RDP 9301: The Response of Australian Stock, Foreign Exchange and Bond Markets to Foreign Asset Returns and Volatilities 1. Introduction

This paper is a study of the relationships between changing volatilities in international asset prices and time-varying correlations of domestic returns in a small open economy with international asset returns. In particular, I examine whether returns in Australian stock prices, long-term bond yields and exchange rates are more highly correlated with corresponding foreign stock prices, bond yields and exchange rates when measures indicate higher than normal volatility in the domestic or international markets.

The paper is data-analytic rather than model-theoretic. I make use of recent time-series methods, involving Schwert measures of volatility, VAR estimation, Granger causality, impulse response functions, variance-decomposition analysis and Kalman filtering to identify and quantify the international factors affecting the volatility of Australian asset prices.

The analysis draws on recent work by Kupiec (1991) as well as that by King and Wadhwani (1988) and King, Sentana, and Wadhwani (1991). While Kupiec drew attention to increasing correlations when market volatility is high, his empirical analysis was based on simple correlation coefficients in different sub-samples, when volatilities were higher than normal. He did not provide a direct test of the influence of volatility measures on time-varying correlations.

The correlation of international asset prices has been explained by the contagion model. King and Wadhwani argue that, in a ‘non-fully revealing equilibrium’, price changes in one market depend on price changes in other markets through structural contagion coefficients. Such an equilibrium exists when the information structure is complex, and when domestic market prices do not reveal all relevant information to agents. Thus, valuable information is contained in prices that other traders in other markets are willing to pay. In this setting, of course, mistakes or idiosyncratic changes may be transmitted from one market to another^[1].

There is also the possibility that, in periods of high or increasing turbulence in domestic or foreign markets, asset prices may also be more strongly correlated with their own past values. The reason for the presence of serial correlation when volatility is high is that during such periods, ‘rational investors may be unwilling to absorb the risk resulting from the transaction that would be necessary to eliminate this arbitrage opportunity’ (King and Wadhwani (1988), p. 24).

I measure the volatility of the domestic and international asset prices by the Schwert (1988) index^[2]. The time-varying estimates of international stock price correlations are computed by a Kalman filtering program described in Hansen and Sargent (1991). With time series for both the volatilities and the return correlations, I first examine the contemporaneous correlations among the volatility indices, as well as temporal patterns of causal interactions through VAR estimation and impulse response functions. I then consider the contemporaneous correlations and causal patterns among the international asset-return correlations and the volatility measures^[3].

The next section is an analysis of the Australian and international stock prices. The two succeeding sections treat foreign exchange rates and bond yields. The final section concludes.

Footnotes

One justification for the non-fully revealing equilibrium offered by King and Wadhwani is the fact that markets are not open around the clock. Thus traders have an incentive to watch other markets for relevant information. [1]

The Schwert volatility index is computed as follows: (1) regress the stationary asset prices on seasonal dummies and lagged returns; (2) take absolute values of the residuals from the first stage, and regress these absolute values on their own lags and seasonal dummies. The predicted values from the second step are estimates of the standard deviation of the asset prices. This procedure is a two-step least-squares ‘short-cut’ for the Bollerslev (1986) GARCH (generalized autoregressive conditional heteroskedastic) maximum likelihood estimation of the volatilities or ‘conditional variances’ of the stock prices. [2]

All calculations for this paper were computed with MATLAB-386. Copies of the programs are available from the author or from the Economic Research Department of the Reserve Bank of Australia. [3]