I don't recall this topic on any exam syllabus unfortunately, so here we are. I'm wondering the best way to back test your trend selections.
I'll make some simplifying assumptions for the below example. First we’ll assume there is no seasonality in loss experience in the past and no reasonable expectation of seasonable loss experience in the future. Second, we’ll assume a pretty stable risk set, so whatever steady distribution of risks across rating variables you have seen in the past, is expected to be similar in the future.
Say two quarters ago your calendar severity was 2000, and your selected prospective annual severity trend was +4%. Today, if your severity is only 2035, this is close to the halfway point of where you would expect your severity to be on the way to 2000*(1+.04)^.5 = 2039. Does this mean on a preliminary basis your trend selection looks pretty good? Can you only truly evaluate an annual severity trend in a year’s time? How does this change if you do have seasonality in your experience? How do you assess a 4% trend selection from 4 quarters ago? How about 6?
Are there any papers on this topic?
Season's Greetings,
Gareth Keenan
I'll make some simplifying assumptions for the below example. First we’ll assume there is no seasonality in loss experience in the past and no reasonable expectation of seasonable loss experience in the future. Second, we’ll assume a pretty stable risk set, so whatever steady distribution of risks across rating variables you have seen in the past, is expected to be similar in the future.
Say two quarters ago your calendar severity was 2000, and your selected prospective annual severity trend was +4%. Today, if your severity is only 2035, this is close to the halfway point of where you would expect your severity to be on the way to 2000*(1+.04)^.5 = 2039. Does this mean on a preliminary basis your trend selection looks pretty good? Can you only truly evaluate an annual severity trend in a year’s time? How does this change if you do have seasonality in your experience? How do you assess a 4% trend selection from 4 quarters ago? How about 6?
Are there any papers on this topic?
Season's Greetings,
Gareth Keenan
Back testing your trend selections?