A Machine learning approach to anomaly detection on derivative pricing

Risk management has become an essential part of the process for any company that trades derivatives. Banks and corporations own risk-management offices dedicated to the development of risk models involving market, credit and operational risk, assure controls are operating effectively, and provide research and analytical support. Their models and operations have to be compliant to current European Central Bank legislation, which supervises the process. In the thesis we provide a new machine learning approach to detect anomalies in option pricing models. The research focuses on a preliminary analysis to assess the applicability of the method, starting with an analysis of call options quoted on the S&P500 index, with historical data from 2010 to 2021. The model consists in forecasting the value of implied volatility and using it to check the results provided by other pricing models used in companies. This is achieved by using three different machine learning models for regression: artificial neural network, random forest and gradient boosting regression. The best results were given by the application of the first model. The analysis of the results shows that our approach can accurately solve the problem and be applicable in a real-world working application, with an integration in the workflow of risk-management offices, providing real-time alerts on errors of various nature that can occur in the pricing process. In the future, this approach could be extended to work on other types of financial derivatives, by collecting the necessary data to train the machine learning algorithms, while another field of research could instead focus on trying out other machine learning models, looking for an improvement of the forecast of implied volatility.