Defensive Online Portfolio Selection

Quaderni del Dipartimento di Informatica, Sistemistica e Comunicazione dell'Università degli Studi di Milano Bicocca. Research Report n. 5, September 2007. The unbiasedness in the small property, belonging to the Game-Theoretic Framework, is exploited to de?ne the class of defensive online portfolio selection algorithms. This new class of online algorithms is oriented to wards ?nite investment horizon and is capable to exploit side-information, whenever available. Successive Constant Rebalanced Portfolios are shown to share the universality property with respect to Sequence Best Constant Rebalanced Portfolios. Thre einstances, be-longing to the class of defensive online portfolio selection algorithms, are presented and theoretically analyzed. Their empirical performance is investigated through a rich set of numerical experiments, concerning four major stock market datasets. The obtained results emphasize the relevance of the new class of investment strategies and underline the central role of their parameters to out perform a new benchmark named Sequence Best Constant Rebalanced Portfolio. This new benchmark is shown to be theoretically non inferior and empirically superior than the Best Constant Rebalanced Portfolio.