Using R in Insurance

Markus Gesmann, GIRO Brussels, 19 September 2012

Hello - About me

Name: Markus Gesmann
Profession: Mathematician, working as an analyst at Lloyd’s
Maintainer and co-author of two R packages:
- ChainLadder: Statistical methods for the calculation of outstanding claims reserves in general insurance
  - Used by actuaries around the world for reserving and teaching
- googleVis: Interface between R and the Google Visualisation API
  - Used for data visualisation on www.lloyds.com (1), (2)
Founder and co-organise of the Cologne R user group
Blogger: mages’ blog

Agenda

My motivation for using R
Brief history of R and its way into insurance
Three R examples
R for actuaries: Where to start?
How I created this presentation with R
Conclusions and discussion

Why I started using R

I started in insurance in 2003, fresh out of university
I had used a variety of software, tools and languages already
But I had no experience with spreadsheets
I was surprised how people worked in insurance
I was really surprised what colleagues did with spreadsheets
I looked for alternative data analysis tools
All the cool kids were talking about R
I wanted to be cool as well

What is R?

R is a free software environment for statistical computing and graphics
It compiles and runs on a wide variety of UNIX platforms, Windows and OS X
Notes on the history of R by Pat Burns

R made its way from academia to industry

R started at universities in the 1990s
It is still largely maintained by academics
R is widely used for research publications
Over 100 text books with R have been published already
Today many graduates leave universities with some R knowledge

R made its way from academia to industry

Many commercial applications support R:
- Oracle, SAP Hana, SAS, IBM SPSS, Spotfire, MS Office, QlikView, etc.
Commercial support is available from third parties, e.g.
- Mango Solutions, Revolution Analytics, Trinostics LLC, Burns Stats
- Or contact me
R is established in many disciplines outside academia, e.g. pharma and finance
The insurance industry is adopting R as well, e.g. Lloyd’s

Why R in insurance?

Why Excel, SAS, SQL, SPSS, Minitab, …?
John D. Cook: Why and how people use R

John D Cook talk about Why and how people
use R

Because it gets the job done!

But why use a computing language?

Geeks vs Non-Geeks

By Bruno Oliveira

Typical use cases for R in insurance

Data transformation
Data analysis
Statistical modelling
Prototyping / ad-hoc work
End user computing
Background statistical engine for applications, e.g. pricing spreadsheet
Reporting and reproducible analysis, e.g. MI, Solvency II documentation
Learning statistical and actuarial skills

Three R examples

Reserving: Mack and Bootstrap chain-ladder
Automated reporting: Create PowerPoint slide with R output
Extracting data from a web page: Display earth quakes of the last 30 days

Reserving: Mack chain-ladder

library(ChainLadder) ## load ChainLadder functions
data(GenIns) ## Famous Taylor / Ashe triangle
## For the purpose of the presentation we change the data slightly
dimnames(GenIns)$origin=2002:2011 ## original 1:10
GenIns <- GenIns/1000 ## Requires less space on slide
round(GenIns, 0) ## Display rounded figures

##       dev
## origin   1    2    3    4    5    6    7    8    9   10
##   2002 358 1125 1735 2218 2746 3320 3466 3606 3834 3901
##   2003 352 1236 2170 3353 3799 4120 4648 4914 5339   NA
##   2004 291 1292 2219 3235 3986 4133 4629 4909   NA   NA
##   2005 311 1419 2195 3757 4030 4382 4588   NA   NA   NA
##   2006 443 1136 2128 2898 3403 3873   NA   NA   NA   NA
##   2007 396 1333 2181 2986 3692   NA   NA   NA   NA   NA
##   2008 441 1288 2420 3483   NA   NA   NA   NA   NA   NA
##   2009 359 1421 2864   NA   NA   NA   NA   NA   NA   NA
##   2010 377 1363   NA   NA   NA   NA   NA   NA   NA   NA
##   2011 344   NA   NA   NA   NA   NA   NA   NA   NA   NA

Reserving: Mack chain-ladder

Chart created with googleVis

Reserving: Mack chain-ladder

M <- MackChainLadder(GenIns, est.sigma="Mack")
M

## MackChainLadder(Triangle = GenIns, est.sigma = "Mack")
## 
##      Latest Dev.To.Date Ultimate    IBNR Mack.S.E CV(IBNR)
## 2002  3,901      1.0000    3,901     0.0      0.0      NaN
## 2003  5,339      0.9826    5,434    94.6     75.5    0.798
## 2004  4,909      0.9127    5,379   469.5    121.7    0.259
## 2005  4,588      0.8661    5,298   709.6    133.5    0.188
## 2006  3,873      0.7973    4,858   984.9    261.4    0.265
## 2007  3,692      0.7223    5,111 1,419.5    411.0    0.290
## 2008  3,483      0.6153    5,661 2,177.6    558.3    0.256
## 2009  2,864      0.4222    6,785 3,920.3    875.3    0.223
## 2010  1,363      0.2416    5,642 4,279.0    971.3    0.227
## 2011    344      0.0692    4,970 4,625.8  1,363.2    0.295
## 
##              Totals
## Latest:   34,358.09
## Dev:           0.65
## Ultimate: 53,038.95
## IBNR:     18,680.86
## Mack.S.E   2,447.09
## CV(IBNR):      0.13

Reserving: Mack chain-ladder

plot(M) ## Default plot of a Mack chain-ladder output

Check for trends in the residuals plots to verify if the Mack-model is reasonable

Reserving: Mack chain-ladder

plot(M, lattice=TRUE)

Plot of developments, forecasts and standard errors

Reserving: Bootstrap chain-ladder

B <- BootChainLadder(GenIns, R=999) ## 999 bootstrap replicates
B

## BootChainLadder(Triangle = GenIns, R = 999)
## 
##      Latest Mean Ultimate Mean IBNR IBNR.S.E IBNR 75% IBNR 95%
## 2002  3,901         3,901       0.0        0        0        0
## 2003  5,339         5,435      95.9      114      151      315
## 2004  4,909         5,392     482.7      232      615      911
## 2005  4,588         5,287     698.5      263      867    1,163
## 2006  3,873         4,860     987.2      296    1,183    1,531
## 2007  3,692         5,133   1,441.6      384    1,692    2,091
## 2008  3,483         5,641   2,158.1      473    2,461    2,966
## 2009  2,864         6,840   3,975.2      793    4,466    5,367
## 2010  1,363         5,613   4,249.5    1,055    4,905    6,110
## 2011    344         5,031   4,687.2    2,001    5,817    8,047
## 
##                 Totals
## Latest:         34,358
## Mean Ultimate:  53,134
## Mean IBNR:      18,776
## IBNR.S.E         2,861
## Total IBNR 75%: 20,596
## Total IBNR 95%: 23,505

Reserving: Boot chain-ladder

plot(B) ## Default plot of a Bootstrap chain-ladder output

Check if the model is reasonable

Automated reporting: Create PowerPoint slide with R output

myfile=tempfile()
win.metafile(file=myfile)
plot(M, lattice=TRUE)
dev.off()
## Load MS Office interface statconnDCOM
library(rcom)
## Run VBA code from R
ppt<-comCreateObject("Powerpoint.Application")
comSetProperty(ppt,"Visible",TRUE)
myPresColl<-comGetProperty(ppt,"Presentations")
myPres<-comInvoke(myPresColl,"Add")
mySlides<-comGetProperty(myPres,"Slides")
mySlide<-comInvoke(mySlides,"Add",1,12)
myShapes<-comGetProperty(mySlide,"Shapes")
myPicture<-comInvoke(myShapes,"AddPicture",
                     myfile, 0,1,100,10)

Check license requirement for commercial use of statconnDCOM interface.

Extracting data from a web page

library(XML)

## Warning: package 'XML' was built under R version 3.4.1

library(googleVis)
## Source data diretly from the web
url <- "http://ds.iris.edu/sm2/eventlist/"
eq <- readHTMLTable(readLines(url),
                    colClasses=c("factor", rep("numeric", 4), "factor"))$evTable
names(eq) <- c("DATE", "LAT", "LON", "MAG",
               "DEPTH", "LOCATION_NAME", "IRIS_ID")
##Format location data
eq$loc=paste(eq$LAT, eq$LON, sep=":")

Display earth quake information of last 30 days

library(googleVis)
## Create a geo chart with the Google Chart API
G <- gvisGeoChart(eq, "loc", "DEPTH", "MAG",
                   options=list(displayMode="Markers", 
                   colorAxis="{colors:['purple', 'red', 'orange', 'grey']}",
                   backgroundColor="lightblue"), chartid="EQ")

plot(G)

Getting started with R in actuarial work

R for Dummies by Joris Meys and Andrie de Vries
Introduction to R for Actuaries by Nigel de Silva
Modern actuarial risk theory using R by Kaas, Goovaerts, Dhaene and Denuit
An Interactive Introduction To R by Michael Driscoll and Dan Murphy
Formatted Actuarial Vignettes in R by Ben Escoto
An Actuarial Toolkit presented at GIRO convention 2006 in Vienna
Using R at Lloyd’s poster at UseR! conference 2011 in Warwick
R-Bloggers

R packages for actuaries on CRAN

actuar: Loss distributions modelling, risk theory (including ruin theory), simulation of compound hierarchical models and credibility theory
ChainLadder: Reserving methods in R
copula: Multivariate Dependence with Copulas
cplm: Monte Carlo EM algorithms and Bayesian methods for fitting Tweedie compound Poisson linear models
evir: Extreme Values in R
fitdistrplus: Help to fit of a parametric distribution to non-censored or censored data
lifecontingencies: Package to perform actuarial evaluation of life contingencies
lossDev: A Bayesian time series loss development model
mondate: R package to keep track of dates in terms of months

Meet the R experts

London R photo

London R user group meeting

How I created this presentation with RStudio, knitr, pandoc and slidy

knitr is a package by Yihui Xie that brings literate programming to a new level
- It allows to create content really quickly, without worrying to much about layout and R formatting
RStudio integrated knitr into its IDE, which allows to knit Rmd-files by the push of a button into markdown
Markdown output can be converted into several other file formats, such as html, with pandoc
slidy is one of the options to create interactive html-slides with pandoc
For more details see my recent blog post and source code of this talk.


Rscript -e "library(knitr); knit('Using_R_in_Insurance_GIRO_2012.Rmd')"
pandoc -s -S -i -t slidy --mathjax Using_R_in_Insurance_GIRO_2012.md 
  -o Using_R_in_Insurance_GIRO_2012.html

Conclusions

R comes with lots of functions for actuarial work
It provides an ideal framework for end user computing
R can automate the production of reproducible reports and presentations
The momentum behind R has grown significantly over the last 5 years
Today R is often known by graduates - open up to their ideas
Many other software products developed R interfaces
New business models have evolved and will evolve

If you liked this presentation …

… you may also like:

Interactive web graphs with R - Overview and googleVis tutorial, Royal Statistical Society Conference, 2012
Overview of Lloyd’s using R and googleVis, R in Finance, 2012
ChainLadder at the Predictive Modelling Seminar, Institute of Actuaries, 2010
How to integrate R into MS Office, LondonR, 2010
Fitting distribution with R, blog post, December 2011
Say it in R with “by”, “apply” and friends, blog post, January 2012

Questions?

Idea: R in Insurance Workshop - Interest?
Contact: markus dot gesmann at gmail dot com

R version and packages used for this presentation

sessionInfo()

## R version 3.4.0 (2017-04-21)
## Platform: x86_64-apple-darwin15.6.0 (64-bit)
## Running under: macOS Sierra 10.12.5
## 
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRlapack.dylib
## 
## locale:
## [1] en_GB.UTF-8/en_GB.UTF-8/en_GB.UTF-8/C/en_GB.UTF-8/en_GB.UTF-8
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] XML_3.98-1.9       tm_0.7-1           NLP_0.1-10        
## [4] wordcloud_2.5      RColorBrewer_1.1-2 googleVis_0.6.3   
## [7] ChainLadder_0.2.4 
## 
## loaded via a namespace (and not attached):
##  [1] biglm_0.9-1        statmod_1.4.30     zoo_1.8-0         
##  [4] slam_0.1-40        reshape2_1.4.2     splines_3.4.0     
##  [7] lattice_0.20-35    colorspace_1.3-2   htmltools_0.3.6   
## [10] stats4_3.4.0       yaml_2.1.14        mgcv_1.8-17       
## [13] rlang_0.1.1        nloptr_1.0.4       tweedie_2.2.5     
## [16] plyr_1.8.4         stringr_1.2.0      MatrixModels_0.4-1
## [19] munsell_0.4.3      cplm_0.7-5         gtable_0.2.0      
## [22] expint_0.1-4       codetools_0.2-15   coda_0.19-1       
## [25] evaluate_0.10      systemfit_1.1-20   knitr_1.16        
## [28] SparseM_1.77       lmtest_0.9-35      quantreg_5.33     
## [31] pbkrtest_0.4-7     parallel_3.4.0     highr_0.6         
## [34] Rcpp_0.12.11       scales_0.4.1       backports_1.1.0   
## [37] jsonlite_1.5       lme4_1.1-13        ggplot2_2.2.1     
## [40] digest_0.6.12      stringi_1.1.5      grid_3.4.0        
## [43] rprojroot_1.2      tools_3.4.0        sandwich_2.3-4    
## [46] magrittr_1.5       lazyeval_0.2.0     tibble_1.3.3      
## [49] car_2.1-4          MASS_7.3-47        Matrix_1.2-10     
## [52] actuar_2.1-1       minqa_1.2.4        rmarkdown_1.6     
## [55] nnet_7.3-12        nlme_3.1-131       compiler_3.4.0