	…	…	…	…	…	…	…	0.51	(1.78)	0.47	(1.69)	…	…
GDP1966	2.56	(3.86)	0.81	(2.52)	…	…	-1.57	(-1.39)	-0.80	(-1.08)	…	…	4.29	(4.30)	4.91	(4.98)	…	…
DEFY (1)	-0.25	…	0.02	(0.67)	-0.05	(-1.24)	-0.17	(-2.36)	-0.08	(-1.43)	0.01	(-0.12)	-0.23	(-1.74)	-0.10	(-0.73)	-0.07	(-1.26)
OPEN	-0.01	(-1.44)	-0.02	(-1.34)	-0.04	(-2.38)	0.02	(1.74)	-0.01	(-1.69)	0.02	(-0.90)	-0.02	(-2.43)	-0.04	(-4.33)	-0.05	(-2.08)
EXR	-7.45	(-7.81)	-3.42	(-7.42)	-2.92	(-6.01)	-4.65	(-7.09)	-3.13	(-6.98)	1.87	(-4.09)	-10.30	(-6.71)	-6.91	(-3.93)	-4.13	(-4.45)
R2	0.28		0.36		0.56		0.36		0.50		0.62		0.25		0.29		0.59
Number of observations	1.061		1.061		1.061		468		468		468		607		607		607

Note: t-statislics are in parentheses. Regression (1) is ordinary least squares. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

View Table

Table 2.

Determinants of Inflation, 1966–89

	Whole Sample						Induslrial Countries						Developing Countries
	(1)		(2)		(3)		(1)		(2)		(3)		(1)		(2)		(3)
CAPCON	2.51	(3.21)	2.89	(8.96)	2.02	(4.12)	3.67	(7.37)	2.56	(8.19)	1.94	(3.92)	1.49	(0.83)	0.90	(0.53)	3.06	(3.19)
CURRCON	1.64	(1.39)	2.14	(4.28)	1.17	(2.41)	2.52	(3.21)	2.16	(3.98)	1.19	(2.19)	0.53	(0.28)	1.58	(0.85)	-1.40	(-1.26)
MULTER	5.38	(4.55)	3.22	(6.77)	3.04	(5.02)	-0.03	(-0.04)	0.57	(1.18)	0.55	(1.37)	6.98	(3.73)	5.35	(2.94)	4.49	(4.02)
LEGAL	-8.83	(-4.11)	-3.66	(-4.06)	-8.13	(-1.03)	-1.61	(-1.16)	-4.11	(-4.77)	6.65	(-1.10)	…	…	…	…	…	…
TURNOVER	13.63	(4.84)	10.20	(6.47)	3.88	(1.82)	…	…	…	…	…	…	6.42	(219)	7.38	(2.54)	2.33	(1.00)
LEFT	2.82	(3.18)	0.93	(2.84)	0.42	(1.35)	1.30	(3.15)	0.70	(2.56)	0.59	(2.20)	3.27	(1.62)	1.88	(0.93)	-0.14	(0.11)
NODEM	6.95	(5.72)	3.72	(5.38)	0.67	(0.73)	…	…	…	…	…	…	7.52	(5.18)	6.47	(4.73)	1.73	(1.32)
COAL	-0.01	(-0.02)	1.09	(2.85)	0.18	(0.35)	0.71	(1.22)	1.22	(3.46)	0.99	(2.10)	-4.86	(-3.01)	-4.58	(-2.96)	-2.46	(-2.16)
TCHANGE	0.05	(1.17)	0.01	(0.23)	…	…	0.03	(0.89)	0.03	(1.13)	…	…	0.11	(0.88)	0.15	(1.22)	…	…
TCOUP	0.91	(3.44)	-0.01	(-0.39)	…	…	…	…	…	…	…	…	0.51	(1.78)	0.47	(1.69)	…	…
GDP1966	2.56	(3.86)	0.81	(2.52)	…	…	-1.57	(-1.39)	-0.80	(-1.08)	…	…	4.29	(4.30)	4.91	(4.98)	…	…
DEFY (1)	-0.25	…	0.02	(0.67)	-0.05	(-1.24)	-0.17	(-2.36)	-0.08	(-1.43)	0.01	(-0.12)	-0.23	(-1.74)	-0.10	(-0.73)	-0.07	(-1.26)
OPEN	-0.01	(-1.44)	-0.02	(-1.34)	-0.04	(-2.38)	0.02	(1.74)	-0.01	(-1.69)	0.02	(-0.90)	-0.02	(-2.43)	-0.04	(-4.33)	-0.05	(-2.08)
EXR	-7.45	(-7.81)	-3.42	(-7.42)	-2.92	(-6.01)	-4.65	(-7.09)	-3.13	(-6.98)	1.87	(-4.09)	-10.30	(-6.71)	-6.91	(-3.93)	-4.13	(-4.45)
R2	0.28		0.36		0.56		0.36		0.50		0.62		0.25		0.29		0.59
Number of observations	1.061		1.061		1.061		468		468		468		607		607		607

Note: t-statislics are in parentheses. Regression (1) is ordinary least squares. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

In order to avoid capturing the effects of outliers, the observations on the dependent variable were restricted to inflation rates below 80 percent.²³ Results for the inflation regressions are presented in Table 2, with the first three columns reporting results using the entire sample. They show that capital controls, current account restrictions, and multiple exchange rate practices are associated with higher rates of inflation. Regressions (1) and (2) also suggest that inflation is higher in countries that have a high turnover of central bankers and a lower degree of legal independence. The coefficients on the central bank independence variables are not significant in the regressions that include fixed country effects (regression (3)). The reason is that the estimate captures only the effects of the time-series variation in the degree of independence, which is very small in the data (for all countries, the index of independence changes only three times or less in the sample). The political variables LEFT and NODEM have a positive sign and are significant in regressions (1) and (2), but not in the regression with country-specific effects. The coefficient on the initial level of income is positive and significant—this result is consistent with the monetization effect outweighing the impact of a less efficient tax system. The coefficient on the lagged budget deficit is significant with the expected sign only in the OLS regression. This result is not too surprising, given the known difficulty in finding an empirically robust correlation between inflation and fiscal imbalances. With regard to external sector variabies, the results show that inflation is significantly lower in countries that are more open and that manage their exchange rate, in accordance to the theory.

We subsequently divided the sample into industrialized and developing countries (see list of countries in Appendix I). For industrialized countries we again find evidence that inflation is higher in countries with capital controls and current account restrictions, while the coefficient on multiple currency practices is not statistically significant. As predicted, inflation tends to be higher under left-wing and coalition governments, as well as when the central bank is less independent. Inflation is lower in countries that manage the exchange rate, but there is no evidence here that openness is associated with lower inflation.

Results for developing countries are presented in the last three columns of Table 2. Not surprisingly, the overall fit of the regressions is worse than for industrial countries. We find that the coefficient on the MULTER dummy variable is positive and significant in all the panel regressions. Also, the capital controls variable is statistically significant in the regressions with country effects. Countries with more turnover of central bankers and with capital controls in place have experienced higher inflation; furthermore, inflation tends to be higher in nondemocratic regimes. Our measures of political instability (COUP, TCHANGE, and COAL) do not provide evidence in favor of a positive link with inflation. Interestingly, developing countries with higher initial income per capita in 1966 have experienced higher inflation (after controlling for the effects of the other explanatory variables). A possible explanation is that poorer developing countries are not fully monetized, and that this effect dominates the seigniorage effect that works through the impact of a less developed tax system on the choice of revenue instruments. The coefficient on external sector variables are significant and with the expected sign in all regressions: countries that are more open and that manage the exchange rate tend to have lower inflation rates.

A similar set of regressions was run using nonoverlapping five-year averages, instead of annual values, for all variables. This procedure reduces the serial correlation problems. The results, not presented for reasons of space, are consistent with those presented in Table 2.

Capital Controls and Real Interest Rates

In the theoretical model, capital controls drive a wedge between domestic and world interest rates. In principle, capital controls may be used either by a country that wants to maintain interest rates that are lower those prevailing on world markets without experiencing capital outflows or by a country that seeks to maintain higher interest rates without experiencing capital inflows. We therefore first considered whether real interest rates differ systematically between countries that impose capital controls and countries that do not. Our sample includes three interest rate measures: interest rates on government bonds, loan rates, and deposit rates, all taken from the IMF’s International Financial Statistics. For reasons of space, we report only the results of regressions that use government bond yields as the dependent variable. Real (ex-post) rates are calculated by subtracting the rate of inflation from the nominal interest rate. Since data on interest rates are not available in a consistent and uniform fashion for the developing countries in our sample, we focus exclusively on industrialized countries.

Among the explanatory variables we include the three measures of controls—CAPCON, CURRCON, and MULTER. The degree of central bank independence, measured by the variable LEGAL, and the political variables LEFT, COAL, and TCHANGE are also included. The first set of regressions comprises data from 1960 to 1989, and therefore excludes MULTER (for which data are available only from 1966). These regressions include the lagged budget balance as an explanatory variable. Ideally, one would want to include the stock of public debt: however, data on public debt for the 1960s are not consistently available for the countries in our sample.

The results, presented in Table 3, show that countries with capital account and current account restrictions have lower real interest rates. One interpretation of the latter finding is that other forms of restrictions on foreign exchange transactions proxy for the intensity of capital controls, and more intensive controls are associated with lower real interest rates. Another, possibly complementary, explanation is that exchange restrictions are capturing the degree of government-imposed distortions, such as financial repression. The coefficient on the degree of central bank independence is positive and significant; countries with a more independent central bank have higher real interest rates. The evidence on political variables is less strong, but there is some evidence that real interest rates tend to be higher under left-wing governments.

Table 3.

Determinants of Real Interest Rates in Industrial Countries. 1960–89

Note: t-statistics in parentheses. Regression (1) is OLS. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

Table 3.

Determinants of Real Interest Rates in Industrial Countries. 1960–89

	(1)		(2)		(3)
	1960–89
CAPCON	-1.87	(-5.90)	-1.07	(-5.12)	-1.11	(-2.89)
CURRCON	-1.48	(-3.02)	-1.13	(-3.50)	-0.71	(-2.01)
LEGAL	2.18	(2.33)	1.93	(3.51)	8.68	(2.13)
LEFT	0.17	(0.53)	0.60	(3.07)	0.80	(4.11)
COAL	0.05	(-0.15)	-0.36	(-1.52)	-0.75	(-2.16)
TCHANGE	0.02	(1.00)	-0.00	(-0.17)
GDP1960	1.27	(2.05)	0.31	(5.56)
DEFY(1)	-0.05	(-1.10)	0.02	(0.75)	0.01	(0.14)
EXR	1.28	(2.85)	1.49	(5.01)	0.48	(1.39)
${\overline{R}}^2$	0.16		0.67		0.72
Number of observations	541		541		541
	1970–89
CAPCON	-1.74	(-4.19)	-0.76	(-3.08)	-0.75	(-2.02)
CURRCON	-4.35	(-4.69)	-3.56	(-7.63)	-1.64	(-3.05)
MULTER	-1.67	(-3.03)	-0.71	(-1.81)	-1.49	(-2.95)
LEGAL	4.03	(3.16)	2.38	(2.98)	2.08	(2.85)
LEFT	0.89	(2.18)	0.99	(3.83)	1.10	(4.60)
COAL	0.54	(1.07)	0.36	(1.27)	-0.20	(-0.59)
GDP1970	-0.07	(-0.08)	0.20	(2.90)
DEBTGY(1)	5.95	(5.61)	2.32	(2.84)	-1.39	(-1.29)
EXR	1.09	(2.16)	0.83	(2.64)	0.30	(0.63)
${\overline{R}}^2$	0.30		0.74		0.81
Number of observations	356		356		356

Note: t-statistics in parentheses. Regression (1) is OLS. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

View Table

Table 3.

Determinants of Real Interest Rates in Industrial Countries. 1960–89

	(1)		(2)		(3)
	1960–89
CAPCON	-1.87	(-5.90)	-1.07	(-5.12)	-1.11	(-2.89)
CURRCON	-1.48	(-3.02)	-1.13	(-3.50)	-0.71	(-2.01)
LEGAL	2.18	(2.33)	1.93	(3.51)	8.68	(2.13)
LEFT	0.17	(0.53)	0.60	(3.07)	0.80	(4.11)
COAL	0.05	(-0.15)	-0.36	(-1.52)	-0.75	(-2.16)
TCHANGE	0.02	(1.00)	-0.00	(-0.17)
GDP1960	1.27	(2.05)	0.31	(5.56)
DEFY(1)	-0.05	(-1.10)	0.02	(0.75)	0.01	(0.14)
EXR	1.28	(2.85)	1.49	(5.01)	0.48	(1.39)
${\overline{R}}^2$	0.16		0.67		0.72
Number of observations	541		541		541
	1970–89
CAPCON	-1.74	(-4.19)	-0.76	(-3.08)	-0.75	(-2.02)
CURRCON	-4.35	(-4.69)	-3.56	(-7.63)	-1.64	(-3.05)
MULTER	-1.67	(-3.03)	-0.71	(-1.81)	-1.49	(-2.95)
LEGAL	4.03	(3.16)	2.38	(2.98)	2.08	(2.85)
LEFT	0.89	(2.18)	0.99	(3.83)	1.10	(4.60)
COAL	0.54	(1.07)	0.36	(1.27)	-0.20	(-0.59)
GDP1970	-0.07	(-0.08)	0.20	(2.90)
DEBTGY(1)	5.95	(5.61)	2.32	(2.84)	-1.39	(-1.29)
EXR	1.09	(2.16)	0.83	(2.64)	0.30	(0.63)
${\overline{R}}^2$	0.30		0.74		0.81
Number of observations	356		356		356

Note: t-statistics in parentheses. Regression (1) is OLS. Regression (2) is weighted least squares, time dummies. Regression (3) is weighted least squares, time + country dummies.

The second set of regressions in Table 2 refers to the period 1970–89, and includes the (lagged) ratio of domestic government debt to GDP (DEBTGY{1}) as an explanatory variable. Results are similar to those for the period 1960–89: capital controls, current account restrictions, and multiple currency practices are associated with lower real interest rates. The debt variable is significant and with the expected positive sign in regressions (1) and (2), but is insignificant and with the wrong sign once we control for country-specific effects. Results are qualitatively similar if we use different real interest rate measures, such as real loan or deposit rates. Overall, the results are in line with those obtained by Cukierman and others (1993), who regress average real interest rates on the degree of central bank independence for a sample of industrial and developing countries. For both samples they find that real rates are significantly higher when the central bank is more independent.

It should be noted that these results do not have implications about the effectiveness of capital controls. Financial repression measures are widespread in some countries with capital controls, so that the interest rates we are measuring are not the relevant ones in determining the effective degree of arbitrage between domestic and foreign financial markets.

Capital Controls and Growth

Once again, theory does not provide unambiguous predictions for the effects of capital controls on growth. On the one hand, capital controls may stimulate capital accumulation and (temporarily) raise the growth rate by lowering real interest rates. On the other hand, in an endogenous growth framework lower real interest rates imply a lower rate of growth. The empirical literature on economic growth is immense.²⁴ A number of empirical studies have examined the relation between growth in real income per capita and political and financial repression variables. For example, Alesina and Perotti (1992) document the relationship between growth and political instability, democracy, and income distribution. Roubini and Sala-i-Martin (1992) find that countries with a more distorted trade and financial system tend to grow more slowly. Cukierman and others (1993) find that central bank independence is positively correlated with growth rates in developing countries.

The empirical growth literature uses mainly cross-sectional estimates or panel estimates that use five- or ten-year averages. Here we present the results of panel regressions that use five-year nonoverlapping averages of all variables. In addition to our data, we use the Barro and Lee (1994) data set, which contains data on schooling and educational attainment that are shown to be correlated with economic growth.

The dependent variable is the rate of growth of real income per capita, taken from Summers and Heston (1991). The independent variables are the three exchange restrictions dummies (CAPCON, CURRCON, and MULTER); the two measures of central bank independence (LEGAL and TURNOVER); the political dummy (NODEM); the average size of the black market premium on foreign exchange (BMP); and the degree of openness of the economy (OPEN). In line with the empirical growth literature, we also include (the log of) initial income (GDP) and the initial level of education (SYRM) among the regressors. The latter is measured using years of secondary schooling of the male population, which is found to be systematically correlated with growth in Barro and Lee (1994). The convergence hypothesis implies that countries with a lower initial level of income should—ceteris paribus—grow faster than richer countries. Finally, we use two measures of government consumption, GCONS and GOVSH. The first one measures the share over real GDP of real government consumption net of spending on defense and education, but is not available for 1985–89. This is the measure used in Barro and Lee (1994), which is found to be negatively correlated with growth. The second, GOVSH, measures the share of real government consumption over real GDP, and is available for the whole sample.

The growth regressions were run using instrumental variable estimation. We use own lagged values as instruments for OPEN and GCONS (GOVSH), while the other variables are their own instruments. Time dummies were also included. Results for the whole sample are presented in Table 4. Regressions (1) and (3) ((2) and (4)) present results without (with) regional dummies. Regressions (3) and (4) also exclude the two measures of central bank independence, TURNOVER and LEGAL. The initial level of income enters significantly in the regression, lending support to the conditional convergence hypothesis. The coefficient on the black market premium is also significant and negative: countries with a more distorted foreign exchange system tend to grow more slowly. The evidence on other determinants of growth in our sample is weaker. For example, the openness variable is significant and with the expected positive sign only in regressions (3) and (4). The coefficient on the capital controls variable is positive in all regressions, but its significance declines once we introduce continental dummies. The coefficient on current account restrictions is negative and significant in regressions (3) and (4), but is insignificant in the others. It should be noted, however, that the black market premium (BMP) is positively correlated with our foreign exchange restrictions dummies, and in particular with CURRCON. Omitting BMP tends to raise the significance of the current account dummy.

Table 4.

Determinants of Growth, 1966–89

(Whole sample, five-year averages)

Note: t-statistics, calculated using White’s heteroscedasticity-consistent standard errors, are in parentheses.

Table 4.

Determinants of Growth, 1966–89

(Whole sample, five-year averages)

	(1)		(2)		(3)		(4)
SYRM	0.002	(1.14)	0.001	(0.90)	0.002	(1.33)	0.001	(0.36)
LRGDP	-0.012	(-3.56)	-0.012	(-2.64)	-0.008	(-3.07)	-0.012	(-3.40)
NODEM	0.005	(0.83)	0.006	(1.10)	0.003	(0.55)	0.005	(1.00)
LOGBMP	-0.032	(-4.37)	-0.029	(-3.84)	-0.026	(-2.47)	-0.021	(-2.09)
GCONS					-0.075	(-1.88)	-0.050	(-1.36)
GOVSH	-0.144	(-2.56)	-0.084	(-1.13)
OPEN	0.004	(1.18)	0.005	(1.25)	0.010	(2.81)	0.011	(2.70)
TURNOVER	-0.017	(-1.82)	-0.009	(-0.89)
LEGAL	0.009	(1.14)	-0.001	(-0.08)
CAPCON	0.006	(1.73)	0.003	(0.67)	0.010	(3.02)	0.006	(1.75)
CURRCON	-0.007	(-1.45)	-0.006	(-1.10)	-0.010	(-2.32)	-0.010	(-2.09)
MULTER	0.003	(0.85)	0.004	(1.23)	-0.004	(-0.92)	-0.000	(-0.01)
AFRICA			-0.017	(-1.44)			-0.021	(-3.01)
INDUST			0.003	(0.42)			0.005	(0.91)
WESHEM			-0.01	(-1.48)			-0.013	(-2.40)
${\overline{R}}^2$	0.35		0.38		0.28		0.32
Number of observations	181		181		238		238

Note: t-statistics, calculated using White’s heteroscedasticity-consistent standard errors, are in parentheses.

View Table

Table 4.

Determinants of Growth, 1966–89

(Whole sample, five-year averages)

	(1)		(2)		(3)		(4)
SYRM	0.002	(1.14)	0.001	(0.90)	0.002	(1.33)	0.001	(0.36)
LRGDP	-0.012	(-3.56)	-0.012	(-2.64)	-0.008	(-3.07)	-0.012	(-3.40)
NODEM	0.005	(0.83)	0.006	(1.10)	0.003	(0.55)	0.005	(1.00)
LOGBMP	-0.032	(-4.37)	-0.029	(-3.84)	-0.026	(-2.47)	-0.021	(-2.09)
GCONS					-0.075	(-1.88)	-0.050	(-1.36)
GOVSH	-0.144	(-2.56)	-0.084	(-1.13)
OPEN	0.004	(1.18)	0.005	(1.25)	0.010	(2.81)	0.011	(2.70)
TURNOVER	-0.017	(-1.82)	-0.009	(-0.89)
LEGAL	0.009	(1.14)	-0.001	(-0.08)
CAPCON	0.006	(1.73)	0.003	(0.67)	0.010	(3.02)	0.006	(1.75)
CURRCON	-0.007	(-1.45)	-0.006	(-1.10)	-0.010	(-2.32)	-0.010	(-2.09)
MULTER	0.003	(0.85)	0.004	(1.23)	-0.004	(-0.92)	-0.000	(-0.01)
AFRICA			-0.017	(-1.44)			-0.021	(-3.01)
INDUST			0.003	(0.42)			0.005	(0.91)
WESHEM			-0.01	(-1.48)			-0.013	(-2.40)
${\overline{R}}^2$	0.35		0.38		0.28		0.32
Number of observations	181		181		238		238

Note: t-statistics, calculated using White’s heteroscedasticity-consistent standard errors, are in parentheses.