Learning or Robust Control

Model mis-specification can cause substantial utility losses in portfolio planning.In this paper, we compare two approaches to cope with this problem,robust control and learning. We derive the optimal portfolio strategies and theutility losses due to model mis-specification. Surprisingly, neither learning norrobust control is uniformly superior to the naive approach where the investorsimply ignores model risk. Furthermore, a comparison of the two approachesshows that learning takes some time to have an impact, so that short-terminvestors are in some (but not all) cases better off with robust control thanwith learning.