--- title: what asset correlation actually measures canonical_url: https://ensurance.app/guide/what-asset-correlation-actually-measures markdown_url: https://ensurance.app/guide/what-asset-correlation-actually-measures.md subtitle: a number can size a sleeve. it cannot deliver the water category: nature-finance --- # what asset correlation actually measures *a number can size a sleeve. it cannot deliver the water* Asset correlation is the cheapest risk report in the building and the most confidently misread. Every cell in the matrix answers one narrow question: over the window someone chose, did these two return streams tend to move in the same direction at the same time? Now set that report beside a physical fact. West of Phoenix, at Tonopah, Arizona, the Palo Verde Generating Station cools itself with treated municipal effluent. There is no river, lake, or ocean at the fence line. The cooling makeup arrives by pipeline, under an agreement in which the cities sell up to 80,000 acre-feet a year, running through December 31, 2050. Upstream of that effluent sits a municipal supply blend: the Salt and Verde rivers, Colorado River water delivered through the Central Arizona Project, and groundwater. Correlation, alpha, and a hedge are statements about return streams. The cooling water a desert plant buys under contract, the rivers the cities drink, and the aquifer under that valley exist whether or not those streams move together. Ensurance is how that living system gets funded. It is not the statistic. :::johnson **a correlation is a number. it is not a river.** — The matrix can tell you how two price series behaved together. It has no cell for the water one of them runs on. ::: ## what the number actually measures ### what is asset correlation? **Asset correlation** is the degree to which two assets' returns moved together over a chosen period, expressed on a scale from -1 to +1. At +1 they moved in the same direction every period. At 0 there is no linear relationship in that sample. At -1 they moved in opposite directions. It is a statement about direction of co-movement, not about magnitude, and not about why. Three choices are doing the work inside that definition: the return series you use (daily, weekly, monthly), the window you measure over (three years, ten years, since inception), and the pair you decided to compare. Change any one of them and the number changes. A correlation is therefore not a property of an asset. It is a measurement of two assets, under a method, across a stretch of time that has already happened. That is also why it earns its place in a risk meeting. Sizing a new sleeve, checking whether four managers are secretly running the same trade, stress-testing a book where everything is supposed to be diversified — the matrix is the right first tool for all of it, and it is fast. Keep it. Just keep it inside its job description. ### does correlation mean causation? No, and the failure runs in both directions. The second direction is the one that gets skipped. The familiar version: two series co-move with no causal link, or both respond to a third thing neither of them touches. Every analyst has been warned about it. The expensive version: a hard causal dependency produces almost no correlation. If a relationship is nonlinear, lagged, or binds only under stress, a coefficient computed over a decade of ordinary months reports something close to nothing. A generating station that needs makeup water every hour of every day has an absolute dependency on that water. The monthly correlation between its owner's equity and anything water-related will not show that dependency across a stretch of ordinary months, because the dependency was never a return stream. A stress window can shove unrelated prices together. It still does not measure the pipe. ### why does correlation change? Because a coefficient is conditional on the regime it was measured in, and regimes are not permanent. Liquidity changes, ownership changes, and the reason two things moved together in 2019 may be gone by 2026 — while a number computed across both periods averages the two states into a figure that describes neither. A published August 2026 note on bitcoin's portfolio case makes exactly this point about its own subject, describing the asset's relationship to risk assets as episodic — a "dual personality" that surfaces in some regimes and recedes in others — rather than a low number to be booked once and carried forward. Take that seriously; it is the honest reading of any coefficient. The sizing argument that follows from it is a separate question, and we took it up in [a bitcoin sleeve is not a supplier](/guide/a-bitcoin-sleeve-is-not-a-supplier?from=guide). This post is about what the coefficient measures, so we will not quote one here. ## three objects that keep getting confused A portfolio conversation tends to slide between three things as if they were interchangeable. They are not. Each has a real job, and each is silent about the others. | the object | what it can do | what it cannot do | |---|---|---| | a price correlation | describe how two return streams co-moved in a chosen window; inform position sizing and concentration checks | explain why they moved, promise the relationship holds, or register a dependency that never prints as a return | | a named hedge | pay when a specified risk hits over a specified horizon, at a cost you agreed to carry | pay for a risk nobody named, or deliver a physical input such as water, fuel, or firm power | | a shared physical cause | fail several holdings at once, through one event, at one location, on nature's schedule rather than the market's | show up as a cell in the matrix, or be sized out of the book by rebalancing | Read the right-hand column downward. The first two rows describe instruments that are honest about their own job and mute about the layer beneath them. The third row is not an instrument at all. It is a condition of the world that the first two assume. ### what can a correlation not tell you? It cannot tell you why two things moved. It cannot tell you whether they will move that way again. It cannot see a relationship that is real but nonlinear or episodic. And it cannot tell you what your holdings physically require in order to keep operating — which river, which aquifer, which transmission node, which growing season, at which location. That last gap is the one worth an hour of a risk committee's time, because it is where two names can share a cause without ever sharing a coefficient. Same watershed. Same snowpack. Same effluent pipe. ## the cause that is not in the matrix Back to the desert, because the specifics are the argument. Treated effluent is used water, so day to day it is more drought-insulated than a river intake at the fence. The cities' drinking supply is still a blend, and the 2010 agreement names when they may refuse delivery: a critical domestic need, other sources above the committed quantity exhausted, conservation steps taken, and notice, after they weigh the need for energy (Section 25 of the effluent agreement). Fire on the Salt and Verde changes the timing of those rivers and the sediment that comes with them. That is a condition of the watershed, not a cell in the matrix. Salt River Project already funds forest work on the Salt and Verde, because its own water and power depend on runoff and sediment. The work is real. No cell in the correlation matrix moves when it happens, or when it stops. So the exposure exists, the remedy exists, and the measurement system used to govern portfolio risk has no field for either one. That is not a flaw in the coefficient. It is a category error about what the coefficient was ever going to cover. Price is a bridge to that work, not a description of it. ## what this is, and what it is not Only now is the instrument worth naming. In our terms, a **specific certificate** is the record of a hold on one named natural asset. Proceeds routed to that place's account fund work on the system. It is not a water right, and it is not a share of anyone's acre-feet. What reporting one looks like is a later post. Here the point is what the matrix cannot say. Now the limits, because they matter more than the pitch. A certificate is not uncorrelated, is not alpha, and does not hedge a token, an equity, or a dry year. Ecological shocks can hit the living system and the operating names at the same time — that is the point of a shared cause, and it cuts both ways. We do not run a correlation model, an alpha product, or a hedge fund. Specific certificates are live and our volumes are small. Certificates are not structured or offered as securities, nothing here is an offer to sell one, and nothing here is investment advice. What a certificate does is pay for work on a system that other instruments assume. That is a different sentence from anything the matrix can say, which is why it belongs in a different column. ## frequently asked questions ### how long should the window be for an asset correlation? Longer than one. Report a full-sample figure alongside sub-periods and at least one stress window, then look at how far apart they are. The spread between those numbers is more informative than any single one of them, because it tells you how regime-dependent the relationship is. A committee that receives one coefficient has received an average of states it was not shown. ### if a dependency is not in the matrix, how would a desk find it? Start from the operating side rather than the price side. For one holding, list the physical inputs that have a location — process or cooling water, firm power at a constrained node, a specific port, a specific aquifer — then name the system that supplies each. Run the same list across the book and look for repeats. The repeats are the shared causes. ### does holding a certificate lower a portfolio's correlation? We will not make that claim. There is no coefficient, no backtest, and no risk model behind it, and anyone offering you one this early is selling a curve fit. The honest framing is narrower: a certificate funds a named living system that specific holdings depend on. That is an exposure being addressed at its source, not a statistic being improved. ## sources [Re-Underwriting Bitcoin: Still a Portfolio Diversifier After the Pullback?](https://www.blackrock.com/us/individual/literature/whitepaper/re-underwriting-bitcoin.pdf) — BlackRock, page dated August 17, 2026. Used here only for its characterization of bitcoin's correlation to risk assets as episodic rather than fixed. Performance shown in that paper is hypothetical and computed retroactively; past performance does not guarantee future results. [Arizona Public Service Company Form 8-K, April 23, 2010](https://www.sec.gov/Archives/edgar/data/7286/000095012310038881/c99652e8vk.htm) — Palo Verde effluent purchase terms: up to 80,000 acre-feet per year through December 31, 2050. Section 25 names when the cities may refuse delivery. Salt River Project — published watershed and forest restoration program materials covering the Salt and Verde watersheds. ## take the next step [see what a named dependency looks like on an investor's page →](https://ensurance.app/solutions/investors?from=guide&topic=correlation-not-the-river) ## the series 1. [what asset correlation actually measures](/guide/what-asset-correlation-actually-measures?from=guide) — this post 2. [alpha is the residual, not the river](/guide/alpha-is-the-residual-not-the-river?from=guide) 3. [a hedge has to name the risk](/guide/a-hedge-has-to-name-the-risk?from=guide) 4. [an alt bucket is not a driver](/guide/an-alt-bucket-is-not-a-driver?from=guide) 5. [the correlation outside the matrix](/guide/the-correlation-outside-the-matrix?from=guide)