Demetra+ - CROS

advertisement
Demetra+
jean.palate@nbb.be
Quick Tour
• Versatile software. Choose the right tool
• Demetra+ main feature: multi-processing
• Demetra+ in production. Understanding the
re-estimation procedure
• The file system of Demetra+.
• Batch processing by means of WSACruncher
• Automation
1. Choose the right tool
Task
Tools
Comments
Interactive analysis
(experts)
Demetra+
Single documents
Black-box SA
Excel add-ins
For unskilled people.
Not suited for repeating
work
Production
Small sets (< 100)
Demetra+
Excel add-in
Excel add-in = light
Demetra+ plugged in Excel
Production
Large sets
Demetra+ (<1000)
WSACruncher
Demetra+ for interactive
improvements,
WSACruncher for bulk
processing
Production. Advanced
(specific handling)
Programming tools
See special topic
2. Demetra+: using multiprocessing
• Choose a default specification
• RSA4 or RSA5(c): automated procedure
• Exception (pre-definition of the specification)
– Specific calendar (National holidays...)
– Specific needs (no outliers detection...)
• Create a new multi-processing
• Use the local menu in the Workspace Window (Multiprocessing)
• Rename the multi-processing, if need be
• Select the series
• Use drag-drop
• Consider the list container (Tools  Containers  List)
to make easier selections
• [Repeat the selection procedure]
• Possibility to mix the specifications
• Possibility to mix the sources
• Run the processing
• Inspect the results
• The summary and the matrix view give useful general
information (see especially the signs of trading days,
leap year, Easter coefficients, the numbers of outliers).
• Sort the results following the quality
– Quality indicator = just an indicator (to be improved)!!
– Inspect bad series firstly, but don't forget the other series
– Use the priority field to mark important series (sort by
priority)
• Inspect and modify bad series
– Problems in diagnostics:
• Spectral td peaks  calendar variables (remove pretests, modify options)
• Residuals
– Skewness: log/level
– Kurtosis: (difficult) critical value for outliers detection
• Spectral seasonal peaks. Difficult
– Seats:  Arima specification (use auto-modelling or Airline; try
(2 1 0)(0 1 1)...). Box-Jenkins could be inappropriate.
– X11:  filters
• Residual seasonality  Series span (shorter) or
– Seats: Arima specification
– X11:  filters
– Other important outputs (personal preferences)
• Charts (main chart, SI-ratio)
• Arima spectrum (→ simplify if possible)
• Regression model (→ checks the significance / the signs
of the coefficients, consider the number of outliers)
• Last residuals
• ...
• Improve by trial and error (see above)
• Generating the output
– Users should prefer csv files
• More efficient, many outputs
• csv format depends on international settings
(Windows)!
• Immediate (or easy) import in numerous software (or
DB)
– Usual options (default folders, presentation...) can
be saved (see "Tools → Options → Outputs")
– Feedback in the log window (location of the output...)
3. Understanding the re-estimation
• Classification of the parameters of a SA processing
– Reg-Arima modelling
• Defining the "domain" of the model
–
–
–
–
Log/level
Type, time span of outliers
Regression variables (calendars...)
Arima model
• Defining the "selection procedure"
– Fct, VA, diffAIC...
• Operational
– TOL, maxiter
– Decomposition
– Diagnostics, output (not used in Demetra+)
Domain
specification
Refresh
(partial)
Estimation
specification
Refresh
(concurrent)
Process
(II)
Process (I)
Refresh
Point (or result) (current)
specification
Refreshing and multi-processing
• Step 1: Definition of the "domain specification"
– limits of the considered models
• Step 2: First estimation
– Results define the "point specification"
• Step 3: Refreshing: definition of an "estimation
specification"
– between the results and the domain (by relaxing some
constraints)
• Step 4: Next estimation
– Results define a new "point specification"
Refreshing
Relaxed constraints (cumulative)
Current adjustment
None
(estimation spec = point spec)
Current adjustment (partial)
Coefficients of the regression variables
Partial concurrent / Parameters
Parameters of the Arima model
Partial concurrent / Last outliers
Last outliers re-evaluated (1 year)
Partial concurrent / Outliers
All outliers re-evaluated
Partial concurrent / Outliers + Arima
Arima model (orders)
Concurrent adjustment
All
(estimation spec = domain spec)
SA Processing and refreshing. Tips
• Keep the domain specification as large as
possible (for significant refreshing)
• Limitations of Demetra+
– Overview of the different specifications (→next
release)
– Possible problem with pre-specified outliers (→
use intervention variables when needed)
– True current adjustment unavailable (→ JDemetra)
– Limited set of refreshing:
4. File system of Demetra+
• Storage of information in xml files
• Separate file for each document
– Contents of the workspace
– TramoSeats / X12 specification
– TramoSeats / X12 single processing
– Multi-processing
– Calendars
– User's variables
Files of a workspace
Contents of
the
workspace
descriptor
Multi-processing file
(in SAProcessing folder)
• Domain specification(s)
• Items
– Attributes
• reference to domain specification[,estimation policy,
quality]
– Identification
– [Data]
– [Point specification]
– [Estimation specification]
• Remarks:
– No results (except the estimated model)
– All information for re-estimating the processing
• Identification
• Data
• All specifications
– Future improvements of Demetra+ (refreshing):
• Comparison of the raw series
• Comparison of the models
5. WSACruncher
• WSACruncher is a console application for
batch processing
wsacrunche
r.params
Read the workspace
Refresh the processing
Generate the output
csv files...
Save the results
Demetra files
(workspace)
• The xml parameters file defines the main
steps:
– refreshing policy
– output (csv, diagnostics)
• The workspace must contain all the files, using
the same structure as Demetra+
• Syntax of the command:
<path>\wsacruncher <workspace> -x <parameters>
Example: "c:\program files...\wsacruncher"
workspace.xml -x wsacruncher.params
6. Automation
• Best solution for production chains
– making your in-house software
• Providing specific Demetra+/WSACruncher modules for
– Accessing the data
– Providing new diagnostics
– Saving the results
(Not discussed here)
• Using building blocks of Demetra+ in a new application
– interacting with WSACruncher through xml files
Advanced?
NO
YES
IT-Team
?
WSACruncher (params)
YES
.NET Applications (C#)
Demetra+
Visual Studio
(express edition)
NO
Light dev.
?
MS-Office
YES
VBA: Excel add-ins...
Advanced users
NO
Others (SAS...),
through files
Xml files, WSACruncher
Automation: comparison
Solution
Advantages
Drawbacks
WSACruncher
•Simple
•Efficient
Limited possibilities
.NET
•Powerful solution
•Complete access to internal
routines
•For "professionals" only
•Uncertain future (JDemetra ?)
VBA
•Light, accessible for non
experts.
•Efficient
•MS solution ( JDemetra ?)
•No full access to routines of
Demetra+
Xml files
•Independent solution
•Integration in legacy
environments
•Complex
•(Rather) inefficient
•Limited access to routines of
Demetra+
Download