Robust Inference for Consumption-Based Asset Pricing

Abstract

The reliability of traditional asset pricing tests depends on: (i) the correlations between asset returns and factors; (ii) the time series sample sizeTcompared to the number of assetsN. For macro-risk factors, like consumption growth, (i) and (ii) are often such that traditional tests cannot be trusted. We extend the Gibbons-Ross-Shanken statistic to test identification of risk premia and construct their 95% confidence sets. These sets are wide or unbounded whenTandNare close, but show that average returns are not fully spanned by betas whenTexceedsNconsiderably. Our findings indicate when meaningful empirical inference is feasible.