Simple Better Market Betas

Abstract

The random-effects estimator in Vasicek (1973) and its variants provide the best predictors of market-beta known to date. However, they require multi-pass procedures (time-series and cross-section) and spookily entangle each stock’s beta estimate with unrelated stocks. Yet, the reason for their good performances has been misunderstood. They work so well because they pull in outliers. Thus, a much simpler robust estimator can predict future market-betas at least as well. It only requires first winsorizing stock returns at -1 and +3 times the market rate of return. A quick WLS decay of historical returns improves performance further. Moreover, my paper shows that the Dimson (1979) and Frazzini and Pedersen (2014) estimators are inferior even to plain OLS estimators. They should be considered de-facto inadmissible estimators of market-beta.

Keywords: market-beta, robust estimation, winsorization

JEL Classification: G1,G12