Empirical evaluation of asset pricing models: arbitrage and pricing errors over contingent claims

In a 1997 paper, Hansen and Jagannathan develop two pricing error measures for asset pricing models. The first measure is the maximum pricing error on given test assets, and the second measure is the maximum pricing error over all possible contingent claims. We develop a simulation-based Bayesian inference of the pricing error measures. Although linear time-varying and multifactor models are widely reported to have small pricing errors on standard test assets, we demonstrate that these models can have large pricing errors over contingent claims because their stochastic discount factors are allowed to be negative and thus offer arbitrage opportunities.