Insurance information

 When I was doing surplus lines insurance, one product was crop insurance, insurance that the season would produce a crop and not suffer some disaster.  The farmers' were not always right, but they seemed to have an uncanny ability there, that of course we lacked, so when we found ourselves swamped with applications, we pulled the reigns.

Yes—that makes perfect sense. What you were seeing was essentially local knowledge functioning as an information advantage.

A farmer doesn't need a formal statistical model to know that a particular year looks ominous. He may have spent forty years watching the weather, the soil, the insects, the crop at a particular stage, and the behavior of neighboring farms. He may also know things that never make it into an insurer's data: “This field has been too wet for three weeks, and we're going to get another storm.”

The insurer, by contrast, sees applications. Collectively, those applications contain information, but individually they may not reveal why the farmers are suddenly applying in large numbers.

And that swamping of applications was itself information. If the volume of people suddenly wanting insurance jumped far beyond normal levels, that was a pretty good indication that the people with their boots in the mud knew something the underwriters sitting at desks didn't.

So “pulling the reins” was actually a very rational response to an information asymmetry. You didn't necessarily know that a disaster was coming; you knew that the people who stood to lose money if it came were behaving as though one might. That is a remarkably useful signal.

It also illustrates something interesting about insurance: sometimes the hardest risk to price isn't random variation, but the possibility that the insured knows more about the risk than you do.

The insured always knows more about the risk; the underwriter is dependent on the law of large numbers.

Exactly. And that gets to the heart of what insurance underwriting really is.

The insured knows the particular risk; the underwriter knows the population of risks. The farmer knows this field, this soil, this crop, this year's weather. The insurer knows that, historically, something like 3% of comparable fields have suffered a particular loss.

The underwriter's great advantage is therefore not superior knowledge of the individual risk but aggregation. One farmer's knowledge may be better than yours, but a thousand farmers' risks can produce a remarkably predictable aggregate result.

Which creates the delicious paradox: the better the insured is at selecting against you, the less useful your statistics become. If the people buying insurance are disproportionately the ones who privately know they are in trouble, your historical loss ratio isn't describing the pool you're actually getting.

That's why your sudden flood of applications was such a warning. The farmers were, in effect, telling you: “We know something.” And if you couldn't distinguish the farmers who knew something from those who didn't, the prudent response was to stop taking so much of the risk.

In a sense, underwriting is a perpetual contest between the insured's information about one case and the insurer's information about many cases. The law of large numbers is the underwriter's counterweight to the farmer's intimate knowledge.