Three quantities wrongly added together

When a property portfolio sees its expected annual loss rise by thirty percent from one year to the next, three possible causes present themselves, and they do not call for the same actions. The peril may have become more violent or more frequent. The exposed property may have become more fragile. Or there is simply more insured value, better recorded, in the same place. These three causes have precise names, hazard, vulnerability and exposure, and most of the modeling craft lies in the discipline of never mixing them.

Hazard is a physical quantity measured at a point, independent of whatever stands there. A water depth, a gust speed ten meters above ground, a ground acceleration, a cumulative moisture deficit over a summer. Hazard knows neither walls nor policies: it would describe the same thing on bare land. Written out this way it looks obvious, and it is nonetheless the first boundary one crosses without noticing, by saying that a zone is risky when one means that it is expensive.

Vulnerability is a relationship, not a number: it states what fraction of an asset's value is destroyed for a given intensity of the hazard. It depends on the construction type, the height of the first usable floor, the location of technical plant, the presence of a basement, the nature of the activity. Two neighboring buildings, under the same water depth, do not have the same vulnerability, and that is precisely what modeling exists to say. A model that does not tell a pallet warehouse from an electronics workshop measures nothing useful.

Exposure is what the insurer actually holds: values, located, described. It looks the simplest of the three and it is by far the dirtiest. A badly geocoded address moves a site eight hundred meters, which is enough to put it on the other bank. An insured value entered at book value instead of reinstatement value understates everything else, proportionally. A multi-building site declared on one line concentrates at a single point what is spread over ten hectares, and that changes both the hazard retained and the aggregation.

The reason for separating the three is not taxonomic, it is operational: each belongs to a different owner and is corrected by a different action. Hazard comes from a scientific source one does not amend oneself, one changes it or accepts it. Vulnerability is corrected by risk knowledge, a survey, a question put to the client. Exposure is corrected by data, an entry control, a geocoding run. A gap attributed to the wrong term leads to paying for the wrong remedy, and most often to raising a rate where a file needed fixing.

This confusion has a recognizable shape, worth naming because it returns at every portfolio review. An underwriter observes that a region's modeled loss has doubled. He concludes that the peril is worsening and raises the rate for the whole region. On checking, half the gap came from a refresh of insured values long left frozen, and the other half from a change of model version on the vulnerability functions. Hazard itself had not moved a millimeter. The decision taken was defensible in its intent and wrong in its object.

The working method that follows is monotonous and it protects. Faced with any gap, the calculation is rerun changing one term at a time: same exposure with the new model, then new exposure with the old model. The two gaps rarely add up to the total gap, and that remainder is exactly the information one was looking for. A portfolio whose variation one cannot decompose is not steered, it is endured, and the first sign that it is endured is that climate comes up at every meeting while address quality never does.