04-15-2014 06:29 AM
What would be the appropriate way to screen out weak predictors so as to later on fit a state space model?
Is there any other test (more advanced and accurate) besides correlations that would help me select the strongest variables in the case where my dependent as well as my independent variables are of continuous type? (lets say sales as Y and different media spends as X(s)?)
Thank you in advance
04-17-2014 12:55 PM
If you have access to SAS Forecast Server - you may want to use the HPFDIAG procedure as a variable selection tool.
Screening out weak predictors for time series models is a bit more tricky than for static regression models. Cross-correlation is an obvious way to assess predictors.