Detecting Structural Shifts of the Risk-Return Tradeoff

Abstract

This paper suggests a detection method of the structural shift points in time of the risk-return relation or market betas. Thismethod sequentially estimates unknown multiple change-pointsof market betas under the circumstances that there are multipleunknown change-points and the number of regimes is unknown.sential not only to obtain more precisely estimated betas in thetwo-pass asset pricing tests, but also to avoid some arbitrarinessin the errors-in-variable correction for the cross-sectional regres-results show that the overall average length of the beta stationaryregimes of all NYSE and AMEX stocks is 55.0 months, which issomewhat close to the arbitrarily chosen 5-year period in estimat-ing betas for the asset pricing tests. Although the average stockreturn on January is known as much greater than on the othercalendar months, there is no signicant di??erence in the frequen-cies of the detected beta change-points across calendar months. Ihave also found that the changes in betas have the same directionas the changes in average returns, and, moreover, the greater thebeta change, the greater the change of average returns, which areas predicted by the beta pricing model.