Scholarlink

<p>Accurate stock trend prediction is vital for financial decision-making, yet existing methods struggle to jointly model multivariate correlations and multi-scale temporal patterns. We propose MA-MSTNet, a novel Mixed Attention-based Multi-Scale Temporal Network that integrates Maximal Information Coefficient-guided attention to quantify feature dependencies while suppressing noise, multi-scale sliding window attention capturing high-to-low frequency features, and dynamic gated fusion for adaptive feature integration. Experiments on four stock indices (S&amp;P 500, Nasdaq, etc.) demonstrate state-of-the-art performance (R<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7700_Article_IEq1.gif" Format="GIF" Height="10" Rendition="HTML" Resolution="72" Type="Linedraw" Width="8" /> </InlineMediaObject> <EquationSource Format="TEX">\(^2\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>2</mn> </mmultiscripts> </math></EquationSource> </InlineEquation> up to 0.984), outperforming seven baselines. Crucially, the model achieves 6.5ms inference latency and 39% faster training per epoch, enabling real-time high-frequency trading.</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

MA-MSTNet: mixed attention-based multi-scale temporal network for stock trend prediction

  • 期刊论文

  • 发表日期：

    2025年8月3日

查看全文
  • Xin Wang,
  • Xiang Zhang,
  • Jifei Liu,
  • Zhenyu Zuo

摘要

Accurate stock trend prediction is vital for financial decision-making, yet existing methods struggle to jointly model multivariate correlations and multi-scale temporal patterns. We propose MA-MSTNet, a novel Mixed Attention-based Multi-Scale Temporal Network that integrates Maximal Information Coefficient-guided attention to quantify feature dependencies while suppressing noise, multi-scale sliding window attention capturing high-to-low frequency features, and dynamic gated fusion for adaptive feature integration. Experiments on four stock indices (S&P 500, Nasdaq, etc.) demonstrate state-of-the-art performance (R \(^2\) 2 up to 0.984), outperforming seven baselines. Crucially, the model achieves 6.5ms inference latency and 39% faster training per epoch, enabling real-time high-frequency trading.