Revisions of National Accounts

Jore, Anne Sofie Research Report Revisions of National Accounts Staff Memo, No. 6/2017 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Jore, Anne Sofie (2017) : Revisions of National Accounts, Staff Memo, No. 6/2017, ISBN 978-82-7553-989-0, Norges Bank, Oslo, https://hdl.handle.net/11250/2506375 This Version is available at: https://hdl.handle.net/10419/210340 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/deed.no Revisions of national accounts STAFF MEMO NO. 6 | 2017 ANNE SOFIE JORE Staff Memos present reports and documentation written by staff members and affiliates of Norges Bank, the central bank of Norway. Views and conclusions expressed in Staff Memos should not be taken to represent the views of Norges Bank. © 2017 Norges Bank The text may be quoted or referred to, provided that due acknowledgement is given to source. Staff Memo inneholder utredninger og dokumentasjon skrevet av Norges Banks ansatte og andre forfattere tilknyttet Norges Bank. Synspunkter og konklusjoner i arbeidene er ikke nødvendigvis representative for Norges Banks. © 2017 Norges Bank Det kan siteres fra eller henvises til dette arbeid, gitt at forfatter og Norges Bank oppgis som kilde. ISSN 1504-2596 (online only) ISBN 978-82-7553-989-0 (online only) Revisions of national accounts Anne Sofie Jore June 8, 2017 Abstract I investigate revisions of growth rates in nominal and real quarterly GDP for mainland Norway, as well as for the GDP deflator, from 2004 to 2016. Several measures from alternative revision periods are computed. Mean revisions of real GDP are small and close to zero, while nominal GDP and the deflator are under-predicted on average when they are first published. For all three variables, mean absolute revisions are quite substantial. Revisions can, in general, be explained by the arrival of new information, although revisions of the deflator in particular are also characterized as reducing noise. 1 1 Introduction Statistics Norway publish national accounts data for each quarter around 50 days after the end of the quarter. When figures for a new quarter are published, previously published figures for earlier quarters are revised. Data for all quarters of year (t−2) are final in August of year (t), when year (t−2) becomes the new base year. The base year is moved forward every year (with some exceptions). Hence, figures for the first quarter of any given year are final after 10 quarters, while figures for the fourth quarter of any year are final after 7 quarters. In this analysis, I regard the figures for the 11th releases as final.1 Monetary policy is conducted in real time. It is therefore important to know to what extent early releases of economic data are likely to be revised between the first and the final release and whether the revisions can be predicted. Revisions of growth rates in nominal and real quarterly GDP for mainland Norway, as well as for the GDP deflator, are investigated here. Mean revisions of seasonally adjusted real growth in GDP are very close to zero, hence there is no tendency for the first released data to either over- or under-predict the final data. First releases of nominal GDP and the deflator somewhat under-predict final figures. In absolute terms, revisions are quite substantial for both real and nominal GDP as well as for the deflator. Revisions are at least as large from the 5th to the 11th release as from the first to the 5th release. When revisions are scaled by the size of the actual growth rates, I find the largest revisions in the deflator, particularly in the four-quarter growth rates. For some measures, results differ depending on whether the data are seasonally adjusted or unadjusted. This is most notable for mean revisions. Seasonal adjustment appears to reduce bias in the first releases for quarterly as well as for four-quarterly growth in real GDP. Even if revisions are sizable, early estimates could still be efficient estimates of the 1Prior to 2016, data were final and the new base year was changed each November. 2 “true” or final data. This will be the case if revisions are characterized as containing news. If, on the other hand, revisions mainly reduce noise in earlier estimates, early information can be used to forecast the revisions and hence improve the estimate of the “true” data. It turns out that revisions of quarterly, seasonally adjusted growth rates from the first release to final data are mainly characterized as containing news, even if some preliminary revisions also reduce noise. This means that the first released growth rates of real and nominal GDP and the deflator can be interpreted as efficient forecasts of the final data. The results depend on whether the data are seasonally adjusted. Revisions of unadjusted growth rates reduce noise to a greater extent than is the case for seasonally adjusted data. In section 2, I first discuss some aspects of the revision process. In section 3, several measures are constructed to illustrate the size and nature of the revisions: mean revision (MR), mean absolute revision (MAR), relative mean absolute revision (RMAR), root mean square revision (RMSR), noise-to-signal ratio (N/S) and sign revision (SR). The measures are computed for alternative revision periods. In section 4, the “news-vs-noise” hypotheses are tested formally. Finally, I summarize the results in section 5. 2 Revisions of Quarterly National Accounts Quarterly national accounts (QNA) series are published as unadjusted data by Statistics Norway (SN). Seasonally adjusted data are also published in order to facilitate analysis of business cycle developments. Until November 2011, data from the start of QNA compilation (first quarter of 1978) to the last published quarter were seasonally adjusted at a disaggregate level (the indirect method), and the adjusted data were then added up to obtain the main aggregates. For the base year, which is changed every year, and the following years, this method ensured consistency between the main aggregate and its disaggregate components. For the period prior to the base year, however, chainlinked disaggregated series were no longer additive. Hence, an increasing wedge between the levels of unadjusted and seasonally adjusted data appeared moving back in time, 3 see figure 1. Also, substantial revisions to quarterly seasonally adjusted data regularly occurred many years after the unadjusted data were understood to be final. 100 150 200 250 300 350 400 450 500 01.03.1978 01.01.1979 01.11.1979 01.09.1980 01.07.1981 01.05.1982 01.03.1983 01.01.1984 01.11.1984 01.09.1985 01.07.1986 01.05.1987 01.03.1988 01.01.1989 01.11.1989 01.09.1990 01.07.1991 01.05.1992 01.03.1993 01.01.1994 01.11.1994 01.09.1995 01.07.1996 01.05.1997 01.03.1998 01.01.1999 01.11.1999 01.09.2000 01.07.2001 01.05.2002 01.03.2003 01.01.2004 01.11.2004 01.09.2005 01.07.2006 01.05.2007 01.03.2008 01.01.2009 01.11.2009 01.09.2010 Real GDP mainland Norway, published August 2011 Seasonally adjusted GDP Unadjusted GDP Figure 1. Seasonally adjusted and unadjusted real mainland GDP There were two main problems associated with the resulting seasonally adjusted data. Econometric models based on seasonally adjusted data would rely on levels that deviated from the levels of the unadjusted data, hence average historical growth rates would be systematically different. Furthermore, turning points of historical economic cycles would move around. In November 2011, SN changed their seasonal adjustment method for obtaining main aggregates by using seasonally adjusted main aggregates (the direct method) for data prior to the base year. The published seasonally adjusted series are now a combination of the direct method, for the years prior to the year before the base-year, and the indirect method, for the year before the base-year and the following years. When the base year is changed, historical seasonal factors are kept unchanged. Revisions in seasonally adjusted 4 final data are thus solely due to revisions in unadjusted data.2 Revisions of unadjusted data are of the following types: •Information-based changes throughout the year •Base year changes every year •Main revisions –Main revisions in the investigated period (2004 - 2016): ∗2006: Changes in the treatment of financial intermediation services ∗2011: New grouping of industries (SN2007). ∗2014: New treatment of research and development expenses (2008SNA and ESA2010). When quarterly national accounts for the second quarter of year t are published in August, figures for the year t-2 become final, and the year t-2 is the new base year. Prior to 2016 this revision occurred in November, when national accounts for the third quarter were published. Hence, for most of the period, data for the first quarter of year t-2 will be final with the 11th release, while data for the fourth quarter will be final with the 8th release. To avoid revisions occurring from 11 quarters onwards, which could be substantial for seasonally adjusted data for most of the period analyzed here, I concentrate on revisions up to 11 quarters after the initial release and regard these releases as approximate final data. The main revisions in 2006, 2011 and 2014 are not treated separately. As a check of the results, part of the analysis was redone, with the vintage prior to the main revision in 2014 as the last vintage. According to Statistics Norway, the 2014 revision increased the level of GDP by 1.5 to 2.1 percent. This exercise did not change any results. See the appendix for details. The revision in 2011 resulted in only minor level changes. Figure 2shows revisions to quarterly, seasonally adjusted GDP (mainland Norway) for 2003, from the vintage released in June 2003 to the vintage released in February 2017. 2This is explained in detail on Statistics Norway’s website. 5 295000 300000 305000 310000 315000 320000 325000 330000 335000 Nominal GDP Mainland Norway 2003 Q1 Q2 Q3 Q4 (a) Nominal level -2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 Nominal GDP Mainland Norway 2003 quarterly growth Q1 Q2 Q3 Q4 (b) Nominal growth -2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 Real GDP Mainland Norway 2003 quarterly growth Q1 Q2 Q3 Q4 (c) Real growth -2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 GDP deflator Mainland Norway 2003 quarterly growth Q1 Q2 Q3 Q4 (d) Deflator growth Figure 2. Revisions of seasonally adjusted mainland GDP, 2003 The upper part of the panel illustrates revisions to nominal GDP levels and quarterly growth. We see shifts in nominal levels connected to the main revisions in 2006 and 2014. The change in the method of seasonal adjustment from November 2011 is likewise clearly evident. Measured as quarterly growth rates, we find the largest revision of the growth rate for the first quarter of 2003, which starts at -1.50 percent and is currently a good 1.50 percent - a change of 3 percentage points. On the other hand, growth in the second quarter starts at 2.3 percent and is currently 0.5 percent. Upwards and 6 YLFN YFN YFNP −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 Mean Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 Mean Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP −0.2 −0.1 0 0.1 0.2 0.3 0.4 Mean Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP −0.2 −0.1 0 0.1 0.2 0.3 0.4 Mean Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 3. Mean revisions (MR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent 13 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Mean Absolute Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Mean Absolute Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Mean Absolute Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Mean Absolute Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 4. Mean absolute revisions (MAR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent 14 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 5. Relative mean absolute revisions (RMAR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Proportion 15 YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 RMSR QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 RMSR QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 RMSR YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 RMSR YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 6. Root mean square revisions (RMSR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent 16 YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios QoQ−Growth Rates, (SA) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios QoQ−Growth Rates, (U) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios YoY−Growth Rates, (SA) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios YoY−Growth Rates, (U) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 7. Noise-to-signal ratios (NS). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent 17 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs QoQ−Growth Rates, (SA) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs QoQ−Growth Rates, (U) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs YoY−Growth Rates, (SA) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs YoY−Growth Rates, (U) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 8. Sign ratios (SR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent 18 4 News and noise The concept of “news versus noise” analysis was originally introduced by Mankiw et al. (1984), with an application to revisions of the money stock. See also Croushore and Stark (2003) for an application to revisions of national accounts data. When a revision to an economic variable is characterized by “noise”, the new release of the data eliminates (or reduces) noise in earlier releases. At the other end of the spectrum, revisions contain “news” if the new release incorporates information that was not available earlier. In this case, the early release is an efficient estimate of later releases. The distinction has important consequences for modeling the revision process. If revisions reduce noise in earlier estimates of the “true” or final data, there is a measurement error in early estimates that is reduced over time. Revisions are correlated with earlier releases and uncorrelated with the revised data. Revisions are then (partially) predictable, and an improved estimate of the final value can be obtained from the history of observed revisions. If, on the other hand, revisions can be characterized as containing news, the revisions are correlated with the revised data and uncorrelated with earlier releases of the data. Hence, revisions cannot be predicted, and the preliminary release can then be interpreted as an efficient estimate of the later releases. These two polar opposites of the revision process can be investigated by running OLS regressions of the revisions on alternative releases of the data. To test the hypothesis that revisions from the p-th published value to the s-th published value are news, and therefore unpredictable, the following regression is run for four alternative revision periods: yt+s t−yt+p t=α1+β1yt+p t+εt.(1) s>p, s = 2, 5, 11, p = 1, 5 and εtis a white noise residual. Again, yτ tdenotes the release of the value for the quarterly variable ytobtained at time τ. The first release of national accounts data is published in the quarter after the actual value, yt+1 t. Unpredictability of revisions, or the news hypothesis, is equivalent to the hypothesis H01= (α1, β1) = (0,0), i.e. revisions are uncorrelated with earlier releases. The Ftest 19 statistic Tˆ β0(ˆ Vˆ β)−1ˆ β 2,(2) is assumed to have a F(2, T −2) distribution, where ˆ Vˆ βis a heteroscedasticity and autocorrelation consistent estimate of the asymptotic variance of √T(ˆ β−β) (Newey and West,1987). If we cannot reject the hypothesis that revisions are uncorrelated with the earlier release, the “news” hypothesis is supported. Similarly, the hypothesis that revisions from the p-th published value to the s-th published value reduce noise, and are therefore partly predictable, is equivalent to stating that the revision is uncorrelated with the later releases. The following regression is run for four alternative revision periods: yt+s t−yt+p t=α2+β2yt+s t+εt,(3) s>p, s = 2, 5, 11, p = 1, 5 and εtis a white noise residual. Predictability of revisions, or the noise-reducing hypothesis, is equivalent to the hypothesis H02= (α2, β2) = (0,0). If we cannot reject the hypothesis that revisions are uncorrelated with the later release, the “noise” hypothesis is supported. If the hypothesis H02is rejected while H01could not be rejected, this will be a strong indication that the revisions are characterized by news and therefore cannot be predicted. The opposite, a rejection of H01combined with an acceptance of H02, is a strong indication that revisions reduce noise and are therefore partly predictable. If both hypotheses are rejected or both hypotheses are accepted, there are no clear indications of either news or noise. The distinction “news versus noise” has implications for the standard deviation of yτ t. If revisions contain news, the standard deviation should be non-decreasing from earlier to later releases. Efficient forecasts incorporate all new information as it becomes available, hence there is a tendency for increased volatility. Standard deviations of releases that reduce noise in earlier releases should, on the other hand, decline. Before presenting results from the formal tests, we will inspect the standard deviations in tables 1(quarterly rates) and 2(four-quarter growth rates). 20 Table 1. Quarterly growth Standard deviations of seasonally adjusted releases 1 2 3 4 5 6 7 8 9 10 11 YLFN 0.92 1.02 0.955 0.995 1.02 0.874 0.885 0.95 1.06 1.21 1.35 YFN 0.484 0.594 0.594 0.622 0.644 0.648 0.581 0.626 0.7 0.762 0.75 YFNP 0.871 0.862 0.86 0.826 0.821 0.707 0.741 0.792 0.864 0.945 1.07 Standard deviations of unadjusted releases YLFN 4.18 4.02 4.02 4.01 4.04 4.08 4.07 4.16 4.21 4.2 4.28 YFN 4.33 4.27 4.25 4.24 4.27 4.34 4.33 4.32 4.35 4.36 4.35 YFNP 0.923 0.952 0.926 0.831 0.803 0.81 0.838 0.837 0.84 0.922 0.925 QoQ growth rates To simplify, I abstract from the intermediary releases and compare standard deviations for three periods: •Initial to fifth release - early period •Fifth to eleventh release - late period •Initial to eleventh release - whole period Standard deviations increase in both the early and the late period for seasonally adjusted nominal and real quarterly growth in GDP, see the upper part of table 1. These results favor the news hypothesis overall. New releases of quarterly seasonally adjusted growth rates of the deflator reduce noise in the early period, but contain news in the late period and over the complete revision cycle. For unadjusted quarterly growth rates in the lower part of the table, new releases tend to reduce noise in the early period, but contain news over the complete revision cycle. The results are similar for all variables. For four-quarter growth rates of nominal and real GDP in table 2, seasonally adjusted and unadjusted, standard deviations increase in the early as well as in the late period, indicating that revisions contain news over both periods. Again, the deflator stands out: Standard deviations decrease in the early period and increase in the second period, 21 Table 2. Four-quarterly growth Standard deviations of seasonally adjusted releases 1 2 3 4 5 6 7 8 9 10 11 YLFN 1.76 1.96 1.97 1.94 1.96 1.94 2.11 2.26 2.45 2.35 2.38 YFN 1.88 1.97 1.99 1.97 1.98 1.99 2.09 2.2 2.22 2.17 2.08 YFNP 1.23 1.13 1.08 1.04 1.02 0.965 1.04 1.17 1.24 1.18 1.15 Standard deviations of unadjusted releases YLFN 2.36 2.38 2.49 2.45 2.47 2.52 2.7 2.86 2.93 2.94 2.84 YFN 2.44 2.45 2.55 2.54 2.52 2.56 2.68 2.8 2.81 2.74 2.65 YFNP 1.15 1.09 1.01 0.994 0.946 0.905 0.899 1.04 1.11 1.08 1.01 YoY growth rates for unadjusted as well as for seasonally adjusted numbers. In contrast to the results for quarterly rates, revisions of the deflator over the whole period tend to reduce noise. Most of the results from these simple standard deviations of alternative releases are in line with the noise-to-signal ratios in figure 7. The formal testing of the “news versus noise” hypothesis is reported in tables 3to 5, one table for each of the three variables. Quarterly and four-quarterly growth rates of seasonally adjusted and unadjusted data are tested. The left-hand side of the table (columns two to five) reports the results for four-quarterly growth rates, and the righthand side of the table (columns six to nine) reports the results for quarterly growth rates. P-values from regressions (1) and (2) for revisions from the initial and from the fifth release are reported. For each revision the two hypotheses H01and H02are tested. I investigate three revisions from the initial release; from the first to the second release, from the first to the fifth release and from the first to the eleventh release. From the fifth release there will be just one revision, from the fifth to the eleventh release. P-values from testing the news hypothesis H01(that revisions are uncorrelated with earlier releases) are reported in the first column and in the last line (except the last entry in the line), i.e. 22 6 Appendix 6.1 Revisions of seasonally adjusted mainland GDP, 2004 295000 305000 315000 325000 335000 345000 355000 365000 375000 Nominal GDP mainland Norway 2004 Q1 Q2 Q3 Q4 (a) Nominal level -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Nominal GDP mainland Norway 2004 quarterly growth Q1 Q2 Q3 Q4 (b) Nominal growth -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Real GDP mainland Norway 2004 quarterly growth Q1 Q2 Q3 Q4 (c) Real growth -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 GDP deflator mainland Norway 2004 quarterly growth Q1 Q2 Q3 Q4 (d) Deflator growth Figure 9. Revisions of seasonally adjusted mainland GDP, 2004 The 4 panels in figure 9illustrate how revisions affect nominal levels and growth rates for the quarters of 2004. The most notable difference compared with revisions of 2003 numbers in figure 2is the impact of the main revision in 2014 on real growth rates. For 2003, real growth rates did not change with this revision. 29 6.2 Measures of revisions for sample ending in 2014 The measures illustrated in the appendix are calculated on a sample where the vintage published in August 2014, prior to the main revision in 2014, is the last one included. To facilitate comparison with results from the full sample, the same scales have been used as in the figures in the text. The main conclusions from comparing the “short” and the “long” samples are: •Mean revisions are larger in the short sample for seasonally adjusted series and approximately the same for unadjusted series. •Mean absolute revisions are larger for the short sample. •Relative mean absolute revisions are of the same size in the two samples. •Root mean square revisions are slightly larger in the short sample. •Noise-to-signal ratios are of the same magnitude in both samples. 30 YLFN YFN YFNP −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 Mean Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 Mean Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP −0.2 −0.1 0 0.1 0.2 0.3 0.4 Mean Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP −0.2 −0.1 0 0.1 0.2 0.3 0.4 Mean Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 10. Mean revisions (MR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent. Sample ends in 2014 31 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Mean Absolute Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Mean Absolute Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Mean Absolute Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Mean Absolute Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 11. Mean absolute revisions (MAR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent. Sample ends in 2014 32 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Relative Mean Absolute Revision YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 12. Relative mean absolute revisions (RMAR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Proportion. Sample ends in 2014 33 YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 RMSR QoQ−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 RMSR QoQ−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 RMSR YoY−Growth Rates, (SA) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 RMSR YoY−Growth Rates, (U) yt+1 t −> yt+11 t yt+1 t −> yt+2 t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 13. Root mean square revisions (RMSR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent. Sample ends in 2014 34 YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios QoQ−Growth Rates, (SA) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios QoQ−Growth Rates, (U) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios YoY−Growth Rates, (SA) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.2 0.4 0.6 0.8 1 Noise−to−Signal Ratios YoY−Growth Rates, (U) yt+1 t −> yfinal t yt+2 t −> yfinal t yt+2 t −> yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 14. Noise-to-signal ratios (NS). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent. Sample ends in 2014 35 YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs QoQ−Growth Rates, (SA) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (a) Quarterly growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs QoQ−Growth Rates, (U) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (b) Quarterly growth, unadjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs YoY−Growth Rates, (SA) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (c) Four-quarter growth, seasonally adjusted YLFN YFN YFNP 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Revision of Signs YoY−Growth Rates, (U) yt+1 t , yt+11 t yt+1 t , yt+2 t yt+2 t , yt+5 t yt+1 t −> yt+5 t yt+5 t −> yt+11 t (d) Four-quarter growth, unadjusted Figure 15. Sign ratios (SR). GDP mainland Norway. Nominal (YLFN), real (YFN) and deflator (YFNP). Percent. Sample ends in 2014 36