Integrated strategy for online portfolio by considering the reversal effect and investor attention
提出一种结合反转效应和投资者关注的在线投资组合策略，通过在线梯度更新整合专家策略，理论证明遗憾值有次线性上界，实证中年化收益率达8.12%至206.35%，多数情况下优于现有策略。
The online portfolio selection (OPS) problem is different from the classical portfolio model problems, as it dynamically adjusts asset positions in response to historical price sequences. Existing OPS strategies based on the reversal effect achieve greater cumulative return than those based on momentum. However, they face theoretical challenges in achieving competitive performance and rely primarily on explicit data variables, such as historical relative prices and trading volumes. Besides, existing research indicates that investor behaviour, an implicit data variable, may influence stock prices. Therefore, we take into account the reversal effect and investor attention to formulate expert strategies, subsequently integrating these strategies through the online gradient update. Theoretically, the regret of our integrated strategy is proven to have a sub-linear upper bound. Empirical results demonstrate that our integrated strategy outperforms existing OPS strategies in most cases. Specifically, across all datasets, annualised returns range from 8.12% to 206.35%. Also, the average Sharpe Ratio, Calmar Ratio, and Information Ratio are 1.2580, 2.8975, and 0.0342, respectively. Moreover, our integrated strategy exhibits robustness under varying parameter settings and remains effective under reasonable transaction costs.