Parameter-free robust optimization for the maximum-Sharpe portfolio problem

The list of datasets and competing methods are presented in Tables 1 and 2 respectively. Figure 1 shows simulation results on the value-weighted datasets. The results are qualitatively similar to those in Figure 3 in the manuscript. Figure 2 shows how the average one-year Sharpe ratio of AlphaRob varies with the training sample size n (results are on real-world monthly returns and not simulated returns). For comparison, we also show the results for CS, Min Var (NLS), and Min Var (L2). AlphaRob generally has the best Sharpe ratio or is close to the best. The sole exception is again the 10FFVW dataset. We note that for small n and large p, the robust optimization of CS becomes infeasible for the parameter values suggested by Ceria and Stubbs (2006).