PROC VARMAX model is not full rank

[itslarajean]

Dear All,

I am attempting to run a VAR model with the PROC VARMAX function with my dataset. My data are all modulus transformed and have two dummy variables and I want to output 1-4 lags. I noticed that some failed at the second lag onwards with the error message: "The model is not full rank. The VARMAX procedure stopped processing further steps." even without the dummy variables.

For example, I have quarterly data from March 2007 to December 2021 and I have 2 variables (including dependent without dummy) I will be able to get output up to 4 lags, but after including the two dummies I can only get 1 lag VAR model with the above error message for the remaining lags.

The second dummy is the subset of the first dummy, i.e. dummy #2 can be 0 or 1 when dummy #1 is 1

Appreciate any kind assistance, thank you.

[sbxkoenk]

Hello,

It seems to me you have a redundant variable (or variableS), no?

Using the default METHOD=ML, you probably have to comment out the XLAG= option and reduce the P= option from higher to lower in order to estimate the model error-free. Correct?

You can also try the newer CML estimation method. 
The parameter estimates associated with redundant terms generated by the XLAG= option are then set to 0, I believe.  

However, you still might need to omit some redundant regressors from the MODEL statement to be able to submit error-free. 

Best,

Koen

[PaigeMiller]

---

@sbxkoenk wrote:

It seems to me you have a redundant variable (or variableS), no?

---

Or much less likely, but also a possible cause, a linear combination of some of the x-variables is perfectly correlated with a linear combination of other x-variables. Example: if x1+x2 is always exactly equal to x3–7*x4, the same error message would likely appear.