MackChainladder with malformed triangle

[jbogaardt]

Another bug I found has to do with the Mack Standard Error, it works when RAA is used but if you delete the lastest origin period being 1990 and Dev year 1990. Version 0.8.8 creates a much larger standard error compared to 0.8.2. I’m not sure why but I’m assuming it has to do with the triangle not being a perfect nxn, Here’s the code that I used after putting RAA data on excel and deleting Origin 1990:

raa_df = table.range('A1').options(pd.Series,index=False, header=True, expand='table').value
raa = cl.Triangle(raa_df, cumulative = True, origin="origin", development="development", columns="value")
mack = cl.MackChainladder().fit(raa)
mack.summary_
mack.total_mack_std_err_.values[0, 0]
Version 0.8.2 Produces : 10009.022070082632
Version 0.8.8 Produces : 76362.26020670775

Originally posted by @rbreen2 in #219 (comment)

[jbogaardt]

@rbreen2, thanks for finding this. I've put this in a separate issue since it is not related to the BootstrapODPSample estimator. This looks like a bug and I trust release 0.8.2 figures more than release 0.8.8.

[jbogaardt]

Offending code:

import chainladder as cl
a = cl.load_sample('raa')
b = a.iloc[:,:, :-1]
# Should produce zeros for all origins except for the last.
cl.MackChainladder().fit(a).parameter_risk_ - cl.MackChainladder().fit(b).parameter_risk_

[rbreen2]

Thank you, do you believe this will be fixed in the next version?

[jbogaardt]

Yes, bugs are given precedence over enhancements so this should be resolved in 0.8.9