gpitt71 / clmplus

Age-period-cohort models for the claim development.

clmplus

Repository GitHub that contains the code for the package clmplus.

About clmplus:

• It adds flexibilty to the chain-ladder model: we provide a new and more versatile framework for claims reserving given the cumulative payments data.

• It achieves the ambitious objective of showing the contact point between models often used in life insurance and non-life insurance models for claims reserving.

• clmplus relies on the powerful StMoMo package. Practitioners can either use the default models we programmed for them or set their own hazard model. Examples of model configurations we support:

Model	Lexis dimension	Claims reserving
a	age	development (chain-ladder model)
ac	age-cohort	development-accident
ap	age-period	development-calendar
apc	age-period-cohort	development-calendar-accident

Installation

The developer version of clmplus can be installed from GitHub.

library(devtools)
devtools::install_github("gpitt71/clmplus")

The current version of clmplus can be installed from CRAN.

install.packages('clmplus')

Get Started

In this brief example, we work with the sifa.mtpl data from the clmplus package. Further examples can be found in the package vignettes. The data set of cumulative claim payments is transformed into an AggregateDataPP object that pre-processes the data for claim development modelling.

library(clmplus)

data ("sifa.mtpl")
dataset = sifa.mtpl
datapp = AggregateDataPP(cumulative.payments.triangle = dataset, eta= 1/2)

Our models can be fit with the clmplus function.

a.model.fit=clmplus(datapp,
                 hazard.model = "a") # age-model replicates the chain ladder
                 
ac.model.fit=clmplus(datapp,
                 hazard.model = "ac")

ap.model.fit=clmplus(datapp,
                 hazard.model = "ap")

apc.model.fit=clmplus(datapp,
                  hazard.model = "apc")

The plot function can be be used to explore the scaled deviance residuals of fitted models. Below, an example for the age-period-cohort (apc) model for the claim development.

plot(apc.model.fit)

Predictions are performed with the predict function.

a.model=predict(a.model.fit)
                 
# clmplus reserve (age model)
sum(a.model$reserve)
#226875.5


ac.model=predict(ac.model.fit,
                 gk.fc.model = 'a',
                 gk.order = c(1,1,0))
                 
# clmplus reserve (age-cohort model)
sum(ac.model$reserve)
#205305.7

ap.model= predict(ap.model.fit,
                 ckj.fc.model = 'a',
                 ckj.order = c(0,1,0))

# clmplus reserve (age-period model)
sum(ap.model$reserve)
#215602.8
          
                 
apc.model= predict(apc.model.fit,
                  gk.fc.model = 'a',
                  ckj.fc.model = 'a',
                  gk.order = c(1,1,0),
                  ckj.order = c(0,1,0))
# clmplus reserve (age-period-cohort model)
sum(apc.model$reserve)
#213821.6

The fitted effect (and extrapolated) effects can be inspected with the plot function. We continue below the example with the apc model.

plot(apc.model)