Four conditions before you price a counterfactual.

There is a proposal circulating in sustainable finance that banks should price the loss a borrower's investment avoids, and let that number cut the borrower's cost of credit. The intention is good and the design flaw it addresses is real. The mechanism does not work, and understanding precisely why it does not work produces a test that is useful far beyond adaptation lending, including for anyone building risk analytics in carbon.

The problem it is trying to solve is genuine

A company that builds a flood defence takes on debt, worsens its leverage ratios, and can end up with a worse credit assessment for having become a safer borrower. The Cambridge Institute for Sustainability Leadership calls this the revenue paradox. It is a genuine design flaw, and the instinct to fix it is sound.

The proposed fix is to quantify the avoided loss and feed it into the credit decision. That single scalar is where the argument breaks, and it breaks in four separate places.

The four conditions

For a priced counterfactual to carry weight in a decision, four things must be true at once. Not sequentially. Simultaneously.

Estimable. Can the number be produced with enough precision to matter? Thirteen specialist providers assessed the same one hundred properties and their damage estimates correlated as weakly as 0.2. Researchers modelling a flood barrier for New York found the answer changes sign between a 3 and a 5 per cent discount rate. When reasonable methods disagree about direction, the estimate is not a measurement.

Attributable. Can the effect be assigned to the intervention? A firm that builds a barrier also automates a line, changes a supplier and renews its insurance. Which of those reduced its default risk, and by how much, has no clean answer, and pretending otherwise transfers the uncertainty into someone's pricing without telling them.

Decision linked. Would the number change behaviour? Thirty years of making sustainability financially legible has not produced the transmission that was supposed to follow. A figure that nobody acts on is not a price signal, it is a disclosure.

Inside the mandate. Is this the institution's job? A bank can judge whether weak adaptation threatens repayment. It cannot work out the correct level of adaptation for a borrower's neighbourhood. Asking a lender to do the second under cover of the first produces bad policy and bad lending.

Fail one, and the honest output is a direction

The useful part of this frame is what it says to do when a test fails, which is not "give up".

Publish the direction, labelled as a direction. A zero coverage stress that shows what the position looks like if a category becomes ineligible carries information and needs no false precision. A reverse stress that asks how much would have to go wrong before a covenant breaches is decision useful and does not pretend to be a point estimate.

The failure mode to avoid is the middle ground: a specific number, presented with implied precision, resting on assumptions that would not survive being written down. That number is worse than a direction because it stops the conversation that should have happened.

Why we adopted this as a hard gate

The temptation for anyone building risk analytics in carbon is obvious and constant. Counterfactuals are the substance of this market. A crediting baseline is a counterfactual. Avoided deforestation is a claim about what would otherwise have happened. Every one of those is an invitation to produce a confident number where the evidence supports a direction.

So the gate applies to us as strictly as to anyone. Any figure that prices a counterfactual for a specific asset must be estimable, attributable, decision linked and inside our mandate, all four. Failing any one, it publishes as a direction and says so on its face.

This is also what keeps an outcome record from quietly turning into another ex ante scoring business. The whole point of recording what happened is that it does not require a counterfactual. A reversal occurred. A regulator retired reserve credits. A methodology left an eligibility list. None of those needs a view about an alternative world, which is exactly why they can be stated with confidence while the estimates around them cannot.

The transmission point, stated plainly

There is one more argument in the adaptation critique that deserves to travel, because it applies to carbon risk too.

The credit channel does not wait for anyone's methodology. Higher assessed risk brings higher prices, more security and shorter maturities, which leaves less capacity to adapt, which raises the risk again. Pricing risk accurately can protect the individual lender while making the system it lends into weaker.

The honest version of what better risk information does is therefore narrower than the marketing usually claims. It lets the party carrying the loss see it earlier and price it deliberately. It does not, on its own, direct capital anywhere. Anyone selling analytics should be clear about which of those two they are actually offering.

The four conditions are a discipline, not a rejection. Most avoided-loss numbers fail at least two of them, and knowing which two is the difference between a defensible figure and a persuasive one.