Closed ThomasLaurentInsee closed 2 years ago
Indeed, it is odd, I'm not sure it has an impact since in the function span_r2jd
(https://github.com/jdemetra/rjdemetra/blob/b6ca1e9e8f0d15eb960c59989cc4a9fa01f202c5/R/spec_rjd.R#L790) only the needed information is used. To be tested.
I confirm it has no impact but I will change it for clarity.
library(RJDemetra)
y = ts(ipi_c_eu[,"FR"],start = 2020, frequency = 12)
spec = x13_spec(estimate.from = "2025-01-01",
outlier.from = "2025-01-01")
mod = x13(y, spec)
mod$regarima$specification$span
#> type d0 d1 n0 n1
#> estimate From 2025-01-01 2020-12-31 0 0
#> outlier From 2025-01-01 2020-12-31 0 0
mod$regarima$specification$estimate
#> preliminary.check span tolerance
#> TRUE From 2025-01-01 1e-07
mod$regarima$specification$outliers
#> enabled span ao tc ls so usedefcv cv method tcrate
#> TRUE From 2025-01-01 TRUE TRUE TRUE FALSE TRUE 4 AddOne 0.7
mod$regarima
#> y = regression model + arima (0, 1, 1, 0, 1, 1)
#> Log-transformation: no
#> Coefficients:
#> Estimate Std. Error
#> Theta(1) -0.7261 0.041
#> BTheta(1) -0.6692 0.046
#>
#> Estimate Std. Error
#> Monday -0.3564 0.316
#> Tuesday -1.7265 0.319
#> Wednesday -0.8741 0.316
#> Thursday 0.8502 0.320
#> Friday 1.0392 0.320
#> Saturday 0.6015 0.326
#> Leap year -1.0575 0.976
#> Easter [1] -2.0775 0.655
#> TC (4-2050) -22.6730 2.593
#> TC (3-2050) -19.7840 2.580
#> LS (11-2038) -11.0964 1.973
#> AO (5-2041) 11.5111 2.283
#> LS (1-2039) -6.8908 1.974
#>
#>
#> Residual standard error: 2.585 on 283 degrees of freedom
#> Log likelihood = -712.2, aic = 1456 aicc = 1458, bic(corrected for length) = 2.186
Created on 2022-03-11 by the reprex package (v2.0.1)
Ok, thanks. But if the parameters are ignored by spec_rjd in that case, wouldn't it be better to keep it at NA ?
For example, if no span is specified in x13_spec, you will only have type ="All" with non-NA values.
> spec = x13_spec()
> mod = x13(y, spec)
> mod$regarima$specification$span
type d0 d1 n0 n1
estimate All 1900-01-01 2020-12-31 0 0
outlier All 1900-01-01 2020-12-31 0 0
Those dates were the default values of JDemetra+. When you change the "Type" value of a span, you have those default dates. In RJDemetra the type parameter is automatically set with the other parameters, but the different dates where only used when needed. But Indeed the results of mod$regarima$specification$span
wasn't clear. However mod$regarima$specification$outliers$span
and mod$regarima$specification$estimate$span
provide a simplified message. I changed it in the last released for clarity:
library(RJDemetra)
y = ts(,start = 2020, frequency = 12)
spec = x13_spec(estimate.from = "2010-01-01",
outlier.to = "2015-01-01")
mod = x13(ipi_c_eu[,"FR"], spec)
mod$regarima$specification$span
#> type d0 d1 n0 n1
#> estimate From 2010-01-01 <NA> 0 0
#> outlier To <NA> 2015-01-01 0 0
mod$regarima$specification$estimate$span
#> [1] "From 2010-01-01"
mod$regarima$specification$outliers$span
#> [1] "Until 2015-01-01"
mod = x13(ipi_c_eu[,"FR"])
mod$regarima$specification$span
#> type d0 d1 n0 n1
#> estimate All NA NA 0 0
#> outlier All NA NA 0 0
mod$regarima$specification$estimate$span
#> [1] "All"
mod$regarima$specification$outliers$span
#> [1] "All"
Hello,
The by default to.date, as defined in spec_transform.R, is passed (31-12-2020). It should be moved forward (2100 ?) or set to a value that implied that the whole serie will be considered (as window(end=NULL) do).