Collective risk assessment in Affordable Care Act markets: A Bayesian hierarchical model

Abstract

The changes that the Affordable Care Act introduced to the US health insurance market have entirely altered the traditional ratemaking process. Precisely, the creation of statewide community rating schemes and a guaranteed issue has facilitated insurance coverage to the high-risk population, leading to massive changes in risk pool compositions. The implementation of Risk Adjustment has neutralized some of the consequences of limiting premium variation in the market. However, setting appropriate rate levels has remained cumbersome due to the uncertainty about the statewide risk pool. Many insurers, who could not quantify the health risk associated with the statewide yearly enrollment, had to face unexpectedly high payments on risk equalization. Natsis (2019) stated that in this environment, the use of traditional univariate techniques to project statewide health care costs could be potentially misleading. This thesis proposes a Bayesian approach to reflect important sources of uncertainty over statewide actuarial estimates. The aggregate loss is modeled with a novel collective risk model based on a Generalized Beta Prime (GBP) distribution, accounting for long tail risks and changes in risk pool compositions. The GBP is presented with a mean-dispersion parametrization, which allows the introduction of a hierarchical prior specification over the state-specific means. This parameter structure, responsible of quantifying uncertainty and sharing information among states, is a cornerstone of the adopted collective risk model. Using the Commercial Health Care data extract published by the Society of Actuaries (2019), the model is applied on the Surgical and Transplant service category. The resulting heavy-tailed posteriors of the nationwide service means illustrate the high variation of inpatient medical costs. Moreover, the posteriors of the statewide aggregate claims remain highly right-skewed, reflecting the risk of facing sicker populations and high-cost treatments at individual claim level.