ICT and Smart Money content is everywhere, but very little of it is tested. We coded dozens of popular ideas and ran them on 98,000 fifteen-minute XAUUSD candles (July 2022 – September 2026), with spread and slippage charged on every trade. Here's what worked, what didn't, and what surprised us.
How we tested
- Every rule written mechanically, so there's no discretion and no hindsight.
- A look-ahead audit, so no decision uses information from the future.
- Rules settled on 2022–2024, then checked on 2025–2026.
- Neighbouring settings tested, to make sure results aren't one lucky number.
- A random-entry benchmark: 400 runs of random trades with the same risk and targets.
1. Implied-volatility "walls": didn't hold up
Expected-move levels built from options volatility are popular on social media. We tested 3,960 variations. Once price touched a wall, the day closed beyond it about 50% of the time, the same as a random line. Nearby non-wall levels reacted just as often. Only 0.1% of variations were profitable in the development period.
2. Imbalance + confirmation: held up
A model built on higher-timeframe imbalances, a visible reaction and a lower-timeframe market structure shift was profitable in more than half of all variants tested, not just the best one. That breadth is what separates a real effect from curve-fitting.
3. Tight stops: a trap
Stops at the swing, FVG candle or 1× ATR look attractive because the reward-to-risk ratio gets bigger. In practice they got hit by normal noise far more often. Typical results were half or less of the wider stop's, with 2–3× the drawdown. One tight-stop version looked brilliant, but a single +68R trade accounted for most of its profit.
4. Early partial profits: expensive
We compared 296 exit plans. Taking profit early, at 1.5–2R or half-way to target, reduced four-year returns by 20–40%. The best plans banked part of the position at a logical target and let the rest run with a trailing stop.
5. Letting trades breathe
Closing trades too early on a time limit cut off winners. Allowing more time before a forced exit improved results by about 10R over four years, with no increase in drawdown.
6. Popular filters: mixed
| Filter | Effect in testing |
|---|---|
| Daily trend bias | Improved quality of a single-timeframe version |
| London / New York killzones | Mixed; cut too many good trades |
| Premium / discount | Mostly worse |
| Higher-timeframe order blocks as zones | Worse than imbalance zones |
| Liquidity sweep entries | Worked on one timeframe, failed on another, so not used |
| Multiple higher timeframes combined | The biggest single improvement |
The finished model
The combination that survived: +91R over 146 trades, profit factor 2.71, maximum drawdown 6.3R, profitable in all five calendar years, and profitable in both the development and check periods. It beat all 400 random-entry runs.
Full details are on the performance page and in the public report.
Most trading ideas fail a fair test. The ones that pass deserve your capital.
Trade the tested model
Use the same rule set as live signals, an MT5 indicator or a fully automated MT5 EA.
Figures in this article are hypothetical backtest and simulation results. Trading involves substantial risk. This is educational content, not financial advice. See the risk disclosure.