"Just re-tune the trigger" — and the 19% that won't go away

Whenever I show that a weather index under-pays the losses it's meant to cover, an experienced actuary gives me the same, reasonable pushback:

"Fine — so re-calibrate it. Move the threshold until the payouts match the losses. Problem solved."

It's the right instinct. So I did exactly that, on real European wind data, and measured what happens. The answer is the useful part: re-tuning helps, and then it hits a floor it cannot cross.

What I did

Using public reanalysis wind data (ERA5) and actual grid generation (ENTSO-E) over a full decade, I swept a wind-drought trigger across every threshold you could reasonably pick, and at each one I asked two questions:

How often does the trigger miss a whole drought the farm actually felt?
Of all the drought-hours the farm suffered, how many go unpaid?

Then I found the settings that make the trigger as fair as possible — the ones an actuary would choose.

What comes out

Two things trade against each other. Push the threshold to catch more droughts, and you start paying out in weather that wasn't really a drought (expensive false alarms). Pull it back to stop the false alarms, and you start missing real droughts. You cannot minimise both at once — improving one worsens the other. There is no threshold that makes both errors disappear, because the problem isn't the threshold. It's that a single regional wind number is an imperfect stand-in for one farm's actual output.

At the best achievable, frequency-matched setting — the fairest tuning available — a residual basis risk survives:

the trigger still misses roughly 6.7% of whole droughts (about 9 of 135 independent events), and
it still leaves about 19% of all drought-hours unpaid (with a confidence range of roughly 16–23%).

Re-tuning moved these numbers. It did not remove them. This residual is irreducible by threshold choice — you cannot calibrate your way out of it. (You might reduce it by redesigning the index — a better-chosen variable, a smarter spatial definition — which is a different and harder piece of work. But no amount of re-tuning the existing trigger gets you there.)

Why the distinction matters to a buyer

"We calibrated it" is often offered as the end of the basis-risk conversation. This result says it's the beginning. Calibration fixes the average; it does not tell you what's left in the tail — and in insurance, the tail is the entire point.

So the question a committee should actually ask isn't "is the trigger calibrated?" (it usually is). It's "after calibration, how much of my loss still won't be paid — and where does it concentrate?" On this wind example the honest answer is: about a fifth of your drought-hours, concentrated in the deepest events. For some buyers that's perfectly acceptable and worth knowing; for others it's a dealbreaker they'd have discovered too late. Either way, they deserve the number before they sign, not after the wind stops.

That number — the residual that survives calibration — is what an independent check is for.

Part of a series measuring how well weather triggers track real losses. Next: what that 19% costs in euros, on a realistic wind cover. If you price or carry this risk, I'd welcome your read on where the residual bites hardest.

Disclaimer — published in a personal capacity, under my own name. This is independent work: produced entirely on my own time and my own equipment, from public data (ERA5 reanalysis; ENTSO‑E generation), using no resources, data, or working time of any employer, and outside my assigned duties. It carries no institutional affiliation and is neither connected with nor endorsed by any employer or institution; the views are mine alone. It is not financial, actuarial, or legal advice, and nothing here is an offer of services.