Matteo Maggiolo

I hold a Master of Science in Engineering (MSEngineering) from SUPSI, where I specialized in financial machine learning and reinforcement learning under the supervision of Dr. Oleg Szehr.

During my studies, I worked on several projects at Dalle Molle Institute for Artificial Intelligence USI-SUPSI. One project involved developing an MDL-based feature selector in PySpark. Another, co-authored with Dr. Oleg Szehr, explored the use of neural networks for financial portfolio optimization. In this work, we introduced a novel statistical method to detect overfitting in end-to-end machine learning-based portfolio optimization and investigated regularization techniques to stabilize neural network training. This research was published in Risk’s Journal of Risk Model Validation, with the editor noting: “This is a valuable paper, and Maggiolo and Szehr shed light on concerns that many of us have about neural networks.” We presented this work at the Bachelier Finance Society 12th World Congress in Rio de Janeiro in July 2024.

My master’s thesis focused on applying reinforcement learning to derivatives hedging. We explored custom versions of MCTS-based algorithms and developed tailored AlphaZero and MuZero variants for hedging European, American, and multi-option strategies. Parts of this thesis are being published as a standalone research paper in collaboration with Dr. Oleg Szehr and Dr. Miroslav Strupl, to be presented at the SIAM Conference on Financial Mathematics and Engineering (FM25) in Miami. You can read the preprint here.

After graduation, I co-founded Vigilant Analytics, a scientific consultancy firm that builds AI-driven solutions for private and public sector clients. Our core expertise lies in machine learning—particularly reinforcement learning, computer vision, and stochastic optimization.

selected publications

Deep Hedging Under Non-Convexity: Limitations and a Case for AlphaZero

Matteo Maggiolo, Giuseppe Nuti, Miroslav Štrupl, and 1 more author

2025

Abs

This paper examines replication portfolio construction in incomplete markets - a key problem in financial engineering with applications in pricing, hedging, balance sheet management, and energy storage planning. We model this as a two-player game between an investor and the market, where the investor makes strategic bets on future states while the market reveals outcomes. Inspired by the success of Monte Carlo Tree Search in stochastic games, we introduce an AlphaZero-based system and compare its performance to deep hedging - a widely used industry method based on gradient descent. Through theoretical analysis and experiments, we show that deep hedging struggles in environments where the optimal action-value function is not subject to convexity constraints - such as those involving non-convex transaction costs, capital constraints, or regulatory limitations - converging to local optima. We construct specific market environments to highlight these limitations and demonstrate that AlphaZero consistently finds near-optimal replication strategies. On the theoretical side, we establish a connection between deep hedging and convex optimization, suggesting that its effectiveness is contingent on convexity assumptions. Our experiments further suggest that AlphaZero is more sample-efficient - an important advantage in data-scarce, overfitting-prone derivative markets.