View Full Version : Trending to AY (head-scratcher)

02-28-2012, 09:01 PM

Something I was wondering... Suppose you have an AY triangle of paid losses separated by 12 mo development intervals. If you are trending the paid losses to current period, each accident year before your current year warrants an additional 1 year's worth of trend.

Why do you not trend losses by CY? I mean, if you think of a simple example where the only factor in trend is inflation, and if you have for AY 2000 at 120 dev month a new incremental paid loss of 1M, does that really warrant 11 years' worth of inflation to apply to that 1M given that the paid loss actually occurred only a year ago in 2010?

I'm assuming the answer lies in the fact that trend doesn't involve soley inflation, but I'd like a different view point if anyone could provide one.


Vorian Atreides
02-29-2012, 09:28 AM
I would suggest re-reading pages 39-43 of Friedland wrt the advantages of AY aggregation (and disadvantages of CY aggregation).

Also, if the situation you mention about payments being paid 10 years (or more) after the end of the AY (and it exists, we have an open WC claim that has been on the books since the 1980's) should be recognized that inflation/trend may need to be applied in a more complex way for some of the longer-tailed lines (which are likely beyond the scope of the current syllabus, so I wouldn't worry about it for the Exam).

But for purposes of the Exam, understand that nearly all property-related lines (like Auto Physical Damage, Homeowners Property) are short-tailed lines and payments are unlikely beyond a few years after the AY.

And for long-tail lines, consider whether or not that $1M payment is "significant" to the overall book. If it is, and this is an isolated case, then remove that payment from the data and deal with it separately before adding it back in to the final analysis. If it's not all that significant (e.g., your reserves are over $500M), then the inclusion/exclusion of the payment will likely be immaterial to the overall analysis.

(FWIW, this could be a good Exam question on the "Synthesis" level of Bloom's.)

02-29-2012, 10:28 AM
You have a little overlap fallacy there.

You're trending the 1M loss to the projected year at the 120 month evaluation. So if you are trending to AY 2011, you are trending the 1M to be the projected amount paid/incurred in CY 2021.

Vorian Atreides
02-29-2012, 10:47 AM
Good point.

Overlap fallacy is a better explaination than what I gave.

02-29-2012, 11:15 AM
okay, I can see what you're both saying. I guess it's still tough for me to wrap my head around it in terms of how that 1M affects our ultimate projection for that AY 2000 in our example.

suppose you know losses are at ultimate at dev mo 132. so, by 2011 our AY 2000 losses will be considered at ultimate. I just can't really grasp how losses paid in CY 2010 need to get 11 years worth of trend to project the ultimate AY 2000 losses, knowing that that 1M is only 1 year away from ultimate. again, this is just me thinking from an inflation stand-point.

Vorian Atreides
02-29-2012, 11:36 AM
What are trying to estimate by trending AY 2000 data to 2011 levels?

Hint: it's not AY 2000 losses.

Are you looking at Ch 8 in Friedland?

02-29-2012, 06:49 PM
Don't think of trending everything to CY 2011.

AY2000 @ 12 months (CY2000) -> 11 years of trend -> AY2011 @ 12 months (CY2011)
AY2000 @ 120 months (CY2010) -> 11 years of trend -> AY2011 @ 120 months (CY2021)

The 11 years has 2 parts: 1 year to bring it to 2011 cost level, 10 years to bring it to the CY where that development comes in

03-09-2012, 11:25 PM
If you're still struggling, try it out in Excel. Build a triangle using the impossible scenario of a constant book of exposures with identical loss experience in every calendar period with the exception of CY trend. Then see how many years you have to trend the ultimates for each accident year to get a column of identical trended ultimates.