Empirical analysis of the forecast error impact of classical and bayesian beta adjustment techniques

The paper presents a comparative study of conventional beta adjustment techniques and suggests an improved Bayesian model for beta forecasting. The seminal papers of Blume (1971) and Levy (1971) suggested that for both single security and portfolio there was a tendency for relatively high and low beta coefficients to over predict and under predict, respectively, the corresponding betas for the subsequent time period. We utilize this proven fact to give a Bayesian adjustment technique under a bilinear loss function where the problem of overestimation and underestimation of future betas is rectified to an extent so as to give us improved beta forecasts. The accuracy and efficiency of our methodology with respect to existing procedures is shown by computing the mean square forecast error.