Gamma and Order Flow: What Confluence Actually Tells You

Scope: a mechanism note on what agreement between an order-flow signal and a gamma level can and cannot tell you. It presents no measurements of its own.

A gamma level can be exactly right about where dealers are obliged to hedge and still describe a price nobody trades. Order flow can show heavy participation at a price with no structural reason to matter. When the two point at the same price, you have two measurements of different things agreeing — which is evidence about which price matters, and no evidence at all about when. Nothing below is a measurement; it is mechanism, and the distinction is the point.

What is a gamma level actually saying?#

That if price arrives, someone is obliged to trade. It is arithmetic on outstanding options positions, not an observation of anything that has happened.

The mechanism is ordinary hedging — with one assumption embedded in it before any level can be drawn. Open interest does not label its counterparties, so every gamma level you have ever seen rests on a convention about which side of the book dealers are sitting on. Grant the convention: the assumed position's exposure to the underlying changes as the underlying moves, and staying neutral means trading the underlying continuously as price travels. Gamma is the rate at which that hedging requirement changes. Where a lot of open interest sits at one strike, the requirement concentrates — and that concentration is what every gamma level, wall, or flip point is describing.

The second qualification belongs next to the first, not in a footnote: the level is conditional. It says what hedging pressure would concentrate there if price arrives. It is not a forecast that price will arrive, and it never was.

What is order flow actually saying?#

That someone did trade, in what size, and in which direction. It is a record, not an obligation — which is exactly the question the options chain cannot answer.

The same level produces three completely different afternoons, and only the tape distinguishes them. Price arrives, size clusters, price turns: the hedging flow met real participation. Price arrives, almost nothing trades, price drifts straight through: the level was arithmetic nobody defended. Or price arrives and size clusters on the other side — someone absorbing the flow, which is often the most informative of the three and is entirely invisible to a positioning-only view. What clustering size predicts on its own is measured separately: a cluster of large trades sizes the next hour's move and says nothing about its direction.

Order flow has its own ceiling, and it is worth stating plainly. The tape is backward-looking. It tells you what executed, never why: an aggressive buy may be a new position, a hedge against something you cannot see, or an algorithm closing out. Reading intent into prints is the order-flow equivalent of assuming which side of the options book the dealer is on.

Why does agreement beat a stronger single signal?#

Because these two measurements could have disagreed. That is the whole of it, and it is an accounting argument rather than a statistical one.

A signal carries information by ruling things out. Two indicators computed from the same price series rule out nearly the same possibilities, so when they agree you have learned one thing twice — the agreement was guaranteed by construction and tells you nothing beyond what either said alone. Positioning and executed trades are built from different inputs, describe different objects, and point in different directions in time: one is conditional and forward-looking, the other is a completed record. To the extent they really are that loosely coupled — and there is one important case, below, where they are not — the set of explanations surviving their agreement is genuinely smaller than either would have left standing.

This is also why "how many signals agree?" is the wrong question and "how many sources agree?" is the right one. Several lines derived from one snapshot of open interest will agree with each other reliably; several readings of the same tape will too. The informative boundary is the one between the sources, not the count of lines drawn on top of them.

What confluence does not mean#

• It does not date-stamp. Agreement raises the odds a price matters; it says nothing about when price gets there, how long any reaction lasts, or whether it happens today at all.
• It is not a direction. A price both measurements agree on can act as a floor or as a trapdoor, and which one depends on the sign of the hedging requirement, not on the fact of agreement. If dealers are net long gamma there, hedging leans against the move and damps it; if they are net short, hedging runs with the move and accelerates it. Both readings inherit the counterparty assumption above, so the sign is an inference, not an observation.
• It is not a probability. No count of agreeing signals converts into a hit rate. Anyone quoting one owes you the observation window, the sample size, and their definition of "held" — without those three the number means nothing.
• It does not outlive its inputs. Positioning is recomputed as price and open interest move — recalculated sequences, not fixed lines on a chart. A level that had agreement this morning may describe a position that has since been unwound.

Where the independence claim breaks down#

Three failure modes, because a page listing only reasons to trust a method is marketing rather than method.

Fake independence. Signals that look different but share an input. It is the easiest one to fall for, because a crowded chart feels like corroboration. Count the sources.

Shared cause — and the one that undercuts this whole note. Independent inputs are not independent of the world. The easy version is external: a scheduled announcement or a large expiry moves positioning and participation together, so both measurements can be right about their own object and still point at a price for the same passing reason.

The hard version is internal — the fake independence above, wearing better clothes. Near a large strike, part of what prints on the tape is the hedging the gamma level predicts. The obligation is upstream, the executions downstream, links in one causal chain: agreement there is partly one mechanism measured twice, not two witnesses corroborating each other.

Nothing separates the two cleanly, but the character of the flow leans one way, and it is worth looking before calling anything confluence:

Qualitative tendencies, not a test. Real flow at a level is usually a mixture of both, and nothing in the tape labels which print was which; the columns say what to look at, not what to conclude.

Flow that starts at the level and dies with it is the weakest agreement on this page, because it is the kind the level itself produced.

Confirmation by patience. Wait long enough at any price and something will eventually agree with something. Confluence means measurements arriving together; agreement assembled over an afternoon is a story told backwards.

What changes tomorrow#

Before treating agreement as a reason for anything, ask three questions of it: how many independent sources agree, when each was last computed, and — the one that does the most work — what would have counted as disagreement. If nothing on the chart could have disagreed, then nothing agreed. That question costs a few seconds and disqualifies most of what gets called confluence.

Related reading#

• One market, many tapes — why evidence from different venues describes one underlying, including the half of that claim that is usually oversold.
• Before you believe a reversal story — the same demand for a control: agreement compared to what?
• Futures trading hours: one answer, and why you will find three others — why a level formed when almost nobody was trading is a hypothesis, not a finding.