UK Experience with Revisions Analysis

1
UK Experience with Revisions Analysis

• UK Office for National Statistics
• 7th October 2004

Emily.George_at_ons.gov.uk Fenella.Maitland-Smith_at_ons
.gov.uk
2
Introduction

• GDP revisions analysis published since mid-80s
• Now major expansion of revisions analysis
• Multiple drivers and multiple aims
• Presentation will cover
• why were expanding
• aims
• programme of work
• some results

3
Statistics Commission Report

• In 2003, ONS and other government departments
  made important revisions to major UK economic
  statistics.
• Widely reported and criticised in British media
• Statistics Commission announced a review of
  revisions to economic statistics

4
Recent GDP revisions

• Larger than expected revisions to GDP published
  in September 2003
• Revisions to historical GDP in the 2003 Blue Book
  came from
• introduction of annual chain-linking
• improvements to imports deflators
• rebased GFCF deflators
• revised imports to account for MTIC fraud
• The latest quarters growth (2003q2) also revised
  significantly
• up from 0.3 to 0.6
• revised construction data

5
Recent GDP revisions
6
Statistics Commission Report cont.

• The main conclusions of the Statistics Commission
  were
• For most economic statistics, revisions are the
  norm.
• The UK record on revisions to GDP is as good as
  that in other countries, and in some respects it
  is better
• Some models used in preliminary estimates were
  criticised.
• Communication about the likelihood of revisions
  needs to be improved, particularly to journalists
• (Annex 2 of paper)

7
Aims of Revisions Analysis at ONS

• Identify where any biases might exist
• Understand any patterns in revisions
• Use that information to improve our forecasting
  and balancing processes
• Make the information available to users to allow
  them to use our datasets more effectively
• Provide measures of data quality (reliability).
• And to.
• build confidence of users
• increase transparency
• help users understand that revisions are
  inevitable

8
What are we doing?

• tables of revisions in First Releases
• estimates of bias (mean revisions) in First
  Releases
• more detailed articles in Economic Trends
• sensitivity analyses - sample dependency, lags,
  etc.
• use of bias adjustments in compilation
• databases of historical revisions on website
• web-pages for revisions analysis, policies, etc.
• seminars with users
• more collaboration with Bank of England and
  others
• international comparisons
• collaboration with other NSOs

9
First Releases - tables of revisions
10

First Releases - summary statistics (bias)

• Started February 2004
• 20 National Accounts First Releases
• Annex 3

11

First Releases - summary statistics (bias)
12
Uses of summary statistics in First Releases

• Primarily aimed at users.
• But ONS may use in management tool Red Amber
  Green
• proportion of significant t tests across all
  variables
• RAG charts created from initial data set.
• future revisions classified as Green (within 2
  sd of mean), Amber (2-3 sd of mean) Red (more
  than 3sd from mean)

13
Results to date (31/08/04)

• 14 (6 of 44) of t tests published most recently
  showed a significant bias.
• 2 of the values lay more then one standard
  deviation from the mean.
• none lay more than two standard deviations from
  the mean.
• so all would lie in the 'green' category.

14
Articles in Economic Trends
15
Articles in Economic Trends

• Displaying revisions graphically
• Examine revisions to GDP by stage of production
  using
• stacked bar charts
• t tests
• but different stages used in articles and First
  Releases
• Examine revisions to GDP by components
• production and expenditure
• t tests
• by component by stage of production
• charts of each component
• Starting work on sensitivity to
• sample
• economic cycle

16
Articles in Economic Trends Graphs to show
patterns
GDP quarterly growth, 1993 to 2001Preliminary
and latest estimates,and difference (revision)
17
Articles in Economic Trends Revisions by stage
Revisions to GDP quarterly growth, 1993 to 2001.
18
Articles in Economic Trends Revisions by stage
19
Articles in Economic Trends Revisions by
component
20
Revisions WebPage
http//www.statistics.gov.uk/about_ns/economic_rev
isions.asp
21
Revisions triangles (real-time databases)
22
Revisions triangles (real-time databases)
Integrated spreadsheets on website containing...

• ..triangles of published estimates...

For all major GDP E and O components aim for
all First Release summary statistics to have
supporting triangles
23
Issues for Discussion

• definition of bias
• treatment of revisions due to changes in scope
  and definition
• treatment of levels series
• length of series - sensitivity analyses
• use of bias estimates in compilation
• archiving
• comparison of different methods of analysing
  revisions
• international comparisons of
• reliability
• bias estimates