Confluence: When Four Reasons Are Really One
Confluence is several separate reasons pointing at the same price level. It is only worth anything when the reasons are independent of one another, and most commonly stacked indicators measure the same underlying thing, which means they add confidence without adding evidence.
How it works
Confluence means several factors agreeing on one price. A moving average sitting where a horizontal level sits where a Fibonacci retracement lands. Three reasons instead of one, and the intuition is that three reasons are stronger.
The intuition is right, with one condition attached, and the condition is the whole subject. Three reasons are stronger if they are independent — if each one could have been wrong without the others being wrong.
Most stacked reasons are not independent. They are the same measurement in different clothes, and the resulting confidence is manufactured rather than earned.
The test that separates the two
Ask one question of every item in a stack: could this have said something different while the others said what they said?
If yes, it is independent and it adds information. If no, it was always going to agree, and its agreement tells you nothing you did not already have.
Work through the common stacks with that test.
Two moving averages of similar length are one observation. A 20-period and a 21-period average of the same closes will agree on essentially everything. Even a 20 and a 50 are computed from overlapping data and move together most of the time. Stacking them is averaging the same numbers twice.
Most oscillators are the same measurement. The do indicators work page measures this directly on the site’s shared history: the relative strength index (RSI) and the moving average convergence divergence (MACD) histogram, computed on the same closes, correlate strongly. When two tools agree, they are frequently agreeing because they were never capable of disagreeing.
The smart-money stack has the same problem in a different vocabulary. An order block, a fair value gap, an imbalance and a mitigation block are frequently drawn across the same three candles. Four names, one region, and calling that four-factor confluence is counting a single observation four times.
What is actually independent
Volume is a different measurement. It counts what traded rather than where price went, so it can disagree with everything on the price side — and a disagreement is only possible when the information is genuinely separate.
A level from a different timeframe is largely independent. A daily support level and a five-minute structure point are derived from different data at different scales, and one can hold while the other fails.
A short list of genuinely independent inputs, then: price structure, volume, timeframe, and anything from outside the chart entirely — a scheduled event, a sector move, a correlated instrument. That is roughly it. Everything else on a typical chart is a transformation of the same price series.
Which caps how much confluence is available. With four genuinely independent sources, four is the maximum, and a stack of nine factors necessarily contains repetition.
In practice: the cost nobody counts
Requiring more agreement makes setups rarer, and rarity has a price that never appears in a trading journal. The trades filtered out do not get recorded, so a stricter filter always looks better in review than it was — the losses it avoided are visible and the winners it removed are not.
Every trade taken still costs 2% of a typical bar’s range in round-trip costs on this site’s shared history, so a filter that halves your trade count halves that bill too. That is a real saving and it has to be weighed against what was filtered out, which requires actually logging the setups you declined.
The failure mode that costs most is motivated stacking. You want to take a trade. You look for reasons. Reasons are always available, because any chart contains a moving average, a prior level, a Fibonacci line and a round number within a short distance of any price. The stack assembles itself around a decision that was already made.
The defence is fixing the list in advance. Decide which factors count, in writing, before looking at a chart. Anything discovered afterwards is a rationalisation regardless of how true it is.
And none of it survives a gap. Six factors agreeing at a level are six factors that are irrelevant when price opens past it. Confluence is a statement about where to act, never about what will happen.
What confluence is not
It is not additive probability. Two correlated signals agreeing does not multiply anything. The maths that would justify treating agreement as compounding evidence requires independence, which is exactly what is usually missing.
It is not a substitute for a reason. A level with five indicators on it and no thesis about why price should react there is a decorated guess.
It is not proportional to effort. Adding a sixth tool to a chart feels like work and adds nothing if the sixth tool is derived from the same closes as the first five.
And it is not the same as agreement across participants. A level everyone is watching does behave differently — that is a real, self-fulfilling effect — but that comes from the level being obvious, not from your having five reasons for it.
When it fails
A range is where confluence is most abundant and least useful. Price oscillates through a narrow band, so every moving average, every prior level and every retracement sits close to every other one. The stack looks strongest exactly where the information content is lowest.
The second failure is the confidence transfer. Five agreeing factors produce a larger position than one factor would, and if the five were really one, the position size has been set by an illusion. That is how a correlated stack turns a small error into a large loss.
A third is not counting what was filtered. Without logging declined setups, there is no way to know whether the filter improved results or merely reduced them.
A fourth is treating round numbers as a factor. They are genuinely watched, so they are not nothing — but they are available everywhere, which makes them the easiest item to add to any stack.
And a fifth is stacking across timeframes without checking direction. A daily level and a five-minute level agreeing on a price is confluence. A daily uptrend and a five-minute downtrend is a conflict, and calling it confluence because both are “on the chart” is reading agreement into disagreement.
The original data
Of the 24,971 videos measured for this site, confluence appears as a supporting idea in a great deal of content and almost never as the subject — which is how a concept that needs a correlation check became a word that means “several things I like.”
The measured claim this page rests on is on the do indicators work page: RSI and the MACD histogram, computed on the same closes on this site’s shared history, correlate strongly enough that their agreement is close to automatic. Run that check on your own stack — two indicators, same data, one correlation coefficient — and you will usually find that what you have been calling four reasons is one or two. It takes an afternoon and it permanently changes how much a stack is worth to you.
Related
Choosing indicators is the measured version of this question. Do indicators work carries the correlation figures. And support and resistance is the level type most stacks are built around.
The chart I was proudest of had six things agreeing on one level and I could not have told you which of the six was doing any work. When I finally checked, four of them were derived from the same twenty closes, and I had been treating one measurement as four confirmations for about two years.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.