ICT Turtle Soup + FVG Carryover Backtest: Does It Actually Work?
We backtested ICT's claim that carrying a Fair Value Gap from a turtle soup during the 2-hour launch period to the next trading day predicts the high or low of the day. 6 months of 1-minute NQ futures data — 183,747 bars — analyzed.
The Claim
In a recent video, ICT (Inner Circle Trader) makes the following claim:
“Go back through your old price action and look at that 2-hour launch period and when there is a turtle soup short or long, find that first fair value gap right before the liquidity is taken, carry that over to the next day. If it trades up in there, you're going to get many times the high or low of the day. Don't take my word for it.”
We took that advice literally — we didn't take his word for it. Here is the video where the claim is made:
Methodology
We downloaded 183,747 one-minute bars of NQ (Nasdaq 100 E-mini) futures data from Databento, covering September 2025 through March 2026 — approximately 6 months and 166 trading days.
Definitions
Results
| Metric | Result |
|---|---|
| Turtle soups detected (launch period) | 172 |
| With qualifying FVG before sweep | 163 |
| Total signals (max 1 per direction/day) | 123 |
| FVG hit next day | 61 (50%) |
| FVG NOT hit (no trade) | 62 (50%) |
When the FVG Is Hit Next Day (61 events)
| Metric | Result |
|---|---|
| Near EXPECTED extreme (HOD/LOD) | 5/61 (8.2%) |
| Near ANY extreme | 8/61 (13.1%) |
| Price reversed after hit | 18/61 (29.5%) |
| Mean distance to expected extreme | 47.4% of day range |
| Median distance to expected extreme | 44.0% of day range |
| Mean distance to nearest extreme | 28.8% of day range |
By Direction
| Type | Events | Near Expected | Near Any Extreme |
|---|---|---|---|
| Bearish TS → expect HOD | 26 | 2 (7.7%) | 3 (11.5%) |
| Bullish TS → expect LOD | 35 | 3 (8.6%) | 5 (14.3%) |
Distribution: Distance to Nearest Extreme
For each event where the FVG was hit the next day, how close was the FVG midpoint to the actual HOD or LOD? Lower is better. A random point within the day's range would average ~25%.
vs. Random Baseline
If the FVG hit price were a random point within the day's range, the probability of landing within 15% of either extreme is approximately 30%.
Our actual “near any extreme” rate: 13.1%.
Edge over random: -16.9 percentage points. The strategy performed worse than random chance at identifying the day's extreme.
Conclusion
Based on 6 months of 1-minute NQ futures data, the turtle soup + FVG carryover strategy as described does not demonstrate a statistical edge. The FVG levels from the prior day's turtle soup consistently land in the middle of the next day's range, not near the extremes.
When the FVG is hit, the average distance to the expected extreme (HOD or LOD) is 47% of the day's range — essentially a coin flip.
Caveats
- ICT's methodology involves many nuances (balanced price ranges, order block overlap, candle body vs wick distinctions, specific gap types) that are difficult to fully codify from a video transcript.
- Only 6 months of data (166 trading days, 61 FVG-hit events) — a longer sample might reveal different behavior in trending vs ranging regimes.
- No higher-timeframe directional bias filter was applied. ICT typically requires alignment with the daily or weekly bias.
- The swing detection and FVG size parameters could be tuned differently. Results may be sensitive to these thresholds.
Data & Reproducibility
Disclaimer: This research is for educational and informational purposes only. It does not constitute financial advice. Trading futures involves substantial risk of loss. Past performance does not guarantee future results. Always do your own research before making trading decisions.