What are the typical deviations between preliminary ”real-time” measurements of key UK macroeconomic data and subsequent, more accurate measurements? And how variable
are the deviations and the revisions processes over time? These questions are the focus of
an analysis of real-time macro data in the UK by Anthony Garratt and Shaun Vahey,
published in the February 2006 issue of the Economic Journal.
These are important issues since in order to aid timely economic decisions, real-time forecasters and policy-makers use preliminary data and make assumptions about the relationships between those and subsequent measurements (the revision). In principle, a
better understanding of the revisions processes can increase the quality of real-time
macroeconomic analysis.
The researchers characterise the relationships between preliminary and subsequent measurements for 16 commonly-used UK macroeconomic indicators drawn from two existing real-time data sets and a new nominal variable database.
The variables comprise the expenditure components measure of real GDP (known as GDPE)
in constant prices: private consumption, investment, government consumption, exports, imports, retail sales, unemployment (total claimant count), M0, M4, industrial production
and average earnings, nominal GDP-E, GDP price deflator, M0 velocity and M4 velocity.
The interest in money velocity stems from its pivotal role in the UK”s 1980s monetary
targeting experiments.
The study finds that the preliminary measurements of most real-side macro indicators are
downwards biased predictors of subsequent measurements. Using a test for multiple
”structural breaks” of unknown timing, it finds that breaks affect the relationships between
early and later measurements for many variables.
These breaks often pre-date the much-publicised late 1980s and early 1990s reforms to UK
statistical reporting stemming from the Pickford Report and subsequently two phases of
”Chancellor”s Initiatives”. Previous studies have noted the predictability property for the
expenditure measure of output and its components but none provide a formal analyses of
structural breaks.
The preliminary measurements of UK monetary aggregates are largely unbiased. In
contrast, initial nominal GDP and GDP price deflator measurements typically understate
final measurements. The revisions to these nominal variables rarely exhibit structural
breaks. The untypical behaviour of monetary aggregate revisions reflects the very different
collection processes for these series.
Often macroeconomic models perform better with revised data than with preliminary measurements. Real-time data sets allow researchers to condition their ex post model analyses on the information set actually available to forecasters and policymakers in real time. But if researchers ignore the predictability in initial measurements, real-time model performance can be misjudged.
To illustrate this, the researchers use a ”vector autoregression” (VAR) in UK real output growth and inflation to forecast the (one-step ahead) probability of above-trend growth – sometimes referred to as the likelihood of ”positive momentum”.
Ignoring the predictability in initial measurements understates the event probability
considerably for the evaluation period. The forecasting example indicates the scope for
improvement in real-time model performance, but the variation in revisions predictability
across variables and through time makes generalisation perilous.
”UK Real-time Macro Data Characteristics” by Anthony Garratt and
Shaun Vahey is published in the February 2006 issue of the Economic Journal.
Anthony Garratt
Birkbeck College | 020 7631 6410 | a.garratt@bbk.ac.uk
Shaun Vahey
Reserve Bank of New Zealand
