rjd3bridge

rjd3bridge is an R package designed to bridge the gap between RJDemetra (based on JDemetra+ version 2) and the modern rjd3 package suite (rjd3x13 and rjd3tramoseats, based on JDemetra+ version 3).

It provides bidirectional conversion tools (as_rjd2() and as_rjd3() functions) to seamlessly translate seasonal adjustment specifications and fitted model objects across both ecosystem versions without needing to rewrite custom configurations from scratch.

Installation

To install the developpment version:

• From r-universe:

install.packages('rjd3bridge', repos = c('https://aqlt.r-universe.dev', 'https://cloud.r-project.org'))

• From GitHub:

# install.packages("remotes")
remotes::install_github("AQLT/rjd3bridge")

Examples

Converting RJDemetra (v2) to rjd3 (v3)

library(rjd3bridge)
y <- RJDemetra::ipi_c_eu[, "FR"]
model_v2 <- RJDemetra::x13(y, spec = "RSA5c")
model_v3 <- as_rjd3(model_v2)

# Convert only a specification object
spec_v2 <- RJDemetra::x13_spec("RSA5c")
spec_v3 <- as_rjd3(spec_v2)

Converting rjd3 (v3) back to RJDemetra (v2)

library(rjd3bridge)
y <- rjd3toolkit::ABS$X0.2.09.10.M
model_v3 <- rjd3x13::x13(y)
model_v2 <- as_rjd2(model_v3)

The modelling_context has to be provided if the specification contains user-defined variables. Otherwise, the conversion will be complete:

library(rjd3bridge)
y <- rjd3toolkit::ABS$X0.2.09.10.M
regs_td <- rjd3toolkit::td(
  s = y,
  groups = c(1, 2, 0, 4, 5, 6, 3),
  contrasts = TRUE
)
colnames(regs_td) <- c("Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday")
context <- rjd3toolkit::modelling_context(variables = list(regs_td))
spec <- rjd3tramoseats::tramoseats_spec() |> 
  rjd3toolkit::set_tradingdays(
  option = "UserDefined",
  uservariable = c("r.Monday", "r.Tuesday", "r.Wednesday",
                   "r.Thursday", "r.Friday", "r.Saturday"),
  test = "None"
)
model_v3 <- rjd3tramoseats::tramoseats(y, spec, context)
model_v2 <- as_rjd2(model_v3, context = context)

For preprocessing models (RegARIMA or TRAMO), the input time series y must be provided to the conversion function. Otherwise the start date of the time series will default to year 1, which may lead to incorrect results (e.g., for outliers).

model_v3 <- rjd3tramoseats::tramo(y)
as_rjd2(model_v3)
#> y = regression model + arima (0, 1, 2, 0, 1, 1)
#> Log-transformation: yes
#> Coefficients:
#>           Estimate Std. Error
#> Theta(1)   -1.1304      0.047
#> Theta(2)    0.3088      0.047
#> BTheta(1)  -0.6523      0.037
#> 
#>             Estimate Std. Error
#> Mean      -3.044e-05          0
#> Week days -6.643e-04          0
#> 
#> 
#> Residual standard error: 0.00614 on 406 degrees of freedom
#> Log likelihood =  1510, aic = -1410 aicc = -1409, bic(corrected for length) = -10.11
as_rjd2(model_v3, y = y)
#> y = regression model + arima (0, 1, 1, 0, 1, 1)
#> Log-transformation: yes
#> Coefficients:
#>           Estimate Std. Error
#> Theta(1)   -0.8278      0.028
#> BTheta(1)  -0.4255      0.045
#> 
#>               Estimate Std. Error
#> Monday      -0.0109445      0.003
#> Tuesday      0.0048940      0.004
#> Wednesday    0.0001761      0.003
#> Thursday     0.0132928      0.004
#> Friday      -0.0024801      0.004
#> Saturday     0.0153509      0.004
#> Leap year    0.0410667      0.010
#> Easter [6]   0.0503888      0.007
#> AO (7-2000) -0.1972355      0.030
#> AO (6-2000)  0.1681679      0.030
#> 
#> 
#> Residual standard error: 0.03616 on 399 degrees of freedom
#> Log likelihood = 781.4, aic =  4199 aicc =  4199, bic(corrected for length) = -6.464