Saturday, November 15, 2008

Adverse selection

Over the past generation, conservatives have managed to popularize the concept of "moral harzard" -- the concept that if you prevent people from feeling the full downside brunt of their poor choices, they will consistently take foolish risks. It's a real problem, and having the prudent subsidize the imprudent is patently unfair.

However, there's an opposite risk associated with trying to make everybody solely responsible for their own risks, and that is the lesser-known problem of "adverse election": namely the problem that you can't run any kind of social program if the people who have less than average need for it are allowed to opt out, since this will steadily erode the quality of the pool that is left.

I never quite realized how these two features connected, until I read this brilliant piece by Robert Solow, which explains exactly how moral hazard and adverse selection are flip sides of the same moral coin, connected by the problem of collective action:

Imagine a population of a million similar families, living in a million more or less similar houses. From long experience it is known that the chance that any given family will suffer a severe fire in any given year is about one in ten thousand. In other words, we can expect about a hundred fires per year. The same experience tells us that the average amount of damage per fire is $200,000. So the total damage per year is some $20 million. Serious house fires are rare, but when one occurs it is devastating to the unlucky family.

The existence of fire insurance makes an enormous difference. If each of the million families pays an insurance premium of $20 a year, all damages can be reimbursed. Major house fires would still not be welcome events; but they would not be financially catastrophic. The small probability of a large loss is eliminated, and replaced by a small but certain cost. Insurance companies would have to charge a bit more than $20 per house, to cover administrative costs and profit. Also companies would have to build up a reserve, to allow for the fact that annual losses would surely fluctuate around the average of $20 million, with an occasional bad year. On the other side of the ledger, investment of the reserves, presumably in reasonably safe and liquid securities, would offset at least some of the costs of the system.

Nevertheless, fire insurance has its problems, two in particular. Notice, first, that the existence of fire insurance does nothing to diminish the number of fires. Insurance is a way of pooling or sharing risks, not of eliminating them. In fact the opposite is true: the existence of fire insurance probably increases the number of fires. In the absence of insurance, one has to expect that home owners will be very careful about loose matches, old soldering irons, and other such dangers. The knowledge that they are fully covered may lead to some carelessness, and to more fires. This sort of effect is called "moral hazard." (It is why subsidized flood insurance encourages people to live in flood plains.) Insurance companies have devices to discourage moral hazard. Deductibles and co-payments are two such devices: no fire is costless to the insuree. Required precautions are another device; every insured home is supposed to have an approved extinguisher and smoke alarms.

The second problem is different. All houses are not alike, after all. Some are more fire-prone than others. To take an extreme case, suppose that 90 percent of the million homes have, for various reasons, essentially no risk of fire. The hundred fires per year all come from the remaining 100,000, each with a probability of one in one thousand. They are responsible for the annual damage cost of $20 million. The 900,000 fire-free home owners very likely know this. They are in effect subsidizing the fire-prone houses, so they will choose not to buy insurance. Only the fire-prone homes will be in the market.

This is called "adverse selection." To be viable, insurance companies will have to charge a premium of $200 per year, and even some of the fire-prone home owners may balk. You can easily imagine how the whole insurance market might unravel if there are houses of many degrees of fire-proneness: each time the rate rises, the least vulnerable, least fire-prone customers may drop out, leading to a still higher rate and still more dropouts. Insurance companies may respond by refusing coverage altogether to very fire-prone houses (or refusing health insurance to people who look as if they might actually become seriously—or expensively—sick). Modern information technology and data-mining techniques make it possible for insurance companies to pinpoint the known risks associated with individual applicants and quote "appropriate" rates.

Naturally, they do; but this only further undermines the insurance principle. Unless something drastic is done about it, adverse selection can lead to a situation in which precisely those who need insurance most cannot get it, or cannot afford it. Keep in mind that it is in the self-interest of the safe or healthy not to be in the same insurance pool, paying the same rate, as the fire- or sickness-prone, because they will be paying in more than the costs they incur, so that others can pay in less. In such cases, if adequate insurance is to be provided, there may have to be external regulation or direct public provision.
Now, the obvious political point here is that the conservative governing philosophy is focused almost exclusively on eliminating moral hazard, but completely ignores the problems associated with permitting adverse selection.

2 comments:

Donald Stevens said...

I am just glad we have insurance. Without it, we would have another major issue to tackle without good solutions. We offer multiple insurance quotes at www.getgliga.com and we are proud to do it.

Michael said...

The best introduction I read to the wider behavioral economics field and applications is the Nobel lectures of Stiglitz, Akerloff (market for lemons paper originally coined adverse selection) and Spence. Long but interesting at least in patches.