ICT Silver Bullet

The ICT Silver Bullet is a time-of-day intraday model associated with Michael J. Huddleston (Inner Circle Trader). Its defining idea is narrow and, unusually for discretionary trading concepts, fairly testable: instead of watching price all session, the trader restricts attention to a fixed one-hour window and looks for one specific sequence inside it. Outside the window, there is no trade — the clock is part of the setup, not a convenience.

The model sits inside the broader ICT vocabulary of liquidity, displacement and price-delivery arrays, but it is usually taught as a self-contained routine. That combination — a small rule set with a hard time filter — is why it appears so often in video form, and why coded versions of it differ so widely from one another.

## The core sequence

Most descriptions follow the same three beats. First, a **draw on liquidity** is identified: an old high or low, a prior session extreme, equal highs or lows, or a relatively obvious cluster of resting stops. Second, inside the window, price is expected to **take that liquidity and then displace** — move away from it with an aggressive, one-sided leg rather than drifting. Third, the trader waits for a **retracement into the imbalance** that displacement created, typically a fair value gap (a three-candle imbalance where the first and third candles do not overlap), and enters there.

The stop usually sits beyond the swept extreme or beyond the origin of the displacement leg. The target is usually the opposing liquidity pool — the high or low the move is presumed to be reaching for — sometimes with partials at intermediate session extremes. Everything else in the model is scaffolding around those three beats.

## The standard windows

The windows are quoted in New York time and correspond to periods of elevated participation: roughly 03:00–04:00 (London), 10:00–11:00 (New York AM) and 14:00–15:00 (New York PM). Some teachings add an Asian-session window. The windows are the most-copied element of the concept and also the most fragile: they are defined against a specific exchange clock, and any version that hard-codes them in a broker or platform timezone will drift twice a year with daylight-saving transitions.

## Main variants

Versions of the model diverge mainly along five axes. **Which window** is traded, and whether more than one is allowed per day. **Which entry array** is used — the fair value gap itself, an order block, a breaker, or an inversion FVG created when a gap is traded through and then respected from the other side. **Whether confirmation is required** before entry: some versions take a resting limit order at the gap, others demand a market structure shift or change in state of delivery on a lower timeframe first. **Whether directional bias is imposed** from a higher timeframe (daily or 4-hour draw, weekly range position, SMT divergence between correlated instruments such as NQ and ES) or whether both directions are taken symmetrically inside the window. And **how the trade is managed** — fixed R multiples versus liquidity-based targets, breakeven rules, partials, and whether an unfilled or open position is closed at the end of the window.

## What typically differentiates implementations

The headline rules are nearly identical across versions; the differences live in the thresholds. What counts as a valid sweep — a wick through the level, or a close beyond it, and with how much tolerance? What counts as displacement — a candle range relative to recent ranges, an ATR multiple, a minimum gap size, or nothing more explicit than "it looked impulsive"? Which timeframe is the gap read on, and is entry at the gap's high, its midpoint (consequent encroachment), or its far edge? Is the stop structural or a fixed distance? Two implementations can share every named component and still produce entirely different trade lists because of these choices. When comparing decoded versions, the threshold definitions are where the real comparison happens.

## Common mistakes

The recurring failure is treating the window as the signal. A window with no clear draw on liquidity, or a gap that formed without a preceding sweep and displacement, is not the setup — and "no trade today" is a legitimate outcome the model produces often. Timezone handling is the second most common defect. Beyond that: taking any imbalance rather than the one displacement created; forgetting that a fair value gap is only identifiable once its third candle has closed, which quietly introduces look-ahead bias in naive backtests; ignoring spread, commission and stop slippage on a model that trades small ranges on low timeframes; and tuning tolerance parameters until a handful of chart examples line up.

## How to evaluate and backtest a version

Start by making the rules mechanical enough that two people would produce the same trade list — sweep tolerance, displacement threshold, entry level, stop, target, and the exact exchange-time window with correct DST handling. You will need 1-minute or tick data with trustworthy timestamps; higher timeframes will not resolve the entry.

Then respect the sample-size problem. At roughly one opportunity per window per day, a single year of a single window on a single instrument is a thin sample for anything beyond a first look. Extend across multiple years, windows and correlated instruments, and split the data out-of-sample by period rather than shuffling. Log every session, including the ones that produced no trade, so the frequency of the setup is measurable rather than assumed.

Useful diagnostics: the distribution of maximum adverse and favourable excursion, which tells you whether stops and targets are placed where the data actually goes; results segmented by regime (trending versus range-bound periods, high versus low volatility); sensitivity of the outcome to each threshold, since an edge that disappears when the displacement filter moves slightly is a fitted one; and a time-matched baseline — random or fixed entries in the same window — to separate what the pattern contributes from what the session timing alone contributes. Finally, note which parts of the original teaching a given version left to judgment. Every coded Silver Bullet is an interpretation, and the interpretation is what you are testing.

Strategies in this concept (4)

Frequently asked questions

What exactly are the Silver Bullet windows, and which timezone are they in?

The commonly cited windows are approximately 03:00–04:00, 10:00–11:00 and 14:00–15:00 New York time, with some teachings adding an Asian-session window. The timezone matters more than it appears: the windows are defined against the New York exchange clock, so a version that hard-codes them in a broker or local platform timezone will be an hour off for part of the year. Any implementation you evaluate should state how it handles daylight-saving transitions.

Do I need to know the whole ICT framework to understand this concept?

No. The Silver Bullet is usually taught as a self-contained routine, and its core sequence — liquidity taken, displacement away from it, entry on the retracement into the resulting imbalance — can be described without the surrounding framework. You do need working definitions of a liquidity pool, displacement, and a fair value gap, because those three terms carry all the specificity in the model. Familiarity with the wider vocabulary mostly helps when a particular version adds bias filters such as SMT divergence or higher-timeframe draw.

Which markets and timeframes is it usually applied to?

It is most often presented on index futures (Nasdaq and S&P contracts) and on forex majors, with gold appearing regularly as well. Entries are typically read on 1- to 5-minute charts, since the fair value gap that anchors the entry is not resolvable on higher timeframes. The instrument choice is not incidental — the model depends on session-driven participation, so it is usually applied to instruments that are actively traded during the chosen window.

How is this different from a generic fair value gap strategy?

A generic gap strategy takes imbalances wherever they appear. The Silver Bullet adds two constraints: the gap must fall inside a fixed one-hour window, and it must be the product of displacement that followed a liquidity sweep, moving toward an identified draw. Those constraints reduce the number of candidates dramatically. If a version accepts any gap inside the window without requiring the sweep-and-displacement sequence, it is closer to a session-filtered gap strategy than to the model as taught — which is a fair thing to test, but worth naming accurately.

Can the concept be fully automated?

Parts of it automate cleanly — the time window, the gap geometry, the stop and target placement. The parts that resist automation are the ones the original teaching leaves to judgment: which liquidity pool is the real draw, what magnitude of move counts as displacement, and how much tolerance a sweep gets. A coded version has to commit to thresholds for each of those, and different threshold choices produce genuinely different strategies. When comparing implementations, the useful question is not whether it is automated but which discretionary decisions were converted into numbers and what numbers were chosen.

What data and sample size do I need to backtest a version properly?

You need 1-minute or tick data with reliable exchange timestamps, and enough history to cover several market regimes. Sample size is the binding constraint: at roughly one opportunity per window per day, a single year on one instrument and one window yields a few hundred observations at best, many of which will be no-trade sessions. Extend across multiple years, windows and instruments, split out-of-sample by period, and log the sessions that produced no setup so the frequency of the pattern is measured rather than assumed. Include realistic spread, commission and stop slippage — on a low-timeframe model, they are not a rounding error.

All strategy concepts · Explore all strategies