Incorrect Cape Cod reserve when incurred loss = 0

[cdietrich215]

I noticed a minor issue when reserving using the Cape Cod. When incurred or paid loss is $0 for an accident period, the Cape Cod method will predict $0 of Ultimate Loss, when theoretically there should still be a reserve > $0. This may matter for some smaller insurers.

The BF and Benktander methods in this package both correctly set ultimate losses to be (a priori) * (sample weight) * (1 - 1/CDF) in these situations. I believe that the Cape Cod method should do the same. This method currently sets IBNR to 0 (or null).

[jbogaardt]

Thanks. I didn't know about this error.

Reproducible example:

import chainladder as cl

triangle = cl.load_sample('raa')
premium = triangle.latest_diagonal * 0 + 50000
triangle.loc[:,:,'1987',48] = 0
cl.CapeCod().fit(triangle, sample_weight=premium).ultimate_

[kennethshsu]

Can close the ticket once #368 is merged.

[jbogaardt]

Reopening until release of 0.8.14