Multiproduct Risk Selection in Medicare Advantage

In 2011, the Centers for Medicare and Medicaid Services (CMS) imposed more stringent access requirements on private fee-for-service (PFFS) plans in certain counties in the Medicare Advantage (MA) market.  In practice, this led to nearly all PFFS plans exiting in affected markets, and thus exogenously decreased the amount of product differentiation available in these markets.  Preliminary results show, however, that this led to exit among plan types that were otherwise unaffected by the access requirements for PFFS plans.  Hence, the shock to PFFS plans affected the equilibrium offerings across the entire MA market. 

I propose a simple model to demonstrate how risk selection may change across plans when the product space becomes limited, as well as the optimal insurer response to these changes.  In particular, this model captures the informational externality of adding additional plans to the market, as mentioned in Rothschild and Stiglitz (1976), which is internalized partially when an insurer offers multiple plans in the same market.  To test the hypotheses generated by this model, I use a difference-in-difference framework to exploit variation among county-level markets in the stringency of the access requirement to explore the effect of limiting product differentiation in a government regulated health insurance market on risk selection and on equilibrium levels of access at the county level. 

To further understand how risk selection changes across plans differentially, depending on whether they belong to an insurer that relied heavily on multiproduct risk selection or not, I will be using a previously unreleased version of the Medicare Current Beneficiaries Survey (MCBS) data with the choice set and choices visible for each beneficiary in order to estimate a discrete choice demand model, allowing me to back out marginal costs of plans.  I compare my marginal cost estimates with MCBS data on cost, utilization, and access at the beneficiary level.  Using the year-to-year estimated marginal costs of plans and the variation across counties in the network requirement, I aim to quantify the size of the informational externality of adding more plans and the effect of limiting product differentiation on insurer profits. 

I also use the estimates from the discrete choice demand model to estimate the effects on consumer welfare.  Theoretically, less product differentiation also implies less plan choice across all plan types, which may decrease welfare, but also decrease insurer mark-ups, which is welfare enhancing.  I also consider using the sufficient statistics approach to calculating welfare from Einav, Finkelstein, and Levin (2010), which uses exogenous variation in prices to estimate welfare and is not subject to the problem in logit demand models, where smaller choice sets imply fewer draws of the logit error, thus automatically leading to a downward biased estimate of consumer welfare. 

This study will evaluate whether the policy in 2011 directed to combat the high costs of Medicare Advantage was effective, whether access and consumer choice was impacted negatively by unanticipated effects from insurer reliance on multiproduct strategies, and whether the government should limit or encourage multiproduct strategies among firms by limiting product differentiation.