Financial Time Series: Multi-Step Ahead Prediction

Forecasting financial time series is one of the most actual tasks of applied mathematics in the field of finance. Indeed, knowledge about changes in the value of a financial indicator in the future and taking timely actions can help to obtain high profits or prevent possible losses. Many of the existing forecasting approaches have good results predicting only one step ahead, and as the number of steps increases, the error increases exponentially. This trend can be explained from the point of view of dynamic chaos theory. This paper takes into account the consequences of the theory of dynamic chaos and improves the algorithm for predicting time series for many steps forward, which is based on the selection of behavior patterns and their subsequent clustering, using machine learning methods. Based on the results of the work, it is possible to improve the accuracy of predictions of the basic approach on the Lorenz series and make acceptable predictions for the real financial series.