Package 'fable.rjd3'

Backcast of the raw series

Description

Backcasts the raw series, i.e. predicts the periods preceding its start, from the pre-adjustment model of a model fitted with X13(), TRAMOSEATS(), REGARIMA() or TRAMO(). fabletools has no backcast generic, so this function plays the role of forecast() for the past. The distribution is built as for forecasts (log-normal for log-transformed series, normal otherwise; see the Distribution section of forecast.jd3_model).

Usage

backcast(object, h, ...)

## S3 method for class 'jd3_model'
backcast(object, h, ...)

## S3 method for class 'mdl_ts'
backcast(object, h, ...)

## S3 method for class 'mdl_df'
backcast(object, h, ...)

Arguments

object	A fitted jd3_model (usually a column of a mable).
h	Number of periods to backcast, immediately before the start of the series.
...	Other arguments passed by forecast() (e.g. times), ignored.

Value

For a jd3_model or a mdl_ts, a tsibble with the h dates preceding the series and a distribution column dist (see distributional). For a mable, a tibble with the key columns, .model, the dates and dist, one block per model. Use mean() or median() on dist to get a point backcast; the median equals JDemetra+'s backcast.

See Also

forecast.jd3_model, jd3_indicator() for the raw indicators (y_b, y_eb...).

Examples

y  <- tsibble::as_tsibble(AirPassengers)
mb <- model(y, x13 = X13(value))

bc <- backcast(mb, h = 12)
bc

# A single model
bc1 <- backcast(mb$x13[[1]], h = 12)
median(bc1$dist)

---

Export a mable of JDemetra+ models as a new workspace

Description

Writes a new workspace readable by the JDemetra+ graphical interface, with one SA-item per model: raw series, estimation specification, reference specification, and the modelling context (calendars and regressors). A workspace has a single context: regressors of different models that share the same name but have different values (e.g. xreg(col) in several series) are renamed ⁠<name>_<sap>_<series>⁠, in the context and in the specifications. Results, comments, priorities and links to data providers are not exported; the series are stored as static data.

Usage

export_jws(mb, file, replace = FALSE, sap = "sap", series = "series")

Arguments

mb	A mable of models fitted with X13() or TRAMOSEATS().
file	Path of the workspace (xml master file) to create.
replace	Replace the workspace if file already exists. By default an existing file is never overwritten.
sap, series	Names of the key columns giving the SA-processing and the SA-item. Without them, a single SA-processing is created and the SA-item names are built from the other keys.

Value

file, invisibly.

See Also

import_jws()

Examples

mb <- model(as_tsibble(AirPassengers), sa = X13(value))
export_jws(mb, tempfile(fileext = ".xml"))

---

Forecast of the raw series

Description

Forecasts the raw series from the pre-adjustment model (RegARIMA / TRAMO) of a model fitted with X13(), TRAMOSEATS(), REGARIMA() or TRAMO(). The values are read from the stored JDemetra+ result, so the seasonal adjustment itself is not affected. For the past, see backcast().

Usage

## S3 method for class 'jd3_model'
forecast(object, new_data = NULL, ...)

Arguments

object	A fitted jd3_model (usually a column of a mable).
new_data	A tsibble of future dates, built by fabletools from the horizon h. It must start right after the end of the series.
...	Other arguments passed by forecast() (e.g. times), ignored.

Value

A distribution vector (see distributional) with one element per row of new_data: log-normal if the series is log-transformed, normal otherwise. Called through forecast() on a mable, the result is a fable.

Distribution

For log-transformed series, the forecast distribution is log-normal, with a median equal to JDemetra+'s forecast (a plain back-transformation of the forecast of the log). Its mean is therefore the bias-corrected forecast, and the difference grows with the horizon. For other series, the distribution is normal and its mean equals the median.

The default point forecast .mean of the resulting fable is the mean of the distribution. Use the point_forecast argument of forecast() to choose another summary, e.g. point_forecast = list(.median = median) to obtain JDemetra+'s forecast.

See Also

backcast(), X13(), TRAMOSEATS(), jd3_indicator() to extract the raw indicators (y_f, y_ef...).

Examples

y  <- tsibble::as_tsibble(AirPassengers)
mb <- model(y, x13 = X13(value))

# The mean is the bias-corrected forecast
forecast(mb, h = 12)

# JDemetra+'s own forecast (median of the distribution)
forecast(mb, h = 12, point_forecast = list(.median = median))

---

Methods for fitted JDemetra+ models

Description

Standard fabletools / generics verbs for models fitted with X13(), TRAMOSEATS(), REGARIMA() and TRAMO(). They can be applied directly on a mable. components() is only available for the seasonal adjustment models (X13() and TRAMOSEATS()).

Usage

## S3 method for class 'jd3_model'
glance(x, ids = NULL, ...)

## S3 method for class 'jd3_sa_model'
components(object, ...)

## S3 method for class 'jd3_model'
model_sum(x)

## S3 method for class 'jd3_model'
tidy(x, ...)

## S3 method for class 'jd3_model'
report(
  object,
  digits = max(3L, getOption("digits") - 3L),
  signif.stars = getOption("show.signif.stars"),
  thresholds_pval = getOption("fable.rjd3.thresholds_pval"),
  ...
)

Arguments

ids	For glance(): named or unnamed character vector of indicator ids (see jd3_dictionary()); the names become the column names. Numbers and strings give one column; tests give two columns, ⁠<name>_stat⁠ and ⁠<name>_pvalue⁠; other indicators give NA. If NULL, the default set of the method is used (information criteria, number of observations, tests on the residuals and on the seasonally adjusted series, M and Q statistics for X-13, SEATS tests for TRAMO-SEATS).
...	Currently unused.
object, x	A jd3_model (usually accessed through a mable).
digits	minimum number of significant digits to be used for most numbers.
signif.stars	logical; if TRUE, P-values are additionally encoded visually as ‘significance stars’ in order to help scanning of long coefficient tables. It defaults to the show.signif.stars slot of options.
thresholds_pval	Threshold used to interpret p-values statistic test. By default c(Severe = 0.001, Bad = 0.01, Uncertain = 0.05, Good = Inf).

Value

• components(): a dable with the response and the columns trend, seasonal, irregular and season_adjust.

• glance(): a one-row tibble of diagnostics.

• tidy(): a tibble with one row per coefficient (ARIMA parameters, regression variables, outliers, calendar effects) and the columns term, estimate, std.error, statistic and p.value. Coefficients fixed in the specification are included, with missing standard errors and tests.

• report(): invisibly, the model; prints the decomposition mode or the transformation, the number of observations, the coefficients, the tests on the residuals and, for seasonal adjustment models, the quality diagnostics (M and Q statistics, residual seasonality tests on the adjusted series).

• model_sum(): a short string describing the model.

Examples

mb <- model(tsibble::as_tsibble(AirPassengers), x13 = X13(value))
components(mb)
glance(mb)
tidy(mb)
report(mb)

---

Import a JDemetra+ workspace as a mable

Description

Reads all the SA-items of a workspace and fits them with X13() or TRAMOSEATS() using the specifications stored in the workspace: the estimation specification is the model specification, the reference specification is kept for the refresh policies (see jd3_model_methods_extra).

Usage

import_jws(file, compute = TRUE)

Arguments

file	Path to the master xml file of the workspace.
compute	Compute the workspace before reading it (required to access the SA-items).

Value

A mable with the key columns sap (SA-processing) and series (SA-item), and a model column sa holding X13 and TRAMOSEATS models. SA-items with another type of specification (e.g. RegARIMA only) are skipped with a warning.

See Also

export_jws()

Examples

file <- system.file("workspaces", "workspace_test.xml", package = "rjd3workspace")
mb <- import_jws(file)
mb

---

List the indicators available in a fitted model

Description

Wraps rjd3toolkit::dictionary() on the Java result of the model. The returned ids can be passed to jd3_indicator().

Usage

jd3_dictionary(object, ...)

## S3 method for class 'jd3_model'
jd3_dictionary(object, ...)

## S3 method for class 'mdl_ts'
jd3_dictionary(object, ...)

## S3 method for class 'mdl_df'
jd3_dictionary(object, ...)

Arguments

object	A fitted mable (mdl_df), a mdl_ts or a jd3_model.
...	Currently unused.

Value

A data frame (class JD3_FULL_DICTIONARY) with one row per indicator: its id (name), description and output type. For a mable, the dictionary of the first model is returned.

See Also

jd3_indicator()

Examples

mb <- model(tsibble::as_tsibble(AirPassengers), x13 = X13(value))
head(jd3_dictionary(mb))

---

Extract any JDemetra+ indicator from a fitted model

Description

Wraps rjd3toolkit::result() on the Java result of the model. If the Java pointer is no longer valid (e.g. after saveRDS()/readRDS()), the model is transparently re-estimated (once per loaded object). The ids available are listed by jd3_dictionary().

Usage

jd3_indicator(object, id, ...)

## S3 method for class 'jd3_model'
jd3_indicator(object, id, drop = TRUE, ...)

## S3 method for class 'mdl_ts'
jd3_indicator(object, id, drop = TRUE, ...)

## S3 method for class 'mdl_df'
jd3_indicator(object, id, simplify = TRUE, drop = TRUE, ...)

Arguments

object	A fitted mable (mdl_df), a mdl_ts or a jd3_model.
id	Indicator id (see jd3_dictionary()).
...	Currently unused.
drop	If TRUE (default) and a single id is given, the indicator itself is returned instead of a list of length one. Several ids always give a named list.
simplify	For a mable only: simplify the result according to the type of the indicator (see Value). FALSE always returns a list-column.

Value

For a mdl_ts or jd3_model, the indicator as returned by rjd3toolkit::result(): a number, a string, a ts, a matrix, a JD3_TEST...

For a mable, a tibble with the key columns, .model, id and:

• numbers or strings: a column value;

• ts series: one row per date, with columns index and value;

• tests: columns statistic, p.value and distribution;

• anything else (or simplify = FALSE): a list-column value with one element per model. With several ids (or drop = FALSE), a named list of such tibbles, one per id.

See Also

jd3_dictionary()

Examples

mb <- model(as_tsibble(AirPassengers), x13 = X13(value), ts = TRAMOSEATS(value))

jd3_indicator(mb, "likelihood.aic")      # one number per model
jd3_indicator(mb, "sa")                  # seasonally adjusted series, long format
jd3_indicator(mb, "residuals.kurtosis")  # test: statistic and p-value

# The raw object of a single model
jd3_indicator(mb$x13[[1]], "sa")

---

Extract the model object from a fitted mable cell

Description

Returns the jd3_model object stored in a fitted model. It is mostly useful for programming; jd3_indicator() and jd3_dictionary() call it internally.

Usage

jd3_model(x)

Arguments

x	A mdl_ts (e.g. mb$x13[[1]]) or a jd3_model.

Value

A jd3_model object: a list holding the data (a tsibble), the series (ts), the final specification (spec), the reference specification (ref_spec), the modelling context, the engine, the name of the response and a cache for the Java result.

Examples

mb <- model(tsibble::as_tsibble(AirPassengers), x13 = X13(value))
m <- jd3_model(mb$x13[[1]])
class(m)

---

Fitted values, residuals, interpolation and updating of JDemetra+ models

Description

Methods for models fitted with X13(), TRAMOSEATS(), REGARIMA() or TRAMO(). They can be applied directly on a mable.

Usage

## S3 method for class 'jd3_model'
fitted(object, ...)

## S3 method for class 'jd3_model'
residuals(object, type = c("innovation", "regression"), ...)

## S3 method for class 'jd3_model'
interpolate(object, new_data = NULL, specials = NULL, ...)

## S3 method for class 'jd3_model'
refit(
  object,
  new_data,
  specials = NULL,
  policy = c("Fixed", "FreeParameters", "Complete", "Outliers_StochasticComponent",
    "Outliers", "FixedParameters", "FixedAutoRegressiveParameters", "Current"),
  period = 0,
  start = NULL,
  end = NULL,
  ...
)

## S3 method for class 'jd3_model'
stream(
  object,
  new_data,
  specials = NULL,
  policy = c("Fixed", "FreeParameters", "Complete", "Outliers_StochasticComponent",
    "Outliers", "FixedParameters", "FixedAutoRegressiveParameters", "Current"),
  period = 0,
  start = NULL,
  end = NULL,
  ...
)

Arguments

object	A fitted jd3_model (usually a column of a mable).
...	Currently unused.
type	Residual type: "innovation" (default) or "regression".
new_data	A tsibble. For interpolate(), the data with missing values (same dates as the data of the model). For refit(), the new full data. For stream(), the new observations only.
specials	Currently unused.
policy	Refresh policy, a character string (see the functions above for the available policies; default "Fixed": the estimated parameters are kept).
period, start, end	additional parameters used to specify the span when policy = "Current" or policy = "Outliers" or policy = "Outliers_StochasticComponent" period: numeric, number of observations in a year (12, 4...), compulsory, if false or missing, re-estimation with refreshed specification won't work. When policy = "Outliers" or policy = "Outliers_StochasticComponent" start has to be specified as the date from which outliers will be re-identified end is not used, if specified it will be ignored. When policy = "Current" start and end have to be both specified and indicate the span on which additive outliers (AO) will be added. Span definition: start and end: defined as arrays of two elements: year and first period (for example, period = 12 and start=c(1980, 1) stands for January 1980)

Details

• fitted() returns the one-step-ahead fitted values of the pre-adjustment model, on the scale of the series. Together with residuals(), it makes augment() and accuracy() work.

• residuals() returns the innovation residuals of the pre-adjustment model (type = "innovation") or the linearized series the ARIMA part is applied to (type = "regression"). Dates for which a residual is not defined (e.g. lost to differencing) are NA.

• interpolate() replaces the missing values of the series by the estimates of the model. new_data must cover the same dates as the data used to fit the model.

• refit() applies the model to new data, using a JDemetra+ refresh policy (see rjd3x13::x13_refresh() and rjd3tramoseats::tramoseats_refresh()): the specification is refreshed from the estimated model and the reference specification stored in the model (see the ref_spec argument of X13()).

• stream() appends new observations to the data of the model and refreshes it in the same way. new_data contains the new observations only.

Value

• fitted(), residuals(): a numeric vector with one value per observation (NA where undefined). Called on a mable, a tibble with the index and the .fitted or .resid column.

• interpolate(): new_data with the missing values filled in.

• refit(), stream(): a jd3_model fitted on the new data. Called on a mable, a mable.

See Also

X13(), TRAMOSEATS(), forecast.jd3_model(), jd3_model_methods

Examples

y  <- as_tsibble(AirPassengers)
mb <- model(y, x13 = X13(value))

# Fitted values and residuals (also used by augment() and accuracy())
fitted(mb)
residuals(mb)                         # innovation residuals
residuals(mb, type = "regression")    # linearized series
augment(mb)
accuracy(mb)

# Interpolation of missing values
y_na <- y
y_na$value[c(20, 21)] <- NA
mb_na <- model(y_na, x13 = X13(value))
interpolate(mb_na, y_na)

# Updating a model when new observations arrive
start_new <- yearmonth("1960 Jan")
y_old  <- y[y$index < start_new, ]
y_new  <- y[y$index >= start_new, ]
mb_old <- model(y_old, x13 = X13(value))

# stream(): append the new observations, estimated parameters kept
mb_ext <- stream(mb_old, y_new)
components(mb_ext)

# refit(): apply the model to the full data
refit(mb_old, y)                      # refresh policy "Fixed" (default)

---

Specials available in the model formula

Description

Specials are used on the right-hand side of the formula of X13(), TRAMOSEATS(), REGARIMA() or TRAMO(), e.g. X13(y ~ transform(fun = "Log") + xreg(strike)). Each one records a call to a function of rjd3toolkit that is applied to the base specification (the spec argument).

Specials

• spec(x): base specification (name or specification object); equivalent to the spec argument of X13(), TRAMOSEATS(), REGARIMA() or TRAMO().

• basic(...): arguments passed to rjd3toolkit::set_basic().

• transform(...): arguments passed to rjd3toolkit::set_transform().

• estimate(...): arguments passed to rjd3toolkit::set_estimate().

• automodel(...): arguments passed to rjd3toolkit::set_automodel().

• outlier(...): arguments passed to rjd3toolkit::set_outlier().

• add_outlier(...): arguments passed to rjd3toolkit::add_outlier() (e.g. ⁠type = "AO", date = "2020-04-01"⁠).

• easter(...): arguments passed to rjd3toolkit::set_easter().

• tradingdays(...): arguments passed to rjd3toolkit::set_tradingdays().

• arima(...): arguments passed to rjd3toolkit::set_arima().

• ramp(...): arguments passed to rjd3toolkit::add_ramp().

• usrdefvar(...): arguments passed to rjd3toolkit::add_usrdefvar().

• x11(...): arguments passed to rjd3x13::set_x11() (X-11 decomposition parameters; use with X13() only).

• seats(...): arguments passed to rjd3tramoseats::set_seats() (SEATS decomposition parameters; use with TRAMOSEATS() only).

• benchmarking(...): arguments passed to rjd3toolkit::set_benchmarking() (use with X13() or TRAMOSEATS() only).

• xreg(..., regeffect = NULL, group = "r", lag = 0, coef = NULL): user-defined regressors taken from columns of the tsibble and added to the context. By default (regeffect = NULL), the regressors are only added to the context (with the group="r"). If regeffect is equals to "Undefined", "Trend", "Seasonal", "Irregular", "Series" or "SeasonallyAdjusted" then variable is also added to the specification using the rjd3toolkit::add_usrdefvar() function with the parameters lag and coeff. If regeffect = "Calendar", then the variable is added to the specification using the rjd3toolkit::set_tradingdays() function.

Precedence

The base spec is applied first, then the specials in a fixed order (basic, transform, estimate, automodel, outlier, add_outlier, easter, tradingdays, arima, ramp, usrdefvar, benchmarking, x11, seats), then xreg(). Later settings override earlier ones. A special that does not apply to the method (e.g. x11() in TRAMOSEATS()) is ignored with a warning.

See Also

X13(), TRAMOSEATS()

---

Identify the outliers of a JDemetra+ model

Description

Returns the outlying observations detected (or specified) in the pre-adjustment model of a model fitted with X13(), TRAMOSEATS(), REGARIMA() or TRAMO(): additive outliers (AO), level shifts (LS), transitory changes (TC) and seasonal outliers (SO). An outlier is dated at one observation, even when its effect (e.g. a level shift) lasts longer.

Usage

## S3 method for class 'jd3_model'
outliers(object, type = NULL, ...)

Arguments

object	A fitted jd3_model (usually a column of a mable).
type	Optional character vector of outlier types to keep, e.g. c("AO", "LS"). NULL (default) keeps all of them.
...	Currently unused.

Value

For a jd3_model, an integer vector with the row positions of the outliers in the data of the model (the contract of fabletools::outliers()). Called on a mable, outliers() returns the corresponding rows of the data, with the key columns and .model. The type, estimate and test statistics of the outliers are available in tidy().

See Also

X13(), TRAMOSEATS(), jd3_model_methods

Examples

mb <- model(tsibble::as_tsibble(AirPassengers), x13 = X13(value))

# Rows of the data that are outliers
outliers(mb)

# Additive outliers only
outliers(mb, type = "AO")

---

Reg-ARIMA Model

Description

Defines a Reg-ARIMA model (the pre-adjustment model used in X-13), without seasonal adjustment, to be used inside fabletools::model(). The model is estimated with rjd3x13::jregarima() and the Java result is kept in the fitted model. Arguments and specials are the same as for X13(), except x11(), seats() and benchmarking(), which do not apply and are ignored with a warning.

Usage

REGARIMA(
  formula,
  spec = c("rg4", "rg0", "rg1", "rg2c", "rg3", "rg5c"),
  context = NULL,
  ref_spec = spec,
  ...
)

Arguments

formula	Model formula: the name of the variable to adjust (y), or y ~ specials (see jd3_specials).
spec	Base specification: a character name (see rjd3x13::regarima_spec() for the available names; the default is the first one) or a specification object from rjd3x13::regarima_spec().
context	A modelling context created with rjd3toolkit::modelling_context() (e.g. for custom calendars), or NULL.
ref_spec	Reference specification stored in the model: it is the reference of the refresh policies used by refit() and stream(). Defaults to spec, that is the specification before its modification by the specials.
...	Currently unused.

Details

There is no decomposition, so components() is not available. The other methods are: fitted(), residuals(), forecast(), backcast(), outliers(), tidy(), glance() and report().

Value

A model definition that can be passed to fabletools::model(). Once fitted, the mable column contains a jd3_model object.

See Also

X13(), TRAMO(), jd3_specials, jd3_model_methods

Examples

y <- tsibble::as_tsibble(AirPassengers)
mb <- model(y, regarima = REGARIMA(value))
tidy(mb)
report(mb)

# Forecasts of the raw series
forecast(mb, h = 12)

---

TRAMO Model

Description

Defines a TRAMO model (the pre-adjustment model used in TRAMO-SEATS), without seasonal adjustment, to be used inside fabletools::model(). The model is estimated with rjd3tramoseats::jtramo() and the Java result is kept in the fitted model. Arguments and specials are the same as for TRAMOSEATS(), except x11(), seats() and benchmarking(), which do not apply and are ignored with a warning.

Usage

TRAMO(
  formula,
  spec = c("trfull", "tr0", "tr1", "tr2", "tr3", "tr4", "tr5"),
  context = NULL,
  ref_spec = spec,
  ...
)

Arguments

formula	Model formula: the name of the variable to adjust (y), or y ~ specials (see jd3_specials).
spec	Base specification: a character name (see rjd3tramoseats::tramo_spec() for the available names; the default is the first one) or a specification object from rjd3tramoseats::tramo_spec().
context	A modelling context created with rjd3toolkit::modelling_context() (e.g. for custom calendars), or NULL.
ref_spec	Reference specification stored in the model: it is the reference of the refresh policies used by refit() and stream(). Defaults to spec, that is the specification before its modification by the specials.
...	Currently unused.

Details

There is no decomposition, so components() is not available. The other methods are: fitted(), residuals(), forecast(), backcast(), outliers(), tidy(), glance() and report().

Value

A model definition that can be passed to fabletools::model(). Once fitted, the mable column contains a jd3_model object.

See Also

TRAMOSEATS(), REGARIMA(), jd3_specials, jd3_model_methods

Examples

y <- tsibble::as_tsibble(AirPassengers)
mb <- model(y, tramo = TRAMO(value))
tidy(mb)
report(mb)

# Forecasts of the raw series
forecast(mb, h = 12)

---

TRAMO-SEATS seasonal adjustment (JDemetra+ v3)

Description

Defines a TRAMO-SEATS model to be used inside fabletools::model(). The model is estimated with rjd3tramoseats::jtramoseats() and the Java result is kept in the fitted model. Arguments and specials are the same as for X13().

Usage

TRAMOSEATS(
  formula,
  spec = c("rsafull", "rsa0", "rsa1", "rsa2", "rsa3", "rsa4", "rsa5"),
  context = NULL,
  ref_spec = spec,
  ...
)

Arguments

formula	Model formula: the name of the variable to adjust (y), or y ~ specials (see jd3_specials).
spec	Base specification: a character name or a specification object from rjd3tramoseats::tramoseats_spec().
context	A modelling context created with rjd3toolkit::modelling_context() (e.g. for custom calendars), or NULL.
ref_spec	Reference specification stored in the model: it is the reference of the refresh policies used by refit() and stream(). Defaults to spec, that is the specification before its modification by the specials.
...	Currently unused.

Value

A model definition that can be passed to fabletools::model(). Once fitted, the mable column contains a jd3_model object. Use components() to get the decomposition as a dable (columns trend, seasonal, irregular and season_adjust).

See Also

X13(), TRAMO(), jd3_specials, jd3_model_methods

Examples

y <- tsibble::as_tsibble(AirPassengers)

mb <- model(y, ts = TRAMOSEATS(value))
components(mb)

# Fit both methods side by side and plot the seasonally adjusted series
mb <- model(y, x13 = X13(value), ts = TRAMOSEATS(value))
if (requireNamespace("ggtime", quietly = TRUE) && requireNamespace("ggplot2", quietly = TRUE)) {
  ggplot2::autoplot(components(mb))
}

---

X-13ARIMA seasonal adjustment (JDemetra+ v3)

Description

Defines an X-13ARIMA model to be used inside fabletools::model(). The model is estimated with rjd3x13::jx13() and the Java result is kept in the fitted model, so that all JDemetra+ indicators can be extracted afterwards (see jd3_indicator()).

Usage

X13(
  formula,
  spec = c("rsa4", "rsa0", "rsa1", "rsa2c", "rsa3", "rsa5c"),
  context = NULL,
  ref_spec = spec,
  ...
)

Arguments

formula	Model formula: the name of the variable to adjust (y), or y ~ specials (see jd3_specials).
spec	Base specification: a character name (see rjd3x13::x13_spec() for the available names; the default is the first one) or a specification object from rjd3x13::x13_spec(). The specials of the formula are applied to it.
context	A modelling context created with rjd3toolkit::modelling_context() (e.g. for custom calendars), or NULL.
ref_spec	Reference specification stored in the model: it is the reference of the refresh policies used by refit() and stream(). Defaults to spec, that is the specification before its modification by the specials.
...	Currently unused.

Details

The specification can be set in two (combinable) ways:

1. Ex ante, with the spec argument: a name or a specification object created with rjd3x13::x13_spec() and modified with the ⁠set_*()⁠ functions of rjd3toolkit.

2. In the formula, with specials such as transform() or xreg() (see jd3_specials). This is the only way to use regressors that are columns of the tsibble.

Specials override the base spec.

Value

A model definition that can be passed to fabletools::model(). Once fitted, the mable column contains a jd3_model object. Use components() to get the decomposition as a dable (columns trend, seasonal, irregular and season_adjust).

See Also

TRAMOSEATS(), REGARIMA(), jd3_specials, jd3_model_methods, jd3_indicator(), jd3_dictionary()

Examples

y <- tsibble::as_tsibble(AirPassengers)

# Default specification
mb <- model(y, x13 = X13(value))
components(mb)

# Specification created ex ante with rjd3toolkit
my_spec <- rjd3x13::x13_spec("rsa3")
my_spec <- rjd3toolkit::set_transform(my_spec, fun = "Log")
mb <- model(y, x13 = X13(value, spec = my_spec))

# Specification set in the formula with specials
mb <- model(
  y,
  x13 = X13(value ~ transform(fun = "Log") +
              outlier(outliers.type = c("AO", "LS")))
)

# Regressor taken from a column of the tsibble
y$shock <- as.numeric(y$index == tsibble::yearmonth("1955 Jan"))
mb <- model(y, x13 = X13(value ~ xreg(shock, regeffect = "Series")))

---