Does the ICT Trading Strategy Work? Our Backtest Verdict
If you've searched "does the ICT trading strategy work?", here's our short answer: in our lab tests, it didn't. We ran four separate tests of Inner Circle Trader and Smart Money Concepts setups, covering fair value gaps, order blocks, liquidity sweeps, market structure shifts and session bias. Every one of them came back dead. The setups did no better than random entries with the same stops and targets. Where they made money, that came from the market's general drift, not from the setup.
That's a blunt verdict, so this post shows how we got there, what each test found and where our testing runs out. We aren't saying the people who teach ICT are dishonest, or that nobody has ever made money trading this way. We're saying that when we wrote the rules down and ran them the way we run every idea, we couldn't find an edge.
How the ICT trading strategy is taught
ICT stands for Inner Circle Trader, the name used by Michael J. Huddleston. His own site offers a free library and mentorship lectures on price action, and points readers to his YouTube channel. "Smart Money Concepts" (SMC) is the wider label that many other teachers use for the same vocabulary. The popular Smart Money Concepts indicator by LuxAlgo on TradingView says the terminology was "coined & popularized primarily within the forex community & by ICT."
The core idea is that big institutions leave footprints on the chart and you can trade alongside them. The building blocks usually look like this:
- Fair value gap (FVG): an imbalance left by a fast move, which price is expected to come back to.
- Order block: a zone just before a strong move, where institutions are thought to have orders resting.
- Liquidity sweep: price pokes past an obvious high or low, triggers the stops sitting there, then reverses.
- Break of structure (BOS) and change of character (CHoCH): swing highs and lows breaking in a way that's read as the trend continuing or turning.
- Session bias: the Asia, London and New York sessions are each given a role, such as London sweeping the Asia low before New York rallies.
To be fair to LuxAlgo, its own page is upfront here. It says these concepts don't guarantee you're trading alongside institutional liquidity and that "there is no supporting data regarding the validity of these teachings." That's the gap we tried to fill.
Is there real research behind any of it?
Some of the ideas touch real findings. A Federal Reserve Bank of New York staff report by C. L. Osler found that in currency markets, stop-loss orders cluster at certain levels and prices move unusually fast once those levels are hit. The effects were significant over hours, not days. So the idea that stops get run isn't made up. But that paper doesn't show you can spot a sweep on a chart and profit from the reversal after costs.
Time of day can matter too. Gao, Han, Li and Zhou (Journal of Financial Economics, 2018) found that the S&P 500 ETF's first half hour of trading helped predict its last half hour, using data from 1993 to 2013. That's a different rule from ICT session bias, though. Neither paper tests ICT setups, and a real effect in the background doesn't mean a chart pattern captures it.
How we tested it
Synergy Trading runs a private research lab that tests popular strategies with the same rules for every idea. About 290 ideas have been through it so far, and only 2 are clean survivors. Every test gets:
Realistic trading costs. A random-entry control: the same trade taken at random times, so it shows what luck alone produces. A look-ahead check, so the rules can't peek at information they wouldn't have had at the time. And a correction for testing many variants, because if you try 40 versions of anything, one will look good by chance.
A few terms you'll see below. A t-statistic tells us how sure we are the average trade isn't really zero; above about 2 is the usual bar. A p value is roughly the chance that luck alone would produce a result this good, so smaller is better. Profit factor is total money won divided by total money lost; below 1.0 means the strategy lost overall. R means multiples of the amount risked, so +0.055R is an average gain of 5.5% of the risk per trade.
What we found, test by test
Test 1: a "Jadecap ICT" model on the Nasdaq 100. Previous day's high or low as the target, power of three, entries on a fair value gap, market structure shift or order block, an ATR stop and a 9:30 to 11:30 window. In sample it looked borderline, with a profit factor of 1.04. On blind out of sample data it took 3,313 trades with a profit factor of 0.62 and a 41.9% win rate. A summary on other instruments showed a profit factor of 6.9, which is far too good to be real and a textbook sign of overfitting. We switched it off on 15 July 2026.
Test 2: Smart Money Concepts across timeframes. We coded swing points, BOS, CHoCH, the first retest of a fair value gap, order block retests and sweep and reclaim entries. Each trade had a structural stop, a 2R target, a 30 bar time limit and costs of 2 basis points per side. The control used random bars, the same direction and the same stop size.
On the top 10 megacap stocks across 15 minute, 1 hour, 4 hour and daily charts, 0 of 40 combinations passed. On 15 minute charts every setup did worse than random entries. The best was 1 hour BOS/CHoCH longs at p 0.07, and that one was negative in sample.
We then checked fresh holdout sets. Only one thing looked good: daily fair value gap longs on ETFs, at +0.055R against +0.002R for random entries, p 0.005. It didn't repeat. It failed on S&P 500 stocks on daily charts (p 0.36), on the top 10 megacaps (p 0.26), and on 4 hour charts for both ETFs and S&P stocks. One pass that won't show up anywhere else is what chance looks like. Overall, longs did about as well as random longs, because they were riding the market's upward drift. Shorts lost everywhere.
Test 3: Asia, London and New York session bias. This is the viral setup where London sweeps the Asia low and you go long in New York. We used NQ 5 minute data (489 days) and ES and NQ 15 minute data (360 days). The headline scenario was negative on all three data sets, with only 10 to 18 trades each. A broader "low sweep, go long" version made small gains: +11 points on NQ 5 minute (t 0.66), +2 points on ES (t 0.53) and +18.6 points on NQ 15 minute (t 0.95). None got anywhere near 2. The fade version lost everywhere. The best of about 40 combinations hit t 1.47, p 0.08, before correcting for 40 tries. That's noise.
Test 4: ICT fair value gap plus golden zone pullback. Tested on gold 15 minute charts, then rerun on SPY, ES and NQ 15 minute charts. Dead on all of them.
What this test can't tell you
Our tests have limits, and you should know them.
The data windows are finite. The session tests covered 489 days of NQ 5 minute data and 360 days of ES and NQ 15 minute data, and the headline session scenario only produced 10 to 18 trades per set. That's too few to prove much either way. The markets were US stocks, ETFs, ES, NQ, the Nasdaq 100 and gold. We didn't test forex, where much of this vocabulary started.
Costs were modelled at 2 basis points per side in the timeframe test. Your real costs could be higher or lower. And ICT is taught with a lot of discretion. We tested rules we could write down; we can't test a trader's judgment. These are historical tests, not live trading results, and past results don't guarantee future ones.
Frequently asked questions
Are fair value gaps a real edge?
In our tests, fair value gaps didn't show a reliable edge. Daily fair value gap longs on ETFs looked good once, at +0.055R against +0.002R for random entries, but the result failed on S&P 500 stocks, megacap stocks and 4 hour charts. A result that won't repeat is most likely chance.
Why do ICT setups look so good on YouTube?
Charts reviewed after the fact make any pattern look clean, because you can see which setups worked. When we wrote the rules down and tested them blind, one model fell from a profit factor of 1.04 in sample to 0.62 out of sample. Hindsight and overfitting can make a strategy look far better than it is.
Does the London sweep of the Asia low lead to a New York rally?
Not in our data. The headline session setup lost money on NQ 5 minute, ES 15 minute and NQ 15 minute data. A broader version made small gains, but none of the t statistics reached 1, well below the usual bar of about 2, so we treat them as noise.
Is smart money concepts trading better for longs or shorts?
Neither showed an edge in our tests. Long setups did about as well as random long entries, because they were riding the market's general upward drift. Short setups lost money everywhere we tested them. Whatever profit appeared came from the market rising, not from the setup itself.
Want a strategy tested?
If there's a strategy you'd like us to put through the same tests, email hello@synergytrading.net with the subject "Verdict request". It's a waitlist: there's no price, no promised date, and nothing is sold yet. Synergy Trading isn't a registered broker-dealer or investment adviser, and nothing on this blog is a recommendation to buy or sell.
Educational research only, not investment advice. Past results do not guarantee future results.
Sources
- The Inner Circle Trader, official site of Michael J. Huddleston
- LuxAlgo, Smart Money Concepts (SMC) indicator on TradingView
- Osler, C. L., Stop-Loss Orders and Price Cascades in Currency Markets, Federal Reserve Bank of New York Staff Report No. 150 (2002)
- Gao, Han, Li and Zhou, Market intraday momentum, Journal of Financial Economics 129(2), 2018
Want your strategy tested?
We run the same test on any rule set: realistic costs, a random-entry control and a look-ahead check. Email us the rules to join the waitlist. Nothing is for sale yet.
Request a verdict
