Abstract

The existing empirical literature fails to agree on the nature of the intertemporal relation between expected return and volatility. The contrary results of either a positive or a negative risk-return relation mainly arise from different ways of empirically modeling the return dynamics in the absence of any theoretical guidance. This paper contributes to the literature by proposing a rigorous statistical framework – Bayesian model comparison approach – to empirically evaluate the models that typically indicate conflicting signs for the risk-return relation. More important, we explicitly assess the scale of the data-based evidence in favor of each model. Model uncertainty is taken into account when necessary. Overall, the evidence in the data, without conditioning on any single predetermined model, points to a negative risk-return relation. The most satisfactory model is identified as one that forecasts volatility using information conveyed in both past returns and exogenous instruments such as one-month interest rates.

Citations