FOLIO 02 — STRATEGY

The lifecycleof an idea

Every strategy we run travels the same road: a written hypothesis, an adversarial validation, an explicit promotion decision, contained execution, and reporting that traces to fills. Most candidates do not finish the journey.

No. 01The pipeline, chapter by chapter

Contents

01Research
02Validation
03Promotion
04Execution
05Reporting

CH. 01 — RESEARCH

The claim comes first.

Every strategy begins as a sentence, not a script. A falsifiable claim about market structure — written down, dated and filed before a single parameter is fit. The claim names its instruments, the regime it expects to work in, and the reason the edge should exist at all.

The reason matters more than the signal. An anomaly with no structural cause is usually a data artefact wearing a costume, so research starts with the mechanism: who is on the other side of the trade, why they are systematically willing to pay, and what would make them stop. If the third question has no answer, the idea is filed, not fitted.

Data hygiene is enforced before any idea touches it. Point-in-time series, venue-native funding and swap rates, spreads as they were quoted rather than as they are remembered. A backtest inherits every sin of its data. We prefer to inherit fewer.

CH. 02 — VALIDATION

Judged where it has never been.

Validation is adversarial by design. The candidate is fitted on one span of history and judged on the span it has never seen — twelve times over, as the windows roll forward.^† A strategy that performs only where it was fitted has memorised, not learned.

Then its trade history is taken apart and reassembled. One thousand Monte Carlo resamplings reorder the trades, redraw the equity curve and put a distribution around every statistic we care about. The question is never what the drawdown was. The question is what it could plausibly have been.

Costs are modelled at their worst, not their average. Spread at the hour the strategy actually trades, slippage scaled to size, funding and swap charged in full. CFD economics are unforgiving of optimism, and the simulation is not allowed any.

CH. 03 — PROMOTION

A decision, not a drift.

Promotion is an explicit act. A candidate that survives validation is put before fixed gates — deflated Sharpe, walk-forward efficiency, drawdown inside mandate, capacity above target — and it either clears them or it does not. Committee enthusiasm is not a substitute for a cleared gate.

What is promoted is more than code. Authority, a risk budget and a rollback policy travel together as one artefact. Before its first live order exists, the strategy knows how much it may hold, how much it may lose, and precisely what happens when either boundary is touched.

Size is earned in steps. A new promotion trades at a fraction of its budget until live fills confirm the simulated costs. Divergence between the two is treated as a finding, not an inconvenience.

CH. 04 — EXECUTION

Limits that route the orders.

In production, boundaries are not advice. Orders carry their limits with them across MT5, cTrader and the digital-asset venues, and the machinery that routes them is the machinery that enforces them. A strategy at its boundary is cut, not consulted.

Exposure reconciles continuously. Every position on every venue maps to a strategy, a mandate and a risk budget; an unmapped position is an incident, not a rounding error. Margin, leverage and venue concentration are watched at the account level, which is where CFD risk actually lives.

Failure is rehearsed. Disconnections, rejected orders, stale prices and funding spikes each have a pre-written response, because the worst time to design a procedure is during the event that requires it.

CH. 05 — REPORTING

One ledger, no adaptation.

Reporting is the pipeline’s memory. NAV, drawdown, fees and attribution trace back to individual fills, and the record the allocator reads is the record we read. One ledger, with no version adapted for the audience.

Attribution is kept honest by construction. Returns decompose to strategy, venue and instrument; costs are stated beside the gross rather than folded quietly into it. In the simulated record the longest drawdown ran 4.1 months,^† and the report of it names where, by how much, and whether the containment held.

The ledger closes the loop. Live fills become the next quarter’s research data, and the pipeline begins again with better evidence than it had before.

12 walk-forward windows.
1,000 Monte Carlo resamplings.
One promotion decision.^†

Correlation to SPX 0.11.^†
Low correlation is tested, not assumed.

6 venues. 0 unmapped positions.^†
Reconciliation is continuous, not end-of-day.

Gate thresholds are printed in full under No. 03.

No. 02Validation in depth

Figure 1a. Walk-forward windows — fitted in sample, judged out of sample.12 WINDOWS · IS / OOS

Each window fits on one span of history and is judged only on the untouched span that follows. The fit is then discarded; the judgement is kept. Twelve verdicts in sequence are harder to flatter than one.^†

Figure 1b. Monte Carlo resampling — the trade history, reordered 1,000 times.1,000 PATHS · 24 SHOWN

Resampling reorders the trade history a thousand times and redraws the equity curve each time. The result is a distribution around every statistic we quote — drawdown included — rather than a single flattering line.^†

No. 03What earns promotion

Deflated Sharpe ratioThe observed Sharpe, discounted for every trial burned in finding it.

DEFLATED SHARPE > 0.95^†PASSED

Walk-forward efficiencyOut-of-sample edge as a fraction of the in-sample edge that promised it.

WALK-FORWARD EFFICIENCY > 0.6^†PASSED

Drawdown inside mandateThe simulated worst passage must fit inside the mandate, with room.

MAX DD WITHIN MANDATE^†PASSED

Capacity above targetModelled capacity at deployed size, market impact included.

CAPACITY ≥ 3× TARGET^†PASSED

Most research dies here. That is the point.

No. 04Containment

We do not ask a drawdown to be small. We decide in advance where it ends.

Containment is the part of the strategy that never appears in a backtest. Each promoted model carries a risk budget it cannot exceed, a drawdown boundary at which it is retired without appeal, and a rollback path rehearsed before its first live order. The limits are enforced by the same machinery that routes the orders, so a boundary is never a matter of opinion at the moment it matters most.

Continued on Folio 03 — Technology

Notes & methodology

Validation methodology: 12 rolling walk-forward windows, fitted in sample and evaluated only out of sample; 1,000-path Monte Carlo resampling of trade sequences; parameter-stability sweeps across a neighbourhood of the deployed configuration. Figures 1a and 1b are schematic illustrations of this process, not records of any single strategy.
Correlation to SPX (0.11), longest drawdown (4.1 months) and related figures are illustrative of the strategy class, derived from simulated performance net of estimated fees, funding, slippage and market impact. Simulated results are hypothetical and do not represent live trading. Venue and reconciliation counts reflect the current production configuration.
Deflated Sharpe ratio follows Bailey and López de Prado (2014), discounting the observed Sharpe for the number of trials consumed in finding it. Walk-forward efficiency is out-of-sample performance divided by in-sample performance. Capacity is modelled from venue depth and estimated impact at target size. Gate thresholds are internal standards, not guarantees of future results. The passed statuses shown are illustrative of the promotion mechanic, not the record of any single candidate.
Trading CFDs, FX and digital assets on margin involves a substantial risk of loss. Capital is at risk. Nothing on this page constitutes an offer, solicitation or investment advice; access is restricted to qualified and professional investors in eligible jurisdictions.