Berquist Sherman Adjustment on Paid Loss Triangles

[IGotCavities]

Was playing around to learn ChainLadder but couldn't seem to find the Paid B/S method that adjusts the paid triangle through disposal rate. Does anyone know if that method exists within ChainLadder package? Can you point me there?

I have seen and gone through the (incurred) Berquist Sherman method that adjusts triangles based on a predetermined trend assumption, which is pretty awesome!

Thanks!

[jbogaardt]

Its been a while since I've looked at BerquistSherman, but the docs do describe the disposal_rate_ property of the esimtator very briefly. I think that if you supply a paid, case incurred, reported count, and closed count triangle, the method should automatically adjust the paids for the disposal rate...i think.

[IGotCavities]

Thanks for the help! Got the initial adjusted paid triangles and am definitely interested in exploring more.

[IGotCavities]

Coming back for more Berquist question if I may. Want to specify the disposal rate to use instead of going for the one based on the default ultimate claim count. I found the "reported_count_estimate" parameter but not sure if I understand what it wants to be fed in. I found the explanation below in the source code, but too rookie to understand what "Estimate" or "Pipeline" means? Can you help understand and perhaps provide an example of using 3-yr volume weighted average to calculate disposal rate? Thanks!

Code and the question is with the reported_count_estimator
berq_paid = cl.BerquistSherman(
paid_amount='paidloss', incurred_amount='incdloss',
reported_count='incdcount', closed_count='cwpcount',
reported_count_estimator = (cl.Development(n_periods=3, average = "volume").fit(CW_tri['incdcount']).cdf_ * CW_tri['incdcount'])
)
berq_paid

Source Code explanation:
reported_count_estimator: Estimator
An IBNR estimator for reported_count used to calculate closed_count
disposal rates. Estimator can be a Pipeline. If None selected then
basic Chainladder model is used."

[jbogaardt]

There is probably opportunity to improve the docs here. Estimator generically refers to any sklearn-like modeling feature in the package. However, in this case, it is limited to estimators that spin off an ultimate_ property. By default, the volume-weighted Chainladder is used, but it could be any other IBNR estimator. If you want to tailor the development patterns of the Chainladder you need to use a Pipeline. A working example with a 3-year volume weighted average reported count ultimate would look like this:

import chainladder as cl
triangle = cl.load_sample('berqsherm').loc['MedMal']

reported_count_estimator = cl.Pipeline([
  ('dev', cl.Development(n_periods=3, average="volume")),
  ('model', cl.Chainladder()),]
)

berq = cl.BerquistSherman(
    paid_amount='Paid', 
    incurred_amount='Incurred',
    reported_count='Reported', 
    closed_count='Closed',
    reported_count_estimator=reported_count_estimator,
    trend=0.15).fit(triangle)

berq.adjusted_triangle_

[IGotCavities]

Thanks for the explanation and example! Pipeline comes in very handy, and the paid B/S adjustment method works well for me now.