Scholarlink

<p>We conduct a quasi-natural experiment that highlights the significance of domain knowledge in forecasting the aggregate bank failure rate using machine learning (ML) algorithms. We find that ML algorithms outperform the predictive OLS (POLS) for a specific set of predictors. When the important variables selected by ML algorithms are incorporated into POLS for an in-sample analysis, all variables demonstrate statistical significance. However, we show that it is important to exercise caution as some ML predictions may contradict economic intuition. This experiment emphasizes the significance of incorporating domain knowledge when utilizing ML techniques for making informed policy decisions and investment choices.</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

Domain Knowledge Matters: Evidence from Bank Failure Rate Predictions with Machine Learning

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Ujjal K. Chatterjee,
  • Joseph J. French

摘要

We conduct a quasi-natural experiment that highlights the significance of domain knowledge in forecasting the aggregate bank failure rate using machine learning (ML) algorithms. We find that ML algorithms outperform the predictive OLS (POLS) for a specific set of predictors. When the important variables selected by ML algorithms are incorporated into POLS for an in-sample analysis, all variables demonstrate statistical significance. However, we show that it is important to exercise caution as some ML predictions may contradict economic intuition. This experiment emphasizes the significance of incorporating domain knowledge when utilizing ML techniques for making informed policy decisions and investment choices.