Transcript assessing the efficiency of early release estimates of

ASSESSING THE
EFFICIENCY OF EARLY
RELEASE ESTIMATES OF
ECONOMIC
STATISTICS
Charles Aspden
Working Party on National
Accounts, October 2008
ASSESSING THE EFFICIENCY OF EARLY RELEASE
ESTIMATES OF ECONOMIC STATISTICS
• Written by Richard McKenzie, Elena Tosetto
(OECD) and Dennis Fixler (United States
Bureau of Economic Analysis)
• Intended as an article for the OECD’s Statistics
Brief, but presented to WPNA instead.
• An earlier version of the paper was sent to all
heads of national accounts in OECD countries
• An extension of the OECD work on revisions
analysis that makes use of the OECD revision
database.
News v. Noise
• Revisions are a fact of life in the QNA, but
are revisions reflecting new information or
are they just fixing problems in the initial
estimates, such as biases, that should not
have been made in the first place? In other
words are they News or Noise?
News v. Noise
Standard revision analysis is commonly
focused on the measurement of bias of
early estimates and the mean absolute size
and variance of revisions. The revision
analysis described here attempts to go
further by making inferences about how
efficiently the initial estimates are
compiled.
News v. Noise
• Noise: all or part of the revision does not
contain any new information
• News: the revision can be attributed to the
incorporation of new information,
implying the early estimates are efficient
forecasts of the later ones
News v. Noise
• Two methods for determining News or
Noise: correlation method and regression
method.
• Usually come up with similar results
because the regression method is really
just a more sophisticated version then the
correlation method.
Correlation Method
Noise: if there is a significant correlation between
the initial estimate and the size of the revision
Rationale: the initial estimate and the revision
should be statistically independent. Example: a
tendency for large initial quarterly growth rates
to be revised down or falls to be revised up
would be indicative of a systematic error
reflected in a negative correlation between the
revision and the first estimate.
Correlation Method
News: a significant correlation between the
revision and the later estimate
Rationale: the opposite of the rationale for
Noise. If the revision contains News then
a significant correlation is expected with
the later estimate. If there is no News then
no correlation is expected and the
revisions and later estimate would be
independent.
Case study
• Revisions to quarterly GDP growth rates of
OECD member countries
• A ‘first update’ revision, which aims to capture
the revision for each data point in the time
series that occurs between its first published
value and the value published with the
1) first release of the next data point in the time
series
2) the revision between first published data and
that published one year later
Case study - Results
• Results suggest that many countries could
improve their quarterly estimation of
GDP.
• Countries are encouraged to undertake
their own analyses and seek to make
improvements according to their findings.