Yesterday afternoon, I was having a beer and a long discussion on Zoom with my friend and most frequent co-author. We’re trying to figure out what our combined research agenda for the next couple of years looks like after we get through a couple of current and near future projects. We think we have some cool questions that both build on our previous work while also being intellectually interesting and extremely policy relevant.
We got cool stuff in the near and long term pipeline.
However, one of the more useful parts of the conversation was the decision to send an idea of mine that has been bugging me for 6+ years to the big farm upstate where good projects go to play with other really good projects. I want to study negative premium plans in the ACA but we can’t figure out the data or the approach. I’m throwing this out there for anyone to steal and do really cool things with as I want to read your paper in 2027 but I don’t know how the hell to write that paper.
We know that zero premium plans are pretty common in the ACA (as well as Medicare Advantage and Medicaid Managed Care). We know token positive premiums create substantial administrative burden. Every ACA paper that I write (including my dissertation) has a line that says something to the effect “premiums are minimally bounded by zero….” That is not quite right! Sometimes there are extremely lagged negative premium plans or in non-econ language — sometimes there are plans where you eventually get paid to buy them.
WHAT?
Yeah, this is weird. It is an artifact of the Medical Loss Ratio(MLR) regulations. MLR rules state that over a 3 year rolling window, an insurer has to spend 80% of qualified premiums on either medical/pharmacy claims OR qualified quality improvement projects. In reality, gross premiums are greater than qualified premiums, so the actual ratio is around 75%(ish). If an insurer over that three year window in a state-market segment has an MLR below 80%, the insurer-segment has to send out rebate checks to enrollees from the last year of the three year window.
Most of the time, MLR rebates are going to be actuarial noise as insurers aim to be just above the cut-off point and sometimes they miss high and sometimes they miss low.
HOWEVER— in 2018, the ACA market was an ungodly mess.
Lots of insurers decided to leave in 2017 for the 2018 plan year as they had no idea if the market would still exist when they had to do their rate filings in summer 2017 due to Repeal and Replace and they had no idea what was happening with Cost Sharing Reduction subsidy payments. The surviving insurers did what insurers and actuarials do when they are scared — massively jack up rates. They also put the cost of CSR payments into their Silver premiums in a move known as Silverloading.
More particularly, a few insurers, including one in Virginia, were local monopolists and realized that they could send premiums through the roof and most enrollees who got subsidies would not be effected (interesting side note — GOLD enrollees facing a massive relative price shock… find new grad student minion to pitch). By summer 2018, it was obvious that insurers were Scrooge McDucking it for 2018 as premiums skyrocketed, enrollment stayed flat-ish and claims were flat-ish.
By the start of 2019 and especially the 2020 Open Enrollment Periods, it was obvious that a couple of insurers would need to issue huge ass MLR rebate checks for their 2018 pricing. And an individual who could buy a low to no premium plan in one of these insurers was effectively buying an option that was very likely to be in the money on one of those huge ass rebate checks. The combination of normal ACA subsidies, and Silverloading would drive down their premiums at the point of purchase and the MLR rebate check would, eighteen months later, make their effective net premium negative by a whole lot. In some cases it could be a several thousand dollar check.
The companies that were very likely to be giving out honking huge rebates was pretty predictable — so did attentive enrollees buy their option?
I want to answer this question but the data has never been good enough….
So feel free to use this set-up as I want to read your paper in 2027.