We have a few new property and liability segments that our company is entering. For now, I have either similar lines or benchmark data that I am using to create development factors for reserving. At some point, I expect us to have enough volume/history to rely on our own experience, but I’m not aware of standard procedures to make that transtion.
In pricing, they have limited fluctuation measures, etc. to credibility weight your own experience with the industry experience, or to determine when you can fully rely on your experience for rating. Does anything similar exist in a reserving framework?
Not in reserving either, but this is a good question. I would think you can use the same credibility standards (for freq and sev or LC) as pricing to determine the appropriate volume you need for reserving.
is there an ASOP to refer to for guidance? other than that, benchmark, similar lines, similar lines in other companies with history, etc. and reasonable ranges will of course be wider, right?
Unqualified to give an opinion, but that has never stopped me before.
Loss Ratio method, until you realize how bad that estimate is.
Also, make sure they (sales) don’t start selling until the beginning of the Fiscal Year. That way, you’ll have more data to use for reserving at year-end.
Nothing that I’m familiar with in the ASOPs or Exam materials that I can think of relating to credibility. I’m currently using benchmarks and similar lines, but it’s not clear to me how to partially incorporate your own age-to-age factors in any justifiable weighting. I’d like something stronger to back me than “I decided to take the average of benchmark factors and own-data factors”.
Not very familiar with MCMC – does anyone know if that is applicable here? The big downside of course would be having to implement something like that from scratch. Doing something with weighted development factors would probably be easier to slide into our existing process.
I would think ASOP 25 would be the one, in terms of using credibility to blend estimates, but I don’t recall the guidance being specific enough to CG’s question.
There are a few papers related to blending industry benchmarks with experience to create LDF patterns. On exam 7, there is the Verrall paper, but more recently there are a few CAS papers and presentations related to using bayesian methods to credibility weight development patterns.
If you are willing to make a few mathematical assumptions, Dave Clark’s paper describes a method to perform credibility weighting using conjugate priors without having to perform any MCMC simulations.
Depends on the tail length of the line. You can probably starting giving weight to line ratios where you have 3 or 4 points of data, assuming they are relatively stable, but how long that take depends on if the line settles after 3 years or 15 or 60.