Scholarlink

<p>Risk prediction in financial markets is a critical task for effective decision-making under uncertainty. The growing complexity of causes of financial risks, driven by the surge in data volume and diversity, presents significant challenges for traditional predictive models that primarily rely on single data sources, struggling to cope with complex and volatile market environments. To address these limitations, we propose a novel Transformer-based financial risk prediction framework (FRP-TNN) that integrates historical stock trading data and financial news sentiment data to examine the risk state characteristics of the financial market from multiple dimensions. By employing causal convolution to enhance feature extraction and introducing a local attention mechanism, the framework is optimized for capturing intricate risk patterns. Comparative experimental results demonstrate that FRP-TNN achieves superior predictive accuracy, exceeding 95% and outperforming other comparative models by an average of 6.12% across key evaluation metrics. This study highlights the potential of advanced deep learning methods to inform strategic financial decision-making, enabling managers and investors to navigate uncertain and dynamic market environments with greater confidence.</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

FRP-TNN: an intelligent framework using transformer for financial market risk prediction to enhance decision-making under uncertainty

  • 期刊论文

  • 发表日期：

    2025年7月24日

查看全文
  • Bo Xu,
  • Lifei Lu,
  • Wenwen Tu

摘要

Risk prediction in financial markets is a critical task for effective decision-making under uncertainty. The growing complexity of causes of financial risks, driven by the surge in data volume and diversity, presents significant challenges for traditional predictive models that primarily rely on single data sources, struggling to cope with complex and volatile market environments. To address these limitations, we propose a novel Transformer-based financial risk prediction framework (FRP-TNN) that integrates historical stock trading data and financial news sentiment data to examine the risk state characteristics of the financial market from multiple dimensions. By employing causal convolution to enhance feature extraction and introducing a local attention mechanism, the framework is optimized for capturing intricate risk patterns. Comparative experimental results demonstrate that FRP-TNN achieves superior predictive accuracy, exceeding 95% and outperforming other comparative models by an average of 6.12% across key evaluation metrics. This study highlights the potential of advanced deep learning methods to inform strategic financial decision-making, enabling managers and investors to navigate uncertain and dynamic market environments with greater confidence.