220 likes | 415 Views
Revisions analysis and the role of metadata. OECD Short-Term Economic Statistics Working Party Meeting Paris, 23-24 June 2008 Andreas Lorenz* Deutsche Bundesbank Statistics Department.
E N D
Revisions analysis and the role of metadata OECD Short-Term Economic Statistics Working Party Meeting Paris, 23-24 June 2008 Andreas Lorenz* Deutsche Bundesbank Statistics Department *This presentation reflects the personal views of the author and not necessarily those of the Deutsche Bundesbank or its staff. Revisions and the role of metadata
Outline • Introduction • Revisions analysis without metadata • Revisions analysis with metadata • Conclusion Revisions and the role of metadata
1 Introduction • Uses of revisions analysis • From the producer perspective: Instrument for quality monitoring • From the user perspective: Tool for building expectations about future revisions of preliminary data Revisions and the role of metadata
Case study: German Index of Industrial Production (IIP) • Real time data from 1999 to 2006 • What does revisions analysis tell about the extent of revisions at different time periods? • Can this information be used to build expectations about future revisions of preliminary estimates? Revisions and the role of metadata
2 Revisions analysis without metadata Classical revision measures • Relative Mean Absolute Revision • Mean revision Revisions and the role of metadata
3 Revisions analysis with metadata • Metadata for the IIP for the overall period 1999 – 2006 • Only the largest firms report monthly • Smaller firms report quarterly • Reporting units for the monthly survey are selected once per year ( dying out sample within a year) • Monthly output figures are benchmarked with results from the full quarterly sample (smaller and larger firms) Revisions and the role of metadata
Metadata regarding the timing of IIP releases • Preliminary release (T + 37 days; ~ 10% missing data) • First revision (T + 57 to 62 days; incorporation of late reports) • Quarterly revision (~2 ½ months after end of reporting quarter, sometimes later) • Annual revision (together with quarterly revision of Q4) • New base year and benchmark revisions (about every 5 years) Revisions and the role of metadata
Metadata specific to single years and sub-periods • 1999: High quarterly revisions; later on: the expected quarterly revisions are included in the estimation of the first monthly estimates. This yields in general to smaller quarterly corrections. • 2002: Missing update of the sample for the monthly survey • 2005: New method for the imputation of missing values for the preliminary release (see next slide) Revisions and the role of metadata
Estimation of missing values for the preliminary release • About 10% of the firms do not report in time for their figures to be included in the preliminary release of the monthly IIP • 1999-2004: The output of the previous month is used as an estimate for the missing value of the current month (no-change assumption). Calendar and seasonal effects are not taken into account. • Since 2005: The output of the firms of the corresponding NACE- Division which report in time is used as an estimate for missing values, which yields to smaller revisions. Revisions and the role of metadata
Metadata for 2007 onwards • More firms report monthly • Reporting units for the monthly survey are updated monthly ( no dying out sample within a year) Revisions and the role of metadata
Results • The first revision to the preliminary release does not follow a systematic pattern any more • Since the sample of the monthly survey is now updated concurrently, there is no systematic quarterly correction to be expected • The yearly revision is not predictable Revisions and the role of metadata
Results • What can be learned in light of available metadata for building expectations about future revisions? • Nothing - the producer of official statistics has already “learned” and improved methodology • Revisions should not be systematic any more Revisions and the role of metadata
4 Conclusions • From the producer perspective: Revisions analysis have motivated the improvement of methodology which, in turn, lead to lower revisions • From the user perspective: Revisions analysis helpful for building expectations on future revisions Revisions and the role of metadata
4 Conclusions • Caveat: Without knowledge of the metadata the use of results of revisions analysis for building expectations can be highly misleading • Metadata (sample methodology, estimation methods for missing values – which later may become available, timing of revisions) are of vital importance for building expectations about future revisions Revisions and the role of metadata