Programming the statistical procedures from SAS

time series with proc reg

Reply
Contributor
Posts: 60

time series with proc reg

Hi,

I am new in time series modeling.

Unfortunately I don't have proc autoreg and I am trying to fit a time series with proc reg.

In my model I am using lag of the dependent and some other x's.
The is no collinearity, dw=1.8 but I still feel that I need to improve the model because the RMSE in my validation is a bit high.

1) I want to add another lag of y, so I will have

y= B0 +B1lag1(y) + B2 lag2(y) +B3X1 +B4X2

What about the Problem of multicolinearity in this case? between lag1(y) and
lag2(y) ? should I be concerned about that? What should I look for in that case.

2) Also, I use durbin watson because in proc reg the godfrey test is not available, someone have a code that imitate the godfrey test? or can pinpoint another way to check autocorellation without proc autoreg because i don't have it in my package.

Hope I will get some answers Smiley Happy
Super Contributor
Super Contributor
Posts: 365

Re: time series with proc reg

Hello SASUser1000,

"What about the Problem of multicolinearity in this case? between lag1(y) and
lag2(y) ? should I be concerned about that? What should I look for in that case"

proc REG has special options to test collinearity, see "Collinearity Diagnostics" topic in SAS help for proc REG.

model y=lag1y+lag2y+X1+X2/ tol vif collin;

switches on the diagnostics.

Sincerely,
SPR
Contributor
Posts: 60

Re: time series with proc reg

Hi SPR,

I know about VIF, I just wanted to know how to deal with the multicollinearity between two lag of y lag1 and lag2? What can I do to reduce the multicollinearity in this case, because the multicolinearity between 2 lags of Y will always be there. How to correct for it?

Best,
Shira
Contributor
Posts: 26

Re: time series with proc reg

You can't do this properly with Proc Reg. You need Proc arima to address the auto correlation between adjacent observations.
Ask a Question
Discussion stats
  • 3 replies
  • 295 views
  • 0 likes
  • 3 in conversation