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


Not looking for a job? Tell us about your ideal job,
and we'll only contact you when it opens up.
https://www.dwsimpson.com/register


Reply
 
Thread Tools Display Modes
  #11  
Old 04-07-2017, 02:24 PM
examsarehard examsarehard is offline
Member
CAS
 
Join Date: May 2011
Posts: 486
Default

Quote:
Originally Posted by hellomath View Post
For these classes (high sev low freq), is it appropriate to move these classes to mean?
Well you asked about credibility models, so what would your complement of credibility be in this case? Absent of any other information, usually it will be the overall mean.

If instead you want your complement to be some specific (higher) value, you could add a dummy indicator variable that aligns with that class, and then fix the dummy variable with an offset.
Reply With Quote
  #12  
Old 04-07-2017, 02:56 PM
Vorian Atreides's Avatar
Vorian Atreides Vorian Atreides is online now
Wiki/Note Contributor
CAS
 
Join Date: Apr 2005
Location: Hitler's Secret Bunker
Studying for ACAS
College: Hard Knocks
Favorite beer: Sam Adams Cherry Wheat
Posts: 57,048
Default

Quote:
Originally Posted by hellomath View Post
For these classes (high sev low freq), is it appropriate to move these classes to mean?
"Movement to the (overall) mean" is one way to "approximate" capping losses for classes with higher overall experience.

But whenever you're dealing with a class with very little data available and you want to utilize credibility-based models to estimate the parameters, you're confronted with the decision of what to use as the complement of credibility.

As examsareeasyhard pointed out, you can create a separate parameter to use to regress toward than an overall mean.

One complement that you could consider is the current relativity for that class.

Another consideration might be to determine an approximate indicated relativity if you "spread out" the infrequent large losses over a longer time horizon (say, 5 years).
__________________
I find your lack of faith disturbing

Why should I worry about dying? Itís not going to happen in my lifetime!


Freedom of speech is not a license to discourtesy

#BLACKMATTERLIVES
Reply With Quote
  #13  
Old 04-07-2017, 06:06 PM
AMedActuary AMedActuary is offline
Member
SOA
 
Join Date: May 2007
College: UCLA Alumni
Posts: 352
Default

This could be handled in a Bayesian framework. If a "normal" market has a mean of X and you believe this other market may be 30% higher, could set the prior to Beta ~ N(1.3X, sigma). Sigma would be set based on how confident you are in the 30% figure (or to put it another way, how much you want the model to "listen" to the data).

Of course a normal distribution may not work either but there are others that can be used instead.
Reply With Quote
  #14  
Old 04-07-2017, 11:16 PM
Marcie's Avatar
Marcie Marcie is offline
Member
CAS
 
Join Date: Feb 2015
Posts: 4,440
Default

This thread is relevant to my interests.
Reply With Quote
  #15  
Old 04-10-2017, 10:33 AM
hellomath hellomath is online now
Member
 
Join Date: Feb 2007
Studying for relaxation
Posts: 4,026
Default

I need to read up on both the paper from examsarehard & the books from AMedActuary.

Still waiting on the Bayesian books from Amazon.

Any thoughts on the high-level differences b/ GLMM & Bayesian?
Reply With Quote
  #16  
Old 04-10-2017, 03:00 PM
examsarehard examsarehard is offline
Member
CAS
 
Join Date: May 2011
Posts: 486
Default

I believe a GLMM can be interpreted as a specific form of the Bayesian GLM where the prior distribution of the parameters is normally distributed with mean 0.

edit: Only the credibility-weighted parameters have normal priors. Any non-credibility-weighted parameters (like in a regular GLM) would have flat priors.

Last edited by examsarehard; 04-10-2017 at 03:59 PM.. Reason: clarity
Reply With Quote
  #17  
Old 07-13-2017, 03:47 PM
hellomath hellomath is online now
Member
 
Join Date: Feb 2007
Studying for relaxation
Posts: 4,026
Default

Quote:
Originally Posted by examsarehard View Post
If instead you want your complement to be some specific (higher) value, you could add a dummy indicator variable that aligns with that class, and then fix the dummy variable with an offset.


I *think* I did that, but the low cred class factors is the same between

1) let the glm solves for the class factor
2) add a dummy indicator variable that aligns with that class, and then fix the dummy variable with an offset

Last edited by hellomath; 07-13-2017 at 11:49 PM..
Reply With Quote
  #18  
Old 07-18-2017, 12:05 PM
v10 v10 is offline
Member
CAS
 
Join Date: Aug 2012
Posts: 33
Default

In R: the bayesglm() function in the arm package is awesome and much quicker than using STAN/JAGS for a simple bayesian GLM.

https://cran.r-project.org/web/packages/arm/arm.pdf
Reply With Quote
  #19  
Old 07-18-2017, 05:30 PM
AMedActuary AMedActuary is offline
Member
SOA
 
Join Date: May 2007
College: UCLA Alumni
Posts: 352
Default

I would also recommend 'rstanarm' for basic bayesian models. See the function 'stan_glm.'

https://CRAN.R-project.org/package=rstanarm
Reply With Quote
  #20  
Old 07-19-2017, 04:55 PM
v10 v10 is offline
Member
CAS
 
Join Date: Aug 2012
Posts: 33
Default

Quote:
Originally Posted by AMedActuary View Post
I would also recommend 'rstanarm' for basic bayesian models. See the function 'stan_glm.'

https://CRAN.R-project.org/package=rstanarm
Thanks for pointing this out AMedActuary!
Reply With Quote
Reply

Thread Tools
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 12:54 PM.


Powered by vBulletin®
Copyright ©2000 - 2017, 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 0.40233 seconds with 11 queries