Consider stock indices and technical indicators to conduct price forecasting for gold and oil: a reinforcement learning perspective

Abstract

Rising oil prices often trigger inflation, whereas higher gold prices reflect risk aversion. Analyzing daily data from 2011 to 2025, this study proposes a novel framework to predict commodity prices: (1) raw materials, stock indices, and technical indicators are integrated as the predictors, (2) key performance indicators (KPIs) are identified using MARS (multivariate adaptive regression splines), RF (Random Forest), and XGB (eXtreme gradient boost), and (3) machine learning (ML), deep learning (DL) models including LSTM (long short-term memory) and GRU (gated recurrent unit), and reinforcement learning (RL) strategies such as A2C (advantage actor–critic) and PPO (proximal policy optimization) are compared. Specifically, one-shot splitting starts from 2011 to 2022 for training and uses 2023–2025 is for testing. In dynamic splitting, four-year training and one-month testing is adopted. Results show that average exchange rate of US dollar (USD), coal, India’s stock, Shanghai’s stock, US10YY (10-year yield rate of US treasury bond), Japan’s stock, DJI (US Dow Jones index), and NASDAQ (US technology index) affect oil prices while silver, S&P 500, German’s stock, SOX (US semiconductor index), India’s stock, and NASDAQ influence gold prices. Dynamic splitting and RL models outperform one-shot splitting because they can capture short-term variations and abrupt changes.