Integrating Prediction in Mean-Variance Portfolio Optimization

Abstract

Many problems in quantitative finance involve both predictive forecasting and decision-based optimization. Traditionally, predictive models are optimized with unique prediction-based objectives and constraints, and are therefore unaware of how those predictions will ultimately be used in the context of their final decision-based optimization. We present a stochastic optimization framework for integrating regression based predictive models in a mean-variance portfolio optimization setting. Closed-form analytical solutions are provided for the unconstrained and equality constrained case. For the general inequality constrained case, we make use of recent advances in neural-network architecture for efficient optimization of batch quadratic-programs. To our knowledge, this is the first rigorous study of integrating prediction in a mean-variance portfolio optimization setting. We present several historical simulations using global futures data and demonstrate the benefits of the integrated approach in comparison to the decoupled alternative.