Incorporating Risk Aversion into Dynamic Programming Models

Most previous stochastic dynamic programming (DP) applications have assumed that decision makers are risk neutral; however, risk permeates both intrayear and interyear relationships in most DP problems. Incorporating risk aversion to intrayear outcomes alone can violate the independence assumption of expected utility and can destabilize long‐run equilibrium returns. Aversion to riskiness of the long‐run returns suppresses the effect of sequential resolution of risk over time. Because tolerance of short‐run versus long‐run risk varies in dynamic situations, procedures for incorporating risk aversion should accommodate this variation. More research on risk averse DP formulations is needed.