Identification and inference in two-pass asset pricing models

We introduce a framework that robustifies two-pass Fama–MacBeth regressions, in the sense that confidence regions for the ex post price of risk can be derived reliably even with weak identification. This region can be unbounded, if risk price is hard to identify, empty, if the model lacks fit, and bounded otherwise. Our framework thus provides automatic weak-identification and lack-of-fit warnings, and informative model rejections. Empirically relevant simulations document attractive size and power properties. Empirical applications with well known models and data sets illustrate practical usefulness and the potential value of additional cross-sectional information.