Evaluating Predictive Power of Fundamental and Technical Indicators Using Machine Learning in Commodity Markets

Abstract

Commodity markets are characterized by high volatility and complex price dynamics, making accurate forecasting crucial for effective trading and investment decisions. This study investigates the predictive power of fundamental and technical indicators in commodity markets using machine learning (ML) approaches to enhance forecasting accuracy and trading decision-making. Historical data from Yahoo Finance for commodities including gold, silver, copper, aluminum, and natural gas were analyzed, encompassing price series, volume, macroeconomic variables, and derived technical indicators such as SMA, EMA, RSI, MACD, ATR, and Bollinger Bands. A structured methodology involving data preprocessing, feature engineering, and exploratory analysis was employed to identify price trends, return dynamics, volatility regimes, and market activity patterns. Rule-based signal generation using RSI and MACD produced buy, sell, and hold signals, while ML models Decision Tree, Random Forest, and Gradient Boosting—assessed predictive performance. Model evaluation metrics included accuracy, precision, recall, F1-score, alongside financial measures such as Sharpe Ratio, maximum drawdown, and cumulative return. Results showed varied predictive performance across commodities: Silver achieved perfect classification (accuracy, precision, recall, F1-score = 1.00), Copper Random Forest yielded 99% accuracy with 0.83 F1-score, and Gold Decision Tree achieved 96% accuracy. Integrating technical indicators with machine learning provides a systematic framework for commodity price forecasting, offering actionable insights for risk-adjusted trading strategies in volatile markets.