Transcript OECD Short –Term Economic Statistics Working Party Revision and

Processes and Policy for Revising
Monthly Production Statistics
(GDP) at Statistics Canada
Prepared by: Michel Girard & Erika Young
Presented by: Michel Girard, Statistics Canada
OECD Short-Term Economic Statistics Working Party
25 – 27 June 2007
Paris, France
Overview
 Revision policies

Source data and monthly GDP
 Integrated revision process


Frequency
Reasons
 Computer system
 Clients
Revision policies
Source data
 Raw and seasonally adjusted data are normally revised over
a 3-month period
 Respondent errors/availability of data
 Annual revision process for monthly data




Coding on central registry
Processing
Accumulation of micro data corrections
Data corrections in other programs (exports/imports, prices,
administrative data)
 Annual surveys: previous year is normally subject to revision
 Discrepancy between monthly and annual estimates becoming a
concern
 Historical process for monthly data
 X12, NAICS, Profiling
Revision policies
Source data
Issues
 Carrying back seasonal adjustments further
back in time
 Micro vs macro adjustments
 Reconciling/benchmarking monthly to annual
results


Will require modifying processes and systems
Calendarization
 Aligning source data and SNA revision policies
Revision policies
Source data
 Project on revision policy
 Description of policies by surveys
 Looking at revision policies in other NSO
 Extent to which policies fits the needs of
clients including SNA
 Recommendations
 Notably reconciliation and benchmarking issues
Revisions to Monthly GDP
 According to frequency
 Monthly, quarterly, annual, historical
 Classification of revisions
 Source data
 Seasonality (including trading-day factors)
 Reconciliation
 Changes to methodologies
 Benchmarking including rebasing
 Classifications
 SNA conceptual revisions
Process of revision
Monthly
Quarterly Annual
Historical
Jan to Dec
Feb, May,
Aug, Nov
September
10 years
In September,
revisions are
carried back 5
years
Tend to go
back to 1961
In August,
revision back
to January of
previous year
Source data
Seasonal
Reconciling
Weights-prices SNA 1993
with quarterly Trading-day
Classifications
GDP
Methodologies
Rebasing
Incorporating Benchmarking
quarterly
Central frame
source data
Computer System
 Menu driven, accessible by many users
 Charts/Reports
 Methodologies compiled for specific period of time
 Database can be decomposed into 4 categories:




Current version
Revised version
Published version
Test/historical version
 Feeder system vs published results
 Avoiding building seasonality
Revisions and dissemination
 Qualitative information
 Source data
 Explained in the context of IO and IE Accounts
 Methodology
 Quantitative information
 Always: growth rates and levels
 Sometime: mean and standard deviations
 Occasionally: bias, dispersion, concordance of
movements
Clients

Survey in 2004
 Satisfaction vis-à-vis aggregates – 77%
 Concern about detail – 33%
 +/- 0.2%: good trade-off on average for aggregates
 Accuracy more important than timeliness – Data released 60
days after reference month
 Appreciate qualitative information, especially when special events
take place
 Demand to incorporate revisions sooner than later – for
modelling and forecasting purposes
 Necessity to revise back when story do not change being
questioned