A late 2009 NBER paper by Alma Cohen and Peter Siegelman reviews the literature on adverse selection in insurance markets. Adverse selection becomes a problem when individuals of higher risk than expected (specifically, than reflected in the premium) purchase coverage. Higher risks lead to higher costs and higher premiums, which induce lower risk individuals to forgo coverage, increasing per insured costs further–a classic death spiral.
Cohen and Siegelman’s paper “Testing for adverse selection in insurance markets” includes a section on health insurance, but covers many other types of insurance as well (auto, life, long term care, crop, etc.). As for health insurance, a large body of empirical work supports the claim that adverse selection exists in health insurance markets. Cohen and Siegelman (hereafter C&S) cite a review article by Cutler and Zeckhauser as one source for accessing this literature.
Some prior reviews of the literature on adverse selection conclude that the evidence is “mixed,” “inconclusive,” or “ambiguous.” C&S
argue that one should not expect the question of whether a coverage–risk correlation exists to be answered identically in all insurance markets or even in all pools within a market. Thus, one should not regard studies that reach opposite conclusions about the existence of a coverage–risk correlation as necessarily in conflict with each other. (© 2009 by Alma Cohen and Peter Siegelman.)
This is a crucial point because it suggests important limitations in our ability to generalize from one market or sub-market to another.
C&S make brief mention of risk adjustment, a means by which the effects of adverse selection can be mitigated by compensating the insurer for the level of risk of its insureds. There is widespread misunderstanding of the degree to which adjustment solves the adverse selection problem. It is never complete, often far from it. C&S tell us why and back it up with literature citations: insurers and their regulators do not always have access to or make use of all information relevant to risk.
The paper includes a section that describes the ways in which adverse selection can fail to appear. One class of ways essentially boils down to policyholders not having an information advantage over insurers. If insurers have sufficient information relative to policyholders they can price their products to account for the expected level of risk each will draw, eliminating the problem of adverse selection.
Another way in which adverse selection can be offset is if a subset of the population that purchases insurance is a source of favorable selection. For instance if cautious people are more likely to buy insurance and more likely to prevent claims they will provide a source of favorable selection. C&S provide other examples like this. It is tempting to view this whole line of reasoning as: selection can fail to be adverse if it isn’t. The value added is the explanation of why it is not adverse.
Selection into an insurance product can be driven by non-personal factors. Institutional or regulatory factors can be the dominant factor in selection. Examples include the employer-based health insurance market, or the high rate of subsidization of Medicare drug plans. One should not expect adverse selection in such cases.
Even when selection is not adverse, coverage can lead to higher utilization of covered services, an effect known as moral hazard. C&S review the literature that attempts to disentangle adverse selection from moral hazard.
Finally, C&S conclude with policy implications of their findings:
Policy discussions should try to tailor themselves to the specific insurance market under consideration, recognizing that adverse selection and coverage–risk correlations vary across insurance markets (and even among pools of risks within a market), and that they do so in ways that are at least somewhat predictable on the basis of existing research.
This is good advice, but no doubt hard to follow. Much of the utility of empirical research flows from its generalization. If such generalizations are shown to be misleading, the import of each piece of work is greatly circumscribed. In practice, judging the credibility of generalizations is as much art as science. Knowing more of the details of the relevant market matters in getting things right (and certainly in getting published), but policy messaging that actually makes a difference rarely relies on the nuances.