Actuarial Outpost
 
Go Back   Actuarial Outpost > Actuarial Discussion Forum > Software & Technology
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
  #11  
Old 12-18-2017, 06:57 PM
DiscreteAndDiscreet DiscreteAndDiscreet is offline
Member
AAA
 
Join Date: May 2016
Posts: 466
Default

Quote:
Originally Posted by Guinness View Post
While you CAN try to predict the underwriting decision, I would not reccomend it (at all...its one of those things where actuaries instinctively want to do it this way...and some companies have...but it is the absolutely the wrong way to do it). It makes much more sense to try to predict the mortality rate directly than to try to predict the underwriting class. Why? First of all, underwriting classes are rather coarse. It makes sense to predict mortality as a continuous attribute so you don't lose any information. Second, and more importantly, traditional underwriting is often flawed. Predictive models will also be flawed, but in different ways, but if done right will be more accurate than traditional underwriting. It makes no sense to constrain your predictive model to match the flaws of traditional underwriting. If you want to predict mortality, then predict it as well as possible.
Even if the underwriter's decision is flawed, the underwriter could give a deposition outlining the basis for the underwriting decision. An underwriting decision that comes purely from a model that is optimized purely for predictive accuracy may not have a communicable rationale for individual decisions. This is an underlying problem with delegating decision-making authority to something that lacks legally recognized agency.

An actuary, on the other hand, is supposed to work from both/multiple ends of this kind of issue. However, I think a responsible actuary will tend not use a model that has a higher predictive accuracy on a held back sample of data if it contradicts traditional models and the actuary can't identify specifically what factors the new model uses that the traditional models don't.
Reply With Quote
  #12  
Old 12-18-2017, 07:05 PM
redearedslider's Avatar
redearedslider redearedslider is offline
Member
CAS
 
Join Date: Oct 2015
Posts: 13,307
Default

Quote:
Originally Posted by DiscreteAndDiscreet View Post
An actuary, on the other hand, is supposed to work from both/multiple ends of this kind of issue. However, I think a responsible actuary will tend not use a model that has a higher predictive accuracy on a held back sample of data if it contradicts traditional models and the actuary can't identify specifically what factors the new model uses that the traditional models don't.
To what extent should we be concerned with deploying a less accurate model because the one with higher accuracy on the holdout sample may identify predictive relationships that we don't understand?
__________________
Quote:
Originally Posted by Abraham Weishaus View Post
ASM does not have a discussion of stimulation, but considering how boring the manual is, maybe it would be a good idea.

Last edited by redearedslider; 12-18-2017 at 07:16 PM..
Reply With Quote
  #13  
Old 12-18-2017, 07:11 PM
DiscreteAndDiscreet DiscreteAndDiscreet is offline
Member
AAA
 
Join Date: May 2016
Posts: 466
Default

Quote:
Originally Posted by Chuck View Post
The mortality rate for an individual record is either 1 or 0 (over some period of time). It seems that to really predict a mortality "rate", by definition you must put the records into classes of some kind whether traditional or otherwise and then predict the rate on the classes. I'm sure there is something here I am not getting my head around.
Identical records should get the same rate and the underlying problem is identifying precisely which records should be regarded as identical.
Reply With Quote
  #14  
Old 12-18-2017, 07:47 PM
DiscreteAndDiscreet DiscreteAndDiscreet is offline
Member
AAA
 
Join Date: May 2016
Posts: 466
Default

Quote:
Originally Posted by redearedslider View Post
To what extent should we be concerned with deploying a less accurate model because it identifies predictive relationships that we don't understand?
What are your specific responsibilities in the task and can you demonstrate that they have been met even if the model underperforms after deployment?

Put another way, outside of accuracy, what are the witnesses to the service you rendered to those responsibilities and what steps have you taken to assure yourself that these witnesses will be accepted by a third party?

Last edited by DiscreteAndDiscreet; 12-18-2017 at 07:59 PM..
Reply With Quote
  #15  
Old 12-19-2017, 06:40 PM
DiscreteAndDiscreet DiscreteAndDiscreet is offline
Member
AAA
 
Join Date: May 2016
Posts: 466
Default

For typical work, actuarial reports with summaries of data that are related to the conclusions of the work constitute a form of proof certificates. If someone points out an error in the data summary, it may very well undermine significant conclusions of the report with the result that the report needs to be revised in order to be acceptable. The report also documents the scope of review and what considerations might be excluded from the analysis but which may ultimately prove to be relevant.

I'm not saying you can't do this type of thing for machine learning models, I'm mostly just saying that you need to give some thought about how you are going to cover these bases.
Reply With Quote
Reply

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 12:59 AM.


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 0.20463 seconds with 9 queries