ICT Power of 3 (AMD) on ES and NQ: Accumulation, Manipulation, Distribution

The first move after the 9:30 AM ET open often looks like a breakout. Sometimes it is the real move. Sometimes price runs through the overnight low, turns, and spends the next hour taking out everyone who sold the break. ICT's Power of 3, also called AMD for accumulation, manipulation and distribution, gives you a vocabulary for that second case. It's a narrative framework, and what it's worth depends on whether you can turn it into rules.

The three phases

• Accumulation: price consolidates in a range, such as the overnight session.
• Manipulation: price makes a move against the eventual direction, usually through an obvious level such as the overnight high or low.
• Distribution: price moves in the direction of the real position for the rest of the period.

The claim is that the manipulation phase looks like a breakout setup, so traders enter it and become the liquidity for the move that follows. That explanation is a story about intent. What you can observe is the price path: a move beyond a visible level, a return inside the range, then travel the other way. The same path can also be a failed breakout with no one orchestrating anything, and that distinction doesn't change how you trade it.

A hypothetical bullish day on ES

Suppose the overnight range is 5,010 to 5,030 and the regular session opens at 5,020. At 9:41 AM ET price drops to 5,006, which is 4 points through the overnight low. A 5-minute bar closes back inside the range at 5,013 by 9:50.

Allow about $29 of friction on ES (a $4 round-turn commission plus a tick of slippage each way). Net reward drops to roughly $771, and a stopped trade realistically costs closer to $579. That's about 1.3 to 1 after costs on the nearer target. The further target at 5,040 is 26 points, or $1,300 gross, and it needs the day to actually trend. The entry came after the return inside the range, so part of the move was already gone by the time the signal printed. That's the price of waiting for confirmation.

Telling the phase apart in real time

The problem is hindsight. On a chart you can always find the manipulation after the distribution. In real time, a drop to 5,006 looks the same whether it reverses or keeps falling. The only things you can write down in advance are rules:

1. Name the level before the open: overnight high, overnight low, or prior session extreme.
2. Define a sweep: price trades at least a set number of points beyond the level. Scale the number to volatility, since NQ points are worth $20 and ES points $50.
3. Define the confirmation: a bar close back inside the range within a set time.
4. Define the invalidation: a close beyond the sweep extreme, or your stop.

If the setup never confirms, there is no trade. If it confirms and then fails, the stop does its job. A stop order can slip, especially around news, so size for a worse fill than the plan shows.

Failure modes

The most common one is the real breakout. Price runs through the low, closes back inside once, then drops through again and keeps going. Your stop below the sweep takes the loss, and the framework would call it a trend day with no manipulation phase. Others: a sweep that takes too long to confirm, a range so narrow that the stop is wider than the target, and scheduled news in the first hour that dominates any pattern.

Where the opening print fits

The regular session open at 9:30 AM ET is the natural midpoint reference for this framework. KPL levels are fixed reference prices anchored to that same opening print, with four levels on the standard ES and NQ map built from full-session and after-hours travel. Since the anchor isn't known until the open, the accumulation phase happens before any KPL map exists. The KPL levels themselves are not predictions, and they don't move intraday. If a sweep and a KPL level overlap, I'd record that as a tag and test it rather than assume it improves anything. See the KPL strategy page for the construction.

Weekly and monthly versions

ICT teachers apply the same pattern to weeks and months, for instance a mid-week extreme followed by a directional finish. I wouldn't size anything off that until you've counted it yourself over a long sample, since weekly counts accumulate slowly and a few weeks prove nothing.

How to test it

Log each trade with the level, sweep size, confirmation time, result in points and net dollars after costs. A set of 30 trades is not a statistically proven edge, and 50 SIM trades is not live proof. A simulated fill is not a test of exchange queue or slippage either. Run the same rules on separate data before putting real size behind them.