Implementation of new techniques

[genedan]

I came across this thread in the CAS General Research forum regarding a few new papers. Since one of our goals was to have this package be at the forefront of actuarial open source, I figured we could collect novel (or older, but not-yet-implemented) techniques in a table, which will be updated as they are discussed and prioritized.

Technique: A name or short-description of the technique
Reference: A link to a paper which details the technique
Difficulty: How hard it would be to implement the technique in chainladder-python
Demand: Public demand for including the technique in the package.

Technique	Reference	Difficulty	Demand
Richman-Wüthrich Individual Claims Reserving	https://arxiv.org/pdf/2602.15385, https://arxiv.org/pdf/2603.11660	TBD	TBD
Frequency-Severity Techniques	https://www.casact.org/sites/default/files/database/studynotes_friedland_estimating.pdf	Low	High
Benktander Golden Stairs	https://www.tandfonline.com/doi/full/10.1080/03461238.2025.2536811?af=R#d1e108	TBD	TBD
Handling Sparse Data for Reserving Using Bayesian MCMC	https://www.casact.org/abstract/handling-sparse-data-reserving-using-bayesian-mcmc	TBD	TBD
Claims Development by Layer: The Relationship between Claims Development Patterns, Trend and Claim Size Models	https://www.casact.org/abstract/claims-development-layer-relationship-between-claims-development-patterns-trend-and-claim	TBD	High
Testing the Assumptions of Age-To-Age Factors	https://www.casact.org/abstract/testing-assumptions-age-age-factors	TBD	TBD
The Market Value Margin Within The Distribution-Free Chain Ladder Model - A Way To Account For Calendar Year Effects And Aggregating Lines Of Business	https://www.casact.org/abstract/market-value-margin-within-distribution-free-chain-ladder-model-way-account-calendar-year	TBD	TBD
A Flexible Framework for Stochastic Reserving Models	https://www.casact.org/abstract/flexible-framework-stochastic-reserving-models	TBD	TBD
The Use of GAMLSS in Assessing the Distribution of Unpaid Claims Reserves	https://www.casact.org/abstract/use-gamlss-assessing-distribution-unpaid-claims-reserves	TBD	TBD
Applications of Reserve Ranges and Variability in Practice	https://www.casact.org/abstract/applications-reserve-ranges-and-variability-practice	TBD	TBD
Testing the Assumptions of Assumptions Testing	https://www.casact.org/abstract/testing-assumptions-assumptions-testing	TBD	TBD

[RonRichman]

Thanks for featuring the first one!

[henrydingliu]

we probably need that migration to tabular backend before we can comfortably tackle no. 1

no. 2 is already possible today. let me know if you want some code snippets.

no. 3 looks doable. always a good sign to have data right in the paper

[genedan]

@RonRichman - Glad to see you here! We also accept pull requests from anyone - no pressure!
@henrydingliu - I definitely think Freq/Sev would be highest priority for the next method rollout, would appreciate a fork. I have also been marking as high priority any papers that are in the exam syllabus.

I've been adding more papers (including @rajesh06's) as I move along, using the super-unsophisticated method of simply searching for "reserving" in the CAS database and reading the abstracts. These examples are only from the first page of results, so as far as I can tell, the sky is the limit when it comes to finding things can be turned into/integrated into a package. So much so, that I think there will always be a niche for Chainladder to fill in the world of open-source.

@kennethshsu - I recall there were some discussions where we couldn't figure out what to do with certain research monies. One idea would be putting a bounty/hackathon on some of these. I think it could attract more contributors and promote the package to the public. We could grade the submissions on criteria including but not limited to:

• Richness of features
• Quality of documentation
• Ease of hand-off to the maintainers (i.e., distilling of difficult mathematics to something maintainable)

[kennethshsu]

Yap, we could. That’s one way. I’ve also seen in some other projects that there are sponsored donations, where features or bugs get voted up through $$.

By the way, I’m prioritizing Friedland pretty heavily, and I think replicating the examples will help us identify gaps in the package. Off the top of my head, I know we are missing one frequency/severity method and have at least one outstanding method. We’ll know more as we work through and reconcile additional chapters.

[rajesh06]

Wow - this is a blast from the past for me. I am actually semi-retired, and one of my plans in semi-retirement was to rewrite that paper.

I wrote it as "research." but I think it could benefit from a rewrite with more of an educational focus. Others disagree on this point, preferring that we teach our students to read/learn from research papers. Anyway, if I rewrite it, I could work on the function at the same time. (I have very good R knowledge, but learning Python is also one of my semi-retirement goals.)

[rajesh06]

One of my suggestions would be to add a "factor picker diagnostic EDA"- basically, a faceted plot with charts for age-to-age factors, with the accident year on the x-axis and the development factor on the y-axis, with the size of the points proportional to the denominator of the development factor. One of the arguments would be the time-weighted and straight averages to include as horizontal lines, and of course the selected development factor as a horizontal line. The first (top left) facet would be a box plot of development factors ~ age interval.