EViews.com

hi.
i want to make a research about determinants of net interest margins in the six banks for the period (2006/2011), and i have 1 dependent variable (nim) and 14 independent variables( 7 of them identify by Cross Section Identifiers - e.g banks).
and when i make a pool estimation to all of the 14 independent variables. i receive a message " near sigular matrix". so as i understand from searching for this problem, that is my independent variables are collinearity together, and when i make a pool estimation of the 12 independent variables, the result is That majority of variables are insignificant, but the F statistics is significant.
According to that i make a correlation matrix between a variables to find this correlation, but it is appear also among the same variable like each bank for the same variable, and i found more than one had a correlation above 0.80, so my question is that correlation matrix make truely image of the correlation?, or there is other way which can i make matrix correlation without appear results by Cross Section Identifiers?
my second question is how can i make less collinearity between variables, without remove the variables, becuase this variables are essential to be in estimated model?
i hope to find solutions for my problem
thanks in advance.

Have you tried searching the forum for similar discussions?

yes, but didnt find my request.
if any one know any post or info can be useful to me?
thanks

You might want to post a version of the workfile that includes the equation object you are estimating that is giving you a problem.

thanks for response,
i am using a pool object not a equation object.
but my theoretic equation should be:
nim = c(1) + c(2) exp + c(3) equ + c(4) crr + c(5) size + c(6) lode + c(7) stat + c(8) des + c(9) gdp + c(10) inf + c(11) exch + c(12) inrate + c(13) stmd + c(14) sbsize + c(15) conc
i hope that i can give your request.

thanks for response,
i am using a pool object not a equation object.
but my theoretic equation should be:
nim = c(1) + c(2) exp + c(3) equ + c(4) crr + c(5) size + c(6) lode + c(7) stat + c(8) des + c(9) gdp + c(10) inf + c(11) exch + c(12) inrate + c(13) stmd + c(14) sbsize + c(15) conc
i hope that i can give your request.

Oh, now I see.
I think the problem is that you have 6 common variables and only six periods. So you really do have multicollinearity. You probably need more data.

this is the problem.
i haven't more data because my sample is modernistic, and there is no more data available.
is there another way except add more data?
any way you make me a favor.
thanks

In principle you might be able to put restrictions on some of the coefficients in your model. In practice, my guess is that you are out of luck. (If you can get quarterly rather than annual data, that might help.)

that's it, i thing the quarterly data is my solution.
thank you again