Sven Karbach

I am an assistant professor for data-driven mathematical modelling and computing in finance at the University of Amsterdam, and the coordinator for the Stochastics and Financial Mathematics Master Program. For questions, please contact me.
I am affiliated with:
- the Stochastics group at the Korteweg-de Vries Institute;
- the Computational Science Lab (CSL) at the Informatics Institute;
- and part of the AI4Fintech initiative.
Ongoing PhD projects I supervise:
- Robust finance in infinite-dimensional models — with Diogo Sousa Franquinho.
- Hedging and trading in energy markets — with Konstantinos Chatziandreou.
- Machine learning for high-frequency financial time-series — with Gianmarco Morbelli.
I recently received a research grant of €25,000 to work on "Deep Spatio-Temporal Hedging for Weather and Climate Risk Mitigation in Renewable Energy Markets" [Read more]
Together with Simon Trimborn, I received a €35,000 grant for our project on "Risk Networks of Renewable Energy Markets". [Read more]
You can find a short CV here and a list of publications here.

A side project — World Cup 2026 odds & simulation dashboard

🏆 Spain beat Argentina 1–0 in the final — my model's number one pick, at 24.99%. Of the six forecasts on the dashboard, mine gave the eventual champion the highest title probability — against a bookmaker consensus of 15.13%, an edge of +9.86 points and the best log score of the six — and it picked 5 of the 8 quarter‑finalists to the market's 4. It also owned the biggest miss of the tournament: runner‑up Argentina at just 4.63%, the lowest of any system. Forecasts were frozen on 8 June 2026 and have not been refitted; the dashboard now scores all six against the full bracket.

For fun, I scraped the official draw, built a Monte‑Carlo simulator of the full 48‑team tournament, and calibrated team strengths from several sources — live bookmaker odds (a multi‑book median consensus), the betting exchange, an Elo model, the FIFA ranking, and squad transfer values. The dashboard compares what each system implies for the title race and the edge versus the market, and now scores all six against the actual bracket. Pick a model, search a team, or sort the table. Open the full dashboard ↗

Probabilities are Monte‑Carlo estimates from 10,000 simulated tournaments per model, built with a small Python engine. The dashboard is a single self‑contained file.