Tests of Asset Pricing Models with Changing Expectations

This paper studies time-variation in the expected returns of common stocks, using linear models with constant “beta” coefficients and the methodology of Gibbons and Ferson. We examine individual stocks and portfolios formed on the basis of firm size and industry. Regression models of the expected returns incorporate information in past returns and other patterns of time-variation documented in previous studies. Our results indicate that the inability of Gibbons and Ferson to reject a single-factor model of the expected returns of the Dow Jones 30 is a sample-specific result. There is strong evidence that expected risk premiums are nonzero in months other than January. The tests indicate that unconditional mean returns and seasonality in “beta” coefficients can be important in latent variable models of expected returns. 21.26.13