Completion Methods for Dealing with Large Claims Tails in Small Cut

[ActuaryGuy23]

Fellow actuarials (this is a joke, don't shoot!), hoping you can point me to SOA papers or other literature for a unique completion factor problem I am encountering.

We have a model that attempts to complete admissions and associated lengths of stay (days) for IP claims. For the LOB, some of these admit types quickly see admission payment activity diminish to a trickle (12-15 months). Problem is we have the anomalous admit out beyond lag 16 that will pop up and bring a LOT of days with it.

These tail payments happen frequently enough to be an annoyance when monitoring restatements, but not often enough that there is any way I can come up with to incorporate into the factors themselves. Our current method uses 6-of-8 factors dropping the high/low... obviously that random admit will rarely/never make it into the completion factors. We discussed removing the high/low condition and just taking a straight 8 average, but it seems like that would create 7 months of favorable pick-ups (not currently occurring in the model) and STILL show a large hit in that one month where the rogue admit with lots of days pops up.

Thoughts?

[AMedActuary]

Do you want the model to not take into account this anomalous admit at all (once it's in your data)? If so, you could just remove it from the data when calculating the completion factors and add it back in after (so that the total still lines up).

The other option would be doing something more complex than a simple average so probably would be harder to implement. If you want the model to give a higher reserve due to this long admit (since apparently it is going to happen and could happen again), you could go this route. I believe there are more longer-tailed distributions that could be tried.

This vignette about using the 'chainLadder' package in R gives a good introduction on different chain ladder methods.

[bluecup129]

Agree with AMedActuary. Just set some sort of cap on the claims that make up your lags (i.e. $250k or something similar) and then separately model those large claims and add to the reserve for that. It could be as simple as saying the average amount of these large, "long" claims is $500k so we will hold that amount for each month.

I believe our reserving actuary will override the completion factors making up the average you describe, effectively smoothing the average. That systematically understates your reserve though, as we do expect those claims to happen periodically (just 1 every 5 months instead of 1/5 every month).

[Dr T Non-Fan]

I'd leave them in, then take a look to see how they affect the Claim Liability (or admits and days).
Also, see if anyone in a claims processing or pre-authorization area has info on admissions. Something that far out might be stuck in limbo inside your company.

[ActuaryGuy23]

Originally Posted by Dr T Non-Fan

I'd leave them in, then take a look to see how they affect the Claim Liability (or admits and days).
Also, see if anyone in a claims processing or pre-authorization area has info on admissions. Something that far out might be stuck in limbo inside your company.

We have looked at putting these "shock" claims back in by removing the high/low factor build and just opting for a straight 8 methodology. When we do that, 7 of those 8 incurred months will see favorable pick-ups in the lag 15-18 range because the shock doesn't repeat in those months. The 8th month will still take a pretty big hit (granted not as big since we reserved SOMETHING).

Also have looked into claims pend/suspend accounts and unfortunately they do not retain the information to get a look at length of stay. Could theoretically determine that an admit is forthcoming, but the LOS is the bigger problem we're dealing with in the model restatements. Auths are also something we leverage heavily, particularly in the early months for getting tabs on the admission volumes. Again unfortunately, the clinicians and nurses working in our auth system are woefully inept at logging consistent, accurate length of stay information on those records.

Great ideas, DTNF... sadly dead ends in our company for this specific issue.

[Dr T Non-Fan]

OK.
You're never going to be able to guess which month the long-tail claim will have happened.

I suggest a lot larger book of business, such that these events happen a little more predictably.