설명 가능한 인공지능을 활용한 단기 콜옵션 가격 예측 연구

초록

This study aims to validate the performance of machine learning-based option pricing prediction model and ensure transparency in the prediction by utilizing explainable artificial intelligence (XAI) technique. I conducted a comparative analysis of the predictive performance between the Black-Scholes and the XGBoost models using call option data from Apple Inc. Both models employed the same input variables including underlying asset price, strike price, days to expiration, time-varying risk-free rate, and 30-day historical volatility, with actual call option prices as the target output variable. The experimental results demonstrate that the XGBoost model significantly outperformed the Black-Scholes model in predictive accuracy. The XGBoost model achieved approximately 87.9% performance improvement based on MSE, with error reductions of 62.9% and 65.3% in MAE and RMSE, respectively. Through actual versus predicted call option price comparisons, I confirmed that the XGBoost model exhibited consistent predictive accuracy across all price ranges. Feature importance analysis using SHAP identified underlying asset price and strike price as the most critical variables for option price prediction. This study distinguishes itself from existing research by employing XAI technique to provide transparency in machine learning-based option price prediction.

키워드