Natural disaster shocks and commodity market volatility: A machine learning approach

Abstract

This study examines the efficacy of machine learning and deep learning techniques for forecasting volatility in commodity prices triggered by natural disasters. By integrating varied natural disaster indicators as exogenous variables, the study trains and evaluates the predictive capability of an array of machine learning methodologies, encompassing tree-based algorithms, support vector machines, and particularly neural networks. The standout performance of neural networks, especially the Nonlinear Autoregressive with Exogenous inputs (NARX) model, underscores their superior accuracy over both other machine learning approaches and conventional statistical models. This superior performance is consistent across different definitions of commodity price volatility, ensuring the robustness of our results beyond the risk of overfitting. The implications of such precise predictive modeling are important, promising to enhance risk management tactics, agricultural planning, and investment strategies in the aftermath of natural disasters. Ultimately, our findings make a compelling argument for the integration of sophisticated predictive analytics into the fabric of economic policymaking and planning, presenting a strategic blueprint for fortifying market resilience against the impacts of natural disasters.

Keywords: Commodity, Market Volatility, Machine Learning