Enhancing Option Pricing and Stock Return Predictions: Integrating Machine Learning with Firm Characteristics and Option Greeks

Abstract

This thesis explores the use of machine learning in financial derivatives, particularly stock options, to improve understanding and prediction of option pricing. It includes three empirical chapters. Chapter 1 evaluates machine learning models that integrate various firm characteristics to predict stock option prices. It introduces two semi-parametric models: a variant of Andreou, Charalambous, and Martzoukos (2010) generalized parametric function model (GPF) and Lajbcygier and Connor (1997)’s hybrid model (HBD), applied to U.S. stock options from 1996 to 2021. The GPF model consistently outperforms the HBD model, with specific firm characteristics emerging as key predictors of option prices. Chapter 2 explores the predictive capacity of option market characteristics, especially implied volatility and Greeks, in forecasting extreme stock returns of the underlying assets. The study employs the LightGBM algorithm, which significantly outperforms traditional logistic regression in predicting stock market trends, emphasizing the value of a comprehensive approach to option dynamics. Chapter 3 builds upon the insights from Chapter 1, focusing on an in-depth analysis of option Greeks and specific firm characteristics within three semi-parametric frameworks: GPF, HBD, and AFFT (Almeida, Fan, Freire, and Tang, 2023). This chapter also explores how these three frameworks maintain consistency with various input features throughout the Pandemic period. Notably, the GPF framework shows exceptional resilience and adaptability when integrated with option Greeks and firm characteristics during the Pandemic. Overall, this thesis underscores the efficacy of incorporating firm characteristics and option Greeks in option pricing and stock return prediction, highlighting the superiority and adaptability of machine learning models in volatile market scenarios.