Full-Scale Optimization (FSO) is a utility maximization approach to portfolio choice problems that has theoretical appeal but that suffers from computational burden in large scale problems. We apply the heuristic technique differential evolution to solve FSO-type asset selection problems of 97 assets under complex utility functions rendering rough utility search surfaces. We show that this problem is computationally feasible and that solutions retrieved with random starting values are converging to one optimum. Furthermore, the study constitutes the first FSO application to stock portfolio optimization. The results indicate that when investors are loss averse, FSO improves stock portfolio performance compared to Mean Variance (MV) portfolios. This finding widens the scope of applicability of FSO, but it is also stressed that out-of-sample success will always be dependent on the forecasting ability of the input return distributions.