TL;DR
Quant interviews test four things: probability and statistics under time pressure, programming, market and risk intuition, and how clearly you reason aloud. Most firms run a timed online test, then interviews that move from puzzles to research or trading judgement.
Interviewers score the method as much as the answer. Stating assumptions, checking edge cases and explaining your reasoning usually matters more than reaching the result fastest.
Key points
- Say your assumptions and approach out loud; silence loses marks even when the answer is right.
- Practise expectation, conditional probability and simple Markov chain arguments until they are automatic.
- Know the limits of your own backtests: overfitting questions come up in almost every research interview.
- Trading interviews test risk discipline, not bravado.
Probability
1. Expected tosses to see two heads in a row. Let E be the expected number of tosses. With probability one half you toss a tail and start again, using one toss. With probability one quarter you toss heads then tails, using two tosses and starting again. With probability one quarter you toss two heads and finish. Solving E = 0.5(1 + E) + 0.25(2 + E) + 0.25(2) gives E = 6.
2. Two fair dice are rolled. What is the probability the sum is 7, given at least one die shows a 3? There are 11 outcomes with at least one 3. Two of them sum to 7, namely (3,4) and (4,3). The probability is 2 in 11.
3. You break a stick at two uniformly random points. What is the probability the pieces form a triangle? The condition is that no piece exceeds half the stick. The probability is one quarter.
4. A random variable is uniform on the interval from 0 to 1. What is the expected maximum of n independent draws? The maximum has distribution function x to the power n, so the expectation is n divided by (n + 1).
Statistics
5. You test 100 signals at the 5% significance level and 7 appear significant. What do you conclude? Under the null hypothesis you would expect about 5 false positives by chance, so 7 is close to what randomness produces. The right response is to correct for multiple testing, use out-of-sample data and check whether any signal has an economic rationale.
6. What happens to a regression when two predictors are highly correlated? Coefficient estimates become unstable with large standard errors, even though the model's overall fit may be fine. Remedies include removing a variable, combining them, or using regularisation.
7. Why can a strategy have a high Sharpe ratio and still be dangerous? Because Sharpe uses total volatility and assumes reasonably symmetric returns. Strategies with rare, large losses look excellent until the loss arrives. See Sharpe and Sortino ratios.
8. How would you test whether a signal's performance is genuine? Hold out data, test across regimes, apply realistic costs, correct for the number of variants tested, and check whether results depend on a single period or instrument.
Brainteasers and estimation
9. You have two ropes that each burn in one hour but not uniformly. How do you measure 45 minutes? Light rope one at both ends and rope two at one end. When rope one is finished, 30 minutes have passed, so light the other end of rope two. It burns out 15 minutes later, giving 45 minutes.
10. Estimate the number of ATMs in a large Indian city. Interviewers want a structured estimate: population, banked share, branches per bank, ATMs per branch, then a sanity check. The number matters less than the structure and your willingness to state assumptions.
Coding
11. Compute a rolling correlation between two long series efficiently. Discuss vectorised operations, avoiding recomputation with running sums, handling missing data, and ensuring no look-ahead bias by using only past data at each point.
12. Given trade data, find the largest drawdown. Track the running maximum of the equity curve and the largest gap below it in a single pass, which is linear in time and constant in extra memory.
Market and risk intuition
13. Your live strategy suddenly loses three times its usual daily loss. What do you do? Reduce risk first, then diagnose: is it a data problem, a system problem, a market regime change or expected tail behaviour? Escalate and document. Interviewers look for risk discipline before curiosity.
14. You are making a market and someone keeps buying from you. What does that tell you? That you may be on the wrong side of informed flow. You would widen your spread or shift your quotes, and reassess your inventory and fair value estimate.
15. How do transaction costs change a signal's value? They reduce net returns, particularly for high-turnover signals, and can invert the ranking of strategies. Strong candidates discuss both explicit costs and market impact.
Behavioural questions
- Tell me about a research idea that failed. What did you learn?
- Describe a time you disagreed with a colleague about a result.
- What is the most interesting thing you have learned about markets recently?
Answer with specifics, including the numbers and the decision you made. Vague answers are read as inexperience.
How interviewers score you
| Dimension | Weak | Strong |
|---|---|---|
| Approach | Jumps to formulas | States assumptions, structures the problem |
| Accuracy | Frequent slips under pressure | Careful, checks results |
| Communication | Silent or vague | Thinks aloud clearly |
| Honesty | Overclaims, bluffs | Says what they do not know |
| Curiosity | Answers only what is asked | Asks good questions back |
How to prepare
Practise timed problem sets daily, work through past projects until you can defend every choice, and do mock interviews where you explain reasoning aloud. For engineering roles, see the HFT C++ and system design interview guide.
Frequently asked questions
What topics do quant interviews cover?
Quant interviews cover six areas in most firms: probability and combinatorics, statistics and inference, brainteasers and estimation, programming in Python and sometimes C++, market and risk intuition, and a detailed discussion of your own projects, research or trading. The weighting shifts by track. Research rounds lean on statistics and research judgement, trading rounds on risk and decision-making under uncertainty, engineering rounds on systems and performance.
How hard are quant interviews?
They are demanding, mainly because of time pressure and depth of follow-up questions. Preparation in probability and coding, plus practice explaining your reasoning, makes a large difference.
Do quant interviews ask brainteasers?
Many quant interviews still include brainteasers, but the purpose is to watch how you decompose an unfamiliar problem rather than to test whether you have seen it before. Estimation questions, sometimes called Fermi problems, are common for the same reason. Interviewers are listening for whether you state assumptions, work in structured steps and sanity-check your own answer. Memorised solutions are usually obvious and count against you.
How should I answer when I do not know?
Say you do not know, then describe how you would work it out. Quant interviewers are explicitly testing calibration, because a researcher who cannot tell a strong result from a weak one is dangerous with capital. Bluffing is nearly always detected and is treated as a serious negative. Stating the boundary of what you know, then reasoning aloud toward the answer, scores better than a confident guess that turns out wrong.