Asset Pricing Tests Using Realized Returns

Results from two-stage (cross-sectional) asset pricing tests vary with first-stage design choices and the test assets employed. First, I argue that time-varying loadings have higher construct validity as explanatory variables for the current expected return in that they do not contain future information and more accurately describe the timevarying risk of the firm. Second, I introduce randomly generated portfolios as test assets which, in sharp contrast to the systematically formed portfolios used in the literature, yield implied factor estimates that are consistent with a reduction in measurement error and robust to the arbitrary level of aggregation. Third, firm-level results on the market factor and HML factor are particularly dependent on the distinction between explaining contemporaneous returns and predicting future returns. This Draft: July 2012