As part of its commitment to produce trusted statistics to a high quality, ONS continuously reviews and improves data sources methods and systems. This article highlights the outcomes of some of the work that has been undertaken to improve the methods used to compile Output in the Construction Industry statistics.
As part of its commitment to produce trusted statistics to a high quality, ONS continuously reviews and improves data sources methods and systems. This article highlights the outcomes of some of the work that has been undertaken to improve the methods used to compile Output in the Construction Industry statistics, previously announced in the Continuous Improvement of Gross Domestic Product: Sources, Methods and Communication – 24 April 2013; in particular the introduction: of a monthly seasonally adjusted series; chained volume measures (CVMs); and referencing and rebasing to 2010=100.
The new and improved construction methods (monthly seasonal adjustment, CVMs, referencing and rebasing) will result in revisions to the estimates of construction outputs, as well as to Quarterly National Accounts and Blue Book 2013. Users are reminded that these changes are only one source of revisions to Blue Book 2013; other revisions will be introduced for example, from artistic originals, own account software and industry reviews as detailed in the Blue Book 2013 scope article (62.7 Kb Pdf) .
The data in this executive summary includes the latest information available for both GDP and Output in the construction industry, including Q1 2013.
The time series used in the detailed analysis in annexe A only includes data up to 2012 Q4 and thus does not contain the latest estimate of GDP Q1 2013 nor the latest estimates of Output in the Construction Industry. The detailed analysis could not be extended to Q1 2013 due to this analysis being conducted before these data were available.
Since 2010, ONS has published Output in the Construction Industry statistics, the release includes non-seasonally adjusted monthly data and seasonally adjusted quarterly data.
In January 2013, three years of monthly data were available and in line with Eurostat guidance on seasonal adjustment, a seasonal adjustment review of the monthly data was carried out. The review found that although a relatively ‘young’ series it was possible to produce seasonally adjusted estimates of monthly data. Three possible methods were reviewed, seasonally adjusting:
The monthly series starting in January 2000 and calculating the seasonally adjusted quarterly series as the sum of the monthly series. A proxy monthly series was calculated to cover the period from January 2000 to December 2009.
The monthly series starting in January 2010 and calculating the seasonally adjusted quarterly series as the sum of the monthly series.
Further details on these methods and the analysis undertaken to determine the most suitable method of seasonal adjustment for these data can be found in Annexe A.
In summary, the analysis showed method 1 as the preferred method (seasonally adjusting the monthly series, from 2000, and summing the monthly series to the quarters), on the basis of the monthly series being ‘young’ and as it matures future revisions are minimised; and that using the proxy series for January 2000 to December 2009 increases the statistical robustness of the seasonal adjustment. However, it should be noted that volatility in the seasonal factors will remain until the series matures i.e. when there is generally 5 to 7 years of data.
Construction output volume indices are currently based on fixed prices for a base period of 2005. Overtime the fixed base period requires updating and this is usually done on a five yearly basis. An improved method, which also brings consistency with international good practice and with National Accounts, would be annual rebasing; this is generally referred to as annual chain-linking.
The June 2013 release of Quarterly National Accounts consistent with Blue Book 2013 will introduce construction CVM’s with the corresponding weights updated from 2005 to 2010. Users can expect no change to published growth rates but will see a change in the published level of these estimates.
The first monthly seasonally adjusted series and CVM’s will be published in the Quarterly National Accounts, consistent with Blue Book 2013, on the 27 June 2013. Each publication of Output in the Construction Industry from this date will include the seasonally adjusted monthly series and chained volume measures.
Seasonally adjusting the monthly construction output estimates will result in revisions to the seasonally adjusted quarterly estimates. The implied revisions to the seasonally adjusted quarterly volume indices currently published in Q1 2013 preliminary estimate of GDP and the new seasonally adjusted series are shown in table 1 below. It must be noted that further revisions may come from the introduction of chained volume measures and revisions since the preliminary estimate of GDP.
Period | Published quarterly growth rate for output in the construction industry Q1 2013 (1) | New seasonally adjusted quarterly growth rate (2) | Implied revisions from monthly seasonal adjustment (p.p.) (2-1) |
---|---|---|---|
2010 Q1 | 2.3 | 3.7 | 1.4 |
2010 Q2 | 6.3 | 5.8 | -0.5 |
2010 Q3 | 2.9 | 1.7 | -1.2 |
2010 Q4 | -1.8 | -2.2 | -0.4 |
2011 Q1 | 0.0 | 1.5 | 1.5 |
2011 Q2 | 1.3 | 1.2 | -0.1 |
2011 Q3 | 0.1 | -1.0 | -1.1 |
2011 Q4 | -0.8 | -0.6 | 0.2 |
2012 Q1 | -5.0 | -4.4 | 0.6 |
2012 Q2 | -2.7 | -2.9 | -0.2 |
2012 Q3 | -2.1 | -2.6 | -0.5 |
2012 Q4 | 0.6 | 0.8 | 0.2 |
2013 Q1 | -2.4 | -2.3 | 0.1 |
As suggested in the May 2013, Economic Review, the underlying trend in GDP in recent years is best described as one that is flat or at best gently rising, and the revisions to GDP from the introduction of the monthly seasonally adjusted volume of output in the construction industry does not change this trend as shown in table 2 below.
Users are reminded that these changes are only one source of revision to the Quarterly National Accounts or Blue Book 2013, other revisions will be introduced as detailed in the
Blue Book 2013 scope article (62.7 Kb Pdf)
, and from other improvements to the methodology used for Output in the Construction Industry and new data which may change the seasonal adjustment further.
The construction industry contributes approximately 6.8% to total GDP when measured from the output approach, using this weight it is possible to estimate the implied revision to GDP (Note: this does not take into account other Blue Book 2013 revisions) as a result of seasonal adjusting the monthly series. These implied revisions to GDP from the seasonal adjustment alone are shown in table 2.
Period | Published quarterly GDP growth rates (p.p.) | Revision to GDP from the change in the seasonal adjustment (p.p.) | Revised quarterly GDP growth rates (p.p.) |
---|---|---|---|
2010 Q1 | 0.6 | 0.1 | 0.7 |
2010 Q2 | 0.7 | 0.0 | 0.7 |
2010 Q3 | 0.6 | -0.1 | 0.5 |
2010 Q4 | -0.4 | 0.0 | -0.4 |
2011 Q1 | 0.5 | 0.1 | 0.6 |
2011 Q2 | 0.1 | 0.0 | 0.1 |
2011 Q3 | 0.6 | -0.1 | 0.5 |
2011 Q4 | -0.1 | 0.0 | -0.1 |
2012 Q1 | -0.1 | 0.0 | -0.1 |
2012 Q2 | -0.4 | 0.0 | -0.4 |
2012 Q3 | 0.9 | 0.0 | 0.9 |
2012 Q4 | -0.3 | 0.0 | -0.3 |
2013 Q1 | 0.3 | 0.0 | 0.3 |
The Quarterly National Accounts, published on 27 June 2013, will be consistent with Blue Book 2013, and include for the first time monthly seasonally adjusted construction estimates. This note explains how and why the seasonal adjustment of construction statistics is changing.
The time series in this analysis uses data up to 2012 Q4 and thus does not contain the latest estimate of GDP Q1 2013 nor does it contain the latest estimates of Output in the Construction Industry. The detailed analysis could not be extended to Q1 2013 due to this analysis being conducted before these data were available.
Data in the executive summary shows the latest picture for GDP i.e. including the preliminary estimate of GDP for 2013 Q1 and the latest estimates of Output in the Construction Industry. The most notable difference is the impact of the new seasonal adjustment on Q1 2012.
Since 2010, the Office for National Statistics (ONS) has carried out a monthly survey of 8,000 construction businesses, collecting data on their construction activity; previously this survey was conducted on a quarterly basis. As the monthly series is relatively new it was not possible to produce seasonally adjusted monthly construction estimates, quarterly seasonally adjusted estimates continued to be produced.
In January 2013, three years of monthly data were available and in line with Eurostat guidance on seasonal adjustment, a seasonal adjustment review of the monthly data was carried out as described below.
On the basis of the seasonal adjustment review, ONS will continue to provide users with quarterly estimates but these will be calculated as the sum of three monthly, seasonally adjusted estimates. This new and improved method will result in revisions to the estimates of construction outputs, as well as Quarterly National Accounts and Blue Book 2013.
ONS examined a number of different options for producing both monthly and quarterly seasonally adjusted data and found that although a relatively ‘young’ series it was possible to produce seasonally adjusted estimates of monthly data.
The chosen method is expected to result in smaller ongoing revisions to the statistics than the alternative methods. Revisions occur when new data are gathered each month and the seasonal factors recalculated; they are a necessary part of the process of producing good quality statistics.
The different options were compared by testing how the size and direction of revisions were affected, and by checking how much volatility was introduced into the seasonally adjusted estimates. The options that were examined were, seasonally adjusting:
The monthly series starting in Jan. 2000, then calculate the seasonally adjusted quarterly series as the sum of the monthly series. A proxy monthly series was calculated to cover the period from Jan. 2000 to Dec. 2009.
The monthly series starting in Jan. 2010, then calculate the seasonally adjusted quarterly series as the sum of the monthly series.
The monthly series and quarterly series separately, then adjust the monthly series so that they add to the quarterly series.
A crucial part of the work was to ensure that there is consistency between the monthly and quarterly statistics, so that the sum of the three monthly estimates for each quarter is equal to the estimate for the quarter. The seasonal adjustment process does not, on its own, guarantee this consistency, but the three composite options do guarantee consistency.
ONS has been collecting construction information monthly since Jan 2010, so the time series are just over three years long. This is a challenge for the seasonal adjustment process because it works by detecting moving patterns in the series over many years. The first method in the list (seasonally adjusting the monthly series, from 2000, and summing the monthly series to the quarters) is ONS’s chosen approach. This method seasonally adjusts the monthly series over thirteen years. The first ten years are proxy monthly values calculated by interpolating the older quarterly series; the monthly series starting in Jan 2010 is then appended to this proxy series. This approach uses some of the information in the longer quarterly series to improve the seasonally adjusted estimates.
Options 1 and 2 will tend to result in smaller revisions than option 3, when new data are added to the series. However, option 3 preserves the existing quarterly series while the first two options introduce a one off revision to the quarterly series. The quarterly series is used in the calculation of the output measure of Gross Domestic Product, GDP(O). The impact of one-off revisions to the quarterly path of GDP is explored in more detail later on in this note. Note that the analysis conducted in this annexe does not include the latest GDP data released in the preliminary estimate of GDP for Q1 2013.
The following charts illustrate the impact of the three methods. The first chart (figure 1) shows the non-seasonally adjusted monthly series, which ONS started to compile in Jan 2010. The series is obviously seasonal, with a trough in construction activity in winter, a rebound in early spring, and a plateau in summer and autumn.
The remaining charts in this note concentrate on the quarterly path calculated using the three alternative methods for seasonal adjustment. Future analysis and subsequent outputs from ONS will provide greater detail about both the monthly and quarterly paths.
The average difference between the current seasonally adjusted quarterly series and those calculated by aggregating a seasonally adjusted monthly series are shown in table 1, in percentage points. The average differences were calculated over the 12 quarters referring to 2010 Q1 to 2012 Q4, and are root mean square values. The values represent the one-off revision to the quarterly series that would be expected if it was decided to calculate it by aggregating the seasonally adjusted monthly series. The two alternative methods of producing the monthly series: by incorporating the quarterly series prior to 2010 (long) and not incorporating the quarterly series (short) are shown.
The following points are important:
The revisions are about the same size as typical quarterly movements in these series (2 to 3 p.p.)
Neither the short or long method for seasonally adjusting the monthly series is uniformly better
Description | Short series | Long series |
---|---|---|
Other new work, ex. Infrastructure, public. | 1.33 | 1.59 |
Other new work, ex. Infrastructure, private commercial | 3.22 | 2.15 |
Other new work, ex. Infrastructure, private industrial | 2.23 | 2.99 |
Other new work, infrastructure | 1.54 | 1.41 |
New housing, public | 2.68 | 0.56 |
New housing, private | 1.43 | 1.15 |
Housing repair and maintenance, public | 1.09 | 2.46 |
Housing repair and maintenance, private | 0.5 | 1.1 |
However, when the individual series are aggregated to the top level construction series it is clear that substituting the short monthly series accumulated to quarters for the current quarterly series results in larger revisions (figures 2, 3 and 4).
From the chart it is clear that the short monthly and long monthly series would both give revisions to the quarterly construction series.
The chart shows the short monthly series results in larger revisions to quarterly construction than the long monthly. Next we track these revisions through to the implied revision to GDP(O). These revisions have been calculated using Construction’s Gross Value Added (GVA) weight of 6.8%.
Adopting the short monthly seasonal adjustment method results in a change to GDP growth to 1 d.p. in all but three of the quarters shown. However in both the long and short methods there are notable upward revisions to GDP in 2011 Q1 and 2012 Q1 and a notable downwards revision in 2012 Q2.
These revisions have been carried through to GDP(O) to determine the effect on this series.
ONS also compared the three methods in terms of ongoing revisions. That is, revisions that can be expected to occur as new data are added to the series. This was achieved by simulating an extra years worth of data, and calculating revisions over that extra year. Several different scenarios were simulated and the results were broadly the same across all the different scenarios.
There was not much difference between method 1 (long monthly) and 2 (short monthly) in terms of mean revisions, which are a measure of bias; but method 3 was worse. In terms of average absolute revision, which is a measure of the average size of revision, method 1 was better in the majority of cases, followed by method 2.
The monthly seasonally adjusted series that will be implemented into the Quarterly National Accounts on June 27 2013 is shown in figure 6, below and is compared to the published non-seasonally adjusted monthly series.
Details of the policy governing the release of new data are available by visiting www.statisticsauthority.gov.uk/assessment/code-of-practice/index.html or from the Media Relations Office email: media.relations@ons.gsi.gov.uk
These National Statistics are produced to high professional standards and released according to the arrangements approved by the UK Statistics Authority.