Published as a TradingView Idea on ETHUSDT. View the chart on TradingView
Traders stack indicators because one tool can be wrong and four agreeing feels harder to argue with. We tested that belief inside a single indicator once before, and the shuffled control won.
There was a reasonable objection to that result. The systems inside one indicator are all reading the same price with related maths, so their agreement may be arithmetic rather than insight. The stronger claim is agreement between genuinely different tools.
This is that test.
What agreement means, precisely
Reduce every tool to one thing per bar: bullish, bearish, or neither. Then on any bar you can count how many separate tools point each way, and subtract.
The belief being tested is that the bars where many tools agree are better bars, in the sense that price subsequently does more of what they were pointing at.
The mistake that makes this test lie to you
A single indicator can fire several alerts on the same bar. One tool producing three alerts is not three-way confirmation, but if you count alerts instead of tools, that is exactly what your data will say.
This is the easiest way to make an agreement test produce the answer you were hoping for, and it is worth knowing before you run one.
The second mistake, which is subtler
Picking the best-looking bucket.
With nine agreement levels, four forward horizons and nine markets, something will look excellent by chance. If the claim is that more agreement is better, then the prediction is not that one bucket excels. It is that the whole ladder is ordered. So that is what gets tested, and each bucket is scored against how that market drifts anyway.
What we found
Thirty six ladders. Twenty four of them sloped the right way, 67% against a coin flip’s 50%, with a mean correlation of +0.237.
That is where most write-ups would stop.
Why we did not stop there
The thirty six are not thirty six independent tests. They are three markets by three timeframes by four horizons, and those four horizons measure overlapping windows on the same bars. The real count is nearer nine, and six of nine is not a finding.
Every strong result is on the daily and weekly. Not one is on the 4H. Which matters, because intraday is where anyone would actually use this.
The strongest ladder is not ordered. Three quarters of its sample sits in the bucket where nothing agrees. The buckets that carry the claim contain between ten and sixty nine bars. And one of them points the wrong way, in the middle.
| count | |
|---|---|
| bars at zero agreement | 754 of 993 |
| bars in the buckets that decide it | 10 to 69 |
| the eye-catching +11.7% weekly bucket | 12 bars |
How rare is real agreement
Across 1,063 bars that carried any event at all, five or more tools agreeing happened 45 times. About 4%.
That is worth sitting with, because agreement is spoken about as though it were a regular occurrence you can wait for. On this data it is a one-in-twenty-five event, and on the timeframe most people trade it did not predict anything.
What we did with the result
Nothing, which is the point.
Two places in our own system called for a confluence reader that was never written. Both now sit behind a check that keeps them idle. This measurement is the reason we did not write one.
The only thing it opened is a better question than the one it was built to answer: our own live setups require four of six tools agreeing before anything is pushed. That threshold was chosen, not measured. The same tool can measure it.
Published as a TradingView Idea on ETHUSDT. View the chart on TradingView
Past chart behaviour and measurement, described after the fact. Nothing here is a recommendation or an indication of future results.