Data-Driven Robust Optimization with Application to Portfolio Management

Portfolio optimization results are strongly dependent on the model parameters. To circumvent this shortcoming, this paper proposes a new modeling approach to address data uncertainty. The model offers full control over the degree of conservatism and underlines its interaction with robustness for a range of extreme market situations. Without prior assumptions on data generating process, we develop a probabilistic guarantee on the optimality of the solution. Unlike previous measures that depend solely on the uncertainty model, our measure is also sensitive to the solution and the investment horizon. We provide an example on international stock indexes investment. Computational experiments and ex-post analysis provide evidence for the potential and the effectiveness of our model.