Selecting risk adjusters with penalized regression and expert judgment: evidence from Colombia

Risk adjustment formulas are essential in health insurance markets, as they mitigate risk selection incentives by aligning revenues with expected healthcare expenses based on enrollee characteristics. However, current formulas can underpredict spending for certain groups, leading to under-compensation for insurers and potentially affecting fairness, quality, and access to care. Many countries are exploring the addition of new variables to improve accuracy, but this can also weaken incentives for cost control. This paper illustrates a methodology approach to risk adjuster selection by using a penalized regression framework that explicitly incorporates the potential downsides of including specific variables. Drawing on a large dataset of over 10 million Colombian health insurance enrollees, we combine statistical estimation with expert assessment of each variable's susceptibility to gaming to construct a specification that limits gaming and maintains predictive accuracy.