Master's student in Finance and Asset Management at Paris 1 Panthéon-Sorbonne.
I work on empirical asset pricing, mostly factor models and regime switching, and on the data problems that come before any of it. Most filings that matter were never meant to be read by a machine, and a good part of my work is making them readable.
These are personal projects. Each one started from a question I ran into in my academic work or during a research internship, then I rebuilt it here from scratch, on public data and with my own code.
edgar-fund-holdings-pre2019 Reconstructs what a US mutual fund actually held, from the raw HTML of its SEC filings. Before N-PORT made this structured in 2019, the schedule of investments sat somewhere inside a filing with no fixed layout. The pipeline finds it, repairs the headers, and classifies the assets. Run at scale it recovered 38 million holdings across 4,700 funds.
regime-switching-portfolio Asks whether letting correlations move with market conditions builds a safer minimum-variance portfolio. Extends the DCC so its two parameters depend on a market regime estimated separately, then scores it on forecast accuracy and on the risk the portfolio actually took. Those two rankings disagree, and the simplest estimator often wins the second one.
climate-nlp-fund-disclosures Measures how much a fund talks about climate change in its shareholder reports, and how it frames it, following the method of Sautner, van Lent, Vilkov and Zhang (Journal of Finance, 2023). Applied to 20,000 filings.
Working on satellite data applied to agricultural commodities, reading observations of the growing regions to build price signals on coffee. Not published yet.