Chapter

Appendix I. Capital Mobility: The Empirical Evidence

Author(s):
Robert Kahn, Adam Bennett, María Carkovic S., and Susan Schadler
Published Date:
September 1993
Share
  • ShareShare
Show Summary Details

The scope for an independent monetary policy or sterilized intervention depends on the degree of capital mobility, that is, the responsiveness of capital flows to changes in relative rates of return. The degree of capital mobility is determined by the substitutability between assets and the existence and effectiveness of impediments to capital. When assets are perfect substitutes and there are no impediments to the movement of capital, relative rates of return will be fixed. In these circumstances, there is no scope for an independent monetary policy or sterilized intervention when exchange rates are pegged or subject to a strict rule; however, when assets are not perfect substitutes or there are effective impediments to capital flows, domestic interest rates can move to and be sustained at levels different from foreign interest rates. The purpose of this appendix is to estimate the degree of capital mobility for the six countries under review using two techniques: the estimation of an offset coefficient measuring the degree to which capital inflows offset changes in net domestic assets (NDA) and a test of movements in domestic relative to foreign interest rates.

The results suggest that during the estimation periods, the mid-1970s through 1991, each of the countries under review had some, but not complete, scope for independent monetary policy. The scope was least in Thailand and comparatively large in Egypt, Chile, and Spain. Given the trend toward greater integration of world capital markets, particularly for the countries under review, however, these results, which reflect the average experience during the estimation period, may overstate the present scope for monetary policy independence.

Offset Coefficients

An offset coefficient measures the degree to which capital inflows offset the effect on money of a change in NDA. Pioneered by Argy and Kouri (1974) and Kouri and Porter (1974), this is a long-standing method of measuring the effectiveness of sterilization operations when exchange rates are fixed. The method is based on a monetary model that includes a simple money demand function and a money supply process. The latter comprises NDA creation (a policy variable) and net capital inflows, which are assumed to be responsive to the interest differential. These components are usually combined in a reduced form expression from which the domestic interest rate is substituted out to give the following expression.

If capital is mobile, a decrease in the NDA of the central bank, resulting, for example, from the sale of bonds, will lead to an offsetting increase in net foreign asset holdings (NFA). The more mobile is capital, the greater the offset. In the limit α, the “offset coefficient,” is –1. Changes in net foreign assets may also be influenced by changes in nominal income (Y), which generate changes in the demand for money, and by changes in the foreign rate of return—a combination of the foreign rate of interest (if) and the expected change of the exchange rate (∊e)

The reduced form approach has the virtue of simplicity. If central banks have routinely sterilized over the estimation period, however, the direction of causation would run from NFA to NDA rather than the other way around and ordinary least squares estimation of the above equation would yield biased coefficients.12 In fact, for none of the countries considered was there evidence of the endogeneity of NDA over the estimation period, suggesting that ordinary least squares would give unbiased results.13

Estimates of the offset coefficient were obtained using quarterly data from 1976–91; for some countries, the estimation period was shorter owing to data limitations.14NFA and NDA were both defined from the point of view of the central bank. The foreign rate of interest (if) was represented by the London interbank offered rate (LIBOR) for three-month dollars. To proxy the expected change of the exchange rate both forward- (rational) and backward-looking models of expectations formation were tried.15 In each of these cases, however, the expectation proxy was either insignificant or had the wrong sign.16 The results reported, therefore, take the rate of return on foreign assets as the foreign interest rate alone—treating the expected change of the exchange rate as immeasurable.17

The ordinary least squares estimates are shown in Table A3. Because of problems with serial correlation of the error, one lagged value for each of the explanatory variables as well as the dependent variable was included. This largely removed the problem for all but two countries. The inclusion of lags also permits a distinction between impact and subsequent effects. In all cases, the estimated offset is negative, and in all except Chile, it is significantly larger than zero. The offset, at –0.7, is highest for Thailand. The other offsets range from –0.1 to –0.5, indicating some scope for independent monetary policy, at least over one quarter. The estimated equations suggest that the impact is little changed subsequently, except in Thailand, where the offset becomes close to complete within six to nine months.18

The Foreign Influence on Domestic Interest Rates

A second approach to the question of capital mobility looks at the strength of the relationship between domestic and foreign interest rates. When capital is highly mobile and the exchange rate is fixed, changes in this relationship should not result from changes in domestic credit conditions but should reflect fundamental changes in factors affecting the risk premium. If the exchange rate is flexible, the correspondence will be with the foreign interest rate adjusted for the expected rate of change of the exchange rate. This suggests viewing the observed domestic interest rate, it, as a weighted average of the interest rate that would exist if the economy were completely open, it* (the foreign interest rate adjusted for expected exchange rate changes, risk premiums, and transactions costs), and the interest rate that would exist if the economy were completely closed, ict (which would reflect domestic money supply and demand).19 With a number of assumptions concerning money demand and the determination of real interest rates, this approach results in the following general model:

Box 5.Data Sources and Transformations

All data are quarterly and, except GDP, were obtained from International Financial Statistics (IFS). Quarterly GDP data are either interpolated from IFS data, using industrial production benchmarks, or are staff estimates. When there is more than one short-term interest rate in IFS, the following is used: Chile, 30–39-day deposit rate; Mexico, treasury bill rate; Spain, rate on interbank operations effected through the Bank of Spain’s cable service; and Thailand, 90-day interbank rate. The inflation rate is the annualized quarterly change in the consumer price index. The rate of change of the exchange rate is the annualized quarterly change in domestic currency price of U.S. dollars. The real money supply is “Money” in IFS deflated by the consumer price index.

Net foreign assets are valued in U.S. dollars (at the end-of-period dollar exchange rate). Assuming that net foreign assets are held primarily in U.S. dollars, changes in their dollar value from the beginning of the period to the end of the period should reflect only transactions, and not revaluations. To calibrate the explanatory variables in the same units, both the change in the local value of net domestic assets and the change in nominal income, measured in local currency, were converted into U.S. dollars at the prevailing average exchange rate for the period.

where i* is the foreign rate of return and equals the foreign rate of interest, if, and the depreciation of domestic currency, ∊e; y is real income; m real money; πe the expected rate of inflation; and (L) indicates a lag operator. This specification permits an investigation of the short- and long-run relationships between domestic and foreign interest rates.

The first step in this more general approach is to regress the domestic rate of interest on the variables identified in equation (2), each entered with the current value only; expectational variables were replaced with the actual outturn during the period.20 The resulting relationships were tested for co-integration—whether each vector on the right side is co-integrated with domestic interest rates—in order to determine whether these variables move together in the long run as described by the equation. The results are shown in Table A4.21 For Chile, Mexico, and Spain the resulting vectors score reasonably highly on the Dicky-Fuller tests, suggesting they are co-integrated with domestic interest rates.22

The signs and significance of coefficients conform to prior expectations to varying degrees. In all cases the “long-run” effect of foreign interest rates is positive, although not always significant. The coefficients are largest for Chile (where they are so large as to be implausible), Mexico, and Thailand. The rate of change in the exchange rate is significant only for Chile. The insignificance of this variable for Thailand may reflect the relative stability of the baht against the dollar. The rate of inflation is significant for Chile and nearly so for Mexico. For Chile, this reflects the fact that most interest rates are indexed, even though the interest rate used here is a nominal quoted rate for short-term deposits. This may have introduced some specification problems into the model, which explain, in part, the unusually large coefficient on LIBOR. The coefficient on the real money stock, where significant, bears a negative sign—more liquidity lowers interest rates; the coefficient on income, where significant, bears a positive sign, as expected.

While the results of the co-integration tests do not strongly indicate that the long-run models assumed are always appropriately co-integrated, error correction mechanisms (ECMs) were nonetheless drawn from each of these equations and included in dynamic models where the same variables were entered in first differences. This permits the estimation of short-run dynamics, while the ECM represents the long-run relationships described in Table A4. In estimating these short-run dynamics, the question of expectations formation had to be addressed explicitly. In the case of inflation, expectations were assumed to be formed in an adaptive or backward-looking manner. Exchange rate expectations were assumed to be forward looking and were modeled along the lines of Wickens (1982). It proved impossible to obtain significant coefficients on the terms representing short-run exchange rate expectations for any of the countries.

The estimated short-run dynamics, after removing insignificant terms, are shown in Table A5. Apart from the long-run ECM effect (which is uniformly significant), foreign interest rates have significant short-run effects in Mexico and Thailand. Inflation expectations (or current price movements if this reflects indexation) are highly significant in the case of Chile and Mexico, and less so in the case of Spain.

Conclusion

The results suggest a wide range in the degree of capital mobility. In Thailand, both approaches indicate high capital mobility—the offset coefficient is large and interest rates move closely with those abroad. In Chile, an extremely low estimate for the offset coefficient suggests virtually no capital mobility, but the estimated influence of foreign interest rates on the domestic interest rate is large—in fact implausibly so. It is likely that the widespread indexation in the financial system distorted the latter result. Both approaches indicate substantial scope for monetary policy independence in Spain and Mexico as does the offset coefficient in Egypt. Both the approaches suggest that the scope for independence is greater in the short run than in the long run, although for Thailand adjustment is relatively swift.

    Other Resources Citing This Publication