Testing seasonal adjustment of the Index of industrial production for 2000-2010, using Demetra+

Testing seasonal adjustment Testing seasonal adjustment of the of the Index of industrial Index of industrial

production for 2000-2010, production for 2000-2010, using Demetra+using Demetra+

Ermurachi GalinaNational Bureau of Statistics, Republic of Moldova

Check the original time series

• Properties of the original time series

• Spectral analysis of the original series

Graphs showing the presence of seasonality and the effect of operating days.

Seasonal adjustment

The approach TRAMO/SEATS was used

Calendar of national holidays was created and used

We started the analysis with the specification RSA4, with change of some options

Used models pre-treatment :

The estimated period : [1-2000 : 12-2010]Pre-processing (Tramo)

Estimation span: [1-2000 : 12-2010]Series has been log-transformedTrading days effects (2 variables)Easter effect detectedNo outliers found

Decomposition

trend. Innovation variance = 0,0845seasonal. Innovation variance = 0,0188irregular. Innovation variance = 0,3407

Dispersion of seasonal and trending components are lower than

fluctuations of components, specifications of which deviate from normal. This means that the obtained stable trending and seasonal components. It can be concluded that the adopted assumption of a canonical decomposition.

Adjustment was applied taking into account national holidays and Easter

Used model type ARIMA (0,1,1)(0,1,1)

Graph of results

Seasonal component is lost in the noise of non-standard component. This means that the number of seasonal variations may be negligible.

Sliding seasonal factors

We noticed unstable and moving seasonal factors.

Апрель 2011

Main quality indicatorsMain results of quality diagnostics

Test for presence of seasonality

The data presented show that there is a moving seasonal component to the 20% level of significance, in the series of industrial production index. Seasonal variations are identified in the original series, but the entire series, nor the last 3 years the series adjusted for seasonal variations, have no residual seasonal fluctuations. The presence of moving seasonally not surprising, given the above described plot ratio of C-H.

Spectral diagnostics

According to the graphics, we can assume that there is no residual seasonal and calendar effects, the residual of the series adjusted for seasonal variations, since the seasonal frequency, and frequency of trading days found no spectral peaks.

Stability of the model

The nearer the point of updates to the red line, the more stable the adjustment. According to the results of this series we have two points that go beyond the deviating values.

Residuals

Residuals are distributed as random normal and independent

Problematic series (if there were any) All the results obtained in the panel

"Diagnostics" shows that they are good (Good), but spectral seasonal peaks are obtained that are uncertain (Uncertain (0,42).

Are these results considered good or not and if not, then what you need to do to fix it?

Questions

More attention should be paid to the results obtained and how to interpret them correctly

Detailed analysis of the residuals

Спасибо за внимание!

E-mail: [email protected]