2023 Exam 7 Progress Thread

I guess not, if it works it works :man_shrugging:

Its good to know LINEST in addition to memorizing the formulas. They can, and have (S18Q4), constructed problems that force you to use the formulas instead of built in Excel or Multiview functions.

If anyone has any excel tips to make our lives easier on the exam please share

I would also add that using excel’s =CORREL() is much quicker and easier than memorizing and carrying out the formulas for correlation. This is relevant for Venter’s correlation test.

2 Likes

Does anyone know if CF/RF have been updated for the new Friedland paper yet?

Not yet, they’re saying it should be updated by the end of January though.

In the Clark paper, when should I be using truncated growth? Is it only when directed to by the question?
(and how would that apply in practice?)

Yes, I would say only when directed to do so by the question, otherwise you are adding unnecessary complexity, which will result in wasted time and a possible error.

In practice I think it comes down to judgement but would be quite apparent that truncation is needed. From the Clark paper, “From this data, the Loglogistic curve estimates that only 77.24% of ultimate loss has emerged as of ten years.” Depending on the line of business, 77.24% reported (ATU 1.295) at 10 years may be obviously unreasonable, so you would need to address this somehow. Truncation or an alternative curve (Weibull) are both options to make the extrapolated tail smaller.

Edit: Maybe they could construct a question where the point is to give you a tail that is obviously too large, and then you’d either have to describe or carry out an alternative method. But if they don’t prescribe a truncation point, then there will be many different correct answers.

1 Like

I would also say only to do it if the question asks for it. If they don’t specifically say that there is truncated growth, I would state my assumption that there is none so that the graders are clear on what you’re doing.

Friedland is now updated on CF/RF-with a day to spare!

1 Like

Adding to this:
Just using =LINEST(Y values, X values) returns the slope. The other arguments both default to TRUE.
You can enter it as an array formula with an adjacent cell to the right to return both slope and intercept.
Alternatively, use =INDEX(LINEST(Y values, X values), 2) to pull the intercept by itself as the 2nd returned statistic.

1 Like

Is anyone using TIA for this? I’ve got some older TIA questions I’m practicing with (from the 2020 TIA manual) and one of the Clark questions doesn’t seem to match what Rising Fellow does. Wondering if someone can help clarify. It’s regarding calculating G(x) for development periods less than a year.

Hi skysn93,

I think we may have already answered you via email so I apologize for the redundancy, but I wanted to reply here for the sake of others.

We double-checked with the author, Dave Clark, on this recently to make sure we’re showing it right and he agreed with our interpretation.

Here’s an example to show what’s going on in Clark Appendix B:

If we are not annualizing, then I don’t think you’d make any adjustment at all. This is basically what Clark says in the first couple sentences of Appendix B.

If we are annualizing, then we would do the adjustment by multiplying G(x) by the %Exposure.

Let’s assume:

  • Losses of $100 at 9 months development for the latest AY
  • G(4.5) from a loglogistic curve is 0.4
  • The %Earned for the total year is 0.75
  • LDF method (to keep it simple)

If we are not annualizing and we just want to calculate the ultimate loss for the 9mo of exposure, it would just be:

Ult = 100 / .4 = 250

If we are annualizing and doing the adjustment in Appendix b, we’d have:

G(t) = .75 * .4 = .3

Ult(annualized) = 100 / .3 = 333.33

This corresponds with the answer above. If we annualized the ultimate after the fact, it’s 250 / .75 = 333.33

Hopefully that makes more sense.

Steve

1 Like

Has anyone happened to have converted old exams (2011-2015) into excel format?

I’m still a bit confused on the issue here. I think it mainly has to do with what value does the question provide when it gives us the G(x) formula. Since in Clark, he gives the formula as G_AY(x) = %emerged times G*(x), how do we know if the G(x) formula given for the question represents G_AY(x), or G*(x)? Would it be acceptable to state my assumption either way? So essentially, if I assume the question gives us G_AY(x), then if we are estimating the annualized ultimate, there would be no adjustment needed for any age, even if it is less than 12 months. Or if I assume the question gives G*(x), then I would make the adjustment to make the estimate annualized. Is that correct?

I suppose it depends on how the problem is given. In the example I gave, if the the loglogistic parameters for t = 4.5 months resulted in a G(4.5) = 0.4, this would be G*(4.5) in Clark’s appendix B notation. The LDF of 1 / 0.4 would develop the $100 of losses at 9 months to ultimate, but only for the 9 months of exposure.

If we wanted to annualize it and estimate the ultimate loss for 12 months of exposure, then we would do what I showed in the example. G(t) = 0.75 * G*(4.5) = 0.75 * 0.4 = 0.3.

The notation is a little confusing in Appendix B, so I would try to think of it more intuitively. On a problem if it’s unclear what the problem is giving, just make sure to state your assumption of how you’re using it.

Sooo what are we thinking might get asked on the Friedland paper? It seems like the paper is mostly review, with the only specifically reserving-related content overlapping pretty squarely with Patrik (data limitations, difficulty of reinsurance reserving) or Exam 5. There’s no more SB or credibility method; just a discussion of the “big three” methods and the cautions to take in a reinsurance context. Does anyone see any potential for questions unlike anything seen before?

The only thing I can think of is written response questions asking for lists of types of reinsurance contracts and their respective primary functions.

This was my expectation as well. Either full essay question or random part e for .25 looking for a definition.

Does anyone know if there is a way to sort values in a list in Pearson Vue? For p-p plots and the K-S test the values need to be in order and I did not see the sort function in the function list here:
Pearson Vue Functions list download

There is a sort function in Pearson Vue, I just tried their sample environment.

2 Likes

Could someone please ELI5 what exactly the decay factors in Sahasrabuddhe are what is is decaying?

The LEV relativity, so connecting it to another paper, remember in Siewart how the severity relativity decreases with age. So in Saharasbuddhe, in the case when we only have the LEVs modeled at ultimate, we’re using the decay factor to more or less estimate the relativity at an earlier maturity.

That’s at least how I’m understanding it, someone else can let me know if I’m off the mark

1 Like