A Decomposition of Conditional Correlations in Us Equities

Abstract

This paper extends the Forbes and Rigobon (2002) adjustment to account for heteroscedasticity and time-varying market volatility biases in conditional correlations for U.S. size and industry portfolios. After the adjustments, it is possible to compare correlations across conditioning sets. While some of the portfolios exhibit downside asymmetric conditional correlations, others show no evidence of correlation asymmetry and some exhibit upside correlation asymmetry. In addition and more importantly, the adjusted conditional correlations during extreme market moves are never statistically above their corresponding unconditional correlation. Thus, any asymmetric conditional correlation pattern is the result of correlations on one side of the distribution falling by more than the correlations on the other. The largest conditional correlations tend to occur when the market experiences average returns. These findings have important implications for asset pricing, market microstructure, and risk management and are generally robust to longer return horizons and conditioning upon prior market performance.

Keywords: Large returns, conditional correlation, equity portfolios, diversification, portfolio performance, heteroscedasticity

JEL Classification: G11, G12