Completion Factor

[T-royBoy]

At what completion factor would you say it is fully credible? If I look at a triangle of data, determine some pattern of payment stream and calculate the IBNR and all, there must be a certain point where, what we call, the "Lag Method" will produce an estimate that is fully credible.

Drug claims are pretty simple and usually they look like this (current month to ultimate) (.800, .99, 1.00, 1.00). I would say with high accuracy that the payment stream is fully credible after the 2nd paid month.

Medical claims can vary but here is what I will use as my example:
(.15, .265, .55, .75, .80, .85, .875, .9, .92, .94, .95, .96, .97, .98, .99, .995, 1.00)

I would say that the lag method would be fully credible at about the .9 level, the 7th lag month. Would you?

[Dr T Non-Fan]

Check the variance of each completion factor. Then use an F-test. I've done that, with reasonable results.
For example, if your 0.9 completion factor is an average of 0.8, 0.85, 0.9, 0.95, 0.99, then I'd conclude (somehow) that 0.9 isn't very credible. If it's tight, the F-test will show this.

The other half of your exercise involves finding complete non-credibility.

I'd punch in at 0.8 full cred, 0.4 full non-cred.

My thoughts behind 0.8 full cred is that if it's off by 0.02 either way (a likely event), then your claim estimate for that month is 2.5% off. At 0.4, being off by 0.02 (which is much more likely, BTW) means your claim estimate is off by 5.2%. I don't think that's acceptable given the possibility.

And it does depend on what you're planning to do with the results. If you're taking 12 months of estimated completed data to price or to analyze certain claims, then the effect of error in one month is diminished (unless the cf process is biased one way).

[Toll Free]

It also depends on the coverage line. In your pharmacy data, for example, I wouldn't necessary call 0.80 a credible factor. For a relatively long-tailed, annuity-type product (such as short-term or credit disability), a factor of 0.20 might be credible.

You also need to consider that, for many lines, trend miss can cause rather large errors in recent months, especially if your projection period (the time with less than full credibility) is long.

That being said, I usually start about 0.70, and work outwards from there - looking at other statistics like restated reserve/exposure, claims pmpm vs a trend model, gut feel...

[mr.c]

That being said, I usually start about 0.70, and work outwards from there - looking at other statistics like restated reserve/exposure, claims pmpm vs a trend model, gut feel...

...clinic days in the month in question...

[Patience]

also you need to consider the consistency of the claims department. Changes in policy, procedure & efficiency can change your expected completion from historical.