NEW Use QuantLogic in Claude or ChatGPTSee how
Research
5 min readNovember 19, 2026FrançaisEspañol

What AI can and can't find in trading

AI is great at finding patterns, including ones that aren't there. Its real value is the opposite of the hype: building an idea on some years, testing it on years it never saw, and killing the ideas that were only luck. That is hard, and most ideas fail. One of ours held.

MethodAI · NQAll researchPDF

The question

An AI can try thousands of trading ideas in minutes. On past data it will always find some that look great. Most of them are luck: they fit the noise of the months they were tested on. The hype is an AI that finds those winners and sells them as picks.

We asked a different question. If you build an idea on one set of years and then test it on years it was never shown, the way a student sits an exam they haven't seen, what is left? And can anything still hold up?

What we ran

  • NQ futures, 1-minute bars, November 2017 to November 2026, after costs.
  • Ideas and their settings were chosen only on November 2017 to October 2023. They were then read once on November 2023 to November 2026, years the choice never saw.
  • We used AI throughout: to write the tests, run them, count how many ideas we tried, and in one test to read charts with the date and price hidden.

Results

What we tested, and what was left on unseen years
TestWhat happened on unseen years
Settings chosen on one recent yearDid worse than using no settings at all
128 one-minute NQ scalping rulesNone made money after costs
An AI reading 600 NQ charts blindMostly said “go with the day”; added nothing a trader can't see
Several independent data points agreeingHeld: 124 of 142 NQ ideas did better than break-even after costs
Past results on market data, not a promise that they repeat.

The first three are what usually happens. A short history picks the wrong things. Very short trades are eaten by costs. And an AI that reads a chart sees what any trader sees.

The last one is why we are still optimistic. When several data points, each with its own market reason, agreed on a side, ideas did better on years the rule had never seen. That rule now drives the QuantLogic NQ indicator. It is a small edge, not a magic formula, and it may fade.

Why it might be so

Intraday futures are close to random. A real edge is a few ticks, and costs eat most of it. That is why raw ideas come out at about zero after costs.

Simple patterns in price and volume on NQ are watched by thousands of firms with faster machines. Anything simple and visible gets traded away.

Nine years sounds like a lot, but markets change, so there are only a handful of truly separate tests. Try enough ideas and about 1 in 20 will pass a luck check by chance alone.

Language models reason about stories, not statistics. They suggest sensible ideas, but not hidden ones. They are very good at writing and running tests, which is where their value is.

What it means when you test

Edges still exist. They hide in combinations of data that few people line up: price, order flow, options positioning, the economic calendar and news, minute by minute, over many years. Big professional firms have that. Most retail traders don't, because it is expensive and slow to build.

That is what QuantLogic builds. Then it adds the honest part: costs, years kept aside, a luck check and a count of how many ideas you tried. Not a formula, a way to test.

When you use AI on your own trading, use it to kill ideas, not to crown them. Ask how many ideas it tried, whether the result holds on years it never saw, and what the market reason is.

Limits

One market for the result that held, NQ, and one way to exit trades. The data points behind the indicator are our own research and are not listed here. Edges decay as others find them, so we keep testing. Past results, not a promise, and nothing here is a trading signal.

Simulated results on past market data. Research only, not trading signals or advice. Past results do not promise future ones; see the hypothetical performance disclosure at the bottom of the page.

Get this note as a PDF

Leave your email and the PDF downloads straight away. Easy to keep, print or pass on.

We use your email only as described in our privacy policy

Test your own ideas the same way

Analyze runs any idea on NQ, ES, gold, oil and Bitcoin with costs, both halves and a luck check, so you can see whether it beats chance before you trade it.

Past results on market data, not a forecast. No advice and no signals.

More research

Educational content, not investment advice. Futures trading involves substantial risk of loss. See the risk disclosure.