This study proposes a new ensemble deep learning approach called LSTM-B by integrating long-short term memory (LSTM) neural network and bagging ensemble learning strategy in order to obtain accurate results of exchange rates forecasting and to improve profitability of exchange rates trading. Previous research literatures have explored exchange rate forecasts, mainly focusing on the validity of forecasts, nevertheless; the precision is only one aspect of exchange rates forecasts. More important than the forecasting performance is how these ensemble learning approaches such as our proposed LSTM-B ensemble deep learning approach can advise professional trading. We extend our forecasts results to examine potential financial profitability of exchange rates between the US dollars (USD) against other four major currencies, such as GBP, JPY, EUR and CNY. The empirical study indicates the effectiveness of our proposed LSTM-B ensemble deep learning approach, which significantly improved forecasting accuracy and potential trading profitability. The proposed LSTM-B ensemble deep learning approach significantly outperforms some other benchmarks with/without bagging ensemble learning strategy under study by means of the forecast performance and the potential trading profitability.

这项研究通过集成长期短期记忆（LSTM）神经网络和袋装集成学习策略，提出了一种称为LSTM-B的集成深度学习新方法，以获取准确的汇率预测结果并提高汇率交易的获利能力。然而，先前的研究文献已经探讨了汇率预测，但主要侧重于预测的有效性。精确度只是汇率预测的一个方面。比预测性能更重要的是，这些集成学习方法（例如我们提出的LSTM-B集成深度学习方法）如何为专业交易提供建议。我们扩展了预测结果，以检查美元对其他四种主要货币（例如英镑，日元，欧元和人民币）的汇率的潜在财务盈利能力。实证研究表明，我们提出的LSTM-B集成深度学习方法是有效的，该方法大大提高了预测准确性和潜在的交易获利能力。拟议的LSTM-B集成深度学习方法通​​过预测的性能和潜在的交易获利能力，显着优于正在研究或不研究袋装集成学习策略的其他一些基准。