Quant FX research and execution
About XTRSK
XTRSK is a Nuremberg-based quantitative FX trading desk. We build statistical systems and run them on our own execution stack, because in modern FX the model is only half the trade. The other half is whether it reaches the market before the market moves.
What we do
Alpha is a measurement, not a story
XTRSK builds trading systems for retail and institutional clients who want exposure to quantitative FX, not another opinion about direction. Every model we run is built from tick data, scored out of sample, and priced against the cost of trading it. If an edge disappears once spread, commission and slippage are subtracted, we do not ship it — no matter how good the equity curve looked in research.
The desk works in one narrow lane on purpose. FX and metals are over-the-counter, fragmented across venues, and priced by a stream rather than a single exchange tape. That fragmentation is the reason a statistical approach can work there at all, and it is also the reason the engineering matters more than the indicator. We research the microstructure, we build the execution path, and then we publish what the account actually did.
Statistical research
Models are fitted on historical ticks, then judged on data they have never seen. A system that only works in the sample it was fitted to is a curve, not an edge.
Execution engineering
We own the order path: FIX 4.4 sessions, colocated infrastructure, fill-or-kill routing and a cost gate that decides whether a signal is worth sending at all.
Published evidence
The live account is public and the drawdown is reported next to the return. Clients do not have to take our word for how the system behaves.
Why the old model stopped working
The half second that decides everything
A conventional expert advisor is built on a request-and-wait loop. The terminal analyses a candle, sends an order, and waits for the broker to answer. A round trip of a few hundred milliseconds is normal; five hundred is common, and on a fast print it can be far worse. Gold can reprice several times inside that window.
By the time the confirmation comes back, the order has answered a question the market stopped asking. That is the uncomfortable truth about traditional EAs: at that latency they are not trading the market they analysed. The logic is often reasonable. The speed makes it irrelevant.
Infrastructure
Speed, tick density and the cost of being late
In FX there is no single price. There is a stream of prints from banks, ECNs and liquidity providers, and the same instrument can be quoted at two different prices at the same moment. Our infrastructure exists to see that stream clearly and to act on it inside the window where the difference is still real.
That means colocated servers close to the liquidity, FIX 4.4 sessions with direct market access, and our own market-data store rather than a charting feed. It also means measuring tick density: how many updates arrive per second, how deep the book is, and whether the tape is thick enough to trust or thin enough to misprice. A model that ignores tick density is trading a smoothed average of a market that never existed.
Colocation
Execution sits next to the liquidity it trades, not on a home connection. Distance is latency, and latency is a cost paid on every trade.
FIX 4.4 and DMA
Orders route directly to the venue with fill-or-kill or immediate-or-cancel handling. No requote loop, no waiting for a dealer to accept.
Tick-level data
Our own store of minute and tick history feeds research and live charts alike, so what we test and what we trade come from the same source.
Cost gate
A signal is only sent if the opportunity clears spread, commission and expected slippage by a multiple. The rest are declined before they exist.
Risk before return
One position at a time, stops written before the entry, nothing carried overnight. Drawdown is designed, then measured.
Monitoring
Every fill, rejection and gap between equity and balance is recorded, reviewed and published. Outages are visible, not hidden.
The team
Mathematicians who ship, engineers who read the tape
XTRSK is a small group of PhDs in mathematics and computer science, working alongside execution engineers. The research side covers stochastic processes, market microstructure, numerical optimisation and statistics. The engineering side writes the FIX sessions, the tick store, the risk gates and the monitoring that keeps them honest.
That combination is deliberate. In this business a model without an execution path is an academic paper, and an execution path without a model is a fast way to lose money. We keep both under one roof because the edge only exists where they meet.
Research
Statistical modelling, signal discovery, out-of-sample testing and the measurement of decay.
Execution
Low-latency routing, broker connectivity, fill quality and the difference between a price quoted and a price done.
Engineering
Data pipelines, monitoring, and the discipline of treating every outage as a trading event.
Transparency
You can check every number we publish
Most vendors show a backtest and ask you to trust the forward curve. We publish the live account instead: the current balance, the closed profit, the drawdown, and the equity curve exactly as the venue records it. Our monthly reports lead with drawdown and put the return second, because that is the order in which the numbers actually matter.
The record is not curated. Closed trades are posted as they happen, the account is flat between trades, and the widest gap between equity and balance is measured rather than described. If the curve retraces, that number appears next to the return in the same size type.
Where we apply it
NIC-X: the system we are running now
NIC-X is our displaced-print system. When the same instrument is quoted at two different prices at the same moment, the system takes the mispriced side, pays the round trip, and keeps the difference. It is closer to statistical arbitrage than to a directional bet, it is not latency arbitrage, and it never relies on price returning to an average.
It runs one position at a time, flat overnight, with the stop agreed before the entry exists and an outer drawdown limit set with the client. That is the discipline the research was built to support — and it is where we would start with you.
Displaced prints
Two prices for one instrument at one moment. NIC-X arbitrages the gap instead of forecasting the next move.
Cost-gated
The gap has to clear the round trip by a multiple before an order is sent. Anything smaller is declined.
Risk first
Low drawdown by construction: one position, defined stops, no grid, no averaging, nothing held into the next session.
Practical questions
How we work with clients
Is XTRSK a fund?
No. XTRSK licenses trading systems and supports the infrastructure around them. Your capital stays in your own account with your own broker; we do not take custody of client funds.
What do you need from me to start?
A broker route that supports FIX API or cTrader, an account size that suits the system, and a server location close enough to the venue to keep execution inside the window.
Why publish the live account instead of a backtest?
Because a backtest measures the model and the live account measures the model plus the execution. Only the second one is a business.
Can you guarantee a return?
No, and we would not work with anyone who wants that in writing. Trading leveraged products carries risk, and the published record is evidence of process, not a promise about the future.
Start with the live record
If the numbers above are the ones you want to interrogate, the fastest route is the NIC-X live account: current balance, closed profit, the drawdown and the full public history in one place. From there, a trial and a conversation about your broker route take about a week.
Past performance is not a guarantee of future results. Trading leveraged products involves risk. Figures on this page are taken from the public record of the live account on the date they are published.
Chat with XTRSK
Chat ready
Start a chat and the XTRSK team will be notified immediately.