High-frequency trading research & systems engineering · for trading teams
Price prediction and the HFT system that trades on it.
Alpha, strategy, risk and production: each on its own, or all four for one market and one horizon. We predict where the price will move next and turn that prediction into an HFT system.
No ready-made signals. No bots. No promised returns.
Each part can be commissioned alone. Together they form one system. Price prediction is the root.
Each part can be commissioned alone. Together they form one system. Price prediction is the root.
Four ways in · come with one question
Four services for HFT teams. One question, or the whole line.
You run a strategy and want more edge in it.
A short-horizon forecast of the next mid-price move for your market and horizon, added to your rules as one term and tested for whether it adds information.
Start here 02 · Strategy designYou have a forecast and no trading rules yet.
When to act, in which direction, at what size, and when to do nothing: rules built around your forecast and tested with fees, queue position and fills.
Start here 03 · Risk controlsYour strategy works, but the risk layer was added after it.
Limits, drawdown, markout and stop conditions redesigned with the strategy, not around it.
Start here 04 · Production systemsSomething in production is limiting research or execution.
The path from a market event to an order, measured, and everything around it that keeps the path running.
Start hereWhere the system starts
Price prediction is the root.
01 · Alpha
Alpha research: the price forecast is a number.
The expected move of the mid price over the next N ticks. It is not a program, not a trade and not an instruction to buy or sell. It is the input the rest of the system is built around.
A model is specific to its market and horizon. What generalises is the process that builds and validates it: order-book reconstruction, feature computation, leakage controls, embargoed walk-forward validation, out-of-sample measurement.
imbalance +0.31depth 4.6 / 2.2trade flow +0.08venue gap −0.6
imbalance +0.31 · depth 4.6 / 2.2 · flow +0.08
selected features × fitted weights for this market, this horizon
selected features × fitted weights
Whether a forecast is good enough to trade is a strategy question.
Whether a forecast is good enough to trade is a strategy question.
02 · Strategy
Strategy design: the number becomes trading logic.
When to act, in which direction, at what size, and when to do nothing. In a market-making strategy, for instance, the forecast becomes one term in the price around which quotes are placed.
From a forecast
Design the rule, then test it under real conditions.
Evaluated with fees, queue position and fill assumptions, latency, slippage, liquidity and failure scenarios.
From an existing strategy
Find what is limiting it before changing it.
Start from the rules you already run and test whether an order-book forecast adds information, before anything is rebuilt.
A better forecast does not automatically make a better strategy. A change in one layer has to make sense next to the trading rules and the risk controls around them.
03 · Risk
Risk controls: bound the damage, alongside the strategy.
Risk decides how much the strategy is allowed to do, and what happens when the market does not behave as forecast. It is designed with the trading logic, not added after it.
What the work can examine
- position and inventory limits
- drawdown
- markout on fills
- trade-level statistics
- explicit stop conditions
Scopeset by the strategy
Boundary
Not a generic risk-control product.
The risk scope is defined around the strategy and the market in question.
Not a generic risk-control product. The scope is defined around the strategy and the market in question.
04 · Production
Production systems: from a market event to an order.
Production has two paths. The hot path runs from a market event to the order and is built for minimal latency. The slow path around it carries control, oversight, computation and dashboards, and is built for correctness. We design both and keep them apart.
For a new system: draw the hot / slow boundary around the strategy first. For an existing stack: find which path is limiting research or execution, and start there.
Hot path
- 01market data in
- 02features and forecast
- 03trading decision
- 04inline risk checks
- 05order out
Minimal hops, bounded work, measured behaviour. No dashboards, historical queries or reporting on this path.
Slow path
- positions and inventory
- limits and oversight
- operator console and dashboards
- audit, replay and research data
Favours correctness, recovery and auditability over microseconds, and never changes what the hot path does.
No latency figure is promised before the venue, network, hardware, workload and measurement method are known.
No latency figure is promised before the venue, network, hardware, workload and measurement method are known.
Full system · all four together
And when you need all four: one HFT system, not four isolated projects.
All four · one system
Starting a new HFT line
Define the whole system before building any of it.
First the frame the whole system has to live within: the market, the trading capital and how it is meant to grow, the technical and legal limits of the venue and the jurisdiction, and what the line is for. The parts come after.
Improving an existing system
Start from the part that is limiting progress today.
The forecast, the trading logic, the risk layer or the production path. A full rebuild is not assumed.
What we do not sell
What HFT Architecture does not sell
Ready-made signals or bots
Every engagement is defined for a market, a horizon and a system.
Performance guarantees
No promised return, Sharpe ratio or latency figure.
Retail products or courses
The work is for professional market participants.
Our research methods
We explain the work without publishing the method behind it.
Questions before we start
Questions trading teams ask before we start
- Is alpha the strategy? #
- No. Alpha is a forecast. Strategy decides whether and how to act on it.
- Does one model work in every market? #
- Not by default. A model is fitted for one market and one horizon; another market needs its own data, connector and features.
- Can work start with a single service? #
- Yes. Start with one of the four, or define the full system.
- Do we have to rebuild an existing system? #
- No. The work can begin with the specific constraint you want examined.
- Do you guarantee profitability or a latency number? #
- No. Trading outcomes depend on markets and involve risk. Latency targets are defined only once the venue, network, hardware and workload are known.
- What should the first message say? #
- What you want to change and where it is stuck: the market, the strategy or system you run today, and the bottleneck in research or execution. Two paragraphs are enough.
Contact
Tell us which part of your HFT system to improve, or the whole line.
Describe what you run today and where it is stuck. That is enough for a useful first conversation.
opens in your mail clientGoes to our inbox as one email.
Opens in your mail client.