Krakowski's paper

[3rdkindactuary]

How do you guys feel about Karkowski's paper? It appears little twisted to me. Only one point has been tested from old exams, I will give up this paper anyway if I can't have enough time left.

I have no way to know the author background, and admire his great work. But
the author is more likely writting a paper in phillosophy but not an actuarial one, some simple concepts have been addressed in a more complicated way then necessary, which leads too much confusion..I got an impression the autor combined actuarial practice and therotical phillosophical approach together to address a practical topic, no way to figure out.

If anyone can give me some suggestions reagrding this paper, I definitely apprecaited. Lange tried to write the increase limit topic in a literature way, which turns out very elegant and informative, but this one is really tough to master...man..or I am too slow?

:wall:

[Vegas or Bust]

No, you're not slow. I was also very frustrated with this paper. I spent about an hour and a half on it, which was about 90 minutes too long. I doubt I'll pick it up again.

[3rdkindactuary]

Please,

also t :rofl: hanks, vegas...

[brente]

Actually, isn't this a newer paper? I think it may have been introduced last year. Last year I believe there was only one question from it, and it was based on the first half of the paper, which generally explains the method and is relatively straightforward. My fear is that this year they will test one of the exhibits, and that is where I have a hard time understanding this paper. They don't seem to be documented well enough to understand what's going on. If anyone has figured this out (or perhaps has a study guide that explains the math on the exhibits), your help would be appreciated! Here's my take on the exhibits:

Exhibit 1: is just taking E(CAT losses)/E(EP) where CAT losses include LAE, and are taken as a percentage of trended exposures, and EP is also trended. My only question: Where is the factor for CAT/AIY coming from? I don't see the "Dev. of Total Catastrophe Provision Exhibit." In any case, if they tested this they would probably just provide the factor.

Exhibit 2: I think I understand what's going on conceptually. Rather than rely on each states' volatile trend indications, we apply a countrywide factor. Using a 30 year period, we fit a line to the indicated CAT ratios for each year, and take the ratio of the fitted ratio at current level to the fitted ratio at the midpoint of the 30 year period. So rather than trend each year individually, we trend the total amount for all 30 years by this one average factor. I tried to fit a line through the data, using x=AIY and y=CAT ratio, and I get reasonably close to what Krakowski shows, but I'm still a bit off. Was anyone able to derive his fitted CAT ratios?

Exhibit 3: This is sort of similar to an off-balance calculation. We have the damage ratios stated for the past 10 accident years, but we want to apply them to just one year, so we adjust the damage ratio for each territory to preserve the overall countrywide damage ratio.

Exhibit 4: I think I understand this one as well. We're combining observations from territories in several different states within a region and ranking them from low to high. We cap these results from above and below. But we have to make up for this difference somehow, so we calculate the excess as basically the difference between the sum of all actual losses minus the sum of all capped losses, where losses = AIY*damage ratio. We then apply this load to each territory. All losses that have been capped are still all equal to each other, but they are all higher now. Again, this is somewhat similar in concept to an off-balance calculation.

Exhibits 5, 6, and 8: And here's where they lose me. There are a few notes in the errata for this text that illuminate a few things, but I'm still confused. For example, the errata explains how the estimated process variance is calculated, and it is the same for all territories. So why does it vary by territory in Exhibit 5? And how is it calculated in Exhibit 6? (The same formula gives the wrong answer.) As to the whole bottom half of the page (everything below the estimated process variance), I'm hopelessly lost.

This was as far as I got before having to crawl up into a little ball on my bed and weep audibly for the better part of an hour. I'll tackle the rest of the exhibits when I have a better feel for the first ones.

[Avi]

Please,

also t :rofl: hanks, vegas... Actually, my part 5 notes are at work, but the truth is, I did not spend much time on this paper. I would suggest knowing the procedures of the current methods and suggested methods, as well as the comments as to which is better (IIRC, none are and one is worse). The actual numbers in the exhibit were near unintelligible when I sat, so I didn't spend any time on that.

Know what single capping is, what dual capping is (and how to do a simple one - NOT the one in the paper :shudder: )

Know the differences between the two credibility-weighting methods and what the reasons for and against them are.

I would not suggest following his exhibits, rather use a manual like All-10 or Feldblum notes. They usually have a simplified version that is easier to follow.

If I can, I'll look over some of my notes and see if anything else stands out.

Good Luck!

[brente]

Thanks, that advice helps. I have the Feldblum and All10 notes, but from the year before Krakowski was on the syllabus. :cry:

[oopsieny]

Thanks guys, I am also fustrated with this paper... you guys just gave me a good reason not to look at it again.