AB-LSTM-GRU: A Novel Ensemble Composite Deep Neural Network Model for Exchange Rate Forecasting

Exchange rate forecasting is always complex and challenging, with great exploration significance and economic value. This study proposes a novel ensemble composite exchange rate forecasting model called AB-LSTM-GRU to improve the exactness and reliability of exchange rate forecasting. This model combines long short-term memory (LSTM) and the gate recurrent unit (GRU) to form a hybrid deep neural network as a weak learner and uses the adaptive boosting (AdaBoost) framework to integrate each weak learner to construct the final strong learner. In various experiments of this study, we conducted a comparative analysis based on the United States dollar/Chinese yuan renminbi (i.e., USD/CNY) historical data from January 1, 2010, to December 31, 2022. We selected the British pound sterling (GBP)/USD and GBP/CNY data within the same time range to demonstrate the robustness of the proposed model. Experimental results manifest that the predicted results of AB-LSTM-GRU are more accurate, and the fluctuation of its accuracy is less than that of other benchmark models. In verifying GBP/USD and GBP/CNY, we found the model has good robustness, making it applicable to different exchange rate data predictions. Overall, AB-LSTM-GRU combines the advantages of LSTM and GRU, which are good at analyzing long short-term and large-span features and the characteristics of low volatility of AdaBoost’s prediction results, and can provide valuable information and reference for decision-making related to investors, enterprises, and countries.