Multivariate Temporal Disaggregaton of a System of Time Series by Regression Models.

Performs simultaneous temporal disaggregation of a system of low frequency time series into higher frequency series, based on the multivariate extension of the Chow-Lin model or the Random Walk approach (Fernandez).

Usage

multivariatechowlin(
  series,
  constant = TRUE,
  trend = FALSE,
  indicators = NULL,
  ccseries = NULL,
  ccdefinition = NULL,
  freq = 4L,
  rhos = 1,
  var = c("fromUnivariate", "allEquals", "userDefined"),
  var.includeCov = FALSE,
  var.shrinkCov = FALSE,
  var.matrix = NULL,
  rescale.variance = FALSE
)

Arguments

series

A named list of ts objects containing the low frequency time series to be disaggregated.

constant

Either a Boolean or a vector of Booleans. If a vector is provided, each element specifies whether a constant term is included in the model for each series, following the order in which they appear in the series object. The length of the the vector must match the number of series. If a single Boolean is provided (default if TRUE), it is applied to all series. Note that this argument is used only with Chow-Lin model (i.e., when rhos values are strictly less than 1). For further details, see the package vignette.

trend

Either a Boolean or a vector of Booleans. If a vector is provided, each element specifies whether a linear trend is included in the model for each series, following the order in which they appear in the series object. The length of the the vector must match the number of series. If a single Boolean is provided (default if FALSE), it is applied to all series.

indicators

a named list of ts objects or a named list of a list of ts objects. Each element represents one or more high-frequency indicator series associated with each series. If an element is NULL, no indicator is used for the corresponding series. The default value is NULL, meaning that no indicators are used for any series.

ccseries

A named list of ts objects containing the contemporaneous constraints. If NULL (the default), no contemporaneous constraints can be considered.

ccdefinition

A character vector defining each contemporaneous constraints. The elements of the vector must be written in the form \(z=w_1 y_1+\ldots+w_n y_n\) or \(c=w_1 y_1+\ldots+w_n y_n\) where:

\(z\) is the name of a contemporaneous constraint,
\((w_1,\ldots,w_n)\) are optional numeric weights,
\((y_1,\ldots,y_n)\) are the names of the time series and
\(c\) is a constant. The default is NULL, meaning that no contemporaneous constraint is considered.

freq

An integer giving the annual frequency of the disaggregated series. This argument is ignored when at least one indicator series is provided for any series.

rhos

Either a numeric value or a vector of numerics. If a vector is provided, each element specifies the value of the rho parameter associated to each series, following the order in which they appear in the series object. The length of the the vector must match the number of series. If a single numeric value is provided (default if 1, corresponding to the Fernandez model), it is applied to all series.

var

A character string specifying the method used to estimate the variance-covariance matrix of the innovations. The default is "fromUnivariate", meaning that it is estimated empirically from the residuals of the univariate models. Others options include "allEquals", which assume a diagonal matrix with identical variances (a strong assumption), and "userDefined", where the matrix is supplied by the user via the var.matrix argument. For additional details, see the package vignette.

var.includeCov

Boolean. Indicates whether non-diagonal elements of the innovation variance-covariance matrix may as well be estimated from the residuals of the univariate models. The default is FALSE, meaning that only a diagonal matrix is estimated. This argument is used only when var = "fromUnivariate".

var.shrinkCov

Boolean. Indicates whether a shrinkage estimator should be used for covariance. See the package vignette for more details. This argument is used only when var = "fromUnivariate" and var.includeCov = TRUE.

var.matrix

The variance-covariance matrix of the innovations. This argument is used only when var = "userDefined" and must be provided in that case.

rescale.variance

Boolean. Indicates whether the variance of the estimates should be rescaled based on the model residuals. The default is FALSE. This option has no impact on the disaggregated series, but affects the standard errors of both the disaggregated series and the estimated coefficients. See the package vignette for more details.

Value

An object of class "JD3_MULTITEMPDISAGG_RSLTS" is returned. The following are returned as a list:

regression [[1]] regression coefficients for each series;
estimation [[2]] disaggregated time series and standard errors, regression effects, smoothing parts, parameters and variance-covariance matrix of the innovations;

Vignette

For more information on the method, its arguments, and the other methods available in the package, see the package vignette:

In R: browseVignettes(package = "rjd3bench")
Online: https://rjdverse.github.io/rjd3bench/articles/rjd3bench.html

Examples