We develop a framework for decision-making under model uncertainty, grounded in Wasserstein distributionally robust optimization. In contrast to standard Kullback-Leibler approaches constrained by absolute continuity, Wasserstein ambiguity sets constructed via optimal transport permit the adversary to physically relocate probability mass to new states, capturing economically relevant support-shifting distortions-structural breaks, regime shifts, and catastrophic tail events-that the reference model excludes entirely. We integrate these state-shift distortions into both static and recursive dynamic decision problems, where agents optimize against the worst-case distribution within a transport budget. A data-driven bootstrap calibration procedure disciplines the degree of robustness by linking the ambiguity radius to statistical confidence sets. Applying the framework to asset pricing, we show that endogenous belief distortions amplify perceived variance, inducing precautionary savings that lower the risk-free rate and widen the equity premium. The geometric uncertainty propagates through persistent state dynamics, generating endogenous stochastic volatility akin to long-run risk. The resulting pricing kernel aligns with empirical Hansen-Jagannathan bounds without requiring implausibly high risk aversion or non-standard preferences.
原文链接
分享
引用
