unece workshop on short-term economic statistics (sts) and seasonal adjustment
DESCRIPTION
UNECE Workshop on Short-Term Economic Statistics (STS) and Seasonal Adjustment. Astana, 14 – 17 March 2011. Workshop on Short-Term Statistics (STS) and Seasonal Adjustment Workshop Purpose and Scope. Astana, 14 – 17 March 2011 Petteri Baer, Marketing Manager, Statistics Finland. - PowerPoint PPT PresentationTRANSCRIPT
March 2011 UNECE Statistical Division 1
UNECE Workshop on Short-Term Economic Statistics (STS)
and Seasonal Adjustment
Astana, 14 – 17 March 2011
March 2011 UNECE Statistical Division 2
Workshop on Short-Term Statistics (STS) and Seasonal Adjustment Workshop Purpose and Scope
Astana, 14 – 17 March 2011
Petteri Baer, Marketing Manager, Statistics Finland
March 2011 UNECE Statistical Division 314 - 17 March 2011
Capacity Building Program on Challenges in STS
• The UNECE organizes with the financial support of the World Bank
• For the Central Asian and other CIS Countries• The program consists of training workshops and study
visits• Builds on the international recommendations
–Training and exercises will be provided –Discusses problems and possible solutions–Exchange of experiences
3Petteri Baer
March 2011 UNECE Statistical Division 414 - 17 March 2011
Workshop I – STS and Seasonal Adjustment
14-17 March, 2011 (Astana, Kazakhstan)
Topics covered:–Why are short-term statistics important? –Use of multiple data sources –Methodology of compilation of STS –Dissemination –Seasonal adjustment in practice –Exercises on seasonal adjustment
• Participants in this Workshop to participate in the next workshop on seasonal adjustment
4Petteri Baer
March 2011 UNECE Statistical Division 514 - 17 March 2011
Workshop II – Challenges in Consumer Price Indices
2011 (Istanbul, Turkey)
Topics covered:–Calculation of elementary and higher-level indices –Coverage of goods and services –Treatment of missing prices and their replacements –Seasonal items –Adjustment for quality changes
• Will be based on the international Consumer Price Indices Manual
5Petteri Baer
March 2011 UNECE Statistical Division 614 - 17 March 2011
Workshop III – Implementation of the 2008 SNA
2011 (Kiev, Ukraine)
Topics covered:–Implementation issues of the 2008 SNA–Discussion of the problematic areas and priority
setting–Support countries to establish implementation plans –Development needs of both national accounts and
the related primary statistics –Building of networks of experts in SNA and NA
6Petteri Baer
March 2011 UNECE Statistical Division 714 - 17 March 2011
Workshop IV – Training in Seasonal Adjustment
2012 (Istanbul, Turkey)
Topics covered:–Tackle the methodological and practical issues of
seasonal adjustment–Seasonal adjustment of problematic time series–Analyses of the quality of seasonal adjustment–Discuss countries’ experiences and problems
7Petteri Baer
March 2011 UNECE Statistical Division 814 - 17 March 2011
Agenda for Monday, 14 March 2011
Session 1
Workshop IntroductionWorkshop Introduction–Welcome–Workshop Purpose and Scope –Challenges and Problems in the area of STS
Session 2
The Aim of STS – User view and Current DevelopmentsThe Aim of STS – User view and Current Developments–General recommendations on STS –Why STS matters? User view on STS
8Petteri Baer
March 2011 UNECE Statistical Division 914 - 17 March 2011
Agenda for Tuesday, 15 March 2011
Session 3
STS Production Methodology STS Production Methodology –STS compilation with multiple data sources –Case studies from countries
Session 4
An Introduction to Seasonality and Seasonal Adjustment An Introduction to Seasonality and Seasonal Adjustment –Components of time series, seasonality and preconditions
for seasonal adjustment –Seasonal adjustment as a process –1. Exercise:1. Exercise: Getting started
9Petteri Baer
March 2011 UNECE Statistical Division 1014 - 17 March 2011
Agenda for Wednesday, 16 March 2011
Session 5
Seasonal Adjustment PracticeSeasonal Adjustment Practice–Why seasonally adjust and how? –Issues on seasonal adjustment in the EECCA countries–Round table: Current state of seasonal adjustment
Session 6
Pre-treatment of Time SeriesPre-treatment of Time Series–Pre-treatment practice for seasonal adjustment including
calendar adjustment –2. Exercise:2. Exercise: Examples of graphical analysis, outlier detection
and the effect of calendar adjustment
10Petteri Baer
March 2011 UNECE Statistical Division 1114 - 17 March 2011
Agenda for Thursday, 17 March 2011
Session 7
Performing Seasonal Adjustment Performing Seasonal Adjustment –Model selection, seasonal adjustment and analyzing results –3. Exercise:3. Exercise: Step-by-step seasonal adjustment
Session 8
Going forward with seasonal adjustmentGoing forward with seasonal adjustment–Disseminating statistical information on economic
development –How to release seasonally adjusted data –Towards the next workshop & closing
11Petteri Baer
March 2011 UNECE Statistical Division 1214 - 17 March 2011
Participants’ expectations
• Gain theoretical knowledge and practical skills
– To carry out seasonal adjustment
– To start providing users with seasonally adjusted data
• To get training and methodological materials
– Exchange experiences with other countries on seasonal adjustment
– Hear about international recommendations on STS and SA
Participant Presentation Round!
12Petteri Baer
March 2011 UNECE Statistical Division 13
Challenges & Problems of Challenges & Problems of Short-Term Statistics (STS)Short-Term Statistics (STS)
Based on the UNECE paper on Short-Term Economic Statistics in the CIS and Western Balkans
Carsten Boldsen Hansen Economic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 14
Agenda
1. Introduction
2. Availability of STS
3. Publication policy
4. Data collection and compilation of time series
5. Seasonal adjustment
6. Conclusions
March 2011 UNECE Statistical Division 15
Introduction
A survey on seasonal adjustment in 2008 Challenges with STS were analyzed in
2007 and 2009 via web sites of NSOs on: Consumer price index Producer price index Producer price index for services Industrial production index Retail trade turnover Turnover of services Volume of services production Wages and salaries
March 2011 UNECE Statistical Division 16
Introduction
Countries included in the assessment: Albania Armenia Azerbaijan Belarus Bosnia and Herzegovina Georgia Kazakhstan Kyrgyzstan Republic of Moldova
Montenegro Russian Federation Serbia Tajikistan The former Yugoslav
Republic of Macedonia Turkmenistan Uzbekistan Ukraine
March 2011 UNECE Statistical Division 17
Introduction
What constitutes international comparability in STS? Coverage Classifications Methodological and computational practices Provision of fixed based and/or discrete time series Provision of long coherent time series Provision of seasonal adjusted series Dissemination of documentation
March 2011 UNECE Statistical Division 18
Availability of Time Series
01
2345
678
91011
12131415
1617
Consumer PriceIndex
IndustrialProduction Index
Producer PriceIndex
Wages andSalaries
Retail TradeTurnover
Num
ber o
f cou
ntrie
s
2007
2009
Availability of time series with more than six observations
March 2011 UNECE Statistical Division 19
Availability of STS on Services
March 2011 UNECE Statistical Division 20
Share of services (incl. trade) in GDP: • 52% 1996 -> 57% 2008 (in the EECCA countries)
Lack of data for services • Problems for estimating GDP
8/17 countries publish wages and salaries Output indicators rarely produced
7 countries publish turnover or volume of services Indicators do not cover the whole service sector
Indicators limited to transport, hotels and restaurants
Availability of STS on Services
March 2011 UNECE Statistical Division 21
01234
56789
101112
1314151617
Wages andSalaries
Number ofEmployees
Turnover in Services Volume of Services Price Information forServices
Num
ber
of c
ount
ries
Availability
Availability of short-term indicators for services (2009)
March 2011 UNECE Statistical Division 22
Publication Policy Issues
Almost all countries publish advance release calendars• A huge step forward in just a few years
Most countries archive releases to websites Time series not easily accessible Few countries have a published revision policy Metadata have been improved Timeliness of releases is often very good
March 2011 UNECE Statistical Division 23
Timeliness
The average timeliness of STS indicators
March 2011 UNECE Statistical Division 24
Publication of Metadata
15 of 17 surveyed countries provide some methodological information in English• Countries subscribing to IMF’s SDDS or GDDS have
more comprehensive set of metadata
For statistics not included in SDDS/GDDS• Very little data for retail trade and services• Many details regarding production methods not
available in English
March 2011 UNECE Statistical Division 25
Publication Methodology
Revisions are a necessary feature of STS Data are rarely published in time series format
• Instead data for a few months is published• Seasonally adjusted data can only be published as
long time series Only half of the countries publish indices with a
fixed reference period Change from previous period should only be
calculated from seasonally adjusted data!
March 2011 UNECE Statistical Division 26
Methodology and Comparability
0123456789
1011121314151617
fixed baseindices for some
indicator
internationallycomparableclassification
seasonallyadjustedstatistics
value added taxdata used for
statistics
Nu
mb
er
of
co
un
trie
s
Production of STS according to international standards (2009)
March 2011 UNECE Statistical Division 27
Many have cut-off samples or totals• Over sampling for some countries and indicators?
Some use registers to reduce sample sizes and increase efficiency
Register data requires substantial IT resources and implementation of new production methods• NSOs may have difficulties in accessing registers?• May provide a solution for developing new statistics?
Data Collection Methodology
March 2011 UNECE Statistical Division 28
Good knowledge of international standards exists Significant methodological differences exist:
• price indices, retail trade turnover, wages and salaries Some incoherences in definitions
• Definitions of turnover, wages and salaries • Treatment of VAT, subsidies and delivery costs • Need to standardize definitions also in the EU
Almost all countries use internationally comparable classifications for economic activities (ISIC/NACE)
Compilation Methodology
March 2011 UNECE Statistical Division 29
Time Series Methodology
Production of cumulative data Suitable for national use only Summarizes development during the current year
• When data are available for April -> information is provided from January to April
• Length of the reference period changes with each release(Jan-Feb > Jan-Mar > Jan-Apr…)
Cumulative data are usually only additional information, not the only type of data• A huge step forward: done by a majority of countries
March 2011 UNECE Statistical Division 30
80
85
90
95
100
105
110
115
120
Jan
Jan
-Fe
b
Jan
-Ma
r
Jan
-Ap
r
Jan
-Ma
y
Jan
-Ju
n
Jan
-Ju
l
Jan
-Au
g
Jan
-Se
p
Jan
-Oct
Jan
-No
v
Jan
-De
c
Jan
Jan
-Fe
b
Jan
-Ma
r
Jan
-Ap
r
Jan
-Ma
y
Jan
-Ju
n
Jan
-Ju
l
Jan
-Au
g
Jan
-Se
p
Jan
-Oct
Jan
-No
v
Jan
-De
c
2008 2009cu
mu
lati
ve
ind
ex
fro
m t
he
co
rre
sp
on
din
g p
eri
od
o
f th
e p
rev
iou
s y
ea
r
industry electricity
Comparison of Series – Cumulative Data
Industrial Production and Production of Electricity in Belarus
Not easy to say which industry is doing better – only the change from previous year visible
March 2011 UNECE Statistical Division 31
40
50
60
70
80
90
100
110
120
130
Jan-
08
Feb
-08
Mar
-08
Apr
-08
May
-08
Jun-
08
Jul-0
8
Aug
-08
Sep
-08
Oct
-08
Nov
-08
Dec
-08
Jan-
09
Feb
-09
Mar
-09
Apr
-09
May
-09
Jun-
09
Jul-0
9
Aug
-09
Sep
-09
Oct
-09
Nov
-09
Dec
-09
ind
ex 2
008=
100
industry electricity
Industrial Production and Production of Electricity in Belarus
Comparison of Series – Monthly Data
Seasonality interferes in comparing monthly data – seasonal adjustment needed
March 2011 UNECE Statistical Division 32
Time Series Methodology
Problems with cumulative data International comparison and analysis not possible Slow identification of turning points
• Change from the previous period in seasonally adjusted data provides faster indications of turning points
User cannot derive a correct monthly time series• Revisions to the earlier periods cannot be matched to the
correct periods of time• Time series from cumulative data have incorrect
seasonality
March 2011 UNECE Statistical Division 33
From Cumulative to Monthly
Cumulative Industrial Production Data (estimates of monthly values)
101
102
102
103
103
104
104
Jan Jan-Feb
Jan-Mar
Jan-Apr
Jan-May
Jan-Jun
Jan-Jul
Jan-Aug
Jan-Sep
Jan-Oct
Jan-Nov
Jan-Dec
cum
ula
tive
ch
ang
e fr
om
th
e co
rres
po
nd
ing
per
iod
o
f th
e p
revi
ou
s ye
ar
Cumulative change
0
200
400
600
800
1000
1200
Jan Jan-Feb
Jan-Mar
Jan-Apr
Jan-May
Jan-Jun
Jan-Jul Jan-Aug
Jan-Sep
Jan-Oct
Jan-Nov
Jan-Dec
mo
nth
ly v
alu
e
101
102
102
103
103
104
104
cum
ula
tive
ch
ang
e fr
om
th
e co
rres
po
nd
ing
p
erio
d o
f th
e p
revi
ou
s ye
ar
Calculated from cumulative values Cumulative change
0
200
400
600
800
1000
1200
Jan Jan-Feb
Jan-Mar
Jan-Apr
Jan-May
Jan-Jun
Jan-Jul Jan-Aug
Jan-Sep
Jan-Oct
Jan-Nov
Jan-Dec
mo
nth
ly v
alu
e
101
102
102
103
103
104
104
cum
ula
tive
ch
ang
e fr
om
th
e co
rres
po
nd
ing
p
erio
d o
f th
e p
revi
ou
s ye
ar
Calculated from cumulative values Real monthly value with revisions Cumulative change
March 2011 UNECE Statistical Division 34
Time Series Methodology
Fixed base indices and/or absolute values for discrete periods are recommended
Time series to be linked or calculated back when base year is changed • Not to shorten the series or to leave breaks
(the series should not start from its b.y.) Previous periods need to be revised to come up
with reliable time series• Currently 10 countries publish time series of more
than 24 observations
March 2011 UNECE Statistical Division 35
Where is the Economy Going?
Frequent changes of base year without links or backcalculation
60
70
80
90
100
110
120
130
2006
m01
2006
m04
2006
m07
2006
m10
2007
m01
2007
m04
2007
m07
2007
m10
2008
m01
2008
m04
2008
m07
2008
m10
2009
m01
2009
m04
2009
m07
2009
m10
2010
m01
2010
m04
2010
m07
2010
m10
pre
vio
us
year
= 1
00
2005=100
2006=100
2007=100
2008=100
2009=100
March 2011 UNECE Statistical Division 36
SA data calculated by 11/17 countries Need for training, materials/guidelines and
support on methodological and practical issues Expansion of number and length of seasonally
adjusted series More metadata on SA needed for the users Development of release practices of SA Standardization of compilation and release
practices would enhance quality of SA
Seasonal Adjustment
March 2011 UNECE Statistical Division 37
Conclusions of the Assessment
1. Need for longer time series• Historical time series to be build and maintained• Backcalculation or linking in base year changes
2. Improve international comparability• Seasonally adjusted data would enable comparison• More comparable information on the service sector
3. Review data collection techniques• Introduce sampling (and allow revisions)• Use administrative sources
4. Publication policy• New release practices (SA, time series, revisions)• More detailed metadata
March 2011 UNECE Statistical Division 38
General Recommendations on General Recommendations on STSSTS
Carsten Boldsen Hansen
Economic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 39
Overview• General guidelines and quality
• Sources for methodology guidelines
• Response burden
• STS vs. SBS
• Time series
• Release Practices
• Metadata
• User consultation
March 2011 UNECE Statistical Division 40
“The use by statistical agencies in each country of international concepts, classifications and methods promotes the consistency and efficiency of statistical systems at all official levels.”The ninth principle of The Fundamental Principles of Official Statistics
in the Region of the Economic Commission for Europe, UNECE
March 2011 UNECE Statistical Division 41
General Guidelines• The Fundamental Principles of Official Statistics (UN)
http://unstats.un.org/unsd/dnss/gp/fundprinciples.aspx
• Quality of Statistics– Data Quality Assessment Framework (IMF)
http://www.imf.org/external/np/sta/dsbb/2003/eng/dqaf.htm
– ESS quality framework (EC) http://epp.eurostat.ec.europa.eu/portal/page/portal/quality/introduction
– OECD quality framework (OECD)http://www.oecd.org/document/43/0,3343,en_2649_33715_21571947_1_1_1_1,00.html
• Handbook of Statistical Organization, The Operation and Organization of a Statistical Agency, 2003http://unstats.un.org/unsd/dnss/hb/default.aspx
March 2011 UNECE Statistical Division 42
The Fundamental Principles
1. indispensable for a democratic society 2. statistical agencies decide methods and
procedures 3. present data according to scientific
standards 4. comment on erroneous interpretation5. statistical agencies choose the data
sources with regard to quality, timeliness, costs and burden
March 2011 UNECE Statistical Division 43
The Fundamental Principles
6. strictly confidentiality of individual data and use exclusively for statistical purposes
7. statistical laws, regulations and measures to be made public
8. coordination among statistical agencies within countries
9. use of international concepts, classifications and methods
10. bilateral and multilateral cooperation
March 2011 UNECE Statistical Division 44
Respondent Burden
• Minimizing respondent burden should be an important objective vs. cut-off sampling
• Coordination of data collections would help reducing response burden and to divide it more evenly among respondents
• Existing sources of information should be used to the largest extent possible– Administrative registers– Commercial datasets– Data collected by other organizations
March 2011 UNECE Statistical Division 45
Coherence
= Degree to which data is logically connected and mutually consistent– Coherence within a data set– Coherence across data sets
• common concepts, definitions, valuation principles, classifications and co-operation
– Coherence over time– Coherence across countries
• Extent to which the recommendations have been adopted
• Link to national accounts important
March 2011 UNECE Statistical Division 46
STS vs Structural Statistics (SBS)
STS = measures economic developmentsSBS = a snapshot describing structure & detail• STS and SBS have different data sources,
definitions, statistical methods, timing and coverage (fiscal/calendar)
• Treatment of changes in the population– SBS: the population in the reference year as it is– STS: makes different time periods comparable
(by correcting for mergers and splits etc) Further improvement of coherence needed!
March 2011 UNECE Statistical Division 47
List of Methodology Guidelines
• Methodology of Short-Term Business Statistics (EC), 2006 http://ec.europa.eu/eurostat/ramon/statmanuals/files/KS-BG-06-001-EN.pdf
• International Recommendations for the Index of Industrial Production (UN), 2010 http://unstats.un.org/unsd/statcom/doc10/BG-IndustrialStats.pdf
• Use of Administrative Sources for Business Statistics Purposes (EC), 1999 http://ec.europa.eu/eurostat/ramon/statmanuals/files/CA-24-99-897-__-N-EN.pdf
• International Recommendations for Distributive Trade Statistics (UN), 2009http://ec.europa.eu/eurostat/ramon/statmanuals/files/Inter_Rec_for_Distribut_Trade_Stat.pdf
• Methodological guide for Producer Price Indices for Services, (EC) 2005 http://ec.europa.eu/eurostat/ramon/statmanuals/files/KS-BG-06-003-EN.pdf
• Evolution of Service Statistics, proceedings of a seminar, (EC) 2002 http://ec.europa.eu/eurostat/ramon/statmanuals/files/KS-BG-02-001-__-N-EN.pdf
• Consumer Price Index Manual, Theory and Practice, 2004 (ILO) http://www.ilo.org/public/english/bureau/stat/guides/cpi/index.htm
• Practical Guide to Producing Consumer Price Indices, 2009 (UNECE/ILO)http://www.unece.org/stats/publications/Practical_Guide_to_Producing_CPI.pdf
• Producer Price Index Manual, Theory and Practice, 2004 (IMF) http://www.imf.org/external/np/sta/tegppi/index.htm
• Export and Import Price Manual, 2008 (IMF)http://www.imf.org/external/np/sta/tegeipi/index.htm
March 2011 UNECE Statistical Division 48
Time Series Recommendations
• Fixed base indices and/or absolute values for discrete periods to be provided
• New series should be linked to the old series to produce continuous series
• Cumulative statistics should be published only as additional information
• If seasonality influences the indicator, seasonally adjusted and trend series to be published
• Reference period should be a year and be updated when the weights are updated
March 2011 UNECE Statistical Division 49
Importance of Long Time Series (1)
• Long and consistent time series important for– International comparison– Analysis – Appraisal of business cycles
• Current practices of countries vary significantly• Currently no international standards on:
– Length of time series – Methods for backcasting– Implementing changes of classifications
March 2011 UNECE Statistical Division 50
Importance of Long Time Series (2)
• STS regulation (EC) requires time series from 2000, Eurostat recommends much longer series
• Methodology of Short-Term Business Statistics:
“To carry out statistical analysis such as seasonal adjustment it is generally considered necessary to have observations for a minimum of 5 years… …for example, in the search for turning points it is important to be able to have data available for several complete cycles.”
March 2011 UNECE Statistical Division 51
Dissemination
• Data should be released as soon as possible– Trade off between timeliness and quality
• Data to be released according to a set timetable• Confidentiality to be secured• Data made available to all users at the same time• Data to be revised as new information is available• Data to be accompanied by explanations• Contact details of relevant statisticians to be given
Source: Index of Industrial Production (UN)
March 2011 UNECE Statistical Division 52
Reference Period
• Some guidelines by IMF and Eurostat:
– Prices, output and sales Monthly
– (GDP), labour variables at least Quarterly (un/employment monthly)
March 2011 UNECE Statistical Division 53
Timeliness Guidelines
• Some guidelines by IMF and Eurostat
Indicator DaysEurostat IMF
output 40 45prices 30-45 30new orders 50sales / turnover 60 60persons employed 60hours worked 90wages and salaries 90 90
March 2011 UNECE Statistical Division 54
Metadata and Dissemination Guides
• UN – The Common Metadata Framework– Making Data Meaningful 1 and 2– Guidelines for Statistical Metadata on the Internet– Terminology on Statistical Metadata
• OECD – Data and Metadata Reporting and Presentation– Glossary of Statistical Terms – Publishing Standards for Datasets and Data Tables
• IMF – Guide to the Data Dissemination Standards
March 2011 UNECE Statistical Division 55
Metadata Management
“data about data”• Identify users and provide only valuable metadata• Ensure metadata is easily available• Reuse metadata for integration and efficiency • Preserve history (old versions) of metadata • Document variations from standards• Make metadata work an integral part of production• Create a coherent system for metadata• Ensure that the metadata for users reflects the real
production process
March 2011 UNECE Statistical Division 56
Contents of Metadata• Purpose and main uses of the statistics• Definitions of the underlying economic concepts• Reference to the respective legislation• Data coverage, periodicity and timeliness• Self assessment of data quality
– revision history and accuracy of concepts– availability, comparability and coherence– limitations in the use of statistics
• Descriptions of the methodologies used– Index formula, weighting and frequency of revising– Frequency of re-basing and linking methods– Treatment of changes in commodities or quality
Source: Index of industrial production (UN) & some NSO’s recommendations
March 2011 UNECE Statistical Division 57
User Consultation
• Create mechanisms to obtain users’ views on regular basis in order to– identify priority areas for improvement– ensure responding to user needs– provide users with advice on the strengths and
weaknesses of your statistics• Important with key indicators (CPI, IPI, GDP...)• Engaging users should be an integral part of the
work in the NSO– Mutual understanding and exchange of information
(=transparency) builds trust– Decisions will be made independently by the NSO
Source: Practical Guide to Producing CPI (UN)
March 2011 UNECE Statistical Division 58
Why are Short-term economic statistics important?
User views on Short-term economic statistics
UNECE Training Workshop “Short-term statistics and seasonal adjustment”
Astana, 14 – 17 March 2011
Petteri Baer, Marketing Manager, Statistics Finland
March 2011 UNECE Statistical Division 5914 - 17 March 2011
Everybody acknowledges the importance of National Accounts
GDP
“- Yes, of course”
• Sure, we need something to measure the economic growth of the society
• Sure, we need international comparability at the level of different economies
• Sure, we need a denominator for the shares of different industries in the economy
59Petteri Baer
March 2011 UNECE Statistical Division 60
What about short-term statistics on economic development?
• It may be another story• Why?
–No real supply provided by official statistics (?)–Too many disappointments, when requesting good
AND timely statistics–Substitutive indicators, guesstimates in wide use–And if there IS some supply
• Often not much information provided to potential users
• Misinterpretations about the consistency of the statistical information due to necessary revisions
Petteri Baer
March 2011 UNECE Statistical Division 61
More and more statistical publication takes place on the internet…
• This is a very positive development
• Availability and accessibility of official statistics has grown substantially
• In the beginning of the year 2008
–>500 Million internet hosts in the world!
• This also increases pressure on timeliness
Internet hosts in the Worldin the beginning of each year
* Millions *
0.00002 0.03 0.716
147
433
0
50
100
150
200
250
300
350
400
450
500
1982 1987 1992 1997 2002 2007
March 2011 UNECE Statistical Division 6214 - 17 March 2011
But note:– There are traps on the way!
• Just putting your information on your web site does not automatically mean it is utilized
• Even though your web information is utilized, it does not mean that your most important users make use of it
March 2011 UNECE Statistical Division 63
All potentialusers
These guysare real users
Heavy user
Traps on the way, continued
You may cover only a tinyshare of your potential users- but not recognize it!
Petteri Baer
March 2011 UNECE Statistical Division 6414 - 17 March 2011
Traps on the way, continued
• Counting the popularity of your web site by “hits” may deceive you because
–a substantial part of the “fabulous growth” comes from search engines checking if you have any new information
March 2011 UNECE Statistical Division 65
Very often you do not really know, who your users are, when you
provide services on the internet
Petteri Baer
March 2011 UNECE Statistical Division 6614 - 17 March 2011
A Problem- as posed in a Pretoria Expert
Group Seminar in 2003 • “Statisticians like to
talk to one another in mysterious ways and with a time clock which runs 10 times slower than everyone else's”
We have to communicate better!
We have to imagine ourselves into the position of the potential users of our information!
March 2011 UNECE Statistical Division 6714 - 17 March 2011
Especially concerning Short-term economic statistics…
• Timeliness is an issue of huge importance
• Trade-off between timeliness and accuracy - always
• Globalization• Modern communication
tools
March 2011 UNECE Statistical Division 68
The voice of a policy maker
•Rob Wright, Deputy Minister of the Department of Finance of Canada, opening in May 2009 the
• International Seminar on Timeliness, Methodology and Comparability of Rapid Estimates of Economic Trends in Ottawa, organized by UNSD and Statistics Canada
Petteri Baer
March 2011 UNECE Statistical Division 69
Rob Wright: The role of Data in Public Policy Development
• “Process of policy development is lengthy–Research and analysis required to understand when and where policy change and reform are needed
–Designing policy–Developing consensus (political, public)
•We rely on high quality and timely data throughout the process”
Petteri Baer
March 2011 UNECE Statistical Division 70
Rob Wright: How can statisticians improve public policy?
• “Identify and close data gaps
•Priority to long term issues and trends
•But there is a need to be able to adapt to current developments, if possible–Identify turning points
•Consult users–Government, private sector, researchers”
Petteri Baer
March 2011 UNECE Statistical Division 7114 - 17 March 2011
The importance of turning points in the assessing
development• In assessing economic
development• In designing financial and
monetary policy decisions
• In making investment decisions
March 2011 UNECE Statistical Division 72
The voice of a businessman and banker
•Don Drummond, Chief Economist of Toronto Dominion Bank, Canada, opening the International Seminar, mentioned above:
–"I would rather have no data than wrong data."
Petteri Baer
March 2011 UNECE Statistical Division 73
Statistics Canada performed a stakeholder need assessment in
2008 (1)• …with the central users
of Short-term economic statistics
• A great deal of trust for Canadian Statistics was expressed in the services of Statistics Canada on STS
• At the same time, the following improvements were required:
• Better timeliness• Additional questions to be
included in the Labour Force Survey
• Better metadata and• Improvements on the web
site and the search engine
Petteri Baer
March 2011 UNECE Statistical Division 74
Statistics Canada performed a stakeholder need assessment in
2008 (2)• Other issues that came
up in this consultation with the main stakeholders:
• A worry that a NSO should not rush into short-term data collection without the usual testing of the instrument
• Short term ad hoc measures will not be useful unless there is time series continuity
• The agency should continue its longer-term investments in quality improve-ments and new data products
Petteri Baer
March 2011 UNECE Statistical Division 75
Conclusions
• Short-term economic statistics ARE important and their importance is growing
–Not only as a service item–But also for building up the image and the reputation of
the statistical agency as an important, accurate and timely information provider
• Identifying turning points in the economic cycle is one of the most important functions
–Availability and comparability of long term time series and seasonal adjustment of fixed term indices are of top importance
• Keep your users well informed about your services• Understand your audiences’ specific needs by better
communication
Petteri Baer
March 2011 UNECE Statistical Division 76
STS Compilation STS Compilation with Multiple Data Sourceswith Multiple Data Sources
Anu PeltolaEconomic Statistics Section,
UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 77
Overview
• Data collection– Sampling – Administrative data– Combining multiple data sources
• Compilation of results– Data editing– Non-response and weighting– Treatment of non-comparable changes
• Publication• Improvement
March 2011 UNECE Statistical Division 78
Theoretical Concept – A Key to Good Quality
• Define the purpose of an indicator • Links to the real world
– What should it describe? – Who are the users/uses (internal/external)?– Possible data sources
• Links to other statistics– Differences in concepts, scope, methods– Goal variables – national accounts/SBS– Regular benchmarking – Follow-up of differences
Act Plan
Check
Do
Continuous improvement Q
uality
TimeBy Deming
March 2011 UNECE Statistical Division 79
Production Process
• Bring the collected data to the level of the intended statistical output!
Publication
Collection of data
Correction of systematic
errors in data
Index calculation
Check for the most important observations
March 2011 UNECE Statistical Division 80
Data CollectionStatistical Units
• Corner stones of business statistics– Legal unit -> enterprise (services) -> enterprise
groups– Establishment (for industry/construction)
• Business registers are fundamentally important– Bridge between administrative and statistical units– The economic activity class (ISIC/NACE)– Improve its comprehensiveness – use as a frame– Examine opportunities to use administrative data– Interactive: update with information from STS
UN: International recommendations for the Index of Industrial Production &EC: STS Metholodological manual
March 2011 UNECE Statistical Division 81
Source: Statistics Finland, Strategy for economic statistics
System of Statistics
March 2011 UNECE Statistical Division 82
• Give clear instructions– Explain the concepts to the respondents
• Revisions to earlier months– Aim to pre-fill the questionnaire with data given earlier – Leave space for reporting revisions
• Always test changes to questionnaires • Inform the respondents of the use of data• Develop useful feedback for respondents
– your company compared to others in the same activity
Data CollectionQuestionnaire Design
March 2011 UNECE Statistical Division 83
• Many surveys are for units above a size threshold– Burdensome and problems with the coverage of small units
• Based on business register and periodically reviewed
• In drawing a sample, special attention to be paid to:– Level of details to be published– Resources available– Accuracy and timeliness required– Response burden
• Simple/stratified sampling by activity and size
Data CollectionSampling in Practice
March 2011 UNECE Statistical Division 84
> Business Register to be kept up-to-date with new units
Total population of unitsin the Business Register
Large units Medium units Small units
Stratification by economic activity
Covered on a complete
enumeration basis
Covered by sampling
Covered mainly by
administrative sources
or administrative sources
March 2011 UNECE Statistical Division 85
• Administrative registers or datasets can be used as:– Single source in their own right– Frame for sampling via the Business Register– Complementary source– Validation– Data source for small enterprises
• For STS limited administrative sources available:– VAT (value added tax)– Social security data (employment and labor cost)– Building permits, etc.
Data CollectionAdministrative Data Sources
March 2011 UNECE Statistical Division 86
+ Reduction of response burden
+ Reduction of costs, data collection and manual work
+ Total populations - detailed classifications/regional indicators
+ Better quality and coverage (of smallest units)
- Data content, units, concepts and definitions may differ- Dependence on few large data suppliers- Timeliness - may require use of estimation- Access and confidentiality- Non-observed economy unlikely to be included- Requires good IT capacity by the supplier and the NSO
Data CollectionPros and Cons of Admin Data?
March 2011 UNECE Statistical Division 87
• National ID-system for enterprises• New production methods:
– to correct for negative values and different concepts– slow accumulation > estimation of missing data
• The most important units to direct collection– Active co-operation with large enterprises
• Development of questionnaires:– Simplification – part of information from registers– Efficiency – electronic data collection
Data CollectionAdministrative Data and Quality
March 2011 UNECE Statistical Division 88
• Compulsory to use existing data (if suitable) in statistics production
• Guaranteed access to administrative sources• State government and social security
institutions obliged to deliver their data to the NSO– Free of charge or compensation of direct costs
– Co-operation in making changes in data collection
• To ensure data confidentiality– Individual data collected for statistics should not be handed
over to any use other than statistics or research!
Data CollectionLegislative Issues
March 2011 UNECE Statistical Division 89
CompilationCentral Role of VAT Data
Source: Statistics Finland
March 2011 UNECE Statistical Division 90
VATe.g. 250 000 units
• Turnover• Estimates for output
and missing data
Business Registere.g. 290 000 units
• Unit IDs• Activity code
• Location• Mergers
• LKAU (regional)
1. release
2. release
revision
optimalsampling
small & mediumenterprises
feedbackto BR
Samplee.g. 2000 units
• Turnover• Mergers
sample,basic info
updates to BRactivity of units
combining
CompilationLinking Admin and Survey Data
March 2011 UNECE Statistical Division 91
CompilationData Control and Editing
• Studying data to identify errors– Detect errors that have a significant influence– Check whether values are within given ranges – Check whether values for related variables are coherent– Compare to past responses (previous months and a year
ago)
• Give top priority to outliers and errors that have the largest impact on the results
• Outlier values require careful treatment– May be correct but caused by unusual circumstances
Source: Methodology of Short-Term Business Statistics, EC
March 2011 UNECE Statistical Division 92
CompilationTreating Non-Response
• Controlling response burden– Better planning of data collection process– Offering various channels for respondents
• Reducing the effect of non-response– Alternative source, e.g. administrative data– Imputation based on historical data– Mean value imputation, donor/nearist neighbour,
regression of variables
March 2011 UNECE Statistical Division 93
CompilationComparing Unit Level Data
0
10000
20000
30000
40000
50000
60000
70000
80000
1 2 3 4 5 6 7 8 9 10 11 12Months
Previous year Current year
Change 115%
March 2011 UNECE Statistical Division 94
-2.40
-1.13-0.58
-2.33
-2.53
-4.1229.70
-1.20
60
80
100
120
140
160
180
1 2 3 4 5 6 7 8Months
Index without a unit Index with a unit
CompilationImpact on the Results
index
March 2011 UNECE Statistical Division 95
CompilationNon-Comparable Changes (NCCs)
• Structural changes in the population: – New units are set up and others stop existing– Units may be taken over, merged or split up – Units may expand, contract or change their activities
• Reasons for large changes1) Errors
2) Actual changes that are comparable
3) Actual changes that are non-comparable
UN Guide on the Impact of Globalization on National Accounts > helps with STS as well
March 2011 UNECE Statistical Division 96
CompilationExample of NCCs
Unit ATurnover = 100 million
Unit BTurnover = 75 million
Exchange of goods50 million
Turnover drops by one third due to a merger!No change in the level of activity!
Unit ABTurnover =
(100-50) + 75 = 125 million
Previous year Current year
March 2011 UNECE Statistical Division 97
CompilationAlternative Treatments of NCCs
1. All changes are recorded as they are (actual)− Contaminated with apparent, non-comparable changes− Difficult to obtain a picture of economic reality+ Simplicity
2. Panel method– Only same units in both periods are included− Start-ups and closures would be cancelled out− Seriously biased results in highly dynamic populations+ Simplicity
March 2011 UNECE Statistical Division 98
CompilationAlternative Treatments of NCCs
3. Overlapping method– Actual comparable changes are not
adjusted– Other changes are made comparable by
a. Collecting comparable information (largest units)
b. Replacing non-comparable figure by an estimate
c. Taken the unit out of calculation (no effect to results)
− Requires more work+ Results reflect actual changes in
economic activity
Firm X
Firm X
March 2011 UNECE Statistical Division 99
CompilationConfrontation with Other Sources
• Regular confrontation may reveal discrepancies
• Aim at coherence: value = price x output
• First at the aggregated level and where necessary at lower levels (largest units)
• Knowledge of differences between statistics helps communication with users
• Quality reviews of indicators to be undertaken
March 2011 UNECE Statistical Division 100
New Requirements for STS?
• Globalization– Internationally comparable data needed– Treatment of more complex business activities
• Increasing amount of services– Output and price measures, industrial services
• Detection of turning points– Longer time series and seasonal adjustment
• Coherence– Compare to National Accounts and
between price/volume/value indicators
March 2011 UNECE Statistical Division 101
Components of Time Series, Components of Time Series, Seasonality and Pre-conditions for Seasonality and Pre-conditions for
Seasonal AdjustmentSeasonal Adjustment
Anu PeltolaEconomic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 102
Overview
• Basic Concepts
• Components of Time Series
• Seasonality
• Pre-conditions for Seasonal Adjustment
March 2011 UNECE Statistical Division 103
Basic Concepts
• Index comes from Latin and means a pointer, sign, indicator, list or register– A ratio that measures change– As per cent of a base value (base always 100)– Each observation is compared to the base value
• Time series are a collection of observations, measured at equally spaced intervals– Stock series = at a point in time (discrete)– Flow series = period in time (continuous)
new observation
old observation
x 100
March 2011 UNECE Statistical Division 104
Components of Time Series
Seasonal adjustment is based on the idea that time series can be decomposed
The components are: Seasonal Irregular Trend
March 2011 UNECE Statistical Division 105
Trend-Cycle Component
50
60
70
80
90
100
110
120
130
Relation of ComponentsComponents of the Industrial Production Index of Kazakhstan
Ind
ex
20
05
=1
00
Original component
50
60
70
80
90
100
110
120
130
140
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
Jan-
08
Jan-
09
Jan-
10
Seasonal component
50
60
70
80
90
100
110
120
Jan-0
0
Jan-0
1
Jan-0
2
Jan-0
3
Jan-0
4
Jan-0
5
Jan-0
6
Jan-0
7
Jan-0
8
Jan-0
9
Jan-1
0
Irregular component
50
60
70
80
90
100
110
120
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
Jan-
08
Jan-
09
Jan-
10
March 2011 UNECE Statistical Division 106
Seasonal Component
= Depicts systematic, calendar-related movements
has a similar pattern from year to year refers to the periodic fluctuations within a
year that re-occur in approximately the same way annually
Is removed in seasonal adjustment
March 2011 UNECE Statistical Division 107
Irregular Component
= Depicts unsystematic, short term fluctuations• The remaining component after the seasonal
and trend components have been removed• Certain specific outliers, such as those caused
by strikes, also belong to this component• Sometimes called the residual component • May or may not be random with random
effects (white noise) or artifacts of non-sampling error (not necessarily random)
March 2011 UNECE Statistical Division 108
Trend Component
= Depicts the long-term movement in a series• A trend series is derived by removing the
irregular influences from the seasonally adjusted series
• A reflection of the underlying development• Typically due to influences such as population
growth, technological development, inflation and general economic development
• Sometimes referred to as the trend-cycle
March 2011 UNECE Statistical Division 109
IPI – KazakhstanAn Example of the Components of Time Series
Ind
ex
20
05
=1
00
50
60
70
80
90
100
110
120
130
140
Jan-
00
Jul-0
0
Jan-
01
Jul-0
1
Jan-
02
Jul-0
2
Jan-
03
Jul-0
3
Jan-
04
Jul-0
4
Jan-
05
Jul-0
5
Jan-
06
Jul-0
6
Jan-
07
Jul-0
7
Jan-
08
Jul-0
8
Jan-
09
Jul-0
9
Jan-
10
Jul-1
0
Original Seasonally adjusted Trend
March 2011 UNECE Statistical Division 110
Causes of Seasonality
= seasons e.g. holidays and consumption habits, which are related to the rhythm of the year– Warmth in summer and cold in winter BUT not
extreme weather conditions (irregular component)
• Seasonality reflects traditional behavior associated with:
• The calendar• Christmas and New Year• Social habits (the holiday season), • Business (quarterly provisional tax payments) and • Administrative procedures (tax returns)
March 2011 UNECE Statistical Division 111
Seasonality
60
70
80
90
100
110
120
130
140
150
1 2 3 4 5 6 7 8 9 10 11 12
2000
2001
2002
2003
2004
2005
2006
2007
2008
Industrial production in Moldova, original series 2000-2008
months
Ind
ex
20
05
=1
00
March 2011 UNECE Statistical Division 112
Seasonal Effect
= Intra-year fluctuations in the series that repeat• A seasonal effect is reasonably stable with
respect to timing, direction and magnitude • The seasonal component of a time series is
comprised of three main types of systematic calendar-related influences: – Seasonal influences– Trading day influences – Moving holiday influences
March 2011 UNECE Statistical Division 113
Trading Day Effect
= The impact on the series, of the number and type of days in a particular month
• Different days may have a different weight• A calendar month comprises four weeks (28
days) plus extra one, two or three days• Rarely an issue in quarterly data, since
quarters have 90, 91 or 92 days
March 2011 UNECE Statistical Division 114
Trading DaysSaturday
Source: Analysis of Daily Sales Data during the Financial Panic of 2008, John B. Taylor (Target Corporation’s sales)
March 2011 UNECE Statistical Division 115
Moving Holidays
= The impact on the series of holidays whose exact timing shifts from year to year
• Examples of moving holidays:– Easter – Chinese New Year - where the exact date is
determined by the cycles of the moon– Ramadan
March 2011 UNECE Statistical Division 116
0
5
10
15
20
25
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
avera
ge w
ork
ing
days
2009
2010
2011
Moving Holidays
Impact of moving holidays to the number of working days
Ascension day Christmas moves between weekdays and weekend
March 2011 UNECE Statistical Division 117
Working Days and Seasonality
Example of average working days in 2009 - 2011
0
5
10
15
20
25
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
aver
age
wo
rkin
g d
ays
2009
2010
2011
March 2011 UNECE Statistical Division 118
Sudden Changes
• Outliers– Extreme values with identifiable causes (strikes or
extreme weather conditions)– Part of irregular component
• Trend breaks (level shifts)– The trend component suddenly increases or
decreases in value– Often caused by changes in definitions (tax rate,
reclassification)• Seasonal breaks
– The seasonal pattern changes, e.g. due to a structural change caused by a crisis or administrative issues such as timing of invoicing
March 2011 UNECE Statistical Division 119
Pre-conditions for Seasonal Adjustment
1. Good quality of raw data– Strange values to be checked (zeros or outliers)– Revision of errors with new acquired data
2. Length of time series 36/12 or 16/4– At least 36 observations for monthly series and
16 observations for quarterly series needed
3. Consistent time series– To provide data according to a base year– Use of comparable definitions and classifications– Remove non-comparable changes
4. Solid structure– Presence of seasonality, moderate volatility– No major breaks in seasonal behaviour
March 2011 UNECE Statistical Division 120
Seasonal Adjustment Process Seasonal Adjustment Process with Demetra+with Demetra+
Anu PeltolaEconomic Statistics Section, UNECE
March 2011 UNECE Statistical Division 121
Overview
Seasonal adjustment process:
• Prepare and check
• Define and adjust
• Analyse and refine
• Document and publish
March 2011 UNECE Statistical Division 122
60
80
100
120
140
160
180
200
Prepare and check
Open Demetra+
Prepare a source file
Import data
Check the original series
March 2011 UNECE Statistical Division 123
Open Demetra+
• The program can be downloaded at: http://circa.europa.eu/irc/dsis/eurosam/info/data/demetra.htm
• Can be used free of charge
March 2011 UNECE Statistical Division 124
Prepare a source file
• Many types of files are suitable
• Excel file either horizontal or vertical
March 2011 UNECE Statistical Division 125
Import data
• You can copy and paste– But you will have to do this
every time
• Dynamic updates possible – Next time you do not have to
import again
1
2
March 2011 UNECE Statistical Division 126
Check the original series
• The quality of seasonal adjustment depends on the quality of the raw data. – accuracy, – length of time series (36/16), – quality of production methods and – consistency of time series
• Demetra+ provides many visual tools, i.e. to test the presence of seasonality
March 2011 UNECE Statistical Division 127
Check the original series
Is seasonality present in the original series?
March 2011 UNECE Statistical Division 128
60
80
100
120
140
160
180
200
Define and adjust
Select an approach
Prepare calendars
Select regressors
Seasonally adjust
March 2011 UNECE Statistical Division 129
Select an approach
• Choose to process only one time series or multiple series
• Select the approach (X12 or T/S)– Working instructions for beginning
seasonal adjustment with Demetra+ written for TRAMO/SEATS
– More instructions for the use of X-12-ARIMA in the Demetra+ Manuals
March 2011 UNECE Statistical Division 130
Prepare calendars
• Demetra+ includes some predefined holidays, but not national holiday calendars of all countries– Collect a time series of your national holidays
• Prepare the calendar by – Selecting from the pre-specified holidays– Defining exact dates – Setting a validity period
March 2011 UNECE Statistical Division 131
Select regressors
= Define the specification
• Start the analysis with the default specifications – choose first either four (RSA4)
or five (RSA5)
Mark your specification
RSA0 level,airline model
RSA1 log/level,outliers detection, airline model
RSA2 log/level, working days, Easter, outlier detection, airline model
RSA3 log/level, outlier detection, automatic model identification
RSA4log/level, working days, Easter, outlier detection, automatic model identification
RSA5log/level, trading days, Easter, outlier detection, automatic model identification
March 2011 UNECE Statistical Division 132
Seasonally adjust
• To launch adjustment – double click on the series or– select from the main menu:
Seasonal adjustment/ Single Analysis/ New
Seasonal adjustment/ Multiprocessing/ New
SAProcessing-xx/ Run
• Save to the workspace double click
March 2011 UNECE Statistical Division 133
Seasonally adjust
• For second adjustment of the same data, decide your update strategy
• Current adjustment = fixed forecasts
• Concurrent adjustment = nothing fixed
March 2011 UNECE Statistical Division 134
60
80
100
120
140
160
180
200
Analyse and refine
Visual check
Models applied
Quality Diagnostics
Refine and adjust
March 2011 UNECE Statistical Division 135
Visual check
Is the seasonal component lost in the irregular?
March 2011 UNECE Statistical Division 136
Visual check
Check the S-I ratio for moving seasonality
March 2011 UNECE Statistical Division 137
Models applied
• Information about pre-processing: – the estimation time span used, – corrections for trading days and Easter,– type of applied ARIMA model, – the dates and types of outliers as well as– the distribution of residuals
• Information about decomposition:– the applied decomposition model, – statistical indicators to validate the model
March 2011 UNECE Statistical Division 138
Quality Diagnostics
• Demetra+ offers a wide range of quality measures
• Verbal description also• Basic checks = annual totals• Residuals, i.e. the part of data
not explained by modelling, should not include any information
March 2011 UNECE Statistical Division 139
Quality Diagnostics
Is there some residual seasonality after adjustment?
March 2011 UNECE Statistical Division 140
Quality Diagnostics
Are there large revisions – is the model stable?
March 2011 UNECE Statistical Division 141
Quality Diagnostics
Do the residuals follow the normal distribution?
Are they random?
March 2011 UNECE Statistical Division 142
Refine and adjust
• Seasonal adjustment is an iterative process
• After the first adjustment, you can try different specifications – For example, from RSA 5 to RSA 4
• In multiprocessing, you can edit a single series by double clicking its name in the results
• You can maintain the previous adjustments for comparison
March 2011 UNECE Statistical Division 143
60
80
100
120
140
160
180
200
Document and publish
Document choices
Export data
Prepare publication
Support users
March 2011 UNECE Statistical Division 144
Document choices
• Document the first page of Main results, Pre-processing, Decomposition and Diagnostics
• Archive the resulting time series for later revision analysis
• Prepare metadata to be published to the users of statistics– ESS metadata template– OECD Data and Metadata Reporting and
Presentation Handbook
March 2011 UNECE Statistical Division 145
Export data
• You can export to several kinds of outputs – Main menu: SAProcessing-xx/Generate output– Copy: TramoSeatsDoc-x/Copy/Results
March 2011 UNECE Statistical Division 146
Prepare publication
• Redesign the content of data releases when you start with seasonally adjusted data
• Consider how to explain revisions, and plan in advance how to revise these data
• The press releases should be simple– Offer some more details on the web site, e.g.
regional or industry level data
March 2011 UNECE Statistical Division 147
Support users
• Seasonal adjustment aims at better service for the users of statistics
• Define a clear framework / policy – choice of method and software– timing of reanalysis of parameters and models – treatment of outliers – practices with revision and dissemination
March 2011 UNECE Statistical Division 148
Why Seasonally Adjust Why Seasonally Adjust and How?and How?
Approaches X-12-ARIMA, TRAMO/SEATS, X-13 A/S and Tools
Anu PeltolaEconomic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 149
Overview
• What and why
• Basic concepts
• Methods
• Software
• Recommendations
• Useful references
March 2011 UNECE Statistical Division 150
A Coyote Moment Did We Notice the Turning Point?
March 2011 UNECE Statistical Division 151
Economic Crises – Statistics
• Did we give any warnings? – A responsibility for the statistical offices? A new task? – Important to all users of statistics
• Not only to politicians, but also to enterpreneurs and citizens
• Statistical offices often have monopoly to analyze detailed data sets– We should not forecast, but draw attention to statistics– Identify changes early, leading indicators, develop more
flash estimates -> quality vs. timeliness
• Otherwise, a risk of marginalisation of NSOs
March 2011 UNECE Statistical Division 152
Economic Crises – Conclusions
• Some limits of official statistics were highlighted by the critics:– lack of comparability among countries– need for more timely key indicators– need for statistical indicators in areas of
particular importance for the financial and economic crisis
Source: Status Report on Information Requirements in EMU
March 2011 UNECE Statistical Division 153
Turning Points Trend vs. Year-on-Year Rate
Volume of Construction
-20%
-10%
0%
10%
20%
30%
40%
20
05
20
06
20
07
20
08
20
09
20
40
60
80
100
120
140
160
Change from corresponding month Trend series
March 2011 UNECE Statistical Division 154
Why Seasonally Adjust?
• Seasonal effects in raw data conceal the true underlying development– Easier to interpret, reveals long-term development
• To aid in comparing economic development– Including comparison of countries or economic
activities
• To aid economists in short-term forecasting• To allow series to be compared from one
month to the next– Faster and easier detection of economic cycles
March 2011 UNECE Statistical Division 155
Why Original Data is Not Enough?
• Comparison with the same period of last year does not remove moving holidays – If Easter falls in March (usually April) the level of
activity can vary greatly for that month
• Comparison ignores trading day effects, e.g. different amount of different weekdays
• Contains the influence of the irregular component
• Delay in identification of turning points
March 2011 UNECE Statistical Division 156
Seasonal Adjustment
• Seasonal adjustment is an analysis technique that:– Estimates seasonal influences using procedures and
filters– Removes systematic and calendar-related
influences
• Aims to eliminate seasonal and working day effects– No seasonal and working day effects in a perfectly
seasonally adjusted series
March 2011 UNECE Statistical Division 157
Interpretation of Seasonally Adjusted Data
• In a seasonally adjusted world:– Temperature is exactly the same during
both summer and winter– There are no holidays– People work every day of the week with the
same intensity
Source: Bundesbank
March 2011 UNECE Statistical Division 158
Filter Based Methods
• X-11, X-11-ARIMA, X-12-ARIMA (STL, SABL, SEASABS)
• Based on the “ratio to moving average” described in 1931 by Fredrick R. Macaulay (US)
• Estimate time series components (trend and seasonal factors) by application of a set of filters (moving averages) to the original series
• Filter removes or reduces the strength of business and seasonal cycles and noise from the input data
March 2011 UNECE Statistical Division 159
X-11 and X-11-ARIMA
X-11• Developed by the US Census Bureau • Began operation in the US in 1965• Integrated into software such as SAS and STATISTICA• Uses filters to seasonally adjust data
X-11-ARIMA• Developed by Statistics Canada in 1980• ARIMA modelling reduces revisions in the seasonally
adjusted series and the effect of the end-point problem• No user-defined regressors, not robust against outliers
March 2011 UNECE Statistical Division 160
X-12-ARIMA
http://www.census.gov/srd/www/x12a/
• Developed and maintained by the US Census Bureau• Based on a set of linear filters (moving averages)• User may define prior adjustments• Fits a regARIMA model to the series in order to detect
and adjust for outliers and other distorting effects• Diagnostics of the quality and stability of the adjustments• Ability to process many series at once • Pseudo-additive and multiplicative decomposition• X-12-Graph generates graphical diagnostics
March 2011 UNECE Statistical Division 161
X-12-ARIMA
Source: David Findley and Deutsche Bundesbank
March 2011 UNECE Statistical Division 162
Model Based Methods• TRAMO/SEATS, STAMP,
”X-13-ARIMA/SEATS”• Stipulate a model for the data (V. Gómes and A.
Maravall)
• Models separately the trend, seasonal and irregular components of the time series
• Components may be modelled directly or modelling by decomposing other components from the original series
• Tailor the filter weights based on the nature of the series
March 2011 UNECE Statistical Division 163
TRAMO/SEATS
www.bde.es
• By Victor Gómez & Agustin Maravall, Bank of Spain• Both for in-depth analysis of a few series or for
routine applications to a large number of series• TRAMO preadjusts, SEATS adjusts• Fully model-based method for forecasting • Powerful tool for detailed analyses of series• Only proposes additive/log-additive decomposition
TRAMO = Time Series Regression with ARIMA Noise, Missing Observations and OutliersSEATS = Signal Extraction in ARIMA Time Series
March 2011 UNECE Statistical Division 164
DEMETRA softwarehttp://circa.europa.eu/irc/dsis/eurosam/info/data/demetra.htm
• By EUROSTAT with Jens Dossé, Servais Hoffmann, Pierre Kelsen, Christophe Planas, Raoul Depoutot
• Includes both X-12-ARIMA and TRAMO/SEATS• Modern time series techniques
to large-scale sets of time series• To ease the access of non-specialists • Automated procedure and
a detailed analysis of single time series• Recommended by Eurostat
March 2011 UNECE Statistical Division 165
X-12-ARIMA vs. TRAMO/SEATS
Source: Central Bank of Turkey (2002): Seasonal Adjustment in Economic Time Series.
March 2011 UNECE Statistical Division 166
Demetra+ software
• Users can choose: – Tramo-Seats model-based adjustments– X-12-ARIMA
• One interface• Aims to improve comparability
of the two methods • Uses a common set of diagnostics
and of presentation tools• Necmettin Alpay Koçak is
a member of the testing group
March 2011 UNECE Statistical Division 167
Common Guidelines
1. Use tools and software supported widely– Demetra+ will be supported by Eurostat– Methodological guidelines will be available– Results will be more comparable
2. Use your national calendars3. Dedicate enough human resources to SA4. Define a SA strategy5. Aim at a clear message to the users
– Consider which series serve the purpose of the indicator– Document all relevant choices and events
March 2011 UNECE Statistical Division 168
Useful references
• Eurostat is preparing a Handbook on Seasonal Adjustment
• ESS Guidelines on Seasonal Adjustment
http://epp.eurostat.ec.europa.eu/cache/ITY_OFFPUB/KS-RA-09-006/EN/KS-RA-09-006-EN.PDF
• Central Bank of the Republic of Turkey (2002). Seasonal Adjustment in Economic Time Series. http://www.tcmb.gov.tr/yeni/evds/yayin/kitaplar/seasonality.doc
• Hungarian Central Statistical Office (2007). Seasonal Adjustment Methods and Practices. www.ksh.hu/hosa
• US Census Bureau. The X-12-ARIMA Seasonal Adjustment Program. http://www.census.gov/srd/www/x12a/
• Bank of Spain. Statistics and Econometrics Software. http://www.bde.es/servicio/software/econome.htm
• Australian Bureau of Statistics (2005). Information Paper, An Introduction Course on Time Series Analysis – Electronic Delivery. 1346.0.55.001. http://www.abs.gov.au/ausstats/[email protected]/papersbycatalogue/7A71E7935D23BB17CA2570B1002A31DB?OpenDocument
March 2011 UNECE Statistical Division 169
Issues on Seasonal Adjustment in the Issues on Seasonal Adjustment in the EECCA countriesEECCA countries
Based on a UNECE Seasonal Adjustment Survey
Carsten Boldsen HansenEconomic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 170
Agenda
• Objective & Scope of the Survey of 2008• Survey Findings
– Current State of Seasonal Adjustment– Dissemination Policies– Applied Seasonal Adjustment Approach– Pre-Treatment of Time Series– Validation of Seasonal Adjustment– Plans for Future Development
• Recommended Future Measures
March 2011 UNECE Statistical Division 171
Objective & Scope of the Survey
• UNECE survey to National Statistical Offices (NSOs) and Central Banks (CBs)– Conducted in October 2008– Targeted to 17 CIS and Western Balkans countries– 11 answers were received from 10 different countries
• To obtain information on the current state in order to offer support
• Main factors identified: – Non-comparability due to lack of seasonal adjustment– Many have started seasonally adjusting– Interest is raising in order to:
• Detect turning points earlier• Enable comparison of sectors of the economy and countries
March 2011 UNECE Statistical Division 172
Current State of SA Practices
• 11/17 countries publish some SA data• Most common statistics seasonally adjusted are:
– GDP (6), Industrial production (3), Exports and imports (4)– NSOs adjust about 30 monthly and 30 quarterly time series
• 9 NSOs reported limited capacity in terms of SA• Unified SA procedure in 5/9 offices• All NSOs that perform SA have some methodological
descriptions about SA• Length of SA time series vary between 48 – 216
months, and quarterly between 16 – 72 quarters
March 2011 UNECE Statistical Division 173
Dissemination of SA Data
• Most publish raw data, SA and trend series– SA series are usually published in regular
publications and/or on the Internet – None of the countries published working day
adjusted series
• Most NSOs publish metadata on the methods of seasonal adjustment
• All eight countries that produce SA series disseminate the data also in paper publications
March 2011 UNECE Statistical Division 174
Seasonal Adjustment Approach
• The most commonly used SA approach is X-12-ARIMA <> TRAMO/SEATS in EU
• Some tendency to shift towards using TRAMO/SEATS and “X-13 ARIMA/SEATS”
• “X-13-ARIMA/SEATS” was expected to be the future tool for seasonal adjustment > Demetra+ software
• 4/9 NSOs currently performing SA have chosen their approaches via international orgs
March 2011 UNECE Statistical Division 175
Seasonal Adjustment Approach
0
2
4
6
8
10
12
14
16
TRAMO/SEATS X12-ARIMA X11-ARIMA X13-ARIMA/SEATS
No
of
co
un
trie
s
CIS and Western Balkans EU countries
Nu
mb
er o
f co
un
trie
s
March 2011 UNECE Statistical Division 176
Pre-Treatment of Time Series
= graphical analysis, calendar adjustment, model selection, filters, outliers and parameters
• Frequency of updating parameters varies– Some do not regularly update the parameters– Some update them, when new data appends– One country updates also in case of atypical weather conditions
• The choice of regressor differs – The most commonly used regressor in seasonal adjustment
is either working days or working days and specific holidays
• Outliers are detected in most cases by tests– Outliers are validated by experts in Armenia, Bosnia and
Herzegovina and Kazakhstan
• It is recommended to document events causing outliers!
March 2011 UNECE Statistical Division 177
• Mostly by using graphical inspection, autocorrelation function and measures of stability over time
• Graphical inspection:– Deviation between the raw and SA series– Standard deviation relative to trend – Intuitive assessment
• Two countries mentioned analysing significant peaks of the autocorrelogram of the raw series
• One uses the Box-Pierce statistics and three use F-tests
• Use of validation measures depends on the SA software used
Validation of Seasonal Adjustment
March 2011 UNECE Statistical Division 178
Validation of Seasonal Adjustment
• Residuals and fit statistics were less used• Four countries mentioned using M-Statistics as
a quality measure of the results• A set of common quality measures are being
constructed– US Census Bureau, Eurostat and the Bank of Spain – For all seasonally adjusted series to be assessed by
the same criteria
> Demetra+ includes some quality measures both from X-12-ARIMA and TRAMO/SEATS
March 2011 UNECE Statistical Division 179
Validation of Seasonal Adjustment
0
1
2
3
4
5
6
7
8
Graphicalinspection
Autocorrelationfunction
Statistics onresiduals
Fit statistics Stability over time
No
ou
t o
f 9 o
rgan
izati
on
s
Used Not used
Nu
mb
er o
f co
un
trie
s
March 2011 UNECE Statistical Division 180
Plans for Future Development
Type of Assistance Needed
1. All countries indicated some need for assistance, because of the lack of resources and knowledge
2. Five countries reported receiving assistance from international organizations
3. Regional training or workshops proposed by most of the respondents
4. Exchange of methodological information and experiences were suggested
Training 10
Technical assistance
9
Methodological material in Russian
7
Methodological material
4
Funding 3
March 2011 UNECE Statistical Division 181
Future Measures
• All reported some need for assistance• UNECE will organize training workshops
in 2010 – 2012• Methodological material and practical guidelines will
be produced and published also in Russian
What are your plans now?– Will be discussed next
March 2011 UNECE Statistical Division 182
Pre-treatment practices forPre-treatment practices for Seasonal Adjustment Seasonal Adjustment
Including Calendar AdjustmentIncluding Calendar Adjustment
Necmettin Alpay KOÇAK
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS)
and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011Astana, Kazakhstan 182
March 2011 UNECE Statistical Division 183
Introduction
• Seasonal adjustment is a statistical procedure with the target of removing the seasonal (and calendar) component from a time series.
• The idea behind is that a series is composed by unobserved components such as trend, cycle, seasonality, irregular
• The seasonal component disturbs short-term analysis, so it is removed from the original series to facilitate the monitoring and interpretation of the economy by analysts
183
March 2011 UNECE Statistical Division 184
First step: the graph of the series
184
Each series must be plotted against time to detect visually whether or not a seasonal component is present (but in some case it is not sufficient!)Example: Industrial production index - Total, Kazakhstan :
Period: 2000M1-2010M10
March 2011 UNECE Statistical Division 185
First step: the graph of the seriesSeasonal graphs are a special form of line graph in which you plot separate line graphs for each season in a regular frequency monthly or quarterly data.Example: Industrial production index - Total, Kazakhstan :
Period: 2000M1-2010M10
March 2011 UNECE Statistical Division 186
The unobserved components
March 2011 UNECE Statistical Division 187
Decomposition scheme
• A time series yt can be decomposed in
Yt = TCt +St+εt
The Additive model
• A time series yt can also be decomposed in
Yt = TCt×St×εt
log(Yt)=log(TCt)+log(St)+log(εt)
Multiplicative model
March 2011 UNECE Statistical Division 188
The REG-ARIMA model
•The REG-ARIMA model is a convenient way to represent a time series with deterministic and stochastic effects. Given the observed time series zt , it is expressed as,
zt = ytβ+xt
Φ(B)δ(B)xt=θ(B)at•where•β is a vector of regression coefficients•yt denotes n regression variables•B is the backshift operator (Bkyt = yt-k )•Φ(B), δ(B), and θ(B) are finite polynomials in B• at is assumed a normal independently identically distributed (NIID) (0,σa
2) white-noise innovation
March 2011 UNECE Statistical Division 189
The regression variables
• The regression variables capture the deterministic components of the series. In TS, these can be of different type:– Calendar effects
• Trading day effect• Easter effect• Leap-year effect• Holidays
– Intervention variables generated by the program– Regression variables entered by the user– Outliers
March 2011 UNECE Statistical Division 190
The ARIMA model
• Model-based-pre-adjustment identifies and fits an ARIMA model on the linearized series (cleaned from deterministic effects). The ARIMA model is composed of three components:– the stationary AR component (polynomial Φ(B))– the non-stationary AR component (polynomial δ(B))– the invertible MA component (polynomial θ(B))
• For seasonal time series, the polynomials are given by:• Φ(B) = (1+ Φ1B + … + ΦpBp)(1+ Φ1Bs + …+ ΦPBs×P)
• δ(B) = (1-B)d(1-Bs)D
• θ(B) = (1+ θ1B + … + θpBp)(1+ θ1Bs + …+ θPBs×P)• A seasonal ARIMA model is identied by the order of its polynomials:
(p;d;q)(P;D;Q)
March 2011 UNECE Statistical Division 191
TRAMO / Reg-ARIMA
• Program for estimation, forecasting, and interpolation of regression models with missing observations and ARIMA errors, with possibly several types of outliers
• The program is aimed at monthly or lower frequency data (quarterly, semester, 4-month, bimonth, semester, year)
• Performs a pretesting to decide between a log transformation and no transformation
March 2011 UNECE Statistical Division 192
TRAMO / Reg-ARIMA
• Identifies the ARIMA model through an Automatic Model Identification (AMI) procedure
• Interpolates missing values
• Detects outliers
• Estimates the REG-Arima model
• Computes forecasts
March 2011 UNECE Statistical Division 193
Automatic model identification
• The ARIMA model can be automatically identified by the program• Two steps
– Obtains the order of differencing – max order ∆2 ∆s
– Obtains the multiplicative stationary ARMA model• 0<=(p;q)<=3• 0 <=(ps;qs )<=1• Chosen with the BIC criterion, favors balanced model (similar orders
of AR and MA parts)• Otherwise, it can be input by the user (parameters P,D,Q, BP,BD,
BQ)• It works jointly with the Automatic Outlier Detection and Correction
(AODC)
March 2011 UNECE Statistical Division 194
Outliers
• They represent the effect on the time series of some special events (new regulation, major political or economical reform, strike, natural disaster). Three possible forms of outliers:– Additive outliers (AO)– Level Shift (LS)– Transitory Changes (TC)
March 2011 UNECE Statistical Division 195
Outliers
March 2011 UNECE Statistical Division 196
Calendar effects
• Calendar adjustment removes those non-seasonal calendar effects from the series, for which there is statistical evidence and an economic explanation. Four possibilities in TS:– Trading days (working/non-working, 6 regressors))– National and moving holidays ((provided by the user))– Leap-year (TS versus X-12-ARIMA)– Easter
• A pre-testing on the presence of these effects.• If trading days are significant, adding the holidays
variable improves significantly the results!
March 2011 UNECE Statistical Division 197
Examples of calendar effects
March 2011 UNECE Statistical Division 198
Trading/Working Day Adjustment
• Aims at a series independent of the length and the composition in days– Length of month, number of working days and weekend days,
composition of working days (Monday/Friday)
• Working or trading-day adjustment is recommended for series with such effects– If effects not present –Regressors should not be applied
• Compile, maintain and update national calendars!– A historical list of public holidays including information on
compensation holidays
March 2011 UNECE Statistical Division 199
Correction for Moving Holidays
• Occur irregularly in the course of the year• Correct for detected moving holidays in series
– Not removed by standard filters– If effects not present –Regressors should not be
applied• These effects may be partly seasonal:
– The Catholic Easter, for example, falls more often in April than in March
• Since the seasonal part is captured by seasonal adjustment filters, it should not be removed during the calendar adjustment
March 2011 UNECE Statistical Division 200
An illustrative example for national calendar regressor
Number of days
Number of Sunday's
Official (fixed) holidays except Sunday's
Religious (moving) holidays except Sunday's
Working days
Average working days (1975-2015)
Mean Corrected working day regressor
Jan.97 31 4 1 0 26 25.14 0.86
Feb.97 28 4 0 2 22 23.76 -1.76
Mar.97 31 5 0 0 26 26.00 0.00
Apr.97 30 4 1 3 22 24.36 -2.36
May.97 31 4 1 0 26 25.14 0.86
Jun.97 30 5 0 0 25 25.24 -0.24
Jul.97 31 4 0 0 27 25.98 1.02
March 2011 UNECE Statistical Division 201
Original vs. Linearized series
March 2011 UNECE Statistical Division 202
Model Selection, Seasonal Model Selection, Seasonal Adjustment, Analyzing ResultsAdjustment, Analyzing Results
202
Necmettin Alpay KOÇAK
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment
14 – 17 March 2011
Astana, Kazakhstan
March 2011 UNECE Statistical Division 203 203
Model Selection
• Pre-treatment is the most important stage of the seasonal adjustment
• X-12-ARIMA and TRAMO&SEATS methods use very similar (nearly same) approaches to obtain the linearized (pre-treated) series.
• Both method use ARIMA model for pre-treatment.
• The most appropriate ARIMA model → Linearized series of top quality
March 2011 UNECE Statistical Division 204 204
ARIMA Model selection
• zt = ytβ+xt
• Φ(B)δ(B)xt=θ(B)at
• (p,d,q)(P,D,Q)s → Structure of ARIMA
• (0,1,1)(0,1,1)4,12
• For the model– Parsimonious– Significance of parameters– Smallest BIC or AIC
• For the residuals– Normality– Lack of auto-correlation– Linearity– Randomness
March 2011 UNECE Statistical Division 205
Diagnostics
• Are there really any seasonal fluctations in the series ?– Seasonality test
• If, yes– Diagnostics based on residuals are the core
of the analysis.
• If, no– No need to seasonal adjustment.
March 2011 UNECE Statistical Division 206
Diagnostics
• Seasonality test– Friedman test– Kruskall-wallis test
• Residual diagnostics– Normality
• Skewness• Kurtosis
– Auto-correlation• First and seasonal frequencies (4 or 12)
– Linearity• Auto-correlation in squared residuals
– Randomness• Number of sign (+) should be equal the number of sign (-) in residuals.
• Final Comment... We select the appropriate model according to the state of the diagnostics.i
March 2011 UNECE Statistical Division 207
Seasonal Adjustment
• 2.1 Choice of SA approach
• 2.2 Consistency between raw and SA data
• 2.3 Geographical aggregation: direct versus indirect approach
• 2.4 Sectoral aggregation: direct versus indirect approach
• (Source : ESS Guidelines)
March 2011 UNECE Statistical Division 208
Choice of seasonal adjustment method
• Most commonly used seasonal adjustment methods• Tramo-Seats• X12ARIMA
• Tramo-Seats: model-based approach based on Arima decomposition techniques
• X-12-ARIMA: non parametric approach based on a set of linear filters (moving averages)
• Univariate or multivariate structural time series models• (Source : ESS Guidelines)
March 2011 UNECE Statistical Division 209
Filtering data: Difference in methods
• X-12-ARIMA use fixed filters to obtain seasonal component in the series.• A 5-term weighted moving average (3x3 ma) is calculated for each month of the
seasonal-irregular ratios (SI) to obtain preliminary estimates of the seasonal factors
• Why is this 5-term moving average called a 3x3 moving average?
3333
1 24121212122433
tttttttttx
tSISISISISISISISISI
S
March 2011 UNECE Statistical Division 210
Filtering data: Difference in methods
• TRAMO&SEATS use a varying filter to obtain seasonal component in the series. This variation depends on the estimated ARIMA model of the time series.
• For example, if series follows an ARIMA model like (0,1,1)(0,1,1), it has specific filter or it follows (1,1,1)(1,1,1), it has also specific filter. Then, estimated parameters affect the filters.
• Wiener-Kolmogorov filters are used in Tramo&Seats. It fed with auto-covariance generating functions of the series. (more complicated than X-12-ARIMA)
• But, it is easily interpreted since it has statistical properties.
March 2011 UNECE Statistical Division 211
Consistency between raw and SA data
• We do not expect that the annual totals of raw and SA data are not equal.
• Since calendar effect exists (working days in a year)
• It is possible to force the sum (or average) of seasonally adjusted data over each year to equal the sum (or average) of the raw data, but from a theoretical point of view, there is no justification for this.– Do not impose the equality over the year to the raw
and the seasonally adjusted or the calendar adjusted data (ESS Guidelines)
March 2011 UNECE Statistical Division 212
Direct and indirect adjustment
• Direct seasonal adjustment is performed if all time series, including aggregates, are seasonally adjusted on an individual basis. Indirect seasonal adjustment is performed if the seasonally adjusted estimate for a time series is derived by combining the estimates for two or more directly adjusted series. The direct and indirect issue is relevant in different cases, e.g. within a system of time series estimates at a sector level, or aggregation of similar time series estimates from different geographical entities.
Industrial Production
Index
Mining and Quarrying
Manufacturing
Electricity, Water, Natural Gas and etc.
EU-27 Aggregate
Germany
France
Spain
Romania
...
March 2011 UNECE Statistical Division 213
Analyzing result
• Use a detailed set of graphical, descriptive, non-parametric and parametric criteria to validate the seasonal adjustment. Particular attention must be paid to the following suitable characteristics of seasonal adjustment series:
– existence of seasonality– absence of residual seasonality– absence of residual calendar effects– absence of an over-adjustment of seasonal and calendar effects– absence of significant and positive autocorrelation for seasonal lags in the
irregular component– stability of the seasonal component
• In addition, the appropriateness of the identified model used in the complete adjustment procedure should be checked using standard diagnostics and some additional considerations. An important consideration is that the number of outliers should be relatively small, and not unduly concentrated around the same period of the year.
March 2011 UNECE Statistical Division 214
Analyzing resultsTime
Series
DeterministicComponents
StochasticComponents
Trend-Cycle Seasonal IrregularOutliersCalendarEffects
Additive Outlier(AO)
Transitory Change(TC)
Level Shift (LS)Final
Trend-CycleFinal
SeasonalFinal
Irregular
Seasonally AdjustedSeries
March 2011 UNECE Statistical Division 215
Revisions to seasonal adjustment
• Forward factors / current adjustment: annual analysis to determine seasonal and trading day factors– Preferable for time series with constant seasonal
factor or large irregular factor causing revision
• Concurrent adjustment: uses the data available at each reference period to re-estimate seasonal and trading day factors
March 2011 UNECE Statistical Division 216
Revisions to seasonal adjustment
• Forecast seasonal factors for the next year (current adjustment)
• Forecast seasonal factors for the next year, but update the forecast with new observations while the model and parameters stay the same
• Forecast seasonal factors for the next year, but re-estimate parameters of the model with new observations while the model stays the same (partial concurrent adjustment)
• Compute the optimal forecast at every period and revise the model and parameters (concurrent adjustment)
Sources: Eurostat working paper on Seasonal Adjustment Policy, ESS Guidelines on Seasonal Adjustment
March 2011 UNECE Statistical Division 217
Evaluation of revision alternatives
• The use of fixed seasonal factors can lead to biased results when unexpected events occur
• Re-estimation in every calculation increases accuracy but also revision
• Re-estimation once a year decreases accuracy but also revision
Re-identification usually once a year However, time series revise in every release
March 2011 UNECE Statistical Division 218
Disseminating statistical information on economic
development
UNECE Training Workshop “Short-term statistics and seasonal adjustment”
Astana, 14 – 17 March 2011
Petteri Baer, Marketing Manager, Statistics Finland
March 2011 UNECE Statistical Division 219
Information services usually provided on economic
development• Opinion surveys (e.g., business and consumer surveys,
purchasing managers surveys, bank lending survey): at least monthly with a high timeliness
• Market data (e.g. stock market data, exchange rates, yields): at least daily, frequently “tick-by-tick”
• Short-term statistics (e.g. Harmonised CPI (HICP), unemployment rate, leading indicators): monthly
• Monetary and financial statistics (e.g. MFI (bank) balance sheets & interest rates, securities, balance of payments): monthly – mainly by central banks
• National accounts (e.g. GDP, sector accounts): quarterly and annually
Petteri Baer
March 2011 UNECE Statistical Division 22014 - 17 March 2011
Operational environment of services on economic statistics
• Growing demand for fresh and timely statistical information on the economic development
–Highlighted by the financial and economic crisis–But existing in all phases of the economic cycle
• Strong influence of rapid communication tools• Growing information overflow• Importance of quality issues
–For the service products themselves–For building up the reputation of official statistics
Petteri Baer
March 2011 UNECE Statistical Division 22114 - 17 March 2011
To develop better interaction with users we need to be proactive
• Who are our present users of economic statistics?
• Who are our present users of short-term economic statistics?
• Who are our POTENTIAL users of short-term economic statistics?
March 2011 UNECE Statistical Division 22214 - 17 March 2011
Nota bene! Users are different: Tourists, Farmers, Miners…
Courtesy to Armin Grossenbacher
We will most likely have to define different ways to approach different users, if we wish to have efficient communication
222Petteri Baer
March 2011 UNECE Statistical Division 22314 - 17 March 2011
So – Who are in need of information on STS? (1)
• The easy reply– Policy makers– Business community– Media and – General public
• But note: Policy makers are much more than only ministries
– Central Bank– High level advisory groups– The district (oblast) level– The local level– Trade unions, lobbies,
NGOs…
March 2011 UNECE Statistical Division 22414 - 17 March 2011
So – Who are in need of information on STS? (2)
• The business community is a much broader target group than often believed
– Banks– Insurance companies– Big corporations– Medium sized enterprises– Chambers of commerce– Branch organizations– Employers organizations– Foreign companies– Etc.
March 2011 UNECE Statistical Division 22514 - 17 March 2011
So – Who are in need of information on STS? (3)
• And numerous target groups mentioned above usually employ
– Business analysts, researchers, economists
• Or make use of– Information brokers– Business intelligence
systems or – Knowledge managements
systems• Do we provide sufficient
information services to them? In proper forms?
March 2011 UNECE Statistical Division 22614 - 17 March 2011
To disseminate economic statistics efficiently: Database services!
• We produce quite a lot of statistical information• Different users have different need structures, they want
information–By industries, By enterprise sizes–By regions–Comparisons over different time periods–International comparisons–And numerous other aspects…
• PC-Axis, PX-Web… User friendly services!
226Petteri Baer
March 2011 UNECE Statistical Division 22714 - 17 March 2011
The importance of good contact information
• User lists–Existing customers and
contacts–Regular and heavy
users of economical statistics
• Contact directories• Feedback contacts Contact / Customer
database Customer Relationship
Management (CRM)
March 2011 UNECE Statistical Division 22814 - 17 March 2011
To improve user interaction we will at least need a Customer Database
• For contacting• For surveys on
satisfaction or dissatisfaction
• For presenting new targeted services
• For other forms of interaction
March 2011 UNECE Statistical Division 22914 - 17 March 2011
Marketing
Customers
Mana-gement
Adminis-tration
Sales
Product development
Developmentprojects
Customerintelligence
Project intelligence
Contactintelligence
Lead intelligence
Informationretrievals Quote
intelligence
Campaignintelligence
Customer Database
Customerservice
Functions Information providers Information users
Projectgroups 229Petteri Baer
March 2011 UNECE Statistical Division 23014 - 17 March 2011
We will profit much from active feedback
• Having more feedback will help us to develop our services
• Interaction with critical customers will help us in having a positive pressure on performing better
• A demanding customer is like a grain of sand within the mussel. It doesn’t feel good but the result may be a beautiful pearl!
March 2011 UNECE Statistical Division 23114 - 17 March 2011
Publicity may sometimes be tough!
• Statistics often tends to attract hostile media coverage…
• “Why does it take so long?”
• “My own perception is different!”
• “Lies, Big lies, Statistics!”Any votes for the Census?But - is there an alternativeto providing better publicity?
March 2011 UNECE Statistical Division 23214 - 17 March 2011
Questions to be raised
• The importance of good and timely statistical information
–How can resources be allocated if basic information on the economic development is based on guesses or too old information?
–Attracting investments, doing good business, developing economic activities needs good infrastructure – reliable official statistics is fundamental
–If the denominators of such as population statistics, GDP etc. are wrong, no real information is reliable
232Petteri Baer
March 2011 UNECE Statistical Division 23314 - 17 March 2011
Take publicity seriously
• Develop useful statistical service products• Make a good plan of what will be published
–Publication calendar• Provide regular Media/Press releases• Make use of your agency’s Press Officer
–Press Conferences from time to time–…but not too often!–Also critical media should always be invited!–Cost plan and budget for publicity activities
• Follow up on media appearance – both quantities and attitudes
233Petteri Baer
March 2011 UNECE Statistical Division 23414 - 17 March 2011
The media is your partner
• In disseminating the main results of all your hard work on statistics
• In making problematic issues known
• Help the media to be well informed!
March 2011 UNECE Statistical Division 23514 - 17 March 2011
Statistical literacy
• Beyond students or school pupils• Special information seminars, breakfast sessions or other
kinds of light information meetings for selected target groups would be advisable
• The web site of the International Statistical Literacy Project of IASE could be helpful in planning
–http://www.stat.auckland.ac.nz/~iase/islp/countries
235Petteri Baer
March 2011 UNECE Statistical Division 23614 - 17 March 2011
Presentation of statistical information
on Short-term statistics• KISS – Keep It Short and Simple• Storytelling approach
–UNECE’s “Making Data Meaningful” materials 1 - 3–Available on the web, also in Russian, so far only 1 – 2
• Focus on turning points• Omit accidental events and “noise”
–The importance of seasonal adjustment
236Petteri Baer
March 2011 UNECE Statistical Division 23714 - 17 March 2011
Conclusions – Five items of great importance
• Existence of good STS services• Packaging the STS services into for the different user
categories relevant service products• Maintaining good accessibility on the web• Obtaining, updating and increasing contact information on
users• Meeting and discussing with and learning from main users
237Petteri Baer
March 2011 UNECE Statistical Division 23814 - 17 March 2011
The efforts you will make will be rewarded, because…
• Only used statistical information is useful statistical information!
• Thank you for your attention!
March 2011 UNECE Statistical Division 239
How to Release Seasonally How to Release Seasonally Adjusted DataAdjusted Data ? ?
(Examples of release practices, metadata, maintenance)
Carsten Boldsen HansenEconomic Statistics Section, UNECE
UNECE Workshop on Short-Term Statistics (STS) UNECE Workshop on Short-Term Statistics (STS) and Seasonal Adjustmentand Seasonal Adjustment14 – 17 March 2011, Astana, Kazakhstan
March 2011 UNECE Statistical Division 240
Overview
• Quality of SA
• Revision policy
• Release practices
• Metadata
• How to get started?
March 2011 UNECE Statistical Division 241
Quality of SA
• Release SA data after you are convinced about the quality
• Explain possible quality issues to users– Quality of original data, length of time-series– Presence of strange features, outliers and
volatility
• Use more time with key indicators• Release documentation of all relevant
seasonal adjustment steps
March 2011 UNECE Statistical Division 242
Timing of Revisions
• SA data usually revised due to– Corrections and accumulation of raw data– Better estimates for the seasonal pattern
• Revisions are welcomed – They derive from improved information set– Forecasts are replaced with new observations in SA
• In SA one new observation can revise the past• Trade off between precision in SA data and
stability of seasonal adjustment pattern• Revision should be scheduled in a regular way
March 2011 UNECE Statistical Division 243
Nature of SA RevisionsIndustrial Production Index, Original Series
Release 9/2008
Release 11/2008
Release 12/2008
Release 10/2008
Source: Statistics Finland
March 2011 UNECE Statistical Division 244
Nature of SA RevisionsIndustrial Production Index, Seasonally Adjusted
Release 12/2008
Release 11/2008
Release 9/2008
Release 10/2008
Source: Statistics Finland
March 2011 UNECE Statistical Division 245
Features of the Trend Series
• End of the trend series may change direction• Problems with defining trends
– How smooth should it be vs. identification of turning points?
– Should it include economic ‘cycles’ or just long-term structural effects?
• Trend can have different annual totals from the original data
• Trend is a good visual tool
March 2011 UNECE Statistical Division 246
50.0
60.0
70.0
80.0
90.0
100.0
110.0
120.0
130.0
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
Published 05/2008
End-point ProblemTurnover in Advertising
50.0
60.0
70.0
80.0
90.0
100.0
110.0
120.0
130.0
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Published 05/2008 Published 03/2009
March 2011 UNECE Statistical Division 247
Releasing Time Series
A. Publish raw and some adjusted data, e.g. one of the following: SA series, SA plus WDA series, Trend-cycle series
B. Include only raw data in press releasesToo limited approach!
C. Present only levels or valuesToo limited approach!
March 2011 UNECE Statistical Division 248
Recommended Release Practices
• Publish rather index numbers than monetary valuesor both
• Month-on-month and change from the same month one year earlier are both useful
• A reference period needs to be determined• Provide long an coherent time series• Present the main contributors to change
– Present products / enterprise groups / industries that are primarily responsible for the monthly movement
Source: Index of Industrial Production (UN)
March 2011 UNECE Statistical Division 249
Revision Policy
• Revisions are inevitable to the quality of data• Be informative about the reasons for revisions
– Methodological, accumulation, errors, changes to classifications– Users should be reminded of the size of the likely revisions
• Correct errors as soon as possible • Revision policy be formulated: regular timing• Revisions to be carried back in time to maintain
consistent time series• Normal revisions: no explicit info & monthly release• Prior information of large scale revisionsSource: Common Revision Policy for STS (EC) & Index of industrial production (UN)
March 2011 UNECE Statistical Division 250
Advance Release Calendars
• Use of advance release calendars is recommended widely
• Reduces the chances of external interference with the release of statistics
• IMF requires the countries that subscribe to the SDDS (but not GDDS) to provide advance release information
• Statistics to be released as soon as the data becomes available and has been processed
March 2011 UNECE Statistical Division 251
Proposed Release Practice
1. Include both raw and SA data in the release, details on the web site– Time series (raw, SA, WDA, trend)
2. Avoid annualized or cumulative growth rates as the only indicator
3. Avoid presentation of trend data in press releases– Trend series are good in graphs!
4. Release several growth rates– “Period on period” growth to be computed on SA data!– Annual growth to be computed on non-adjusted data
March 2011 UNECE Statistical Division 252
Effect of Moving Holidays
1.2
8.1
5.1
0
1
2
3
4
5
6
7
8
9
Seasonally adjusted, from March 2008
Original data, from April 2007
Working day adjusted, from April 2007
Output of the national economy grew in AprilSeasonally adjusted output rose by 1.2 per cent in April from the month before. Year-on-year the increase amounted to 8.1 per cent according to the original series. April 2008 included three working days more than the comparison month of the previous year. Adjusted for working days, the year-on-year growth was 5.1 per cent.
March 2011 UNECE Statistical Division 253
Comparison of CountriesChange from Previous Period
Source: OECD statistical news release
March 2011 UNECE Statistical Division 254
Detecting Turning PointsChange from the Previous Period
Source: UNECE weekly
Seasonally adjusted industrial production growth rates from the previous quarter
-12%
-10%
-8%
-6%
-4%
-2%
0%
2%
4%
20
05
Q1
Q2
Q3
Q4
20
06
Q1
Q2
Q3
Q4
20
07
Q1
Q2
Q3
Q4
20
08
Q1
Q2
Q3
Q4
20
09
Q1
Q2
Q3
Q4
20
10
Q1
Q2
% f
rom
pre
vio
us
qu
art
er
CIS countries ¹ ² EU27 countries North America ²
March 2011 UNECE Statistical Division 255
Source: NSO of Tajikistan
Smoothening of DevelopmentWhat is the direction of development?
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
20
05
20
06
20
07
20
08
20
09
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
month-on-month
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
2005
2006
2007
2008
2009
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
month-on-month 3 months' moving average
March 2011 UNECE Statistical Division 256
Source:ONS
March 2011 UNECE Statistical Division 257
1. Non-technical explanation of SA2. Enough metadata for assessment of reliability3. Metadata to enable repetition of SA:
– Method and software used – Decision rules, aggregation policy– Outlier detection and correction methods– Revision policy– Description of working day adjustment– Contact information
A metadata template is annexed to ESS guidelines on seasonal adjustment!
Metadata for Different Users
Source: OECD data and metadata reporting and presentation handbook
March 2011 UNECE Statistical Division 258
Suggestions for Starting with SA
• Assess your and users’ needs• Define a clear SA policy, covering:
– Method, software, reanalysis, outliers, revision etc.– Choose simple and reliable method and software
• Allocate sufficient resources and time• Train staff• Inform users:
– about major events affecting seasonal adjustment– easy access to all relevant metadata
• Do not publish until confident with the results