Power of Three / AMD

Power of Three, usually shortened to PO3 and used interchangeably with AMD, describes a directional move as three phases rather than as a single event: accumulation, manipulation, distribution. Price builds inside a range, trades against the eventual direction to reach the orders resting beyond one side of it, then delivers the move toward the other. It comes from the ICT and Smart Money Concepts tradition and behaves less like a strategy than a narrative template — a claim about the order in which things happen, which each version has to turn into rules.

## How the model works

The name comes from reading a candle from the inside. A closed bullish candle usually opens near its low, wicks below the open, and closes near its high; the wick is read as manipulation and the body as distribution. Because a higher-timeframe candle is, on a lower timeframe, a finished sequence with internal structure, the model is applied fractally: the container is defined above — a daily or 4-hour candle, a session, a prior day's range — and the phases are looked for below.

The other common framing is time-anchored. Asia builds a comparatively tight range (accumulation), London trades through one side of it (manipulation), New York delivers the day's real leg (distribution). Both framings share a weak point: "manipulation" only means something if you decided in advance which side is the false one. That decision is external to the model and comes from a directional bias — higher-timeframe draw on liquidity, position within a weekly range, prior-day extremes, correlated-asset divergence.

Entries are rarely taken on the phase label alone. Most versions wait for the sweep to complete and then require a lower-timeframe trigger — a market structure shift, a displacement leg, a fair value gap, an order block — with the stop beyond the manipulation extreme and the target at opposing liquidity or the opposite side of the container.

## Main variants

Higher-timeframe alignment versions read PO3 on the daily or weekly candle and trade only in the direction of its presumed close, often with a tool that projects the unfinished candle's open, high and low onto the entry chart. Session or AMD versions keep the Asia–London–New York schedule and make the clock a hard filter. Pure price-action versions drop indicators and anchor everything to reference opens — midnight, daily, session — treating the phases as a timetable. Pattern-combination versions use PO3 to time a specific reversal entry, most often a failed breakout of a prior extreme acting as the manipulation phase itself.

## What typically differentiates implementations

Two versions can share the vocabulary and produce unrelated trade lists. The differences that matter: what defines the container (closed candle, session range, prior-day range, opening range) and how long it stays valid; how directional bias is established, and whether that is a written rule or a judgment call; what counts as a valid manipulation (a wick through a reference, a close beyond and a reclaim, a minimum penetration); whether the sweep has a window it must occur in and a deadline after which distribution is abandoned; whether entry is at the swept level, at a retracement, or only after confirmation; and where stop and target sit.

## Common mistakes

The dominant error is retrospective labelling. On a finished chart almost any completed swing can be split into three phases, so the model looks universal until it has to name the manipulation side in advance. Closely related is assuming that phase is mandatory — plenty of sessions expand out of the range and never come back, and a version with no rule for that case takes the trend day as a loss. Others recur: hard-coding the session template without checking the instrument respects it; handling the clock carelessly, since Asia, London and New York are defined against exchange time and shift with daylight saving; stacking confluences that derive from the same displacement leg, which adds no independent evidence and only shrinks the sample; and treating tendencies such as "the day's low forms in London" as rules rather than as measured frequencies.

## How to evaluate and backtest a version

Write the rules out and mark the bias step honestly — that is where the discretion usually hides. If a version cannot state how direction is chosen, what remains testable is the conditional claim: given a bias, does the sequence add anything to it?

Before testing entries, measure base rates. How often is the Asian range extreme swept during London on this instrument? How often does the day's high or low form inside the stated window? How often does the distribution leg reach the opposite side of the container? Cheap to compute, and often decisive: if the template does not describe the instrument, entry refinement will not rescue it.

Then fix the session offset and re-run with a different one — a result that survives only at one daily open is anchored to a timestamp, not to behaviour. Compare against a time-matched baseline, the same stop and target taken at the session open with no phase logic, so the pattern's contribution is separated from the hour's. Include spread, commission and slippage, and resolve bars that touch stop and target together pessimistically. Finally, respect the sample: at roughly one opportunity per session, log no-trade days too, extend across instruments, years and volatility regimes, and check sensitivity to the sweep threshold and the window boundaries.

Strategies in this concept (10)

Frequently asked questions

Are Power of Three and AMD the same thing?

In practice they are used interchangeably, and most material treats them as one concept. Where a distinction is drawn, Power of Three refers to the candle reading — open, wick, close, with the three phases visible inside a single bar — while AMD refers to the phase sequence itself, most often mapped onto sessions. The underlying claim is identical: a range is built, one side of it is taken, and the move is delivered toward the other.

How is this different from Candle Range Theory?

They describe the same behaviour at different levels of specification. Power of Three names the phase order without committing to where the range begins and ends; CRT pins the container to one closed candle's high and low and treats the opposite extreme as the default objective. Because of that, CRT setups can be written out in advance more easily, while PO3 versions still have to state what the container is before the model becomes testable.

Does the manipulation phase always happen?

No, and a version that assumes it does will be short a rule for the days it matters most. Trend days expand out of the accumulation range and never return; news-driven sessions can skip the sequence entirely. The useful figure is the frequency, not the assumption — measure how often the pattern actually completes on your instrument and session before deciding how much of the model to build on it.

Which sessions and timeframes is the model applied to?

The Asia–London–New York mapping is the most common because it gives the three phases natural boundaries, but the template is used on weekly and daily candles as well, and shrunk onto intraday cycles inside a single session. What matters more than the choice itself is the ratio between the container timeframe and the entry timeframe, and whether the instrument is genuinely session-driven — the mapping assumes participation shifts at those hours, which holds for index futures and major FX far better than for assets that trade evenly around the clock.

How much of the concept can be automated?

The geometry automates cleanly: session ranges, sweeps of a defined level, stop and target placement, time windows. The part that resists automation is the directional bias — deciding which side of the range is the manipulation before it happens. A coded version has to replace that judgment with a rule, such as higher-timeframe structure or position within a prior range, and the rule chosen matters more to the results than any of the entry mechanics.

What should I check first when backtesting a version?

Base rates and the clock. Compute how often the accumulation range is swept in the stated window and how often price then reaches the opposite side, before testing any entry logic — if those numbers are unremarkable, the entry refinement is decorating a template the instrument does not follow. Then confirm the session times are handled in exchange time with daylight saving applied, and re-run with a shifted offset; a version that only works at one anchor is fitted to a timestamp.

All strategy concepts · Explore all strategies