We show that benchmark-linked convex incentives can lead risk-averse money managers aware of mispricing to overinvest in overpriced securities. In the model, the managers’ risk-seeking behavior varies in response to the interaction of mispricing with convexity and benchmarking concerns. Convexity effects can exacerbate the manager’s overinvestment in overvalued nonbenchmark securities. In contrast, they potentially offset the benchmarking effects studied in the literature, leading to underinvestment in overpriced benchmark securities. Under correlated mispricing across assets, our model rationalizes positive positions in nonbenchmark, negative risk premium (i.e., “bubble”) securities and “pairs trading” in two overvalued securities. Our findings help explain several empirical puzzles.

We conduct a controlled laboratory experiment in the spirit of Merton (1971), in which subjects dynamically choose their portfolio allocation between a risk-free and risky asset. Using the optimal allocation of an investor with hyperbolic absolute risk aversion (HARA) utility, we fit the experimental choices to characterize the risk profile of our participants. Despite substantial heterogeneity, decreasing absolute risk aversion and increasing relative risk aversion are the predominant types. We also find some evidence of increased risk taking after a gain. Finally, the session level risk attitudes show a different profile than the individual descriptions of risk attitudes.