Abstract

We extend the Rothshild and Stiglitz (1976) model to two sources of risk -inpatient and outpatient risk- to better proxy real-world health insurance markets. We uncover an interesting theoretical possibility: Take individuals A and B, who are low risks in, say, the inpatient dimension but A is riskier in the outpatient dimension. Then, A may enjoy less coverage than B in the inpatient dimension (coverage reversal). This phenomenon indicates that when testing for adverse selection in a given dimension, one has to treat individuals who differ in the other dimension separately. With this insight in mind, we adapt the Chiappori and Salani & eacute; (2000) positive correlation test to this multi-dimensionality and use it to test for adverse selection using individual-level claims data for the privately insured in Chile. This empirical analysis indicates that overlooking the aforementioned need of separating samples can potentially lead to biased conclusions.