Improving stock trend prediction through financial time series classification and temporal correlation analysis based on aligning change point

In order to improve the accuracy of stock prediction, people major in computer science and technology begin to apply their techniques to the financial market. In the financial market, there are many similar but not simultaneous fluctuations caused by different reaction efficiencies to the same event. Therefore, quickly reflected stocks' trends could improve trend predictions of similar slowly reflected stocks. To find the temporal correlation between stocks in the same securities exchange, a financial time series classification approach based on aligning change points is proposed to help investors discover hidden temporal correlations, which could improve stock trend prediction, to adjust portfolios. Firstly, the securities index of the securities exchange is chosen to be the benchmark, and the important change points are screened out to mark the essential fluctuations. Secondly, the points of all the constituent stocks of the same securities index which could be aligned to the important change points of the index are screened out and aligned through the aligning algorithm. Then the number of aligned stocks' points in different types helps to divide stocks into lead class and lag class. Temporal correlation and time difference are obtained through the temporal correlation analysis algorithm. Finally, four different prediction models are used to verify whether the classification information and time difference obtained from temporal correlation analysis could improve the stock trend prediction. The results show that our work could reveal potential connections among stocks as a bridge to introduce valid exogenous information, which is promising for stock trend prediction studies.