Online Portfolio Selection Strategy with Side Information Based on Learning with Expert Advice

Online portfolio selection is a decision-making problem of how to dynamically adjust asset positions based on historical price sequences. The existing works on online portfolio strategy are based on integrating static experts whose advice does not change with market states, which is inconsistent with the advice from dynamic experts in reality. In this paper, we propose a new online portfolio strategy by using side information states to characterize market states, getting advice from dynamic experts who dynamically adjust the strategy based on side information states and integrating their advice. To do so, we first regard all the state constant rebalanced portfolio strategies, which make portfolios dynamically according to side information states, as expert advice. Second, we apply the exponentially weighted average (EWA) algorithm to integrate the advice from the experts and propose a new strategy. Then, we prove that the strategy has the same asymptotic average logarithmic return rate as the best state constant rebalanced portfolio (BSCRP) strategy, i.e., it is universal. Finally, we conduct numerical experiments by using actual financial data from Chinese and American markets. The results indicate that our proposed strategy has good competitive performance.