Actuarial Outpost
 
Go Back   Actuarial Outpost > Exams - Please Limit Discussion to Exam-Related Topics > SoA/CAS Preliminary Exams > MAS-I
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 our jobs meet your skills and objectives.


MAS-I Old Exam S Forum

Reply
 
Thread Tools Search this Thread Display Modes
  #151  
Old 09-17-2019, 06:05 PM
Abraham Weishaus Abraham Weishaus is offline
Member
SOA AAA
 
Join Date: Oct 2001
Posts: 7,250
Default

Quote:
Originally Posted by The Disreputable Dog View Post
This seems trivial, but I can't seem to get a handle on it, and it's driving me nuts. "Quantiles" - how on earth do we talk about these with respect to observations in a sample?


Where yj is the jth largest unique observation in the sample and gj is the count of all observations (unique or otherwise) less than or equal to yj.

So if I have n=4, and I look at the third largest observation y3 in a sample where there are no ties (gj=3), this is the (3/(4+1)) = 0.60-quantile (or the 60th percentile).

Calling it "the sixtieth percentile" makes sense to me - it's denoting where 60% of the probability mass gets cut off. But how do you refer to the quantile? Is it the "zero point six quantile"? My inclination was to call it the "third quintile", with the understanding that each quintile picked up 20% of the probability mass, but I'm only running across sources referring to this type of thing as the "0.60-quantile". What does everyone here call it?
Yes, .6 quantile = 60 percentile. Divide your percentile number by 100 to get the corresponding quantile.

Quote:
I'm also struggling to reconcile this with ASM's description.

So again, if I have n=4, and I look at the third order statistic (k=3), this is the 3(4+1) = 15 - what? 15-quantile? Or did I read that wrong? Are we actually saying that x(k) is the kth (n+1)-quantile? So here we have the 3rd (5)-quantile (or the "third quintile")?
I should divide, not multiply: k/(n+1) quantile. I'll post an erratum.
Reply With Quote
  #152  
Old 09-17-2019, 06:09 PM
Abraham Weishaus Abraham Weishaus is offline
Member
SOA AAA
 
Join Date: Oct 2001
Posts: 7,250
Default

Quote:
Originally Posted by The Disreputable Dog View Post
Dr. Weishaus, problem 28.25 on pg. 389 is another where we're asked to find a percentile that lies between tied ordinal statistics. The solution says the 75th percentile is 4, but I'm thinking Tse would calculate it as 3.75.
Yes. I'm going to remove the smoothing part from the question. The original exam question based on the textbook used then used a simpler smoothing method - they had 10 observations, with the first one being the 10th percentile, the second the 20th percentile, etc. So it's not the first time I'm changing to question to the current textbook's method.
Reply With Quote
  #153  
Old 09-19-2019, 10:32 AM
Acebelladona Acebelladona is offline
Member
CAS
 
Join Date: Apr 2019
College: Georgia State University, Alumni
Posts: 51
Default

Likely a dumb question, but something I was curious about. I'm working through the Time Series portion of Mahler's manual and I came to a problem similar to 5/18 Q44.
You're given a fitted AR(1) model along with the Mean Squared Error and asked to calculate the two-step ahead forecast error.

With the assumption that the first term of the series is known, I understand the procedure for solving this problem; the variance of the fitted value is driven by the variance of the white noise terms but I was thrown off by being given the MSE.

Is the thought process here that the white noise terms should be equivalent to a residual error series, the variance of which is equal to the MSE?
Reply With Quote
  #154  
Old 09-29-2019, 03:41 PM
Acebelladona Acebelladona is offline
Member
CAS
 
Join Date: Apr 2019
College: Georgia State University, Alumni
Posts: 51
Default

How is everyone feeling at this point? I took my first practice exam yesterday (Fall 2018) and just barely passed.

I feel like so long as I keep on top of doing practice problems, drilling flashcards, and a few more practice exams, I should be able to pull out a pass assuming the real thing doesn't throw out too many wildcards.

Coming to this after taking Exam 5 it doesn't feel anywhere even remotely close in terms of difficulty and the amount of advanced critical thinking required. The primary roadblock to getting a pass seems to just be remembering all the different formulas and more advanced concepts.
Reply With Quote
  #155  
Old 10-02-2019, 09:39 PM
sant93 sant93 is offline
Member
CAS
 
Join Date: Jan 2019
Posts: 32
Default

Not feeling good. its my 3rd time around... I do practice exams and score very well.. around 75%-80%. But when the exam day comes... I just cant perform...its very frustrating...sometimes i feel at a loss...my hours and hours of practice does not pay off... how is everyone else doing?
Reply With Quote
  #156  
Old 10-03-2019, 05:07 PM
SinisterRobert SinisterRobert is offline
Member
CAS
 
Join Date: Apr 2015
Location: Texas
Studying for MAS-I
College: University of Texas at Austin
Favorite beer: Karbach Blood Orange Radler
Posts: 94
Default

I'm feeling pretty good about my chances, I've studied for this exam for more time and more consistently than any other exam that I've passed.

I'm going back through Mahler practice exams and marking specific questions that I get wrong under exam conditions.
A few topics that I'm pretty weak on are MVUE, and the different types of GLM models like 'cumulative logit' models that are very hard for me to tell apart.
I also struggle with some of the more complicated order statistics stuff and maximum likelihood on grouped data due to algebraic complexity, but those problems are infrequent and I can handle simple questions on them.

Good luck to everyone during the crunch time!
Reply With Quote
  #157  
Old 10-04-2019, 01:21 AM
Copper525's Avatar
Copper525 Copper525 is offline
CAS
 
Join Date: Aug 2019
Location: Chicago
Studying for MAS-I
Posts: 2
Default

Quote:
Originally Posted by sant93 View Post
Not feeling good. its my 3rd time around... I do practice exams and score very well.. around 75%-80%. But when the exam day comes... I just cant perform...its very frustrating...sometimes i feel at a loss...my hours and hours of practice does not pay off... how is everyone else doing?
I feel the same... I really need to pass this time, but man, is this exam is hard. Took my first Mahler exam and got a 4, hopefully a month more of them will do me good.
__________________
1[✔] 2[✔] 3F[✔] MAS-I[] MAS-II[ ] 5[ ] 6[ ] 7[ ] 8[ ] 9[ ]

VEE: Accounting and Corporate Finance[✔] Economics[✔]

Online Course: 1[✔] 2[ ]

Course on Professionalism[ ]
Reply With Quote
  #158  
Old 10-04-2019, 07:30 AM
stan1236 stan1236 is offline
CAS SOA
 
Join Date: Sep 2018
College: Waterloo
Posts: 7
Default Maximum Covered Loss vs. Policy Limits in MLE

I'm having some trouble wrapping my head around how Maximum Covered Loss, Deductibles and Policy Limits work together when calculating MLE. I'm referencing Mahler Practice Exam #1, question 7 here. http://www.howardmahler.com/Teaching...S1Exam%231.pdf

I know maximum covered loss (MCL) as policy limit (u) - deductible (d). So for the two losses coming from the segment with MCL = 25,000 and d = 1000, i would've calculated their contribution to the likelihood function as S(24,000)^2, i.e. two losses above the policy limit of 24,000. In the solutions though, their contribution to the likelihood function is [S(25,000)/S(1,000)]^2. Can someone please explain to me why this is in an intuitive way?
Reply With Quote
  #159  
Old 10-04-2019, 08:47 AM
Abraham Weishaus Abraham Weishaus is offline
Member
SOA AAA
 
Join Date: Oct 2001
Posts: 7,250
Default

The 2 losses are for amounts above 25000. If there were no deductible, their likelihoodx would be S(25000). However, due to the deductible, you receive no information about losses below 1000, so your observation of these two losses is conditional on them being greater than 1000. By the usual formula for conditional probability, their likelihoods are therefore Pr(X>25000 and X>1000)/Pr(X>1000), which is S(25000)/S(1000).
Reply With Quote
  #160  
Old 10-04-2019, 05:20 PM
The Disreputable Dog's Avatar
The Disreputable Dog The Disreputable Dog is offline
Member
CAS
 
Join Date: Dec 2011
Studying for MAS-1
College: Somersby School
Favorite beer: Bulldog Root Beer
Posts: 1,041
Default

Quote:
Originally Posted by ASM pg. 431
For a binomial distribution, the maximum likelihood estimate of mq is the sample mean. Given m, the maximum likelihood estimate of q is x-bar/m.
Instead of x-bar, should this be the sum of the xi?

Maybe I'm misunderstanding what x-bar is for binomial results. If you flipped 10 coins and got 5 heads, and then flipped 10 coins again and got a 7, is x-bar 12/2 = 6? Or is it 12/20 = 0.6? I had assumed the former.
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 10:00 AM.


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 0.32062 seconds with 11 queries