Application of Transformer-DDQN Model in Straddle Options Trading Strategy

Machine learning strategies in algorithmic trading are becoming critical in today's volatile financial markets, where data randomness prevails.Existing research has predominantly focused on predicting stock prices, a method vulnerable to extreme market conditions resulting in significant losses.On the other hand, straddle options, which combine call and put options, exhibit stability during normal market fluctuations and seek superior benefits during dramatic market swings.Our innovative approach includes a high-fidelity simulation of the real options trading environment that accurately reflects significant market shifts.Furthermore, we have refined the model's efficiency by incorporating resistance levels and volatility data into the reward function, thereby enhancing exploratory behaviors and learning effectiveness.Accordingly, we develop a straddle options trading strategy that capitalizes on price volatility to earn profits, regardless of the market's directional movement.In detail, we first created an interactive environment to simulate the real options trading market, capturing significant market fluctuations with high fidelity.Then we construct a Transformer-DDQN model to efficiently automate the decision-making for the optimal points to open and close positions and execute trades.Finally we enhance our model by integrating resistance levels and volatility information, refining the reward function to foster exploration and effective learning.Compared with traditional stock price prediction strategies, our straddle options trading strategy demonstrates its superiority through higher returns, improved Sharpe ratios, and reduced maximum drawdown rates, especially in markets with significant volatility.