The Economics of Costly Risk Sorting in Competitive Insurance Markets

In the last twenty years, state and federal policy makers have often been asked to restrict the criteria, or tests, on which insurance companies may base their risk categorization. Insurance companies have responded that it is fair and reasonable to charge groups with different probabilities of death or accident actuarially fair rates for each group. The companies argue that if they are not allowed to make these distinctions, the result will be adverse selection with low-risk individuals self-insuring to a greater extent than is efficient and high-risk individuals overinsuring. This paper analyzes the efficiency of insurance companies’ incentives to use certain types of sorting devices-those that are themselves costly to society. Genetic screening, tests for the HIV (AIDS) virus, and criteria that require extensive monitoring of behavior (e.g., cigarette smoking habits) all use real resources to separate risk groups. Such sorting devices do not necessarily enhance efficiency. Though the decline in adverse selection does increase total surplus, this increase may be more than offset by the resource costs associated with determining individuals’ risk levels. Still, the test may be used in private insurance markets, because the gains to low-risk individuals from the sorting come not just from the increase in total surplus but also include a transfer of wealth from highrisk people. So long as the low-risk people would be made better off by distinguishing themselves, they will do so, and competitive firms will find it profitable to offer them lower insurance rates. The analysis is in the tradition of Spence’s (1973) signalling theory.’ Just as education yields a greater private return than social return in Spence’s model, the next section demonstrates that risk-sorting yields a greater return to low-risk individuals than to society as a whole. The basic point of inefficient costly risksorting has been made by Cracker and Snow (1986) who have also shown that price/quantity bundling to induce self-selective sorting, as proposed by Rothschild and Stiglitz (1976), may be imposed inefficiently. Section II gives some intuition for these results. Sections III and IV extend the Cracker and Snow analysis in two directions. Section III considers the role of a consumer’s information about his own risk