SAEG, 15 March 2018 Item 2.2 Development: progress and next steps. Contribution of the NBB jean.palate@nbb.be
Outline JD+ 2.2.1 JD+ 3.0 Final remarks NBB contribution R interface Milestones (initial planning / current status) Architecture, design, Main challenges API, tests, technical documentation Final remarks
1.1 JD+ 2.2.1 Algorithms GUI Others No impact on the results Additional features Tramo-Seats: extension of the forecasts horizon X11: pseudo-additive decomposition (no pre-processing) 𝑌=𝑇× 𝑆+𝐼−1 , 𝑆𝐴=𝑇×𝐼 GUI Zoom of HTML output Others Cruncher (bugs) XML, SDMX: security issues
1.2 JD+ and R New complete R package for Tramo-Seats and X13 Developed in the context of the Schuman project (ECB – NCBs) Schedule: beta version in June 2018.
2.1 JD+ 3.0: Broad outline Complete refactoring of the code → Maintenance, reusability … Larger scope High-frequency time series Statistical domain: Modelling: improved REGARIMA, STS… Decomposition: new filters… Benchmarking, temporal disaggregation, calendarization Rapid estimates: nowcasting
2.1 JD+ 3.0: Broad outline (cont.) External use WEB services R (actual packages) Excel Documentation Statistical methods API Schedule End of 2019.
2.2 JD+ 3.0: Milestones Actions Initial schedule (3/2017) Status Architecture 2017/Q1-Q2 OK (see below) Normalized design patterns Development of the basic modules (including the new generic time series) 2017/Q2-Q3 OK WEB Service (ESSNET) 2017/Q2 JD+ 2.2 R interface (Schuman project ?) 2017/Q3-Q4 In progress 2018/Q1-Q2 Refactoring of the existing core modules (algorithms) In progress (90%) Refactoring of the existing graphical modules 2017/Q4-2018/Q1 In progress (75%)
2.2 JD+ 3.0: Milestones (cont.) Initial schedule (3/2017) Status Common REGARIMA (tests) 2016/Q4-2017/Q3 In progress (50%) 2018/Q4 Common REGARIMA (implementation) 2017/Q4-2018/Q1 Integrated two-steps SA 2018/Q2 SEATS OK, X14 (?) Benchmarking, temporal disaggregation (JEcotrim) 2018/Q1-Q3 In progress (60%) Benchmarking OK Nowcasting… 2018/Q3-Q4
Benchmarking, temp. disagg. 2.3 JD+ 3.0: Architecture Skeleton of the main modules CORE Modelling Advanced Seasonal adjustment Seasonal adjustment Advanced Modelling Benchmarking, temp. disagg. High-frequency DFM From 1 main algorithmic module to ±15 modules
2.3 JD+ 3.0: Design Adoption of some common design patterns: Immutable objects Builder design pattern … Improved readability, maintenance
2.4 JD+ 3.0: Main challenges Compatibility with the current implementation Systematic comparison of the results New API Separate namespaces (→ old API still valid) Bridge between the two APIs External tests Systematic use of (temporary) R interfaces Technical/statistical documentation Public API should be fully documented Detailed statistical documentation of all algorithms (specific WEB site)
Example of R interface See https://github.com/nbbrd/jd3-rtests
Example of technical documentation
3. Final remarks Huge work; end 2019 is a sustainable planning only for the main algorithms/documentation Impact of the guidelines on benchmarking / temporal disaggregation still to be analysed. Balance between graphical interfaces / R interfaces to be discussed. Long term maintenance of the tool (mainly algorithmic part) is still an open issue.