Does a wick rejection mean anything?

No, or at least not in the way the shape implies. A wick through a level tells you the bar was long enough to reach it. It does not tell you who was there, what they did, or whether the level mattered — and I can show that, because I built a stand-in with no participants in it at all and got the same answer.

I call this a Geometry Ghost: an effect that is real, large and overwhelmingly significant, produced entirely by the geometry of the measurement rather than by anything happening in the market.

The split that looks like proof

Take sixteen years of one instrument, define a level as reached when any part of the bar’s range crosses it, and record what happened next. Bars that touched with a wick and closed back absorbed at roughly 8%. Bars that reached by closing beyond absorbed at roughly 46%.

That is a gap of nearly forty points, at z = +198.

A result like that does not need defending. It replicates, it is nowhere near the boundary of significance, and no amount of extra data would weaken it. If you found it in your own testing you would reasonably conclude that wick rejections are a distinct and meaningful event.

The stand-in that has no market in it

Here is the test that decides it. Build a surrogate series matched to the real data on two properties only — its volatility and its wick geometry. Nothing else carries over. There are no participants, no orders, no levels anybody is defending, no intent of any kind. It is shape without a market.

Then run the identical measurement on it.

The surrogate produced +39%. The real data produced +38%. The excess — the part of the effect that belongs to the market rather than to the shape of the bars — was approximately zero.

The reason is mundane once seen. A bar with a long wick reaches further than a bar without one. Reaching further means touching levels that are further away, and a level further away is in a different state, tested under different conditions, than one price closed through. The split is real. Its cause is that long bars are long.

Why significance could not have caught this

A significance test asks whether an effect differs from zero. That is almost never the question. The question is whether the effect differs from what the measurement itself would produce given the same data shape — and no p-value, at any sample size, addresses that. Enormous n makes this worse rather than better, because it drives the p-value toward certainty on an effect whose cause was never in dispute.

The comparison that answers it costs a morning: construct a series that shares the shape and contains none of the supposed mechanism, and run the same code over it. Where the surrogate matches, the significance was never load-bearing.

This is not an exotic technique. It is standard practice in fields that got burned earlier than mine.

The definitional trap underneath

There is a second finding worth having, because it affects any level-based study whether or not wicks interest you.

Recounting the same sixteen years by close rather than by extreme took arrivals from 200,369 to 100,708. Roughly half of every “level touch” my engine had ever scored was a wick brushing past, not price trading through.

Neither count is wrong. They answer different questions, and both are defensible. The problem is that the choice is usually implicit, so two people testing the same idea on the same instrument can be working from samples that differ twofold before a single outcome is measured. State the reach rule first.

Limits

The result covers one instrument, on bar data, over sixteen years. A single-instrument result establishes nothing about any other, and I have not tested whether the same geometry dominates elsewhere — though the mechanism is arithmetic rather than market-specific, which is a reason to expect it travels and not evidence that it does.

More importantly, this kills a particular reading of wicks — that the shape carries information about participation — and nothing broader. Order-flow and book-based measures of what actually happens at a level remain entirely open; they are simply not what a candlestick shows you.

Why this one is worth internalising

Every trader eventually finds an effect that looks like this: large, consistent, statistically overwhelming, and visible in the data for as long as they care to look. Most of them are real in exactly the sense this one is real.

The question that separates the useful ones is never “is it significant”. It is “would a thing with no market in it produce this too”. That test is cheap, it is available to anyone with the data they already have, and it is the one almost nobody runs.