Statistical Office of the European Communities / Directorate General Statistics
163 April 2007
TF-SAQNA-17
Task Force on Seasonal Adjustment
of Quarterly National Accounts
Second Meeting
Agenda Item 4
______
Case Study on Differences in Seasonal and Calendar Pattern
First Results
1.Background
The mandate of the Task Force on Seasonal Adjustment of Quarterly National Accounts (TFSA QNA) requires to “examine further, for selected components of ESA Table 1, the underlying causes of the significant differences in the seasonal and working day correction factors currently used”. For this purpose, a questionnaire was sent to all members of the Task Force on 15 March 2007. It was dividedintothreecountry groups in order to broaden the scope of the comparison, and to limit the response burden. As of 13 April replies have been received from eleven Task Force members. Replies from seven countries are pending (BE, GR, FR, MT, SE, FI, UK). The analysis of the answers presented below is therefore provisional and will be completed. It is planned that a summary of the main results will be included to the final report or the Task Force.
This first review of the results follows the structure of the questionnaire. The answers received were supplemented with further data from ECB/Eurostat databases. The focus is on the determinants of the quarterly profile of national accounts, common factors and major differences. The note is divided in three main sections: Section 2 on Value Added in NACE C-E, G-I and J-K; Section 3 on Household and Government Consumption; Section 4 on Value Added in NACE F and Gross Fixed Capital Formation.
For all seasonal factors quoted in this note see the tables in the annexed questionnaire which was sent to Task Force members in March 2007.
For the meeting of the Task Force it is planned to present an overview of the following detailed comparisons. All Task Force members are invited to comment in the meeting on the results, and add experiences from their own countries.
Summary of general results and issues
Despite all country-specific issues there are strong common determinants of the seasonal components across countries (e.g. variations of energy use, tourism season, weather conditions, salary bonuses) which are quoted by all countries.
Differences in QNA seasonal patterns may also be due to compositional differences of the economy across countries (e.g. the weight of the tourism sector). Different income levels may equally have implications for the distribution across the year (e.g. tourism expenditure may be more equally spread across the year when income is higher).
The monthly and quarterly indicators used for deriving QNA data vary significantly across countries. To some extent this can explain differences in the seasonal patterns of the QNA variables (e.g. when labour input or turnover instead of production indices are used for estimating value added). However, it can also not be excluded that two alternative and equally suitable sources (e.g. retail trade turnover surveys and Household Budget Surveys may show a different seasonal and calendar pattern. Furthermore, the frequency of the indicator sources differs, and some countries rely more on quarterly sources than others. This has implications for calendar adjustment techniques (for which monthly data is the preferred source). Even when monthly source are available, such as monthly production or retail turnover indices, these are not always used for QNA estimates.
Some of the data provided indicate rapid changes and sometimes volatility of the estimated seasonal components. This may be due to changing economic conditions, but may also be due to a different understanding of what the seasonal component should ideally reflect.
The available data show that the magnitude and even the sign of calendar effects may vary across countries and across GDP components. For example, using the variation of the number of working days as a rough indicator of calendar effects, e.g. in private consumption,positive and negative correlations to the calendar factors have been found. Given the weight of private consumption in GDP this is also a main factor for different GDP calendar effects.
The survey showed many differences in the practices for calendar adjustment. It appears very likely that this has an impact on the results and implies that current calendar adjustments are not sufficiently comparable across countries (which is important given the forthcoming mandatory provision of calendar adjusted GDP data under the new ESA Transmission Programme). Sometimes the quality of the source data and the available staff and software resources determine the practice. However, there seems to be also some differences in the definition of the calendar component. It was also mentioned that the in some cases calendar effects are decreasing in recent years.
2.Value added
Q1: What are the main monthly or quarterly statistical sources and indicators that are used for derivingthe QNA variable?
CZ / EE / HU / SE / SK / FINACE C-E
Enterprise surveys (q) for output and intermediate consumption / Production indices (m); Industry turnover / Production at current prices deflated / Surveys (q) and administrative data (m/q)
NACE G-I
See before / Retail index (m); car registrations (m) ;nights spent * average costs;
Turnover and transport statistics / See before / See before
NACE J-K
See before / Profit and loss account for NACE J (q) ; Production; dwelling new orders / Profit and loss account for NACE J (q) ;
Production at current prices deflated / See before
Findings and open issues
Many sources are at quarterly frequency only and monthly industrial production indices for NACE C-E are not always the preferred source (which may have implications for the recommendations on calendar adjustment). Only in some cases variations of intermediate consumption are explicitly taken into account in the QNA series, while other indicators (industrial production) usually imply the assumption of constant intermediate consumption shares within a give year.
Q2&Q3:What are the main factorsdetermining the seasonal pattern of the QNA variable?Please describe possible reasonsthat may help to explain the most significant differences between the seasonal pattern of QNA data for your country and the European aggregates
CZ / EE / HU / SE / SK / FINACE C-E
Determined by the sources;
Peak in Q2 due to lower intermediate consumption to output ratio;
Trough in Q3 due to NACE E / Main impact from strong peaks of NACE E in Q4/Q1 / Trough in Q3 due to summer vacation
Q4 peak due to Christmas / Main impact comes from NACE E in Q1 (heating season) and reparations (=high intermediate consumption) in Q3)
NACE G-I
Peak in Q 2 due to lower intermediate consumption to output ratio, and vice versa in Q4 / Q4 Christmas consumption;
Q3 tourism expenditure / Mainly determined by retail sales before Christmas, counter reaction in Q1;
Lower income implies high spending around Christmas, and higher savings in Q1;
Lower income implies only one holiday season per year in summer / Special seasonal factors due to NACE G due to [fiscal changes?]
NACE J-K
Trough in Q3 due to high ratio of intermediate consumption / Trough in J in Q4 due to high costs / Similar pattern as for G-I due to similar factors, mainly in NACE 74 (other business activities) / Mainly due to NACE J peak in Q1 (compensation) and trough in Q4 (insurance payments)
Findings and open issues
The seasonal pattern of NACE E has a strong impact on the seasonal pattern of value added in industry (see following table). High seasonal factors in Q4 and Q1 as well as low seasonal factors in Q2 and Q3 influence the aggregate despite the fact that NACE E contributes only about 1/6 to total industry (see second table).
Industrial Production NACE E(energy) - seasonal factors(period averages, 1998-2006)*
CZ / EE / HU / SK / SE / FI
Quarter 1 / 113.3 / 121.8 / 122.3 / 113.3 / 114.1 / 116.3
Quarter 2 / 93.2 / 83.0 / 86.3 / 90.0 / 92.2 / 92.0
Quarter 3 / 89.3 / 81.6 / 79.3 / 85.6 / 83.9 / 84.1
Quarter 4 / 103.9 / 113.8 / 112.2 / 111.1 / 109.2 / 109.9
Source: Eurostat, ECB calculations using monthly data. Data refer to NACE section E except
for Slovakia and Sweden (MIG Energy); * Sweden and Finland starting 1995q1
Value added by NACE sections
(current prices, percentages of whole economy)
C / D / E / F / G / H / I / J / K
CZ / 1.5 / 25.9 / 4.2 / 6.6 / 12.4 / 2.0 / 10.3 / 3.3 / 13.8
EE / 1.0 / 16.8 / 3.4 / 7.3 / 15.2 / 1.6 / 12.1 / 3.8 / 18.8
HU / 0.2 / 22.2 / 2.9 / 4.8 / 10.9 / 1.6 / 7.6 / 4.6 / 17.5
SK / 0.6 / 23.4 / 4.9 / 6.8 / 14.4 / 1.3 / 10.4 / 4.2 / 14.3
SE / 0.3 / 19.7 / 3.0 / 4.5 / 10.7 / 1.5 / 8.0 / 4.5 / 19.8
FI / 0.3 / 23.1 / 2.1 / 5.9 / 10.7 / 1.5 / 10.4 / 2.3 / 18.5
Source: Eurostat, ECB calculations. Data refer to 2005 except Sweden (2004)
Several countries mention the impact of varying intermediate consumption shares on the seasonal pattern of value added and seemto have available the required infra-annual information. The CzechRepublicprovided quarterly data for output, intermediate consumption and value added. According to this, these ratios may vary 3-4 points within the year, and for all three branches the ratio is the highest in the fourth quarter. It would be interesting to know more about the treatment of intermediate consumption and its effect for other countries.
Czech Republic - Ratios intermediate consumption to gross output
NACE C,D,E / NACE G, H, I / NACE J,KAverages 2002-2006 / Q1 / 0.723 / 0.570 / 0.644
Q2 / 0.728 / 0.577 / 0.651
Q3 / 0.734 / 0.579 / 0.655
Q4 / 0.754 / 0.608 / 0.675
Source: CZ
Furthermore, the data provided by CZ shows a very rapid change in the seasonal pattern of value added components (therefore our approach to compare average seasonal factors over 10 years may be somewhat misleading).
NACE G-I has a strong seasonal pattern, which is due in particular to NACE G&H. In almost all countries, these two sections have a clear trough in the first quarter (after-Christmas decline of retail turnover, little tourism activities)
For explaining seasonal differences in NACE G and H,Hungary points to the effect of different income levels across the EU. When income is low, most exceptional purchases e.g. for consumer durables are done before Christmas; furthermore tourism expenditure is concentrated in the summer months, since consumers cannot afford two vacations in a year.
From the 3 branches compared, the seasonal variation in NACE J-K is smallest due to a small seasonal variation in NACE K (which includes real estate). However, there are likely to be relatively significant differences in the seasonal component of NACE J. The comparison of EE and SK confirms this.
Slovakia and Estonia - Average seasonal factors of value added (NACE sections)
C / D / E / G / H / I / J / KSlovakia (average 1995-2006)
Q1 / 98.15 / 97.95 / 142.72 / 85.70 / 85.64 / 100.01 / 130.82 / 99.89
Q2 / 101.02 / 103.79 / 90.82 / 124.06 / 111.63 / 99.99 / 94.01 / 100.02
Q3 / 100.02 / 100.10 / 78.19 / 116.29 / 108.97 / 100.02 / 97.79 / 100.02
Q4 / 109.28 / 98.10 / 90.63 / 74.08 / 93.31 / 99.91 / 73.61 / 100.09
Estonia (average 2000-2006)
Q1 / 101.2 / 94.0 / 124.7 / 91.7 / 76.7 / 97.1 / 99.6 / 99.8
Q2 / 96.0 / 105.5 / 84.3 / 103.7 / 111.2 / 105.6 / 102.8 / 102.2
Q3 / 106.2 / 99.4 / 75.0 / 101.7 / 119.7 / 103.3 / 103.4 / 96.5
Q4 / 96.4 / 101.0 / 116.0 / 102.5 / 92.1 / 94.1 / 94.5 / 101.4
Source: EE and SK
Accordingly, Slovakia underlines the high seasonal component of Financial intermediation, which explains most of its exceptional seasonal pattern in NACE J-K. The reasons given are compensation paid in the first quarter and insurance claims in the fourth quarter. For NACE I and K the data for Slovakiado not show a significant seasonal component, while the seasonal pattern of Estonian value added is significant. Information that may explain this would be welcome.
Overall, while there are plausible explanations for the seasonal differences across countries, in some cases the difference might also reflect differences of chosen source data, rather than differences in value added. This may raise questions as regard the comparability of the sources and indicators used for compiling unadjusted quarterly value added.
Q4 Calendar adjustment is carried out for themonthly or quarterly source data, or the quarterly QNA data, which calendarregressors are used (e.g. Mo-Fr, Saturdays etc.…)?
CZ / EE / HU / SE / SK / FINACE C-E
1 regressor, no leap year correction, CZ calendar / No calendar adjustment yet / 1 country specific working day regressor / Between 0-7 regressors in DEMETRA
NACE G-I
1 regressor, no leap year correction, CZ calendar / No calendar adjustment yet / Only for retail trade: 1 country specific working day regressor / Between 0-7 regressors in DEMETRA
NACE J-K
1 regressor, no leap year correction, no adjustment for NACE J, CZ calendar / No calendar adjustment yet / No significant calendar effects / Between 0-7 regressors in DEMETRA
Findings and open issues
Considerable differences in the practices for calendar adjustment exist. Estoniadoes not yet perform calendar adjustment; the CzechRepublic performs calendar adjustment for all except NACE J, using one working day regressor which considers the country-specific public holidays.Hungary has tested for working day effects, however, only in NACE C-E they proved to be significant. A country-specific calendar is used.Slovakiauses the default options available in DEMETRA. It seems that none of the countries uses calendar factors coming from monthly indicator time series.
Q5&Q6. Please provide us with information that roughly quantifies the calendar effect (e.g. “…one additional working day in the quarter is estimated to increase the series level by X % …”). Please explain any special features in your country thatmay be relevant for explaining the calendar component of the series.
CZ / EE / HU / SE / SK / FINACE C-E
C: 0.03%
D: 0.22%
E: -0.50%
Country-specific public holidays may explain differences to other countries / No calendar adjustment yet / n.a.
NACE G-I
G: -0.12%
H: -0.54%
I: 0.04% / No calendar adjustment yet / Calendar effects only in retail trade / n.a
NACE J-K
K: -0.22% / No calendar adjustment yet / No significant calendar effects / n.a.
The data for the CzechRepublic underline the role of calendar adjustment for QNA data, including for services output. The effects are most significant in manufacturing, energy; hotels and restaurants and business services. The negative signs of the coefficients for NACE E and K are somewhat counterintuitive.
Both for CZ and HU calendar factors have been provided. Calendar effects are common in industry and retail trade; however, for NACE H and K there is a significant calendar component in data for CZ, but no calendar adjustment in Hungary.
For CZ, the calendar factors are shown in the table below. The long-term averages of these factors differ from a value of 100, which suggests that the calendar component contains as seasonal element (e.g. in NACE C-E all calendar factors for the first quarter in the years 1996-2006 are > 100). This feature could also be observed in some other countries and should be clarified by the Task Force.
Czech Republic, Calendar factors, averages 1996-2006
NACE C,D,E / NACE G, H, I / NACE J,KQ1 / 100.42 / 99.64 / 99.29
Q2 / 100.08 / 99.94 / 99.97
Q3 / 100.38 / 99.76 / 99.61
Q4 / 99.63 / 100.30 / 100.66
Source: CZ
3.Household and government final consumption
Q1: What are the main monthly or quarterly statistical sources and indicators that are used for derivingthe QNA variable?
BE / DE / ES / FR / IT / NL / GBPrivate Consumption
Monthly retail, wholesale, hotels and restaurants turnover
Q-crafts reports
Car registration
Energy reports
Tobacco tax reports / Enterprise tax data
Monthly food consumption
Retail sales
Energy reports
Passenger transport
Other services indicators / Q-Household Budget Survey (HBS) is a main source for almost all components.
For food, tobacco and alcohol also retail trade
Turnover statistics for transport, communication
Car registrations
Overnight stays in hotels / M retail trade and commercial and personal services.
Energy reports; consumption by non-residents and by resident households abroad.
Consumption of other services, e.g. medical services and welfare, transport , communication .
Statistics on housing and population.
Turnover of supermarkets and sales of vehicles.
50% for services consumption, a part of it is extrapolated
Government Consumption
Q reports of government and social security statistics / National Audit Office provides all data / Q data from government on cash and accrual basis, refers to compensation, social transfers and intermediate consumption / Q-labour input, compensation and social benefits in kind
Findings and open issues
For private consumption each country uses a variety of monthly and quarterly sources. The role of monthly retail turnover appears to differ (prominent source in DE and NL) while other countries seem to predominantly use other sources which are often only at quarterly frequency (Household Budget survey and administrative/tax data). It would be interesting to know whether these sources could show different seasonal or calendar influences.
For government consumption the main source are government reports, social security data and labour input measures. Sources are generally at quarterly frequency
Q2&Q3: What are the main factorsdetermining the seasonal pattern of the QNA variable? Please describe possible reasonsthat may help to explain the most significant differences between the seasonal pattern of QNA data for your country and the European aggregates.
BE / DE / ES / FR / IT / NL / GBPrivate Consumption
Retail trade Christmas and post-Christmas effect
Pattern is close to EU pattern / Twoextra wages in June and Dec have important effect, e.g. on vehicle purchase retail trade and restaurants; Christmas and post Christmas effect;
High consumption in restaurants&bars in third quarter compensated by low food consumption in the same quarter;
School year starts in Sep (expenses in 4th Q);
Spaniards spend holidays in Spain. / Peak/trough in Q3/Q4 due to hotels/tourism;
Clothing peaks in Q3/Q1 due to seasonal sales
Higher share for hotels and restaurants may explain differences to other countries / At aggregate level the seasonal pattern of the components (e.g. energy) largely cancels out
Government Consumption
Employee bonus in Q4;
Vacation bonus in Q3;
Funds spent at end of budget year;
Anticipated purchases due to changes in next year;
All effects of decreasing magnitude / Twoextra wages in June and Dec have important effect / 13th month salary in December;
Other components less affected;
No extra salary in June / Holiday allowance in May, end-of year bonus;
End-of year expenditures
Findings and open issues
For private consumption most replies mention three common factors: extra salaries and bonuses, Christmas turnover as well as tourism/climatic conditions.
Extra (e.g. 13/14th salary) are important for general retail turnover, and the purchase of consumer durables (e.g. cars). In Spain, two extra salaries (June/December) are paid. In Germany, at least one increased salary used to be paid in November/December, but the additional component has been kept constant in absolute terms or even reduced in recent years in many branches.
The trough in the first quarter in most countries is explained by the after-Christmas decline of retail trade, and weak tourism activity in this quarter. There may be also special tax or administrative effects which tend to increase private consumption at the end of the year (e.g. VAT increases in January), and cause a decline in the first quarter.
The strong seasonal profile of retail trade consumption is also confirmed by the seasonal factors of national retail trade turnover as shown in the following table. All countries have seasonal troughs in Q1 and pronounced seasonal peaks in Q4. An important difference concerns however the change of retail trade turnover from the third to the fourth quarter: while in BE, DE, ES, FR and NL the turnover in the fourth quarter is about 10 percent above the turnover of the third quarter, Italian and UK turnover data are 26% and 17% above the third quarter turnover.