XTX Markets is London's largest homegrown systematic trading firm and one of the biggest FX liquidity providers in the world, trading tens of thousands of instruments across more than eighty venues. It is a machine learning driven firm, and the interview reflects that far more than it reflects the trading-floor stereotype.
Set your expectations before you start. XTX hires a small number of researchers and engineers each year against a very high volume of applications, and its online assessment is widely considered one of the hardest in the industry, with reported pass rates around 5 to 10 percent. This is not a process you can walk into unprepared.
XTX is not an options market maker in the mould of IMC, Optiver, SIG or Flow Traders. There are no make-me-a-market dice games, very few classic brainteasers, and rapid mental arithmetic is not the point. The firm tests genuine depth in machine learning, statistics and coding instead.
That distinction should shape your preparation. If your background is fast mental maths and market making intuition, you have a lot of ground to cover. If it is statistical learning, applied probability and serious programming, XTX may suit you better than most trading firms. Numeracy practice still builds the foundation, but the interview itself looks nothing like the ones covered elsewhere in our knowledge base.
Worth reading alongside this guide: the official XTX careers page, which sets out their side of the process.
- 1Online assessment. Hard algorithmic, or machine learning and probability, problems under time pressure.
- 2Phone screen. Around 60 minutes with a working researcher or engineer, technical from the first minute.
- 3Onsite super day. 5 to 6 back-to-back technical interviews in a single day at the London office.
- 4Decision. Offers typically arrive within 1 to 2 weeks.
The content depends heavily on the role, so this guide covers both the quantitative researcher and software engineer tracks.
Stage 1: Online assessment
Sent to strong candidates once the hiring manager has reviewed your CV. The format depends on the role.
| Track | What you sit |
|---|---|
| Software engineer | Two or three hard algorithmic problems in a 90 to 120 minute window. Difficulty sits at the harder end of LeetCode, with a premium on clean efficient solutions rather than brute force |
| Quantitative researcher | A combined machine learning and probability test of similar length, often with a small data analysis component where you work with a dataset directly |
| Trading analyst | A home assignment on trading knowledge and Python, rather than a timed coding test |
This stage is the great filter. The problems are genuinely difficult and the time pressure is real.
How to prepare
- Engineers: drill hard algorithmic problems against a strict time budget, since speed and correctness are both being measured. Our mental arithmetic practice keeps your raw numeracy sharp alongside the coding work.
- Researchers: be fluent in core machine learning rather than able to call library functions. Rebuild the probability and statistics foundations with our quant interview questions guide and the probability questions.
Stage 2: Phone screen
Around 60 minutes, by phone or video, with a working researcher or engineer. There is very little HR involvement: the technical bar is set from the first conversation.
- Engineers. Live coding, usually with a focus on efficiency and edge cases.
- Researchers. A discussion of machine learning and statistics that probes how deeply you understand the methods you claim to know.
How to prepare
Practise talking through your reasoning out loud while you code or derive, because the interviewer is watching how you think rather than collecting an answer. Expect to be challenged: XTX interviewers push back on correct answers to see whether you can defend or refine them.
Stage 3: Onsite super day
Five to six interviews in a single day at the London office, each 45 minutes to an hour, conducted by working researchers and engineers throughout. It is sustained and intense, and fatigue is part of the test.
The rough mix candidates report, by track:
| Area | Researcher | Engineer |
|---|---|---|
| Machine learning theory and practice | 40% | |
| Probability and statistics | 25% | |
| Coding | 25% (Python and data manipulation) | 50% |
| Systems design | 30% | |
| C++ specifics | 10% | |
| Behavioural | 10% | 10% |
The quantitative researcher track
XTX research is more machine learning driven than at most of its peers. Three areas come up again and again.
Machine learning
Neural networks, ensemble methods, regularisation and the practical pitfalls of training models on financial data. You should be able to explain why ridge regression has a closed-form solution while lasso does not, why batch normalisation works, and how a model's predictions actually turn into trading decisions.
Validation
A classic XTX area. Be ready to explain why standard K-fold cross-validation leaks future information in a time series, and why forward-walking validation, training up to time T then testing on the following window before advancing, is the correct approach.
Overfitting
Given a model with many features and few genuinely independent observations, diagnose the overfitting and reason about L1, L2 and elastic net regularisation, along with effective sample size when your samples are clustered.
Probability and statistics
Clean derivation-based questions rather than trick puzzles. A representative one: if X and Y are independent standard normals, what is the conditional expectation of X given that X plus Y equals z?
Derive it from the joint Gaussian properties rather than recalling the result.
The probability questions are the right drill for the underlying reasoning, even though XTX asks for derivations rather than puzzles.
Coding
Coding appears on both tracks. A common type is streaming statistics: implement a class supporting add and standard deviation over the last N observations in O(1) amortised time, using a circular buffer with running sums of x and x squared.
The software engineer track
A FAANG-style technical interview with a low-latency twist. C++ is heavily favoured and some teams use Rust. Expect lock-free data structures, the C++ memory model and performance-critical design: sketching a single-producer single-consumer lock-free queue, explaining the acquire-release ordering the memory model requires, and trading systems design.
Behavioural
Light but real. A common prompt is to describe a model or system you built that failed in production: what you expected, what went wrong, how you found it and what you changed. XTX values post-mortem rigour, so avoid any story where the failure was someone else's fault.
For the standard motivation questions, prepare why you want to work here and what makes you stand out.
Stage 4: Decision
Offers usually arrive within 1 to 2 weeks of the super day, and the whole process from application to offer typically runs 4 to 6 weeks. If you are rejected, XTX accepts reapplications after 12 months and takes repeat candidates seriously, particularly those who can show measurable improvement.
What XTX looks for beyond technical skill
Three traits separate offers from rejections.
- Genuine depth, not surface familiarity. XTX hires people who can derive methods from first principles rather than list techniques. Explaining why something works, not just that it works, is the differentiator.
- Quiet confidence under pressure. Interviewers challenge correct answers and push into edge cases. Treat that as a prompt to refine your reasoning rather than as an attack, because candidates who get defensive do badly.
- Genuine interest in markets. Despite the machine learning content, XTX is fundamentally a market making firm. Candidates who can say why markets work the way they do, beyond wanting to apply machine learning to a hard problem, stand out.
One encouraging note: XTX is known for hiring more meritocratically than most. A strong assessment and demonstrable signals such as competition results, Putnam, ICPC or machine learning competition wins, are a real route in from a non-target background.
Closing remarks
The XTX interview rewards depth over breadth. The firm is small, focused and almost monastic about its research process, and the interview is a concentrated test of whether you can do the actual work: building and validating predictive models on messy financial data, or engineering the low-latency systems that turn those models into trades.
The common mistake is preparing for XTX as though it were a generalist trading interview. The make-me-a-market and rapid mental maths drills that serve you well at an options market maker will not carry you here. Put the time into machine learning fundamentals, statistical reasoning and serious coding practice instead.
These roles are exceptionally competitive and the firm takes only a small number of people each year. Go in with deliberate preparation, an honest assessment of your gaps, and the composure to think clearly under sustained pressure. Work through everything we have for XTX Markets, and read our full interview guide for how these stages compare across firms.