Evaluating the Accuracy of Beta Forecasts

Abstract

The authors present a methodology to compare the accuracy of beta forecasts. They show that residual volatility is minimized when true (unobservable) betas are used. The authors relate the level of error in the beta estimate to the realized residual volatilities and then use their approach to compare the accuracy of several beta estimates. In particular, they find that the predicted betas from a factor risk model are significantly more accurate than the historical betas derived from a time-series regression.