Wyckoff and the Power of Three (AMD) Explained, and Tested

Wyckoff is one of the oldest ways of reading a trading range, and it is the root of the Smart money ideas many newer traders use today. ICT’s Power of Three, or AMD, is a short version that squeezes the same story into a single candle. This article explains both step by step, then does what most articles skip: it writes them as measurable rules, counts how often they show up in prices that are pure chance, and looks at our own backtests on gold.

How Wyckoff saw the market

Richard D. Wyckoff was an American trader and writer in the early 20th century. He taught students to picture the market as if one large player were running it, which he called the Composite operator (or Composite man). This player quietly buys while price is cheap without letting it run, then slowly sells while price is expensive to the people chasing the news.

Wyckoff did not claim such a person really exists. It is a model that keeps you asking: “If I were the big player, what would I want small traders to do right now?” The method rests on three laws.

1. Supply and demand. Price rises when buying pressure beats selling pressure and falls when the reverse is true. It sounds obvious, but everything else follows from it.
2. Cause and effect. A long sideways range (the cause) builds energy for the move that follows (the effect). The wider and longer the range, the further the breakout tends to travel. Wyckoff followers use Point and Figure charts to count a range and set targets.
3. Effort vs result. Effort (volume) should match the result (how far price moves). If volume surges but price barely moves, someone on the other side is absorbing it.

Accumulation phase by phase: from PS to LPS

Accumulation is a range that forms after a long decline, which Wyckoff saw as the period when large players build positions. The Phase A–E labels and the schematic drawings you see today were organised by later teachers who carried on his method.

Phase	Event	What it means
A: the fall stops	PS (Preliminary support)	First real buying appears and the decline slows
	SC (Selling climax)	Panic selling, a sharp drop on high volume, then buyers absorb it
	AR (Automatic rally)	A strong bounce once selling is exhausted. Its high becomes the top of the range
	ST (Secondary test)	Price revisits the SC area, usually on lighter volume
B: building the cause	Several STs	A long swing inside the range while large players accumulate
C: the final test	Spring, then a Test	A brief drop below the range low that snaps back inside, taking out the stops of buyers and luring in sellers
D: an uptrend begins	SOS (Sign of strength)	A strong rally on rising volume, through or close to the range top
	LPS (Last point of support)	A shallow pullback that holds, the entry Wyckoff followers like
E: uptrend	Markup	Price leaves the range and trends

The Spring gets the most attention, because if you read it right you enter near the bottom of the range with a tight Stop loss below the Spring low. Wyckoff’s own teaching does not say every range has a Spring. Some ranges move out of Phase C without ever breaking the low.

Distribution: the same picture upside down

Distribution is a range after a long rise, where in Wyckoff’s picture large players sell to the people chasing price. The events mirror accumulation.

• PSY (Preliminary supply): first real selling appears and the rise slows
• BC (Buying climax): heavy buying on high volume makes the top
• AR (Automatic reaction): a drop once buying is exhausted. Its low becomes the bottom of the range
• ST: a retest of the BC area, sometimes with a UT (Upthrust) that pokes above the top and falls back
• UTAD (Upthrust after distribution): the mirror of the Spring. A late break above the top that looks like a breakout, then falls back inside
• SOW (Sign of weakness): a sharp drop through or close to the range low
• LPSY (Last point of supply): a weak rally that fails to reach the top, before the markdown begins

Effort vs result, and the volume problem in forex

Wyckoff built his method on the stock market, where every order goes through a central exchange, so volume is a real number everyone sees. Spot forex trades in a decentralised way across many banks and liquidity providers (more in FX market structure and liquidity), and nobody sees the volume of the whole market.

The volume MT5 shows for currency pairs and gold CFDs is tick volume: the number of times the price in that broker’s feed changed during the bar. It tells you when the market is busy or quiet, but not how much money changed hands, and two brokers can show different numbers.

In practice:

• Use tick volume to compare bars on the same chart, for example whether the SC bar was clearly busier than its neighbours
• Don’t read more detail than the data holds. A rule like “ST volume 30% below SC confirms it” means nothing with tick volume
• Tick volume tends to jump at the London and New York opens and on news. That is a time-of-day pattern, not a footprint of accumulation

ICT’s Power of Three (AMD)

ICT took the idea of “accumulate, trick, then run” and compressed it into three moves, called the Power of Three or AMD.

1. Accumulation. Price swings in a narrow range near the open, usually tied to the Asian session.
2. Manipulation. Price runs against the real direction to sweep stops on one side. ICT calls this false move the Judas swing. It is usually tied to the London open.
3. Distribution. Price makes the real move and closes near the other end of the candle, usually tied to the New York session.

On an up day, price first drops below the open to make the low of the day, then rallies. The result is a bullish daily candle with a lower wick. Down days are the mirror image. ICT applies the same frame to weekly and H4 candles too.

Watch the clash of words. Wyckoff distribution is a range at a top where large players sell. AMD distribution is the phase where price makes its real move. They are completely different events.

The bigger problem with AMD is candle geometry. Almost every bullish daily candle has a lower wick, which means price traded below the open first. We simulated 100,000 days of a pure random walk, 96 steps per day (the same as 15-minute bars over 24 hours). No big player, no manipulation. The result:

• 50,153 days closed up
• Of those, 38,960 days, or 77.7%, dropped below the open and made their low before their high, meaning they “have the AMD shape”
• 69.5% of up days also made that low in the first half of the day

So once you know a day closed up, looking back you will see “manipulation” on most up days, even when prices are random. The useful question is not “can I see the AMD shape?” but “while price is below the open, can I know in advance that today will close up?”

Worked example: write the Spring as a rule and count it in a random walk

Wyckoff takes interpretation. The only way to know whether it adds information is to write it as a clear rule. We defined a Spring like this (price units are the “average step” of one bar):

1. Range: the highest close minus the lowest close of the previous 40 bars is no more than 8 units (a narrow range)
2. Spring: a bar closes at least 0.5 units below the range low
3. Confirmation: within 3 bars, a bar closes back above the range low. Buy at that close
4. Exit: target is the range top. Stop loss is the lowest close during the Spring

Then we ran it on prices made purely by chance (each bar moves up or down at random with mean 0 and a standard deviation of 1 unit): 200 series of 5,000 bars, 1,000,000 bars in total. No big player, nobody accumulating.

• The narrow-range condition held on about 45% of all bars
• The rule found 6,164 Springs, about 6.2 per 1,000 bars. On an H1 chart of a market open 24 hours, 5 days a week (120 bars a week), 1,000 bars is about 8 weeks, so you would see around 6 “Springs” in two months from chance alone
• 6,149 finished (15 ran out of data). 1,563 reached the target and 4,586 hit the Stop loss
• Win rate = 1,563 ÷ 6,149 = 25.4%

25.4% sounds poor, but it has to be read with the size of wins and losses. Winners made 6.04 units on average and losers cost 2.13 units.

1. RR = 6.04 ÷ 2.13 = 2.84
2. Break-even win rate = 1 ÷ (1 + 2.84) = 26.0% (the formula is explained in win rate vs break-even)
3. The actual 25.4% is 0.6 points below break-even
4. The average result per trade is −0.051 units with a standard error of 0.047, about one standard error from zero, so it can’t be told apart from zero. Profit factor 0.97

That is what we expect, because a random walk has no edge to capture. The point is that these numbers are the baseline. If you test the same rule on real prices and get a 26–27% win rate at a similar RR, that is still not evidence that Springs work. It has to clear break-even clearly after spread and beat what random prices deliver.

If a pattern shows up in random prices about as often as in real ones, seeing it on your chart tells you nothing about big players.

Our backtests: PO3/AMD on gold

In our backtests we wrote the Power of Three as rules and ran them in the MT5 Strategy Tester on gold (XAUUSD) with real tick data. These are backtest results. Past results do not predict future results.

Set 1: PO3 entries on M1, January to August 2026. The box was 40 bars that were narrow relative to ATR. A strong enough push out of the box counted as “manipulation”. We tested 3 entries: fade the break (the AMD idea), follow the break, and a strict version that waits for price to close back inside the box before fading. Combined with several exits, that made 10 variants.

• No variant made more net profit than its own maximum drawdown
• The best was fading with the original exit (Stop loss 5 × ATR, target 0.86 × the Stop loss): +351 dollars, with a maximum drawdown of 597 dollars over 1,424 trades at 0.01 lot
• Its 53.8% win rate sounds good, but break-even at RR 0.86 is 1 ÷ 1.86 = 53.8%, exactly the same
• The strict AMD version that waits for a close back inside lost with both exits (−831 and −719 dollars)

Set 2: AMD on boxes of several sizes, January to September 2026. Boxes on M5, M15 and H1. Wait for a poke out that closes back inside, fade it, and target the opposite edge and 161.8% of the box. All 10 variants lost. Profit factor was 0.80–0.82 for the M5 boxes and 0.96–0.98 for the M15 boxes. The H1 boxes gave only 34–84 trades and won just 0–14%.

Another count of ours helps explain this. On gold M1 in 2026, when price broke out of a narrow box, it continued in the breakout direction 56% of the time versus 50% for a random baseline (7,800–25,000 events per box size). On this data, breakouts leaned towards being real rather than fake, the opposite of AMD’s core assumption.

Limits: this is our own interpretation as fixed rules, on one market in one year. Someone reading Wyckoff with judgement and a bigger context may get different results. What it does show is that “box, fake-out, reversal”, in a form that can be written as rules, has no evidence of an edge so far. See more ICT tests in ICT concepts on gold and Smart Money Concepts tested.

Why schematics always look right in hindsight

Wyckoff’s schematic is very flexible. Phase B can last any length, there can be any number of STs, and the Spring may or may not appear. If the range breaks down instead, it gets relabelled as re-distribution. That flexibility means almost any chart “fits Wyckoff” once you know the ending, which is the same trap as hindsight bias.

Here is a simple check. While a range is still unfolding, write in your journal which phase you think it is in and what would prove you wrong. If you can’t write that down, you are using the schematic to tell a story after the fact, not to make decisions.

Wyckoff is useful as a map that helps you ask good questions. But SC, Spring or UTAD labels added after price has already moved are not evidence.

Using Wyckoff in a way that helps

• Use it as context, not a stand-alone signal. Wyckoff-style questions are genuinely useful: is the decline slowing, where was the busiest bar, did a break below the range come back quickly? Combine them with measurable support and resistance
• Write the definition as numbers first. How many bars make the range, how far below counts as a Spring, how many bars to get back inside. If the definition changes with each chart, you can’t measure it
• Place the Stop loss at the structure, then size from that distance. Below the Spring low or above the UTAD high. The distance differs each time, so use position size from Stop loss to keep the risk per trade constant
• Compare with break-even, not 50%. Springs often give a high RR because the stop is tight, so a low win rate is normal. Use the RR calculator to find your break-even line
• Don’t try many definitions and keep the best. The more variants you try, the more likely one looks good by luck (multiple testing). And remember stops really do sit below old lows (round numbers and stop clustering), so a poke below and back can happen without anyone planning a trick

Common mistakes

• Labelling phases after the fact. Call a range that breaks up accumulation and one that breaks down distribution, and you are never wrong, but it never helps you decide either
• Buying every break below the low. A real Spring has to close back inside the range. A break that stays out is a breakout, and on our data breakouts more often continued than reversed
• Reading tick volume like stock volume. It can compare activity on the same chart, but it is not the amount of money traded and it differs between brokers
• Looking only at the win rate. Winning 75% can still lose if the target is much smaller than the Stop loss, and winning 25% can break even when the RR is high. Always compare with 1 ÷ (1 + RR)