Time-varying correlations between asset returns and implications to risk management.

Abstract

This paper studies the quarterly and overall performances of equal-weighted portfolios created through the removal of an asset with high correlation in each quarter based on different criteria, and that of minimum-variance portfolios with and without short-sale constraints as an alternative strategy for investors, using out-of-sample tests. Daily returns of ten prominent market indices across equities, bonds, commodities and real estate spanning different geographical regions are used to compute correlation matrices over 158 quarters starting from the year 1972. Results indicate that both 1/M portfolios which have one asset removed each underperform the benchmark 1/N portfolio as well as the minimum-variance portfolios according to the corresponding Sharpe ratios and Jensen’s alphas. The 1/M portfolios display similarly high fluctuations in performances as the benchmark, while the minimum-variance portfolios exhibit stable performance and lower draw-downs especially during financial crises but at the expense of lower returns. An analysis on the dispersion and composition of time-varying correlations show that rising correlations in recent years lead to larger market fluctuations and the lack of appropriate assets leads to suboptimal diversification.