A hybrid approach to option pricing integrating unsupervised and supervised machine learning models

This thesis explores a novel approach to option pricing by integrating unsupervised and supervised machine learning models. The objective is to assess whether these models can outperform the traditional Black-Scholes-Merton model and to investigate the effect of clustering on prediction accuracy. The analysis uses data from the Ivy DB US database, focusing on S&P 500 options traded on the CBOE from December 30, 2019, to December 30, 2022. The methodology involves applying K-means clustering to segment the dataset, followed by training Random Forest and Deep Neural Network models on these clusters. The models' performances are then compared against non-clustered machine learning models and the BSM model. The results show that the hybrid machine learning approach improves option pricing accuracy. Specifically, the Deep Neural Network models achieve a median improvement of 39.1%, while the Random Forest models show a median improvement of 5.2%. This suggests the potential of integrating advanced clustering techniques in financial modeling for more precise option pricing.