Seasonal adjustment with Demetra

1
Seasonal adjustment with Demetra

• Ajalov Toghrul,
• State Statistical Committee of the
• Republic of Azerbaijan

2
Check the original time series

• The duration of the time series (1/2000 -
  12/2010)
• Time series used were retail trade indices
• Base year 2005 100

3
Original data in graphs

• The original data includes seasonality

4
The choice of approach and predictors

• Method used, TRAMO/SEATS
• National holidays were defined
• Selected specification was RSA 5

5
The model applied

• Pretreatment
• Estimation span (1-200012-2010)
• The effect of operating days is not observed
• 6 outliers identified
• Innovation
• Trend - innovation variance 0.0024
• Seasonal - innovation variance 0.4094
• Irregular - innovation variance 0.0254
• Type of model used ARIMA (2,1,0) (1,1,0)
• Deviating values

Value Std error T-Stat P-value
AO12-2007 -0,0348 0,0038 -9,14 0,0000
AO4-2009 -0,0367 0,0038 -9,68 0,0000
AO7-2005 -0,0258 0,0035 -7,30 0,0000
AO10-2001 -0,0209 0,0039 -5,36 0,0000
LS1-2009 -0,0199 0,0043 -4,66 0,0000
AO11-2002 -0,0131 0,0036 -3,60 0,0005
6
Graphs of the results

• Seasonal component is not lost in the irregular
  component

7
Check for a sliding seasonal factor

• In December, highly volatile seasonal variation
  present

8
The main quality diagnostic

• Referring to the estimated values ??of we can
  determine the quality of the results
• The overall summary quality diagnostics are good

9
Residual seasonal factors

• There are no peaks in the seasonal and trading
  day frequencies, this indicates that there is no
  residual seasonality in the results

10
Model stability

• Regardless the four points beyond the red line
  you can come to the conclusion that the model is
  stable

11
Residuals

• The residuals are
• distributed
• as random,
• normal and
• independent

12
Questions

• Innovation
• Trend - innovation variance 0.0024
• Seasonal - innovation variance 0.4094
• Irregular - innovation variance 0.0254
• The innovation variance of the irregular
  component is lower than the variance of the
  seasonal component, in this case are the results
  questionable?

13
Questions
Why indicators of kurtosis and normality are
highlighted in yellow? Does it mean that there is
an asymmetry in the distribution of residual
values???
14
Questions

• What if I get undefined, erroneous diagnosis or
  severe final result? In this case, should we
  revise source data series or what can be done?
• Do diverging values influence the final results?

15

• Thank you for your attention!