Harry Markowitz's insight got compressed into a slogan: portfolio diversification is the only free lunch in finance. The slogan traveled much further than the fine print.
The fine print is one word — correlation. And correlation is not a feature you buy. It is an output of how holdings behave together, measured after the fact, in whatever regime you happened to be standing in. In 2022 that regime shifted: U.S. equities and investment-grade bonds fell together, and "bonds are the hedge" stopped being a free assumption for anyone who had to explain it to a board.
The scramble that followed produced a thousand third legs. Most of them are a new label on an old engine. This post gives you the filter that tells the difference — four questions you can run Monday morning on any alternatives pitch on your desk, including the one at the end of this page.
portfolio diversification: correlation is an output, not a property
Separate two words most decks blur. A diversifier improves the whole portfolio because its returns are driven by something other than what drives the rest of your book. A hedge is narrower and stricter: it pays when a specified exposure loses. A diversifier that flattered your risk-adjusted returns in one decade is not a hedge, and calling it one is how allocators end up short protection at exactly the wrong moment.
2022 is not proof that diversification is dead, and it is not a supercycle sermon. It was one regime break — one year in which the two engines most books relied on turned out to answer to the same variable, a rapid repricing of real rates. The honest lesson is narrower and more durable than the doom version: most books did not own a diversifier. They owned an inherited correlation regime. The difference is underwriting. A correlation regime is something you observe. A diversifier is something you can explain causally, in advance, without a spreadsheet.
three things "uncorrelated" can mean
When a manager says uncorrelated, they are making one of three very different claims. Only one is a portfolio property.
| the claim | what's actually happening | the tell | behavior under stress |
|---|---|---|---|
| exogenous driver (the real one) | payment is decided by an event outside financial markets — a peril, a statute, a contract, a biological process | they can name the deciding event in one sentence without using the word "uncorrelated" | may hold, if the payor and the wrapper hold |
| unmarked correlation | same driver as the rest of the book, priced less often | measured correlation falls as marking frequency falls; comps are chosen by the manager | reprices late, then all at once |
| crisis-correlated liquidity | diversified until someone is forced to sell | the wrapper promises more liquidity than the underlying asset has | fails precisely when the diversification was needed |
All three can produce a low correlation number. Cliff Asness named the second one volatility laundering — the observed calm is a reporting artifact, not a risk property. The third shows up in gate notices and redemption queues rather than in the correlation table. So the statistic cannot arbitrate between them. Only the causal question can.
the filter: four questions
Here is the tool. It is not sophisticated, which is the point — it works in a meeting, without data, out loud.
| question | what a usable answer sounds like | what disqualifies the pitch |
|---|---|---|
| 1. what is the payoff? | a contracted cash flow, a contingent payment on a defined trigger, or capital appreciation on a real asset — named, not blended | "total return" with no distinction between coupon, mark, and exit |
| 2. what is the causal driver? | the specific event that decides whether the payment arrives: a peril, an offtake, a rate, a harvest, a regulatory obligation, a beneficiary's dependency | "it's private," "it's real assets," "it's alternative" — those are distribution categories, not drivers |
| 3. what is the liquidity, really? | how the position is marked, how often, by whom, and what happens to it in a forced-seller quarter | a redemption promise the underlying asset cannot fund |
| 4. what is the failure mode? | the regime or event that breaks the thesis, named by the seller before you find it | "we don't see a scenario where this doesn't work" |
Two rules make the filter bite. First, the driver has to be nameable without the word "uncorrelated." If the answer to question two is a synonym for question two, there is no answer. Second, whoever cannot name the failure mode is not carrying it — you are.
Three answers should end a meeting: it's uncorrelated because it's private (that's question three failing, dressed as question two), our marks have been stable (that's the tell, not the evidence), and it's a hedge offered without a named risk, horizon, and trigger.
the filter on alternatives you already know
Run it quickly and the field sorts itself, without anyone needing to invent a beta.
Insurance-linked securities — the institutional-scale event-risk market, catastrophe bonds and their cousins — score well on construction: the payoff is contractual, the driver is a named peril that does not consult the equity market, the liquidity is defined, and the failure mode is stated on the cover — a big event, or a model that mispriced it. That is what a genuinely exogenous driver looks like. It is also worth noticing what an event-risk security does not do: a catastrophe bond does not reduce hurricanes. It moves who absorbs the loss. Hold that thought.
Contracted infrastructure often scores well because availability payments are decided by a contract and an operating standard rather than by volume or sentiment. Private credit frequently fails question two — the driver is usually the same credit cycle already in the book — and then fails question three when marks are the actual source of the smoothness. Gold has no answer to question one by design, which is fine if you wanted a monetary hedge and a problem if you were sold a return stream. Farmland and timber score partially: biological growth is a real exogenous driver, but commodity prices, financing conditions, operating skill, and appraisal cadence sit on top of it, and the appraisal cadence is question three wearing a disguise.
None of that requires a track record to evaluate. That is the value of a causal filter: it works on instruments that have no history yet, which is exactly when allocators are most exposed to a good story.
nature is not first a diversifier
Now the part that is easy to get backwards, including in our own materials.
Diversification assumes failures are independent. It breaks when holdings share a hidden common cause — one physical dependency underneath positions that look unrelated on a factor sheet. The World Economic Forum's Nature Risk Rising put roughly $44 trillion of global GDP as moderately or highly dependent on nature. For an owner of a broad book, that is not an ESG observation. It is a statement about correlation you cannot diversify away by adding a ticker, because the dependency is upstream of every ticker.
Nature enters a portfolio first as a common cause, not as a diversifier. The first honest role is risk reduction: an allocation that lowers the physical loss sitting under the assets you already hold. That is a different line on the term sheet than return diversification, and the two should never be summed into one claim. The dependency argument is worked through in portfolio ensurance and you can't sell your ecosystem exposure.
when nature becomes a diversifier
Nature becomes a candidate diversifier under two conditions, and both have to hold.
The driver has to be ecological. Sun, rain, biological growth, flood attenuation, aquifer recharge, pollination, and a beneficiary's physical dependency on those functions are not decided by the equity market. That satisfies question two — as candidacy, not as proof.
The coupon has to be contracted. An intact wetland that prevents flood losses is economically real and is not income. Without an underwritten payor — a utility, a municipality, an insurer, a corporate beneficiary, or a foundation deploying program-related capital — there is no answer to question one. A protocol balance sheet can sit in that seat only as a reflexive payor: primary sales and trading activity the protocol routes as proceeds, not an independent third-party obligation. Value without a payment is a dependency, not a position. Do not score the ecological driver (question two) unless this payor exists. Sun and rain are a candidate cause. They are not a diversifier credit until someone is contracted to pay.
This is where the two versions of nature investing separate sharply, and the separation matters more than any sleeve name. Working nature — farmland, timber, water-linked assets — already holds a seat in institutional books as real-asset beta, with decades of return history and its own well-documented risks. Pure protection does not inherit that history. A protected wetland is not farmland with a nicer story, and citing farmland performance to support a protection allocation is a question-two failure that happens to be flattering. For what nature can and cannot claim by hunt, see the five things allocators hunt; for the class-formation question, is nature an asset class yet?
running the filter on ensurance, in public
A filter that exempts its author is marketing. So here is the same card, applied to us.
Quick gloss: ensurance is a protocol for funding protection of natural assets before loss occurs, rather than compensating after it. Two instruments carry capital — coins (general ensurance: fungible, protocol-wide funding and discovery) and certificates (specific ensurance: one certificate per agent).
A certificate holder holds a recorded, transferable claim on the protection funding of that agent's named place, people, or purpose. It is not title, not residual equity, and not an indemnity. Coins are not a portfolio diversifier position — they are the protocol-wide funding and discovery layer, and their price is reflexive to trading. Run the four questions on a named certificate, not on a coin.
| question | honest answer | where it is weak |
|---|---|---|
| payoff | a contracted protection premium where a payor is underwritten; otherwise funded protection with no coupon | most protection today is the second case. No premium, no income claim |
| causal driver | ecological function plus a beneficiary's dependency — only if question one already has a payor | without a payor this is candidacy, not a scored driver. Do not award Q2 with no Q1 |
| liquidity | certificates are hold-side by design; coins trade and provide edge liquidity and discovery | they are different claims, not interchangeable ones. Neither is the land, and a certificate is not residual title |
| failure mode | no payor, weak measurement, or a wrapper that promises liquidity the underlying work cannot fund | also: an ecological shock can damage a natural asset and conventional holdings in the same quarter |
You are probably thinking: an instrument with no track record has no business being granted a portfolio role. That is the correct instinct, and it is the reason a causal filter exists rather than a performance table. The filter tells you what would have to be true — a named payor, a measured service, a term, a wrapper honest about its own liquidity — and lets you check each one on a specific asset instead of taking a category on faith. Where those are not yet true, the honest label is risk reduction, not diversification. Nothing about being early changes that.
Three things we will not claim, because the filter forbids them.
This is not a catastrophe bond. Event-risk securities pay after an independent peril and are collateralized against it. Ecological degradation is gradual, correlated across regions, and often irreversible — the opposite risk shape. The strongest honest crisis claim is regime resilience: protection may matter more as water stress, heat, insurance withdrawal, and supply disruption make old assumptions weaker. Regime resilience is not a positive return in a market drawdown, and anyone selling it as convexity is selling a lottery ticket that was never printed.
RealValue is measurement, not correlation. RealValue is our natural capital accounting engine — ecosystem service value per acre per year measured against ecological condition across fifteen stocks and nineteen service flows. It makes the condition of a place legible to capital, which is the precondition for underwriting a premium. It does not establish a beta, a correlation, or a return history, and no amount of measurement substitutes for a payor.
The chain is settlement, not the thesis. Strip Base, the coins, and the certificates out of this argument and it survives intact: living systems are a candidate exogenous driver, a contract is what makes the driver holdable, and protection reduces a common-cause loss whether or not it is recorded onchain. Onchain infrastructure makes claims, proceeds routing, and provenance verifiable and composable. It does not create diversification, and any post that implies otherwise has failed its own question two.
two jobs, two lines of the term sheet
Keep these apart in every memo, because merging them is the most common way honest people oversell nature.
The first job is portfolio return improvement — a stream whose driver is decided outside markets, which requires a contract and a payor. The second job is reducing the common physical dependency beneath the book — which does not require a coupon at all, and which a catastrophe bond, a trend program, and a put option cannot do at any price. Most hedges never touch the loss; they rearrange who eats it. Funding protection changes the odds of the loss occurring.
An allocator can honestly want the second job while the first is still under construction. What is not honest is charging the second job's cost to the first job's return line.
This is educational content on portfolio construction and natural capital — not investment advice, not an offer or solicitation of any instrument, and not a recommendation to buy or sell. Ensurance is not insurance: a certificate is not an indemnity and does not promise to make a holder whole after a loss. Certificates and coins described here are not offered as investments on this page.
what to read next
If the filter was the useful part, take it and go. Nothing here needs a decision.
- the position that funds the thing that shrinks the loss — when the test is done and you want the holdable claim
- a hedge against what? — hedge, diversifier, third leg, and protection are four different jobs with four different tests
- your third leg needs a different engine — which economic event decides whether each candidate pays
- five tests for an alternative that actually diversifies — the full diligence scorecard, applied to familiar alternatives
- the five things allocators hunt — and where nature actually fits — what nature may and may not claim, by hunt
- portfolio ensurance — the common-cause dependency argument in full
- divest, reprice, or protect — the three doors once the exposure is named
To see what the instruments actually look like rather than read about them, general ensurance is the open surface, and ensurance.app is the protocol overview.