Decentralized Online Portfolio Selection with Transaction Costs

Online portfolio selection provides sequential decision-making on asset allocation without estimating the distribution function of asset return in advance, which is suitable for complex financial market. However, transaction costs generated during the portfolio update process have a significant impact on the return of investment strategy. In addition, some existing online portfolio models are prone to centralized investment. Thus, this paper firstly proposes a decision-making framework aiming to balance transaction costs and investment returns by using L1 norm (i.e., the Manhattan Distance) as the regularization term of the objective function. Second, a framework of decentralized online portfolio with transaction costs (DTC) is provided by taking the entropy of the portfolio as a constraint. Third, using two approaches of predicting asset return, two strategies of DTC1 and DTC2 are proposed correspondingly. Lastly, the performance of DTC1 and DTC2 is verified by using historical data from financial market. The results show that DTC1 and DTC2 can effectively deal with reasonable transaction costs and improve related strategies in the case of non-zero transaction costs.