Conditional Factor Models and Return Predictability

Abstract

This paper develops a new approach to examining the time variation of risk premia within the framework of conditional asset pricing models. By combining conditional factor models with approximate present-value relationships we derive a linear relationship between the log stock price and investors' expectations of future factor loadings, risk premia, and cashlows. This framework allows us to estimate conditional risk premia from a cross-sectional regression of log prices on proxies for expected factor loadings and cashflows. We apply this technique to various factor specifications including the CAPM, the three factors advocated by Fama and French (1996), and a five-factor model with economically motivated factors similar to Chen et al. (1986). Consistent with rational pricing we find that, for the majority of the risk factors, the estimated risk premia contain significant information about the future expected returns of the factor portfolios over the sample 1938-2004. Our framework abstracts from the use of ad-hoc conditioning variables, and offers a theoretically appealing approach to modelling the predictable components of stock returns. In recent samples (1978-2004) our estimates of the market risk premium prove to be better forecasters of market returns than the dividend-price ratio and other commonly used forecasting variables. Results from the economic factor model provide evidence that current levels of treasury and corporate bond yields are embedded in the cross section of equity market prices.