1. Successful economic policy relies on good data, and on the whole, the United Kingdom’s data are of high quality. However, sizeable revisions to national accounts data have sometimes occasioned a significant re-assessment of the cyclical position, and thus of the stance of policy. Notably, recent revisions to figures for 1994 and 1995 increased the latest estimate of GDP by almost 1 percent relative to the initial estimate; revisions ran up to 2 percent during the boom of the 1980s. Figure 1 suggests that revisions to GDP data have been biased upward and have had a marked cyclical pattern—i.e., they have been larger in upswings.

Abstract

1. Successful economic policy relies on good data, and on the whole, the United Kingdom’s data are of high quality. However, sizeable revisions to national accounts data have sometimes occasioned a significant re-assessment of the cyclical position, and thus of the stance of policy. Notably, recent revisions to figures for 1994 and 1995 increased the latest estimate of GDP by almost 1 percent relative to the initial estimate; revisions ran up to 2 percent during the boom of the 1980s. Figure 1 suggests that revisions to GDP data have been biased upward and have had a marked cyclical pattern—i.e., they have been larger in upswings.

I. National Accounts Revisions and the Economic Cycle1

A. Introduction

1. Successful economic policy relies on good data, and on the whole, the United Kingdom’s data are of high quality. However, sizeable revisions to national accounts data have sometimes occasioned a significant re-assessment of the cyclical position, and thus of the stance of policy. Notably, recent revisions to figures for 1994 and 1995 increased the latest estimate of GDP by almost 1 percent relative to the initial estimate; revisions ran up to 2 percent during the boom of the 1980s. Figure 1 suggests that revisions to GDP data have been biased upward and have had a marked cyclical pattern—i.e., they have been larger in upswings.

FIGURE 1.
FIGURE 1.

UNITED KINGDOM: REVISIONS IN GDP DATA

(In percent of GDP)

Citation: IMF Staff Country Reports 1998, 004; 10.5089/9781451814071.002.A001

Sources: Office for National Statistics; and staff estimates.

2. To measure GDP, the Office for National Statistics (ONS) uses each of three basic approaches, based on output, expenditure, and income. In theory, these measures should be equal, but in practice, measurement difficulties generate discrepancies that could persist even after the information on the different measures is finalized. Such discrepancies result in relatively large balancing items, and in revisions up to two or three years after the initial estimates have been reported. Measurement uncertainties have been compounded by the increasing importance of the service sector—now constituting % of GDP—where measurement issues are harder to tackle.

3. At the same time, the ONS has introduced several methodological improvements. One would expect the revisions to have become smaller and less systematic as these improvements take effect—as suggested by a casual inspection of Figure 1. Here, the key issue is whether these improvements have outpaced the changes in the economy that make measurement more difficult.

4. This chapter uses a simple regression model to examine these issues more systematically. It concludes that upward revisions to GDP data are positively correlated with economic activity, in particular, with growth in its domestic component, and that although revisions may have become smaller in recent years (controlling for the stage of the cycle), the procyclical bias in the data revision has not been eliminated. The regression model employed also suggests that GDP growth in 1996—currently estimated at 2.4 percent—could be as much as 0.6 percentage point higher than that estimate.

B. Revisions to Data and Economic Activity

5. An investigation by the ONS identifies biases in the first estimates of GDP and suggests that these may have a systematic tendency to be larger in the expansion phase of the cycle.2 This is especially true for the expenditure measure, where a major focus of systematic downward bias is in the initial estimates of gross domestic fixed capital formation. This bias reflects procyclical increases in the number of new businesses, which are not included in the sample used for initial estimates, and in the number of outliers during an upswing. According to this study, among other components of alternative measures of GDP, revisions in imports (on the expenditure side), company profits (on the income side), and manufacturing production (on the output side) have also been relatively large.

6. This section uses a simple regression model to examine the relationship between revisions to GDP figures and the stage of the cycle. Revision to GDP, denoted by r, and defined as percent difference between the initial and final estimates of GDP, is regressed against GDP growth, g and the output gap (actual minus potential output) in percent of potential output, gap.3 The data are annual and cover the more recent 1985-95 period, which constitutes roughly one complete cycle.4 Estimating a linear relationship yields:

r^=0.45+0.11gap+0.18g(1.57)(1.75)(2.15)(1)

Period 1985-95 R˜2 = 0.49 D.W. = 1.46 s.e. = 0.590

χ2 test for serial correlation: χ2(1) = 0.85 [0.36]

χ2 test for functional form: χ2(1) = 5.88 [0.01]

χ2 test for normality: χ2(2) = 1.06 [0.59]

χ2 test for heterscedasticity: χ2(1) = 0.48 [0.49]

where numbers in parentheses are t-ratios, and the numbers in square brackets are acceptance probabilities for the diagnostic tests.5 The estimation results suggest that the output gap and GDP growth significantly (at 10 percent) and positively influence revisions to GDP data. A 1 percentage point increase in the output gap is associated with an upward revision in GDP of about 0.1 percentage point, while a one percentage point increase in GDP growth implies an upward revision of 0.2 percentage point. The estimated equation passes standard diagnostic tests except that for functional form.6 Actual and fitted values are plotted in the middle panel of Figure 1.

7. It can be argued that data revision is influenced differently by the different components of GDP. To test this, GDP growth is separated into its domestic component, c/, and foreign component,/ A regression of data revisions on these variables and the output gap yields estimated coefficients that are generally insignificant; growth in domestic demand is marginally significant at 10 percent critical value:

r^=0.50+0.07gap+0.15d-0.09f(1.80)(1.05)(1.75)(-0.41)(2)

Period 1985-95 R˜2 = 0.54 D.W. = 1.54 s.e. = 0.562

χ2 test for serial correlation: χ2(1) = 2.81 [0.09]

χ2 test for functional form: χ2(1)= 1.89 [0.17]

χ2 test for normality: χ2(2) = 0.81 [0.66]

χ2 test for heterscedasticity: χ2(1) = 0.00 [0.99]

when f is dropped from the regression (results not reported here), d becomes significant at 5 percent. These results provide support for the hypothesis that the bias in the initial estimates is mainly associated with growth in domestic activity.7

C. Improvements in Data Quality

8. In recent years, a number of steps have been taken to improve the quality of GDP data and to reduce discrepancies. Since 1991, rebalancing techniques, using improved and annually updated input-output tables, have been used to make the final estimates of the various measures consistent and to obtain a single measure of GDP.8 The balancing methodology attempts to reconcile estimates of industry value added between the income-based and output-based approaches, reconcile supply and demand for each product, and ensure consistency of GDP deflators with the current and constant price estimates of GDP and its components.

9. Moreover, new statutory inquires and measurement procedures have been introduced, aimed at enhancing the quality and frequency of data collection, notably in the areas of company data and service sector output. In 1990, existing voluntary inquiries into quarterly capital expenditure and stockbuilding, and provision of monthly retail sales figures became compulsory; statutory quarterly inquiries into company profits and annual surveys of financial asset and liabilities of larger companies and share registers were introduced; new statutory quarterly inquiries covering selected parts of the service sector were also introduced; and there were quality improvements in a number of areas, including quarterly family expenditure surveys, and on the external side, in the measurement of trade in services and foreign direct investment. The ONS is considering further actions to improve and expand service sector statistics.9 Measures taken to enhance the quality and coverage of data, together improvements in input-output balancing techniques, are expected to reduce discrepancies between alternative measures of GDP over time.

10. With these improvements in data collection and compilation, revisions should have become smaller in recent years, as also suggested by a casual inspection of Figure 1. A simple test of this hypothesis would be to re-estimate (1) using data for the period prior to the reforms and test the predictive power of the estimated equation for the latter period. Assuming that reforms take time to be effective, the 1994-95 period is picked as the forecast period.10 The equation is estimated using data for 1985 to 1993:

r^=0.41+0.07gap+0.24g(1.44)(1.07)(2.61)(3)

Period 1985-93 R˜2 = 0.58 D.W. = 1.76 s.e. = 0.579

χ2 test for serial correlation: χ2Q) - 0.00 [1.00]

χ2 test for functional form: χ2(1) = 7.06 [0.08]

χ2 test for normality: χ2(2) = 1.16 [0.56]

χ2 test for heterscedasticity: χ2(1) = 1.01 [0.31]

χ2 test for predictive failure: χ2(2) = 2.31 [0.32]

Testing for predictive failure does not suggest a structural break in the equation, despite the fact that the equation overpredicts data revisions for the 1994-95 period by relatively large amounts (Figure 1, bottom panel). Given the simplicity of the model and the small number of observations, the possibility that changing practices may have reduced the magnitude of data revisions cannot be considered as conclusively rejected by this test. However, these results do suggest that, despite the undoubtable methodological improvements in the national accounts that have taken place, the systematic procyclical bias in the data revisions has not been eliminated.

D. Estimated Revisions to 1996 GDP

11. Equation (1) can be used to derive a forecast for revisions to GDP growth in 1996—initially estimated at 2.1 percent and currently at 2.4 percent. To do this, one needs to take account of the fact that equation (1) describes the relationship between the final estimates of the variables involved so that any further revisions would affect the variables on both sides of the equation. Taking this into account, the final estimates of GDP growth g* can be predicted based on the initial estimates of GDP growth and the output gap, as follows:11

g*=0.63+0.155gap+1.25g(4)

Applying this formula to the initial estimate for 1996 GDP growth of 2.1 percent and an associated initial output gap of -1.3 percent, the final estimate of GDP growth for 1996 is predicted to be 3.0 percent. The standard error of this forecast is 0.6, or that the final estimate of GDP growth in 1996 is in the range 2.4-3.6 with 95 percent confidence. The point estimate for 1996 of 3.0 percent is 0.6 percentage point higher than the current estimate of 2.4 percent. This estimate, of course, is subject to the qualifications already noted in the previous section.

E. Concluding Remarks

12. This chapter has discussed measurement problems in GDP and has examined the link between revisions to GDP data and the stage of the cycle. The analysis, which is based on a simple regression equation—and, therefore, is intended to be only suggestive—indicates that upward revision to GDP data is positively correlated with activity, in particular, with growth in domestic demand; that the evidence in favor of a reduction in the size of the revisions in recent years is not conclusive; and that GDP growth in 1996—currently estimated to be 2.4 percent—could be as high as 3 percent. The possibility that the cyclical pattern of data revisions will repeat itself thus still constitutes an important upside risk to projections of economic activity and inflation.

1

Prepared by Hossein Samiei.

2

See U. M. Rizki, “Testing for Bias in Initial Estimates of the Components of GDP,” Economic Trends, No. 514, August 1996.

3

Data sources are the ONS, except for potential output, which is based on staffs calculations; GDP is the average measure as defined by the ONS; and GDP revisions are calculated as the difference between the initial and final estimates of GDP for a particular year in successive issues of Economic Trends.

4

Although using a longer series would be desirable in terms of enhancing the regression’s degrees of freedom, the difficulty of obtaining consistent data on GDP revisions and changes in data quality and definitions over time, argue in favor of focusing on the more recent data.

5

The econometric package used for estimation is Microfit 4.0 by M. H. Pesaran and B. Pesaran, OUP, Oxford: United Kingdom.

6

A simple elaboration of the above specification to allow for non-linearities by including quadratic terms (results not reported) passes the functional from test, but suggests that the square of GDP growth positively affects revisions. This implies that revisions would be large and positive when growth is large in absolute terms, so that recessions are also associated with large revisions. Examining Figure 1 suggests that this could reflect revisions in one year only, namely 1991, and the possibility is not examined further here.

7

Attempting to decompose the effects of the individual components of domestic demand was not successful due to the small number of observations. A more detailed approach, beyond the scope of the present study, would be to examine the link between revisions in the individual components of GDP and economic activity.

8

See Sanjiv Mahajan, “Balancing GDP: U.K. Annual Input Output Balances,” Economic Trends, No. 519, January 1997. •

9

See Bill Cave, “The President’s Task Force on Service Sector Statistics,” Economic Trends, No. 519, January 1997, which describes proposals by the Department of Trade and Industry.

10

Using 1985-91 as the estimation period did not change the results, nor did the more restrictive methodology of including an intercept dummy for the latter period in the full-sample estimation of the equation.

11

Denote by g*, r* and gap*, the final estimates of GDP growth, data revisions, and the output gap for 1996. From (1):

r* = 0.45 + 0.11 gap* + 0.18 g*

Moreover: r* = g* - g, i.e., percent revision in GDP is the difference between current and final estimate of GDP growth. Similarly: gap* = gap + r* « gap + (g*-g). Substituting for r* mdgap* gives:

g* - g = 0.45 + 0.11 [gap + (g* - g)] + 0.18 g*

which yields Equation (4) upon simplification.

United Kingdom: Selected Issues
Author: International Monetary Fund