Why do correlated pairs decouple on low timeframes?
Asked as: why do correlated pairs decouple on low timeframes
Below a measurable interval, two instruments barely share prints — most fine-grain decoupling is the measurement dissolving, not the relationship breaking.
The short answer
Because below a certain sampling interval, the two instruments barely share prints — and a correlation computed where prints don’t overlap is mostly measuring the sampling, not the relationship. I call that interval the coherence floor: above it, co-movement is observable and a divergence means something; below it, most of what a chart shows is asynchrony — returns from one instrument compared against interpolated silence from the other. The academic record for this is the Epps effect, documented in 1979 and replicated ever since, and it is almost entirely absent from the way retail analysis talks about correlated pairs.
The practical inversion is the point: fine-grain decoupling is the default state of the measurement, not an event in the market. The event would be coherence down there.
WHAT THIS IS — AND WHAT IS NOT PUBLISHED. This article is method resting on published academic literature, cited above. No measured dissolution curves for specific pairs are published here — my own measurements of where the floor sits, per pair and per venue, publish under the methodology’s rules when they publish, with their artifacts. Status of any measured claim: NOT YET COMPUTED.
Prerequisite Knowledge
You need price data you can resample — any platform’s export of two instruments you believe are correlated, at the finest resolution it offers, over the same window. The method below is a resampling exercise in a spreadsheet or a few lines of any language; no special access, no indicator packages.
What a correlation number actually claims
A correlation is not a property of two instruments; it is a property of two instruments at a sampling interval, over a window. The daily figure answers “did their daily closes move together this quarter?” The one-minute figure answers a different question — and mostly fails to answer it, because a one-minute bar frequently contains a print from one side and nothing from the other. The formula doesn’t know that: it compares a real move against a flat interpolation and reads independence. Stack thousands of those comparisons and the measured correlation slides toward zero as the interval shrinks — smoothly, systematically, and without anything happening in the market at all.
That slide has a name and a literature. Epps documented it in equities in 1979; it holds in FX, futures and crypto; and the fix the literature built — estimators that only compare genuinely overlapping observations — is itself the confession of what the naive number does below the floor.
The floor, and what lives below it
The coherence floor is where the attenuation takes over: the interval below which the majority of your reading is asynchrony.
It is not one universal number — it depends on how actively both instruments print, on the venue’s feed behaviour, and on the window — which is exactly why it is a measurable property rather than a rule of thumb. Below the floor live most of the dramatic intraday stories: the “decoupling” of index futures on the one-minute chart, the pair that “broke correlation” for twenty minutes, the divergence read as a hidden hand. Some of those stories may even be true — but the honest sequence is to measure what asynchrony alone produces at that grain first, because that is the null the story has to beat.
How to find your own floor in ten minutes
Take both instruments’ data over the same window. Compute the correlation at a ladder of intervals — one minute, five, fifteen, sixty, daily. Plot the ladder. The curve you get — high at coarse grains, dissolving as you descend — is the dissolution profile of that pair on that data, and the knee of the curve is your working floor.
Then apply the one rule that separates measurement from narrative: a fine-grain divergence is only an event if the coarse-grain relationship moved too. If the daily correlation is intact, you’re below the floor, watching fog. The regime-shift entry covers the genuine article; your data’s provenance bounds what any of it can mean, since feed staleness and aggregation add artificial asynchrony on top of the real kind.
The Observable Mechanism
Everything here is computable from data you already hold, and the central claim is falsifiable in a spreadsheet: if the attenuation were not real, your resampling ladder would show flat correlation across intervals. The literature has said it won’t since 1979 — but you don’t have to take the literature’s word for it any more than mine.
What This Does Not Establish (The Limits)
This article establishes why fine-grain correlation readings understate and where the boundary concept comes from — not where the floor sits for any specific pair, venue or window, which is a measurement this page deliberately does not fake. It also does not establish that no fine-grain divergence is ever informative: lead-lag structure below the floor is a real research subject with its own estimators — the claim is only that the naive reading is dominated by asynchrony, and that any story built below the floor owes the asynchrony null an answer first.
Where this leads
The measured version of everything above — where the floor actually sits per pair, how retention decays by grain, what your own venue’s feed adds to it — is measurement work this house does under its published method, and its results publish with their artifacts or not at all. The Microstructure Dashboard is where co-print behaviour lives in the instrument line, and Data Forensics is where a file’s fitness to carry such measurements gets established. If the vocabulary here is useful before any of that ships, it is in the lexicon — the floor, like every term this research coins, is defined once and pinned to the argument that earned it.
Claims examined
Claim 01§ claim-5da1b845
When ES and NQ decouple on the one-minute chart, that's smart money showing its hand.
The observation is real — at fine grains, correlated instruments visibly disagree — but the conclusion skips the boring explanation that accounts for most of it. Two instruments almost never print at the same instant, and below a measurable interval the correlation between them is mostly undefined: the divergence you see is largely asynchrony, a measurement effect documented since 1979, before any story about intent is needed. Whether a specific divergence exceeds what asynchrony produces is a measurable question — and a manipulation reading taken without that measurement is a narrative wearing a chart.
Claim 02§ claim-658b74b5
These pairs are ninety percent correlated, so they move together on every timeframe.
A correlation number is bound to the sampling interval it was computed at, and the same pair over the same window produces materially different numbers at different grains — systematically lower as the grain gets finer, a finding replicated across equities, FX and futures since 1979. The tools display this without explaining it: the hourly tab and the daily tab of the same correlation matrix disagree, and neither is wrong — they are answers to different questions. A single correlation figure quoted without its interval is not yet a fact.
Claim 03§ claim-139b3d14
The correlation broke down, so the relationship between the pairs has ended.
Two different phenomena share the word 'breakdown'. A relationship genuinely changing — a regime shift — shows up at grains where the correlation is well-measured: the daily and hourly numbers move. Sampling attenuation shows up only below the coherence floor: the fine-grain number was always low, and looking there mistakes the permanent fog for a fresh event. The check is one comparison: if the coarse-grain correlation is intact while the fine-grain reading looks broken, you have walked below the floor, not witnessed a divorce.
Each claim above has a permanent address — the § link — whose canonical home is the refutation index, where it carries its variant phrasings and the true proposition stated on its own feet; this article is the evidence behind it. If a claim's text ever changes, it becomes a new claim at a new address, and the old one stops resolving rather than silently meaning something else.
Explore further
Instruments
Indicators whose taught claim it examines
Research
- Is my volatility regime just telling me the time?Asked as:
is my volatility indicator just measuring time of day