Video summary
Claude Tested Over 25,000 ICT Strategies (Here's What Works)
Main summary
Key takeaways
Summary
The video reports a large backtest of the ICT Silver Bullet day-trading strategy, using 16 years of one-minute futures data for NQ and ES. The presenter says the tests found that liquidity sweeps followed by short-term reversals were the most useful elements. Many other commonly taught rules—including specific trading windows and a 15-minute directional bias—did not improve results.
Instruments and scope
- NQ and ES futures were tested.
- Forex and gold were not tested.
- Other untested elements included order blocks as entry signals, intrahour macro windows, and news-day filters.
Testing methodology
The presenter says the dataset contained more than 11 million one-minute bars, spanning 16 years, and was used to simulate roughly 10 million trades.
Because ICT rules are subjective, the test converted choices such as what counts as liquidity, gap size, entry, stop, trading window, and bias into configurable parameters. The presenter estimates these choices could produce about 258 million strategy variants. Rather than test all of them, the project ran a uniform random sample of 25,000 configurations, plus three named versions:
- 1-minute textbook
- 5-minute textbook
- “As traded”, intended to reflect common online-trader practices
The backtest included commissions and additional slippage on stop-outs. If a one-minute candle touched both a stop and target, it was counted as a loss.
Results were compared with a same-risk random benchmark and adjusted for multiple testing. Strategies that survived were then tested on ES without changes. The final two years of data were held out and reportedly used only once for the final test.
Results reported
Of the 25,000 configurations:
- 9,481 made at least 100 trades.
- 6,190 were profitable after expenses.
- 2,191 passed a risk-management threshold.
- 1,614 passed the random-adjustment test.
- 420 remained profitable on ES without reconfiguration—about 1.7% of the original sample.
The 5-minute textbook version passed early filters but failed the risk stage. The 1-minute textbook version reached the ES test but failed there. The “as traded” version passed the reported test sequence.
The claimed 70–80% win rate was not typical of the tested strategies:
- Better-performing versions generally showed 34–48% win rates.
- The strongest configurations clustered around 52%.
- Only five of 25,000 configurations reached a 70% win rate; the presenter says these traded only about 20 times per year.
The presenter says the popular time windows were not uniquely advantageous. The 2 p.m. New York window and the London window performed comparatively well, while the heavily followed 10 a.m. New York window was slightly unprofitable.
A rule-by-rule test reported:
Rule Reported figure Fair-value-gap entry alone −0.33 Adding a liquidity sweep 0.69 Adding a 15-minute directional filter 0.16The subtitles do not identify the units for these figures.
On the held-out two-year period, the best-performing configuration reportedly made $455,000 on one contract, with a Sharpe ratio of 13 and all 24 months profitable. The presenter explicitly calls this an unrealistic, highly optimized result.
The “as traded” version reportedly made $251,000 in that period. The presenter cautions that the period featured strong market growth and that the results should be viewed as an upper-bound illustration, not an expected outcome.
The presenter also reports that a top configuration had 19 losing days in a row. High-frequency versions reportedly had 62–68% profitable months, but that does not establish a stable income stream.
Elements the presenter says mattered
The video’s proposed surviving framework is:
- Require a liquidity sweep.
- Trade against the side whose liquidity was just taken.
- Use a one-minute gap formed during the reversal for entry.
- Take a high number of setups, since the claimed per-trade edge is small.
- Avoid the 15-minute directional bias filter.
- Do not rely on special trading windows.
The presenter characterizes the surviving idea as similar to mean reversion: a sweep followed by a snapback toward the mean.
Risks and caveats
- The reported results are backtests, not live trading results. The presenter notes that the model assumes orders are filled when touched; in live markets, losing stops may execute while some profitable limit entries do not.
- Strong performance in the held-out period may reflect favorable market conditions, and the best result was described as optimized and unrealistic.
- The presenter says discretionary judgment would need to filter out 17.2% of losing trades to turn a textbook version into a reliable strategy, but says this cannot be demonstrated through backtesting.
- The subtitle’s trade-frequency figures are internally inconsistent: it reports about 1,906 trades per year for the “as traded” version and also says about 68 trades per day.
- Disclaimer: The presenter says the video is not financial advice and makes no guarantee of profits.
Presenter and sources
The presenter is Brendan of AI Pathways. Claude was used to help build the testing system. The strategy examined was the ICT Silver Bullet framework.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.