Does ICT Work on Crypto? Automating Smart Money 24/7

The Verdict on Whether ICT Works in Crypto

Yes — with a split. Roughly half of what you currently call ICT survives the move to crypto untouched, and the half that does not is, awkwardly, the half most traders port over first.

It helps to stop treating ICT as one method and start treating it as two stacked claims:

• A claim about liquidity and order flow. Price gravitates toward pools of resting orders, absorbs them, and leaves a readable trace of where a large participant was filled. Nothing in that sentence names an asset class. It describes how any continuous limit order book behaves when somebody who needs size has to go and find it.
• A claim about who, and when. That the large participant is an interbank dealer. That its activity clusters into the London and New York windows. That the trading day has a defined close, which is what gives your higher-timeframe candles their meaning. That is a description of the foreign exchange market, not a law of markets.

The first claim transfers to crypto essentially intact — and in one respect it arrives in better condition than it left. In forex, "there are stops resting under that low" is an inference. In crypto, leveraged positions carry a published liquidation price and exchanges run deterministic engines that close them at it, so the pool under the low stops being a belief and becomes an inventory of forced orders with known addresses.

The second claim does not transfer as-is. Crypto has no interbank tier to concentrate flow into two windows. It has no exchange-wide close, so the daily candle you are reading is a timezone convention your data provider chose. And its version of "smart money" is a rotating cast of exchange market makers, basis and funding desks, and liquidation machinery that never goes home for the weekend.

So the practical verdict: automating ICT on crypto is a defensible use of a build cycle if your Expert Advisor (EA) encodes the order-flow layer and re-derives the timing layer from crypto's own structure. It is a poor use of a build cycle if the plan is to take a forex ICT EA, change the symbol to BTCUSD, and assume the killzone filter still means something.

If you want the full structural comparison sitting underneath that split — session structure, who provides liquidity in each market, and what "the close" even means — that is the 24/7 playbook framework for running ICT across crypto and forex, and this article assumes it rather than re-deriving it.

Which ICT Concepts Transfer Cleanly to Crypto

These are the pieces you can port with no translation layer, because none of them was ever really about forex.

Liquidity pools above and below obvious levels. Equal highs, the low of an overnight range, the wick everyone can see — these are pools because a crowd of traders puts protective orders in the same obvious place, not because a bank decided so. Crypto keeps that behaviour and adds a mechanical version on top: every leveraged position has a liquidation price, and when price reaches it the exchange closes the position at market whether the trader is watching or not. Sell-side liquidity under a low is therefore not just stop orders — it is stop orders plus a queue of forced market sells that will fire on contact. If anything, the concept is more literal in crypto than in the market it was named for.

Displacement and the order block. The reasoning behind an order block is an absorption story: somebody filled size against the last push in the opposite direction, and price left that area in a hurry. That story requires a continuous order book and a participant too large to fill in one click. Both exist on any major crypto pair. Nothing in the logic depends on London.

Imbalance and fair value gaps. When price moves faster than resting limit orders can refill the book, it leaves a thin patch. Crypto does this constantly, especially in a cascade where forced closures consume several levels of depth at once. Whether those gaps then get revisited at the same rate they do in forex is its own question — worth reading up on whether fair value gaps fill the same way on crypto before you make a fill assumption load-bearing in your rules.

Market structure shifts. Higher high, higher low; the failure of that sequence; the reclaim that follows. This is geometry over a price series. It is as valid on a chart of freight rates as on EUR/USD, and it is valid on BTC/USD.

Premium and discount within a defined range. The ratio logic holds anywhere. The catch is the word defined — in forex the range often comes free from the session; in crypto you have to declare the range yourself, which lands us in the next section.

Where Crypto Breaks the Forex-Native ICT Assumptions

Every item below is a place where a copied forex EA quietly does the wrong thing rather than failing loudly.

The daily close is a convention, not an event. A forex daily candle closes because a market convention says the day ends at 5pm New York, and enough participants act on it that the boundary is real. Crypto has no exchange-wide close. Your daily candle boundary is whatever your feed picked — usually midnight UTC — and a "daily order block" drawn on a UTC chart may not exist at all on a chart cut at 5pm New York. Two fixes are workable: declare one boundary in the EA and never change it, or lean on intraday structure and use higher-timeframe context only as a bias filter. The one genuine analogue worth knowing is the regulated bitcoin futures contract, which does close and does leave weekend gaps — a real session boundary in a market that otherwise has none.

Killzones proxy something crypto does not have. London and New York killzones are shorthand for "this is when interbank flow concentrates". Crypto flow does still thicken around US and Asian working hours, scheduled macro releases, funding settlements and large options expiries — but not with the reliability that lets you hard-code an hour range and forget it. Rather than importing 08:00–11:00 New York into a crypto EA, measure where range and volume actually cluster on the specific pair and venue you trade, and express the result as a trading session filter you can re-fit — or replace the clock entirely with a volatility condition, which is what the clock was approximating in the first place.

"Smart money" is a different cast. The classic framing imagines a dealer with a mandate and an intent. In crypto the participants large enough to matter are exchange market makers running inventory, basis and funding arbitrage desks, treasuries rebalancing, and liquidation engines that have no intent whatsoever. The useful adaptation is to drop intent from your reasoning altogether. You do not need a story about who wants your stops. You need the observation that somebody large enough will take price to where the forced orders are, because that is where the fills are.

There is no single tape. A crypto pair trades on many venues at once, spot and perpetual prices differ, and the candles differ with them. An order block is only real on the venue whose candles you drew it on. If you backtest on one exchange's data and execute on another's — or analyse a spot chart and trade a perpetual — you have introduced a mismatch your test will never show you. Pick the venue you execute on and analyse that one.

The weekend is a regime, not a gap. Forex hands you a closed market and a Monday gap. Crypto hands you a thin book. Sweeps still happen, and some of them are genuine, but a rule tuned on weekday depth behaves differently in weekend depth. An EA that trades all week without knowing the difference is running two strategies and reporting one.

Can a Discretionary ICT Edge Survive a Bot-Dominated Market?

This is the question every general crypto-automation guide steps around, and it is usually asked in the wrong shape. The fear is that so much of crypto order flow is algorithmic that any readable pattern has already been competed away. Three things are worth separating.

Competition erases speed edges, not location edges. What an army of automated participants genuinely arbitrages out is latency and quoting: being first to a stale quote, capturing a spread, front-running a slow order. ICT is not in that business. It does not race anyone to a fill — it decides where to put risk and where to be wrong. You are not competing with a market maker for a queue position; you are choosing a price zone hours in advance and sizing for it.

The automated flow is not covering the footprint — it is the footprint. A liquidation engine is the least discretionary participant in any market: it fires at a published price, in size, without hesitation, with no ability to wait for a better fill. That is the exact behaviour ICT calls a liquidity grab, except in crypto it is machinery rather than intent. Automation makes the reaction more repeatable, not less — which is precisely the property a rule-based strategy needs. The mechanics of how those sweeps resolve when there is no daily close to reference are worth a separate read on how liquidity sweeps behave without a daily close.

What does get crowded is the obvious entry. When enough retail bots place a limit at the edge of every four-hour order block, the reaction at that edge gets shallower and the sweep-of-the-sweep gets more common. That is a real effect and it is not the market being unbeatable — it is one specific execution style being over-subscribed. The adaptation is ordinary: require confirmation rather than anticipation. A structure shift on a lower timeframe after the sweep, instead of a resting limit at the zone boundary, costs you some entry price and removes a large share of the traps set for the anticipating crowd.

Made concrete: suppose BTC/USD grinds sideways for several hours under a well-defined range low, then breaks it in one push, tags the resting stops and the liquidation cluster beneath, and closes back inside the range within the same hour. The last down-close candle before that reversal becomes the origin zone, price retests it a few candles later, and the move continues into the range high. An EA never "sees" any of that — it matches conditions: a new low beneath the prior swing, a close back above it within N candles, then a limit inside the origin candle's body with the stop below the sweep wick. Same trade, no narrative.

The honest limit is not the bots — it is the translation. The parts of ICT that resist automation are the parts you resolve by eye: was that displacement decisive or just noise, is the higher-timeframe bias genuinely bullish, does this sweep look engineered or accidental. An EA has no judgment; it has thresholds. Automating ICT means converting every one of those calls into an explicit, testable number — how far beyond the swing counts as a sweep, how many candles the reclaim may take, how large the displacement body must be relative to recent range. Your automated version will be a coarser instrument than your eye. That conversion, not the density of machines on the other side, is where most ICT automation projects actually fail.

What Backtest-to-Live Decay Means for an ICT EA Running 24/7

Assume you get the translation right and the historical curve looks convincing. Expect live results to be worse than the backtest, and expect the gap to be wider on crypto than it would be on a session-gated forex version of the same rules. Five reasons, none of them exotic:

• You are trading the least liquid moment in the sequence. The sweep candle is exactly where depth is thinnest and forced orders are firing. A tester fills you at the price you asked for; the exchange fills you at the price that exists. Slippage on this specific strategy is not an average-case cost, it is concentrated at the entry your whole edge depends on.
• 24/7 multiplies everything, including costs. A killzone-gated forex EA gets a handful of chances a week. The crypto version is awake all the time and will trade far more often. If the per-trade friction estimate is slightly optimistic, that error compounds much faster.
• Every threshold you tuned is a parameter. Sweep depth, reclaim window, displacement size, retest tolerance — each one you nudged until the curve improved is a chance to fit history rather than structure. Overfitting does not announce itself; it announces a beautiful equity curve.
• Regime is not a constant. A rule set fitted on a trending quarter behaves differently in a chopping one, and crypto changes regime without asking.
• Test data and execution data may not match. Back to the fragmented-tape problem: candles from one venue, fills on another.

The defences are unglamorous and they work. Reserve untouched data for an out-of-sample test instead of judging on the period you optimised. Run a forward test on live data at minimum size before real capital, because it is the only stage that prices in slippage and exchange behaviour honestly. And treat the Strategy Tester curve on your MetaTrader 5 terminal as one sample of what could have happened, not a forecast.

That last point is worth seeing rather than reading. Take a modest hypothetical edge — say a rule set that wins slightly less than half its trades at a reward-to-risk near 1.6 — and run it many times. The median path looks fine. The band around it is enormous, and any single run inside that band is a plausible "backtest" someone could show you.

Size for the band, not the median: work out your per-trade risk with a position size calculator using the worst plausible run rather than the average one, and read our risk warning before any of this touches funded capital. Historical and backtested results describe the past, and a 24/7 market gives them a lot of opportunities to stop describing the present.

See ICT Setups Forming on Live Crypto Signals

There is a cheap step between "the verdict makes sense" and "I am committing a build cycle": watch the pattern occur on a live market for a couple of weeks and see whether you can call it before it resolves.

That is what our crypto live signal feed is for. It publishes setups on major crypto pairs as they form — the sweeps, structure shifts and origin zones this article has been describing — and you can watch it without signing up. Reading a few dozen of them against the chart is the fastest way to find out whether your rule thresholds match what actually happens on the venue you plan to trade, before you have written a line of MQL.

Follow one pair to start; the BTC/USD signal page is the natural choice since it is the most liquid book and the one where structure reads cleanest. To be clear about what this is: it is a live feed for watching and acting manually, not a turnkey automated EA. If you have already decided to fully automate, you still need to choose and configure the bot itself.

Next Steps: Choosing and Building Your ICT Automation

The verdict is settled, so the remaining questions are build questions, and each has its own answer worth reading properly.

1. Decide the venue and the contract before the strategy. Spot and perpetual behave differently around the exact moments your rules trigger, and the venue determines which candles are real for you.
2. Choose the automation vehicle. Comparing what runs where is a decision in its own right — start with choosing a bot once you have decided ICT is worth automating on crypto, exchange by exchange.
3. Wire it up on the exchange you picked. The connection work — a scoped API key, the symbol mapping between your terminal and the exchange, a first verified test order — is a walkthrough of its own, and if the exchange is Binance the steps are laid out in how to run an ICT EA on Binance, step by step.
4. Fix your timing layer explicitly. Replace the inherited killzone with something you measured on your pair, or with a volatility condition, and write down which one you chose and why.
5. Budget the honest test. Out-of-sample first, forward test second, small live size third. Skipping straight to the live account is the expensive path.

FAQ

Does ICT work better on Bitcoin than on small-cap altcoins?

Generally yes, and for a structural reason rather than a mystical one. ICT logic depends on a book deep enough that a large participant cannot fill in one go, which is what produces displacement and leaves an origin zone behind. On a thin altcoin, a single order can move price several percent, so the "footprint" you are reading may be one participant's clumsiness rather than absorption. The deeper the book, the more the patterns behave the way the method assumes.

Do ICT killzones work in crypto?

Not as inherited constants — the killzone paragraph above covers why those windows lose their referent once there is no interbank tier to concentrate flow. The build question is what you do with whatever replaces them. Treat an hour window as a parameter like any other: fit it on one stretch of data, then keep it only if it still earns its place on a stretch you did not fit it on. A window that improves the curve solely on the period you tuned is a fitting artifact, not a session.

Can an EA trade ICT without discretionary judgment?

It can trade a version of it, and that version is always coarser than the one in your head — the translation problem described earlier is why. The useful move is to measure the gap instead of arguing about it: run your thresholds over a stretch of chart you have already marked up by hand, then compare the two lists of trades. The bot will take setups you would have passed on and pass on setups you would have taken, and the shape of those disagreements tells you what to fix — a threshold set too loose, or a condition you were applying by eye without ever writing it down.

Does the missing daily close actually break order blocks?

It breaks higher-timeframe order blocks, not the concept. An intraday order block on a one-hour chart is unaffected, because the hour boundary is the same everywhere. A daily or weekly order block, on the other hand, exists only relative to the day boundary your data provider chose, and a different boundary can produce a different candle and therefore a different zone. If your rules lean on daily structure, declare the boundary explicitly and keep it constant across backtest and live execution.

Is a crypto ICT bot the same product as an MT5 ICT EA?

Functionally it is the same strategy running in a different place. An EA is a program running inside a MetaTrader terminal; reaching a crypto exchange from there requires a bridge between the terminal and the exchange account, and how that bridge handles symbols, contract sizes and order types matters more than most strategy parameters. A native exchange bot skips the bridge but gives up the MetaTrader tooling. Neither is inherently better — pick the one whose execution path you can actually inspect when a trade behaves unexpectedly.

Sources & Further Reading

Want to go deeper? These independent, authoritative sources shaped this guide — each one is worth reading in full:

• CFTC advisory on commodity trading systems sold online — how to interrogate a system's claims
• Research paper on early-warning signals across crypto-perpetual liquidation cascades
• Wikipedia — Market microstructure, the study of how order books and trading rules shape price