Testing Seasonal Adjustment with Demetra

1
Testing Seasonal Adjustment with Demetra

• Levan Gogoberishvili
• Head of National Accounts Division
• National Statistics Office of Georgia

2
The original series

• GDP by production approach in constant prices
  (1996, base year)
• Quarterly time series starting with 1st quarter
  of 2000.
• GDP by production approach is calculated by
  several types of economic activities. Production
  of such activities like Agriculture, Construction
  services or Electricity supply and other services
  have seasonal features.

3
Methods and Specifications used

• X12 method were used for seasonal adjustments of
  following quarterly series
• GVA of Agriculture in constant prices
• GVA of Industry in constant prices
• GVA of Construction in constant prices
• Total GDP at constant prices
• We received the most stable results (according
  to automatic quality checks made by software
  itself) by using automatic RSA1 specification.

4
Pre-processing and Diagnostics
5
Graph of the results
6
Check for moving seasonality
7
Main Quality Diagnostics
8
Residual seasonality
9
Stability of model
10
Residuals
11
Some problematic series

• There is a problem of spectral peaks in the time
  series of industry.
• We explained it as a result of aggregation
  (Industry Mining and quarrying Manufacturing
  Electricity, Gas and Water production and
  distribution)

12
Possibilities to publish the results

• At this stage following seasonally adjusted
  time series are published
• Industrial production Indices (Base year 2011,
  quarterly)
• GDP in constant, 2003 prices (quarterly)
• Industrial production Indices are available on
  Geostat website and also in annual publication
  National Accounts of Georgia
• Seasonally adjusted GDP are not published on
  web-site in order to avoid confusion of some
  users. However, this time series are included in
  annual publication. Some users with advanced
  knowledge of economics use this time series for
  analysis. There are some users, which prefer to
  receive original time series and do seasonal
  adjustments themselves.

13
Conclusions

• Before the first meeteng we had used Demetra 2.0
  software for seasonal adjustment of IPI time
  series.
• Now we use Demetra version. Despite the method
  of Tramo/Seats is the same, the results are a
  little different because of changes in
  specification.
• For GDP time series we started to use X12 method.
  Some of the results are shown in this
  presentation.
• The basic problem for us is, that our time series
  are quarterly and we do not take into account
  Easter and Trade days effects.
• We hope that, experts in the second seminar will
  additionally prepare specific recommendations for
  countries with only quarterly time series.

14
Conclusions

• Another problem is that, some users do not
  understand the meaning of seasonal adjustments.
  If we publish both the original and seasonally
  adjusted GDP time series at the same time, it
  might not be understood properly.
• We are planning to prepare some metadata and
  explanations of seasonal adjustments for users.
• On the other hand we will continue work for
  improving SA work. For this purpose the planned
  workshop will be very useful.

15

• Thank you!