Evaluating the Specification Errors of Asset Pricing Models

This paper examines the specification errors of several asset pricing models using the methodology of Hansen and Jagannathan (1997) and a common data set. The models are the CAPM, the Consumption CAPM, the Jagannathan and Wang (1996) conditional CAPM, the Campbell (1996) dynamic asset pricing model, the Cochrane (1996) production-based model, and the Fama-French (1993) three-factor and five-factor models. We use returns on the Fama-French twenty-five portfolios sorted by size and book-to-market ratio and the risk-free rate as our test assets. The sample is 1952 to 1997. We allow the parameters of the models' pricing kernels to fluctuate with the business cycle which we measure in two ways. One uses the Hodrick-Prescott (1997) filter applied to either industrial production for monthly models or real GNP for quarterly models. The second approach for quarterly models uses the consumption-wealth measure developed by Lettau and Ludvigson (1999). While we cannot reject correct pricing for Campbell's model, a stability test indicates that the parameters may not be stable. None of the models correctly prices returns that are scaled by the term premium.