Actuarial Outpost
 
Go Back   Actuarial Outpost > Exams - Please Limit Discussion to Exam-Related Topics > SoA/CAS Preliminary Exams > Exam PA: Predictive Analytics
FlashChat Actuarial Discussion Preliminary Exams CAS/SOA Exams Cyberchat Around the World Suggestions


Fill in a brief DW Simpson Registration Form
to be contacted when new jobs meet your criteria.


Reply
 
Thread Tools Search this Thread Display Modes
  #251  
Old 06-02-2020, 09:31 AM
freddowell freddowell is offline
Member
SOA
 
Join Date: Jul 2017
Posts: 80
Default

At the end of the ACTEX manual, it says two original practice exams will be released. Does anybody know where to find those?
__________________
Exams: P, FM, MFE, LTAM*, C, PA*

VEEs: Economics, Corporate Finance, Applied Statistics

FAP: 1, 2, 3, 4, 5, IA, 6, 7, 8, FA

APC
Reply With Quote
  #252  
Old 06-02-2020, 03:39 PM
Gmrgms Gmrgms is offline
SOA
 
Join Date: Jan 2020
College: Valparaiso University
Posts: 6
Default

Quote:
Originally Posted by freddowell View Post
At the end of the ACTEX manual, it says two original practice exams will be released. Does anybody know where to find those?
Took me a minute to figure this out when I went to find them too.
Go to actuariallearning.com, login, click on "ACTEX Study Manual for SOA Exam PA, Spring 2020 [R Markdown Files for Printed Customers]" and you will find the Rmd files for the manual, practice exams, and commentary from the December 2019 exam.
Reply With Quote
  #253  
Old 06-02-2020, 06:26 PM
Yossarian Yossarian is offline
Member
SOA
 
Join Date: Jun 2011
Location: SoCal
Posts: 140
Default

Quote:
Originally Posted by Gmrgms View Post
Took me a minute to figure this out when I went to find them too.
Go to actuariallearning.com, login, click on "ACTEX Study Manual for SOA Exam PA, Spring 2020 [R Markdown Files for Printed Customers]" and you will find the Rmd files for the manual, practice exams, and commentary from the December 2019 exam.
Only Sample Exam #1 is out.

Sample Exam #2 will be released June 8th.

From Ambrose, post from a few days ago:
"Default Practice Exam 2
Latest update: Practice Exam 2 and some supplementary notes will be released on June 8 (Monday). This practice exam will provide excellent last-minute preparation for those who will take the exam on June 16-19."
Reply With Quote
  #254  
Old 06-02-2020, 06:42 PM
ambroselo ambroselo is offline
Member
SOA
 
Join Date: Sep 2018
Location: Iowa City
College: University of Iowa
Posts: 308
Default

Quote:
Originally Posted by freddowell View Post
At the end of the ACTEX manual, it says two original practice exams will be released. Does anybody know where to find those?
All supplementary files of this manual are posted on actuariallearning.com...

Quote:
Originally Posted by ambroselo View Post
I wanted to give everyone an update: I have just finished writing Practice Exam 1, which will be available on actuariallearning.com shortly.
Reply With Quote
  #255  
Old 06-02-2020, 06:45 PM
ambroselo ambroselo is offline
Member
SOA
 
Join Date: Sep 2018
Location: Iowa City
College: University of Iowa
Posts: 308
Default

Quote:
Originally Posted by Yossarian View Post
Hi Ambrose,

ACTEX Sample Exam 1 Task 6

The model solution does not appear to mention the Poisson distribution. Isn't overdispersion important because the mean equals variance in a Poisson distribution?

Would this be a valid response to Task 6, or am I misunderstanding something:

"Overdispersion can be an issue when using a Poisson distribution in modeling. The Poisson distribution assumes mean is equal to the variance for the target variable, so if variance is much higher than the mean, you have overdispersion. As it applies to clm_count, the mean is .0698 and the variance is 0.0757, so overdispersion does not appear to be a significant problem."
Your response is well-written () and it is really a good idea to mention the Poisson distribution when describing overdispersion. Even though the choice of the target distribution is the theme of the next task, overdispersion is closely related to the Poisson distribution.
Reply With Quote
  #256  
Old 06-02-2020, 08:04 PM
sowamazing97 sowamazing97 is offline
SOA
 
Join Date: Jan 2020
College: Nebraska
Posts: 18
Default

Hi Ambrose,

Why in the June 2019 PA exam do we not binarize the categorical variables before using stepAIC()? I thought that was standard but I was surprised it wasn't really mentioned in the SOA solution or your comments. Was there something different about this situation that I am missing or do we not use binarization on PA unless specifically told to?

Thanks in advance!
Reply With Quote
  #257  
Old 06-03-2020, 01:35 AM
debratee debratee is offline
SOA
 
Join Date: May 2020
College: University of Michigan, Ann Arbor
Posts: 10
Default

Hi Dr Ambrose,

For December 2019 exam, I noticed that in the end, the model was not fitted to the full dataset in the solution and also in your commentary. When I practiced using this exam, I performed Task 11, as well as my executive summary using the model fitted to the full dataset. Is that acceptable? What would you recommend? I thought we always need to fit the model to the full dataset in the end, to simulate the real life situation where we would be using new unseen data for predictions?
Reply With Quote
  #258  
Old 06-03-2020, 04:23 AM
ambroselo ambroselo is offline
Member
SOA
 
Join Date: Sep 2018
Location: Iowa City
College: University of Iowa
Posts: 308
Default

Quote:
Originally Posted by sowamazing97 View Post
Hi Ambrose,

Why in the June 2019 PA exam do we not binarize the categorical variables before using stepAIC()? I thought that was standard but I was surprised it wasn't really mentioned in the SOA solution or your comments. Was there something different about this situation that I am missing or do we not use binarization on PA unless specifically told to?

Thanks in advance!
This was mentioned in the model solution, which says that "Some candidates binarized the factor variables so that the stepwise selection could work with individual factor levels. This was not required and candidates did not gain or lose points for doing so." The focus of the task is on the selection process (forward vs. backward) and selection criterion (AIC vs. BIC).
Reply With Quote
  #259  
Old 06-03-2020, 01:48 PM
Yossarian Yossarian is offline
Member
SOA
 
Join Date: Jun 2011
Location: SoCal
Posts: 140
Default

Hi Ambrose,

(edit: similar to Debratee's question above, #257)

In Task 8 of Sample Exam #1:

Once the model has been tested, it would seem to be a valid step to re-run the model using the full binarized data set in order to get the coefficients used in the write-up.

I would not want to lose points for not doing that, but I also would not to unnecessarily waste time re-running. So ... do you think it's safe to use the coefficients derived from the training set? How to discern when the SOA thinks it's a good practice to use the full data set for coefficients and not?
Reply With Quote
  #260  
Old 06-03-2020, 02:34 PM
windows7forever windows7forever is offline
Member
SOA
 
Join Date: Apr 2016
Posts: 530
Default

I have read your exam analysis on June and December 2019 exams and their take-aways as well as the official solutions. I want to know if it's acceptable to just paste R output into word document on the target variable and predictors that need transformation, level combination, and missing values removal only. For all the rest predictors, can we just comment them briefly in bullet points without R output?

That will be the game changer to save at least an hour to answer a few tasks later on. When I did the similar way as SOA solution had in December 2019 exam, I spent over 1 hour on each of Tasks 1 and 2 that added a lot pressure to complete the remaining tasks within the 5 hour limit. I also did more variable transformations than the solution did.

1. Is it common that we should not log transform age as a predictor even if it shows right skewness? If we do log transformation on age, do we have to remove the original age variable? I am skeptical to do so since there's no collinearity between age and log(age).

2. I also decide to merge the Group NA to Group 2 in occupation, because Group NA only represents 4% of total observations that's the least among all levels. I consider Group NA as a sparse level and merge it with Group 2 due to similar mean of target or proportion of high value individuals.


In addition, for Task 2's best tree selection, SOA tested several scenarios and commented the tree with 5 branches. I got the idea that the interpretation focused on the level of interest on how changes in predictors affect the high value's prediction accuracy. However it would be easier to do a 4 branch tree and give same kind of interpretation.

Also, do you recommend to complete R coding on all tasks except market demo type prior to do writings? Will that save more time than do each question with coding and writing sequentially?

Other than first two tasks (I got 7 and 10), I really did not have issue to complete other tasks within the time limit on the points they were assigned. I made some mistake earlier due to time pressure, so I asked you some weird questions on Tasks 10 and 11 earlier. When I redid both versions of exams, I could work out all questions similar to the solution.

Almost all the reasoning behind each task can be found somewhere in your manual.

Thank you.

Last edited by windows7forever; 06-03-2020 at 02:37 PM..
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 06:28 AM.


Powered by vBulletin®
Copyright ©2000 - 2020, 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 0.32468 seconds with 12 queries