The Research And Application Of Learning To Trade Algorithm Based On The Universal Portfolio Selection

This paper focuses on the similarity-driven learning-to-trade algorithm of universal portfolio selection problem. As to the defects of the original algorithm Bk and CORN, our paper introduces a new strategy Bv based on the new definition of vector similarity. Further, we make the numerical simulation of Chinese stock market base on all strategies of Bk, CORN and Bv. The numerical results show that:1) New strategy Bv is able to earn a stable and outstanding return of long term in Chinese stock market.2) The growth rate of new strategy Bv is higher than that of the original strategy Bk and CORN.3) The new strategy Bv is less sensitive to the parameters than the original strategy Bk.4) All strategies are sensitive to the transaction costs and fat-tail.The numerical results indicate that the new strategy Bv improve the defects of the original strategy Bk and CORN, and show better effectiveness and robustness. The results also indicate that like the developed stock market, the return of Chinese stock market present the characteristic of fat-tail and the transaction cost restrict the performance of the strategies.