At a glance
The short version
What you can backtest
ICT is a big framework, but the load-bearing concepts are testable if you define them precisely: liquidity (resting orders above old highs and below old lows), order blocks (the last opposing candle before a strong move), fair value gaps (the imbalance a fast move leaves behind), and market structure (break of structure and change of character). Nearly every ICT setup is a combination of these. We cover the full process in How to Backtest ICT Concepts.
Which ICT concepts are actually testable, and which are not
Not all of the framework survives contact with a rule. Some of it is precise enough that a stranger could apply your definition and mark the same candle you did; some of it is a way of describing a chart after the event. The first kind can be backtested. The second kind cannot be, and pretending otherwise is where most ICT backtests go wrong before the first trade is logged.
This is the split worth being honest about before you start. It is not a criticism of the framework — it is the difference between the parts that produce a number and the parts that produce a narrative.
| Concept | Testable? | What has to be pinned down first |
|---|---|---|
| Order block | Yes | Last opposing candle, impulsive move away, and a structure break. Plus one zone convention (open-to-low, or full range) applied every time. |
| Fair value gap | Yes | Minimum gap size, and whether a partial fill counts. Without a size floor every chart has hundreds. |
| Liquidity sweep | Yes | What counts as taking the level — a wick through, or a close beyond — and how long price has to reject before it is a sweep rather than a break. |
| Break of structure / CHoCH | Yes | Which swing points count, and whether a body close is required. See market structure explained. |
| Kill zones | Yes | Exact windows in New York time, and what you do with a setup that starts inside one and completes outside it. |
| Premium / discount | Yes, with care | Which swing points define the dealing range. Two traders drawing different ranges get opposite answers on the same candle. |
| Draw on liquidity | Partly | Testable only once you fix the timeframe it is read from and commit to it before the entry, not after. |
| Market maker models / narrative | No | Descriptive rather than falsifiable. Useful for reading a chart, not something a rule can be written against. |
The rule of thumb: if you cannot write it down tightly enough that someone else marks the same candle, you are not backtesting it — you are remembering it.
Why backtest ICT concepts?
On a chart, after the fact, every sweep and reversal is obvious. Live, in real time, none of it is — that's the problem a backtesting tool solves. Testing forces you to answer what actually decides profitability: how often a clean setup (say, a liquidity sweep into an order block) appears, your real win rate across a few hundred of them, and which killzone and instrument it works best in.
Log the reward:risk with every result. ICT models tend to run high reward and modest win rates, which is exactly the combination people misread — 30% winners looks like a broken model until you notice the average winner paid 3R. The risk/reward calculator turns the pair into the one number that settles it: the win rate that ratio needs to break even.
Replaying bar by bar also stops you from cheating. Scroll a static chart and you can quietly use information you wouldn't have had live — which turns a 40% setup into a 90% fantasy. Candle-by-candle replay makes that impossible.
How to backtest ICT in CRTLAB
Pick one setup, define it as a rule, then replay real price one candle at a time and mark your order blocks, FVGs and structure shifts as they form. Log every occurrence — including the losses — and review the sample after 50–100. That's how you turn "I understand ICT" into a measured edge.
In practice that is five steps, and the order matters. One. Write the setup down as a rule a stranger could apply, using the table above to check every term in it is actually pinned. Two. Pick one instrument and one timeframe and stay there — an ICT model that works on NAS100 in the New York session is a different claim from one that works on EURUSD in London, and mixing them produces a sample that answers neither. Three. Set the chart to a date you have not studied and step forward bar by bar, marking levels only from what is on the screen. Four. Log every occurrence that met the rule, including the ones you would rather not have taken; the skipped losses are what turn a 40% model into a 90% fantasy. Five. Read the sample as a whole, not trade by trade.
All 38 markets and the full history on each are open from the first chart, and so is every concept overlay the method uses — fair value gaps, order blocks, liquidity sweeps, market structure shifts, killzone shading. None of it is a paid add-on. What you spend is candles: 10,000 free sample first, or unlimited at $14.99 a month.
Open an ICT model already set up on the chart
Most ICT strategies are one of a handful of shapes, and each shape has a market and a window where it makes a sensible first test. Every link in the table opens a chart already on that market and timeframe, with the overlays the model reads switched on, so the first thing you do is pick a date and press play rather than spend ten minutes configuring.
Nothing about the model is filled in for you. The overlays mark where the candles could qualify; whether a setup meets your rule is still your call, taken before the next candle prints. Treat each row as one test. A sample that mixes two rows answers neither question.
| ICT model | A sensible first test | Pin this down first | Chart |
|---|---|---|---|
| New York kill zone sweep into a fair value gap | NAS100, 5-minute, 09:30 to 11:00 New York time | Which liquidity counts (previous day, Asia or London high and low), and whether the FVG has to form on the displacement that followed the sweep. | NAS100 · 5m |
| London order block retest | Gold, 15-minute, London session | Open-to-low or full-range zone, and whether a structure break after the block is required before it is tradeable. | Gold · 15m |
| Opening range break and retest | US30, 5-minute, New York open | Range length (first 15 or 30 minutes), and whether the retest candle has to close back inside the range. | US30 · 5m |
| Equal highs or lows raid | GBPUSD, 15-minute, London into New York | How close two highs have to be to count as equal, in pips, fixed before the first trade. | GBPUSD · 15m |
| New week opening gap as the draw | EURUSD, 1-hour | Whether the target is the edge of the gap or its midpoint, and how many sessions price gets to reach it. | EURUSD · 1h |
The markets and windows are starting points, not recommendations. Test the one you actually trade.
What to log, and what the numbers should tell you
ICT models tend to produce low win rates and high reward:risk, which is the combination people misread most often. A 38% win rate looks broken until you notice the average winner paid 3.2R. Log enough that the pair can be read together, and the honest verdict falls out of the sample rather than out of how the last five trades felt.
Six fields carry almost all of the information: the date and session, the instrument, which concept triggered the entry, the planned reward:risk, the realised R, and one line on whether the setup was clean or a stretch. That last field is the one that pays for itself — sort by it after a hundred trades and the difference between your A-setups and the ones you talked yourself into is usually larger than the difference between your instruments.
Then check the sample is worth acting on. Thirty ICT trades constrains almost nothing, and a high-R model needs a larger sample than a high-win-rate one because its results swing further. The backtest sample size calculator gives you the number for your own win rate and R, and the risk of ruin calculator answers the question that comes immediately after it: at that win rate, what position size survives the losing streaks a low-win-rate model guarantees?
Why hindsight is the specific problem with ICT
Every framework suffers from hindsight bias in review. ICT suffers from it more than most, for a structural reason: its concepts are identified by what happened next. An order block is only an order block because the move after it broke structure. A sweep is only a sweep because price rejected. A fair value gap matters because it got filled. On a completed chart, all three are visible instantly — and none of them were visible at the moment you would have had to act.
That is why scrolling back and marking setups produces win rates ten to twenty points above the truth, and why it feels like confirmation rather than error. The only fix is to take the decision before the outcome exists, which is exactly what stepping forward one candle at a time enforces. Everything else — the journal, the analytics, the sample size — is downstream of that one constraint.
Frequently asked questions
Can I backtest ICT for free?
CRTLAB is free to try. You get any of the 38 markets on the full history, with every tool the product has, for your first 10,000 candles of replay — roughly two hours of stepping through a chart, no card needed. After the sample, unlimited replay is $14.99 a month, cancel any time. Every ICT overlay the chart draws — FVGs, order blocks, sweeps, market structure, killzones — is on from the start. None of them is a paid feature.
Which ICT concepts can I backtest?
Anything you can define as a rule — liquidity sweeps, order blocks, fair value gaps, market structure shifts, opening ranges and gaps, and the setups built from them. The chart can draw each of those as an overlay so you can see where the candles could qualify, but deciding whether a setup meets your rule stays with you, and that decision is the skill you are actually training for live trading.
What's the best way to backtest ICT?
Bar-by-bar replay on real data, testing one setup at a time and logging every occurrence. Stepping through candle by candle stops hindsight from inflating your results — the single biggest thing that makes fake backtests look good.
How many ICT trades do I need before I can trust the result?
More than for most models, because ICT setups typically run a low win rate and a high reward:risk, and results with that shape swing further around their average. Fifty is enough to notice something; a few hundred is where the number starts to mean what you think it means. The backtest sample size calculator gives you the figure for your own win rate and R rather than a rule of thumb.
Can I backtest ICT on TradingView instead?
You can, with bar replay, and plenty of traders do. The differences that matter in practice are that the replay is not free, the free tier restricts the intraday history you can rewind to, and nothing is logged for you — so the journal and the win rate are a spreadsheet you maintain by hand beside the chart. CRTLAB replays all 38 markets with nothing withheld and scores the trades as you take them.
Which markets are best for backtesting ICT?
Whichever one you actually trade — the model does not transfer between instruments as cleanly as it is usually taught. In practice the index futures (NAS100, US30, SP500) and the forex majors carry the most ICT-specific behaviour around session opens, and gold is the one people most often assume behaves like the majors and find does not.
Do I need to pay to backtest ICT properly?
No. There is no version of the product with more markets, more history or more overlays in it — everything opens on the first chart. The only thing you can run out of is candles, and 10,000 are free sample first. Where people do pay is depth of sample: testing a concept across trending, ranging, high- and low-volatility conditions takes tens of thousands of bars, and that is what unlimited replay at $14.99 a month is for.