Time series- Seasonal ARIMA model

[Monali Parab]

Hi All,

 

Can anyone help me on Seasonal Arima model for Time series?

 

I am interested in knowing how we can decide the model parameter & what other things need to look out in order to improve the model?

Also for the seasonal decomposition, data requirement for the additive model & for the multiplicative model. i.e. how shall i decide whether go for additive model or the multiplicative model?

 

I am using SPSS 21 Forecasting module for the same.

 

Thanks in advanced.

 
Thanks,
Monali

[David Marso]

What is the nature of the data?
How many time points?
What do the ACF and PACF plots look like?
---

Monali Parab wrote

Hi All,

Can anyone help me on Seasonal Arima model for Time series?

I am interested in knowing how we can decide the model parameter & what
other things need to look out in order to improve the model?
Also for the seasonal decomposition, data requirement for the additive
model & for the multiplicative model. i.e. how shall i decide whether go
for additive model or the multiplicative model?

I am using SPSS 21 Forecasting module for the same.

Thanks in advanced.



Thanks,
Monali

[David Marso]

If you can follow this then there is a basis for further discussion.  If not then you need to do further reading to grasp basics.
http://people.duke.edu/~rnau/seasarim.htm
 

Monali Parab wrote

Hi All,

Can anyone help me on Seasonal Arima model for Time series?

I am interested in knowing how we can decide the model parameter & what
other things need to look out in order to improve the model?
Also for the seasonal decomposition, data requirement for the additive
model & for the multiplicative model. i.e. how shall i decide whether go
for additive model or the multiplicative model?

I am using SPSS 21 Forecasting module for the same.

Thanks in advanced.



Thanks,
Monali

[tomdireill]

Monali,

You will also need to be on the lookout for outliers.  You should be searching for pulses outliers, level shifts, seasonal pulses and changes in trend.

These are not often discussed and when ignored can make you think you should be differencing when in fact if you had just built a level shift you would have the correct model.

Tom

[David Marso]

<RANT>
FWIW:  Time series analysis is a non trivial, relatively advanced topic in the statistical sciences.  Once upon a time in a previous century when I took a full semester course in ARIMA models, transfer functions and the like were relegated to the last few weeks of study AFTER the less complicated issues were well understood.
Before taking this course one had to have about 4 prerequisite courses under their belts including a full semester in GLM/Regression/ANOVA which followed introductory and intermediate courses.
Somehow it seems that statistics has become so much 'simpler', that anyone can push a button and be a 'statistician'... Real Stats Real Easy... Caveat Emptor!!!  GIGO...
This applies even more to the wanna be SEM aspirants... Draw a pretty picture and BOOM!!!
If you think it is expensive to hire a professional wait until you engage a novice!
</RANT>
---

tomdireill wrote

Monali,

You will also need to be on the lookout for outliers.  You should be searching for pulses outliers, level shifts, seasonal pulses and changes in trend.

These are not often discussed and when ignored can make you think you should be differencing when in fact if you had just built a level shift you would have the correct model.

Tom