Transcript Introduction to the International Family of Classification

Testing seasonal adjustment
with Demetra+
Zavadskaya Oksana
The National Statistical Committee, Republic of Belarus
Check the original time series
In this report, was carried the adjustment of volume indices of gross domestic
product (GDP), calculated to the average quarter of 2005 in 2005 prices
October 2011
Graphs showing the presence of
seasonality
In the original time series a seasonal factor is present
October 2011
The choice of approach and regressors
 The approach TRAMO/SEATS was used
 Pre-defined holidays and national holidays were used
 Our own specification was used :
- Transformation, Function : AUTO
- Calendar effect, Operating days, Type :
Calendar, Belarus
- Details, Operating days:
td2
- The Easter effect, Options:
Pretest
- Modeling ARIMA, Automatic modeling, Enabled : True
October 2011
The applied models
Pre-treatment:
The estimated period: [I-2005 : IV-2010]
Logarithmically transformed series
The effects of operating days (2 variables)
No effects of Easter
Deviating values not found
Decomposition:
Trending: innovation dispersion = 0,3802
Seasonal: innovation dispersion = 0,0000
Irregular: innovation dispersion = 0,1441
Used model type: ARIMA (0,1,1)(0,1,1)
October 2011
Graph of results
Seasonal component in the irregular component is not lost
October 2011
Check on a sliding seasonal factor
Graph of the ratio of seasonality and irregularity in III and IV quarters
October 2011
The main diagnostic quality
Final diagnosis: in general, the results are good, there are the unstable spectral calendar peaks
in the residuals of regarima
Октябрь 2011
Residual seasonal factor
There are no indications of residual seasonal and calendar effects, in time series adjusted
on seasonal fluctuations
Октябрь 2011
Stability of the model
Visual assessment of series allows you to conclude that the model is
stable
October 2011
Analysis of the residuals
Residuals analyzed on randomness, normality and independence
October 2011
Questions
1.
2.
What’s the meaning of «Friedman statistic = -12666373951979500,0000»
in nonparametric tests for seasonal fluctuations in the Friedman test (Friedman
test)
How to interpret the absence of Autoregressive spectral graph in residuals :
October 2011
Questions (continuance)
3. The program did not perform an assessment of nonlinearity. Why?
Analysis of the residuals
4. How to interpret the straight line on the S-I ratio graph?
October 2011
Questions (continuance)
5. How to explain the "zero" in the test for autocorrelation? Should we consider
this as a problem?
October 2011
Problems
 1. Lack of Russian interface
 2. Lack of detailed user's manual of the program in Russian
 3. Lack of detailed step by step instructions for novice on
decoding results
October 2011