Simulation-based-Estimation in Portfolio Selection

This paper provides a simulation-based approach to optimal portfolio selection. We take a Bayesian approach as it naturally accounts for estimation risk, i.e., parameter uncertainty, learning of state variables and models, and can incorporate prior beliefs about future return distributions. We specifically highlight two implementations with great potential in portfolio selection. First, for competing models of predictable returns, we show how filtering techniques can be used to compute time varying model probabilities. Second, we show how simulation methods can be used to obtain maximize expected utility, bypassing computationally awkward gradient methods. We illustrate these methods in the classic risky stock allocation framework with the dividend-to-price regressor as an exogenous predictor.