Quantitative Systems · Execution Infrastructure

Institutional-grade quantitative systems and AI infrastructure.

We build measurement-first trading infrastructure: systematic multi-market portfolios, algorithmic execution against live broker venues, and the instrumentation that tells you whether an edge is real before capital is committed.

0.49

Sharpe on a sealed 13-year holdout · PSR 0.9594

0.385 bps

Measured round-turn execution cost, XAU

Instruments across six sectors, cost-screened

906

Automated tests gating every change

About

Forge Trade House is a quantitative research and execution firm. Our work rests on a single conviction: in retail-accessible markets, most apparent alpha is measurement error, and the hard engineering problem is falsification, not prediction.

Every design is pre-registered with an expected outcome range before code is written. Results that land outside that range trigger a defect hunt before they are believed — favourable surprises most of all. The discipline has retired the majority of our own ideas, including a trailing-stop momentum rule that looked profitable at +215 $/trade until live fill measurement collapsed the edge to +5 $/trade with zero inside the confidence interval.

What survives that process is small and durable. Our validated core is a daily-horizon, volatility-targeted multi-market momentum portfolio, positive in every five-year sub-period across 26 years and clearing the statistical-power threshold on untouched data. Around it sits an execution stack measured at tick resolution against a live broker venue.

Core capabilities

Algorithmic Execution

Pull-architecture bridge between signal engine and MetaTrader 5, with layered risk gates — kill switch, drawdown and daily-loss ceilings, consecutive-loss locks, session and news windows — enforced in code, not convention.

Node · MQL5 · HTTP

Order Book & Microstructure Analytics

Depth recording with authenticity diagnostics. We test a venue's book against a randomised control before trusting it — a discipline that has caught synthetic volume ladders presented as real liquidity.

Depth · Delta · Spread

Systematic Multi-Market Portfolios

Daily-horizon momentum across metals, energy, agriculture, equity indices, rates and FX. Four independent alpha families combined by z-score with a 15% annualised volatility target and sector-balanced weighting.

26-year validation

Execution Measurement

Every fill decomposed into signal staleness, half-spread and execution error with a strict adverse-is-positive sign convention. Trailing-stop fill gaps measured at tick resolution against the level that triggered them.

Signed · Tick-level

AI Research Integration

Claude-driven authorship and adversarial review of measurement code. Our agents have caught look-ahead bias that inflated a measured edge sixfold and a sign error that overstated execution cost by 3.2× — errors they had made themselves.

Anthropic Claude

Custom Financial Tools

Bespoke instrumentation for desks that need it: spread and depth samplers, execution ledgers, statistical-power calculators, pre-registration tooling and hotkey execution surfaces for discretionary overlay.

Python · Node · MQL5

How we qualify an edge

A design must survive all four gates in order. Most do not reach the third.

GATE 01

Pre-registration

Expected win rate and profit factor written down before any code exists. Results outside the band are treated as suspect, not as discovery.

GATE 02

Cost wall

Round-turn cost computed before the signal. If raw edge sits below it, the design stops there — no parameter sweep closes that gap.

GATE 03

Statistical power

How many observations would distinguish this effect? When the answer exceeds the data available, the verdict is unknowable, not no edge.

GATE 04

Sealed holdout

Selection happens on development data only. The holdout opens once, is recorded, and is never reused for a second candidate.

04 · Contact contact@forgetradehouse.com

For research collaboration, infrastructure engagements and institutional enquiries.