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


Upload your resume securely at https://www.dwsimpson.com
to be contacted when new jobs meet your skills and objectives.


Reply
 
Thread Tools Search this Thread Display Modes
  #1  
Old 06-13-2018, 04:18 PM
Marco_park Marco_park is offline
CAS
 
Join Date: Jan 2018
Posts: 9
Default Ratemaking Model

Hi All,

we are trying to assess the financial impact deriving from an improvement in a ratemaking model.

In particular we are moving from a standard GLM approach towards a more sophisticated one using Machine Learning.

We have also estimated the increase of the Gini index of both, the current and the proposed pricing approach.

Our doubts are related to how to quantify the financial impact in terms of possible monetary gain due to the model change.

We are aware that our proposed model is more precise, but how to quantify the possible financial impact from a monetary point of view?

Is there any theoretical and practical framework to address this?
Reply With Quote
  #2  
Old 06-14-2018, 08:53 AM
sticks1839 sticks1839 is offline
Member
CAS AAA
 
Join Date: Jul 2005
Posts: 1,600
Default

Think about what will happen if/when you implement the new model. Some customers will see prices go up and some will see prices go down. Assuming they are price elastic, you should see a mix shift in your book. If the new model is truly better, then that mix shift will be favorable because it will lessen the opportunity for adverse selection.

You can calculate this impact by making assumptions about that elasticity and the future state of your book. Overall, whether via mix or price corrections, you should see an improved loss ratio.
Reply With Quote
  #3  
Old 06-18-2018, 12:29 PM
DeepPurple's Avatar
DeepPurple DeepPurple is offline
Member
 
Join Date: Jun 2004
Posts: 4,154
Default

I'm curious about the answers too...

When I have heard practicioners talk about this subject, they use the word "lift" to convey a better, less biased rating plan.

Is there a consensus on the calculation of "lift" or is that a relative term that is just bandied about?


And how is the economic principal of Gini Index used for a goodness of fit test on a rating plan?
__________________
Come on. Let's go space truckin'. Come on!
Reply With Quote
  #4  
Old 06-18-2018, 01:17 PM
Vorian Atreides's Avatar
Vorian Atreides Vorian Atreides is offline
Wiki/Note Contributor
CAS
 
Join Date: Apr 2005
Location: As far as 3 cups of sugar will take you
Studying for ACAS
College: Hard Knocks
Favorite beer: Most German dark lagers
Posts: 62,738
Default

Gini Index shouldn't be used as a "goodness of fit" statistic, IMO.

It would be a good measure of how well a rating plan/model differentiates risk.

you can find a model with a very good fit stastic, but a very poor Gini Index.
__________________
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
  #5  
Old 06-18-2018, 01:22 PM
Vorian Atreides's Avatar
Vorian Atreides Vorian Atreides is offline
Wiki/Note Contributor
CAS
 
Join Date: Apr 2005
Location: As far as 3 cups of sugar will take you
Studying for ACAS
College: Hard Knocks
Favorite beer: Most German dark lagers
Posts: 62,738
Default

Quote:
Originally Posted by Marco_park View Post
Hi All,

we are trying to assess the financial impact deriving from an improvement in a ratemaking model.

In particular we are moving from a standard GLM approach towards a more sophisticated one using Machine Learning.

We have also estimated the increase of the Gini index of both, the current and the proposed pricing approach.

Our doubts are related to how to quantify the financial impact in terms of possible monetary gain due to the model change.

We are aware that our proposed model is more precise, but how to quantify the possible financial impact from a monetary point of view?

Is there any theoretical and practical framework to address this?
How, exactly, is the "new" model an improvement over the "old" model?

And what, exactly, do you mean by the phrase, "our proposed model is more precise"? More precise in what way?


Do you have any models that address price sensitivity on conversion of new business and/or lapse/retention of current business?
__________________
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
  #6  
Old 06-19-2018, 09:25 AM
sticks1839 sticks1839 is offline
Member
CAS AAA
 
Join Date: Jul 2005
Posts: 1,600
Default

Quote:
Originally Posted by DeepPurple View Post
I'm curious about the answers too...

When I have heard practicioners talk about this subject, they use the word "lift" to convey a better, less biased rating plan.

Is there a consensus on the calculation of "lift" or is that a relative term that is just bandied about?
Generally lift is simply the spread of your modelled metric, like pure premium or loss ratio, across a defined grouping like deciles. So lift would be the worst decile average over the best decile average. Similar to Gini in that it represents the ability of the model to differentiate between high/low risks. It's meaning, like many model related terms, is only useful in comparison to other models calculated in a similar manner.
Reply With Quote
  #7  
Old 06-19-2018, 02:38 PM
DeepPurple's Avatar
DeepPurple DeepPurple is offline
Member
 
Join Date: Jun 2004
Posts: 4,154
Default

Quote:
Originally Posted by sticks1839 View Post
Generally lift is simply the spread of your modelled metric, like pure premium or loss ratio, across a defined grouping like deciles. So lift would be the worst decile average over the best decile average. Similar to Gini in that it represents the ability of the model to differentiate between high/low risks. It's meaning, like many model related terms, is only useful in comparison to other models calculated in a similar manner.
If your old model undercharges low risks and overcharges high risks, then a better, improved model would then give lower lift. No?
__________________
Come on. Let's go space truckin'. Come on!
Reply With Quote
  #8  
Old 06-19-2018, 02:46 PM
Heywood J Heywood J is online now
Member
CAS
 
Join Date: Jun 2006
Posts: 4,040
Default

Quote:
Originally Posted by DeepPurple View Post
If your old model undercharges low risks and overcharges high risks, then a better, improved model would then give lower lift. No?
No. Lift measures the spread of actual losses when ordered by predicted losses. It does not measure the spread of predicted losses when ordered by predicted losses.
Reply With Quote
  #9  
Old 06-19-2018, 03:31 PM
DeepPurple's Avatar
DeepPurple DeepPurple is offline
Member
 
Join Date: Jun 2004
Posts: 4,154
Default

Yeah, ok. Thanks HJ, I get it.

Although, I would still note that you could potentially make great strides in the middle classifications without affecting your lift if your lift is based on the top decile relative to the bottom decile.


I don't do this (ratemaking design), but my actuarial gut instinct would be to define a better plan by reducing the mean squared error between the actual vs predicted losses, weighted across all insureds.
__________________
Come on. Let's go space truckin'. Come on!

Last edited by DeepPurple; 06-19-2018 at 03:41 PM..
Reply With Quote
  #10  
Old 06-19-2018, 06:10 PM
Heywood J Heywood J is online now
Member
CAS
 
Join Date: Jun 2006
Posts: 4,040
Default

Quote:
Originally Posted by DeepPurple View Post
Yeah, ok. Thanks HJ, I get it.

Although, I would still note that you could potentially make great strides in the middle classifications without affecting your lift if your lift is based on the top decile relative to the bottom decile.
If you use Ginis to measure your lift, the middle improvement will be reflected. I don't think anyone actually measures lift merely by comparing the last bucket to the first bucket, that's just an oversimplification for the sake of quick explanation.
Quote:
I don't do this (ratemaking design), but my actuarial gut instinct would be to define a better plan by reducing the mean squared error between the actual vs predicted losses, weighted across all insureds.
MSE is not a good measure of deviation of actual from predicted in insurance applications, since insurance outcomes tend to be very skewed.
Reply With Quote
Reply

Tags
machine learning, pricing, ratemaking

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 01:44 PM.


Powered by vBulletin®
Copyright ©2000 - 2018, 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 0.31894 seconds with 9 queries