A Hybrid Deep Reinforcement Learning Approach for Algorithmic Trading in Commodity Futures Markets

Abstract

The financial markets have been affected extensively in recent years, due to algorithmic trading, predominantly concerning commodity futures. This research provides a novel hybrid approach combining classical machine learning algorithms (random forest) with modern deep reinforcement learning algorithms (deep Q-network, proximal policy optimization, and soft actor–critic) for algorithmic trading in the commodity futures market. In contrast to prior investigations which mostly focus on equities, we applied this hybrid model to precious metals (gold) futures using a novel dataset of gold prices augmented with geopolitical risk indices, resulting in outstanding performance on financial risk metrics. The evaluation done suggests that the soft actor–critic agent modeled outshines conventional approaches (buy-and-hold and sell-and-hold) along with deep Q-network and proximal policy optimization, in relation to cumulative profit, Sharpe ratio, maximum drawdown, and annualized volatility. Our results enhance the algorithmic trading methodology growth by supporting problem reduction commonly associated with classical trading techniques or approaches. This study provides significant value for traders and financial institutions intending to implement artificial intelligence in their cognitive trading strategies.