Transcript Introduction to the International Family of Classification

Testing Seasonal
Adjustment with
Demetra+
Ariunbold Shagdar
National Statistical Office, Mongolia
April 2011
The original series, real GDP by 2005
price
• We used the data of GDP (quarterly nominal
and real GDP at 2005 constant price).
– accuracy,
– From 2000 to 2011, by quarter
– quality of production methods,
– consistency of time series
– issues to be improved in the future
Nov 2011
Real GDP
Yes, seasonality is
Nov 2011
Describe the chosen approach
and regressors
• I used TRAMO/SEATS
• Didn’t choose some predefined
holidays or national holidays
 I was selected automatic
specification(RSA1)
Nov 2011
Models applied
• Give information about pre-processing:
– the estimation time span used,
– (if) applied corrections for trading days and
Easter,
– type of applied ARIMA model (p,d,q)(P,D,Q)s
– the dates and types of outliers as well as
– distribution of residuals
Nov 2011
Graph of the results
The seasonal component is almost lost in the irregular.
Nov 2011
Check for moving seasonality
Forth quarter where the moving seasonality is quite evident.
Nov 2011
Main Quality Diagnostics
• The result of the test is good
(except the spectral td
peaks/spectral analysis and
regarima residuals/)
Nov 2011
Diagnostics summary was good. The definition(0.000) and
annual totals(0.008) were very close to zero.
In series, there may not be peaks at seasonal or trading day
frequencies.
visual spectral analysis
spectral seas peaks: Good
spectral td peaks: Bad
April 2011
The result of the remained test was good.
residual seasonality
on sa: Good (0.734)
on sa (last 3 years): Good (0.983)
on irregular: Good (0.638)
outliers
number of outliers: Good (0.023)
seats
seas variance: Good (0.779)
irregular variance: Good (0.534)
seas/irr cross-correlation: Good (0.419)
April 2011
Residual seasonality
There is some residual seasonality after adjustment.
Nov 2011
Stability of model
• The revision dots are to the red line not much closer,
the model is relative stable after the adjustment.
Nov 2011
Residuals
• Residuals almost follow the
normal distribution. But we
have some problems.
• They are random.
Nov 2011
Assess possibilities to publish
the results
• I think so, it is possible to publish, but we
need improve in the future.
Nov 2011
Conclusions
• Training course on seasonal adjustment
would be an important tool for improving the
quality of statistics by fostering the
exchange of good practices.
Nov 2011
• Things to consider in the future
– How improve the seasonally adjusted data?
– Trading day and Holiday effect.
• Problems to solve
- Didn’t specified the calendars.
- Can’t distinguished the regressors.
- If have the bad results then how adjust
the time series?
• Questions to the trainers in workshop II
April 2011