Scholarlink

<p>Understanding what drives precious-metals pricing has become increasingly complex amid intensified geopolitical uncertainty and heterogeneous market information. Traditional econometric models often fail to reconcile high-frequency financial signals with low-frequency macroeconomic fundamentals, limiting their explanatory and predictive power. To bridge this gap, this study proposes an explainable mixed-frequency machine learning framework, the Variable-Weight Mixed-Frequency Data Sampling Multi-Output Support Vector Regression (VW-MIDAS-MSVR) model, which dynamically integrates multi-scale data through a state-adaptive weighting mechanism guided by the Geopolitical Risk (GPR) index. Using monthly and daily data for palladium, platinum, silver, and gold from 2010 to 2024, the model captures nonlinear interactions among geopolitical risk, market sentiment, and metal-specific fundamentals. SHAP and PDP/ICE analyses provide interpretable evidence of key price drivers, revealing threshold-type effects of lagged returns, mining-sector indicators, and GPR shocks. Empirical evaluation shows that VW-MIDAS-MSVR consistently outperforms benchmark models in accuracy and robustness, while portfolio backtests confirm substantial improvements in risk-adjusted returns. The findings indicate that geopolitical uncertainty and sectoral linkages jointly shape the heterogeneous behavior of safe-haven and industrial metals. Overall, this research takes precious-metal forecasting as an interpretable learning process, demonstrating how adaptive mixed-frequency modeling can uncover the underlying economic mechanisms driving price dynamics in complex global 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

What drives precious metals pricing? An explainable Mixed-frequency machine learning approach

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Xianning Wang,
  • Jian Wu,
  • Yuchen Pan,
  • Ni Zhang,
  • Tuo Wang,
  • Juan Liu

摘要

Understanding what drives precious-metals pricing has become increasingly complex amid intensified geopolitical uncertainty and heterogeneous market information. Traditional econometric models often fail to reconcile high-frequency financial signals with low-frequency macroeconomic fundamentals, limiting their explanatory and predictive power. To bridge this gap, this study proposes an explainable mixed-frequency machine learning framework, the Variable-Weight Mixed-Frequency Data Sampling Multi-Output Support Vector Regression (VW-MIDAS-MSVR) model, which dynamically integrates multi-scale data through a state-adaptive weighting mechanism guided by the Geopolitical Risk (GPR) index. Using monthly and daily data for palladium, platinum, silver, and gold from 2010 to 2024, the model captures nonlinear interactions among geopolitical risk, market sentiment, and metal-specific fundamentals. SHAP and PDP/ICE analyses provide interpretable evidence of key price drivers, revealing threshold-type effects of lagged returns, mining-sector indicators, and GPR shocks. Empirical evaluation shows that VW-MIDAS-MSVR consistently outperforms benchmark models in accuracy and robustness, while portfolio backtests confirm substantial improvements in risk-adjusted returns. The findings indicate that geopolitical uncertainty and sectoral linkages jointly shape the heterogeneous behavior of safe-haven and industrial metals. Overall, this research takes precious-metal forecasting as an interpretable learning process, demonstrating how adaptive mixed-frequency modeling can uncover the underlying economic mechanisms driving price dynamics in complex global markets.