XTX Markets Interview: Process, Questions and How to Pass 2026

What to Expect from an XTX Markets Interview

XTX Markets is London's largest and most successful homegrown systematic trading firm. Founded in 2015 by Alex Gerko (formerly of GSA Capital) and a small team, the firm now has roughly 250 employees and is one of the largest FX market makers in the world. XTX is famously focused, almost monastic in its approach to research - the firm publishes very little, recruits a small number of researchers and engineers each year, and pays exceptionally well. The interview process reflects all of those characteristics: short, intense, and unusually focused on machine learning fluency.

This guide covers XTX's process for quantitative researcher and software engineer roles, the question types that come up most, and a six-week preparation plan that maps onto the Quantt coding tests.

---

XTX Markets at a Glance

• Founded: 2015 by Alex Gerko
• Headquarters: London, UK
• Size: ~250 employees globally
• Offices: London, New York, Singapore, Yerevan, Mumbai, Bristol
• What they trade: FX (largest single dealer globally for some currency pairs), equities, fixed income, futures
• Roles they hire: Quantitative Researcher, Software Engineer, FPGA Engineer, Trading Systems Engineer
• Application route: careers page or via the XTX Markets firm page on Quantt

For broader context on London quant jobs, see the London quant jobs page.

---

The Interview Process

Stage 1: Online Assessment

For software engineer candidates, two or three hard algorithmic problems in 90 to 120 minutes. For quantitative researcher candidates, a combined ML and probability test of similar length, often with a small data analysis component.

XTX's OA is widely considered one of the harder ones in the industry. Pass rates are reportedly around 5 to 10%.

Stage 2: Phone Screen

A 60-minute call with a working researcher or engineer. Live coding for engineers; ML and statistics discussion for researchers.

Stage 3: Onsite Super Day in London

Five to six interviews in a single day at XTX's London office. Each session is 45 minutes to an hour. The interviewers are working researchers and engineers throughout - HR plays a minimal role in the technical content.

Stage 4: Decisions

Offers usually arrive within 1 to 2 weeks of the Super Day.

---

How Interviews Differ by Role

Quantitative Researcher

XTX's research is heavily ML-driven, more so than at most peers. Expect deep questions on neural networks, ensemble methods, regularisation, the practical pitfalls of training models on financial data, and the mechanics of how predictions get translated into trading decisions.

Typical split: 40% machine learning theory and practice, 25% probability and statistics, 25% Python and data manipulation, 10% behavioural.

Software Engineer

A FAANG-style technical interview with a low-latency twist. C++ is heavily favoured. Expect questions on lock-free data structures, the C++ memory model, and trading-systems design.

Typical split: 50% coding, 30% systems design, 10% C++ specifics, 10% behavioural.

FPGA Engineer

A specialist track. Expect Verilog/VHDL, FPGA architecture, deterministic-latency design, and questions about the hardware-software interface.

---

Real Question Types

Machine Learning (Researcher)

Question 1: Cross-validation in time series You have a time series of FX prices. You want to validate a predictive model. What is wrong with K-fold cross-validation, and what should you do instead?

Approach: Standard K-fold randomly shuffles observations across folds, which leaks future information into the training set. For time series, use forward-walking validation: train on data up to time T, test on T+1 to T+H, then advance the window. Be explicit about the validation horizon and the assumed holding period.

Question 2: Regularisation You have a linear model with 1000 features and 10000 observations. The unregularised model achieves an R-squared of 0.6 in-sample but 0.05 out-of-sample. What is going on, and what would you do?

Approach: Severe overfitting. Try L1 (lasso) for feature selection - many of the 1000 features are likely noise. Try L2 (ridge) for shrinkage if features are correlated. Try elastic net for both. Cross-validate the regularisation parameter using forward-walking validation. Also consider whether 10000 observations is genuinely independent (clustered samples reduce effective sample size).

Probability and Statistics

Question 3: Conditional expectation X and Y are independent N(0, 1). What is E[X | X + Y = z]?

Approach: By symmetry and linearity, E[X | X + Y = z] = z / 2. The interviewer is testing whether you remember the result and can derive it (use the joint Gaussian property: conditional expectation is linear).

Coding

Question 4: Streaming statistics Implement a class that supports add(x) and stddev() over the last N observations, both with O(1) amortised complexity.

Approach: Maintain a circular buffer of the last N values, plus running sums of x and x^2. On each add, subtract the value being evicted and add the new one. Variance = mean(x^2) - mean(x)^2.

Question 5: Lock-free SPSC queue Sketch a single-producer single-consumer lock-free queue. What memory orderings do you need, and why?

Approach: A ring buffer with two atomic indices: head (written by producer, read by consumer) and tail (written by consumer, read by producer). Producer uses release on its store of head; consumer uses acquire on its load of head. The acquire-release pair ensures the consumer sees the data the producer wrote before updating head.

Behavioural

Question 6: Failure Tell me about a model you built that failed in production. What did you do?

Approach: XTX cares about post-mortem rigour. Describe the model, what you expected, what went wrong, how you discovered it, and what you changed. Avoid stories where the failure was someone else's fault.

---

How to Prepare - A Six-Week Plan

Weeks 1-2: Foundations. For research candidates, work through The Elements of Statistical Learning (Hastie, Tibshirani and Friedman) chapters 1 to 9, plus our statistics for quantitative trading and machine learning finance guide. For engineering candidates, 100 LeetCode hard problems with a strict time budget.

Weeks 3-4: XTX-specific depth. For research candidates, read Advances in Financial Machine Learning (Marcos Lopez de Prado) and Pattern Recognition and Machine Learning (Bishop). For engineers, deepen your C++ knowledge - Effective Modern C++ (Meyers) and C++ Concurrency in Action (Williams).

Weeks 5-6: Mock interviews and timed practice. Three full mock onsites under realistic conditions. The XTX Super Day is intense - the only way to inoculate yourself is to practise tired.

For broader context, see our high-frequency trading guide and latency arbitrage explained.

---

What XTX Looks For Beyond Technical Skill

Three traits separate XTX offers from rejections.

Genuine ML depth, not surface familiarity. XTX hires people who can derive ML methods from first principles, not just import sklearn classes. Candidates who can explain why ridge regression has a closed-form solution while lasso does not, or why batch normalisation has the effect it does, do better than candidates who can list ML techniques without depth.

Quiet confidence under pressure. XTX's interviewers will challenge correct answers and push deeper into edge cases. Candidates who treat the pushback as a signal to update or refine their reasoning do well. Candidates who become defensive do not.

Genuine interest in markets. Despite the ML-heavy interview content, XTX is fundamentally a market-making firm. Candidates who can articulate a real interest in why markets work the way they do, beyond just "I want to apply ML to a hard problem", outperform.

For broader context, see our quantitative analyst career guide.

---

Compensation & recruiting notes

Pay ranges in this guide are illustrative estimates from publicly discussed bands and anecdotal reports - not official figures from the employer. Packages vary widely by desk, office, performance, and year. Hiring processes change; nothing here guarantees an interview, assessment format, or offer.

---

Frequently Asked Questions

How hard is it to get an XTX interview?

Extremely competitive. XTX hires roughly 30 to 50 researchers and engineers globally each year and receives a very high volume of applications. The OA is the great filter - typical pass rates are reported around 5 to 10%.

Does XTX hire from non-target universities?

Yes, more meritocratically than most. The firm's founder Alex Gerko has spoken publicly about wanting to hire raw mathematical talent regardless of background. Strong OA performance and demonstrable signals (Putnam, ICPC, ML competition wins) are the route in for non-target candidates.

What programming languages should I know for an XTX interview?

For research roles, Python is required - and the depth tested is genuine, not surface (expect questions on numpy broadcasting internals, pandas memory layout, vectorisation). For engineering roles, C++ is essentially required. Some teams use Rust.

How does XTX's compensation compare to other quant firms?

XTX is widely discussed as one of the highest-paying market makers, particularly for research roles. Publicly discussed bands for London graduate researchers have sometimes been cited around £200,000 to £350,000 total compensation - figures vary widely by team and year and should be treated as illustrative only. Senior researchers earn into seven figures, often substantially so - XTX has a reputation for paying its top performers exceptionally well. See our quantitative analyst salary guide for cross-firm context.

How many interview rounds does XTX have?

OA, one phone screen, and a final-round Super Day with five to six interviews. Total time from application to offer is typically 4 to 6 weeks.

What is the difference between XTX and Citadel?

Both pay extremely well and both run highly automated trading. Citadel is a multi-strategy hedge fund and market maker; XTX is purely a market maker (predominantly FX). Culturally Citadel is much larger and more corporate; XTX is small, focused and almost monastic about its research process. We have a dedicated Citadel interview guide.

Can I reapply if rejected?

Yes, after 12 months. XTX takes repeat candidates seriously, particularly those who can show measurable improvement.