An Indicator of Monthly GDP and an Early Estimate of Quarterly GDP Growth∗: Discussion Paper 127 (revised)

A range of monthly series are currently available giving indications of short-term movements in output in the United Kingdom. As the only available information, these indicators are routinely exploited in various ways although they only provide an incomplete picture of gross domestic product (GDP). The main aim of this paper is to suggest a formal and coherent procedure for grossing these monthly data up to represent the whole of GDP. Although the resultant estimates of GDP would be worse than those obtained by direct measurement, they should be more satisfactory than simply making an informal inference from whatever monthly data are available. Our examination of the efficacy of the method for estimation of the state of economic activity indicates a rather satisfactory outcome. ∗This paper summarises the results of a project initially undertaken for the Office of National Statistics (ONS) of the United Kingdom, and jointly supported by HM Treasury. A steering committee provided very useful feedback during the early part of the work. Richard Clare, Geoff Reed, Keith Vernon and Colin Yeend were particularly helpful in providing both data and suggestions. We also wish to thank Stephen Hall for helpful comments. The current state of the work described here at which estimates of UK real GDP are produced on a monthly basis was achieved with support from ESRC research grant L116251012. Richard Smith also gratefully acknowledges partial financial support for this and an earlier revision from the ESRC under the project R000237334 “Automatic Leading Indicators” and a 2002 Leverhulme Major Research Fellowship. Any errors remain our own. 1 The Motivation for Monthly Estimates of GDP Macro-economic policy-making in real time faces the perennial problem of uncovering what is actually happening to the economy. Movements of seasonally adjusted real GDP (referred to subsequently simply as GDP) and related estimates of the output gap are widely regarded as important predictors of future inflation and thus are relevant to the problem of inflation targeting. Estimates of GDP are typically produced quarterly with the first estimates in the United Kingdom available about twenty five days after the end of the quarter to which they relate.1 In many countries including the UK monetary policy is set more frequently than quarterly and in order to do so policy makers need not only to anticipate first estimates of GDP growth but also to estimate what is happening within each quarter. A range of monthly series are currently available giving indications of short-term movements in output. As the only available information, they are already exploited in various ways: financial commentators routinely examine monthly data on retail sales, the trade figures, and the output of the production industries in order to assess the state of the economy and likely developments in monetary policy; academic researchers exploiting high frequency econometric techniques make use of one or other of these series as the best available proxy for a broader measure of demand or output. If these monthly data are to be used to draw inferences about the state of the economy as a whole, then it is desirable that there should be some formal procedure for grossing them up to represent the whole of GDP. Such a procedure is likely to produce estimates of GDP which are worse than those which might be produced by direct measurement. On the other hand, it would certainly be more satisfactory than simply making an informal inference from whatever happen to be the latest numbers available. The Office for National Statistics (ONS) in the United Kingdom used to publish lagging, co-incident and leading cyclical indicators for the UK. These indicators were constructed (O’Dea 1975) using a variant of the methods developed by the United States’ Bureau of Economic Analysis and now maintained there by the US Conference Board. In their earlier stages it was unclear what represented the ‘business cycle’ but by the end of their life the indicators were intended to represent cyclical movements in real GDPand the co-incident indicator was constructed by interpolating quarterly GDP. The interpolation method used industrial production and retail sales as indicators but in a manner which was uninformed by statistical theory. The indicator system was abandoned mainly because the forecasting power of the leading indicator was negligible. However, a logical step was to see whether the co-incident indicator could be developed into an indicator of monthly seasonally adjusted GDP in constant prices estimated using clearly specified statistical methods. Longer lags are typical elsewhere although EUROSTAT is aiming for a reduction to forty-five days for members of the European Union.