General Insurance Data
Hi All, Thank you all for your responses. What I have in mind is a package to help with the initial stage of getting claims data into R and ready to use (formatting etc) before any modeling is done. Effectively to provide a workflow for working with actuarial claims data. My past experiences, using GI reserving projection programs, have usually involved a stage to summarise the data in some format then later feed this data into a program, be it a proprietary program or one of the R packages. The key stage I wish to contribute to here is where the data has to be fed in to R. MRMR and ChainLadder do allow for reshaping of the data (and loading data) but I would like to build on the functionality in these packages while not reinventing the wheel. I will take some time to digest the MRMR package as this does seem to have a lot of functionality that I could use for manipulating the data. A package providing a workflow to get data into a basic tabular structure would act as a stepping stone for getting claims data available and cleaned for use in packages such as ChainLadder or MRMR. It would even be easier to perform other more basic analysis. As I see it at the moment when someone gets to the stage that the data is loaded and formatted in R they are basically ready to go and use MRMR or ChainLadder _*but*_ this is a very big hurdle for novice R users who typically have data stored elsewhere (SAS, SQL Excel or other flat files). Even moving from Excel to R, where data can be stored in objects as opposed to tables is a giant leap in its own right and I fear this is the stage where a lot of novice R Actuarial users are lost or lose confidence in R. I am not a professional programmer and I am not very knowledgeable about different types of connections or porting files from one format to another so I was hoping to leave that up to packages such as RODBC or xlsx. My idea is to provide a workflow structure to load claims data into a data.table using a simple GUI interface. Its worth mentioning, I have come across RExcel which allows you to call R from Excel but what I am talking about is somewhat different in that your using R as the backbone, rather than Excel, for storing and manipulating the data. This would at the same time leave the data in familiar tabular format but helper functions could pivot the data and allow novice R users to create triangles intuitively or extract useful representations such as latest diagonals cumulative data, in year data etc. It would still be easy to manipulate the basic underlying data structure if desired as it is just a data.table but helper functions would give it intuitive Actuarial functionality. The goal would be to reduce the time to close to zero for a novice R Actuarial user to go from using existing data (e.g. Excel or SAS) to a data object that could then be passed to ChainLadder, MRMR or ggplot2. The package Rattle,, has a lovely GUI interface for loading data. I have something like this in mind to get people started. A GUI interface could be used to specify key fields and formats such as origin period etc. The data stored in R as a data.table would still be in a very familiar tabular format and look similar to what novice R users are used to when they have worked with data stored elsewhere in SQL, SAS or Excel files. This would provide a bit of reassurance to get working in R. It would also facilitate novice R users to plot the data using ggplot or googlevis, reshape it, summarise it in tables for printing and you also get the benefit that the data.table structure is quite efficient for working with big in memory data. My recent background is claims reserving in a GI company and claims data is what I have initially in mind to tackle. Other data such as pricing data, time series etc would also certainly lend themselves to more specialised Actuarial data structures and I think I could work on these down the road. From my description above the package might just sound like its to load and format the data. This is a non trivial exercise for novice R users and such a package would reduce the barrier for people adopting R especially at work. At the optimistic end of the scale I also think it could make development data analysis accessible to non Actuaries and could even be useful in other fields. I would love the help from anyone that is interested in contributing as I am working on this solo at the moment. I wish to work a bit on a basic proof of concept first and when I get it working I will load it to git hub and post some links to this mailing list. I wanted first to check that such a package does not exist but it will take a bit of time to get something useful and working. I also appreciate any comments, contributions or feedback. Thanks again for the help Regards Edward Roche
On 29/09/14 02:03, Brian Fannin wrote:
Edward, It may not be exactly what you're looking for, but MRMR could have some things you might like. My primary aim was to develop a package which made data manipulation for loss reserving a bit more efficient. However, it has application in a more general context as well. It begins with a clear, robust yet flexible notion of an origin period (accident/underwriting/report year). For each origin year, one may store measured observations, which have an arbitrarily complex set of dimensions. For example, written premium is allocated to territory, line of business, currency, etc. For loss reserving, we go a step further and allow those measures to change at regular evaluation points. That's that for the data structure. I've tried to build in reasonable support for simple multidimensional/multivariate visualizations. I've got a lot more work to do here, but it's a good start. There's also a new generic method "write.excel" which stores information to a spreadsheet in a reasonable way. This is a pragmatic feature to facilitate data exchange with folk who aren't yet using R. The loss reserving methods presume a linear model framework with chain ladder as a special case. We use mixed effects regression for "loss reserving with credibility". Attached is a very short vignette which may help you to understand how the package is meant to be used. I'm happy to talk more about it, if you're interested. Note that I'm describing features on the development version of MRMR. The CRAN version is more limited. I hope to get the new version submitted and approved before the end of the year. Regards, Brian Fannin -----Original Message----- From: Markus Gesmann [mailto:markus.gesmann at googlemail.com] Sent: Sunday, September 28, 2014 6:36 AM To: Christophe Dutang Cc: Edward Roche; R-Sig-Insurance at R-Project. Org; Brian Fannin Subject: Re: [R-sig-ins] General Insurance Data Hi Edward, Have you looked at Brian Fanin's MRMR package,http://cran.r-project.org/web/packages/MRMR/index.html orhttps://github.com/PirateGrunt/MRMR? The package provides quite a bit on infrastructure for dealing with triangle like data. Regards Markus On 27 Sep 2014, at 14:08, Christophe Dutang<dutangc at gmail.com> wrote:
Dear Edward, To my knowledge, there is no single package doing data manipulation in the insurance context. There are package for manipulating from database manager and other proprietary softwares : see http://cran.r-project.org/doc/manuals/r-release/R-data.html Maintaining the CASdatasets package, I face to some of your problem to read SAS or excel files. Typically the case, where there is a separator for thousand. I did write some R functions : - readfunc(filelist, yearlist, sheetname, pattern, nrow, ncol, colname, col2conv, echo=FALSE, row2rm=NULL) : read a list of excel file to extract the same sheet and concatenate the whole - str2num(x) : convert a character string to a numeric dealing with thousand separator - concatmultcol(df, nbinfo, nbblock, col2foot, col2conv, col2rmpre, cm2rmpost, cname2trunc=NULL, echo=FALSE) : concatenate blockwise data into a rowwise dataframe etc... In such function, I read excel files with gdata package. As you say, for manipulating triangles, there are functions in ChainLadder. By the way, I propose cum2incr and incr2cum to M. Gesmann. Another function shrinkTriangle was not kept in this package: this function computes an annual triangle from a monthly or quarterly basis. For manipulating string, there is a good book by Gaston Sanchez, see http://gastonsanchez.com/work/ At the last R in insurance conf, there was a good presentation of data manipulation (seehttp://www.cass.city.ac.uk/news-and-events/conferences/r-in-insurance-2014/r-in-insurance,-conference-programme and ask for presentation file). Do you have a detailed list of what type of data you want to manipulate? for which purpose? (is it only for reserving purpose?) If you are willing to create a package, I will be happy to contribute. Regards, Christophe - Christophe Dutang LMM, UdM, Le Mans, France web:http://dutangc.free.fr Le 27 sept. 2014 ? 01:16, Edward Roche<ed.roche at yahoo.co.uk> a ?crit :
Hi all, I am interested in a package for working with General Insurance data, i.e. something that deals with the initial stage of getting GI data into R, reshaping it and subsetting it before performing projections using packages such as ChainLadder or claim development visualisations using googlevis. When working with GI data before you can do any meaningful analysis or projections you need to get your data in the right format. Typically you might start with incremental long format data showing claim transactions and this needs to be reshaped and summarised into triangle matrices (e.g. paid and incurreds). You can then perform projections with packages such as ChainLadder or other simple development methods. This initial manipulation stage is not always easy to do especially for novice users of R. There are helper functions in the ChainLadder package such as incr2cum, as.triangle etc that let you perform conversions on the fly but I cant find anything for the wholesale restructuring or manipulation of GI data such as converting it from Annual Quarterly to Annual Annual etc. This type of manipulation is very easy to do in some proprietary software (not R based) that I use on a daily basis. I am considering working on a package that would generally provide helper functions to load untidy GI data into R and let a novice R user perform restructuring and manipulation on the fly. I envisage a GUI such as is available in Rattle to load the data and specify the key variables and formats. Once the data is loaded intuitive helper functions would let you manipulate it on the fly. For example you might wish to pick out paid and incurred triangles, subset the data in some way or convert it to Annual Annual or from cumulative to incremental. My question for this mailing list is are there any such packages out there? or is anyone working on something like this? I would love to get involved if so. Any other thoughts? Thanks Ed
_______________________________________________ R-SIG-insurance mailing list R-SIG-insurance at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-insurance
_______________________________________________ R-SIG-insurance mailing list R-SIG-insurance at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-insurance