Last year, model looked at the following table and came up with a 2025 projected frequency of 207. Exposure is not something that has any trend to consider.
| Exposure | Freq | |
|---|---|---|
| 2018 | 100 | 100 |
| 2019 | 100 | 109 |
| 2020 | 100 | 114 |
| 2021 | 100 | 137 |
| 2022 | 100 | 158 |
| 2023 | 100 | 171 |
Since that projection, the new final 2024 claims year final figure came in at 221.
Anyone have any ranges they’d care to put out for a 2026 program year projection of frequency?
Just on the information you’ve given: 261, with a range of possible values of 254-265.
is there an overall cap? I.e. in life insurance there’s a max of 1,000 / 1,000, so while an exponentially increasing rate can continue for a while, it cannot continue forever. Is there a max frequency that would apply?
otherwise, I’d estimate 253, +/- 10 ( [243, 263] ), just based on geometric average growth rate, with error bounds at +/- 0.5 st.dev. of growth rate.
Tree fiddy.
the downtown estimate is always at the ready
there is not. I mean there is a theoretical cap (every member of the population). But the population (used a standby of 100 here) is not in play at all.
@Bro and @Ted_Hoffman - those projections/ranges are for 2026, 2 years after the listed experience periods, correct?
Merde, that’s right - I need to forecast out 2 years, not 1.
Guesstimate: 308
Low end: 285
High end: 345
With trends both this steep and volatile, it’s going to be a huge range. Can you reveal anything about the product and what might be driving these jumps?
I asked Gemini, and it came up with 2025=281 and 2026=351. That’s obviously giving a great deal of weight to the most recent data point. That may or may not be appropriate.
This is a rate of high cost medical claims from an employer stop loss book of business. Sun Life, one of the main writers of it produces a study annually based on their experience. High cost is defined as total paid >$1M. So the threshold of $1M in total paid remains the same for all years. There is standard severity trend pushing some of the increase, as another year of 7% ground up trend brings more over the line. There are also some treatment plan changes (new innovative treatments) that emerge, albeit slowly. The exposure they use in the paper is “per 1M employees”.
The most recent year over year jump did possibly two things. One, it definitely increased any multi-year average that includes the latest year. Two, if you use most recent as a launch to the future it is higher. While the 29% increase this year is highest, there was a 20% and a 15% in the sample period also. (and a 9%, 5%, and 8%)
So while last year’s projection (from me pulling numbers from the air) using a certain methodology projected 2025 at 207. But that isn’t holding up well!
ETA - the 2020 plan year only went up 5% from 2019. Covid suppressed some care for sure, but people with the real bad shit still need care. We likely lost “non-emergency knee replacement gone wrong through sepsis” that would have generated some claims in 2020. and maybe that suppressed care leads to some cancers that didn’t get caught early enough and drove some worse outcomes 2 years later. But the overall pattern is more than that IMO
The numbers make a bit more sense knowing it’s an excess trend. One of the main drivers in excess trend is new claims penetrating the layer due to ground up inflation.
Reminds me of figure 11 from this paper from my exam days: https://www.casact.org/sites/default/files/database/proceed_proceed88_88049.pdf
Given the thinness of the data, you might consider adding an additional method where you apply selected ground up trends to an estimated loss distribution to see the implied leverage on the excess layers.
How much data do you have / can you get on the history of sub-$1m losses. Would it be possible to get data on losses/claims that were, say, $100k or $250k or greater?
Also, do you have access to an external/industry assessment on the expected frequency of claims excess of, say, $250k and $500k in addition to $1m?
I’ve never worked directly with Medical Excess…so I don’t know what you might have available. However, your problem sounds somewhat analogous to what I deal with when I parameterize claim frequency for large losses in my (P&C) capital model. The information I mention would be analogous for the starting point for the approach I take. If you’ve got something like that, I can probably provide a few hints that wouldn’t run afoul of IP concerns.
nope, I missed that. I was just doing 1 year forward look. My bad. I’ll see myself out now that there’s more qualified people here.
Reinsurance?
Does the cost trend affect the frequency trend?
Asking for a friend.
Ah, Post #9 explains this.