Neural Networks for Financial Hedging: A Deep Hedging Implementation

The main goal of this thesis is to develop a method for hedging a port- folio of derivatives using neural networks and deep reinforcement learning techniques. The first four chapters introduce the theoretical foundations of the tools and methods used, while the last one presents the results of their practical implementation, showing how the proposed hedging approach is supported by both theory and results obtained. The algorithms underly- ing the deep hedging strategy are model-independent, and in this work, the Heston model is used to generate asset prices in a realistic yet controlled setting. This approach shows how modern machine learning techniques can offer flexible and efficient solutions for managing the risk of complex port- folios, making it a promising tool for both academic research and practical applications in finance.