Repository navigation
[BUG] Incorrect scale parameter computed with ClarkLDF Cape Cod method #1501
Description
Are you on the latest chainladder version?
- Yes, this bug occurs on the latest version.
Describe the bug in words
ClarkLDF using the Cape Cod method on genins results in a scale parameter of 53,822 instead of 61,577 in the paper (see bottom of page 68)
How can the bug be reproduced?
There is a typo in Clark, so editing one cell to match:
import numpy as np
import pandas as pd
import chainladder as cl
tri = cl.load_sample("genins").copy()
tri.values[0, 0, 0, 3] = 2_182_708
tri["premium"] = tri["values"] * 0 + np.array([10_000_000 + 400_000 * i for i in range(10)])[:, None]
premium = tri["premium"].latest_diagonal
cape_cod = cl.ClarkLDF().fit(tri["values"], sample_weight=premium)
print(cape_cod.scale_)
53821.777605What is the expected behavior?
assert cape_cod.scale_ == pytest.approx(61,577, abs=atol)Would you be willing to contribute this ticket?
- Yes, absolutely!
- Yes, but I would like some help.
- No.
Metadata
Metadata
Assignees
Labels
No labels
Type
Fields
Priority
High
Effort
Low
Scope
Codebase
Activity
scks1 commentedon Nov 1, 2026
Opened #1520 for this. The Cape Cod fitted incrementals were built from the LDF-implied ultimate instead of premium × ELR, which threw off
scale_andnorm_resid_. With the fix, the reproduction above gives a scale of 61,577, matching the paper. The regression tests apply the one-cell typo correction from Clark's table, andgeninsitself is unchanged.Disclosure: the fix and this comment were written with Claude Opus 5.5 (Claude Code), working under my direction.