<aside> 🔍
This is a case study. It maps a domain and points to where the real, messy problems live. It deliberately stops before telling you what to build or which problem to pick, because that investigation is the track. If we handed you a tidy problem statement, we'd be doing the exact legwork Fundamentum asks you to do yourself.
</aside>
This year, Fundamentum is about quant trading firms. Pick one foundational problem a quant firm actually lives with, investigate it more deeply than anyone else in the room using real sources and real people, and build something your research tells you is right.
Quantitative trading firms (Jane Street, Two Sigma, Optiver, Akuna, IMC, the local prop shops) look, from the outside, like they've solved everything. Smartest people in the room, best-resourced, most-optimised corner of finance. That's exactly why they're a good subject: the interesting problems aren't the ones on the recruiting posters. They're the unglamorous, foundational ones that persist because everyone assumes a firm that sophisticated must have handled them already.
And a lot of you are the quant talent pipeline. You're in the Quantitative Trading Society or UNIT, you've done the Akuna and Optiver trading games, some of you have interviews lined up. So you have something most teams researching an industry don't: direct access to people who live inside it. Your peers, the society alumni now a year into these desks, the recruiters chasing you. Use that.
Your job is not to out-trade a quant firm. You can't, and a weekend won't change that. Find one foundational problem these firms live with, understand it more deeply than a surface reading allows, and build something grounded in that understanding.
Skip the two obvious ideas. Almost everyone first reaches for latency ("make trades faster") or alpha models ("find a better signal"). You can't build them in a weekend (sub-millisecond execution takes years and millions in hardware; a novel signal is what these firms hunt full-time), and you can't research them either (the details are the most guarded IP in the industry, so you'd be left paraphrasing blog posts). The good problems are less obvious: real to practitioners, researchable by an outsider willing to do the legwork, and prototype-able at hackathon scale.
These are starting points. Part of your research is confirming whether the problem is even real. Start here, then narrow to one.
Talent and knowledge transfer (start here). The one direction where you have the best access in the building. Quant firms run on high churn, and when a quant leaves, their undocumented knowledge often leaves with them. There's a cluster of problems here: how juniors actually get up to speed, how research survives turnover, how the hiring process looks broken from the candidate's side. You can interview real people about all of it: QTS and UNIT peers, alumni now on desks, recruiters. That's the heart of what Fundamentum rewards.
The reproducibility gap. Most academic trading strategies don't survive contact with real capital, largely from overfitting and lookahead bias. Firms fight a quieter internal version: a strategy that shone in someone's notebook can't be rebuilt months later because the data, parameters, and code weren't captured. A strategy no one can reconstruct is effectively unusable, however good its backtest looked. The tooling problem here is experiment tracking, data versioning, and provenance is genuinely prototype-able.
Data infrastructure under stress. Market-data systems that are fine on a calm day fall over in high volatility, exactly when they matter most. The underlying tension is concrete: balancing speed, cost, and coverage. There are public industry reports to use as primary sources, and "detect and handle bad or dropped market data" is buildable on free historical feeds.
Alpha decay and model drift. Signals crowd and die as more players exploit them, and shift as markets change regime. The researchable slice isn't building a better model but a fundamentally monitoring problem: knowing a strategy is decaying before it costs you.