A cursory examination of the betas provided by different investment services reveals that these betas vary greatly, even when they are calculated on the basis of historical returns. How can one determine which beta is best? Using a mean square error test, the author evaluates betas from various commercial sources in terms of their accuracy in predicting subsequent betas and in predicting subsequent returns.

In general, the longer the horizon and the larger the portfolio, the better the forecast accuracy of any beta. On the other hand, all the methods tested left the possibility for over- and underestimations of troubling magnitudes.

More specifically, in terms of predicting ensuing betas, the Value Line forecasts exhibited the lowest mean square errors for a sample of utility stocks, but a naive forecast (assuming betas would equal one) performed almost as well. The Rosenberg long-term fundamental beta proved superior in forecasting the betas of a sample of industrial stocks. Performance in predicting return varied greatly, depending on the model parameters for market return and risk-free rate. A naive model performed best for the utility sample, but no one beta proved best in the case of the industrial sample.