13-Jul-07

European Statistical System guidelines on seasonal adjustment: a major step towards PEEIs harmonisation

C. Calizzani – G.L. Mazzi – R. Ruggeri CannataEurostat

Quality 2008 , Rome 8 -11 July 2008

ESS guidelines on seasonal adjustment

Introduction (1)

Crucial role in the production process of infra-annual statistics– Reliability– Comparability

Seasonally adjusted data: reference key indicators for analysis and forecasting exercises

Several aspects:– Treatment of calendar effects– Outliers– Temporal and sectoral reconciliation– Revisions policy– Etc.

ESS guidelines on seasonal adjustment

Introduction (2)

Well known tools: – TRAMO-SEATS – Census II X-12 ARIMA– Unobserved components based decomposition

Same seasonal adjustment tool can produce quite different seasonally adjusted results

Need for harmonisation

ESS guidelines on seasonal adjustment

ESS specificities (1)

More than 27 members plus Eurostat– Different characteristics of national statistical systems

– Different level of expertise

– Different internal organisations

Legal acts as the major instrument for harmonisation of statistical production– Rarely giving clear rules for seasonal adjustment

Seasonal adjustment performed on the basis of sectoral and national practices – Lack of comparability

ESS guidelines on seasonal adjustment

ESS specificities (2) European aggregates derived from national data

– Aggregation– Estimation– Aggregation/estimation

Crucial role of harmonisation for the quality of European aggregates

Harmonisation of seasonal adjustment needed– Relevant discrepancies in:

• calendar adjustment • seasonal adjustment• Revisions policies

ESS guidelines on seasonal adjustment

ESS specificities (3)

Several recommendations for the harmonisation of seasonal adjustment practices– ECOFIN Council– Economic and Financial Committee (EFC)– Committee for Monetary, Finance and Balance of payments

statistics (CMFB) Key points:

– High degree of harmonisation of seasonal and calendar adjustment practices for Principal European Economic Indicators (PEEIs) needed

– Convergence of revisions policy for seasonal adjusted data– Improvements on the communication on seasonally and calendar

adjusted data

ESS guidelines on seasonal adjustment

ESS specificities (4) Some already existing guidelines on seasonal

adjustment – U. S. Census Bureau – Statcan– ONS

Synthetic versus detailed guidelines – Complexity of the harmonisation problem

• Sectoral level• Geographical level

Privileging detailed guidelines– Eurostat guidelines 2006 starting point

ESS guidelines on seasonal adjustment

ESS specificities (5) European Statistics Code of Practice: definition of good

practices covering the institutional environment, the statistical processes and its outputs– Principle 7: Sound methodology must underpin quality

statistics. This requires adequate tools, procedures and expertise

– Principle 14: European statistics should be consistent internally, over time and comparable between regions and countries…

– Principle 15: European statistics should be presented in a clear and understandable form, disseminated in a suitable and convenient manner, available and accessible on an impartial basis with supporting metadata and guidance

ESS guidelines on seasonal adjustment

Main characteristics

Sound methodology Completeness Flexibility Pragmatism Clarity User-oriented Transparency of seasonal adjustment practices Expertise development and capacity building

ESS guidelines on seasonal adjustment

Guidelines Table of Contents (1) 0 – SEASONAL ADJUSTMENT BENEFITS AND COSTS 1 - PRE-TREATMENT

– 1.1: Objectives of the pre-treatment of the series– 1.2: Graphical analysis of the series– 1.3: Calendar adjustment

• 1.3.1: Methods for trading/working day adjustment• 1.3.2: Correction for moving holidays• 1.3.3: National and EU/euro area calendars

– 1.4: Outlier detection and correction – 1.5: Model selection – 1.6: Decomposition scheme

2 - SEASONAL ADJUSTMENT– 2.1: Choice of seasonal adjustment approach – 2.2: Consistency between raw and seasonally adjusted data– 2.3: Direct versus indirect approach

• 2.3.1: Direct versus indirect approach: dealing with data from different agencies

ESS guidelines on seasonal adjustment

Guidelines Table of Contents (2) 3 - REVISIONS POLICIES

– 3.1: General revisions policy– 3.2: Concurrent versus current adjustment – 3.3: Horizon for published revisions

4 - QUALITY OF SEASONAL ADJUSTMENT– 4.1: Validation of seasonal adjustment– 4.2: Quality measures for seasonal adjustment – 4.4: Comparing alternative approaches and strategies– 4.5: Metadata template for seasonal adjustment

5 - SPECIFIC ISSUES ON SEASONAL ADJUSTMENT– 5.1: Seasonal adjustment of short time series– 5.2: Treatment of problematic series

6 - DATA PRESENTATION ISSUES– 6.1: Data availability in databases– 6.2: Press releases

ESS guidelines on seasonal adjustment

Chapters’ structure

Chapters subdivided into specific items describing different steps of the seasonal adjustment process

Items presented in a standard structure providing: – Description of the issue– List of options which could be followed to perform the

concerned step– Prioritized list of three alternatives from the most

recommended one to the one to be avoided (A,B, and C)– A synthetic list of main references

Added value: – Conceptual framework and practical implementation steps– Both for experienced users and beginners

ESS guidelines on seasonal adjustment

Example: 2.12 - SEASONAL ADJUSTMENT

2.1 – Choice of seasonal adjustment approach

Description TRAMO-SEATS and X-12-ARIMA are currently the most commonly used seasonal adjustment approaches. TRAMO-SEATS is based on a parametric approach while X-12-ARIMA is based on a non-parametric approach. Structural time series models represent a reasonable alternative, provided they allow for a complete calendar and outlier treatment and include an adequate set of diagnostics. The consistent use of a common set of seasonal adjustment packages will improve transparency and comparability of seasonally adjusted time series across countries.

Options X-12-ARIMA;

TRAMO-SEATS;

Structural time series models.

Alternatives * A) TRAMO-SEATS, X-12-ARIMA together with well-documented and stable interfaces to these tools should be used for seasonal adjustment. The choice between TRAMO-SEATS and X-12-ARIMA can be based on past experience, subjective appreciation and characteristics of the time series. Production tools should be updated on a regular basis after satisfactory testing. Methods and tools versions currently used in data production should be clearly communicated to users. B) Use of structural time series models based on simultaneous representation of the unobserved components of the series. The chosen software has to estimate calendar and outlier effects with diagnostics for all components and effects. For mass data production the chosen software should offer automatic modelling procedures that can reliably identify the presence of the effects mentioned. C) Use of other production tools.

ESS guidelines on seasonal adjustment

Pre-treatment - Key topics

Removal of non-linearity and deterministic effects affecting a proper identification of the seasonal component

Detailed graphical analysis as essential starting point for the detection of all effects

Linearization of the series– Calendar effects– Outliers– Modelling and extrapolating time-series – Identification of ARIMA models

ESS guidelines on seasonal adjustment

Pre-treatment – Main implications

Use of national calendars to improve and better tune calendar adjustment

Estimation of proper calendar effects represented by the deviation of the number of working or trading days from their long-term monthly/quarterly average– Part of calendar effects are seasonal and do not have to

be removed Statistical and economic validation of size and sign of

regression coefficients Accurate identification and correction of outliers by type

– More conservative approach recommended at the end of the series

ESS guidelines on seasonal adjustment

Seasonal adjustment – Key topics

Identification of recommended filters to remove seasonality – TRAMO-SEATS – Census II X-12 ARIMA– Structural time series models

Relationship between non seasonally adjusted data, calendar adjusted data and seasonally adjusted data– Time consistency

How to impose to a set of seasonally adjusted data the same aggregation constraints corresponding to raw data– Direct versus indirect

ESS guidelines on seasonal adjustment

Seasonal adjustment – Main implications

Focus on TRAMO-SEATS and X-12 ARIMA: widely used and most developed methods– Structural models also acceptable if well-defined pre-treatment

module available– Other approaches discarded

No guidance on which method to prefer and why– applying the same method to a given set of related series

No methodological rational in imposing time consistency between raw, calendar and seasonally adjusted data– Strong users’ request for time consistency, especially in some areas

Direct approach to be preferred when component series show similar seasonal patterns, indirect otherwise– Check of residual seasonality– Considering users’ preferences for sectoral and geographical

consistency

ESS guidelines on seasonal adjustment

Revisions policy – Key topics Causes of seasonally adjusted data revisions

– Raw data revisions

– Revisions specific to seasonal and calendar adjustment methods

Need for a general policy for seasonal adjustment– Transparent

– Coherent

– Publicly available

Timing of revisions based on trade-off between precision and accuracy – Current versus concurrent adjustment

Depth of revisions

ESS guidelines on seasonal adjustment

Revisions policy – Main implications

Seasonal adjusted data published according to the scheduling of raw data – Release calendar

Most appropriate strategy for re-identification and re-estimation of parameters and filters based on:– Number of periods revised in raw data– Stability of the seasonal component– Presence of benchmarking constraints

Depth of revisions should take into account:– Depth of raw data revisions– Number of periods needed to stabilise the seasonal filters

results

ESS guidelines on seasonal adjustment

Quality of seasonal adjustment – Key topics

Validation of seasonal adjusted data before their dissemination – Check for absence of residual seasonal and calendar

effects– Stability and reliability of estimates

Definition of appropriate quality measures to assess the quality of seasonal adjustment

Defining a common set of quality measures– Comparing alternative seasonal adjustment methods– Comparing alternative seasonal adjustment strategies– Documenting all seasonal adjustment steps

ESS guidelines on seasonal adjustment

Quality of seasonal adjustment – Main implications

Validation of results by using a large set of quality measures – Specific measures to each method– Additional measures – Detailed graphical analysis

Identification of a common set of quality measures – Helping users in comparative analysis

• TRAMO-SEATS versus X-12 ARIMA• Direct versus indirect

Definition of an harmonised metadata template for seasonal and calendar adjustment

ESS guidelines on seasonal adjustment

Specific issues on seasonal adjustment – Key topics

Overall quality of seasonal adjustment affected by:– Length of time-series– Presence of strange features

• Non-linearity• Outliers• Volatility

Special treatment needed – Particular attention to key indicators

ESS guidelines on seasonal adjustment

Specific issues on seasonal adjustment – Main implications

No seasonal adjustment for series shorter than 3 years

Awareness of the instability of seasonally adjusted data for series of 3 - 7 years length – Assessing a specific strategy for re-identification and re-

estimation of filters and parameters– Users information

Case by case approach for series with high degree of irregularity– Use of standard tools

ESS guidelines on seasonal adjustment

Conclusions (1)

Major step towards the harmonisation of PEEIs production

Enhancing Quality– Improvement of comparability– Robustness and reliability of European aggregates– Transparency

Promoting best practices Great contribution to the international methodological

and empirical debate Largely supported inside and outside European Union

ESS guidelines on seasonal adjustment

Conclusions (2)

Efforts required to Eurostat production units and Members States to become compliant with the guidelines– Based on voluntary commitment– Implementation plan to be developed

Monitoring strategy – Regular reporting to institutional bodies– Collecting information inside and outside Eurostat on

seasonal adjustment practices• Metadata template

ESS guidelines on seasonal adjustment

Thank you for your attention!