Actuarial Outpost
 
Go Back   Actuarial Outpost > Actuarial Discussion Forum > General Actuarial
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.


General Actuarial Non-Specific Actuarial Topics - Before posting a thread, please browse over our other sections to see if there is a better fit, such as Careers - Employment, Actuarial Science Universities Forum or any of our other 100+ forums.

Reply
 
Thread Tools Search this Thread Display Modes
  #1171  
Old 06-14-2019, 01:10 PM
Arroway's Avatar
Arroway Arroway is offline
Member
CAS SOA AAA
 
Join Date: Oct 2011
Location: In a house
Studying for Done, bois!
College: UConn
Favorite beer: Barq's or Mug, or any local Root Beer with caffeine
Posts: 1,185
Default

Quote:
Originally Posted by Vorian Atreides View Post

But far more important, understanding DS techniques does not guarantee that the practitioner will have a qualitative understanding of the underlying data and potential (or actual) relationships between variables within the dataset.

Nor does it guarantee an understanding of how those relationships may change (or drift) over time.
100% agree. At a prior company, we were approached by data scientists who had just finished a consulting project doing similar work for Procter & Gamble. They had heard about predictive analytics, and assumed the skills transferred to insurance easily. Yay, a new source of consulting revenue!

So, they offered to give us a freebie. They would look at data for one of our business units and use their skills to help us understand the most important underwriting variables. Then when they were successful, we'd hire them to do the rest of our business units, if we liked their work. We gave them data on our Legal Professional Liability book and they set to work.

When it came time to present their results they were very excited to tell us they had found a very useful variable in predicting profit -- and that variable was policy year! I had to explain to them what policy year meant (they didn't know) and that if I could accomplish time travel, underwriting insurance policies would not be my first thought of what to do with that skill.

And that's the difference between a data scientist and an actuary.
__________________
Arroway, Injustice 2 Raid Leader

All opinions are either my own, or made up for dramatic and/or comedic effect.

Reply With Quote
  #1172  
Old 06-14-2019, 01:26 PM
act_123 act_123 is offline
Member
CAS
 
Join Date: Dec 2013
Posts: 4,841
Default

Quote:
Originally Posted by Arroway View Post
Good points. It's been "a minute" since I took exams (ok, decades) and I remember never using all of the allotted time, or at least never running out of time. I can only think of one exception (Life Contingencies, which was an SoA exam anyway, before I switched) where more time would have made a difference, and in that case it was only because I sat for the last hour trying to remember how to do one problem and couldn't remember, then suddenly remembered with not quite enough time to actually do the calculations. I got a 5 on that exam and I'm still convinced that one question would have gotten me a 6.

I also don't remember being an exception, but that's more subjective. So maybe the length has become more of a problem since I finished the exams.

It also comes down to how they set the pass mark. If you take the assumption that from sitting to sitting the candidate pool is equally prepared, passing the top x% makes some sense. I know that's not how they do it, but in that case it wouldn't matter why you left it blank, because the less prepared students would have left even more blank (on average). But it's definitely less "clean" when a blank has two different meanings.
I agree with this. If they had a set percentage, I would be fine with that. I think the mqc is a bit all over the place.

I am sure CAS has proofs showing how it is a great measure. They aren't too transparent about it. They also don't always accommodate for difficult tests. In the past I thought they did a pretty good job. However, last sitting they seemed to throw that out the window.
__________________
ACAS 7 8 9
Reply With Quote
  #1173  
Old 06-14-2019, 01:29 PM
bjc2142's Avatar
bjc2142 bjc2142 is offline
Member
CAS
 
Join Date: Aug 2014
Location: Boston
Studying for Python
Favorite beer: Asahi Draft
Posts: 1,193
Default

Quote:
Originally Posted by Arroway View Post
When it came time to present their results they were very excited to tell us they had found a very useful variable in predicting profit -- and that variable was policy year!
__________________
FCAS - Hawaii 2019
Reply With Quote
  #1174  
Old 06-14-2019, 01:34 PM
Lucy's Avatar
Lucy Lucy is offline
Member
 
Join Date: Sep 2001
Posts: 2,643
Default

Quote:
Originally Posted by bjc2142 View Post
If you are from P&C, could you name few of those non-DS technical areas besides what I have here if I'm missing something?

(excluding things that are considered not too technical and done in Excel)
1. excess/reinsurance pricing using distributions and simulations
2. stochastic reserving
3. capital modeling
4. cat modeling (probably not anymore if running vendor software is job)

Most of technical work exclusive to actuaries seem to be stochastic stuff. That's where DS presence is weak, and also explains life actuaries seem to be less threatened by DS profession.
Quote:
Originally Posted by Vorian Atreides View Post
Any frequency model ("Will a claim occur?" as well as "How many claims for this risk/risk-profile?").

Large loss propensity model (related, but different from excess pricing).


But far more important, understanding DS techniques does not guarantee that the practitioner will have a qualitative understanding of the underlying data and potential (or actual) relationships between variables within the dataset.

Nor does it guarantee an understanding of how those relationships may change (or drift) over time.

A key part of actuarial ratemaking is to take historical data and adjust it to reflect what is expected in the future timeframe. This adjustment is 97% subjective and relies heavily on several assumptions. This is the part that most DS practitioners lack.
Also estimating the impact of observed claim statistics of changes in claim department practices, working with experienced underwriters to select credibility formulas that will work in weird situations, generally understanding what factors are likely to be influencing results so as to develop models (GLMs or simple models) that reflect actual drivers. Being the subject matter expert in a modeling project that includes data scientists is a technical role but not a data science role.

Quote:
Originally Posted by Arroway View Post
...They would look at data for one of our business units and use their skills to help us understand the most important underwriting variables.
.. When it came time to present their results they were very excited to tell us they had found a very useful variable in predicting profit -- and that variable was policy year!...

Thus, the value of subject matter experts.
Reply With Quote
  #1175  
Old 06-14-2019, 01:54 PM
yoyo's Avatar
yoyo yoyo is offline
Member
CAS
 
Join Date: Dec 2001
Posts: 23,703
Default

Quote:
Originally Posted by Vorian Atreides View Post
Any frequency model ("Will a claim occur?" as well as "How many claims for this risk/risk-profile?").

Large loss propensity model (related, but different from excess pricing).


But far more important, understanding DS techniques does not guarantee that the practitioner will have a qualitative understanding of the underlying data and potential (or actual) relationships between variables within the dataset.

Nor does it guarantee an understanding of how those relationships may change (or drift) over time.

A key part of actuarial ratemaking is to take historical data and adjust it to reflect what is expected in the future timeframe. This adjustment is 97% subjective and relies heavily on several assumptions. This is the part that most DS practitioners lack.
used DS skills to get this #
Reply With Quote
  #1176  
Old 06-14-2019, 01:56 PM
yoyo's Avatar
yoyo yoyo is offline
Member
CAS
 
Join Date: Dec 2001
Posts: 23,703
Default

similar to Arroway's story, we had an outfit tell us that falls from height was a significant severity driver (our specialty is residential construction)
Reply With Quote
  #1177  
Old 06-14-2019, 02:45 PM
NAMAK NAMAK is offline
Member
CAS
 
Join Date: Nov 2013
Posts: 273
Default

Quote:
Originally Posted by yoyo View Post
similar to Arroway's story, we had an outfit tell us that falls from height was a significant severity driver (our specialty is residential construction)
Maybe it's just me and I haven't been in the industry long enough or right companies, but I have yet to see a valuable model produced by the "modeling (DS) team" that was implemented in a reasonable amount of time and had a meaning impact on loss ratio.

Unless analysis is done by the business actuaries, I feel like it takes too long, results aren't relevant anymore, or doesn't add any insight.

This may change yet, but that has been my experience and I've been around a few of these projects.
Reply With Quote
  #1178  
Old 06-14-2019, 03:38 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: 66,064
Default

I work (indirectly) with a "DS" team and have worked to try to help them see the difference between their academic training and business-practicality.

They're a good group and open to input from the business area; but they're finding out just how important qualitative factors (regulatory constraints, cost-benefit trade off, etc.) come into the equation when evaluating "how good" a model is for a given business purpose.

As one illustration, they were concerned with how "stale" data was that is (essentially) PY 2015-2017. When I asked the individual how much we know about all of the risks for someone written in Dec 2018, after a 3 minute discussion, the light bulb went off (without my direct prompting).

Then there was a follow up discussion of "how much" data was needed to get a "reasonably full" distribution of losses.
__________________
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
  #1179  
Old 06-14-2019, 09:11 PM
yoyo's Avatar
yoyo yoyo is offline
Member
CAS
 
Join Date: Dec 2001
Posts: 23,703
Default

Quote:
Originally Posted by NAMAK View Post
Maybe it's just me and I haven't been in the industry long enough or right companies, but I have yet to see a valuable model produced by the "modeling (DS) team" that was implemented in a reasonable amount of time and had a meaning impact on loss ratio.

Unless analysis is done by the business actuaries, I feel like it takes too long, results aren't relevant anymore, or doesn't add any insight.

This may change yet, but that has been my experience and I've been around a few of these projects.
yup, the practice is still immature, but i expect that will improve over time
Reply With Quote
  #1180  
Old 06-15-2019, 04:58 PM
Marcie's Avatar
Marcie Marcie is offline
Member
CAS
 
Join Date: Feb 2015
Posts: 9,716
Default

Quote:
Originally Posted by Vorian Atreides View Post
I have no desire to become that poster as well.
This guy gets me.
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 12:32 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 0.23202 seconds with 12 queries