Scholarlink

A classic problem in financial econometrics is predicting stock returns from noisy & high frequency data. Though a strong alternative to traditional linear approaches is presented by Machine Learning (ML) models, the vast majority of applications concentrate on broad-market indices and may overlook important sector-specific trends. In this paper, a sector-specific approach to daily stock return prediction is proposed. It incorporates a Random Forest regressor, customized features, and a three-factor asset pricing model (Market risk premium, SMB & UMD). This framework is employed for the August 2019 to August 2025 Indian banking sector stocks (Bank Nifty). The results show modest explanatory power (highest R-squared of 3.86%) and strong out-of-sample directional accuracy, consistently exceeding 65% and reaching up to 71.29%. In contrast to normal goodness-of-fit, the results suggest that sector-level attention increases the capabilities of ML models, particularly when calculating directional accuracy. In addition to offering a framework that may be used to other industries like IT, pharmaceuticals, or FMCG, the research focuses on sector-level prediction in emerging market strategies.

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

Recasting Fama-French Factors as Predictive Tools: A Sector-Focused Machine Learning Framework for Stock Returns

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Divye Shah,
  • Aarya Dama,
  • Pratik Kanani,
  • Deepali Patil,
  • Shruti Dodani,
  • Prakshal Doshi

摘要

A classic problem in financial econometrics is predicting stock returns from noisy & high frequency data. Though a strong alternative to traditional linear approaches is presented by Machine Learning (ML) models, the vast majority of applications concentrate on broad-market indices and may overlook important sector-specific trends. In this paper, a sector-specific approach to daily stock return prediction is proposed. It incorporates a Random Forest regressor, customized features, and a three-factor asset pricing model (Market risk premium, SMB & UMD). This framework is employed for the August 2019 to August 2025 Indian banking sector stocks (Bank Nifty). The results show modest explanatory power (highest R-squared of 3.86%) and strong out-of-sample directional accuracy, consistently exceeding 65% and reaching up to 71.29%. In contrast to normal goodness-of-fit, the results suggest that sector-level attention increases the capabilities of ML models, particularly when calculating directional accuracy. In addition to offering a framework that may be used to other industries like IT, pharmaceuticals, or FMCG, the research focuses on sector-level prediction in emerging market strategies.