A portfolio of climate risks is not judged risk by risk. Two hundred sites each of which is acceptable can form a whole that is not, if a single event touches them all. That is the very definition of accumulation, and it is the one question individual underwriting cannot ask, because it looks at one file while accumulation shows on the whole portfolio. A department that does not measure its accumulation does not know what it sells, whatever the quality of each of its decisions.
The first difficulty is choosing the unit on which to count, and the default choice is almost always the wrong one. One spontaneously counts by commune, by department or by postcode, because those are the fields available. Yet a natural event knows none of those boundaries: a flood follows a catchment, a windstorm follows a track, a drought follows a geological formation. Counting by department means measuring accumulation with an instrument whose graduations are not those of the phenomenon, and the result can be wrong in either direction.
The relevant unit is therefore the peril's, not the administration's. For flooding, the catchment and the sub-catchment. For windstorm, a wide grain, because the footprint is large and the whole territory is concerned to varying degrees. For drought, the map of clay formations. For earthquake, the seismic zone. These divisions exist and are public; what is usually missing is not the data but the decision to attach each site to something other than its commune.
Then comes the question of what is counted, and it is less obvious than it looks. Counting a zone's insured values gives an upper bound that will never occur: an event does not destroy everything it touches. Counting modeled loss gives a sound quantity that depends on the model and its assumptions. Reasonable practice keeps both, because they answer two different questions: the first says what has been put in the same room, the second what one expects to lose from it, and a management that holds only the second loses the ability to see a concentration the model understates.
One particular accumulation must be named here, because it escapes every geographic grain: that of sites which are not in the same place but depend on the same place. An insurer may hold a single plant in a valley and find itself with fifteen claims, because fourteen of its insureds depend on that plant for a component, and their policies carry a supplier failure extension. That accumulation reads on no map: it reads in the extensions written, and nobody aggregates them.
Controlling accumulation finally requires an organizational decision, failing which the measurement serves no purpose. Measuring accumulation after the fact produces an observation; holding it requires a limit per zone, known to the underwriter at the moment of saying yes, and a mechanism that informs him when the zone fills. Without that, the measurement arrives at a quarterly committee, by which time the business has been written, and the only possible action is non-renewal, that is, correcting a concentration twelve months late.
What this discipline costs must be owned, because a lesson that does not say so describes a trade that does not exist. A zone limit leads to declining business that is good in itself, because it arrives after other business. That is unpleasant to explain to a broker and it is exactly the point: zone capacity is a finite resource, and the only way never to decline a good risk is to discover in one night that everything was put in the same place.
A commercial insurer measures its flood accumulation by department. In the department where it is most present, it shows 1.8 billion euros of insured values across 240 sites, and its internal limit is set at 2 billion per department. Attaching sites to sub-catchments, done for the first time, gives another picture: 62 of those sites, that is 780 million in values, lie in the same sub-catchment, and 41 of them are downstream of the same protective structure. In addition, 18 contracts in the portfolio, located in four other departments, carry a supplier failure extension naming a logistics site that lies in that sub-catchment. What does this attachment reveal, and what must change?
The departmental limit was not too high, it was measuring something other than what it thought it measured. A department is not a peril unit: a flood follows a catchment, and a portfolio spread over two catchments of one department does not carry the same accumulation as a portfolio concentrated on one. Here the sub-catchment attachment shows that 43% of the department's values are exposed to the same event, which the previous measurement could not see and never will, whatever limit is given to it. The second finding is harder still: 41 sites downstream of the same protective structure are not 41 risks, they are a single risk carried by the holding of a structure the insurer neither maintains nor controls. A hazard map built assuming the structure functional will say nothing of that scenario, which is nonetheless the one where the burden is made. The third finding is the one nobody was looking for: the 18 supplier failure extensions naming a site in the sub-catchment add to the accumulation contracts located hundreds of kilometers away, and that share reads on no map since it sits in the extensions written. What must change comes to three decisions, none of which is a number. Attach each site to its sub-catchment rather than to its commune, which is a bounded data task done once. Set the limit on the peril's unit rather than the administrative one, and make it known to the underwriter at the moment of binding, not at the quarterly committee that records. And aggregate supplier failure extensions by named site, because an accumulation one does not count is not a smaller accumulation, it is one that will be discovered at the moment of paying.
- 01Two hundred risks acceptable one by one can form a whole that is not: accumulation is the one question individual underwriting cannot ask.
- 02The counting unit is the peril's, catchment, clay formation or seismic zone, never the administration's.
- 03Insured values and modeled loss answer two questions: what has been put in the same room, and what one expects to lose from it.
- 04Sites downstream of one protective structure are not several risks, they are one risk carried by that structure holding.
- 05A zone limit bites only if the underwriter knows it at the moment of binding: measured at committee, it merely records.