Quant Interview

The hiring filter quantitative trading firms use, and — going by the MIT guide built around it — the single most documented artifact of the field from outside. Firms publish almost nothing about their systems; what leaks out is what they ask candidates.

Four kinds of question

• Brainteasers and probability. Expectation, conditional probability, order statistics, optimal stopping. A representative Jane Street sequence: price a d6 roll (3.5), then the same roll with one optional re-roll (4.25 — re-roll only below 3.5), then with the re-roll costing $1, then with infinite $1 re-rolls, where the answer requires seeing that the continuation value is the game’s own value and solving X = (3.5 − 4.5p)/(1 − p). Recursion on your own valuation, under time pressure.
• Market-making games. Quote a two-sided market, get traded on, requote — see market-making. Typical prompts: regular-season wins for a baseball team, the population of Tanzania, the number of windows in the interviewer’s own building. What is scored is theo-value generation with stated assumptions, confidence intervals sized to genuine ignorance, “trader memory” (position and PNL tracked across the whole sequence), and updating on what each of the interviewer’s trades reveals.
• Research case studies. Open-ended modelling: predict NY housing prices, forecast opera-house attendance, model CitiBike demand. Graded on feature brainstorming and the reasoning about each feature’s hazards — cyclicality, collinearity, geography — then on the model discipline: z-scores per variable, F-tests to drop groups of correlated insignificant ones, and regularized regression (ridge / lasso / elastic-net) or stepwise selection for dimensionality reduction.
• Coding. Increasingly conventional data-structures-and-algorithms screens; the guide names Two Sigma, HRT, Akuna and Belvedere as shops that give them, and puts Cracking the Coding Interview on the reading list.

The filter shapes the field

Two things follow from what gets asked, and both are visible in this wiki’s other sources:

• The idiom is statistics, not machine learning. The preparation canon is Elements of Statistical Learning, an econometrics course, and stochastic processes. Nothing in the guide’s research sections asks for a neural network. The gap between that and the corpus’s software — RL agents (reinforcement-learning-trading), boosted-tree pipelines (supervised-learning-trading), transformer forecasters (kronos-financial-foundation-model) — is a gap between how the field hires and what its open-source edge builds. Some of it is the guide’s 2022 vintage; some of it is that the interview tests reasoning under a whiteboard’s constraints.
• Pricing under ignorance is the actual skill. Nearly every format above rewards producing a defensible number from thin information, stating the assumptions, and revising as evidence arrives. That is the same loop backtesting automates and the same one an after-action-review runs by hand — the discretionary version of this wiki’s central discipline.

Preparation as an economy

The guide’s own reading list is evidence of how institutionalized the filter is: Heard on the Street, Mark Joshi’s Quant Job Interview Questions, Xinfeng Zhou’s “green book,” Mosteller’s Fifty Challenging Problems, Natenberg on options (the guide claims Optiver teaches directly from it), plus Glassdoor question postings, LeetCode and Kaggle. A body of published material exists whose only purpose is passing this interview, and MIT students map it onto a course-road years in advance.

The guide’s stated motive is that the filter’s difficulty is partly self-fulfilling — the field reads as an in-group for geniuses, so only people who already believe they belong apply. Whether preparation guides widen the gate or just professionalize it is not something a single self-interested source can settle.

Related

quant-bible-mit-sloan · market-making · quant-trading-firm · algorithmic-trading · supervised-learning-trading · synthesis