Allow different policy length in ParallelogramOLF

[kennethshsu]

Description

Discussed in #684

^{Originally posted by andyhufford-HARC March 25, 2026}
Is there documentation for the ParallelogramOLF the takes into account the length of the policy term (i.e. semi-annual vs. annual)?

approximation_grain was not described in the API Reference, but that appears to be monthly or daily grain noted from another forum (#497). Is there another parameter for policy-term length that is an input?

class chainladder.ParallelogramOLF(rate_history=None, change_col='', date_col='', approximation_grain='M', vertical_line=False)

Is your feature request aligned with the scope of the package?

• Yes, absolutely!
• No, but it's still worth discussing.
• N/A (this request is not a codebase enhancement).

Describe the solution you'd like, or your current workaround.

No response

Do you have any additional supporting notes?

No response

[andyhufford-HARC]

Happy to contribute to this effort. ParallelogramOLF

[kennethshsu]

Thanks @andyhufford-HARC.

So I think something that could be helpful is building an Excel version showing the calculation in varying policy term that is not annual. Perhaps a 6 months term or even make that a variable?

I wonder if we should support both months and days to match the approximation_grain. So basically have two new parameters policy_length as 1, 6, 12 , 180, 365, etc, and policy_length_unit as M or D? So you can have any length policy in month and any length policy in days. Not sure if I am overcomplicating this. But by doing this we can match the approximation_grain.

What do you think?

[andyhufford-HARC]

In general yes - making it consistent with the approximation_grain makes sense.
Creating a variable version in Excel seems simple enough. I don't want to over promise on timeline, but I will try to have something by mid-day Wednesday (4/1).
Is best practice to upload the excel file right into a comment?

[kennethshsu]

No rush, I am actually going to be away for the next few weeks. But having the sample calculations in will make implementation much easier, and will actually give me more motivation to get this feature implemented.

You can upload the file directly here.

[kennethshsu]

@andyhufford-HARC would you mind taking a look at the open PR? This only takes care of the docstring documentation, not the implementation of what we discussed here. Can you see if you can approve the PR?

#687

[andyhufford-HARC]

@kennethshsu
I have created an excel file that mimics the on-leveling (parallelogram method) examples from the Werner & Modlin text.
In this example we have annual (12 month) policies and semi-annual (6-month) policies and produces the on-level factors by CY. (I used Claude to create the file, but I have tech reviewed the formulas).

Let me know what the next follow up will be.

OnLevel_6Month_Policies_1.xlsx

[kennethshsu]

Thanks, I just finally got a chance to look at this, and it looks like we fully reconcile against the 12 month annual policies.

rate_history = pd.DataFrame(
    {
        "EffDate": ["2010-07-01", "2011-01-01", "2012-04-01"],
        "RateChange": [0.05, 0.1, -0.01],
    }
)
data = pd.DataFrame(
    {"Year": [2010, 2011, 2012, 2013, 2014], "EarnedPremium": [10_000] * 5}
)
prem_tri = cl.Triangle(data, origin="Year", columns="EarnedPremium")
prem_tri = cl.ParallelogramOLF(
    rate_history, change_col="RateChange", date_col="EffDate"
).fit_transform(prem_tri)
prem_tri.olf_

Let me see how to best implement the 6 month policy scenario.