R packages

In this chapter

Core JDemetra+ Java algorithms can be accessed in R. This chapter provides an overview of the R ecosystem related to JDemetra+ version 2.4 and 3.x, still under construction. More details on specific functions are available in the relevant chapters in the Algorithm part of this documentation. Help pages, Readme pages and vignettes relative to each package provide a detailed description of all available functions: purpose, arguments, output and examples. There are directly linked to this documentation whenever their use in a selected algorithm is mentioned.

Configuration needed

• To run packages based on version 2.x: Java 8 or higher

• To run packages based on version 3.x: Java 17 or higher: how to get a portable Java and link it with R on your computer is explained here

Packages based on JDemetra+ version 2 algorithms

Packages corresponding to version 2.x core routines:

• RJDemetra on CRAN or https://github.com/jdemetra/rjdemetra
• rjdworkspace on https://github.com/InseeFrLab/rjdworkspace
• JDCruncheR on https://github.com/InseeFr/JDCruncheR
• ggdemetra on https://github.com/AQLT/ggdemetra
• rjdqa on https://github.com/AQLT/rjdqa

Packages based on JDemetra+ version 3 algorithms

Packages corresponding to version 3.x core routines:

• rjd3toolkit at https://github.com/rjdverse/rjd3toolkit
• rjd3x13 on https://github.com/rjdverse/rjd3x13
• rjd3tramoseats at https://github.com/rjdverse/rjd3tramoseats
• rjd3providers at https://github.com/rjdverse/rjd3providers
• rjd3workspace at https://github.com/rjdverse/rjd3workspace
• rjd3highfreq at https://github.com/rjdverse/rjd3highfreq
• rjd3filters at https://github.com/rjdverse/rjd3filters
• rjd3x11plus at https://github.com/rjdverse/rjd3x11plus
• rjd3bench at https://github.com/rjdverse/rjd3bench
• rjd3revisions at https://github.com/rjdverse/rjd3revisions
• rjd3nowcasting at https://github.com/rjdverse/rjd3nowcasting
• rjd3sts at https://github.com/rjdverse/rjd3sts
• rjd3stl at https://github.com/rjdverse/rjd3stl
• ggdemetra3 on https://github.com/AQLT/ggdemetra3

Algorithms available in R

Seasonal adjustment

Using JDemetra+ version 2.x

Algorithm	Package	Comments
X13-ARIMA	RJDemetra	Reg-Arima and X-11 decomposition available independently
TRAMO-SEATS	RJDemetra	Tramo available independently

Using JDemetra+ version 3.x

Algorithm	Package	Comments
X13-ARIMA	rjd3x13	Reg-Arima and X-11 decomposition available independently
Extended X-11	rjd3x11plus	Extended for high-frequency (infra-monthly) data and Trend estimation with local polynomial filters
TRAMO-SEATS	rjd3tramoseats	Tramo available independently
Extended Tramo	rjd3highfreq	Extended for high-frequency data
Extended Seats	rjd3highfreq	Extended for high-frequency data
STL+	rjd3stl	Airline based preadjustment and high-frequency data extension
Basic Structural Models	rjd3sts	State space framework

Version 3.x includes Revision Policies in X-13 and Tramo-Seats.

More details on functions parameters and retrieving output in the chapter dedicated to Seasonal Adjustment

Filtering and Trend estimation

Algorithm	Package
Moving average functions	rjd3filters
Local Polynomial Trend Estimation	rjd3filters, rjd3x11plus

Benchmarking and Temporal disaggregation

Algorithm	Package
Denton	rjd3bench
Cholette	rjd3bench
Cubic splines	rjd3bench
Temporal Disaggregation	rjd3bench

Utility functions available in R

The packages listed below contain utility functions useful when running a production process with massive datasets.

Running the cruncher and generating a quality report

JDemetra+ cruncher is an executable module designed for mass production of seasonally adjusted series .

Package	JD+ version	Comments
rjwsacruncher	2.x	estimation update and output
JDCruncheR	2.x	all the above + Quality Report

Wrangling JD+ workspaces

A workspace is a specific JDemetra+ data format (xml files) allowing to use the graphical user interface (GUI) and the cruncher.

Package	JD+ version	Comments
rjdworkspace	2.x	update meta data, set specifications, merge workspaces
rjd3providers	3.x	update metadata
rjd3workspace	3.x	set specifications, merge workspace

Generating enhanced output in SA estimation

This additional packages produce enhanced plots and diagnostic outputs.

Package	JD+ version	Comments
rjdmarkdown	2.x	enhanced print of diagnostics
ggdemetra	3.x	plots based on ggplot
ggdemetra3	3.x	plots based on ggplot
rjdqa	2.x	visual dashboard on one series

General structure

The R object resulting from an estimation is a list of lists containing raw data, parameters, output series and diagnostics.

RJDemetra output structure

Organised by domain:

V2 SA structure

To retrieve any element just navigate this list of lists.

rjd3x13 output structure

Results and specification are separated first and then organised by domain.

sa_x13_v3 <- RJDemetra::x13(y_raw, spec = "RSA5")
sa_x13_v3$result
sa_x13_v3$estimation_spec
sa_x13_v3$result_spec
sa_x13_v3$user_defined

To retrieve any element just navigate this list of lists.

Installation procedure

version 2

install.packages("RJDemetra")
remotes::install_github("InseeFrLab/rjdworkspace")
remotes::install_github("InseeFr/JDCruncheR")

version 3

To use version 3 based R packages you need Java 17 or higher. How to configure this in your R set up is explained here.

# install.packages("remotes")
# install.packages("devtools")
remotes::install_github("rjdverse/rjd3toolkit")
remotes::install_github("rjdverse/rjd3x13")
remotes::install_github("rjdverse/rjd3tramoseats")
remotes::install_github("rjdverse/rjd3providers")
remotes::install_github("rjdverse/rjd3workspace")
remotes::install_github("rjdverse/rjd3filters")
remotes::install_github("rjdverse/rjd3sts")
remotes::install_github("rjdverse/rjd3highfreq")
remotes::install_github("rjdverse/rjd3x11plus")
remotes::install_github("rjdverse/rjd3stl")
remotes::install_github("rjdverse/rjd3bench")
remotes::install_github("rjdverse/rjd3revisions")
remotes::install_github("rjdverse/rjd3nowcasting")
remotes::install_github("AQLT/ggdemetra3")

rjd3 suite of packages: overview

The sections below provide an overview of each package based on version 3.x of JDemetra+. For detailed description refer to the package’s own R readme file and documentation pages as linked below.

rjd3toolkit

Contains utility functions used in other rjd3 packages and has to be systematically installed before using any other rjd3 package. From a user point of view, it allows to:

• customize specifications in rjd3x13 and rjd3tramoseats

• generate user-defined regressors for calendar correction

• generate auxiliary variables (outliers, ramps..)

• run arima model estimations

• perform tests (seasonality, normality, white noise)

• access general functions such as autocorrelations, distributions

Documentation here

rjd3x13

rjd3x13 gives access to X13-ARIMA seasonal adjustment algorithm.

• Specification: created with spec_x11_default(), spec_x13_default(), spec_regarima_default() and customized with rjd3toolkit functions + set_x11()

• Apply model with x11(), x13(), fast.x13(), regarima(), fast.regarima()

• Refresh policies: regarima.refresh() and x13.refresh()

Documentation here

rjd3tramoseats

rjd3tramoseats gives access to TRAMO-SEATS seasonal adjustment algorithm.

• Specification: created with spec_tramoseats_default(), spec_tramo_default() and customized with rjd3toolkit functions + set_seats()

• Apply model with tramoseats(), fast.tramoseats(), tramo(), fast.tramo()

• Refresh policies: tramo.refresh() and tramoseats.refresh()

Documentation here

rjd3sts

Gives access to structural time series and state space models.

Documentation here

rjd3highfreq

Depends on rjd3sts

Seasonal adjustment of high frequency (infra-monthly) data:

• fractional airline based reg-Arima pre-adjustment

• fractional and multi airline decomposition

Documentation here

rjd3filters

The rjd3filters package allows to:

• easily create/combine/apply moving averages moving_average() (much more general than stats::filter()) and study their properties: plot coefficients (plot_coef()), gain (plot_gain()), phase-shift (plot_phase()) and different statics (diagnostic_matrix())

• trend-cycle extraction with different methods to treat endpoints:

• lp_filter() local polynomial filters of Proietti and Luati (2008) (including Musgrave): Henderson, Uniform, biweight, Trapezoidal, Triweight, Tricube, “Gaussian”, Triangular, Parabolic (= Epanechnikov)
  

• rkhs_filter() Reproducing Kernel Hilbert Space (RKHS) of Dagum and Bianconcini (2008) with same kernels
  

• fst_filter() FST approach of Grun-Rehomme, Guggemos, and Ladiray (2018)
  

• dfa_filter() derivation of AST approach of Wildi and McElroy (2019)

• change the filter used in X-11 for TC extraction

Create moving averages

library("rjd3filters")

m1 <- moving_average(rep(1, 3), lags = 1)
m1 # Forward MA
m2 <- moving_average(rep(1, 3), lags = -1)
m2 # centred MA

m1 + m2
m1 - m2
m1 * m2

Can be used to create all the MA of X-11:

e1 <- moving_average(rep(1, 12), lags = -6)
e1 <- e1 / sum(e1)
e2 <- moving_average(rep(1 / 12, 12), lags = -5)

# used to have the 1rst estimate of the trend
tc_1 <- M2X12 <- (e1 + e2) / 2
coef(M2X12) |> round(3)
si_1 <- 1 - tc_1
M3 <- moving_average(rep(1 / 3, 3), lags = -1)
M3X3 <- M3 * M3

# M3X3 moving average applied to each month
coef(M3X3) |> round(3)
M3X3_seasonal <- to_seasonal(M3X3, 12)
coef(M3X3_seasonal) |> round(3)
s_1 <- M3X3_seasonal * si_1
s_1_norm <- (1 - M2X12) * s_1
sa_1 <- 1 - s_1_norm
henderson_mm <- moving_average(lp_filter(horizon = 6)$filters.coef[, "q=6"],
    lags = -6
)
tc_2 <- henderson_mm * sa_1
si_2 <- 1 - tc_2
M5 <- moving_average(rep(1 / 5, 5), lags = -2)
M5X5_seasonal <- to_seasonal(M5 * M5, 12)
s_2 <- M5X5_seasonal * si_2
s_2_norm <- (1 - M2X12) * s_2
sa_2 <- 1 - s_2_norm
tc_f <- henderson_mm * sa_2

par(mai = c(0.3, 0.3, 0.2, 0))
layout(matrix(c(1, 1, 2, 3), 2, 2, byrow = TRUE))

plot_coef(tc_f)
plot_coef(sa_2, col = "orange", add = TRUE)
legend("topleft",
    legend = c("Final TC filter", "Final SA filter"),
    col = c("black", "orange"), lty = 1
)

plot_gain(tc_f)
plot_gain(sa_2, col = "orange", add = TRUE)

plot_phase(tc_f)
plot_phase(sa_2, col = "orange", add = TRUE)

Apply a moving average

y <- retailsa$AllOtherGenMerchandiseStores
trend <- y * tc_1
sa <- y * sa_1
plot(window(ts.union(y, trend, sa), start = 2000),
    plot.type = "single",
    col = c("black", "orange", "lightblue")
)

rjd3x11plus

Depends on rjd3filters

• Extension of X-11 decomposition with multiple non integer periodicities

• Trend estimation with local polynomial based filters

Full documentation here

rjd3stl

rjd3stl contains usual STL functions and an airline model based pre-adjustment module. Is also tailored to handle high-frequency data.

Full documentation here

ggdemetra3

ggdemetra3 uses ggplot2 to add seasonal adjustment statistics to your plot (Like ggdemetra but compatible with version 3.x.). Also compatible with high-frequency methods:

library("ggdemetra3")

spec <- spec_x13_default("rsa3") |> set_tradingdays(option = "WorkingDays")
p_ipi_fr <- ggplot(data = ipi_c_eu_df, mapping = aes(x = date, y = FR)) +
    geom_line() +
    labs(
        title = "SA - IPI-FR",
        x = NULL, y = NULL
    )
p_sa <- p_ipi_fr +
    geom_sa(
        component = "y_f(12)", linetype = 2,
        spec = spec
    ) +
    geom_sa(component = "sa", color = "red") +
    geom_sa(component = "sa_f", color = "red", linetype = 2)
p_sa
p_sa +
    geom_outlier(
        geom = "label_repel",
        coefficients = TRUE,
        ylim = c(NA, 65), force = 10,
        arrow = arrow(
            length = unit(0.03, "npc"),
            type = "closed", ends = "last"
        ),
        digits = 2
    )

rjd3bench

Tailored for Benchmarking and temporal disaggregation

rjd3revisions

Revision analysis, more info here