Multivariate Time Series Study of Excess Returns on Equity and Foreign Exchange Markets

Abstract

This paper employs a multivariate time series approach to study long-horizon predictability of equity and currency market returns in the US, Great Britain, Japan and Germany. Following Bekaert and Hodrick (1992) we examine the predictability of returns by estimating VARs for pairs of countries. To highlight the dynamic structure in returns and to isolate important information variables that capture long-run predictability we use parsimonious versions of VARs and generalize the canonical analysis of Box & Tiao (1977). For forecast horizons between 1 month and 5 years we find that currency returns are more predictable than equity market returns and tend to reach peak forecastability at shorter horizons than country stock markets. US stock returns are primarily predicted by past US dividend yields and forward premia. Foreign stock markets are mostly related to past U.S. stock returns. The US dividend yield plays a central role in all the three systems. Thus the US stock market seems to influence other markets, but is not influenced by other stock markets.

JEL Classification: C22, F31