EUROPEAN COMMISSION DG ECFIN Business and Consumer Surveys

1
EUROPEAN COMMISSION - DG ECFIN Business and
Consumer Surveys
Seasonal Adjustment of Business and Consumer
Survey Series
Pilar Bengoechea October 2004
2
Introduction

• In BCS series, seasonality should not be present
  because respondents should answer questions of
  surveys taking into account the particular time
  of the year
• However, very often it is (for reasons not well
  understood)

3
Introduction

• Since 2003, the project of SA of the series in
  the Joint Harmonised EU Programme of BCS has been
  considered as one of the main priorities in order
  to
• Provide general guidelines concerning the
  Seasonal adjustment of BCS
• Study the possibility of taking into account a
  more decentralised approach to the SA

4
Methods and Softwares for SA

• Parametric (Specification of a Model)
• The seasonal component is stochastic
• There are unit roots at all seasonal frequencies
• Arima Models (Tramo/Seats (1997) , X-11 Arima
  (1988) and X-12 Arima (1996))
• Structural Models (Stamp6.0 (1996))
• There are some units roots Specific models
• The seasonal component is Deterministic
• Based on local regressions Dainties (1979) and
  BV4 (1983)
• Non parametric (implicit models), e.g. X-11

5
DG ECFIN publication practice

• Series derived from BCS are compiled by DG ECFIN
  from national data
• Aggregations and SA data are carried out by DG
  ECFIN
• Seasonal adjustment is done by DAINTIES
• Pre-adjustment is carried out only in exceptional
  cases
• Raw data are not published
• There are some discrepancies (due to the
  different methods used for sa ) between the sa
  series published by the institutes and those
  published by DG ECFIN
• There is a trade-off between coherence of DG
  ECFIN publications and institutes publications.

6
Important point to recall

• In difference to Eurostat, which operates under a
  Regulation on Short-term Statistics, no such
  regulation exists for DG ECFIN publications
• Therefore, the primary concern is the search for
  the best approach to SA
• Moreover, in order to ensure the quality and
  comparability of the data a common approach is
  desirable

7
DG ECFIN uses DAINTIES for all BCS series

• Advantages
• All series adjusted with the same procedure
• Due to the ease of calculation very fast (50,000
  series in a couple of hours) important for
  timely publication
• No revisions (revisions may confuse users as
  recent analysis presented in press releases and
  other publications could be reversed)
• In general, it performs well on BCS data because
  seasonal component is generally small
• Disadvantages
• Not used outside DG ECFIN
• Assumes seasonality without testing
• Use of an asymmetric moving average (phase shift)
• No preadjustment (outliers etc.) which may
  seriously distort the estimated seasonal
  component

8
Should DG ECFIN change to another method of SA?

• Against criticism of the use of Dainties, DG
  ECFIN has undertaken to evaluate Dainties in
  comparison with other methods and softwares
• Clearly, any change in the method needs to be
  carefully evaluated and justified in order to
  avoid reducing the reliability of data, breaks in
  historical data, comparability, etc.
• ? Creation of a study group in 2003

9
Study Group 2003

• Participants DG ECFIN, Eurostat, ECB, OECD
• Objectives
• To study the technical and empirical aspects of
  the seasonal adjustment of BCS
• To review DAINTIES
• To consider alternative standard methods
• TRAMO and SEATS (TS)
• X-12-REGARIMA (X12)

10
Study Group 2003 (contd)

• Selected Methodological issues studied
• Automatic model identification
• Direct versus indirect seasonal adjustment
• Outlier treatment
• Revisions
• Impact of sa and aggregation method on the
  determination of turning points
• Smoothing of series

11
Study Group 2003 (contd)

• Reports
• DG ECIFN
• Technical and empirical aspects of sa methods in
  the Joint Harmonised EU Programme of BCS
• The impact of sa and aggregation methods on the
  determination of turning points
• A Butterworth filter to smooth series from BCS
  and short term quantitative indicators
• Eurostat
• Direct and indirect approaches to sa of
  qualitative surveys
• European Central Bank
• The impact of alternative sa methods on EC survey
  data for the euro area
• OECD
• Detection of phase shift in the determination of
  turning points for BCS tendency data after sa

12
Study Group 2003 (contd)

• Main conclusions
• Presence of seasonality
• - More than half of the series do have a seasonal
  component,
• - 15 series could be more suitable for
  smoothing
• Automatic Model identification
• - TS performs better than X-12-REGARIMA
• Direct vs indirect seasonal adjustment
• - Direct method is preferable (less revisions,
  smoothness, etc.)
• - No big differences between TS and X-12-REGARIMA
• Outlier treatment
• - Outliers were found in many cases (about 45 of
  the series)

13
Study Group 2003 (contd)

• Main conclusions (contd)
• Revisions approaches
• model identification and parameter estimation
  every month
• model identification once per year and
  parameter estimation every month
• model identification and parameter estimation
  once a year.
• The last approach is preferable
• The impact of sa and aggregation methods on the
  determination of turning points
• If series have a small irregular component
  alternative methods of sa have little impact on
  the dating of turning points
• If not, TS and X-12 give same results but some
  differences for DAINTIES.
• Direct vs indirect method does not affect turning
  points

14
Study Group 2003 (contd)

• Main conclusions (contd)
• The impact of alternative sa methods on
  relationship between quantitative indicators and
  qualitative variables
• Cross-correlations between quantitative and
  qualitative variables are not affected by
  alternative methods of sa
• Direct vs indirect method does not affect
  cross-correlation between quantitative and
  qualitative variables
• The nature of the seasonal component
• A significant part of the series do not have unit
  roots at ALL seasonal
• frequencies

15
Discussion at last years workshop

• Main conclusions were presented at last years
  workshop
• As an outcome of the discussion at the workshop
  DG ECFIN examined the possibility to publish
  seasonally adjusted data as provided by the
  institutes

16
SA methods across institutes
DG ECFIN has asked all institutes about the
method of SA currently used and on possible
future changes (see Annex)
17
Results

• More than 50 of the institutes do not use SA
• Among those who do, methods differ, sometimes
  even across surveys of the same institute
• X-11/X-12 ARIMA is the method most used
• Some institutes have adopted a mixed strategy,
  using two methods for the same survey
• Some institutes plan to introduce or migrate to
  X-12 Arima or TS in the near future
• 2/3 of the institutes that do not carry out SA do
  NOT intend to introduce SA in the near future

18
Where does this leave us?

• Use of seasonally adjusted data from institutes
  is not possible at this stage
• Is there an alternative method to Dainties which
  retains its advantages (e.g. ease of use) but
  eliminates its disadvantages?

19
Study commissioned by DG ECFIN

• In the 2003 study by DG ECFIN, it was implicitely
  assumed that the seasonal component of BCS (if
  present) was stochastic
• As a result of the 2003 study by Eurostat and
  further analysis by DG ECFIN, a new study has
  been launched which focuses on the nature of the
  seasonal component (deterministic or stochastic)

20
Objectives of the 2004 Study

• Nature of the seasonal component (if present) of
  BCS?
• Is it deterministic or stochastic?
• If deterministic
• Specification of the best appropriate model
• Comparison with Dainties, TS and X-12 Arima
• Effect of proposed method on revisions, outliers
  and aggregation method (direct versus indirect)
• If stochastic
• Test whether seasonal unit roots in all seasonal
  frequencies
• Specification of the best model to remove the
  seasonal component
• Comparison with Dainties, TS and X-12 Arima
• Impact of this specification on revisions,
  outliers and aggregation method ( direct versus
  indirect)

21
Prospective outcomes of the study

• Results from the study should be available in
  early 2005
• Based on these results, DG ECFIN intends to
  provide general guidelines concerning the SA of
  BCS, taking into account
• already existing Eurostat recommendations on SA
  in general and
• the practical implementation of the suggested
  methods in a statistical production process

22
EUROPEAN COMMISSION - DG ECFIN Business and
Consumer Surveys
Thank you very much