Actuarial Outpost
 
Go Back   Actuarial Outpost > Actuarial Discussion Forum > Property - Casualty / General Insurance
FlashChat Actuarial Discussion Preliminary Exams CAS/SOA Exams Cyberchat Around the World Suggestions

Salary Surveys
Property & Casualty, Life, Health & Pension

Health Actuary Jobs
Insurance & Consulting jobs for Students, Associates & Fellows

Actuarial Recruitment
Visit DW Simpson's website for more info.
www.dwsimpson.com/about

Casualty Jobs
Property & Casualty jobs for Students, Associates & Fellows


Reply
 
Thread Tools Search this Thread Display Modes
  #1  
Old 12-24-2018, 05:40 AM
bigalxyz bigalxyz is offline
 
Join Date: Aug 2013
Posts: 14
Default Selection of Time Periods in GLM Pricing Work

Hello forum,

Hoping to find a bit of guidance. I’m in the middle of a GLM pricing project on a (fairly small) portfolio of household insurance. I’m quite experienced with GLMs as a pricing tool, although most of my experience relates to motor rather than household.

I’m working with quite a clean data set, and for the most part it behaves well in the GLM analysis and gives sensible results. I’d like more data to work with (who wouldn’t?!), but other than that it works well.

Except for one thing…the “time period” aspect of the models.

I’ve used underwriting year (inception year) as a time basis for the work. Ordinarily I would prefer to work on an accident period basis, but to do that would have meant additional IT development, which wasn’t practical because the time available for the work was very short.

Using underwriting year as a factor in my GLMs, there is a very strong trend over time – much, much more than could be explained away by claims inflation, etc. I do know that the make-up of the account has changed quite a lot over the years, so this could have affected claims experience (over and above what the other rating factors would predict, of course). It makes it tricky to select a “base period” from which I can roll things forward (claims inflation + loss development loadings, etc.) to produce rate recommendations for 2019.

After some discussions with the underwriting team I have only included data from the 2016 and 2017 underwriting years in the GLMs. 2018 is very immature of course (exposure and claims data are as at 31 October 2018) so I have ignored that. Volumes of data from 2015 and prior are quite small (the account has grown quickly).

Having created GLMs that I’m happy with, I rolled forward the rates into 2019 using assumptions about loss development and claims inflation, using 2017 as the base period. Fairly standard stuff I think. However, it would be possible for me to do the same thing using 2016 as the base period. Because of the strong time trend in the data, this makes a big difference to the outcome.

Does anyone have any views on the pros and cons of basing my work on 2016 or 2017? One might argue that 2017 is a little immature I suppose (eg when the data extract was created, everything was earned as at 31/10/18 so there was still unexpired risk from policies incepting in Nov and Dec 2017). On the other hand, in a portfolio that has changed a lot over time (underwriting philosophy, risk acceptance, distribution methods, etc.), I’m keen to use the most recent data as a (perhaps) more reliable guide to how the account will look in 2019.

The reason for my concern is that the final results of the analysis look a little out of balance: there are two separate GLMs (for two separate cover types). Looking at the competitive position of the implied rating structure, one looks too cheap and one looks too expensive. Changing the analysis to use 2016 as a base period instead of 2017 would largely fix this problem, but I’d like my choice of method to be based on a more solid foundation that just “it gives me a result that I like”!

Part of the answer here will be to talk some more to the underwriting team in the new year, to understand a little more about exactly what has changed in the account over time and how that might have affected claims experience. That may help me to interpret the model outputs better. But in the mean time I’m keen to hear from anyone else who has been in this position, to find out how they’ve approached things.

Thank you.

ps I’m in the UK and I suspect most readers of this will be in the USA, so apologies in advance for any unfamiliar terminology!

Last edited by bigalxyz; 12-24-2018 at 06:49 AM..
Reply With Quote
  #2  
Old 12-24-2018, 11:43 PM
AbedNadir's Avatar
AbedNadir AbedNadir is offline
Member
CAS SOA
 
Join Date: Mar 2014
Studying for enlightenment
Posts: 2,170
Default

You could weight the results judgementally. Also, figure out if the growth will continue and choose weights based on that. Gl
__________________
Reply With Quote
  #3  
Old 12-26-2018, 06:25 PM
therealsylvos's Avatar
therealsylvos therealsylvos is online now
Member
CAS
 
Join Date: Mar 2014
Posts: 16,924
Default

Quote:
I do know that the make-up of the account has changed quite a lot over the years, so this could have affected claims experience (over and above what the other rating factors would predict, of course).
Yea, impossible to say without knowing more about the change you mentioned. What's driving the change? Was it something you're own team did, or are you being adversely selected against?

If your book changed rapidly in 2016 and 2017, but you don't expect changes to continue in 2019, I would think just using 2017 data is fine. But given that you already think your data is thin, I'd try and dive deeper into what changed and try and come up with an appropriate adjustment for the 2016 data.
__________________
Quote:
Originally Posted by ao fan View Post
i like laying in the tube and pretending that i'm dead.
Reply With Quote
  #4  
Old 12-28-2018, 06:54 AM
bigalxyz bigalxyz is offline
 
Join Date: Aug 2013
Posts: 14
Default

Exploring a little more closely now. Volumes of business written jumped sharply in mid 2016 (deliberate decision to go for growth, it seems), and claims experience (in respect of business written in mid 2016 onwards) worsened.

This growth stopped in 2017 - volumes of business roughly constant during the year.

I think the way ahead here could be to rerun the GLM fits but using half-years as time periods in my GLMs, to allow me to keep 2016 H1 and 2016 H2 separate.
Reply With Quote
  #5  
Old 12-30-2018, 08:45 PM
mattcarp mattcarp is offline
Member
CAS
 
Join Date: Sep 2016
Studying for Exam 6
College: UC Berkeley
Posts: 334
Default

Analyze the trend with traditional methods then include in GLM as an offset?
__________________
P FM VEE MFE C S OC1 5 OC2 6 7 8 9
Reply With Quote
Reply

Tags
glm, pricing

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


All times are GMT -4. The time now is 06:13 PM.


Powered by vBulletin®
Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
*PLEASE NOTE: Posts are not checked for accuracy, and do not
represent the views of the Actuarial Outpost or its sponsors.
Page generated in 0.36606 seconds with 9 queries