Oskam et al have a new health insurance paper in the European Journal of Health Economics that looks at the risk of risk even after pretty good risk adjustment in the context of the Dutch health insurance market. The Dutch market has some similarities to the US ACA market. I read this paper last week and it has been deeply marinating in my head each time I’ve reread it.
We think that risk adjustment, compensatory payments to insurers that cover people with predictable high costs is needed to discourage or eliminate selection and screening incentives when we have a system of community rated and guaranteed issue insurance. Risk adjustment is a group level compensator. It pays an insurer the average incremental cost generated by a specific marker of predictable cost. However there is variance within group. Some people have cheaper than average experiences with a specific marker of predictable costs and other people have really expensive relative to the average experiences.
The easiest American example is hemophilia where the “cheap” years are $150,000 to $250,000 while the rare OMG year is equal to the pay of a competent veteran Major Leaguer. Yet risk adjustment will give the same transfer for hemophilia for both individuals. Insurers have strong incentives to try to figure out who could cost the equivalent of someone who can hit a curve reliably so that they can be avoided while the insurer would love to cover a bunch of hemophiliacs who are very unlikely to have million dollar bleeds because they risk adjust very well.
The reason is that the uncertainty about residual spending (i.e., spending net of risk equalization) differs across groups, e.g., the risk of substantial losses is larger for the chronically ill than for the healthy. In a risk-rated market, insurers are likely to charge a higher profit mark-up (to cover uncertainty in residual spending) and a higher safety mark-up (to cover the risk of large losses) to chronically ill than to healthy individuals. When such differentiation is not allowed, insurers face incentives to select in favor of the healthy….
In a third step, to quantify the heteroscedasticity of residual spending, we estimate the standard deviation (as a proxy for the uncertainty in residual spending) and the 99.5th percentile of residual spending (as a proxy for the risk of ruin) for each risk group. Variation in these measures across risk groups provides an indication of the heteroscedasticity in residual spending across risk types….
Although the mean residual spending equals €0 for both risk groups, the distributions are quite distinct. In addition to the visual contrast between the two distributions, some metrics are presented to summarize the divergence. The red [morbidity indicated — DA added ] distribution has a standard deviation of €11,378, far larger than that of the total population (€6978), while the blue distribution has a smaller standard deviation of €4646. The results for the group without morbidity are more concentrated around €0 than those for the group with morbidity, implying that insurers face more uncertainty regarding the ex-post financial result for individuals with a morbidity flag…
Insurers are in the business of aggregating individual level variance into far more stable . The more variance in a distribution, the more expensive it is to control even if the aggregate mean is the same as a lower variance distribution.
The authors also find that as the number of morbidity flags increase, the variance increases as well. People who are very sick are both very expensive (which is fine as that is what risk adjustment and reinsurance are there for) and very variant — which produces selection incentives. For conditions where there is a fairly large population like diabetes, an insurer that is big enough can eat enough risk to tamp down the individual level variance within a population but for conditions that are both expensive and rare, variance would still dominate a national single payer. Insurers have strong incentives to either segment the market in some way to force the high cost high variance people to choose higher premium plans OR to avoid the segment entirely even if risk adjustment is perfect at the group level.
Baud
There are or used to be high-risk pools for people who couldn’t get private insurance, at least pre-Obamacare. To my knowledge, those didn’t work that well. But does it make sense, in a system where we still rely on private insurance, to have an preliminary “Medicare for All” system that pays for illnesses that fit the profile of what you’ve described?
David Anderson
@Baud: A well funded automatic high cost risk pool system can work — there are several key words in that phrase.
High cost is not particularly a problem per se as long as risk adjustment works reasonably well — it is the variance that exists after risk adjustment that creates the problem. Here reinsurance on the residual makes a lot more sense where the reinsurance kicks in only after the risk adjustment payment is accounted for.
StringOnAStick
New word alert: heteroscedasticity! I read a lot so it’s rare to see a word I can truly say I’ve never seen before, so thanks for the info and a new word.
gene108
@Baud:
Medicare pays for dialysis, kidney transplants, and immunosuppressants, for three years after transplant, so I think Medicare already dipped its tow into insuring a high risk population.
I think the issue is are we willing to see an increase in taxes to cover other organ transplants and other high risk conditions? I don’t think so, primarily because the savings insurers get from the transfer of risk will not be passed on to consumers.
Andrya
Thanks Mr. Anderson! I teach “Introduction to Statistics” at a community college, and this post will make a great “think about this” problem for classroom discussion. One of my challenges is to get my students to realize how very, very important variability is.