Dear Reserving Subgroup,
We need to decide what the output of our subgroup will be. Lee Bowron
suggested the following as a rough outline:
1. Identifying suitable reserving data sets
2. Reviewing reserving methods available in R.
3. Using at least one method to analyze a data set.
For (1) above, here are some basic triangle reserving data sets that
I'm aware of:
- Jed Frees's R data includes several triangle sets
- The Chainladder package has some reserving data
- We could always input annual statement data or type in sample data
from a published paper.
About (2), here are some methods already implemented and freely
available in R:
- Mack, Bootstrap, and Munich Chain Ladder in the Chainladder package
- Glenn Meyers' Collective Risk Model
- Frank Schmid and Chris Laws' lossDev method
About (3), it seems we have to choose which method to employ and how
to package our work. As to the method, I think there is a tradeoff
between being sophisticated and showing off what R can do (e.g. the
lossDev package) and doing something simpler that all actuaries are
familiar with and can understand (e.g. Chain Ladder).
My current inclination is to rework the FAViR basic reserving paper,
and present the Chain Ladder method (either Mack or Bootstrap) and the
Additive Loss Method (i.e. the Cape Cod Method). Actuaries will be
familiar with (at least with the non-stochastic components of) these,
and we can show how R can present the old LDF vs BF battle in a
regression context to analyze fit and provide reserve ranges. It
seems if we use a more sophisticated method, actuaries would want to
see how the results compare to a more traditional chainladder and BF
We can pick a data set to use as an example, but creating an Excel
template (like the Parallelogram example I sent to all of you) would
allow actuaries to plug in their own data and get rolling with these
methods without having to type any source code or install anything.
Hopefully actuaries will see that it works and makes nice graphs, and
will then be motivated to install and start using R.
Any thoughts? Does anyone know of any more data sets or reserving methods available in R? How do you feel about the basic plan above?
Links for data/methods mentioned above:
Jed Frees's data files, including triangle data:
Chainladder package and reserving data:
Glenn Meyers' collective risk model paper with code:
Example FAViR template:
FAViR "Basic Reserving" paper: