The Centers for Medicare and Medicaid Services released the 2023 risk adjustment report earlier this week. Risk adjustment is intended to make insurers risk agnostic. Transfers are paid within state between insurers with no additional federal dollars entering the market. The one exception is an insurer funded catastrophic risk pool that covers a portion of claims over $1 million dollars. My friend and colleague Wesley Sanders combined the data with a few different sources of enrollment data to produce a per member per month (PMPM) transfer calculation.
I have a few observations that I want to pull out from the Top 10 recipients and payers of risk adjustment.
| Largest Reciepents of RA PMPM | Largest Payers of RA PMPM | |||
| MM | PMPM Transfer | MM | PMPM Transfer | |
| 1 | 68,267 | 738.62 | 45,130 | -209.73 |
| 2 | 53,801 | 722.59 | 878,706 | -184.31 |
| 3 | 62,943 | 522.43 | 306,751 | -178.48 |
| 4 | 19,085 | 465.33 | 619,292 | -169.04 |
| 5 | 334 | 375.05 | 76,501 | -124.72 |
| 6 | 5,079 | 352.79 | 1,975,515 | -116.11 |
| 7 | 5,555 | 290.13 | 1,489,464 | -106.33 |
| 8 | 99,663 | 254.91 | 109,178 | -103.41 |
| 9 | 30,680 | 219.52 | 7,331,524 | -102.75 |
| 10 | 98,887 | 213.65 | 13,659,170 | -102.37 |
First, look at the difference in the size of the transfers. The #1 payer pays $209 PMPM into risk adjustment. This is smaller than the PMPM of the 10th largest recipient. The tails are very different in the PMPM size of the transfers.
More interesting to me is the sheer difference in member months. An enrollee can have anywhere from 1 to 12 enrolled months in a year. Dividing total member months by 10 or 11 probably gives a decent approximation of total unique individuals enrolled. The Top 10 net recipients are SMALL to TINY insurers. I am trying to figure out how insurer #5 with 334 enrolled member months generates enough revenue to justify hiring an actuary to file their rates.
The net risk adjustment payers range from small-ish to really BIG. This is interesting. It is not unusual to see extremes of a distribution to be heavily composed of smaller entities because each enrolled individual has more variance in a small cluster than a big cluster. But we don’t see that here. Instead, we see some national and regional carriers having substantial payables.
We probably should think about the market dynamics where the big companies are actively seeking to enroll low risk enrollees and the small companies get extreme tail risk.
Gin & Tonic
Thank you for continuing to post your work here, now that you’re a PhD and all.
I took a quick look at your question about insurer #5, and I’m wondering if that’s just a data reporting issue? They seem to be part of a substantial insurer group, operating in several states (although not long ago sanctioned in one for operating without appropriate authority.) So maybe their data are incomplete, or are showing you one state only, or something like that. Not having access to your data sources I can’t pursue this further.
BradF
David
Conjecture on the 334. What kind of carrier and what kind of patients? Is this an outlier coverage entity that specializes in rare diseases? What block of business is this?
Brad
Keaton Miller
Interesting stuff. In the Medicare Advantage context, the gestalt is that the government overpays relative to true risk, at least in part due to upcoding. Here, scale incentives might generate some of the patterns you are talking about — it’s even cheaper for big firms to insure low-risk folks.
David Anderson
@Keaton Miller: I think the dynamics of zero sum ACA risk adjustment versus external money MA risk adjustment really matter. In MA, every insurer has every reason to STFU about hacks and advantages as they can benefit from running something into the ground while in the ACA, some insurers have strong incentives to call bullshit and get formulas changed as they’re paying for a red queen race.
I agree with you, scale probably matters a lot at the low end of the market which is only buying on premium.
Schtreaky
@Gin & Tonic: I had the same thought. A number like that in this grouping might be a data error — or maybe revealed fraud?