Abstract
Hansen and Jagannathan (1997) introduced a measure of model misspecification based on the L2-norm. It is however well-known that L1-norm methods may show good properties in the presence of non-normal distributions. In this paper we therefore introduce some L1-norm based measures of misspecification. We also provide an easy algorithm which simplifies the computation of the gain-loss ratio presented by Bernardo and Ledoit (2000). Two Monte Carlo simulations are undertaken to assess the performance of the measures under varying distributional assumptions. We provide evidence that L1-norm based measures tend to perform better in small and non-normally distributed samples.