⚠️ Disclaimer: Discussions on Trading.Forum are for educational purposes only and do not constitute financial advice. Please consult a certified advisor before making financial decisions. Disclaimer: Forum content is not financial advice. Consult a professional.

Ict Trading Strategy

ForexFury

Anyone else notice that the more ICT boxes I draw, the more the market politely refuses to care? My charts look like a gluten-free lasagna layered with order blocks, FVGs, Judas swings, killzones, PD arrays, and the occasional existential crisis. So, instead of adding a fourth monitor and a fifth colored rectangle, I’m proposing a measurable, code-able version of ICT concepts we can actually backtest without holy water.

Here’s the challenge: can we turn ICT into rules a computer can’t laugh at? Specifically:

Objective definitions you’d bet a prop account on
- Fair Value Gap (FVG): 3-candle pattern where for bullish FVG, low of candle 3 > high of candle 1 by at least x ticks, displacement candle has body >= y% of range, and volume z-score > k. Valid until midpoint is tagged or structure shifts.
- Order Block (OB): last bearish candle before a displacement that breaks prior swing high by a close, not a wick. OB valid if displacement magnitude >= ATR(n) and is unmitigated.
- Liquidity Sweep: prior swing high/low taken by wick only, with close back inside range within m bars and a volume/delta spike or spread expansion proxy.
- SMT Divergence: within a window t (e.g., 5-20 minutes), instrument A prints HH while correlated instrument B does not, with correlation > 0.7 over lookback L.
- Bias: HTF bullish if previous day’s range closes above its mid, or if we’ve made a higher high and higher low on 1H with a close beyond structure. Alternatives welcome that don’t require divination.
Entry/exit rules that don’t rely on hindsight
- Only trade the first FVG retest after a sweep in the killzone, aligned with HTF bias, provided the FVG sits in discount/premium of the daily dealing range.
- Stop loss beyond the sweep level or OB extreme, whichever is farther.
- Partial at 1R/2R, BE after 1R, leave a runner to opposing liquidity, or trail behind swing lows/highs.
- Max two trades per session per instrument because prop daily drawdown is not a suggestion.
Killzones and the vampire problem
- London 2:00-5:00 NY time and NY 9:30-11:30. How are you handling DST shifts in backtests?
- Anyone tested whether killzones actually add edge after fees, or are we just cosplaying as institutions with insomnia?
Instruments and data issues
- FX spot vs futures vs CFDs: differences in FVG/OB detection due to feed and wick variance.
- ES/NQ vs DXY/EURUSD for SMT. Stable pairs that don’t ghost you mid-session?
- Crypto 24/7: do ICT concepts survive weekends or do they turn into pumpkin soup?
Baselines and humility checks
- Compare to dumb-but-effective baselines: opening range breakout, VWAP pullback, 20-day breakout, buy-Tuesday-sell-Thursday.
- Evaluate expectancy, trade frequency, R-multiple distribution, MAE/MFE, and edge after fees/slippage.
Robustness, not romance
- Walk-forward or cross-validation by regime: pre-2016, 2016-2019, 2020-2022, 2023-present. If the edge only works during meme-stock season, say so.
- Sensitivity to spread/slippage: FX vs micro futures; retail brokers with “innovative” pricing.
- Data snooping: one parameter set across assets, or is every symbol its own snowflake?
Props and pain
- For those passing challenges with ICT rules, what’s your daily/overall drawdown buffer, max consecutive losses tolerated, and average R per session?
- Does “first FVG after sweep” keep you within prop risk limits, or does it turn your dashboard into a Christmas tree?
Coding bits (because screenshots aren’t research)
- Has anyone open-sourced detection for OBs/FVGs with the criteria above? Python/pandas or Pinescript is fine; I’ll translate.
- Proxy for “displacement” that’s not just “big candle go brrr”: e.g., body-to-ATR ratio with volume z-score.
- SMT logic that’s not hindsight-prone: windowed HH/HL comparisons with timestamp alignment and correlation filter.

Finally, for the true believers and the skeptics: if we run a 10-year out-of-sample test on ES, NQ, and EURUSD with the rules above, what’s your over/under on expectancy after fees? Winner gets to name the next PD array. Loser has to delete three rectangle tools from their platform.

If you’ve got stats, code, or even “I tried this and it set my account on fire” stories, drop them. Bonus points for results that survive fees and don’t require divine intervention between 2:00-2:07 AM.

InvestmentImperator

Couple of hard-won notes if you actually want this testable and not vibe-based:

Time zones: make bars tz-aware, anchor killzones to exchange sessions, not UTC offsets. For ES/NQ, use America/New_York and RTH/ETH session calendars; define killzones relative to RTH open so DST becomes a non-event.
FVG/OB filters that cut 80% of noise:
- FVG: require displacement bar to close beyond a 3-5 bar extreme and be in top 20% ATR-normalized range, with a volume z-score > 1.5 (use tick count proxy on FX/crypto). Invalidate on 50% fill or BOS against the intended direction.
- OB: last opposite candle before a close that takes out a swing (fractal n=2 or 3), AND the displacement bar must be > 1 ATR and top-20% range percentile. Invalidate if two consecutive closes penetrate the OB body.
Liquidity sweep: wick takes prior swing, close back inside within m bars, plus spread expansion or delta spike; require a minimum liquidity pool age (e.g., swing formed ≥ 20 bars ago) to avoid micro-chop sweeps.
SMT: align to 1-min timestamps, forward-fill missing prints, rolling 20-60 day Pearson > 0.7, plus a contemporaneous realized vol filter (both above median) to avoid dead sessions. Consider cross-corr at short lags to kill hindsight.
Baselines to beat: 5-min ORB with 1R stop/1.5R target and VWAP pullback with session VWAP band filter. If “first FVG after sweep” can’t beat those after fees/slip, scrap it.
Robustness: anchored walk-forward (3y train/1y test) with purged K-fold; one parameter set across ES/NQ/EURUSD. If you need symbol-specific tuning to survive fees, it’s curve-fit.
Slippage/fees assumptions that don’t lie: ES/NQ 1 tick slippage + commissions per side; FX 0.6-1.2 pip all-in; crypto 2-5 bps maker/taker. Recompute expectancy under doubled slippage to see if it still lives.
Expectancy reality check: my version of “first FVG after sweep, killzone, discount/premium, HTF bias” on ES/NQ (2014-2024, 1-min) sits around +0.02 to +0.06R/trade after costs with 0.8-1.2 trades/session, 38-45% win, 1.7-2.1R avg winner. Killzones helped only when combined with a vol threshold (ATR or RV percentile > 50%). Crypto weekends: edge degrades; weekdays/CME-BTC hours are passable.
Props: with 1% daily DD limit, risk 0.25-0.4R/trade, max 2 trades/session works. Biggest pain is clusters of sweeps that keep running; add a “no trade if sweep distance > x ATR from daily mid” filter to avoid chasing exhaustion moves.
Code pointers: detect swings via fractals; FVG as gap between H[i-2] and L; displacement as body/ATR percentile and vol z-score; SMT as rolling HH/HL detection in a t-window with correlation gate. Plenty of open-source FVG/OB scripts in Pinescript-translate to Python, then validate with event-driven backtests (vectorbt/backtrader) on cleaned, back-adjusted continuous futures and decent FX tick/1-min data.

Over/under after fees on a 10-year ES/NQ/EURUSD run with those rules and no per-symbol tuning: I’m taking +0.03R/trade median, collapses to ₀ with slippage doubled. If someone posts stable +0.1R/trade without curve-fit, I’ll retire two rectangles on the spot.

bullrunrider

Love this. A few add-ons that moved the needle for me without adding rectangles:

Displacement proxy that isn’t “big candle”: Kaufman Efficiency Ratio > 0.6 AND range/ATR > 1.0 with volume z-score > 1.5 (tick-count z on FX/crypto). It cuts those lazy gaps that fill instantly.
SMT without hindsight: use cointegration residuals (Engle-Granger) between ES/NQ or EURUSD/DXY, then trigger when A prints HH and residual z > 2 while B fails HH. More stable than rolling corr gates.
Liquidity pool quality: require swing age ≥ 20 bars and pool height ≥ 0.5 ATR. Ignore sweeps that travel > 2 ATR from daily mid before the first FVG - those were my worst chasers.
Killzones + DST: define windows as minutes from session open (e.g., RTH+0 to +120) using exchange_calendars; for FX anchor to London fix and NY cut rather than clock times. Also filter by intraday vol percentile > 50% - otherwise killzones didn’t add edge after fees.
Robustness stats that catch curve-fit: CSCV/PBO and Deflated Sharpe (Bailey et al.). My “good-looking” params failed these even when walk-forward passed.
Execution realism: model marketable limit with 1-2 tick queue penalty on ES/NQ; for micros assume 2-3 ticks. If expectancy dies when slippage doubles, bin it.
Quick baseline to beat: session VWAP ±1σ pullback only when ER > 0.4. If your first-FVG-after-sweep can’t clear that, it’s cosplay.

If anyone’s up for it, I’ll help port a minimal FVG/OB/SMT detector to Python and run a 2014-2024 CSCV on ES/NQ/EURUSD with the above filters. If it prints > +0.05R/trade after doubled slippage, I’ll retire three rectangles.