HOW TO USE ucm OUTPUT TO SCORE NEW DATASETS WITH SEASONALITY

[lucky66]

Hi all, 

I have a time series data and trying to use proc ucm to generate some output, but I am not sure how to implement those output for seasonality into new datasets.  The output related to seasonality is like this: Anyone knows how to implement seasonality into to new data forecasting/scoring? Thank you. 

Significance Analysis of Components (Based
on the Final State)
Component	DF	Chi-Square	Pr > ChiSq
Irregular	1	0.3	0.2
Season	4	20	<.0001
Summary of Seasons
Name	Type	Season Length	Error Variance
Season	TRIG	12	0

and SAS code like this:

proc ucm data=a;
	id date interval=month;
	irregular;
	season length=12 variance=0 noest type=trig keeph=1 2 plot=smooth print=smooth;
	model b=var1 var2 var3;
	estimate plot=wn plot=panel plot=model outtest=outdata;
run;

[lucky66]

I have a time series data and trying to use proc ucm to generate some output, but I am not sure how to implement those output for seasonality into new datasets.  The output related to seasonality is like this: Anyone knows how to implement seasonality into to new data forecasting/scoring? Thank you. 

Significance Analysis of Components (Based
on the Final State)
Component	DF	Chi-Square	Pr > ChiSq
Irregular	1	0.3	0.2
Season	4	20	<.0001
Summary of Seasons
Name	Type	Season Length	Error Variance
Season	TRIG	12	0

and SAS code like this:

proc ucm data=a;
	id date interval=month;
	irregular;
	season length=12 variance=0 noest type=trig keeph=1 2 plot=smooth print=smooth;
	model b=var1 var2 var3;
	estimate plot=wn plot=panel plot=model outtest=outdata;
run;

[rselukar]

I don't fully follow your question.  In any event, I am offering an answer based
on what I think you are asking.  The suggested answer is only an adhoc process
and I don't have personal experience of using this technique.

I have heard of some cases where seasonal indices of one series are used as a
proxy for seasonal indices of other series, presumably because of the lack of sufficient
data in the other series.  Of course, this would make sense only if the two series
are similar in some sense, e.g., monthly sales of an item of two different brands.  Even
in such cases, because of the differences in the scales of mean patterns of the
two series it is important to model the series in the log scale (multiplicative form in
the original scale)
so that the seasonal indices have the interpretation of percentage change.  Assuming that
all these conditions are met, you could use this strategy as follows:

1. Fit a seasonal model to the log-transformed version of series 1 (the
one that has sufficient data).  Read off the smoothed seasonal component (the s_season
column) from the OUTFOR= data set in the FORECAST statement.
2. Fit a non-seasonal model to the log-transformed version of series 2.  Add the
smoothed seasonal component of the first series to the forecasts of the second series.
This would adjust the forecasts of the second series according to the seasonal
pattern of the first series.

Hope this helps.