FORECASTING COMMODITY MARKET VOLATILITY: A COMPARATIVE ANALYSIS OF TRADITIONAL AND MACHINE LEARNING MODELS
Abstract: This study examines how conventional and machine learning models perform in predicting and forecasting volatility in the Indian commodity market and study involved seven key commodities i.e., gold, silver, crude oil and metals. The study applied GARCH, GARCH plus LSTM, and machine learning models like Random Forest, BiLSTM and XGBoost. The study also uses Principal Component Analysis. This study employs daily data for the period spanning from January 2003 to December 2025. Data was sourced from the Multi Commodity Exchange (MCX). While traditional GARCH models are helpful in capturing volatility persistence, they are insufficient for accurately forecasting volatile commodity markets. It was concluded that Machine learning models performed better, especially hybrid ones that mix GARCH and LSTM. Findings matter for investors, policymakers and analysts who need better ways to understand risk in fast-changing markets. It supports the use of data driven tools to manage market risk in a volatile environment, specifically focusing on Indian commodity market.
Keywords: Volatility Forecasting, Commodity Markets, Machine Learning, GARCH, Financial Risk