Transcript JDemetra+

SASG, 16/10/2012

What's new?
◦
◦
◦
◦
◦

Graphical interface
Statistical issues
I/O features
Technology
Tests
Final remarks
Same main principles as .NET version
• Interactive processing
• Rich results
• ...
Many improvements and goodies
See the presentation

Core engines (Java implementation)
◦ X12/X13:
 Small improvements in comparison with the previous
Java version
◦ Tramo-Seats:
 ≈ future release of Tramo-Seats (FORTRAN):
 New AMI procedure
 New strategy in the case of non decomposable models
 Automatic choice of the "best" TD/WD effects (in progress)
 ! Stochastic TD not integrated !

Other SA methods
◦ Structural models, Generalized airline, Airline + seasonal
noise

Benchmarking
◦ Univariate Cholette's method (see X13)
◦ Extension to the multi-variate case with contemporaneous
and/or temporal constraints

Seasonality tests
◦ See Tramo

Direct/indirect comparison
◦ Use of multi-variate benchmarking if need be.

Calendars
◦ Weights on holidays

Improved access to external data
◦ ODBC, Spreadsheet (Excel or OpenOffice)...

More output (larger set of results)
◦
◦
◦
◦
Csv
Csv matrix
Excel
Txt
JTsToolkit
Core algorithms
External
packages
Peripheral
modules
In house
developments
Pure Java
NetBeans modules
NetBeans
JDemetra+ plugins
Third party
plug-ins

JDemetra+ = NetBeans application

What is NetBeans ?
 IDE for Java developments (and others)
 Framework for extensible applications
 Sponsored by Oracle

Benefits of using NetBeans?
◦ Extensible architecture
 Allows independent development teams
 New features = new plug-ins (no impact on the existing
modules)
◦ Numerous functionalities
 Rich graphical interface
 Management of the plug-ins
 Automatic updates...
◦ (Well) documented framework
◦ Large developers' community
◦ So, JDemetra+ = { plug-ins}






Core SA (TramoSeats, X13...)
Advanced SA
Benchmarking
...
Data providers
...

Profound re-engineering of Demetra+ (.NET)
◦ Better design of the algorithmic modules
 Extensible algorithms
 Faster processing (huge use of multi-processing)
◦ Improved I/O of the high-level components
 Generic xml serialization
 Facilities for providing results to other environments
 Designed for
 (Quasi-)immediate WEB services
 Development of macro-languages (?)
◦ New design of the other modules to fit higher
(NetBeans) modularity

Direct calls to the algorithmic routines
◦ Should be the preferred solution in many cases.
◦ From simple high level modules...
◦ ...to all details
◦ Depends only on 1 library (jtstoolkit.jar)
◦ Documentation based on examples (in progress)
◦ Training in November

Contents of the JTsToolkit API (≈50% code)
◦ Basic mathematical tools
 Matrices, polynomials, filters, optimization, ...
◦ Advanced time series model
◦ Statistical tools
 Descriptive statistics, statistical distributions, ...,
state-space framework

Designed to develop rapidly new features
◦ For examples:




New tests (Canova-Hansen...)
Time-varying TD.
Temporal disaggregation (largely developed)
...


Must be NetBeans modules (-> for advanced
users/developers)
Possible extensions (open list)
◦ Modification of the existing features (graphical
components, menus...)
◦ New data providers, new diagnostics, new output
◦ New seasonal adjustment methods (automatic
integration in batch processing...)
◦ New statistical methods (integration in the main
menus, in the workspace...)
Such extensions will be installed as plug-ins, without
modification of the existing modules.

Needs for test
◦ validation of the statistical methods

Type of algorithms
◦ "Fuzzy" problems:
 Example: seasonal adjustment methods...
 Difficult to develop a true strategy.
◦ " Implicit" problems:
 Example: non linear optimization (ML estimation)...
 Possibility to compare the results. No way to guarantee an
optimal algorithm
◦ "Analytical" problems:
 Example: Estimation of a Reg-Arima model...
 Possibility to validate an algorithm

Main tests developed in JDemetra (type 3)
◦ Likelihood computation of Reg-Arima models
 Comparison of Kalman filter (Tramo), Ljung-Box algorithm( X13)
and others (Ansley...).
◦ Solution of linear models
 Comparison of QR, LU, singular value decomposition...
◦ Canonical decomposition
 Verification through decomposition/aggregation
◦ Many small other tests
 Benchmarking, statistical distributions, auto-correlations, linear
filters, Easter...
 Verification by:
 comparison of alternative algorithms
 logical tests (checking of some constraints...)
 tests by reciprocal transformation
◦ Further tests needed


Huge developments supported by NBB and
Eurostat
NBB will ensure a minimal service of
maintenance:
◦ Correction of bugs
◦ Training
◦ Add-ins following the needs and the agenda of NBB

Necessity to organize the future