Statistics Quant Interview Questions
Statistics interview prep for quant roles: distributions, variance, covariance, sampling, regression intuition, and inference.
Quant researcher and trader candidates refreshing statistics for interviews.
What statistics rounds test
Statistics questions test whether you understand uncertainty in data, not whether you can recite formulas. Expect distribution intuition, variance and covariance, sampling error, regression interpretation, hypothesis testing, and - above all for research roles - judgment about when a result is simply noise. The question that separates candidates is rarely "what is the p-value" and almost always "what would make you stop believing this."
Concrete example
Two uncorrelated bets, each with variance 1, give a sum with variance 2. Give them covariance 0.5 and the sum has variance 3, because Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y) and the cross term is counted twice. This is not trivia - it is the entire reason diversification works. Average ten uncorrelated unit-variance signals and the average has variance 1/10, so a standard deviation of about 0.32; average ten perfectly correlated ones and the variance is still 1, so you have gained nothing at all. Most portfolio and signal-combination follow-ups are a version of that arithmetic.
How to practice
Practice explaining formulas in words, because the interview is testing interpretation rather than recall. For a regression coefficient, say what it means, what assumptions make it useful, and what would break it - omitted variables, non-stationarity, or the fact that the specification was chosen after looking at the data. For distributions, connect shape to mean, variance, and tail behavior instead of memorizing densities. One drill is worth more than the rest: take any statistical claim and generate the two most plausible alternative explanations before you accept it.
Common mistakes
Three errors recur. Candidates apply tests without checking their assumptions. They confuse independence with zero correlation, when two variables can have correlation exactly zero and still be perfectly dependent - X and X squared are, for any X symmetric about zero. And they treat a backtest as proof, forgetting that a strategy selected from hundreds of candidates looks good by construction. Whatever the question, say what evidence would change your confidence; that sentence is very often the thing being tested.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.