Asset Pricing with Dynamic CAPM: An Application to 49 US Industry Portfolios

Abstract

The Sharpe-Lintner Capital Asset Pricing Model (CAPM) has been widely criticised from Finance researchers on the basis that the expected returns of assets cannot be explained by the single market factor. This is due to the fact that market beta as predicted by the CAPM is stochastic and not constant which can result in deficiencies in prediction or inability to explain the cross-section of asset returns. In this paper we propose a CAPM assuming that market beta follows either an autoregressive stochastic process with one lag or a pure random walk process. In addition, we extend the latter model to incorporate heteroscedastic errors. We use two Markov Chain Monte Carlo algorithms and we find that stochastic beta models generate highly precise estimates of parameters in a simulation study. More specifically, our models' estimates are significantly more precise when compared with rolling betas regressions. We also test the stochastic beta models on real data using a 20-year sample of 49 value-weighted industry portfolio returns from the US market. Our results confirm previous studies that industry betas are not stable over time since the underlying beta process of most portfolios shows evidence of high persistence. Furthermore, models that account for time variation of market beta outperform both simple regressions and rolling beta regressions in in-sample and out-of-sample tests. Finally, our findings indicate that stochastic beta models can adequately explain the cross section of sector returns. This is especially essential because it implies that the stochastic beta models can be used as effective asset pricing tools without the adoption of other factors which are difficult to identify and construct.