Development Correlation test

[chang48]

Greetings! First of all, thank you for this wonderful package. It provides pretty much everything for reserve estimation. I'm still learning about it and have a question about the development_correlation test. The Chain Ladder method operates under the assumption that (subsequent) development factors are independent. The method development_correlation is designed for checking the assumption.

When I applied this test on the sample data set in Mack's 1994 paper (the data is available in this repo), the test result indicates that the development factors are correlated, as shown in the screenshot:

In Appendix G of the paper, Mack reached the following conclusion on the same data set

the hypothesis of having uncorrelated development factors is not rejected.

I'm wondering what causes the difference. Is it because I interpreted the test results incorrectly or I did not set the p_critical properly? Thank you!

[henrydingliu]

Mack's original data is already cumulative. The correct triangle declaration will produce the desired result.

[chang48]

Okay, that makes total sense. I went back to the paper and found that Mack mentioned clearly in Section 6 Numerical Example that the numbers "are cumulative incurred case losses." This is my oversight and thank you for catching it. Very much appreciated.