Many papers on health insurance worry about the problem of adverse selection. Critics of HMOs claim that the fact that HMOs have lower costs is not due to more efficient provision of services nor the limitation of the provision of services, but instead largely caused by the fact that the people who choose to enroll in HMOs are healthier on average. Shen and Ellis (2002) aim to test whether or not costs incurred from modeling an individual’s risk lead to higher profits. A patient’s risk is measured in 4 different ways:
- using age-sex categorization
- using prior medical service utilization
- using Adjusted Clinical Groups (ACG) [example]
- using Diagnostic Cost Groups (DCG)
The data used in the paper come from the Mercer privately insured dataset for the years 1992 and 1993. The authors use OLS estimation and justify this econometric methodology by claiming that when data sets are large, OLS is nearly as good an approximation as cell-based or non-linear models. The authors find that categories 2, 3, and 4 provide significant information regarding the probability that an individual will incur future medical costs (R-squared between 0.079 and .106). Category 1 also improves cost prediction over the pricing with no categorization, but the age-sex category is less accurate (R-squared of 0.019). The authors find ACG generally has the highest gross profit rates for insurance companies regardless of the information which the payer has. Gross profit ranges between 24% and 60%. Overall profit increase ranges from $68 to $260 million depending on the risk selection system used and the information the payer has about their own propensity to use medical services.
In order to obtain these results, the authors used a simulation methodology and assumed that it was costless to drop any unprofitable payer. The authors also looked only at one-year profit maximization schemes and did not look at any reputation effects. Shen and Ellis conclude that this additional information does help increase an insurance company’s bottom line. The major contribution of this paper is that they quantify how much additional profit can be gained from these risk selection mechanisms, even if this quantification is done in a static environment in which customers can be costlessly dropped from coverage. The authors do not mention whether or not risk selection is good or bad for society since the utilty loss of those dropped form coverage is not modeled. The only area of study is insurance company profits.
Shen and Ellis (2002) “How profitable is risk selection? A comparison of four risk adjustment models,” Health Economics, Vol 11, pp. 165-174.