Scholarlink

<p>Exchange rate prediction has consistently been a popular research topic in the financial domain. In 2020, in response to the COVID-19 pandemic, the United States implemented large-scale quantitative easing policies. However, in 2022, to address domestic inflation, the United States began a series of significant interest rate hikes. Under these circumstances, the exchange rates of various currencies have experienced substantial fluctuations. In this study, we propose a novel hybrid model based on the Convolutional Long Short-Term Memory (CNN-LSTM) model, combined with a Residual Network (ResNet), aiming to improve the accuracy of exchange rate predictions. By integrating signal processing techniques such as Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Savitzky-Golay (SG) filters with an innovative dual-stream (DS) architecture, our model (DS-ResNet-LSTM) demonstrates outstanding performance across multiple metrics, significantly outperforming the traditional LSTM, CNN-LSTM and others. The experimental results indicate that the DS-ResNet-LSTM model exhibits strong robustness, high generalization capability, and clear advantages in numerical prediction, demonstrating its potential in financial time series analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of a Dual-Stream Hybrid Network for Exchange Rate Prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月28日

查看全文
  • Si-Qi Chen,
  • Chien-Hsiu Lin,
  • Szu-Lang Liao

摘要

Exchange rate prediction has consistently been a popular research topic in the financial domain. In 2020, in response to the COVID-19 pandemic, the United States implemented large-scale quantitative easing policies. However, in 2022, to address domestic inflation, the United States began a series of significant interest rate hikes. Under these circumstances, the exchange rates of various currencies have experienced substantial fluctuations. In this study, we propose a novel hybrid model based on the Convolutional Long Short-Term Memory (CNN-LSTM) model, combined with a Residual Network (ResNet), aiming to improve the accuracy of exchange rate predictions. By integrating signal processing techniques such as Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Savitzky-Golay (SG) filters with an innovative dual-stream (DS) architecture, our model (DS-ResNet-LSTM) demonstrates outstanding performance across multiple metrics, significantly outperforming the traditional LSTM, CNN-LSTM and others. The experimental results indicate that the DS-ResNet-LSTM model exhibits strong robustness, high generalization capability, and clear advantages in numerical prediction, demonstrating its potential in financial time series analysis.