XTX Markets Interview Process & Prep
XTX Markets is a London-headquartered electronic market maker that trades almost entirely off the strength of its statistical and machine-learning models — and became one of the world's largest FX liquidity providers on that basis. Its interview is unusually research-heavy: machine learning, statistics, linear algebra, regression and clean numerical coding are the center of gravity, and the online assessment is widely reported as one of the hardest single filters in the industry. Depth over breadth — interviewers pick one or two areas and probe to the edge of what you know.
The XTX Markets interview funnel
1. Online Assessment
~60–120 min · timed · reportedly TestGorilla or HackerRank-style (role-dependent)XTX's first filter, and the one candidates talk about most — widely reported as one of the hardest online assessments in quant, with pass rates often cited around 5–10% (a community estimate, not an official figure). The format varies by role and year, and sources genuinely disagree on the exact platform, so the only safe move is to prepare for the TOPICS rather than betting on a specific tool. Researchers report a combined machine-learning and probability test, often with a small data-analysis component bolted on. A commonly described general / quant version is roughly 20 questions in about 60 minutes spanning probability, linear algebra, calculus, algorithms and brain-teasers, hosted on TestGorilla. Engineering candidates report HackerRank-style timed algorithmic coding under a hard clock. Two details recur in candidate write-ups: the per-question time budget is tight enough that you cannot afford to re-derive fundamentals on the spot, and you reportedly cannot return to a question once you move past it — so read each one fully, commit, and pace yourself. Because the assessment is the single biggest gate in the whole funnel, most of your preparation time is best spent making the core machinery (MLE, the OLS normal equations, ridge/lasso, PCA and the SVD, numerically stable softmax) fast and automatic rather than chasing exotic topics.
2. Recruiter / First Technical Screen
45–60 min · videoA 45–60 minute remote call after you clear the OA, usually a mix of a recruiter / hiring-manager component and a first technical screen. It blends background and motivation ('why XTX', 'walk me through your research') with one or two technical questions tailored to your area — typically a statistics or ML concept for researchers, or a coding question for engineers. Candidates repeatedly describe XTX interviewers as unusually civil and genuinely curious for a top quant shop: the goal of the conversation is to understand how you actually think and to find the edge of your knowledge, not to trap you with gotchas. That tone cuts both ways — because they probe deeply on whatever you claim, it pays to be precise about your own work and candid about its limitations. Expect to spend real time on one research project or system you built: what the problem was, why you chose the method you did, how you validated it, and what you would change. Engineers should be ready to write and reason about correct, efficient code live, talking through complexity as they go.
3. Take-Home / Technical Data Round (QR track)
Multi-hour take-home or live technical · researcher-heavyResearchers commonly report a take-home assignment somewhere in the middle of the funnel — an open-ended statistical / ML problem on a dataset, sometimes a model-building or data-analysis exercise, sometimes a more theoretical ML question. It is deliberately less about a single right answer and more about how you frame an ambiguous problem, build a sensible baseline before anything fancy, validate out-of-sample, and avoid the classic traps of modeling noisy financial data: overfitting, multiple-testing inflation when you screen many features, look-ahead leakage, and non-stationarity. XTX trades on tiny, fleeting statistical edges, so the firm cares a great deal about whether your conclusions would survive contact with live data rather than just fitting the sample in front of you. Treat the deliverable as work you would hand a colleague — clear structure, reproducible code, honest caveats — and be ready to defend every modeling choice in a follow-up conversation, because they will ask. Engineering tracks may get a longer technical coding exercise in place of the data take-home.
4. Onsite — Coding
60 min · live coding (Python / C++)Part of the onsite 'super day' — several back-to-back 45–60 minute interviews, usually run on-site or over video late in the funnel. The coding round is typically one or two algorithmic problems, medium-to-hard, graded on correctness, edge-case handling, time and space complexity, and — distinctively for XTX — numerical stability. For research-adjacent roles the coding often has a numerical or data flavor: implement an estimator, a numerically stable softmax / log-sum-exp, a clustering step, or a streaming statistic, rather than a pure puzzle algorithm. The signal they are reading is whether you can turn math into code that is both correct and well-behaved under the messy realities of floating point and large inputs. State your complexity up front, confirm the constraints before you start typing, and narrate your edge cases; clean, readable, correct code that you can fully explain beats a terse 'optimal' solution you cannot defend.
5. Onsite — Machine Learning / Research Deep-Dive
60 min · whiteboard / discussion · QR & ML rolesThe round XTX is known for, and the one that most sets it apart from market-making peers. A deep dive into machine learning and statistical modeling: regularization and the bias–variance tradeoff, ensembles (bagging, random forests, boosting), kernels and Gaussian processes, PCA and the EM algorithm, and — above all — the practical pitfalls of training models on financial time series. Interviewers reportedly pick one or two threads and push to the very edge of your understanding, so a candidate who knows three topics cold will do better than one with shallow exposure to ten. Expect 'why does this hold' far more than 'name the algorithm': be ready to derive the OLS normal equations, the first principal component, or why an L1 penalty induces sparsity, and to justify how you would pick a regularization strength or diagnose overfitting. Because XTX is fundamentally a forecasting business operating in extremely low signal-to-noise conditions, the conversation almost always returns to validation: out-of-sample testing, guarding against multiple-testing and leakage, and reasoning honestly about how much of an apparent edge is real. Knowing what you do not know — and saying so — is reportedly a positive signal here, not a weakness.
6. Onsite — Probability, Statistics & Linear Algebra
60 min · whiteboardA super-day round on the mathematical foundations that everything else at XTX is built on. On the probability and statistics side: MLE and the common distributional families, sampling and estimator properties, why we divide by n−1, hypothesis testing and the multiple-testing corrections that matter when you screen many signals. On the linear-algebra side: covariance and positive-semidefinite matrices, the SVD and best low-rank approximation, eigenstructure (including the eigenvalues of structured, equicorrelated matrices), and orthogonalization via QR or Gram–Schmidt. Expect to derive results from first principles and to reason about why an estimator behaves the way it does, not just to recall a formula — and to move fluently between the geometric picture, the algebra, and, when asked, a few lines of code. A reliable way to score well is to set up the objective or probability space explicitly, turn the crank, and then sanity-check your answer with limits, symmetry, and degenerate cases; catching your own error in front of the interviewer reads better than landing a lucky final number.
Net it out: XTX is a research-led, ML-first market maker that interviews more like a top machine-learning lab than like a fast-math trading floor. The online assessment is the dominant gate, so the highest-leverage preparation is making the statistical and linear-algebra core fast and automatic — MLE, the OLS normal equations, ridge and lasso, PCA and the SVD, numerically stable softmax — and being able to derive, not just recall, each one. Past the OA the rounds reward depth and intellectual honesty: pick a few areas you genuinely understand to the bottom, prepare one project or model you can defend end to end including how you validated it out-of-sample, and be candid about the limits of what you know. Practice the representative problems for each stage below — every one carries a full worked solution — and lean hardest on the machine-learning, statistics and linear-algebra sets, which is where XTX concentrates its bar.
XTX Markets interview — FAQ
What is the XTX Markets interview process?
XTX Markets is a London-headquartered electronic market maker that trades almost entirely off the strength of its statistical and machine-learning models — and became one of the world's largest FX liquidity providers on that basis. Its interview is unusually research-heavy: machine learning, statistics, linear algebra, regression and clean numerical coding are the center of gravity, and the online assessment is widely reported as one of the hardest single filters in the industry. Depth over breadth — interviewers pick one or two areas and probe to the edge of what you know. The loop runs 6 stages: Online Assessment, Recruiter / First Technical Screen, Take-Home / Technical Data Round (QR track), Onsite — Coding, Onsite — Machine Learning / Research Deep-Dive, Onsite — Probability, Statistics & Linear Algebra.
How many rounds does XTX Markets have?
6 stages in total, starting with the Online Assessment and ending with the Onsite — Probability, Statistics & Linear Algebra.
How do I prepare for the XTX Markets interview?
Work the stage notes above, then drill the XTX Markets interview-questions set and the XTX Markets online-assessment practice — each problem has a full worked solution.