Finger does mention that CWP's and nuisance claims tend to reduce the goodness
of fit to a lognormal curve. However, he also states that if the implied CV's
are approximately the same at each interval (attachment point times mean) then
one can infer that the distribution is lognormal. If the CV's come out
similar using 30,000, we can assume that the distribution is lognormal even
with the CWP's in the data.
--------------------------( Forwarded letter 1 follows )---------------------
Date: Tuesday, 29 September 1998 2:30pm HPDesk
To: External.EMAIL
From: EXTERNAL.EMAIL
Subject: Finger 1996 #42
You are a BCC recipient of this message
******************************************************************************
/External
/From Marybeth_Waak@STPAUL.COM
/To studygroup9@lists.casact.org
/End
******************************************************************************
Part 3.
I was having trouble with this problem and wanted a second opinion.
#42. from the 1996 exam:
Interval #claims Losses
0 2648 0
1-50K 1500 25,001 K
50-100K 400 30,000 K
100-200K 274 45,748 K
250-400K 100 30,000 K
400-750K 22 12,474 K
750-2500K 3 2,621 K
2500K + 1 2,586K
Total 4948 148,440K
In calculating the average claim size, should you be using
148440K/4948 = 30000
Or
148440K / (4948-2648) = 64539?
Finger mentions removing closed without pay in order to make the log-normal
distribution fit the data. The answer given by the Study Manual and in the
CAS Sample Answers uses 30000.
Do you think they would have accepted either answer as long as you state
your assumption that the table was calculated after removing closed without
pay? The 30,000 works better for pulling the values from the table to
calculate the requested ILF, so that's the mean they were expecting you to
calculate.
Thanks for your help.