How to perform a What-If Analysis in SAS/ETS

[LawrenceYu01]

Hi,

I want to see how sensitive of each variable to a response variable in percentage for a time series model generated by an ARIMA procedure in SAS/ETS, for instance, I have a model as follows:

Y(1) = AR(1) + MA(1) + aX + bZ + cR,

where Y(1) is a first-order difference term of the response variable, AR(1) is a first-order autocorrelation term,  MA(1) is a first-order moving average term, and X, Z, R are the predictor variables in the model.

Each of them has its own coefficient, and I want to see the percentage change of the response variable while each predictor variable changes for 1%, for instance,  

Variable Sensitivity- Impact of One Point Change on Y(1)

Predictor Variable	Sensitivity
AR(1)	30%
MA(1)	1.5%
X	-19%
Z	-11%
R	-30%

How can I generate this kind of table for a time series model in SAS/ETS?

Thank you for your help!

Lawrence

[user24feb · Accepted solution]

Not sure, if I understood you correctly:

ODS Output ParameterEstimates=Est;
Proc ARIMA Data=A;
  Identify Var=x CrossCorr=(y z r);
  Estimate p=1 q=1 Input=(y z r);
Run;
ODS Output Close;

Data Sensitivity1Pct (Keep=Parameter Variable Sensitivity);
  Set Est;
  Sensitivity=Estimate*0.01;
Run;

[LawrenceYu01]

Thank you!

After you got the data set Sensitivity1Pct how do you estimate the effect of a change of an independent variable on the dependent variable while keeping all the other terms fixed?

Thank you for your help!