Domestic and external factors in interest rate determination

Transcription

1 Applied Financial Economics, 1997, 7, Domestic and external factors in interest rate determination GUGLIELMO MARIA CAPORALE and NIKITAS PITTIS Centre for Economic Forecasting, ondon Business School, Sussex Place, Regents Park, ondon N¼1 4SA, ºK, and Department of Economics, ºniversity of Cyprus, 75 Kallipoleos Avenue, PO Box 537, Nicosia, Cyprus This paper develops a theoretical framework which allows for both domestic and external factors in the determination of interest rates. We argue that if capital controls are imposed, or if the risk premium is a function of disequilibria in the money market, domestic factors should also play a role. We assess empirically the role of both domestic monetary conditions and open economy factors in the US, Japan and Germany, France and Switzerland. Our results suggest the following. Concerning external factors, monetary policy in Germany affects significantly both US and Japanese interest rates; Germany, on the other hand, is rather responsive to developments in continental Europe, in spite of its dominance of the EMS. As for domestic influences, inflationary expectations are an important factor in explaining interest rate behaviour in Japan, and the budget deficit plays a role in the US. Also, the US monetary authorities adopt a more accommodating policy than the Bundesbank. I. INTRODUCTION This paper analyses interest rate determination in five major industrial economies, namely the US, Japan, Germany, France and Switzerland. Since the early 1960s, when the Mundell Fleming model introduced capital mobility in the form of a flow theory into the textbook analysis of open economies, a great deal of attention has been paid in the literature to the cases of perfect and zero capital mobility, even though in reality most economies can be characterized as partially open, in the sense that regardless of whether or not capital controls are imposed the domestic money market appears to affect interest rates. For this reason, the present study aims to develop a theoretical framework which allows for both domestic and external factors to which interest rates might respond. Unlike most of the existing empirical literature, we avoid the two extreme assumptions that the economy is completely closed or fully open, and we are therefore able to derive an empirical model which can be used to assess the role of both domestic monetary conditions and open economy factors in the determination of interest rates For a clear exposition, see Mundell (1968). See Edwards (1985) for an analysis of interest rate determination in the same spirit. The paper is structured as follows. Section II presents a theoretical model which analyses interest rate behaviour in an economy where it takes time for interest parity to hold, and internal monetary disequilibria also affect interest rate movements. Section III reports the estimation results and a number of tests of the importance of domestic versus external factors in interest rate determination. The implied structural parameters and the forecasting power of the empirical model are also discussed. Some conclusions are offered in Section IV. II. INTERNAL MONETARY CONDITIONS AND OPEN ECONOMY FACTORS IN INTEREST RATE DETERMINATION Generally, it is assumed that the economy is either fully closed or fully open. However, an intermediate situation in which capital controls exist and domestic monetary policy has some effect on the short-run behaviour of interest rates is typical of most economies; furthermore, it is often found that, even in countries which have not experienced capital Routledge 465

2 466 G. M. Caporale and N. Pittis controls, monetary disequilibria significantly affect interest rates. Therefore, our aim in this section is to develop a small analytical model which considers the role of both domestic and external factors in explaining interest rate behaviour. In such an economy the following expression could be used to describe interest rate determination i "λ [m!m ]# φ [i*#(e (s I )!s )!i ] (1) where m and m stand for money demand and money supply respectively, s is the log of the nominal effective exchange rate, i* represents the interest rate in the foreign country j, and 0(φ(1. It should be noted that under full open arbitrage with bilateral exchange rates there is a complex set of relations between bilateral interest rates the range of interest rates used in Equation 1 is a way of approximating an effective interest rate index for each country. The above specification encompasses the two extreme cases of an economy in which only external factors matter, corresponding to λ"0, and of one in which only domestic factors are significant, which amounts to φ"0. On the other hand, both parameters should be different from zero if both foreign and domestic factors affect interest rates. Equation 1 implies a deviation from uncovered interest parity (UIP), which can be plausibly attributed to the existence of a risk premium ρ. It can be seen as an adequate characterization of interest rate behaviour in the presence of effective capital controls, which make domestic monetary disequilibria play a role. However, under risk aversion, such a specification might be valid even if capital controls are not imposed, provided that the risk premium is a function of excess money supply (demand). It can be argued that this is a valid assumption to make on the grounds that an excess in the supply over the demand for money may signal a tightening in the future which leads to an increase in the risk premium. In other words, it is assumed that ρ "λ [m!m ]. The justification for the two terms on the right-hand side of Equation 1 is the following. In a closed economy, short-term nominal interest rates can be modelled as responding to disequilibria in the money market, i.e. to excess supply or demand for money i "λ [m!m ] (2) This implies that financial markets adjust slowly; a more general equilibrium model would be required under the assumption that goods prices react sluggishly to excess money demand, whilst a gradual adjustment in the money supply would be inconsistent with the monetary authorities controlling the money stock through a reaction function as below. We assume that demand for real money balances depends on real output y and the nominal short-term interest rate, i.e. m "αγ!βi (3) and that the money supply is controlled by the monetary authorities through a policy reaction function m "γz #δe (π I ) (4) where π is the rate of inflation, I is the agents information set, and z stands for the budget deficit. The parameter γ should have a positive sign if the government finances the deficit by issuing money, whereas if a tighter monetary policy is pursued when the budget is not balanced γ will be negative. As for the other parameter δ, if keeping inflation under control is a primary policy objective then δ will be negative, whilst if monetary policy is accommodating δ will be positive. Conversely, in an open economy the domestic interest rate should equal the world rate of interest plus the expected rate of devaluation. However, it is possible that, because of impediments to capital movements, short-run deviations from the equilibrium condition might occur, which can be expressed in the following way i "φ [i* #E (s I )!s!i ] (5) If more than one foreign interest rate affects domestic rates, the equation should be rewritten as i " φ [i*#e(s I )!s!i ] (6) Equation 6, though, still does not give any role to domestic monetary disequilibria. Under the assumptions of relative purchasing power parity (PPP), i.e. E (s I )!s " E(π I )!E(π I ), and of a unity income elasticity in the money demand equation (i.e. α"1), substitution of Equations 4 and 5 into 1 and some algebraic manipulations yield i "(λ/(1#λβ))y!(λγ)/(1#λβ)) z # φ!λδ /(1#λβ)) E(π I ) See, for example, Boughton (1988). For the theoretical foundations of so-called disequilibrium or fixed price equilibrium macroeconomics and applications to other markets, see Henderson and Quandt (1971), Peston and Quandt (1986), and Quandt (1988). The assumption that the choice between money and bonds and that between interest-bearing securities are separable in the agents utility function allows us to eliminate long-term rates from the equation (see Tobin, 1969). As for the inflation rate, it was left out of the equation for real money balances for the following reason: as this variable appears in the policy reaction function, its inclusion would not affect the number of reduced form parameters, but would result in one more structural parameter, and therefore in under-identification,

3 Factors in interest rate determination 467 # φ /(1#λβ)) r* # 1! φ /(1#λβ) i where r* is the real interest rate, given by i*!e (π I ). It should also be noted that the presence of the last term on the right-hand side means that the reduced form Equation 7 can be interpreted as a partial adjustment model. It can be shown that a sufficient condition for a unique stable equilibrium is the following (7) 1! φ /(1#λβ) (1 (8) This amounts to the interest rate exhibiting a low degree of persistence given the particular multivariate information set. In the next section we present estimates of the reduced form equation describing the behaviour of interest rates, and test the significance of the various parameters in order to assess the role of open economy factors and internal monetary conditions. Recovering the structural parameters also enables us to make some statements about policy objectives in the five countries we study. III. EMPIRICAL RESULTS The model of interest rate determination presented above was estimated for five countries (US, Japan, Germany, France and Switzerland). All the series in our dataset are from the OECD Main Economic Indicators. Short-term rates are either Treasury Bill rates or 3-month interbank rates. Consumer price indices were used to compute the inflation rate. The regressor for output is nominal GDP detrended. Finally, the budget deficit variable was defined as the ratio of government expenditure minus tax revenue to GDP. Instrumental variable (IV) methods were required for the estimation of Equation 7 as one of the regressors is expected inflation. The results are reported in Table 1. The instrument set used in the first stage to generate a series for expected inflation included lagged inflation and lagged money growth. In the case of the US, Japan and Germany, an MA(q) process was estimated for the error term since autocorrelation was detected in the residuals; this was not necessary for France and Switzerland. In the case of the US, both German and Japanese interest rates were initially chosen to represent the role of open economy factors. Similarly, the interest rate equations for Japan, France and Switzerland were specified to include both German and US rates. On the contrary, German Table 1. ºnrestricted I» estimates of the partial adjustment interest rate equations with an MA(q) error specification. Dependent variable: i Constant !0.002 (0.008) (0.002) (0.005) (0.0023) (0.004) Domestic y 0.201!0.035! (0.037) (0.002) (0.022) (0.042) (0.037) Domestic i (!1) (0.108) (0.047) (0.073) (0.069) (0.094) r (0.047) (0.080) (0.048) (0.057) r! (0.065) (0.055) r (0.070) r (0.072) (0.034) (0.081) (0.120) Domestic z! (0.152) (0.080) Domestic π ! (0.080) (0.031) (0.061) (0.403) (0.167) R LM(4) p-value ARCH(4) p-value (i) Standard errors in parentheses. (ii) No MA(q) specification was required in the case of France and Switzerland.

4 468 G. M. Caporale and N. Pittis Table 2. ¹esting alternative hypotheses: the role of domestic versus external factors in interest rate determination. ¼ald test on the significance of the regressors (p-value) r "r "0 0.04* r " Domestic y"z"π" * 0.011* 0.000* * Domestic y"z" * * * Domestic π" * 0.000* 0.000* r "r " * 0.000* r " * r "0 0.03* 0.04* 0.000* 0.000* r "r "r " * r "r " r " * z " An asterisk indicates rejection of the null at the 5% level. interest rates were postulated to react not only to US and Japanese interest rates, but also to a weighted average of other European interest rates. We selected one EMS and one non-ems country; more specifically, we used French and Swiss real rates with relative weights 0.7 and 0.3, which yields: r "0.7r #0.3r. Alternative definitions of European rates gave similar results. The inclusion of the variable r in the German equation was decided on econometric grounds, although it is questionable whether the Bundesbank is responsive to pressures coming from other European countries towards a looser German monetary stance, given the common finding of German dominance in the EMS (see for example Giavazzi and Giovannini, 1988, Artis and Nachane, 1990; Caporale and Pittis, 1993). Wald deletion tests of the significance of the various regressors are presented in Table 2. It appears that, in the case of the US, all domestic fundamentals (income, budget deficit and inflation) play an important role in interest rate determination. Of the external factors, only the German real interest rate is significant. As for Germany, there is no evidence that the evolution of the budget deficit affects interest rates, whilst output, expected inflation and the constructed real interest rate variable all play a role. External factors are also significant. In the case of Japan we find no role for GDP or US real rates, although German rates are significant; of the domestic variables, only expected inflation seems to affect interest rates. Concerning France, again we find a role for all domestic fundamentals; US and German real rates both play a role. Finally, domestic factors are not significant in Switzerland, whilst the German real rate is significant. The statistically insignificant regressors were then dropped and a restricted version of the model was estimated (see Table 3). Table 4 reports the implied long-run elasticities (LRE), which can be computed from the restricted reduced form equations. These elasticities were defined as α/(1!ρ), where i"α #α π#βr*#2#ρi. Alternative estimates are provided in Table 5, where we show that the series are cointegrated and then appeal to the superconsistency theorem (see Stock, 1987) to argue that the estimated parameters can be legitimately interpreted as long-run elasticities. They should not differ significantly from those retrieved from the reduced form dynamic equations, which is confirmed by a comparison between Tables 4 and 5. The cointegrating vectors are given by i "c #c y #c r #c E (π I ) #c r (9) i "d #d y #d E (π I ) #d r (10) i "k #k E (π I ) #k r (11) It is interesting to note that inflation has the biggest impact on interest rates in the long run in the case of France and Germany, and the least impact in the US. Also, German rates affect Japanese rates more than US rates, whilst they react only to European rates. The dominant external factor for French interest rates is the German real rate, the elasticity being almost equal to one. Similarly, the German rate appears to be the main driving force of Swiss interest rates. Finally, the long-run elasticity with respect to output is highest in Germany. Where possible, the estimated coefficients from the restricted reduced form equations were then used to recover Other studies, however, conclude that a weak version of the German Leadership Hypothesis (GLH) is a better description of the workings of the EMS (see, for example, Fratianni and Von Hagen 1990). OLS rather than IV methods were used for Switzerland since expected inflation is not a significant regressor and does not enter the equation.

5 Factors in interest rate determination 469 Table 3. Restricted IV estimates of the partial adjustment interest rate equations with an MA(q) error specification Constant ! (0.0079) (0.002) (0.0046) (0.002) ( ) Domestic y (0.033) (0.022) (0.035) Domestic i (!1) (0.107) (0.045) (0.073) (0.062) (0.087) r (0.050) r (0.068) r (0.070) (0.032) (0.082) (0.078) Domestic z!0.291 (0.124) Domestic π (0.079) (0.024) (0.049) (0.123) R LM(4) p-value ARCH(4) p-value (i) Standard errors in parentheses. (ii) No MA(q) specification was required in the case of France and Switzerland. (iii) The equation for Switzerland was estimated by OLS since expected inflation is not a significant regressor and does not enter the equation. Table 4. Estimates of long-run interest rate elasticities based on the restricted partial adjustment equations Domestic y Domestic z!0.407 Domestic π r r r r Sample period 73:3 93:3 73:3 93:3 68:3 93:3 i/ y is a semi-elasticity. the structural parameters of the model (see Table 6) and also to conduct a forecasting exercise (see Table 7). It can be seen from Table 6 that, in the case of the US, the parameter δ in the policy reaction function is positive, which implies that, in the US, monetary authorities increase the money supply in the presence of higher expected inflation, whereas the opposite is true in the case of Germany, where the Bundesbank appears to pursue a counter-inflationary policy. As for the coefficient φ, it measures the speed at which uncovered interest rate differentials will be corrected. The estimated values indicate that US rates react rather quickly to changes in German rates. The semielasticity of the demand for money relative to the nominal interest rate, corresponding to the parameter β, is found to be higher in Germany that in the US. Finally, the parameter γ in the policy reaction function appears to be positive, which implies that an expansionary monetary policy is pursued by the Federal Reserve in the presence of a deteriorating budgetary balance. Concerning the forecasting performance of the empirical models, we compared its out-of-sample accuracy with that of a simple random walk model by using some standard measures of predictive ability (see Table 7). It should be noted that all estimated equations outperform the simple random walk model in terms of forecasting power, and that the model for German interest rates has the highest degree

6 470 G. M. Caporale and N. Pittis Table 5. Cointegration tests and long-run elasticities based on the estimates of the cointegrating relationships y Domestic z!0.42 Domestic π r r r R CRDW * 0.65* 0.56* ADF!5.52*!3.86*!5.08*!4.08*!4.94* J * 29.33* 26.24* 31.73* 34.40* J * 43.05* 50.11* 49.47* 83.09* (i) J1 and J2 are the maximum eigenvalue and trace statistic tests for the number of cointegrating vectors developed in Johansen (1988) and Johansen and Juselius (1990). (ii) An asterisk indicates rejection of the null of no cointegration at the 5% level. Table 6. Retrieving the structural parameters from the estimated reduced form interest rate equations US GE λ β γ 1.44 δ 0.358!0.511 φ φ In the case of Japan, France and Switzerland, the number of structural parameters exceeds that of the reduced form parameters, and hence they cannot be retrieved. of accuracy, whilst the equation for Japanese rates yields the least accurate forecasts. IV. CONCLUSIONS In this paper we have argued that a better understanding of the behaviour of interest rates can be gained by considering the role of both domestic and external factors, and can therefore avoid the unrealistic assumptions that the economy is either completely closed or fully open. Even in relatively open economies there is a role for domestic factors if there is some degree of control on across-the-border capital movements, or if the risk premium is a function of disequilibria in the money market. Hence our theoretical model encompasses the two extreme cases normally analysed in the literature. We estimated it for the US, Japan, Germany, France and Switzerland because, in spite of the relatively high degree of openness of these economies (only Japan and France imposed capital controls in some periods), in all of them domestic monetary conditions could have affected interest rates. The empirical results suggest that the behaviour of interest rates does reflect both internal and external influences in all five countries. In particular, monetary policy in Germany has important effects on the US and Japanese economies, with higher German interest rates being quickly translated into higher US and Japanese rates. Germany, on the other hand, appears to be responsive to policy developments in continental Europe, in spite of its dominance of the European Monetary System. German monetary policy appears to be the main external factor determining interest rates not only in France, but also in a non-ems country such as Switzerland. Table 7. Ex-post prediction analysis ( forecasts are based on the restricted reduced form interest rate equations). Forecasting period: 89:493:3 US JA GE FR SZ Model RW Model RW Model RW Model RW Model RW RMSE MAE Theil s U (i) Future values for the exogenous variables have been replaced by their actual values. (ii) RW stands for random walk, RMSE for Root Mean Square Error, and MAE for Mean Absolute Error.

7 Factors in interest rate determination 471 Perhaps surprisingly, Japan seems to be more closely integrated with Germany than the US; also, inflationary developments are closely monitored there. Another interesting finding is that the evolution of the budget deficit affects interest rates in the US, but not in Germany. In the US, the monetary authorities fail to adopt a tighter policy stance in the presence of a looser budgetary stance. Finally, it appears that the relative weights of output and price stabilization in the policy makers objective function are different in these two countries: the preferences of monetary authorities are such that the fight against inflation is the foremost concern in Germany, but not in the US. ACKNOWLEDGEMENT We are grateful to Stephen Hall for useful comments and suggestions. The usual disclaimer applies. REFERENCES Artis, M. J and Nachane, D. (1990) The European Monetary System: an evaluation, Journal of Policy Modelling, 9, Boughton, J. M. (1988) Exchange rates and the term structure of interest rates, IMF Staff Papers, 35(1), Caporale, G. M. and Pittis, N. (1993) Common stochastic trends and inflation convergence in the EMS, ¼eltwirtschaftliches Archiv, 129(2), Edwards, S. (1985) Money, the rate of devaluation, and interest rates in a semiopen economy, Journal of Money, Credit and Banking, 17(1), Fratianni, M. and Von Hagen, J. (1990) German dominance in the EMS: the empirical evidence, Journal of International Money and Finance, 9, Giavazzi, F. and Giovannini, A. (1988) Models of the EMS: is Europe a greater Deutschemark area? in Global Macroeconomics, eds R. C. Bryant and R. Portes. St Martin s Press, New York. Henderson, J. M. and Quandt, R. E. (1971) Microeconomic ¹heory, 2nd edition, McGraw-Hill, New York. Johansen, S. (1988) Statistical analysis of cointegrating vectors, Journal of Economic Dynamics and Control, 12, Johansen, S. and Juselius, K. (1990) Maximum likelihood estimation and inference on cointegration with applications to the demand for money, Oxford Bulletin of Economics and Statistics, 52, Mundell, R. A. (1968) International Economics, MacMillan, New York. Peston, M. H. and Quandt, R. E. (1986) Prices, Competition and Equilibrium, Philip Allen. Quandt, R. E. (1988) ¹he Econometrics of Disequilibrium, Basil Blackwell, Oxford. Stock, J. H. (1987) Asymptotic properties of least squares estimators of cointegrating vectors, Econometrica, 55, Tobin, J. (1969) A general equilibrium approach to monetary theory, Journal of Money, Credit and Banking, 1,