Alongside academic research in probability and random networks, I work on quantitative modelling and machine learning in financial contexts. My contribution is typically methodological: translating a financial question into a well-defined stochastic or statistical problem, examining model assumptions, constructing and validating features, and assessing robustness and interpretability.
Quantitative modelling and model governance
I serve on the Advisory Board of Proficiency S.r.l. as Scientific Advisor in Quantitative Modelling and Model Governance. The role brings an independent academic perspective to fund selection and asset allocation, drawing on probability, stochastic systems, complex networks, machine learning and AI.
Typical methodological questions:
- whether a model’s assumptions match its intended use;
- robustness to sampling choices, regime changes and distribution shift;
- validation design, and the separation of in-sample fit from decision-relevant evidence;
- interpretability of features and model outputs;
- risk modelling and stress scenarios;
- governance of quantitative, data-driven and AI-based tools.
The aim is to make modelling choices explicit, testable, and proportionate to the decisions they support.
Regime-aware machine learning for asset allocation
Since 2023 I have collaborated with Alessandro Greppi and Francesco Mercatelli on machine-learning approaches to portfolio allocation that combine time-series forecasts with indicators of market regime and dependence.
The work treats market regime as a latent, evolving state describable through time-series, network and information-theoretic features — recurrent architectures for forward-looking allocation objectives, sentiment information, the principal eigenvalue of a rolling equity-correlation matrix as an indicator of market synchronisation, and multivariate dependence measures such as O-information and transfer entropy.
The research interest is not a single forecast but the construction of a system whose features, objectives and risk controls can be validated separately, and whose behaviour can be compared against transparent benchmarks.
No investment advice is provided on this page, and no performance figures, portfolio weights or proprietary data are published here.
Teaching in quantitative finance
I taught Clustering Methods within SIAT’s Data Science for Asset Management course, and have supervised student projects on graph neural networks for cryptocurrency markets, data-driven portfolio allocation, and European corporate-bond pricing.
See the CV for the full teaching and supervision record.