Given a set of asset returns, an information-theoretic approach is used to estimate non-parametrically the pricing kernel to price the given cross-section out-of-sample. Compared to leading factor models, this information SDF delivers smaller pricing errors and better cross-sectional fit, and identifies the maximum Sharpe ratio portfolio out-of-sample. Moreover, it extracts novel pricing information not captured by Fama–French and momentum factors, leading to an ‘information anomaly.’ A tradable information portfolio that mimics this kernel has a very high out-of-sample Sharpe ratio, outperforming both the 1/N benchmark and the Value and Momentum strategies combined. These results hold for a wide cross-section of assets.
Financial Markets Group Discussion Papers DP 749