Detection of Multiple Beta Shifts in Monthly Returns Data

Abstract

This abstract was created post-production by the JFI Editorial Board.

This study has examined the power and type I error rate of four methods of testing for regression parameter changes when applied to detecting beta changes in monthly stock return series. The study used simulated stock return series with known betas, error variances, beta change dates, and error term distributions. In summary, it appears to be nearly impossible to detect or find the location of small or moderate beta changes in monthly stock return series. This suggests that the market model parameter changes reported by Hays and Upton (1986) are most likely not beta changes. However, they find of market model nonstationarity in almost all of the stocks in their sample-far more than this study finds. This suggests that if non-normality of stock returns accounts for the results obtained by Hays and Upton, the Stable Paretian 1.95 distribution does not adequately explain monthly stock returns.