#537

[cdietrich215]

This PR accomplishes two things:

1. Ensures that the paid Berquist-Sherman adjustment continues to work when the lookup (based on the adjusted paid claim count) exceeds the maximum actual paid claim count in the row. It accomplishes this by capping the lookup value.
2. Adds an additional Berquist-Sherman unit tests to ensure adjusted values are as expected on the Berquist Sherman dataset. This test passes both before and after the change.

To replicate the error in question:

import chainladder as cl
triangle = cl.load_sample("berqsherm").loc['MedMal']
xp = triangle.get_array_module()
# Purposefully cause negative closed development towards the end for demonstration
triangle.loc["MedMal", "Closed", "1969" , 84] = triangle.loc["MedMal", "Closed", "1969" , 72] * 0.9
triangle.loc["MedMal", "Closed", "1969" , 96] = triangle.loc["MedMal", "Closed", "1969" , 84] * 0.9
triangle.loc["MedMal", "Closed", "1970" , 84] = triangle.loc["MedMal", "Closed", "1970" , 72] * 0.9

berq = cl.BerquistSherman(
    paid_amount='Paid', incurred_amount='Incurred',
    reported_count='Reported', closed_count='Closed',
    trend=0.15)
# Transform triangle
berq_triangle = berq.fit_transform(triangle)

[jbogaardt]

@cdietrich215, welcome to the list of contributors to the chainladder-python package. Your improvements are very much welcome and appreciated. Thank you for your hard work on this. Impressively thorough for a first PR. Nicely done. We'll get a release going so that your code is in an official release. There are a couple minor issues I will need to resolve to get there, but be on the look out for a new version over the coming week.

[cdietrich215]

Thanks @jbogaardt . Of course, happy to help on the package!