Presentation is loading. Please wait.

Presentation is loading. Please wait.

Towards a seasonal adjustment and a revision policy Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February.

Similar presentations


Presentation on theme: "Towards a seasonal adjustment and a revision policy Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February."— Presentation transcript:

1 Towards a seasonal adjustment and a revision policy Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February 2012, Ankara, Turkey

2 UNECE Statistical Division Slide 2February 2012 Overview  How to organize seasonal adjustment?  Designing a policy  Contents of the Policy  ESS Guidelines on Seasonal Adjustment  Examples of policies and changes in them  Revision policy is an essential part  Examples of revision policies

3 UNECE Statistical Division Slide 3February 2012 Delegated or Centralised? 1) Production units perform seasonal adjustment to their own data  How to ensure sufficient knowledge?  How to avoid mistakes?  How to ensure consistency between statistics? 2) One methodology unit performs all seasonal adjustments that are published  How to manage all series in the hectic schedule?  How to ensure knowledge of industry specific issues, such as causes of outliers? 3) A policy guides the production units and directs them into cooperation with methodology experts

4 UNECE Statistical Division Slide 4February 2012 Reflect National Conditions  The policy cannot be exactly the same for all offices/countries  Find out about users’ preferences  Consider resources available in your office Staff time Computer resources Release schedules  Consider advantages and disadvantages of seasonal adjustment Prepare to face them in the policy

5 UNECE Statistical Division Slide 5February 2012 Prepare a Policy in Stages  Learn and gather experience: Test seasonal adjustment comprehensively Involve colleagues of other statistical areas in seasonal adjustment if useful for their statistics Study international guidelines  Define the basic choices first : Software and method, timing of revisions, release and metadata guidelines  Expand the policy later: Guidelines for problematic series, breaks in time series, times of economic uncertainty

6 UNECE Statistical Division Slide 6February 2012 Contents of the Policy  Method and software choice for seasonal adjustment, dissemination and storage  Methods and timing of re-analysis and revisions  Means of aggregation of series  Treatment of outliers  Requirements for documentation both internal and for users  Guidelines for releasing seasonally adjusted

7 UNECE Statistical Division Slide 7February 2012 ESS Guidelines on Seasonal Adjustment  Helps define the seasonal adjustment policy  Following them improves international comparability  Gives tips for alternative methods in: Outlier detection Calendar adjustment and moving holidays Seasonal adjustment approach Consistency between raw and adjusted data Indirect vs. direct aggregation Revision of seasonally adjusted data Quality measures  Touches also the issues of more problematic series

8 UNECE Statistical Division Slide 8February 2012 Statistics Canada – SA Policy  Scope and purpose  Method chosen  Principles of seasonal adjustment  Seasonal adjustment guidelines  Quality indicators  References http://www.statcan.gc.ca/pub/12-539-x/2009001/seasonal-saisonnal-eng.htm

9 UNECE Statistical Division Slide 9February 2012 Statistics Canada – Guidelines  Seasonality needs to be identifiable for adjustment  No residual seasonality in the adjusted data  10 to 15 years of data ideal, 5 years minimum  RegARIMA model to extrapolate the series to reduce revisions  Options reviewed periodically - not in-between Factors and the regARIMA model parameters recomputed every time Exceptions only when the most recent observations have been historically subjected to large revisions > forecasted factors  Aggregate (direct or indirect) checked for residual seasonality  Revisions published according to a official revision policy  Month-to-month rates computed on seasonally adjusted data Use with caution if the time series has high volatility  Year on year same-month rates computed on calendar adjusted data, or, in absence of calendar effects, on raw data.  Users have access to the historical raw series, seasonally adjusted and, upon request, to the adjustment options

10 UNECE Statistical Division Slide 10February 2012 Statistics Finland – SA Policy  What is seasonal variation and why should it be removed?  What are the components of a time series?  Functioning principles of the method applied  All forecasts contain statistical uncertainties!  Seasonal adjustment practices applied Models are kept fixed for one year but parameters of them are re-estimated in each calculation round Models used checked once a year Details of the models are freely available to anybody http://www.tilastokeskus.fi/til/tramo_seats_en.html

11 UNECE Statistical Division Slide 11February 2012 Direct or Indirect Aggregation  Direct approach means that the aggregate time series are seasonally adjusted independently  Indirect approach - by aggregating the seasonally adjusted series of the component time series by using a weighting scheme  Direct approach preferred for transparency and accuracy Especially if component series show similar seasonal patterns  Indirect approach may be preferred when components show significantly differing seasonal patterns Useful in addressing strong user requirements for consistency Presence of residual seasonality needs to be monitored carefully

12 UNECE Statistical Division Slide 12February 2012 Bank of England – Change in Policy  Started deriving quarterly series from the monthly seasonally adjusted series in 2007 – and stopped separate adjustments  Informed the users with a brief article: Explained the background Reasons behind the change:  Users were confused: M and Q series did not match  Deriving Q from M in line with international best practice  No need to review the Q series separately – less resources Effects on the data Implementation http://www.bankofengland.co.uk/statistics/ms/articles/art2apr07.pdf

13 UNECE Statistical Division Slide 13February 2012 Bank of England – Change in Policy Effect of the change in policy on the annual growth rate of household sector

14 UNECE Statistical Division Slide 14February 2012 Revision policy - OECD/Eurostat Guidelines  Provides the users with the necessary information to cope with revisions: Defines a predetermined schedule for revisions Is reasonably stable from year to year Is transparent Gives advance notice of larger revisions due to conceptual or methodology changes Offers adequate documentation of revisions  Carry revisions back several years to give consistent time series

15 UNECE Statistical Division Slide 15February 2012 Documentation on Revisions Such documentation should include:  Clear identification of preliminary (or provisional) data and revised data  Advance notice of major changes in concepts, definitions, and classification and in statistical methods  The sources of revision explained  Information on breaks in series when consistent series cannot be constructed  Information on the size of possible future revisions based on past history

16 UNECE Statistical Division Slide 16February 2012 Size of the Likely Revisions  Information to judge reliability and accuracy  Do periodic analyses of revisions Investigate the sources of revision from earlier estimates Make statistical measures of the revisions  Publish the historical revision data for major aggregates

17 UNECE Statistical Division Slide 17February 2012 Canadian System of National Accounts Revision Policy  Status of data clearly indicated – preliminary / final  Revisions are carried out regularly To incorporate current information from censuses, annual surveys, administrative sources, public accounts, etc. To implement improved estimation methods  Number of revision times per year  The period open for revision – e.g. four years  Historical revisions conducted periodically To improve estimation methods To introduce conceptual and classification changes To revise data to be in line with new international standards  Dates for revision schedule for the next year  Approximate historical revisions of the aggregates http://www.statcan.gc.ca/pub/13-605-x/2011001/article/11414-eng.htm

18 UNECE Statistical Division Slide 18February 2012 Statistics Finland – Level Shift  Started to treat the economic slowdown as a level shift The aim is to improve the quality of seasonal adjustment The crisis can be seen as abnormal observations Eurostat and ECB suggest treating the slowdown as outliers

19 UNECE Statistical Division Slide 19February 2012 Statistics Finland – Level Shift Level shift shows a sharper change in the series

20 UNECE Statistical Division Slide 20February 2012 Revision Policy for Seasonal Adjustment  The policy should address: Select methods for refreshing the seasonally adjusted data Set the timing for refreshing the adjusted data Define the time period over which the raw and the seasonally adjusted data will be revised Convey the relative size of revisions of the seasonally adjusted data and the main causes of revisions Set the timing of publication of revisions to the seasonally adjusted data and publication of the revisions to the raw data

21 UNECE Statistical Division Slide 21February 2012 Methods for Refreshing the Seasonally Adjusted Data  The quality of forecasts used in seasonal adjustment increases with the frequency of updates A trade-off between the cost and the quality Very frequent updates of the seasonal model could also lead to weaker stability of results and revisions in opposing directions  Current adjustment strategy minimizes the frequency of revision - concurrent generates the most accurate data but will lead to more revisions

22 UNECE Statistical Division Slide 22February 2012 Select Between the Alternative Strategies  Balanced alternatives may provide better quality of adjustment  Partial concurrent adjustment is widely used Keeps the model, filters, outliers and calendar regressors fixed until the annual or biannual re-identification Re-identifies parameters and factors every time new or revised data become available  ESS Guidelines suggests using balanced options  But if you find a problem between the updates, it should be promptly corrected  The choice depends on the properties of the series For series shorter than seven years, re-identification could be done more often, for example twice a year

23 UNECE Statistical Division Slide 23February 2012 Balance Between Accuracy and Stability  If a different model is selected in the annual update, examine the diagnostics to find out whether it‘s notably better than the previous one  Consider the time period over which the results are revised A full revision from the beginning of the series, promotes a methodically uniform treatment Some question whether a new figure contains relevant information for revisions in the historical seasonal pattern  Some offices limit the period of revision of the seasonally adjusted data to a period that is about four years longer than the revision period for the original data

24 UNECE Statistical Division Slide 24February 2012 Publishing Revisions to the Seasonally Adjusted Series  In general publish revisions to the seasonally adjusted series at the same time as you add new observations  Link a change of the seasonal adjustment method to other methodological revisions of statistics, such as changes of base year or the economic activity classification  Give advance information about the forthcoming methodological changes  To correct notable mistakes additional release may be needed


Download ppt "Towards a seasonal adjustment and a revision policy Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February."

Similar presentations


Ads by Google