Press release 8.12.2009
Statistics Finland regards the economic depression as a level shift which influences the picture of the economy obtained from seasonally adjusted and trend time series. The aim is to draw a more truthful picture of the economy and improve the quality of seasonal adjustment. A level shift can be seen in the time series of quarterly national accounts released on Tuesday, 8 December 2009.
Statistic Finland's task is to produce statistics which are comparable with figures from other EU countries. Eurostat, the Statistical Office of the European Communities, issues guidelines and recommendations to ensure achievement of this comparability. Therefore, many decisions concerning the production of statistics are based on guidelines and recommendations issued by Eurostat. This is also the case with seasonal adjustment.
Eurostat recommends that national statistical institutes should use the method called Tramo/Seats for seasonal adjustment. Eurostat has also given guidelines on how to use this method.
Major falls have been evident in the indicators describing Finland's economy since the latter part of 2008. Statistics Finland is changing the treatment of these outlying values in its seasonal adjustment. The change, which is explained in detail below, is initially implemented in quarterly national accounts.
Figures which are seasonally adjusted by using Tramo/Seats are especially uncertain if the data contain abnormal observations, outliers. Therefore, Eurostat expects these outliers to be taken into account in seasonal adjustment.
The depression caused by the financial crisis can be seen as abnormal observations in many economic time series. Experts in seasonal adjustment from Eurostat and the ECB have recently suggested that EU member states should treat these observations as outliers.
Statistics Finland seasonally adjusts its time series with the Tramo/Seats method which can capture three different types of outliers. These outlier types can be differentiated by the way the time series moves back to its original level. In the case of an additive outlier the time series moves immediately back to its normal level. A transitory change is an outlier where the level of the time series changes abruptly but moves gradually back to its original level over the next few reference periods. The third outlier type is a level shift, where the time series moves suddenly to a different level and does not move back over the next few reference periods.
It is very challenging to examine outliers at the end of a time series. To be able to identify the type of outlier one needs to know whether the time series will move back to its original level and, if so, how much time this will take. A level shift cannot be identified until it is seen that the time series does not move back to its original level within the next few observations.
A sharp fall was observed in the source data of GDP in the last quarter of 2008. The strong fall of GDP continued through the first quarter of 2009.
At that time it became clear that those observations were outliers. However, it was impossible to say how the GDP series would evolve in future. Therefore, the outliers of the turn of the year could not be corrected for.
Now, in the third quarter of 2009 it can be seen that the time series has not moved back to its original level after the sharp fall. Instead, the economy has remained at approximately the level of year 2006. Based on this, the sharp fall at the turn of the year can be identified as a level shift.
Taking the level shift into account has an effect on seasonal adjustment and on the interpretation of the economic situation (cf. figure).
Without the level shift the seasonally adjusted time series (grey line) would suggest that the economic downturn started in the second quarter of 2008 and would still continue strongly. The source data support neither of these statements: according to them the strong downturn started in the last quarter of 2008. In addition, the year-on-year changes, i.e. changes from the same quarter of the previous year, have not been as negative in the third quarter of 2009 as they were in the second quarter of 2009.
When the level shift is used in seasonal adjustment (blue), the downturn of the economy can be timed to the last quarter of 2008. According to the seasonally adjusted time series the economy has bottomed out.
Using level shifts has a strong impact on the results of seasonal adjustment. Therefore, thorough analysing of time series is necessary before outliers can be classified as level shifts. The following principles apply to all time series that are seasonally adjusted at Statistics Finland.
Level shifts are not used in seasonal adjustment if no explanation can be found for them in the statistical topic. This means that the direction and timing of a level shift must have a rational interpretation. In addition, the data must provide clear evidence of the existence of a level shift. This is examined by using seasonal adjustment software which checks the significance of the level shift.
Eurostat expects the EU countries to evaluate the quality of seasonal adjustment against diverse statistical tests. An outlier is classified as a level shift if this improves the quality of seasonal adjustment as measured with these quality indicators.
In other words, level shifts are used in seasonal adjustment to get a truthful picture of the economic situation and to improve the quality of seasonal adjustment. The use of level shifts has a strong impact on the results of seasonal adjustment. Therefore, there must be both empirical evidence based on data and intuitive rationale for the use of level shifts in seasonal adjustment. All these conditions hold true for GDP time series.
It must be underlined that an outlier cannot be identified as a negative level shift until it is seen that the time series has not returned to its original level over the next few observations. In the case of GDP, this happened in the third quarter of 2009.
Inquiries: Mr Faiz Alsuhail, +358 9 1734 2921