-
Notifications
You must be signed in to change notification settings - Fork 0
/
README.Rmd
111 lines (74 loc) · 4.95 KB
/
README.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
---
output: github_document
bibliography: "vignettes/biblio.bib"
---
<!-- README.md is generated from README.Rmd. Please edit that file -->
# Revisiting Whittaker-Henderson Smoothing
```{r setup, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "",
fig.width = 10,
fig.asp = 0.618,
out.width = "100%",
fig.align = "center",
fig.path = "man/figures/README-"
)
library(WH)
```
<!-- badges: start -->
[](https://github.com/GuillaumeBiessy/WH/actions/workflows/R-CMD-check.yaml)
<!-- badges: end -->
## What is Whittaker-Henderson smoothing ?
Whittaker-Henderson (WH) smoothing is a gradation method aimed at correcting the effect of sampling fluctuations on an observation vector. It is applied to evenly-spaced discrete observations. Initially proposed by @whittaker1922new for constructing mortality tables and further developed by the works of @henderson1924new, it remains one of the most popular methods among actuaries for constructing experience tables in life insurance. Extending to two-dimensional tables, it can be used for studying various risks, including but not limited to: mortality, disability, long-term care, lapse, mortgage default, and unemployment.
## How to use the package?
The `WH` package may easily be installed from CRAN by running the code `install.packages("WH")` in the $\mathsf{R}$ Console.
It features two main functions `WH_1d` and `WH_2d` corresponding to the one-dimensional and two-dimensional cases respectively. Two arguments are mandatory for those functions:
* The vector (or matrix in the two-dimension case) `d` corresponding to the number of observed events of interest by age (or by age and duration in the two-dimension case). `d` should have named elements (or rows and columns) for the model results to be extrapolated.
* The vector (or matrix in the two-dimension case) `ec` corresponding to the portfolio central exposure by age (or by age and duration in the two-dimension case) whose dimensions should match those of `d`. The contribution of each individual to the portfolio central exposure corresponds to the time the individual was actually observed with corresponding age (and duration in the two-dimension case). It always ranges from 0 to 1 and is affected by individuals leaving the portfolio, no matter the cause, as well as censoring and truncating phenomena.
Additional arguments may be supplied, whose description is given in the documentation of the functions.
The package also embed two fictive agregated datasets to illustrate how to use it:
* `portfolio_mortality` contains the agregated number of deaths and associated central exposure by age for an annuity portfolio.
* `portfolio_LTC` contains the agregated number of deaths and associated central exposure by age and duration (in years) since the onset of LTC for the annuitant database of a long-term care portfolio.
```{r fit-1d}
# One-dimensional case
d <- portfolio_mort$d
ec <- portfolio_mort$ec
WH_1d_fit <- WH_1d(d, ec)
```
```{r fit-2d}
# Two-dimensional case
keep_age <- which(rowSums(portfolio_LTC$ec) > 5e2)
keep_duration <- which(colSums(portfolio_LTC$ec) > 1e3)
d <- portfolio_LTC$d[keep_age, keep_duration]
ec <- portfolio_LTC$ec[keep_age, keep_duration]
WH_2d_fit <- WH_2d(d, ec)
```
Functions `WH_1d` and `WH_2d` output objects of class `"WH_1d"` and `"WH_2d"` to which additional functions (including generic S3 methods) may be applied:
* The `print` function provides a glimpse of the fitted results
```{r print}
WH_1d_fit
WH_2d_fit
```
* The `plot` function generates rough plots of the model fit, the associated standard deviation, the model residuals or the associated degrees of freedom. See the `plot.WH_1d` and `plot.WH_2d` functions help for more details.
```{r plot}
plot(WH_1d_fit)
plot(WH_1d_fit, "res")
plot(WH_1d_fit, "edf")
plot(WH_2d_fit)
plot(WH_2d_fit, "std_y_hat")
```
* The `predict` function generates an extrapolation of the model. It requires a `newdata` argument, a named list with one or two elements corresponding to the positions of the new observations. In the two-dimension case constraints are used so that the predicted values matches the fitted values for the initial observations [see @carballo2021prediction to understand why this is required].
```{r predict}
WH_1d_fit |> predict(newdata = 18:99) |> plot()
WH_2d_fit |> predict(newdata = list(age = 50:99,
duration = 0:19)) |> plot()
```
* Finally the `output_to_df` function converts an `"WH_1d"` or `"WH_2d"` object into a `data.frame`. Information about the fit is discarded in the process. This function may be useful to produce better visualizations from the data, for example using the ggplot2 package.
```{r}
WH_1d_df <- WH_1d_fit |> output_to_df()
WH_2d_df <- WH_2d_fit |> output_to_df()
```
## Further WH smoothing theory
See the package vignette or the upcoming paper available [here](https://hal.science/hal-04124043)
## References {-}