Skip to content

JDemetra+ v2.2.3
Compare
Choose a tag to compare
@charphi charphi released this 07 Jul 10:36
v2.2.3
8d5edfc

This is the release of JDemetra+ 2.2.3.
Java SE 8 or later version is required to run it.

Release notes

Statistical Algorithms

  • Fix error in TradingDaysSpec
  • Correction for the extreme value detection if the difference between the boundary and the value is below machine precision.
  • Fix for LogAdd for half yearly data with bias correction legacy.
  • Fix for AutoHenderson for halfyearly data.
  • Add epsilon for standard deviation
  • Fix bug in automatic differencing (diagnostics)
  • Fix bug in Tramo (automatic choice of trading days with holidays)
  • Add new revision policy (current [AO])
  • Fix bugs with User-defined holidays
  • Correction in regression specs

Other

  • Fix parsing of numbers that use non-breaking spaces as grouping separator
  • Fix InvalidPathException
  • Fix drag & drop on Java11
  • Fix rendering on high dpi
  • Fix Windows search scope
  • Add native support of Java 11+ thanks to NetBeans Platform 11
  • Add new open document engine (.ods files)
  • Add generation of checksum on binaries
  • Fix SQL keyword list
  • Fix parsing of some dates in Excel ( .xlsx)
  • Fix native library issues in some restricted environments

Some explanations on the new revision policy: current[AO]

The new revision policy fixes the model (including all the parameters) and handle any new observation as an additive outlier.
Those “pseudo outliers” are in fact introduced in the model as intervention variables (to avoid possible confusion).

It is easy to show that, in the case of model-based decompositions (canonical decomposition, structural model…), this solution is exactly the same as using forecasts (some small discrepancies can appear, due for instance to the bias correction in multiplicative models).
In X13, that approach is not exactly the same as the use of forecasts, due to the fact that the projected seasonal factors are not necessary coherent with the ARIMA forecasts. In most cases, the differences between the use of “Current[AO]” and “Current[forecasts]” are small.

The main advantages of the “Current[AO]” approach are:

  • No modification of usual routines
  • Usual diagnostics (+ estimation of the size of the AO)
  • Possibility to fine tune the model (removal of some “AO”…)

The generated intervention variables are kept in the model (and perhaps re-estimated) when other revision policies are applied on subsequent periods, except in the case of the concurrent (=complete) policy; in that latter case they are removed from the model. Statisticians must then decide on the best handling of the series (use of automatic outliers detection, of ramps, of specific regression variables…)

Assets 8
Loading