🤖 AI Summary
This study addresses the estimation risk inherent in parametric portfolio strategies, which traditionally rely on return-generating models and struggle to effectively integrate prior knowledge with empirical data. The authors propose a generalized Bayesian framework that updates investor beliefs about feature tilts and out-of-sample returns solely through the utility function, bypassing the need to specify a return-generating process. A key innovation is the introduction of a unique belief-updating rule aligned with investor utility, coupled with the KNEEDLE algorithm that endogenously selects the optimal scaling parameter λ* without requiring out-of-sample validation. Theoretical analysis reveals a direct link between λ* and both risk aversion and higher-order moments of returns. Empirical results using U.S. equity data from 1955 to 2024 show that feature-based predictability is largely concentrated before 2000, and that adaptive selection of λ* substantially enhances portfolio robustness.
📝 Abstract
Parametric portfolio policies may experience estimation risk. I develop a generalized Bayesian framework that updates priors, delivering a posterior distribution over characteristic tilts and out-of-sample returns that is the unique belief-updating rule consistent with the investor's utility function, requiring no model for the return generating process. The Gibbs posterior is the closest distribution to the prior in Kullback-Leibler divergence subject to utility maximization. The posterior's scaling parameter $λ$ controls the weight placed on data relative to the prior. I develop a KNEEDLE algorithm to select optimal $λ^*$ in-sample by trading off posterior precision against numerical fragility, eliminating the need for out-of-sample validation. I apply this to U.S. equities (1955-2024), and confirm characteristic-based gains concentrate pre-2000. I find that $λ^*$ varies meaningfully with risk aversion and depends on higher-order moments.