---
title: "Introduction to actuaRE"
author: "Bavo D.C. Campo"
date: "`r Sys.Date()`"
output:
bookdown::html_document2:
toc: true
fig_caption: yes
bibliography: references.bib
latex_engine: xelatex
biblio-style: "apalike"
vignette: >
%\VignetteIndexEntry{actuaRE}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
pkgdown:
as_is: true
---
```{r setup, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
options(rmarkdown.html_vignette.check_title = FALSE)
```
```{r logo, echo=FALSE, out.width="25%", fig.alt="actuaRE package logo"}
knitr::include_graphics("./actuaRE.png")
```
In this document, we give you a brief overview of the basic functionality of the `actuaRE` package. For a more detailed overview of the functions, you can consult the help-pages. Please feel free to send any suggestions and bug reports to the package author.
# Summary of available functions
The `actuaRE` package provides a comprehensive toolkit for handling both single-level and hierarchical multi-level factors:
| Model Type | Function | Description |
|------------|----------|-------------|
| Single-level credibility | `buhlmannStraub` | Buhlmann-Straub credibility model |
| Single-level GLM + credibility | `buhlmannStraubGLM` | Combines Buhlmann-Straub with GLM (fixed $p$) |
| Single-level GLM + credibility | `buhlmannStraubTweedie` | Combines Buhlmann-Straub with GLM (estimates $p$) |
| Hierarchical credibility | `hierCredibility` | Jewell's hierarchical credibility model |
| Hierarchical GLM + credibility | `hierCredGLM` | Combines hierarchical credibility with GLM (fixed $p$) |
| Hierarchical GLM + credibility | `hierCredTweedie` | Combines hierarchical credibility with GLM (estimates $p$) |
| Generalized Linear Mixed Model | `tweedieGLMM` | Tweedie GLMM with credibility-based initial estimates |
All models support both additive and multiplicative formulations and share common S3 methods: `summary`, `fitted`, `predict`, `fixef`, `ranef`, `weights`, and `plotRE`.
# Handling multi-level factors using random effects models
Multi-level factors (MLFs) are nominal variables with too many levels for ordinary generalized linear model (GLM) estimation [@Ohlsson]. Within the machine learning literature, these type of risk factors are better known as high-cardinality attributes [@Micci2001]. This package allows to incorporate both a single MLF and MLFs that have a hierarchical structure. A typical example of the latter, within workers' compensation insurance, is the NACE code with a hierarchical structure of industry and branch. Further, the package allows to combine the MLF with contract-specific covariates.
## Random effects model structures
The `actuaRE` package supports two types of random effects structures: single-level random effects and hierarchical or nested random effects.
### Single-level random effects
For a single MLF (e.g., clusters, states, or lines of business), we can fit models of the form:
\begin{align*}
g(E[Y_{ijt} | U_j]) &= \mu + \boldsymbol{x}_{ijt}^\top \boldsymbol{\beta} + U_j \\&= \zeta_{ijt}.\\
\end{align*}
Here, $Y_{ijt}$ denotes the loss cost of risk profile $i$ (based on the contract-specific risk factors) operating in cluster $j$ at time $t$. We calculate the loss cost as
\begin{align*}
Y_{ijt} = \frac{Z_{ijt}}{w_{ijt}}
\end{align*}
where $Z_{ijt}$ denotes the total claim cost and $w_{ijt}$ is an appropriate volume measure (i.e., the exposure). $g(\cdot)$ denotes the link function (for example the identity or log link), $\mu$ the intercept, $\boldsymbol{x}_{ijt}$ the contract-specific covariate vector and $\boldsymbol{\beta}$ the corresponding parameter vector. The random effect $U_j$ captures the unobservable effect of cluster $j$. We assume that the random cluster effects $U_j$ are independent and identically distributed (i.i.d.) with $E[U_j] = 0$ and $\mathrm{Var}(U_j) = \tau^2$.
### Hierarchical (two-level) random effects
For hierarchically structured MLFs, we fit models with two nested levels. In our illustration, we work with a hierarchical MLF that has two hierarchical levels: industry and branch. Figure 1 visualizes this hierarchical structure with a hypothetical example.
```{r hMLF, fig.align = 'center', fig.cap = "Figure 1: Hierarchical structure of a hypothetical example", fig.topcaption = TRUE, echo = FALSE, out.width="100%"}
knitr::include_graphics("./HierarchicalStructureAdj.png")
```
The hierarchical model has the functional form:
\begin{align*}
g(E[Y_{ijkt} | U_j, U_{jk}]) &= \mu + \boldsymbol{x}_{ijkt}^\top \boldsymbol{\beta} + U_j + U_{jk} \\&= \zeta_{ijkt}.\\
\end{align*}
Here, $Y_{ijkt}$ denotes the loss cost of risk profile $i$ operating in branch $k$ within industry $j$ at time $t$. $U_j$ denotes the industry-specific deviation from $\mu + \boldsymbol{x}_{ijkt}^\top \boldsymbol{\beta}$ and $U_{jk}$ denotes the branch-specific deviation from $\mu + \boldsymbol{x}_{ijkt}^\top \boldsymbol{\beta} + U_{j}$. We assume that the random industry effects $U_j$ are i.i.d. with $E[U_j] = 0$ and $\mathrm{Var}(U_j) = \tau^2$. Similarly, the random branch effects $U_{jk}$ are assumed to be i.i.d. with $E[U_{jk}] = 0$ and $\mathrm{Var}(U_{jk}) = \nu^2$.
This package offers three different estimation methods to estimate the model parameters:
- Credibility models: Buhlmann-Straub [@Buhlmann2005] for single-level and hierarchical credibility [@JewellModel] for two-level structures
- Combining credibility models with a GLM [@Ohlsson2008]
- Mixed models [@Molenberghs2005]
## Just the code please
### Example data sets
To illustrate the functions, we use different data sets. We illustrate the credibility models using the Hachemeister [@Hachemeister] data set. The GLM-based methods and GLMMs use the `tweedietraindata` and `tweedietesdata` data sets.
### Buhlmann-Straub credibility model (single random effect)
To estimate the parameters for a single-level random effects structure, we use the Buhlmann-Straub credibility model using the function `buhlmannStraub`. By default, the additive credibility model is fit:
\begin{align*}
E[Y_{ijt} | U_j] &= \mu + U_j.
\end{align*}
```{r, message = FALSE}
capture.output(library(actuaRE), file = tempfile()) # suppress startup message
library(actuar)
data("hachemeister")
# Reshape to long format for single state analysis
X = as.data.frame(hachemeister)
Df = reshape(X, idvar = "state",
varying = list(paste0("ratio.", 1:12), paste0("weight.", 1:12)),
direction = "long")
fitBS = buhlmannStraub(ratio.1, weight.1, state, Df)
fitBS
```
To get a summary of the model fit, we use the `summary` function.
```{r}
summary(fitBS)
```
To obtain the fitted values, we use the `fitted` function
```{r}
head(fitted(fitBS))
```
and we use `ranef` to extract the predicted random effects.
```{r}
ranef(fitBS)
```
We can visualize and the predicted random effects using the function `plotRE`.
```{r, fig.show = 'hold', fig.width = 6, fig.height = 4, fig.alt = "description"}
plotRE(fitBS, plot = FALSE)
```
To obtain predictions for a new data frame, we use the `predict` function.
```{r}
newDt = Df[sample(1:nrow(Df), 5, FALSE), ]
predict(fitBS, newDt)
```
With the package, we can also fit the multiplicative Buhlmann-Straub credibility model
\begin{align*}
E[Y_{ijt} | \widetilde{U}_j] &= \tilde{\mu} \ \widetilde{U}_j
\end{align*}
by specifying `type = "multiplicative"`.
```{r multBS}
fitBSMult = buhlmannStraub(ratio.1, weight.1, state, Df, type = "multiplicative")
fitBSMult
```
### Hierarchical credibility model (two-level random effects)
To estimate the parameters for a hierarchical structure using the hierarchical credibility model, we use the function `hierCredibility`. By default, the additive hierarchical credibility model [@Dannenburg] is fit
\begin{align*}
E[Y_{ijkt} | U_j, U_{jk}] &= \mu + U_j + U_{jk}.
\end{align*}
```{r}
data("hachemeisterLong")
fitHC = hierCredibility(ratio, weight, cohort, state, hachemeisterLong)
fitHC
```
To fit the multiplicative hierarchical credibility model [@OhlssonJewell]
\begin{align*}
E[Y_{ijkt} | \widetilde{U}_j, \widetilde{U}_{jk}] &= \tilde{\mu} \ \widetilde{U}_j \ \widetilde{U}_{jk}
\end{align*}
you have to specify `type = "multiplicative"`.
```{r, eval = FALSE}
fitHCMult = hierCredibility(ratio, weight, cohort, state, hachemeisterLong, type = "multiplicative")
fitHCMult
```
Similarly to before, we use the `summary` function to get a summary of the model fit.
```{r}
summary(fitHC)
```
To obtain the fitted values, we use the `fitted` function
```{r}
head(fitted(fitHC))
```
and we use `ranef` to extract the predicted random effects.
```{r}
ranef(fitHC)
```
We can inspect the predicted random effects using the function `plotRE`.
```{r, fig.show = 'hold', fig.alt=c("description 1", "description 2")}
ggPlots = plotRE(fitHC, plot = FALSE)
ggPlots[[1]]
ggPlots[[2]]
```
To obtain predictions for a new data frame, we use the `predict` function.
```{r}
newDt = hachemeisterLong[sample(1:nrow(hachemeisterLong), 5, FALSE), ]
predict(fitHC, newDt)
```
### Combining credibility models with a GLM
To allow for contract-specific risk factors, we extend the credibility models. For single-level structures, we have:
\begin{align*}
E[Y_{ijt} | \widetilde{U}_j] &= \tilde{\mu} \ \gamma_{ijt} \ \widetilde{U}_j = \gamma_{ijt} V_{j}
\end{align*}
For hierarchical structures, we extend the multiplicative hierarchical credibility model to:
\begin{align*}
E[Y_{ijkt} | \widetilde{U}_j, \widetilde{U}_{jk}] &= \tilde{\mu} \ \gamma_{ijkt} \ \widetilde{U}_j \ \widetilde{U}_{jk} = \gamma_{ijkt} V_{jk}
\end{align*}
where $\gamma_{ijkt}$ (or $\gamma_{ijt}$) denotes the effect of the contract-specific covariates.
#### Single random effect with GLM
To estimate the single-level model using Ohlsson's GLMC algorithm [@Ohlsson2008], we use the function `buhlmannStraubGLM` or `buhlmannStraubTweedie`. `buhlmannStraubGLM` allows the user to specify the power parameter $p$, while `buhlmannStraubTweedie` estimates $p$ along with the other parameters using the `cpglm` function from the `cplm` package.
```{r}
# Add a time factor to the reshaped Hachemeister data
Df$time_factor = factor(Df$time)
fitBSGLM = buhlmannStraubGLM(ratio.1 ~ time_factor + (1 | state), Df,
weights = weight.1, p = 1.5)
summary(fitBSGLM)
```
We use the same syntax as used by the package `lme4` to specify the model formula. Here, `(1 | state)` specifies a random effect $U_j$ for `state`. We extract the estimated parameters using `fixef` (contract-specific effects) and `ranef` (random effects).
```{r}
fixef(fitBSGLM)
ranef(fitBSGLM)
```
#### Hierarchical random effects with GLM
For hierarchical structures, we use `hierCredGLM` or `hierCredTweedie`. `hierCredGLM` allows the user to specify the power parameter $p$, while `hierCredTweedie` estimates $p$ along with the other parameters using the `cpglm` function from the `cplm` package.
```{r}
data("tweedietraindata")
fit = hierCredGLM(y ~ x1 + (1 | cluster / subcluster), tweedietraindata, weights = wt)
summary(fit)
```
Here, `(1 | cluster / subcluster)` specifies a random effect $U_j$ for `cluster` and a nested random effect $U_{jk}$ for `subcluster`. We extract the estimated parameters using `fixef` and `ranef`.
```{r}
fixef(fit)
ranef(fit)
```
In addition, the same functions as before can be used.
```{r }
head(fitted(fit))
predict(fit, newdata = tweedietraindata[1:2, ], type = "response")
ggPlots = plotRE(fit, plot = FALSE)
```
### Mixed models
Alternatively, we can rely on the mixed models framework [@Molenberghs2005] to estimate the model parameters. We use the `cpglmm` function to estimate a Tweedie generalized linear mixed model. Fitting the model, however, can take some time if we have a large data set. We can speed up the fitting process by providing initial estimates from the credibility models, and this is exactly what the `tweedieGLMM` function does!
The `tweedieGLMM` function **automatically detects** whether you have a single random effect or nested random effects based on your formula:
#### Single random effect GLMM
```{r, eval = FALSE}
# Single random effect - uses Buhlmann-Straub for initial estimates
fitGLMM_single = tweedieGLMM(y ~ x1 + (1 | cluster),
tweedietraindata, weights = wt, verbose = TRUE)
```
#### Nested random effects GLMM
```{r, eval = FALSE}
# Nested random effects - uses hierarchical credibility for initial estimates
fitGLMM_nested = tweedieGLMM(y ~ x1 + (1 | cluster / subcluster),
tweedietraindata, weights = wt, verbose = TRUE)
```
Note: Even with the initial estimates, the fitting process can take some time with large data sets. Setting `verbose = TRUE` allows you to monitor the progress.
### Balance property
For insurance applications, it is crucial that the models provide us a reasonable premium volume at portfolio level. Hereto, we examine the balance property [@Buhlmann2005][@Wuthrich] on the training set. That is,
\begin{equation}
\begin{aligned}
\sum_{i, j, k, t} w_{ijkt} \ Y_{ijkt} &= \sum_{i, j, k, t} w_{ijkt} \ \widehat{Y}_{ijkt}\\
\end{aligned}
\end{equation}
where $i$ serves as an index for the tariff class. GLMs with a canonical link function fulfill the balance property when we use the canonical link (see [@Wuthrich]). For linear mixed models (LMMs) and hence, the credibility models, this property also holds. Conversely, most GLMMs do not have this property. To regain the balance property, we introduce a quantity $\alpha$
\begin{equation}
\begin{aligned}
\alpha &= \frac{\sum_{i, j, k, t} w_{ijkt} \ Y_{ijkt}}{\sum_{i, j, k, t} w_{ijkt} \ \widehat{Y}_{ijkt}}\\
\end{aligned}
\end{equation}
which quantifies the deviation of the total predicted damage from the total observed damage. In case of the log link, we can then use $\alpha$ to update the intercept to $\hat{\mu} + \log(\alpha)$ to regain the balance property.
By default, the intercept is updated when fitting models using `buhlmannStraubGLM`, `buhlmannStraubTweedie`, `hierCredGLM`, `hierCredTweedie` and `tweedieGLMM`. If you do not wish to update the intercept, you can set the argument `balanceProperty = FALSE`.
```{r}
fitnoBP = hierCredGLM(y ~ x1 + (1 | cluster / subcluster), tweedietraindata, weights = wt, balanceProperty = FALSE)
yHatnoBP = fitted(fitnoBP)
w = weights(fitnoBP, "prior")
y = fitnoBP$y
fitBP = hierCredGLM(y ~ x1 + (1 | cluster / subcluster), tweedietraindata, weights = wt, balanceProperty = TRUE)
yHatBP = fitted(fitBP)
sum(w * y) / sum(w * yHatnoBP)
sum(w * y) / sum(w * yHatBP)
```
Alternatively, you can use the built-in function `BalanceProperty`. You can use this function with any object that has the slots `fitted`, `weights` and `y`.
```{r}
BalanceProperty(fitnoBP)
BalanceProperty(fitBP)
```
# References