Hi, The examiners answer to this question was actually not very satisfactory to me and I wondered if the tutors and students on this forum can help me and verify my approach to the problem. The question gives a cumulative paid and notified triangle with 5 development years for a household insurance LoB of a mid sized insurer. Part 1 asks about the reasonableness of the development patterns and is a fairly straightforward question. part 2 of the question states: "Discuss how the ultimate claim cost estimated by the Basic Chain Ladder method would be affected if during 2000 there had been a major hurricane that had caused the paid losses in underwriting year 2000 to be 50 instead of 11 and the notified losses to be 100 instead of 46." However, the examiner's answer only briefly touches upon the actual impact of the hurricane on the BCL method and goes on to discuss topics that are (in my humble opinion) not relevant to the actual answer. The way I see it, it comes down to the following points: 1.The hurricane happened in 2000. This means that UW year 1999 and UW year 2000 policies would be affected, assuming we are dealing with 1year policies 2. The fact that UW 1999 shows only a small spike in notified claims and paid claims in development period 1 suggests that the hurricane took place towards the end of 2000 (if it had taken place at the beginning, there would be more 1999 policies in force, which would mean a greater spike in dev period 1 for UW 1999 cohort) 3. Thus, we can say that the impact of the 2000 hurricane will only majorly affect the UW2000 cohort 4. Since this is the latest development period for the triangle, this would not be captured in the ultimate calculations while using the BCL method (since the BCL method ignores the latest cohort in its CDF calculations) 5. Thus, the overall calculations would be unchanged if the same BCL was used, without making any allowance for large and catastrophe claims 6. I would make a recommendation to split cat claims separately and create a separate cat reserve to make the reserve calculation accurate Is my answer right here? The answer key speaks about payment patterns of hurricane claims, which I believe is very tenuous and not at all in line with what the question is asking. @Ian Senator your inputs would be greatly appreciated
Hi The keywords are discuss: present different angles to the issue. ultimate cost: the value of claims when fully runoff. BCL: stable claims development pattern, homogeneous data, and claims to date can be used to predict future claims. Hurricane data: different type of claim from the ones normally experienced, is data homogeneous ie does it have the same development pattern as non-hurricane data? Numerical data: is there any way I can check if the hurricane data has a similar development pattern to the non-hurricane data? I can follow the examiner's thought process. Maybe you can explain your thought process a bit more. I am not sure about 2 given that we have been supplied with new data for UY 2000. We probably need new data for UW1999 too. Regarding 4, the claims to date for UY2000 will have hurricane claims. If we apply the same CDFs it means we expect hurricane claims to have a similar development pattern to the non-hurricane claims. Have you proved that? 5 would follow if you have proven 4. PS: This is a 303 paper which is recommended for SP7 in the SP7 FAQ.
Hi @Busy_Bee4422 , thanks for the response. My thought process was basically: 1. The question has given an updated claim number for only UW 2000 and not 1999 - So I am basically assuming that the numbers for 1999 are correct and include the hurricane figures, which is what led me to conclude the hurricane happened in late 2000 and not early 2000 2. w.r.t same CDFs, my argument is that if we use the BCL on the updated claims triiangle, we will arriive at the same CDFs since the historical data has been unchanged. So our data with hurricane claims and without will result in the same ultimate (since UW2000 is the latest cohort and BCL ignores the data in the latest cohort). So my answer was basically "if we continue to use the same method, the answer will remain unchanged" Is my thinking correct? >PS: This is a 303 paper which is recommended for SP7 in the SP7 FAQ. Sorry about this... I made a mistake when typing out the title. Yes I was attempting this question for practice for SA3 since we can expect any type of question
Hi @padasala 1. ok. 2. Yes the computed cdfs will be the same but they will be applied to 50 not 11 for the paid triangle and 100 in the notified claims triangle. This means we expect the hurricane claims to develop in a manner similar to non-hurricane claims. This needs to be confirmed from the numerical data provided. The examiners used a ratio of paid to notified to see if it is the same between the non-hurricane and hurricane claims. From the ratio it seems hurricane claims are being processed faster than non-hurricane claims. It therefore seems inappropriate to use the same cdfs on hurricane data.
This question is effectively about diagnostics. We cover it in SP7 tutorials. This part of the question is asking whether the BCL will still be appropriate if we get this catastrophe. The answer is probably not, because cats (can) develop differently to attritional claims - they're generally reported more quickly but settled more slowly. So if you're relying on a stable claim development (the main assumption in the BCL) then you're going to get your reserves wrong if you leave the cat data included. Note also that, as Busy-bee points out, looking at the paid/notified figures suggests that the hurricane occurred early in the year compared to non-hurricane claims. Which is at odds with the fact that 1999 figures haven't changed. It's the queries around this that the examiners were after, rather than a definitive conclusion.
