Relative value, engineered.

A systematic equity market-neutral firm built on proprietary AI and data infrastructure. We turn large, carefully engineered datasets into long/short portfolios that earn from the dispersion between stocks, not the direction of the market.

Long bookShort bookIllustrative
01Approach

Every position is paired against risk it doesn't intend to take. What remains is the view we actually hold: that some stocks will outperform others.

Our portfolios hold offsetting long and short positions across a broad universe of liquid equities. Market, sector and style exposures are neutralised in construction, so returns depend on stock selection rather than on where the index goes.

Neutrality is measured, not assumed. Exposures are recomputed continuously and compared with the targets set in research, and any gap between intended and realised risk is treated as a finding to explain.

Strategy

Equity market neutral

Method

Fully systematic

Universe

Liquid developed-market equities

Horizon

Mid-frequency

02Data & AI infrastructure

The edge is in the infrastructure.

Signals decay; well-built systems compound. We invest first in the platform that turns raw data into trustworthy features, and in the AI that learns from them, so every model is faster to build, harder to fool and fully reproducible.

Platform architectureSimplified
01 SOURCESMarket dataFundamentalsText & alt data 02 INGESTNormalisationEntity mappingAutomated QC 03 STOREPoint-in-time lakeFeature storeVersioned snapshots 04 LEARNML signal modelsNLP on textEnsembles 05 CONSTRUCTRisk modelOptimiserCost model 06 MONITORExposuresModel driftAttribution FEEDBACK TO RESEARCH LINEAGE & GOVERNANCE Code version · data snapshot · parameters · model registry · review sign-off · audit trail
A · Data platform

Point-in-time, by construction.

Every dataset passes through the same pipeline, so research never sees information that wasn't available on the day.

  • Values stored with the date they became known, including restatements
  • Survivorship-free universe with delisted and merged companies
  • Security master that maps identifiers through corporate actions
  • Automated checks for missing, stale and anomalous data
  • Corrections versioned, never overwritten
B · Machine learning

Models that learn, tested like everything else.

Machine learning finds structure that linear factors miss. It is also easy to overfit, so it gets the strictest validation we have.

  • Gradient-boosted trees, regularised linear and neural models
  • Ensembles that blend models to reduce dependence on any one
  • Purged, embargoed cross-validation to prevent leakage
  • Feature importance and stability checks across regimes
  • Every model compared against simple, transparent baselines
C · Language & unstructured data

Reading at the scale of the market.

Much of what moves stocks is written down first. Language models turn filings, transcripts and news into timestamped, structured features.

  • Tone, topic and change detection across company disclosures
  • Entity linking from text to the security master
  • Strict publication timestamps to keep features point-in-time
  • Human review of model outputs before features enter research
D · Compute & MLOps

Research at production standard.

The same engineering discipline applies from first experiment to live portfolio, so nothing changes silently between research and trading.

  • Elastic cloud compute for parallel backtests and model training
  • Experiment tracking and a model registry for every run
  • Automated tests on research and production code
  • Versioned deployments with rollback
  • Live drift monitoring against each model's research profile
03Research process

A process designed to say no to most ideas.

I

Hypothesis

Each idea is written down, with its economic rationale and expected behaviour, before any data is examined.

II

Data & features

Inputs are drawn from a point-in-time store, so research only ever sees what was knowable on the day.

III

Validation

Walk-forward and out-of-sample testing, realistic costs and capacity, and explicit correction for the number of ideas tried.

IV

Review

Peer review and written sign-off before any signal reaches a portfolio. Rejections are recorded with their reasons.

V

Monitoring

Live behaviour is tracked against research expectations. Decay triggers review, and retired models get a written postmortem.

Many independent signal sources, no single bet.

Fundamentals

Valuation and quality measures built from point-in-time financial statements.

Price dynamics

Momentum and reversal effects across horizons, measured against sector and risk peers.

Expectations

Analyst revisions, earnings surprises and corporate events, captured as they were known.

Alternative data

Non-traditional datasets, admitted only after licensing, coverage and bias review.

Machine learning

Regularised non-linear models that combine features, held to the same out-of-sample standard as every other signal.

Cost & capacity

Market-impact models that decide whether an effect can be traded at size, and how much.

04Risk

Constraints first. Then optimisation.

Risk limits define the space portfolios are built in. They are not an afterthought applied to them.

  • Market & sectorNet dollar, beta and sector exposures held close to zero in the optimiser.
  • Style factorsExposures to common factors bounded and monitored daily.
  • LiquidityPositions sized to traded volume, with market impact priced into every rebalance.
  • DrawdownPre-set thresholds trigger exposure reduction and a documented review.
  • Model riskWritten decay and retirement criteria; changes to live models need sign-off.
  • CrowdingOverlap with common factor portfolios and short-interest concentration tracked.
05People

Built by practitioners.

Olivier Chevillon

Founder · Head of Research

More than twenty years in quantitative research and portfolio management at leading hedge funds. MBA, INSEAD; MSc, ENSAE.

VisiumMaven Securities

Hadrien Darmon

Partner

Portfolio manager with experience across multi-strategy and systematic platforms. Graduate of CentraleSupélec and Université Paris-Assas.

VeritionVisiumTriumMethod Investments
06Contact

For professional investors and research partners.

management@diptechtrading.com
Restricted · Strategy

Professional investors only.

This section is reserved for eligible professional investors and counterparties. Enter the access password provided by the DIP Tech Trading team, or request access at management@diptechtrading.com.