The short answer is yes, but only after stripping out most of the mythology.
That is a less satisfying answer than either camp wants. The ICT crowd wants a clean confirmation that liquidity sweeps, fair value gaps, and SMT divergence are universal truths. The anti-ICT crowd wants the whole thing dismissed as chart cosplay. The evidence does not support either extreme.
What I found is narrower and more useful.
Crypto markets do show the raw ingredients that make ICT-style ideas plausible: pronounced time-of-day structure, repeatable liquidity and order-flow effects, short-horizon mean reversion, and cross-asset relationships strong enough for non-confirmation logic to at least make sense.[1][2][3][4][5] What I did not find is serious evidence that the retail, educator version of ICT and SMT, copied directly from forex and narrated with enough conviction, is already a proven edge in crypto.
That distinction matters. A market can contain the mechanics that inspired a trading framework without validating the framework in its retail form. I care about that distinction because I am already building a Rust statistical arbitrage portfolio bot, and the difference between a plausible story and a tradable system gets painfully concrete once you have to make the runtime, the research process, and the live evidence line up.
What ICT and SMT are really claiming#
Stripped to basics, the claim is this.
Price moves toward obvious pools of liquidity, raids them, leaves a footprint, then reverses or continues in a way that can be structured. In ICT language, that footprint might be a liquidity sweep, a fair value gap, an order block, or a market structure shift. SMT adds a second chart and says: if two assets usually move together, and one pushes to a new extreme while the other refuses to confirm it, the move is suspect.
That is not a ridiculous claim. It is just a badly specified one most of the time.
The usual problem with ICT discussion is that it mixes three very different things together as if they were the same:
- a story about how markets move,
- a chart vocabulary for describing that story,
- and an actual tradable rule set.
The story can be directionally right while the vocabulary is loose and the rule set still loses money.
That is exactly why this question has to be answered empirically.
What the published research actually supports#
The strongest evidence I found does not come from papers saying "ICT works." It comes from market microstructure papers describing the underlying behaviour in more sober language.
A large cross-exchange study found that crypto trading activity, volatility, and illiquidity are not evenly spread through the day. They peak around 16:00 to 17:00 UTC, despite crypto trading around the clock.[1] A 2026 paper on Binance perpetuals found predictable bursts in activity and return behaviour around quarter-hour boundaries, with opening returns forecastable out of sample and boundary order imbalance carrying information further out.[2]
That does not prove ICT killzones. It does support the milder claim that timing matters, and that crypto is not as temporally uniform as people pretend.[1][2]
A separate 2026 paper on short-horizon mean reversion found that, at 15-minute horizons, directional mean reversion is much stronger in crypto than in US equities under one matched out-of-sample protocol. But it also found the gross edge peaked around 1.3 basis points per trade against about 5 basis points of round-trip cost. The signal was real enough to measure and too small to clear ordinary capture costs.[3]
That is a recurring pattern in this literature. There is often something there, but the distance between statistical predictability and a tradable retail edge is large.
The same caution shows up in a 2026 Frontiers paper on crypto microstructure. It found leakage-controlled predictive content in minute-level features like spread proxies, realised volatility, momentum, and order-flow imbalance, but concluded that no strategy based on those 5-minute forecasts survived realistic exchange fees and slippage.[4] Another 2026 paper in the Journal of Financial Markets found that order flow helps explain and predict cryptocurrency returns, especially once you separate its transitory and more permanent components.[5]
That is much closer to the kind of evidence I would trust. It says crypto does exhibit repeatable behaviour around liquidity, flow, timing, and reversion. It does not say that drawing a fair value gap on a chart automatically turns that behaviour into an edge.[1][2][3][4][5]
The main thing the ICT crowd gets right#
The part I would defend is the basic intuition that obvious levels matter because people cluster around obvious levels.
That is not mystical. It is market structure.
If a lot of traders anchor risk above the same high or below the same low, then moves through those levels can trigger stops, forced exits, and reactive momentum. The exact same phenomenon can be described as a liquidity sweep, a stop cluster, a short squeeze, or a temporary dislocation in inventory and flow. The name is less important than the fact that clustered behaviour exists.
The published evidence is stronger for that broad interpretation than for any one retail pattern package.[1][4][5]
So if someone says "price moves into crowded areas and sometimes snaps back," I think that is plausible. If they say "this means my exact three-candle FVG entry with a London killzone filter wins at 82%," I stop listening until they show the data.
The main thing the ICT crowd gets wrong#
They often act as if seeing the pattern is the same thing as proving the trade.
It is not.
The biggest trap in this family of strategies is that they are easy to explain after the fact. Once you know where price turned, it is trivial to point to the sweep, the non-confirmation, the imbalance, and the structure break. The chart starts telling a beautifully coherent story. The problem is that a coherent story is not the same as a rule that survives fees, slippage, sample changes, and live execution.
That is why I care more about what survives an untouched holdout than about how elegant the chart explanation looks.
In my own trading project, an earlier ICT-style setup looked strong on the research window and then sagged badly on a proper untouched holdout. The later version improved the in-sample picture, but the holdout was already spent, so the documentation now says very plainly that testnet forward trading is the only clean evidence left. That is how these strategies should be talked about: not as revealed truth, but as hypotheses with a provenance trail. It is the same instinct behind What a Go Scaffold Should Look Like: the interesting part is never the slogan on the box, it is whether the thing survives contact with the failure cases.
What my own local checks showed about SMT#
I also ran a read-only check on my local history database, just to answer the basic prerequisite question: are the usual crypto SMT pairs even correlated enough for the idea to make sense?
For BTC and ETH, yes. On my local Binance/Bybit-style history set, the return correlation stayed around 0.81 to 0.82 across 5-minute, 15-minute, 1-hour, 4-hour, and daily horizons. The 90-bar rolling median correlation stayed roughly in the mid-0.8s as well. So BTC/ETH is absolutely close enough for cross-asset non-confirmation logic to be a reasonable thing to test.
Then I ran one deliberately simple exploratory check on 15-minute data: when one asset made a fresh rolling high or low and the other failed to confirm, did the expected relative move follow?
The answer was messy.
BTC downside non-confirmations had some positive follow-through. Other cases were flat, asymmetric, or contradictory. Some had slightly positive medians with negative means. Others had hit rates barely above coin-flip. In other words, naive SMT by itself did not come out looking like a clean standalone edge.
That is not a failure of the idea. It is just a reminder that SMT is more likely to be a filter than an entire strategy.
This is where a lot of retail discussion goes off the rails. SMT is usually presented as if the divergence itself is the trade. I think the more serious interpretation is narrower: SMT tells you something about relative strength, leadership, or non-confirmation between correlated markets. That may help you choose a direction or an instrument. It does not automatically hand you a complete, fee-aware trade plan.
Where ICT and SMT probably can work in crypto#
If I had to defend the strongest version of the case, it would look like this.
First, treat ICT concepts as candidate event labels, not as doctrine. A liquidity sweep is just one way of saying price pushed through an obvious prior extreme and then closed back through it. A fair value gap is just one way of labelling a three-candle imbalance. Those are testable definitions.
Second, treat SMT as a cross-asset filter. If BTC and ETH are normally correlated and one pushes to a new extreme without confirmation from the other, that can help identify relative weakness or strength. But I would want it stacked with something else, not traded naked.
Third, use crypto-native context instead of pretending crypto is forex with weekend candles. Funding, liquidation maps, venue fragmentation, and order-flow state are all more native to crypto than the dramatic language around "smart money". The external literature is much stronger on those variables than on retail pattern names.[2][4][5]
Fourth, keep turnover under control. A lot of minute-level predictability in crypto is real and still too weak to survive costs. If an ICT or SMT expression works, I would expect it to work as a selective, lower-frequency filter layered onto a broader process, not as a machine that trades every tiny structural disagreement.[3][4]
That is the serious path. Not rejecting the concepts outright, but forcing them to pass through a proper quant filter before they are trusted.
Where they probably do not work#
I would not trust the raw, retail version of the framework.
By that I mean:
- fixed session dogma imported from forex and pasted onto a 24/7 market,
- unsourced claims about 80% or 90% win rates,
- chart examples selected after the move,
- no realistic fee model,
- no untouched holdout,
- no separation between descriptive language and trading rule.
That version of ICT can always explain the last move. I have not seen credible evidence that it can always pay for the next one.
The uncomfortable but useful conclusion#
So, can ICT and SMT win in crypto?
Yes, in principle. Crypto clearly contains the ingredients: clustered liquidity, time-of-day structure, short-horizon mean reversion, periodic algorithmic behaviour, and enough cross-asset correlation for relative-strength logic to make sense.[1][2][3][4][5]
But the edge, if there is one, is probably not living where most of the marketing says it is.
I would look for it here instead:
- quantified sweep definitions,
- SMT as a filter rather than a trigger,
- order-flow and liquidity-state context,
- realistic cost assumptions,
- strict out-of-sample and forward validation.
That is a less glamorous answer than "yes, smart money leaves footprints everywhere." It is also the only version I trust.
A framework does not become true because it is visually compelling. It becomes useful when it survives contact with data, fees, and a market that does not care about the story we tell over the chart.
Sources#
[1] Brauneis, Mestel, Theissen, The crypto world trades at tea time: intraday evidence from centralized exchanges across the globe. https://doi.org/10.1007/s11156-024-01304-1
[2] Kim and Hansen, The Quarter-Hour Effect: Periodic Algorithmic Trading and Return Predictability in Cryptocurrency Futures. https://arxiv.org/html/2607.09426v2
[3] Short-horizon mean reversion in cryptocurrency markets: a matched cross-market measurement. https://arxiv.org/abs/2608.21888
[4] Microstructure alpha: hierarchical learning and cross-asset transfer in cryptocurrency markets. https://www.frontiersin.org/journals/blockchain/articles/10.3389/fbloc.2026.1811716/full
[5] Anastasopoulos, Gradojevic, Liu, Maynard, Tsiakas, Order flow and cryptocurrency returns. https://www.sciencedirect.com/science/article/pii/S1386418126000029
Local notes#
The post also draws on my own read-only checks run on 2026-09-06 against a local historical database and project notes:
- BTC/ETH return-correlation check across 5m, 15m, 1h, 4h, and 1d horizons
- exploratory 15m BTC/ETH non-confirmation test for simple SMT-style follow-through
- local strategy notes from the project documenting prior ICT-style holdout failure and later testnet-only status