Cream skimming under “perfect” risk scoring

Beyond Statistics: The Economic Content of Risk Scores

By Liran Einav, Amy Finkelstein, Raymond Kluender, and Paul Schrimpf
National Bureau of Economic Research, June 2015, NBER Working Paper No. 21304

From the Conclusions

Our objective in this paper was to highlight the fact that risk scores that are commonly used in credit and insurance markets are not merely statistical objects, as they are generated by economic behavior. We illustrated this point empirically in the specific context of Medicare Part D, the public prescription drug insurance program that covers over 30 million individuals, and explored their implications theoretically. We exploited the famous “donut hole” where insurance becomes discontinuously much less generous at the margin.

Using this research design, we empirically illustrated two conceptual points. First, analyzing the average demographic and health characteristics of individuals as a function of annual drug spending, we showed that spending differences across individuals reflect not only heterogeneity in underlying health but also heterogeneity in the underlying behavioral response to the insurance contract. Second, we show that the current (statistical) risk scores - which are designed to predict spending under a given contract - do not capture this second dimension of heterogeneity.

In the second part of the paper, we use a highly stylized theoretical example to explore some of the potential implications of these findings for the standard use of risk scores: to predict outcomes out of sample under other contracts and use these predictions to set reimbursement rates. We showed that standard risk scoring can create incentives for private insurers to cream-skim individuals whose (unpriced) behavioral response to the contract they offer will make them lower cost than what is predicted by the risk score that was generated under a different contract. A key point is that, when there is heterogeneity in the behavioral response to the contract, these cream-skimming incentives can still exist even in the presence of “perfect” risk scoring under a given contract.



By Don McCanne, MD

A innate characteristic of a fragmented, multi-payer system of health care financing is that health risks vary between insurance pools, creating an opportunity for private insurers to profit by cream skimming healthier population subsets. Risk scoring is used to compensate for this heterogeneity, but this study shows that even when risk scoring is “perfect” under current standards, opportunities for cream skimming still exist.

A single payer system creates a single “homogenized” risk pool which eliminates cream skimming - another compelling reason to dismiss the private insurers and expand an improved version of Medicare to cover everyone.