Beta Regime-Switching Hedge Funds and their Clones

Abstract

In this article, the authors introduce clones of beta-switching funds that utilize the most-recent state to set their factor exposures. The optimal weight on this state depends on whether the goal is to track an index or produce a timing signal. In the latter case, full weight on the recent state is optimal, but timing gains disappear with even minimal data lags. When tracking funds, it is optimal to shrink recent beta toward its long-run value and adjust exposures for alpha differences in the two states. In most of the equity-sensitive hedge fund indexes tested, dynamic mixture clones have lower mean square error than rolling-regression clones of single-factor and Fama–French–Carhart models. The authors measure dynamic hedge fund exposures with a beta regime-switching model. Once estimated, this becomes a dynamic mixture model, where high-beta probabilities serve as mixing fractions. The covariance of estimated high-beta probabilities with market returns yields a market-timing contribution; however, few hedge fund indexes show significant equity timing ability, either positive or negative. In many cases, timing conclusions differ from those of the classic Merton–Henriksson [MH] model. The authors decompose the MH timing coefficient, assuming the fund follows a dynamic mixture model. In the decomposition, only one term corresponds to factor timing, and it rewards the timing of large returns while penalizing timing of moderate returns.