Skip to content
Delta Investment Group

We look for what predicts, and we do not believe it until it is measured.

A person proposes what to measure. The data decides. Everything that follows is built on that one rule.

Read the principlesSee what we study
Delta Investment Group

The invariants the work is built on.

These are not slogans. Each one changes what the work does, and each has a page that says how.

What counts as true

How a result earns the right to be believed.

  1. 01Proof before beliefIntuition proposes. Measurement decides.
  2. 02The default is no edgeEvery positive result carries the burden of proof.
  3. 03Judge only what could have been tradedWe measure at prices someone could actually have had, after cost.
  4. 04Only what was knowable at the timeNothing borrows from the future.
  5. 05Unseen data is spent onceSearching is a screen. Confirmation happens once, on data nobody has searched.
  6. 06Every look is countedEach pass over the data spends some credibility.
  7. 07We do not run tests that cannot answerSome questions cannot be settled with the data at hand.
  8. 08Measure outcomes, not activityIdeas tested and features built are not results.

Where edge comes from

What we look for, and what we refuse to assume.

  1. 09Edge is discovered, not designedA person proposes what to measure. The data decides what predicts.
  2. 10Many small edges, combinedNo single measurement is the insight.
  3. 11Every edge has a counterpartyWe name who is on the other side, and why they are wrong.
  4. 12The market is taken outA result is what is left once the tide is removed.
  5. 13Stability counts as much as sizeA small edge that holds beats a large one that comes and goes.
  6. 14Condition on context, never discard itEvery claim carries the kind of market it was measured in.

How the practice runs

The habits that keep the work honest as it grows.

  1. 15Influence is earned and can be lostWeight follows the evidence, and goes when the evidence goes.
  2. 16What was turned down is dataWe keep what we refused, and study what failed.
  3. 17Nothing is overwrittenEvery result can be reproduced.
  4. 18Every source clears the same barA new kind of data gets no exemption.
  5. 19The vocabulary is the modelA term has one meaning, because a loose word hides a loose idea.
  6. 20Question first, delete second, simplify thirdSpeed and automation come last.

How risk is held

The limits that sit above every forecast.

  1. 21Risk sits above the forecastLimits can halt the book, and nothing in the research can override them.
  2. 22No single bet mattersMany small positions, never one large one.

The ideas the rest depend on.

01

Proof before belief

We do not call anything an edge because the argument for it is good. Every idea meets the same measurement, and until it survives that measurement it does not exist as far as capital is concerned.

Where it shows upA candidate with a good story and a weak measurement waits. The story is kept with the result.

02

The default is no edge

We start from the assumption that an idea predicts nothing. Markets are crowded, most inputs are public, and an edge that looks easy has usually been found and traded away already.

Where it shows upA strong early result is a reason to look harder for the mistake.

05

Unseen data is spent once

Data we have searched cannot also confirm. Results on it only decide which ideas deserve a real test. The real test is a single run on data nobody has looked at, applied to the whole book, meaning every idea together and how they are combined, not to a favorite.

Where it shows upA book is frozen before its test, and the verdict is recorded whichever way it falls.

10

Many small edges, combined

Each measurement predicts a little. The edge is in the combination of many, weighted by the data. A hundred views that move together are worth far less than a hundred independent ones.

Where it shows upSeveral readings of the same price move are weighed as one view, not several.

11

Every edge has a counterparty

A lasting edge comes from better information, better processing, someone who has to trade for reasons other than price, a behavior that persists, or a risk others will not carry. Every family of ideas we study states which one, and who pays.

Where it shows upEvery family on the research page names its counterparty and what would make it stop paying.

15

Influence is earned and can be lost

No idea has influence by default. An idea earns weight with proof, and the work is designed to take that weight away when the measured edge fades, without anyone having to notice.

Where it shows upBy design, a fading idea is cut first and has to prove itself again to come back.

A research practice built to improve from its own record.

Most ideas fail. The practice is built so that a failed idea costs nothing, a surviving idea has to prove itself twice, and every result, good or bad, makes the next proposal better.

The research loop

  1. 01

    Propose

    A person proposes what to measure. Candidates that fail are kept, not discarded.

  2. 02

    Measure

    Each candidate is tested on data it has not seen, using only what was knowable at the time.

  3. 03

    Run aside

    A new candidate is designed to run on paper first, where it can be watched and cannot move a decision.

  4. 04

    Earn weight

    Weight is designed to follow evidence that holds, and to leave on its own when the evidence fades.

  5. 05

    Write back

    What was learned, including what was refused, is designed to be written back into the machinery that proposed the idea.

Families of ideas, each with a stated reason to exist.

Open the research ledger

The shape of the work is a triangle.

  1. 01

    Many candidates are proposed

    From many directions, none favored.

  2. 02

    Fewer survive measurement

    Tested on data they have not seen.

  3. 03

    A few hold

    Only these are meant to be given weight, and only while the evidence holds.

Retiring an idea is not a failure of the practice. It is the practice working, and the reasons are kept. Read why ideas leave.

Depth of history, measured name by name.

2000
The longest daily history begins in
2006
The longest intraday history begins in
Daily, Hourly, 15-minute, 5-minute
Bar intervals studied

Being built

The aim is for agentic workflows to drive the research, inside the same gates as everything else. This is being built.

Read the direction

Delta Investment Group invests its own capital through three companies kept apart on purpose. Read about the group.