Measuring State-Dependent Risk Aversion Using Data Augmentation

We propose a continuous-time consumption-based capital asset pricing model in which the representative agent's preferences display state-dependent risk aversion. Since fluctuations in marginal utility can be ascribed to variations in levels of risk aversion as well as in levels of consumption, we obtain a valuation equation in which the vector of excess returns on equity reflects both consumption risk as well as the risk associated with variations in the state variables. We develop a simple model that can be estimated without specifying the functional form linking risk aversion with wealth and state variables. Our estimates are based on a Bayesian analysis of exact discrete-time parametrisations for continuous-time processes. Since consumption risk is not forced to account for the entire risk premium, our results contrast sharply with estimates from models in which risk aversion is state- independent. We find that relaxing fixed risk preferences yields estimates for relative risk aversion that are centered at 0.9 and are highly counter-cyclical, i.e. decreasing in unanticipated shocks to consumption.