Comparative Evaluation of Hybrid Deep Reinforcement Learning Models with Fixed Risk-Reward Ratios for Intraday Gold Trading

Ravi Prajapati, Kobid Karkee

Abstract

This study proposes a hybrid trading framework that combines a Rule-Based Trading Strategy with a Deep Reinforcement Learning (DRL) agent for intraday gold trading on a 15-minute timeframe. The Rule-Based Trading Strategy generates rule-based signals, while the DRL agent determines whether to execute or skip the generated trading opportunities. The framework evaluates three separate DRL models using fixed risk–reward (RR) ratios of 1:1, 1:2, and 1:3. Volatility-adjusted stop-loss levels are determined using the Average True Range (ATR), with a fixed risk of 0.5% per trade. Each model is trained and evaluated independently under identical trading conditions, enabling a systematic comparison of the effects of different fixed RR ratios on DRL-based trading performance. The trading environment incorporates transaction costs and risk-based position sizing, while the reward mechanism reflects the realized outcome of each trade under the corresponding fixed RR configuration. Model performance is evaluated using cumulative return, maximum drawdown, win rate, profit factor, and average return per trade. The study compares the three fixed-RR models to determine which risk–reward configuration provides the most favorable balance between profitability, risk, and trading consistency in intraday gold trading.

Source: semanticscholar · PDF

Read the AI summary, key takeaways and discussion on WOBR Quant Research.