We introduce a flexible utility-based empirical approach to directly determine asset allocation decisions between risky and risk-free assets. This is in contrast to the commonly used two-step approach where least squares optimal statistical equity premium predictions are first constructed to form portfolio weights before economic criteria are used to evaluate resulting portfolio performance. Our single-step customized gradient boosting method is specifically designed to find optimal portfolio weights in a direct utility maximization. Empirical results of the monthly U.S. data show the superiority of boosted portfolio weights over several benchmarks, generating interpretable results and profitable asset allocation decisions.

Keywords: utility maximization, return predictability, machine learning, gradient boosting

JEL Classification: C22, C53, C58, G11, G17