Independent strategy research

Does ICT / Smart Money Concepts actually work? We tested it like everything else.

Q: Does ICT / Smart Money Concepts actually work?

Partly, and that is the honest answer. In our testing about 73% of ICT/SMC strategies were rejected — most fail, but the category does not collapse to zero the way grid/DCA does. A minority showed a real conditional edge on specific instruments and sessions.

Q: Is 20-plus tests a big enough sample to trust?

It is smaller than our hundreds-deep categories, and we treat it that way. We report the pattern and the honest caveat rather than pretending two dozen tests carry the weight of two hundred. The direction was consistent enough to publish, but we weight it less than trend-following.

About 73% of the ICT and Smart Money Concepts strategies we tested were rejected — one of the lower reject rates in the whole audit. That surprised us. ICT gets written off as pure narrative, yet as a category it held up better than the breakout and mean-reversion piles, and it is nowhere near the grid bots that fail 100% of the time. So no, it isn’t all smoke. Most of it fails. A minority doesn’t. And a good chunk of that minority falls apart the second you shuffle the bars.

The category doesn’t die at 100% the way grid/DCA does. A minority clears the first cut — then most of that minority performs the same on scrambled data, which means it was never reading the market. Illustrative proportions from a deliberately small sample.

20+ICT / SMC strategies tested

~73%rejected after real costs

A fewshowed a conditional edge — none we’d sell as deploy-and-forget

Smaller sample than our big categories — 20-plus ICT/SMC tests against hundreds of trend-following ones. We weight it accordingly, and we say so out loud instead of dressing two dozen tests up as a law of nature.

What ICT and SMC actually claim

Order blocks. Fair value gaps. Liquidity sweeps and market-structure shifts. The premise is that price hunts pools of stop orders, and if you can read where the big money sits you trade with it instead of getting run over by it. It is a compelling story and it draws clean charts.

We did not test the philosophy. We tested the code. The only question our pipeline asks is whether a set of rules built on those ideas makes money after real spreads, commissions and slippage, on data the author never saw. The story does not get a vote.

What we found

We ported 20-plus ICT/SMC strategies and ran them through the same process as everything else. Roughly 73% were rejected. Lower than the mean-reversion and breakout categories, and a different planet from grid and DCA, where every single bot we tested failed. For a set of ideas the internet loves to dunk on, that is a better showing than we expected going in.

Then comes the part that separates ICT from the honestly-built strategies. A good number of them passed the eye test — clean entries, sensible structure, a curve that looked real. So we shuffled the bars. Under a placebo test, feeding the same rules permuted and randomized data, a lot of those survivors kept “working.” A rule that scores the same on scrambled prices was never reading the market. It has no real edge. It found a shape in noise, and noise is everywhere.

A rule that performs the same on scrambled data was never reading the market. It was reading you.

Why most ICT strategies fail

Three failure modes did most of the damage, and they are all downstream of one thing: ICT is taught through hand-picked chart examples where the winning order block is always obvious after the candle closes.

Hindsight labeling. Turn a taught example into code and you tend to fit the label to the outcome you already know. That is textbook overfitting — the rule memorizes the training charts and has nothing left for new ones.
Look-ahead in the eye test. You mark the liquidity sweep because price already reversed. Backtests that “confirm” the setup often lean on look-ahead bias — information the trade could not have had when it was live.
Description dressed as prediction. A chunk had no edge before a single cent of cost. They do not forecast direction; they relabel what already happened in a more impressive vocabulary.

The minority that survived — the honest exception

Not all of it died, and this is where ICT earns the nuance that grid/DCA never gets. A few — well under a third of what we tested — showed a real, conditional edge. Specific instrument. Specific session. Tight risk. Under the hood these looked less like a mystical read of institutional intent and more like a disciplined structure filter bolted onto an ordinary trend or reversal trade. The filter was doing honest work. The mythology around it was optional.

Deployable? A couple, in a narrow lane, sized small and risk-capped. Not the do-everything system the courses sell you, and not “quit your job” material. That is the real difference between this category and the pure bot categories. Grid dies at 100%. ICT does not. Most of it fails. Some of it, tested honestly, survives — as a component, not a religion.

How we test

Every strategy is ported to Python and run against real costs — spreads and commissions modeled from tick data, not a guessed flat number. Futures come from Databento (13 years of CME), FX from Dukascopy with real bid/ask, stocks with liquidity-aware fills, crypto as spot and perps. A fast model does the bulk porting; the strongest model then tries to break every apparent survivor, hunting look-ahead bias and impossible fills, and the placebo/permutation pass catches rules that only work on the exact history they were drawn on. We hash the code, so an ICT script re-published under three names gets tested once, not three times. It is the same process that rejects roughly 78% of everything we test — and about 73% of the ICT/SMC entries.

Want the whole map? The research hub has the by-type breakdowns for the rest of the categories.

Research and education, not financial advice. No signals, no return promises. Independent, and not affiliated with TradingView.

Which ICT scripts survived, and who published them?

You now know the shape of the category: most fails, a minority holds up, and the placebo test is where the pretenders drop. What this page does not give you is the names — which specific published ICT/SMC scripts we tested, who wrote them, and the exact after-cost verdict on each one. That is The No List: every strategy we audited, named, with the reason it lived or died.

Get The No List →

FAQ

Does ICT / Smart Money Concepts actually work?

Partly, and that’s the honest answer. In our testing about 73% of ICT/SMC strategies were rejected — most fail, but the category does not collapse to zero the way grid/DCA does. A minority showed a real conditional edge on specific instruments and sessions.

Why does ICT backtest so well but fail your test?

Because the eye test smuggles in hindsight. Order blocks and sweeps look obvious after the candle closes, so the code fits the label to a known outcome. Add a placebo test on scrambled data and most of the “wins” keep appearing, which means the rule was reading noise, not the market.

Is any ICT strategy worth trading?

A few survived as a risk-capped structure filter on a normal trend or reversal trade — narrow instrument, tight sizing. Not a do-everything system. The value was the discipline, not the mythology bolted around it.

Is 20-plus tests a big enough sample to trust?

It’s smaller than our hundreds-deep categories, and we treat it that way. We report the pattern and the honest caveat rather than pretending two dozen tests carry the weight of two hundred. The direction was consistent enough to publish; we’d weight it less than trend-following.