THE EFFECTS OF REVISIONS TO CENTRAL STATISTICAL OFFICE DATA FOR THE U.K. ECONOMY*
| Author | P. A. ORMEROD |
| Published date | 01 May 1978 |
| DOI | http://doi.org/10.1111/j.1468-0084.1978.mp40002005.x |
| Date | 01 May 1978 |
THE EFFECTS OF REVISIONS TO CENTRAL
STATISTICAL OFFICE DATA FOR THE U.K. ECONOMY*
By P. A. ORMEROD
The aim of this paper is to examine various aspects of revisions to CSO quarterly
and annual data series which are of particular interest to those who construct or
use short-term forecasts. The construction of a forecast involves a blend of the
use of a set of formal equations which summarize the dominant characteristics of
the past behaviour of the economy and a complex set of judgements, contained in
which is the need to formulate a view of where the economy has stood in the recent
past. This paper first reviews the literature on the effects of data revisions on the
estimated parameters of econometric equations which make up the formal model,
and then considers to what extent initial estimates of national accounts series
distort our view of the current position of the economy. Finally, some examples of
the size of revisions to UK data are presented.
The literature on the effects of data revision on estimated parameters seems to
be quite thin, and the results somewhat contradictory. Denton and Kuiper'
considered a simple, six-equation model of the Canadian economy using current
price, first-differenced annual data from 1949 through 1958. They estimated the
model using two estimation techniques with preliminary, mixed and final data,
where preliminary data are the initial estimates of the variables in a particular
year, final data are the latest available estimates, and mixed data are a com-
bination of the two. Their finding was that the variations in estimated para-
meters are greater between different sets of data than between different methods
of estimation. Holden,2 using a four-equation model of the UK estimated by
two-stage least squares and ordinary least squares from 1951 through 1961 using
annual data at 1958 prices, reached the different conclusion that the variations (in
parameters) due to the use of different estimation methods were greater than those
due to data revisions.
Both these studies used a very small number of observations, so the con-
clusions are sensitive to the existence of even a few large revisions to some of the
observations. The same criticism can be levelled at a more recent study by
Denton and Oksaven.3 They fitted a three-equation model to the economies of
twenty-one countries using current price annual UN data, from 1955 through 1964.
Their conclusion was that, although changes in the data did affect the magnitude
of estimated parameters, the latter changed sign very rarely and there was no
evident tendency for the variation in estimates to be in one direction or another.
* The author is grateful to J. Boreham of the CSO and G. F. Ray of NIESR for comments.
Any errors or omissions remain the responsibility of the author.
1Denton, F. T. and Kuiper, J., 'The effect of measurement errors on parameter estimates
and forecasts: a case study based on Canadian preliminary national accounts'. Review of
Economics and Statistics, 47, 1965, pp. 198-206.
2Holden, K., 'The effect of revisions to data on two econometric studies', Manchester
School, 37, 1969, pp. 23-36.
Denton, F. T. and Oksaven, E. H., 'A multi-country analysis of the effect of data revisions
on an econometric model', Journal of the American Statistical Association, 67, 1972, pp. 286-291.
165
To continue reading
Request your trial