JDemetra + (1.2.1)
-
Upload
phillip-gilliam -
Category
Documents
-
view
73 -
download
4
description
Transcript of JDemetra + (1.2.1)
![Page 1: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/1.jpg)
JDemetra+ (1.2.1)Luxembourg, 16/4/2013
![Page 2: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/2.jpg)
Data providers IT improvements Methodological improvements
What's new ?
![Page 3: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/3.jpg)
SDMX◦ .STAT (OECD) compatible
Automatic change of frequency ◦ Excel, ODBC...
Optimization◦ Caching...
Plug-ins ◦ Access databases
File-based (Access not needed)◦ Random Arima◦ SAS
Data providers
![Page 4: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/4.jpg)
Correction of bugs, improvements of many features◦ Workspaces (storage...)◦ Graphical components (charts, grids...)◦ Properties Window◦ ...
Calendars and user-defined variables (graphical interface)◦ Demetra+ (not yet in the cruncher)
IT improvements
![Page 5: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/5.jpg)
X11◦ Diagnostics
Calendars◦ Documentation
Arima estimation◦ Stdev of parameters ( + scores)◦ Optimisation procedure
Methodological improvements
![Page 6: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/6.jpg)
Problem:◦ The likelihood function of complex models (AR
and MA parameters) have often several local maxima.
◦ Tramo, X12 and JD+(1.1.0) can lead to different solutions
◦ No "best" solution (with acceptable performances)◦ The solution is more dependant on the starting
point than on the Levenberg-Marquardt variant.
Solution in 1.2.1◦ Several starting points
Optimisation procedure
![Page 7: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/7.jpg)
Comparison between: JD+ / TS JD+ / X12 TS / X12
Model: Arima [+calendar effects]
JD+ better
TSbetter = JD+
betterX12 better = TS
betterX12 better =
(0,1,1)(0,1,1)+TD7 0% 0% 100% 0% 0% 100% 0% 0% 100%
(1,1,1)(1,1,1)+TD7 3% 0% 97% 2% 0% 98% 0% 2% 98%
(2,1,1)(0,1,1)+TD7 4% 1% 95% 7% 0% 93% 4% 1% 95%
(3,1,1)(0,1,1) 18% 2% 80% 11% 1% 88% 6% 11% 83%
(1,1,3)(0,1,1) 19% 1% 80% 14% 1% 85% 5% 11% 83%
[1] "Better" means significantly higher likelihood (and thus different estimates).
![Page 8: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/8.jpg)
Regular polynomials Seasonal polynomial
Log-LikelihoodAuto-regressive Moving
averageMoving average
φ(1) φ(2) φ(3) θ(1) Θ(1)
X12 -0.449 0.124 -0.079 -0.715 -0.733 -706.27
Tramo 1.089 0.424 0.136 0.931 -0.746 -703.31
JD+ 0.761 0.326 -0.031 0.622 -0.779 -702.57
Estimation for a (3 1 1)(0 1 1) model
![Page 9: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/9.jpg)
![Page 10: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/10.jpg)
01/0
1/19
91
01/1
0/19
91
01/0
7/19
92
01/0
4/19
93
01/0
1/19
94
01/1
0/19
94
01/0
7/19
95
01/0
4/19
96
01/0
1/19
97
01/1
0/19
97
01/0
7/19
98
01/0
4/19
99
01/0
1/20
00
01/1
0/20
00
01/0
7/20
01
01/0
4/20
02
01/0
1/20
03
01/1
0/20
03
01/0
7/20
04
01/0
4/20
05
01/0
1/20
06
01/1
0/20
06
01/0
7/20
07
01/0
4/20
08
01/0
1/20
09
01/1
0/20
0980
90
100
110
120
130
140
TramoJD+
Tramo-Seats and JD+. SA series based on the same model (different parameters estimation)
![Page 11: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/11.jpg)
Comparison is not so simple Impact of the estimation problem on the
whole AMI◦ Differencing: (1 x 1)(1 x 1)◦ Arma identification◦ Last resort model (3 1 1)(0 1 1)
Comparability depends on the set of series:◦ Simple models (airline...) -> Highly comparable
results◦ Complex models -> Many different results
Consequences
![Page 12: JDemetra + (1.2.1)](https://reader036.fdocuments.us/reader036/viewer/2022082505/568130b7550346895d96d800/html5/thumbnails/12.jpg)
Tramo-Seats◦ Integration of the last modifications of the core
engine. ?
Next steps