Risk Aversion in a Dynamic Asset Allocation Experiment

We conduct a controlled laboratory experiment where subjects dynamically choose their portfolio allocation between a safe and a risky asset. We first derive analytically the optimal allocation of an expected utility maximizer with HARA utility function. We then fit the experimental choices to this model to assess the risk attitude of our subjects. Despite the substantial heterogeneity across subjects, decreasing absolute risk aversion and increasing relative risk aversion are the most prevalent risk types, and we can classify more than 50% of the subjects in this combined category. We also find evidence of increased risk taking after a gain but the effect is small in magnitude. Overall, our robustness tests show that the behavior of subjects is generally well accounted for by the HARA expected utility model. Finally, the analysis at the session level suggests that the behavior of the representative agent is less heterogeneous and closer to (though statistically different from) constant relative risk aversion.