Option pricing and hedging with market friction using reinforcement learning

• Main Author: Jiang, Zixing
• Other Authors: Bo An
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Option pricing; Option hedging
• Online Access: https://hdl.handle.net/10356/181126
• Institution: Nanyang Technological University

This paper presents a discrete European option pricing and hedging model under an environment with market friction using Q-Learning in Reinforcement Learning. The research novelty lies in the comparison of performance by various transaction cost models. Reinforcement Learning is implemented with the Q-Learning Black Scholes (QLBS) model. This model is not only presented as an alternative to the canonical Black Scholes (BS) Model, but also offered as an extension to pricing models that commonly overlook realistic market conditions. The QLBS model is augmented to account for the trader’s risk aversion, transaction costs and market impact costs to reflect a risk-adjusted price and an optimal hedge. The baseline BS model is then used as a benchmark to illustrate the how the QLBS model can address the practical considerations of a trader’s hedging needs. The performance of two solutions, namely Dynamic Programming and Fitted Q Iteration, will be evaluated. By going model free and retaining the model’s simplicity solely built on linear algebra, this data driven approach can depart the academic limit of the Black Scholes continuous time framework and be extended as a practical tool for further strategy testing and optimisation.