A New Boosting Algorithm for Online Portfolio Selection Based on dynamic Time Warping and Anti-correlation

Online portfolio selection focuses on maximizing cumulative wealth and outputs a portfolio in each period. Anticor is a state-of-the-art algorithm in this area, but the similarity calculation method between long-period time series of different assets in the Anticor algorithm cannot effectively reflect the correlation of long-period time series with different stocks. The new algorithm(Anticor-DTW) proposed in this paper improves the original similarity distance calculation method of the algorithm by introducing dynamic time warping(DTW), which can identify similar shapes and spatial differences between different series by aligning the shortest paths. We not only simulated Anticor-DTW and Anticor on four classic stock datasets, NYSE(N), NYSE(O), TSE, and MSCI but also conducted experiments on new untested stock datasets HuShen300 and NASDAQ. All experiments indicated that Anticor-DTW outperforms Anticor. Moreover, we conducted a transaction costs experiment with exponential Ornstein-Uhlenbeck process, and the result also proved the great practicability of the Anticor-DTW algorithm in the real asset market.