Asset Pricing Tests Using Random Portfolios

Results from two-stage asset pricing tests vary with first-stage design choices and test assets. First, I argue that time-varying betas have higher construct validity than constant loadings as they are not conditional on future (returns) information and more accurately capture time-varying expected returns. Second, randomly generated portfolios as test assets yield, in contrast to the systematically formed portfolios in the literature, implied factor premia that are consistent with a reduction in measurement error and robust to the arbitrary level of aggregation. Third, the market and HML factor premia are strongly dependent on the distinction between explaining and predicting returns.