This text is from the answer 3.6i BF can used to stripping individual large claims out of the triangles that can't develop further Could you please explain this in details for me?
A basic chain ladder will project whatever claims are included in the triangle. Let’s assume that attritional claims are still developing on the early years, but that large losses have already reached ultimate. In this scenario, we’ll be projecting further development of all losses, including the large losses … hence we’ll be over-reserving. Hence, the solution is suggesting that the BF method is preferable since we know the basic chain ladder is not accurate. (Mind you, the BF method isn’t ideal either, I’d much rather strip out the distorting effect of the large losses. But at least the BF method allows us to place less reliance on the chain ladder ultimate.)
I have a few questions: 1. You say BF is more preferable than BCL. But this case we are comparing with a case by case reserving looking at each one, not BCL? 2. I have question about case by case reserving looking at each one, does that mean looking at all in force policies and set the reserve for each one? Or looking at each reported claims and set the reserve for each claims (case estimate- made by claim adjudicator)? If the former, who is the one is doing that? Underwriter? 2. Let's go back to what you say, BF is more preferable than BCL in term of attritional claims are mixed with large losses. But both BF and CL use the development factors (2/1, 3/2...) and all affected (Only that BF used a credibility weighted Ultimate losses so that the affect is smaller)? Is this what you mean?
The Course Notes say it can be used 'as an alternative to stripping large claims out of the triangles...'. From this context we can infer that we are comparing the BF method and the BCL method. If you did decide to look at claims on a case by case basis (although this is not the intention here), then you would use your second suggestion, set the reserve for each reported claim individually. Each insurer would have its own method, but you'd probably only do this for large losses, it'll be too onerous to do it for all attritional claims. It'll probably be the responsibility of the claims team, and they'll ask for input from various experts including their legal team and the underwriters. Quite right. I did say it's not ideal. As I said, I'd much rather split out the large losses.
So case by case reserving here means case estimate for reported claims. Case by case estimate can't estimate Incurred But Not Reported, therefore isn't it not very equivalent to compare BF (used to estimate IBNR) and case estimates (used to estimate reported claims). The fact that case estimate is included in the Reported Run off triangle to calculate development factor used in BF and CL. I feel that the question asks us to compare the in-comparable . Let me know your thought!
That's right. Case-by-case estimate always refers to outstanding reported claims. In my experience it never refers to IBNR claims. I can only repeat: we are not comparing BF with case-by-case reserving. We are comparing the BF method and the basic chain ladder method and as I said, it's not ideal, but we have to answer the question we have been asked. By all means, you can discuss the limitations of the method and hope to pick up marks, ... but this question asks specifically for a comparison of BF against BCL so make sure you focus on that.
In the question 3.6ii, the question asks candidate to compare between BF and case-by-case reserving, that's why I asked before (sorry I did not mention question 3.6ii and that's why you're confusing)
Ah I see, yes now I understand your question! You make an excellent point. A BF calculation (based on an accident year triangle) will indeed include IBNR claims, whereas a case-by-case exercise would not. Well spotted! I think you would be well justified in hoping for a mark for that, I'll add it to our marking scheme for next year. Of course, you could carry out a BF calculation using a reporting year triangle instead. That would exclude IBNR claims and give you a like-for-like comparison. Mind you it would be a pretty unusual methodology.
