Deep learning stacking for financial time series forecasting: an analysis with synthetic and real-world time series

Abstract

The forecasting problem is one of the main applications arising from the synergy between finance and artificial intelligence. With the advancement in the field of deep learning, some ANN achieved very satisfactory results and gained more attention. One approach to increase the time series forecasting model’s performance is ensemble models, combining each model’s prediction (stacking). However, there are some difficulties in combining and evaluating these models for a good performance in financial time series. We use synthetic and real-world time series to evaluate the model stacking, trying to understand the main financial time series components. Using this ensemble method, we reduced the prediction error for both scenarios.