This paper compares the ability of four discount rate models to explain the cross-sectional and time-series variation of stock returns in the U.S., Japan, England, Germany, and Canada. The data consist of quarterly returns (in dollars) on Morgan Stanley's Capital International indices for the period 1972 through 1991. The following four models are considered: (1) A consumption CAPM model, linking the discount rate to the intertemporal marginal rate of substitution in consumption, (2) A production CAPM model, linking the discount rate to the intertemporal marginal rate of transformation in production, (3) A traditional CAPM model, linking the discount rate to the stock returns themselves, and (4) A constant (identical) discount rate model, which rules out risk premia, and therefore provides a useful benchmark. ; The main result is that a production-based CAPM (Cochrane(1991)) performs at least as well as the more standard consumption-based CAPM in explaining variation in national equity markets. However, as with the CCAPM, there is something of an `Equity Premium Puzzle' associated with a simple frictionless PCAPM. In particular, very high depreciation rates or low elasticities of substitution between capital and labor are required to fit the data. The intuition is as follows - Viewed from either the consumption or production side, the challenge is to explain why small changes in quantity variables are associated with large changes in relative prices. In the CCAPM this requires very curved indifference curves (i.e., a large coefficient of relative risk aversion), whereas in the PCAPM this requires very curved isoquants (i.e., a low elasticity of substitution between capital and labor).
