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.matrix = NULL
)

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 is is estimated 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.matrix

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

Value

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

regression [[1]] regression coefficients for each series;
estimation [[2]] disaggregated Time-Series and standard deviation for each series, regression effects, smoothing part, parameter and variance-covariance matrix;

Examples

# Low-frequency data
Y1 <- ts(c(30.0, 30.6, 31.2, 31.6), frequency = 1, start = c(2010,1))
Y2 <- ts(c(80.0, 81.2, 82.5, 82.6), frequency = 1, start = c(2010,1))
Y3 <- ts(c(8.0, 8.1, 8.2, 8.2), frequency = 1, start = c(2010,1))
lf_series <- list(y1 = Y1, y2 = Y2, y3 = Y3)

# Contemporaneous constraint
z <- ts(c(27.1,29.8,29.9,31.2,29.4,27.9,30.9,31.7,29.2,30.2,30.6,31.9,29.3,30.4,30.7,32.0), frequency = 4, start = c(2010,1))

# High-frequency indicators
x11 <- ts(c(7,7.2,8.1,7.5,8.5,7.8,8.1,8.4,8.6,7.8,8.0,8.3,8.7,7.9,8.0,8.6), frequency=4, start=c(2010,1))
x12 <- ts(c(18,19.5,19.0,19.7,18.5,19.0,20.3,20.0,18.6,19.5,20.4,20.1,18.7,19.1,20.4,20.8), frequency = 4, start = c(2010,1))
x2 <- NULL
x3 <- ts(c(1.5,1.8,2,2.5,2.0,1.5,1.7,2.1,2.1,1.6,1.6,2.2,2.3,1.7,1.9,2.3), frequency = 4, start = c(2010,1))
indic_series = list(y1 = list(x11, x12),
                    y2 = NULL,
                    y3 = x3)

# Check consistency between temporal and contemporaneous constraints
rowSums(cbind(Y1,Y2,Y3)) - stats::aggregate.ts(z) # ok!
#> Time Series:
#> Start = 2010 
#> End = 2013 
#> Frequency = 1 
#> [1]  0.000000e+00  1.421085e-14  0.000000e+00 -1.421085e-14

# Estimate models and get results

## Mix Chow-Lin - Fernandez
rslt1 <- multivariatechowlin(series = lf_series,
                             constant = c(FALSE, FALSE, TRUE),
                             trend = c(FALSE, FALSE, FALSE),
                             indicators = indic_series,
                             ccseries = list(z = z),
                             ccdefinition = "z=y1+y2+y3",
                             freq = 4L,
                             rhos = c(0.85, 1.0, 0.9),
                             var = "fromUnivariate",
                             var.matrix = NULL)

d1 <- do.call(cbind, rslt1$estimation$disagg)
ed1 <- do.call(cbind, rslt1$estimation$edisagg)

## Fernandez only (Random walk model) with user-defined variance-covariance matrix
rslt2 <- multivariatechowlin(series = lf_series,
                             constant = FALSE,
                             trend = FALSE,
                             indicators = indic_series,
                             ccseries = list(z = z),
                             ccdefinition = "z=y1+y2+y3",
                             freq = 4L,
                             rhos = 1.0,
                             var = "userDefined",
                             var.matrix = diag(c(0.003,0.01,0.001)))

d2 <- do.call(cbind, rslt2$estimation$disagg)
ed2 <- do.call(cbind, rslt2$estimation$edisagg)

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

Usage

Arguments

Value

See also

Examples