Skip to content
← The wizards
DeclassifiedThe Systematic Volatility Trader

David Harding

Operationalized by

Regime Classification Engine — the state machine

Integration rate

59% — computed, with the gaps below

The record

What is publicly documented.

Verified against multiple public sources. Where sources disagree, the disagreement is stated rather than resolved silently.

Born
Oxford, England, 24 August 1961
Education
Cambridge — first-class honours, natural sciences (theoretical physics), 1982
First role
Sabre Fund Management — designing trading systems
AHL
Co-founded Adam, Harding & Lueck, 1987 — the H
AHL sold
To Man Group, 1994 — now Man AHL, the group's flagship systematic business
Winton founded
February 1997 — began trading that October
Winton at launch
$1.6 million, two employees, a mews office in Kensington
Reported performance
≈ 15% net per year on average as of 2013 (Wikipedia, citing published sources)
Peak scale
Widely reported above $25bn; ≈ $10bn as of late 2023
Notable drawdown
Above 20% in Q1 2002 — held the system without intervening
Honours
Knighted · £100m+ in science philanthropy, including a major 2019 Cambridge gift

Harding read theoretical physics at Cambridge and took a first, and the framing never left him. He went to Sabre Fund Management to build trading systems, met Michael Adam and Martin Lueck, and in 1987 the three launched Adam, Harding & Lueck — a commodity trading advisor running systematic trend following at a moment when that was still an unusual thing to do in London. Man Group bought AHL in 1994, and it remains the group's flagship systematic business today.

He founded Winton in February 1997 and began trading that October out of a mews house in Kensington with $1.6 million and two employees. What separated Winton from the CTA field was not the initial strategy — it was recognisably the AHL approach — but what he did next. Unimpressed by the industry's deference to efficient-market theory, he recruited outside the usual finance pools, hiring scientists trained to find signal in noisy data, and made research spending the firm's defining commitment. The firm has been widely reported above $25 billion at peak and at roughly $10 billion as of late 2023.

The episode that best describes the method is a negative one. In the first quarter of 2002 Winton was in a drawdown above 20%, and the temptation for any systematic manager in that position is to intervene in the code. Harding did not. A systematic operation is only systematic in the quarters where being systematic is expensive, and that quarter is the evidence. He has since been knighted, and his foundation has given substantial sums to scientific research, including a major gift to Cambridge in 2019.

The framework

The idea underneath the method.

Not the trades — the reasoning that decides which trades are permitted, at what size, and when they stop.

The rules do not bend; the allocation does

The distinction the entire dossier turns on. A rule that is overridden when conditions look unusual is not a rule, and a system whose parameters are adjusted after a bad quarter is a discretionary system with extra steps. Harding's answer is to keep the rules rigid and make the allocation across them responsive: detect what state the market is in, and shift weight toward the rule sets that state historically rewards. Nothing is overridden. The mix changes.

Research is the product

Winton's structural bet was that a continuous, well-staffed research operation would out-compete a firm with a better initial idea, because edges decay and research replaces them. This is a statement about where a trading business's value actually sits — not in the current parameter set, but in the machinery that produces the next one. It also implies a discipline most operations lack: changes are proposed, tested against evidence, and promoted deliberately rather than adopted because the last month was disappointing.

Empiricism over narrative

Harding has been consistently uninterested in explanations of why a market should do something. The question is whether a pattern is present in the data with enough persistence to survive out of sample. Hiring scientists rather than traders was an expression of that: people trained to be suspicious of their own results, working on noisy data, in a field where a satisfying story is the most common way to be wrong.

Multiple horizons, blended

The flagship programme blends short-term trading with long-term trend following across more than a hundred futures markets, running several systems and several timeframes simultaneously. The blend is the diversification: not just across instruments but across the horizons on which an edge is being expressed, so that a regime hostile to one is not automatically hostile to all.

Mechanics

How it actually runs.

The operating detail, stated the way a reference describes a technique.

Universe

Over a hundred futures markets globally, long or short, with no reliance on general asset appreciation.

Signals

Fully systematic and research-derived. Statistical evidence of persistence is the admission criterion, not economic narrative.

Regime layer

Market state is classified, and allocation across strategies and horizons is adjusted to match it.

Horizons

Short-term systems blended with long-term trend following, deliberately, so no single horizon carries the programme.

Drawdown behaviour

Held through. The 2002 drawdown above 20% was ridden without intervening in the system.

Research cadence

Continuous. A large research staff exists specifically to replace decaying edges before they matter.

Divergence

Where MARS does something else.

Listed first, and at length, because a mapping that only claims similarity is a poster. Some of these are scale limits; at least one on every dossier is a deliberate refusal.

Who generates the signal

David Harding

The system does, entirely. Entry, exit and instrument selection are all outputs of research-derived rules with no human input in the loop.

MARS

The operator does. MARS never issues a signal — it governs what may be deployed once the operator has found something. This is the deepest divergence in the file and it caps the integration rate on its own.

Research apparatus

David Harding

A large permanent staff of scientists whose only job is finding and validating new edges before the current ones decay.

MARS

A scenario contract and promotion workflow that one person can run. The discipline transfers; the throughput does not, and no framework can manufacture a research department.

Breadth and horizon blending

David Harding

A hundred-plus markets across multiple timeframes simultaneously, which is what makes regime-linked reallocation meaningful — there is always somewhere to move weight to.

MARS

A retail book on a few instruments. Regime classification changes posture and branch weighting rather than reallocating across a hundred markets.

What the regime layer controls

David Harding

Capital allocation across strategy families and horizons.

MARS

Branch selection, variant permission and management posture. The regime engine informs how a trade is managed, not which of a hundred programmes receives funding.

Replication

What the rail carries over.

Each mapping names the module that performs the function, so the claim can be checked against the product rather than taken on trust.

Regime as a formal state machine

Regime Classification Engine

The direct mapping. Market state is classified into named regimes by measured inputs rather than by impression, transitions are defined, and the classification is an input to what the system permits. Harding's insight — that the same rules deserve different weight in different states — is the engine's entire reason for existing.

Rigid rules, responsive posture

Branch architecture · the variant matrix

MARS implements the rules-do-not-bend principle structurally. The four branch architectures are declared at entry and frozen; identity drift is logged as a breach even when it wins. What varies is the management posture inside the branch, selected by the seven-lens matrix from measured conditions. Rigid identity, responsive management — the same shape as rigid rules, responsive allocation.

Research before deployment

Scenario Manager · promotion workflow

Every candidate change becomes a scenario contract with stated expected behaviour, is run against evidence in a sandbox, is compared to the standard baseline over a persistence window, and is promoted deliberately with a pre-written rollback trigger. That is Winton's research discipline compressed to one operator's scale.

Holding the system through the drawdown

The gate ladder · the review cadence

The 2002 non-intervention is the behaviour MARS is engineered to make automatic. A red week adjusts posture; only a pattern across periods justifies a structural change. The gate reduces deployment mechanically so the operator is protected while the evidence accumulates — which is precisely what allows the structural question to be answered slowly.

Integration rate

Scored, with the shortfall shown.

Each dimension is judged separately and the headline is their mean — recomputed at render, so it cannot be hand-set. Every dossier in this library carries at least one dimension below 35%. Four uniformly high scores would be marketing.

Integration rate

What MARS actually reproduces

Regime Classification Engine — the state machine

59%

mean of 5 dimensions

  • Regime detection as a formal layer

    90%

    Named states, measured inputs, defined transitions, feeding what the system permits

  • Rules rigid, posture responsive

    85%

    Branch identity frozen at entry; the variant matrix moves, the branch does not

  • Research-before-deployment discipline

    80%

    Scenario contracts, persistence windows, deliberate promotion, rollback clause

  • Multi-market, multi-horizon blending

    30%

    A retail book cannot run a hundred markets across several timeframes at once

  • Fully systematic entry signals

    10%

    MARS entries stay discretionary — it governs deployment, it does not generate signal

Why it is not higher

The 10 is the honest ceiling on the whole profile. Harding's method is systematic end to end and MARS is deliberately not — it is a governance layer wrapped around a human's trade selection. Everything downstream of the signal maps closely; the signal itself does not map at all, and no amount of module coverage changes that.

David Harding built the governance function privately, because nothing off the shelf existed to buy. So did every other trader in this library. That private apparatus — the constraint stack, the sizing authority, the rule about when to stop — is the part that never gets published, and it is the part that separates a documented edge from a surviving account.

The claim on this page is not that MARS outperforms anyone. It is narrower and considerably more useful: their method is public, their apparatus was not, and this is the apparatus — at a scale one person can actually run.

Sealed · nearest neighbours

Three dossiers sit next to Harding's.

These mappings are not in the bundle and not rendered anywhere on this site. They are the closest methodological neighbours to the dossier you have just read — which is precisely why they are the ones held back.

Eleven dossiers remain sealed, each at this depth — the framework, the record, the divergences, and the integration rate with its shortfall shown. Waitlist registration unlocks all fifteen.

Join the waitlist — unlock all 15

Also declassified

The other three open dossiers.

Sources and standing notice

David Harding has no affiliation with Aura Logic Systems or the Montex AlphaRail System, and nothing on this page constitutes an endorsement. This dossier summarises publicly documented method and publicly reported career facts, in the way a reference work describes a technique. It contains no quotations. Performance figures are as reported by the sources listed below, are historical, and are not audited by us; past performance of any trader, fund or method does not indicate future results. Nothing here is investment advice.

The doctrine, made executable.

Eleven governed modules that turn documented method into arithmetic you can actually run.