—
- Find a range — 40 hours where price went sideways instead of trending.
- Price drops below the bottom of that range, then closes back inside within three bars. The breakdown was fake.
- Buy that closing price.
- Stop just below the low it made on the way down. Target three times what you risked.
- Wyckoff said never trade one signal on its own. His two extras: the drop happened on quiet volume, and the bar that closes back inside is a strong bounce bar.
- The third toggle is ours, not his: how wide the range is compared to its own average hour.
- Buy only. The sell version is the same thing upside down, and it loses — see below.
- Our choices, not the books': the range must be no wider than 2.5× the 20 bars before it (so it is a range, not a trend); a trade still open after 30 hourly bars is closed; the “wide range” filter (range above 7.68 average bars) is a cut-off we found in our own earlier tests, so it is fitted to this data and should be read that way.
| trades | win | a trade | a random line | verdict | |
|---|---|---|---|---|---|
| The dip on its own | 192 | 27.6% | $5.43 | $21.14 | loses |
| + quiet drop + strong bounce bar | 66 | 42.4% | $133.08 | $32.93 | beats it |
| + wide range | 33 | 48.5% | $214.64 | $27.26 | beats it |
| Selling the mirror setup instead | 132 | 20.5% | −$9.75 | $21.09 | loses |
| The same rules on gold | 248 | 33.9% | $270.49 | $369.46 | adds nothing |
Think our rules are wrong? Change them. Ours is one version out of 192 we scored, each against its own randomly placed line. Every one is here, including the ones that fail.
Stacking is the whole thing. Wyckoff's instruction is never to trade one signal alone, and that is exactly what the numbers say. Require the two confirmations he names and the average goes from $5.43 to $133.08 a trade. It survives every check we run: out of sample, a random-level control at p<0.0001 across 192 searched variants, drop the best year, and a bootstrap that excludes zero.
The filter that helps most is our own measure of something he already said mattered. He sizes up a range with a Point and Figure count. We used how wide it is next to its own average hour. Same idea, a different ruler — and not something anyone tells you to check. Require it and 33 setups remain at $214.64 a trade, 48.5% win, positive in five years out of five.
The hardest test we ran, and it cuts the claim down. A spring sits at a low. So we compared every one against the other lows right next to it that were not springs — same days, same trend, same risk. On Nasdaq the structure is worth $9 a trade over an ordinary nearby dip, which at p=0.44 is nothing. On gold it is worse than the dips around it. The money in this setup comes from buying a dip, not from the range low or the confirmations. Everything else on this page is still true — the range low really does beat an arbitrary level, and that is worth knowing — but if you are deciding whether to trade it, this is the number that matters.
Gold makes money and still fails. The setup earns $235 a trade there, so why is it a failure? Because gold rose $301,260 a contract over the test period, and a long entered at a random hour with the same stop earns $214. The setup is worth $22 a trade, which is inside the noise. You would have made money on gold and credited the method, when almost all of it was simply owning gold in a bull market. On Nasdaq the same comparison gives the setup $96 a trade over a random entry, and it beats 98 random books out of 100. That gap is the whole finding.
There is no sell side. The mirror setup loses on Nasdaq in every version we tried, and the books' own advice for it — wait for the pullback rather than selling the breakout — makes it worse, not better.
Everything else we tested
The exit is not what is carrying it. The 3x target was our choice, not his, so we re-ran the same 66 trades under eight exits — six from his own book: risk one to make two, stop to breakeven, trail it, take half the prior move, take two-thirds, be flat by the close. 7 of the eight pass every check. The best is targeting the far side of the range at $210.79 a trade. We kept the 3x target anyway: it is what the published numbers were checked under, and picking the best of eight afterwards is how a curve fit starts. The sell side passes under none of the eight, and neither does gold.
Entering later, where the method says to, does not pay. The books describe a sequence: the fake break, a test making a higher low on lighter volume, a strong push up, then a pullback they call the best entry. We ran all four. The test entry earns about the same on barely half as many trades; the push-up and pullback entries leave too few trades to judge and lose money on those. On the sell side the same advice takes it from −$9.75 a trade to −$117.79 waiting for the push down and −$7.96 waiting for the pullback.
We built the range properly, and it barely exists. The strongest objection to this page is that a real range is a structure — a selling climax, the rally off it, then a retest on lighter volume — and not just 40 quiet hours. So we coded that. Across four years of hourly Nasdaq it finds 10 ranges, giving 5 trades. Loosening every threshold to the point where a "climax" is only an average-sized bar on average volume still gives 19 ranges and 11 trades. On daily bars, the timeframe Wyckoff actually wrote for, it finds 2. The handful of trades it does produce are profitable, but five trades prove nothing either way. The honest reading: the structure the books describe hardly ever forms on an index future, which is a fair argument that the method belongs on individual shares — not evidence that our lookback is generous.
Gold is here as the control, and the reason it fails is drift. The same rules off-venue lose to a randomly placed line at every level of stacking, and stacking makes gold worse while it makes Nasdaq better. The mechanism is not mysterious: over the test period gold rose $301,260 a contract, so any long with a stop below and a target above made money. A random-hour long earns $213.60; the spring earns $235.40. Twenty-two dollars of difference, better than 59% of random books — noise. On Nasdaq the same measurement gives $32.38 random against $128.79 for the spring, better than 98% of random books. A setup that works everywhere is usually measuring nothing; this one does not work everywhere, which is the reason to believe the Nasdaq result.
What this does not say. It does not say nobody makes money trading gold. It says that on this instrument, over this period, the signal was worth about nothing — so if you traded it and did well, the credit belongs to the trend, not the rule. Everything a real trader does that this test holds fixed — how much they size, which setups they skip, when they add, when they cut a loser early — is where the rest of the outcome lives. We hold those constant on purpose, because they are what makes a claim untestable. What we can tell you is whether the entry signal itself carries any weight. On gold it does not.
What did not survive: the parts Wyckoff stresses hardest. Timing the setup late in the range tested negative, as did prior-trend context and confluence with an older level. Two ideas of our own — a slower bounce back, and the bounce bar closing in the top third of its own range — failed validation and are not used here.
What this test does not do
We tested one mechanical reading of one setup. We did not test the Wyckoff method, which is a discretionary way of reading a chart and is far bigger than any rule you can code. Where our version and yours differ, ours is the one on trial.
- The range is a lookback, not a structure. We call 40 sideways hours a range. The books identify one by what happens inside it — a climax of selling, an automatic rally, then a retest on lighter volume. We built that version too, and it produced 10 ranges in four years — 5 trades. Loosen the definition as far as it will honestly go and you get 19 ranges and 11 trades. You cannot test anything on that. See below.
- We do not read the phases. No accumulation is confirmed before the trade. Every qualifying dip is taken, whether or not the range looks like buyers were absorbing supply.
- Wyckoff wrote about shares, not index futures. His whole model is one large operator quietly building a position. Nobody accumulates the Nasdaq. It is fair to argue the method does not transfer, and this page cannot settle that.
- It does not beat an ordinary dip. Against nearby local lows that were not springs, the structure adds $9 a trade on Nasdaq (p=0.44) and is negative on gold. It beats an arbitrary level; it does not beat the dip twenty bars away.
- We ignore what the wider market is doing. The method says to trade only in harmony with the trend and to pick the stronger instrument. Our test takes every setup regardless.
- 33 trades is a small sample for the fully stacked version. It clears every check we run, but a small sample is a small sample.
- One index, 2022-08-19 to 2026-09-01. Gold is here as a control, not as a second confirmation. More markets would make the result stronger.
If you think a rule is wrong, say which one and what it should be. We will test it against the same gates and publish whatever comes out, including if it beats ours.
Everything above is taken from the source and frozen. The chart tests those exact words — nothing added, nothing softened.