Commentary

Accuracy-of-Computation Tests

Topic One – Commentary

Accuracy-testing must be treated as an essential part of the claim reserve computational protocol. The model work-product, Attachment A, sets forth such accuracy test on Page Three where the following metrics are shown for a particular computed reserve and for a particular trun-out test period: (a) expected paid claims, (b) actual paid claims and (c) ratio of actual/expected. Details in support of Items (a) and (b) are set forth in the Intermediate Values Report attached as later pages of the Work-Product. A/E testing may be done in the early as well as the later months of the run-out period; this is done by dividing (or allocating) the reserve into parts (Part One – paid in first run-out year; Part Two – paid in second run-out year, etc.

The lessons learned from this run-out testing are numerous:

Lesson One – The provided-data (a) is inconsistent and incomplete and (b) the populations, from whence the data was obtained, disparate in practices.

Lesson Two - The computational protocols are, for the most part, new-millennial, and yield results which are reasonably creditable but not up to expected standards.

Lesson Three – The provided-data issue is not correctible which means that we (a) live with the problem or (b) modify our computational protocol. Such modification may be (a) as part of the basic computation (i.e., non-transparent) or (b) separate from the basic computation (i.e., transparent). If, transparent, the work-product would b in two parts: (a) Basic Reserve and (b) Reserve Amendment. The Reserve Amendment would any and all computational adjustments needed to reach the goal of the A/E being 100% both in theory and in practice.



Topic Two – Summary of Reported Accuracy Tests


Number of Results of Accuracy Tests

Study No.

Reserves

Over

Under

Mean

1

7

3

4

1.01

2

9

7

2

1.36

3

120

78

42

1.11

4

71

27

44

.99

5

14

10

4

1.04

6

45

26

19

.89

Composite

266

151

115

1.07



Notes: (1) Each of the Study Groups had its own unique set of claims processing rules and practices which left its “lag print” on the A/E results;

Topic C - Conclusion

If these six tests were an unbiased sample, we could expect for any computed reserve a run-out of 107%. The assertion of this E-Text is that this sample is not unbiased and therefore traditional statistical inferences are not available. However, the results are reasonable and acceptable and as good as we are going to get until some major changes in the claim database are made.