Quant Interview Questions: Trading, Research and Dev

Work out which of the three you are interviewing for

Job descriptions are unhelpfully similar and the titles are not standardised, so read for the verbs rather than the nouns. A posting that talks about quoting, positions, risk limits and P&L is a trading seat. One that talks about signals, features, backtests and horizons is research. One that talks about latency, throughput, deployment and infrastructure is development.

This matters because the failure mode is asymmetric. Arriving at a trading interview having read three statistics textbooks and no market-making practice is a bad afternoon. Arriving at a research interview able to quote a fast market but unable to say why a backtest overstates a Sharpe ratio is a worse one.

• Trading — speed, decisions, market-making games, immediate scoreboard
• Research — probability and statistics, signal construction, longer feedback loops
• Development — data structures, complexity, systems, low-latency reasoning

Mental arithmetic, and why it is tested at all

Nobody does long multiplication on a modern desk, and firms know that. It is tested because it is a cheap, hard-to-fake proxy for working at speed without becoming careless, and because it is almost perfectly correlated with having practised. That last part is the point: it is one of the few components of the process you can improve reliably.

The standard is higher than most candidates assume. Two-digit multiplication, percentages of awkward numbers, and fractions to decimals should be automatic rather than derived. Where candidates lose is not accuracy but hesitation, and hesitation is what the exercise is measuring.

• Practise to a timer, not to correctness, because correctness comes first and then stops improving
• Learn the shortcuts properly: differences of squares, complements to 100, halving and doubling
• Say answers out loud. Silent practice does not transfer to a room with someone watching

Probability and expected value

The classic problems recur because they are compact tests of whether someone reasons about uncertainty or recites results. Conditional probability, Bayes, expected value of a game with an option to re-roll, and the expected wait for a pattern of coin flips will all appear in some form.

The commonest error is not arithmetic. It is answering a different question: computing the probability of the evidence given the hypothesis when asked for the reverse, which is exactly what the medical-testing question is built to catch. State which conditional you are computing before you compute it.

• Expected number of flips to see two consecutive heads — set up the recursion, do not guess
• A 99% accurate test for a 1-in-10,000 disease — the answer is counter-intuitive and the point is that it is
• A die you may re-roll up to three times — work backwards from the last roll, always

Market-making games

You are asked to quote a two-way price on something uncertain, the interviewer trades against you, and the exercise continues. It is not really about the estimate. It is about whether your spread reflects your uncertainty, whether you move your quote after being hit, and whether you notice what being filled instantly tells you.

If someone lifts your entire offer without hesitation, the information content of that is the whole exercise: they know something, or you are mispriced, and either way your next quote should move. Candidates who hold their quote after being run over are demonstrating the trait the game exists to detect.

• Quote wider when you know less. A tight spread on a number you cannot estimate is not confidence
• Update after every fill, and say why you are updating
• Manage the position you have accumulated, not the one you meant to have

Narration is half the assessment

On a desk you think out loud so someone can interrupt you before the loss gets larger. Interviews test the same thing, which is why a correct silent answer often scores below a slightly wrong narrated one. The interviewer is trying to find out whether they could work next to you.

Being wrong well is a specific skill and it is being watched for. The failure modes are symmetric: defending a first answer against a good objection, and abandoning it at the first push. What is wanted is updating on the argument rather than on the social pressure.

For research seats: what a backtest cannot tell you

A research interview will find out quickly whether you have ever been fooled by your own results. Overfitting, look-ahead bias, survivorship in the universe, and the multiple-comparisons problem created by trying two hundred signals are the standard territory, and the question is usually posed as an invitation: here is a strategy with a good Sharpe, what would stop you trading it?

The strong answer is not a list of biases. It is naming which one you would check first and how, because that is the actual work.

• How many things did you try before this one worked?
• Does the signal survive out of sample, and was the out-of-sample period genuinely untouched?
• What are the transaction costs, and does the edge survive them at realistic size?

Timelines run early

Quant recruiting starts ahead of banking and closes on a rolling basis rather than a published date. Several of the largest trading firms open summer applications a year or more ahead and stop once the class is full, which means a candidate working to a banking calendar can find the process finished before they begin.

Check each firm directly rather than assuming a cycle. The tracker on this site carries live openings and their stated deadlines where a firm publishes one.