# 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. - Canonical: https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/ - Published: 2026-08-03 - Author: Hadal Research - Answers the question: "why do correlated pairs decouple on low timeframes" - Coins the term: **The Coherence Floor** — The sampling interval below which two instruments' relationship stops being defined: finer than the floor, most of what a chart shows is asynchrony — prints that never co-occurred — and a reading taken there is about the sampling, not the relationship. --- ## 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](/glossary/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](/glossary/regime-shift/) entry covers the genuine article; your [data's provenance](/glossary/tick-data-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](/instruments/microstructure-dashboard/) is where co-print behaviour lives in the instrument line, and [Data Forensics](/instruments/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](/lexicon/) — the floor, like every term this research coins, is defined once and pinned to the argument that earned it. --- ## Claims examined ### Claim 01 — canonical: https://hadalinstruments.com/refutations/#claim-5da1b845 > "When ES and NQ decouple on the one-minute chart, that's smart money showing its hand." — our reading: Misleading 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. **What is true:** Fine-grain divergence between correlated instruments is the expected output of non-synchronous trading; a divergence is evidence of something unusual only after it exceeds what asynchrony alone produces at that interval, which is a measurable threshold. Evidence: https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-5da1b845 ### Claim 02 — canonical: https://hadalinstruments.com/refutations/#claim-658b74b5 > "These pairs are ninety percent correlated, so they move together on every timeframe." — our reading: False 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. **What is true:** Correlation is interval-specific: the daily number describes daily co-movement and says progressively less as the observation grain gets finer, so any correlation figure is meaningful only with its sampling interval attached. Evidence: https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-658b74b5 ### Claim 03 — canonical: https://hadalinstruments.com/refutations/#claim-139b3d14 > "The correlation broke down, so the relationship between the pairs has ended." — our reading: Misleading 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. **What is true:** A real relationship change moves the coarse-grain correlation; a fine-grain reading that looks broken while the daily number holds is the measurement dissolving at the floor, and the two cases are separated by checking both grains. Evidence: https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-139b3d14 ## Cite This Article APA BibTeX HTML Hadal Research. (2026). Why do correlated pairs decouple on low timeframes?. Hadal Research. https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/ Version e472963, 2026-09-14. @misc{hadal_2026_why-do-correlated-pairs-decouple-on-low-timeframes, author = {Hadal Research}, title = {Why do correlated pairs decouple on low timeframes?}, year = {2026}, url = {https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/}, howpublished = {Hadal Research}, version = {e472963}, note = {Published: 2026-08-03; version dated 2026-09-14} } Source: Hadal Research, Why do correlated pairs decouple on low timeframes?. Original Research Copy Citation **Version e472963** identifies the commit that last changed this page in Hadal's content repository. That repository is not public, so the identifier does not resolve externally — it is published so a citation pins one specific state rather than a moving page. To obtain the exact version cited, use the [press and research route](https://hadalinstruments.com/press/). ## Explore further ### Instruments - [Data Forensics](https://hadalinstruments.com/instruments/data-forensics/) - [Microstructure Dashboard](https://hadalinstruments.com/instruments/microstructure-dashboard/) ### Concepts - [Epps Effect](https://hadalinstruments.com/glossary/epps-effect/) - [Regime Shift](https://hadalinstruments.com/glossary/regime-shift/) - [Tick Data Provenance](https://hadalinstruments.com/glossary/tick-data-provenance/) ### Indicators whose taught claim it examines - [SMT Divergence Indicator](https://hadalinstruments.com/ict/smt-divergence-indicator/) ### Research - [Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/) Asked as: is my volatility indicator just measuring time of day [All Hadal research](https://hadalinstruments.com/research/)[This article as plain markdown](https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes.md) --- ## Raw artifact — NOT PUBLISHED FOR THIS PAGE No downloadable artifact ships with this page. Eight published measurements do, each content-hashed so a reader can verify the figures independently. Where a measurement is published here without one, that is a gap rather than a policy, and it is stated rather than left to be noticed. [Measurements that ship their data](https://hadalinstruments.com/research/)