Quantitative Researcher - Deep Learning in London

The FirmXTX Markets is a leading algorithmic trading firm. We build models that forecast prices for more than 53,000 financial instruments across equities, fixed income, currencies, commodities and crypto. Those forecasts are used to trade on exchanges and other venues, and to provide liquidity to clients. The firm trades over $250bn a day across 35 countries, with over 300 people in London, Singapore, New York, Paris, Bristol, Mumbai, Yerevan and Kajaani. XTX is a technology business. Trading here is driven by research, large-scale computation and careful engineering.

Over the last decade our models have moved from the econometric methods that gave the firm its name, through trees and neural networks, to modern deep learning. We expect that evolution to continue, and we are looking for researchers who can push it. The research platform is large by any standard, and particularly unusual in trading: well over 25,000 GPUs and more than 1 exabyte of usable storage, with further compute being built in Finland. The point is simple: good ideas should not be bottlenecked by data, infrastructure or compute.

This role sits in XTX's central Quantitative Research team, the group responsible for the firm's core forecasting research. The team reports directly to XTX's founder and is intentionally flat. There is little ceremony and little hierarchy; researchers are expected to have ideas, test them properly, argue from evidence and get useful work into the system. The work is modern ML research, not finance by another name. You do not need to have worked in trading. The best candidates usually have strong research taste, mathematical depth and the ability to make models work in practice. An interest in markets helps, because the domain is full of useful structure, but trading intuition can be built here.

Trading is a hard setting for machine learning in ways that are not always obvious from outside the industry. The data is enormous, messy, non-stationary and produced by other agents. Signals are weak and often short-lived. Evaluation is unforgiving: costs, latency, risk and capacity all matter, and a model that looks good in a backtest may still be useless in the market. The time scales also range from microseconds to weeks, so there is no single right abstraction. Your job will be to do original research that improves XTX's forecasting and trading systems. Depending on your background, that may mean:

• Designing and training deep learning models for large-scale time-series and cross-sectional prediction.
• Improving representations, objectives, optimisation, regularisation, uncertainty estimates, evaluation methods or scaling.
• Making models more robust under distribution shift and feedback from the market.
• Working with engineers and trading technologists to turn research into production systems.
• Learning enough market structure to ask better modelling questions.

This is not about applying standard methods to tidy datasets. It is a place to do serious ML research on difficult data, with enough compute and infrastructure for ambitious ideas to be tested properly.

What We Are Looking For

• Evidence of original technical ability in ML, statistics, mathematics, computer science, physics or another quantitative field. Publications are one route, but not the only one; strong industrial research or major open-source contributions can also be good evidence.
• Deep practical understanding of modern deep learning: architectures, objectives, optimisation, scaling, evaluation and failure modes.
• Strong mathematical and statistical foundations.
• Excellent programming ability and comfort working close to the details of implementation.
• Good experimental taste: forming hypotheses, designing clean tests, reading ambiguous results and changing your mind quickly when the evidence changes.
• High standards, intellectual honesty and the ability to work in a flat team where ideas are challenged directly.
• No finance background is needed; the important thing is being willing to learn the domain.

Beneficial Experience

• PhD or equivalent research experience in a relevant technical field.
• Experience training large models or building ML systems at meaningful scale.
• Work in time-series modelling, online or continual learning, reinforcement learning, causal inference, probabilistic modelling, optimisation, compilers, HPC or low-latency systems.
• Experience taking research beyond a prototype.

Benefits

• Onsite gym, sauna and fitness classes.
• Extensive medical benefits, including onsite doctor and therapist.
• Breakfast and lunch provided daily.
• Support for caregivers, including emergency dependent care.
• King's Cross office.
• 25 days paid holiday per year, plus statutory holidays and paid sick days.
• Generous pension contributions.