International asset allocation with time varying risk: an analysis and implementation

Abstract

This paper investigates the problem of optimally allocating funds in an investment portfolio among major classes of U.S. and Japanese assets, given that the asset risk parameters vary over time. We devise an econometric specification in the ARCH/GARCH family to model the evolution of the returns covariance matrix and find substantial time variation in both returns variances and correlations. The fitted covariance matrices are then used to analyze optimal asset allocation portfolios. The time patterns in the portfolio weights and risk assessment make sense: They show the change in character of the U.S. Treasury bond market in 1979 and the increase in stock market risk in October 1987, for example. However, while the models capture some of the time variation in asset risk parameters, the improvement in portpolio performance is limited. We also examine investment performance under different portfolio constraints, and find that allowing international diversification makes the biggest impact, especially for a U.S. investor. Prohibiting short sales and the hedging of currency risk seem to have little impact. We find that post-sample performance of the model deteriorates relative to the fixed variance model.