Selection of Market Window Size in Portfolio Strategies

We propose a feasible selection method of market window size for four state-of-the-art portfolio strategies. Market window size is a common parameter in the machine learning strategy for portfolio selection. However, in previous researches, the selection of market window size often lacks the guidance of scientific theories. In this paper, we analyze the sensitivity of market window size for four strategies on six benchmark data sets respectively. We study the distribution rule of the best market window sizes, which bring the peak total wealth, and then present the market window size selection method that is effective whether there is ample history data or not. What's more, to appraise the result of our method, we divide the benchmark data sets into two parts. We select the appropriate window size in the first part by our method, while the second part is used to test. By comparing with the wealth achieved in the second part using original method, we find that our selection method can effectively optimize the final results.