Time variation in the correlation structure of exchange rates: high‐frequency analyses

Abstract

The correlation structure of asset returns is a crucial parameter in risk management as well as in theoretical finance. In practice, however, the true correlation structure between the returns of assets can easily become obscured by time variation in the observed correlation structure and in the liquidity of the assets. We employed a timestamped high-frequency data set of exchange rates, namely, the US$-deutsche mark and the US$-yen exchange rates, to calibrate the observed time variation in the correlation structure between theirThe authors would like to thank an anonymous referee and the editor, Robert Webb, for their suggestions and recommendations. returns. We also documented time variation in the liquidity structure of these rates. We then attempted to link the observed correlations with the liquidity via an application of an illiquid trading model first developed by Scholes and Williams (1976). We show that the observed correlation structure is strongly biased by the liquidity and that it is possible to effect at least a partial rectification of the otherwise downward-biased observed correlation. The rectified sample correlation is, therefore, more appropriate for input into models used for forecasting, option pricing, and other risk management applications.

Implied correlation index: A new measure of diversification

“…The significant time-variation in correlation has been reported in previous studies e.g. Muthuswamy, Sarkar, Low, and Terry (2001) and Bali, Genberg, and Neftci (2002). An interesting phenomenon shown in the figure is that the daily DJCIX level exhibited periodic spikes.…”

Cross‐market correlations and transmission of information

“…First, cross-market correlations are important because they affect the volatilities of portfolios and, therefore, have implications for asset allocation. Second, cross-market correlations are important for the pricing of derivative securities, especially exotic options, whose payoffs depend on more than one underlying asset price (see Muthuswamy et al, 2001, and additional references therein for details). Third, prices of specific futures contracts capture overall market dynamics very well, so results obtained in this study can be generalized to spot and other related markets.…”

“…For the empirical analysis we develop a new multivariate GARCH model that allows us to test hypotheses on the nature of time-varying correlations. A multivariate GARCH framework is desirable given the evidence of time variation in conditional volatility (see , for a survey), and time variation in the conditional covariances (Klaassen, 1999;Longin & Solnik, 1995;Muthuswamy et al, 2001). But existing parameterizations of multivariate GARCH models either directly parameterize the conditional covariance, which make testing hypotheses on the conditional correlation inconvenient, or specify constant correlations.…”

Statistical Causes for the Epps Effect in Microstructure Noise

“…Initially, we made the assumption of an underlying time series of prices, which is correlated and which exists on a smaller time scale. Equation (6) does no longer depend on the time scale of the hypothetical underlying time series. Neither does it depend on the actual prices on the underlying time series.…”