do.BackTesting: Perform Out-of-Sample Testing of Mortality Forecasts Over One...

In mpascariu/MortalityForecast: Standard Tools to Compare and Evaluate Various Mortality Forecasting Methods

Description

Perform Out-of-Sample Testing of Mortality Forecasts Over One Time Period

Usage

do.BackTesting(data, data.B = NULL, x, y.fit, y.for, data.in = c("qx",
  "mx", "dx", "lx"), models, level = 95, jumpchoice = c("actual",
  "fit"), verbose = FALSE, ...)

Arguments

data	A data.frame or a matrix containing mortality data with ages x as row and time y as column. This object should contain the data to be used in fitting (training) and validation process as well.
data.B	A data.frame or a matrix containing mortality data for the benchmark population. This dataset is needed only in the coherent mortality models (e.g. LiLee, OeppenC). Must be the same format as in data;
x	Numerical vector indicating the ages in input data. Optional. Default: NULL.
y.fit	Years to be considered in fitting.
y.for	Years to be forecast.
data.in	Specify the type of input data. Various life table indices are accepted: "qx", "mx", "dx", "lx".
models	One or several mortality models to be estimated. The following options are available: "MRW" – The Multivariate Random-Walk (w/o Drift); "MRWD" – The Multivariate Random-Walk with Drift; "LeeCarter" – The Lee-Carter Mortality Model; "LiLee" – The Li-Lee Mortality Model; "HyndmanUllah" – The Hyndman-Ullah Mortality Model; "RenshawHaberman" – The Renshaw-Haberman Mortality Model; "Oeppen" – The Oeppen Mortality Model; "OeppenC" – The Coherent Oeppen Mortality Model; "MEM2" – The Maximum Entropy Mortality Model of order 2; "MEM3" – The Maximum Entropy Mortality Model of order 3; "MEM4" – The Maximum Entropy Mortality Model of order 4; "MEM5" – The Maximum Entropy Mortality Model of order 5; "MEM6" – The Maximum Entropy Mortality Model of order 6; "MEM7" – The Maximum Entropy Mortality Model of order 7;
level	Significance level of the confidence interval.
jumpchoice	Method used for computation of jumpchoice. Possibilities: "actual" (use actual rates from final year) and "fit" (use fitted rates).
verbose	A logical value. Set verbose = FALSE to silent the process that take place inside the function and avoid progress messages.
...	Arguments to be passed to or from other methods.

See Also

do.BBackTesting evalAccuracy.BackTesting

Examples

library(magrittr)

x  <- 0:100            # Ages
y1 <- 1980:2000        # Training period
y2 <- 2001:2016        # Validation period
y  <- c(y1, y2)        # entire period
K <- "GBRTENW"         # Country to be analysed

# Mortality data for country: K
mx.country <- HMD_male$mx[[K]][paste(x), paste(y)] %>% replace.zeros

# Create a mortality data for a benchmark population
# to be used in "LiLee" and "OeppenC" models
average_mx <- function(w) prod(w, na.rm = TRUE)^(1/(length(w)))

Mx <- HMD_male$mx %>% 
  lapply(as.matrix) %>% 
  lapply(replace.zeros) %>% 
  simplify2array() %>% 
  apply(., 1:2, FUN = average_mx) %>% 
  as.data.frame()

mx.benchmark <- Mx[paste(x), paste(y)]

# Select various mortality models
MM <- c("MRWD", "LeeCarter", "LiLee", "HyndmanUllah", 
        "Oeppen", "OeppenC", "MEM5", "MEM6")
# Fit & Forecast the models 
B <- do.BackTesting(data = mx.country,
                    data.B = mx.benchmark,
                    x = x,
                    y.fit = y1, 
                    y.for = y2,
                    data.in = "mx",
                    models = MM)

# Compute accuracy measures for resulted life exectancies
A <- evalAccuracy(B, data.out = "ex")
A
# Rank the model's performance.
R <- do.Ranking(A)
R
R[, c(1:3, 20)]

# Visualize the results
plot(B, data.out = "ex", facet = "x", 
     which = c(0, 20, 40, 60, 75, 90))
plot(B, data.out = "mx", facet = "y", which = 2016) 

mpascariu/MortalityForecast documentation built on Sept. 28, 2020, 2:40 p.m.