Actuarial Discussion Forum

Actuarial Jobs from Actuary.com    Submit Your Actuarial Resume Anonymously
Actuarial Blog - Actuarial Employment Articles, Help, Jobs and News
Other Insurance Jobs    Other Financial Jobs    Other Health Jobs    Other IT Jobs    Other Jobs, Careers and Employment    Actuarial News
Directory of Actuarial Exam Study Courses - Online    Directory of Actuarial Exam Study Materials    Directory of Actuarial Exam Study Seminars - Live
Directory of Actuarial Recruiters    Directory of Actuarial Schools    Actuarial Grads Network    Actuary.com 



D.W. Simpson & Co, Inc. - Worldwide Actuarial Jobs
Life Jobs 
Health Jobs Pension Jobs Casualty Jobs Salary Apply
Pauline Reimer, ASA, MAAA - Pryor Associates
Nat'l/Int'l Actuarial Openings: Life P&C Health Pensions Finance
ACTEX Publications and MadRiver Books
Serving students worldwide for over 40 years
Advertise Here - Reach Actuarial Professionals
Advertising Information
Actuarial Careers, Inc. - Actuarial Jobs Worldwide
Search positions by geographic region, specialization, or salary
Ezra Penland Actuarial Recruiters - Top Actuarial Jobs
Salary Surveys  Apply Online   Bios   Casualty   Health   Life   Pension

+ Reply to Thread
Results 1 to 10 of 10

Thread: Loss Triangles - Forcasting

  1. #1
    Actuary.com - Level I Poster
    Join Date
    Sep 2006
    Location
    WV
    Posts
    9

    Question Loss Triangles - Forcasting

    Hi Everyone,

    I work in Worker's Compensation, I have passed 2 exams, and I am studying for the VEE exams. I am also the only actuary in my company, so help is not easy to come by. I have a question. I frequently create loss triangles, and I am learning to forcast those triangles using regression. However, the books do a poor job of explaining some of the formatting issues.

    I understand that you must be using incremental triangles, divide by the volume, and take the log of the values before the regression. I have two questions. First, when dividing by the volume, do you use # claims, # of policies, or does it matter. Second, how do you get back to monetary values after the regression? Do you simply take the exponent?

    I would appreciate anyone's help.

    Thanks,
    Chris Johnson

  2. #2
    Actuary.com - Level VI Poster Ken's Avatar
    Join Date
    Mar 2005
    Location
    Thailand
    Posts
    833
    I'm not sure what you're forecasting. Are you trying to find LDFs, tail factors, number of claims, number of policies, development of losses for recent accident years, amount of losses for future accident years...
    Whether you are the lion or the gazelle, when the sun comes up, you better be running.

  3. #3
    Actuary.com - Level I Poster
    Join Date
    Sep 2006
    Location
    WV
    Posts
    9
    Development of losses for recent accident years.

  4. #4
    Actuary.com - Level VI Poster Ken's Avatar
    Join Date
    Mar 2005
    Location
    Thailand
    Posts
    833
    Sherman and/or Diss have an article on the CAS website about how to fit an inverse power curve to your development factors I believe. Is that one of the "books" you're looking at and trying to mimic?
    Whether you are the lion or the gazelle, when the sun comes up, you better be running.

  5. #5
    Actuary.com - Level I Poster
    Join Date
    Sep 2006
    Location
    WV
    Posts
    9
    I got this article from NEAS -
    [URL="http://www.neas-seminars.com/discussions/download.aspx?id=1420&MessageID=6116"]http://www.neas-seminars.com/discussions/download.aspx?id=1420&MessageID=6116[/URL]
    This is what I am trying to do. Specifically, I am referring to the end of the article where it talks about using real data, taking the log, dividing by volume, etc. and using the lognormal dist to get the values back.

  6. #6
    You spam? I ban! Irish Blues's Avatar
    Join Date
    Mar 2005
    Location
    WI
    Posts
    1,303
    Ken: Sherman's paper (the one I think you're talking about) discusses how to fit an inverse power curve to get CLDF's given data obtained in various ways. It's something we explored in a project recently; the main problem with the example that fit our interests was that, after going through the steps to approximate and smooth, he fits one IPC to the entire range of data from 12-84 months. That's entirely inappropriate, especially for recent years - true development from 12-84 months doesn't fit any one curve appropriately over the entire range ... and we tried over a dozen curve shapes trying to find one.

    Chris: is there a specific reason why you're wanting to use regression for loss forecasting instead of the traditional method? In other words, do you have reason to believe that the typical development method is so inaccurate that you need to find another way to do this, or are you just trying this approach purely to see whether you can do it or not?

    I'm still not completely clear on what exactly you're shooting for as a final result. If this is better explained via PM, go ahead and shoot me one.
    "You better get to living, because dying's a pain in the ***." - Frank Sinatra

    [url]http://www.hockeybuzz.com/blogger_archive.php?blogger_id=174[/url] - where I talk about the Blues and the NHL.

  7. #7
    Actuary.com - Level I Poster
    Join Date
    Sep 2006
    Location
    WV
    Posts
    9
    Here is the story. I work in Worker's Compensation, and I am the only actuary. Most of our actuarial work is contracted out. I have passed exams 1 and 2, but working alone makes it difficult to get the exerience I need. Complicating things, we are a new private insurance company formed from a state comp system, so everything is starting over.

    I create many loss triangles. I create paid and incurred losses for medical, permanent partial, excess liability, etc on a quarterly basis. However, the anaylsis is being done by our contracted actuaries. I am now getting ready to take the statistics VEE exam, (I have a Masters in Math). I have come across articles on NEAS about using regression, and since this is required, I would like to be able to apply it. I came across the triangles done by my predecessor, and they were forecast using regression, but not done the way the articles describe.

    So, I guess my question is, is regression commonly used for this? This was my impression. In what other areas would you use regression. The past work I saw did not take the log of the values (paid amount), did not divide by the volume, and it was done on a cummulative triangle. All three I thought were wrong?

    I would appreciate any help / advice / explanation you can give. I appreciate your interest in helping.

    Thanks,
    Chris Johnson

  8. #8
    Actuary.com - Level VI Poster Ken's Avatar
    Join Date
    Mar 2005
    Location
    Thailand
    Posts
    833
    Irish Blues, I also found that the inverse power curve is a poor fit at earlier evaluation dates, but there are areas where it fits very well. If I remember correctly, Sherman's point is that the inverse power curve is very good for estimating tails. Have you tried using two different curves, one for the first so many years and the inverse power curve over the remaining years.

    Chris, I don't think that you can directly apply anything you learn on a preliminary to a real life problem. I see that you're working in WV and I find that very interesting since they'll be opening their state to the public I think. On the other hand, because their state has been closed, I think there could be data quality issues. If you have very little quality data, I could see you trying to fit a curve to the few points you have. If you have enough data, I think you should look into using the chain ladder approach. I'm not sure how similar WV is to other states, but you could consider credibility weighting with other sources, possibly NCCI. I thought that NCCI will soon be involved with all the WV stuff.
    Whether you are the lion or the gazelle, when the sun comes up, you better be running.

  9. #9
    You spam? I ban! Irish Blues's Avatar
    Join Date
    Mar 2005
    Location
    WI
    Posts
    1,303
    Ken: the inverse power curve is pretty solid for tails; for that matter, so is a bondy curve. The trick is getting the ATU factors for in between valuations; if you know say 48-ULT and 60-ULT, how do you get 55-ULT? There, how you get 55-ULT may not be appropriate for how you get 38-ULT or 68-ULT, because of how the development curve is shaped - one curve over that whole range won't cut it.

    I actually worked on this problem late last year and earlier this year, looking at various ranges, various curves, and various limits to see what worked and what didn't. We studied Sherman's methods and tried them out, and decided we did not like it as he presented it. After about 21-24 months, a curve can be found to fit to the data in 12-month increments; prior to that, it's difficult and depending on the data, there can be a lot of volatility - enough that we're not comfortable trying to fit a curve.


    Chris: you'll find as you go through the exams that people will suggest a method and you'll be tested on it, but you'll never see it used in real work or it'll be something that's later been proven to be unreliable - but because it's on a paper that's on the syllabus, you'll have to know it. A great example of this is Feldblum's paper on CAS Exam 5 on asset share pricing; we don't use it for P&C work on new policies, period - but it's on the syllabus, and I'll have to know it on the exam in another month. Just use the traditional chain-ladder approach (selecting link ratios), especially for the more recent years. Realize that you're not going to get "the" correct answer now anyway, because of changes in development, late reporting claims, slower/faster settlement on claims, and other stuff like that.
    "You better get to living, because dying's a pain in the ***." - Frank Sinatra

    [url]http://www.hockeybuzz.com/blogger_archive.php?blogger_id=174[/url] - where I talk about the Blues and the NHL.

  10. #10
    Actuary.com - Level I Poster
    Join Date
    Sep 2006
    Location
    WV
    Posts
    9

    Smile

    Thanks very much to you both. Working in my situation, advice from other actuaries is invaluable. I have a great deal to learn.

    Thanks again,
    Chris

+ Reply to Thread

Thread Information

Users Browsing this Thread

There are currently 1 users browsing this thread. (0 members and 1 guests)

     

Similar Threads

  1. Ex 6.8 in Loss Models
    By douglaslam in forum SOA Exam MLC - Actuarial Models, Life Contingencies - with practice exam problems
    Replies: 4
    Last Post: June 22nd 2006, 03:20 PM
  2. Replies: 0
    Last Post: September 14th 2005, 02:57 PM
  3. Replies: 0
    Last Post: September 14th 2005, 02:55 PM
  4. Replies: 0
    Last Post: September 14th 2005, 02:46 PM
  5. Replies: 0
    Last Post: September 14th 2005, 02:44 PM

Bookmarks - Share

Bookmarks - Share

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts