You won't be able to run the test with 11 variables if your number of observations is insufficient. If you take those data on a yearly basis, you'll only get 6 observations for all of the variables, which isn't enough to run the test and will result in a "Near Singular Matrix" error each time you try to estimate the model. You could either try increasing the number of observations or reducing the number of variables.

Hi, I will help you out.

You are having the "near singular matrix" problem because:

the number of variables (13) in the model are more than the number of cross sections (12 banks).

Solution:

1. Reduce the control variables to about 5 variable.

2. You can use Principal Component Analysis to determine the CV to drop from the model.

3. Ensure that number of variables < number of cross-section (banks)

4. Re-estimate the model.

**Problem solved**

Technically speaking, to get estimates, regression requires the computation of the inverse of the matrix: (XTranspose*X)^-1. Inverse involves a division. But as we know one cannot divide by zero, so you get the error you noted.

Some suggestions to get around the problem s reduce some of the independent variables to one index through factor or Principal Component analysis. Alternatively, increase the number of observations. [Note also how multi-collinearity of the independent variables can make you thing that you have more independent variables than the number of observations]

It's possible there is command commission . You need to check that.

Also, it could be the way you specified your variables and how it was entered and commanded in the system.

Likewise, estimation has steps and stages, if one is skipped, near singular error can come up. For instance, you ARDL and you are jumping Lag Selection ...etc

I presume you work with E-Views.

There are two possible reasons: 1) Perfect Collinearity; 2) Insufficient degrees of freedom.

P.S. If you address the first problem (perfect collinearity) and you work with other software (Stata, SPSS), instead of error message, the software will simply omit one of the explanatory variables.

One variable may be explaining the other. Or some related variables are missing

This happens frequently when there is insufficient length of data.