Scholarlink

<p>This study explores Bitcoin’s role as a conditional safe haven during times of U.S. macroeconomic stress, providing one of the first divergence-based, explainable machine-learning frameworks for this purpose. Using daily data from January 2020 to March 2025, we combine Bitcoin, Gold, the S&amp;P 500, the VIX, CPI, and the Federal Funds Rate into a macro-sensitive classification system. A new safe-haven flag identifies episodes when Bitcoin appreciates while both Gold and the S&amp;P 500 decline, capturing flight-to-safety behavior. We compare Logistic Regression, Random Forest, and XGBoost classifiers, showing that XGBoost consistently outperforms traditional models on recall, F1, and ROC-AUC metrics even under class imbalance. SHAP-based explainability reveals that divergence indicators between Bitcoin and traditional assets dominate predictions, followed by lagged macroeconomic factors. The findings show that Bitcoin does not act as a constant hedge but exhibits episodic, stress-driven safe-haven characteristics. Our XAI framework, based on divergence, gives investors and regulators a clear, repeatable way to track systemic shocks and pinpoint safe-haven behavior under different conditions. By combining substitution and divergence logic with explainable machine learning, it takes theory forward. It also informs practice by giving risk managers probabilistic decision bands and guides policy by showing how interpretable AI can improve oversight of new digital asset markets.</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

Classifying bitcoin’s safe haven role using explainable AI: evidence from US macroeconomic stress episodes

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • B R Manjunath

摘要

This study explores Bitcoin’s role as a conditional safe haven during times of U.S. macroeconomic stress, providing one of the first divergence-based, explainable machine-learning frameworks for this purpose. Using daily data from January 2020 to March 2025, we combine Bitcoin, Gold, the S&P 500, the VIX, CPI, and the Federal Funds Rate into a macro-sensitive classification system. A new safe-haven flag identifies episodes when Bitcoin appreciates while both Gold and the S&P 500 decline, capturing flight-to-safety behavior. We compare Logistic Regression, Random Forest, and XGBoost classifiers, showing that XGBoost consistently outperforms traditional models on recall, F1, and ROC-AUC metrics even under class imbalance. SHAP-based explainability reveals that divergence indicators between Bitcoin and traditional assets dominate predictions, followed by lagged macroeconomic factors. The findings show that Bitcoin does not act as a constant hedge but exhibits episodic, stress-driven safe-haven characteristics. Our XAI framework, based on divergence, gives investors and regulators a clear, repeatable way to track systemic shocks and pinpoint safe-haven behavior under different conditions. By combining substitution and divergence logic with explainable machine learning, it takes theory forward. It also informs practice by giving risk managers probabilistic decision bands and guides policy by showing how interpretable AI can improve oversight of new digital asset markets.