A General Stationary Stochastic Regression Model for Estimating and Predicting Beta

A beta regression model is proposed where the coefficients follow a general class of stationary stochastic processes. The procedure identifies the process and estimates the parameters of the model simultaneously from the information contained in the return series. The returns of each of the Dow Jones 30 securities are examined. Betas of 5 of the securities are nonstationary and do not appear to follow a particular form of nonstationarity. Conclusions of many earlier studies may be suspect since they are based on procedures tailored to adoption of a specific form of beta nonstationarity and, thereby, based on an erroneous a priori assumption regarding such form. The ordinary least squares model is also found to be quite robust, providing reliable beta and intercept estimates not materially different from the more complex procedure with 25 of the return series.