Research / Quant
Quantitative trading systems for FX and metals
Quantitative trading means deciding what to trade from evidence rather than opinion: a rule with parameters, tested on data, sized by a risk model, and judged on results that were not used to build it. Here is how XTRSK applies that to FX and metals — and why the execution path, not the signal, is usually what decides the outcome.
What makes a strategy quantitative
The distinguishing feature is not the language or the platform. It is that the decision rule is explicit, so it can be measured, challenged and falsified. A chart drawn by hand cannot be tested; a rule can. That is also what separates quantitative trading from ordinary algorithmic trading — an algorithm can automate a hunch, while a quant rule has to survive data it has never seen.
Explicit rules
Entry, exit, size and stop are written down precisely enough for a computer to execute them without interpretation.
Testable claims
Every claim is a hypothesis that data can reject. “It works on trends” is not quantitative; “it is positive when the weekly and daily trend agree” is.
Risk as a parameter
Loss per trade and total exposure are inputs to the model, not afterthoughts applied when a position goes wrong.
The pipeline XTRSK uses
The work is a sequence, and skipping a step is how a backtest becomes a story.
1. Data
Historical candles and tick data for the pairs in scope, cleaned and gap-checked before anything is measured on them.
2. Hypothesis
A structural reason the edge should exist — a liquidity pattern, an execution effect, a persistent behavioural bias — stated before it is tested.
3. In-sample fitting
Parameters chosen on a training window, deliberately kept away from the data used to judge the result.
4. Out-of-sample and walk-forward
The rule is then applied to periods it has never seen. A strategy that only works on its training data is discarded, not tuned.
5. Cost modelling
Spread, commission and slippage are charged to every trade. This is the step that most retail backtests quietly omit.
6. Live verification
Finally the rule runs on a real account with real routing, and the result is published where it can be checked.
Why the cost model decides the answer
On a short horizon the edge per trade is measured in pips and the cost of a round trip is also measured in pips. If the cost floor is higher than the average stretch you are trying to capture, the strategy cannot work no matter how good the signal looks. That is not a tuning problem; it is arithmetic.
It is also why XTRSK cares about FIX API connectivity and direct market access rather than chart features. Execution is what determines whether the modelled edge survives contact with a real account.
The number that matters
Cost per round trip versus average captured move. If the first is close to the second, the system is a coin flip with a commission.
The number that misleads
Win rate. A high win rate with an unbounded tail is a losing system that has not lost yet.
The number that survives
Worst month, worst drawdown and whether the result holds out of sample. Those are what a live account actually experiences.
Latency and colocation are part of the model
A quant model that assumes it will be filled at the quoted price is making an execution assumption, not a market assumption. Where the strategy is short-horizon, that assumption is worth modelling explicitly: how far the price can move during your round trip, and what happens to the edge when it does.
Latency is the input to that calculation, and colocation is the lever that controls it. Placing the server in the same facility as the liquidity — LD4 or LD5 in London, Equinix in New York — removes the transport component of the round trip, which in turn shrinks the slippage assumption the model has to defend. Read alongside what makes a system high-frequency, this is the part of quantitative work that most retail research never reaches.
Model the round trip
Decision time plus transport plus venue is the window in which the price can move against you. Shorten it or widen your stops; do not pretend it is zero.
Refuse bad fills
A fill-or-kill order turns slippage from a variable cost into a decision. That is a modelling improvement, not a setting.
Measure, then claim
An execution path that reports its own round-trip time is the only version of “low latency” that can be verified after the fact.
What a quantified FX book looks like
A defensible FX and metals book is usually a small number of rules over a small number of instruments, each with a stated reason to exist, a bounded loss, and a plan for what happens when the month goes badly. Size is a function of the cost floor, not of conviction. And the evidence is a live account that anyone can inspect — not a curve fitted to history.
Few instruments
Liquidity and spread quality matter more than breadth. A pair with a wide spread cannot host a thin edge.
One position at a time
Overlapping exposure makes results unattributable and drawdown unpredictable.
A stated worst case
The stop and the monthly limit are written before the first trade, so the model has a defined failure mode.
The systems this produced
Two live systems came out of that pipeline. PULSE-X is an HFT system on the FIX protocol with fill-or-kill execution and a live verified account. NIC-X looks for displaced prints in over-the-counter FX and metals and takes them with FOK orders, with no resting orders and one position at a time.
The backtesting method, including how the samples are split and what is deliberately not optimised, is documented on the backtesting page. The live result is on the live account page.
Quantitative trading questions
What is the difference between quantitative and algorithmic trading?
Algorithmic trading automates execution; quantitative trading derives the rule from data and validation. Most serious systems are both — a quantified rule, executed by an algorithm.
How much history do you need to validate a rule?
Enough to cover more than one market regime, split so the rule is judged on data it never saw. Length matters less than the honesty of the split.
Why does latency belong in a quant model?
Because the model assumes a fill price. The longer the round trip, the more the market can move inside it, so latency is part of the cost the model has to overcome.
Can a quantitative strategy work on a small account?
Often not, for cost reasons rather than signal reasons. If the round-trip cost exceeds the average captured move, size does not rescue the arithmetic.
How is the XTRSK research verified?
The method is published (see backtesting) and the outcome is tracked on a live account, which is the only evidence that cannot be fitted after the fact.
Further reading
For the conduct and market-structure background to trading FX and metals, see the Global Foreign Exchange Committee and its FX Global Code. For the execution protocol the systems above use, see the FIX Trading Community standards.
Quantitative methods discipline a strategy; they do not remove risk. Trading leveraged FX and metals can result in losses exceeding your expectations, and past performance, modelled or live, does not guarantee future results.
Chat with XTRSK
Chat ready
Start a chat and the XTRSK team will be notified immediately.