Change Point Detection and Trend Analysis for Financial Time Series

Change point detection of time series plays a vital role in financial data analysis for trading decision support. Traditional methods generally assume that time series follows a normal distribution, while the distribution of financial data usually has specific statistical characteristics with a higher peak and fatter tails. In this paper, we propose an online change point detection and trend analysis method, FinCPD, for financial time series. We apply a mixture normal distribution to specify the leptokurtic characteristics of the financial time series. A novel measurement, Accumulative Advantage, is presented to evaluate the stability of a financial time series based on the likelihood-ratio, which is further used for online change point detection and trend analysis. Experiments of simulated trading based on real-world financial data show that FinCPD outperforms the traditional methods in terms of a series of metrics.