Carbon has no actuarial table. Someone has to build it.

Every insurance market begins the same way. Capacity arrives before data, premiums are set by conviction, and the first carrier to build a real loss history quietly starts pricing everyone else's book. Fire insurance was written for decades before anyone had credible fire statistics. Carbon insurance is at that exact moment now: cover is scaling, tenors are stretching toward ten years, capacity is multiplying, and the loss history underneath it all is a few years old and thin. This is not a criticism of the insurers. It is a description of where the value is about to concentrate.

What is actually being underwritten

01 · Four perils, and they travel together

Strip the product names away and carbon cover is exposure to four perils. The physical peril: the forest burns, the drought bites, the stored tonne comes back, which the market calls reversal. The paper peril: a methodology ruling, an eligibility delisting or an integrity reassessment reprices an entire cohort overnight, with no watchlist and no grace period. The performance peril: a project delivers fewer tonnes than the curve it was contracted on, year after year, in a forward market now worth multiples of the spot market. And the counterparty peril: the developer behind a decade-long delivery promise is often a young company in a hard place.

The uncomfortable actuarial fact is that these perils are correlated. Credits fail together. A drought year drives fire across whole regions at once. A single methodology decision moves every project issued under it. A book that looks diversified by project count can be concentrated to a handful of shared factors, and a buffer pool calibrated to average years can be badly short in a correlated one.

02 · The inputs an underwriter is offered today

Ask what sits behind a carbon quote today and the answer is a patchwork. A project rating, which is an independent opinion at a point in time, ordinal rather than probabilistic, and not designed to be a frequency or a severity. Registry data, which proves issuance and custody and says nothing about what has happened at the site since. And the developer's own documentation, which is the insured party's account of the risk being insured. Each input is legitimate. None of them answers the underwriter's actual question, which is a frequency and severity question: how often does this class of risk fail, how badly, and how correlated is it with the rest of my book.

The underwriter's question is a frequency and severity question. Nobody in the carbon market currently sells the answer.

What an actuarial table for carbon looks like

03 · Observation, record, calibration

An actuarial table is not a clever model. It is a discipline applied over time, and for carbon it has three parts. First, continuous observation of the physical asset itself: every covered project watched on the cadence of the satellites, so that reversal, degradation and underdelivery are events with dates and evidence rather than surprises at the next verification. Second, a risk-event record: every reversal, every reclassification, every delivery shortfall, every near-miss, captured consistently across projects, methods, geographies and years, in a form that can be interrogated the way a loss triangle is interrogated. Third, calibration: the predictions the engine made, kept honestly against what the world then did, published as a record rather than asserted as a claim. A stated probability has to mean what it says, and an underwriter is entitled to see the receipts.

04 · Independence is not optional

There is a structural requirement underneath all three: the read has to come from a party with no position in the transaction. The developer's monitoring is the insured describing its own risk. The broker's view is attached to the placement. A read produced by the party selling the credit, or paid by it, cannot anchor a premium, a reserve or a claim determination, for the same reason an issuer-paid grade could not anchor a mortgage book. The reading party must hold no title, sell no credits, and be paid by the capital that carries the risk. That is not a marketing stance. It is what makes the number usable inside an underwriting file.

Where this goes

The carbon insurance market is forecast to grow by an order of magnitude as compliance regimes harden and forward books professionalise. Every policy written between now and then either contributes to someone's loss history or it does not. The carriers that underwrite on an independent, calibrated, continuously observed read will select better risks, price them tighter, and hold the record that the rest of the market eventually has to buy. The ones that do not will discover their book's correlation structure the way markets usually do.