Change Point Detection in Financial Time Series

The current paper provides an analysis of Changepoint Detection (CPD) algorithms in financial time series. We explained that the use of existing approaches is impractical in the financial sphere for the task we set, therefore, a new approach for detecting critical points in the future was proposed, based on methods of time series forecasting. The advantage of the new approach is not only its ability to detect a changepoint, but also unequivocally decide on the purchase or sale of shares at a critical point in time. The hypothesis that a causal relationship between different financial time series and between a single time series and a macroeconomic indicator positively affects the forecasting of changepoints was also tested in the work. The proposed approach was tested in the format of a trading strategy on a portfolio of US stocks and demonstrated acceptable results.