Mixed Data Sampling "Near Singular Matrix"

[Stegmann]

Hi,

I am trying to estimate the MIDAS models of clements and galvao (2008) in order to estimate GDP growth using the leading indicators.
The leading indicator I am using to estimate gdp growth is new building permits and the equation i am trying to estimate is equation 3 in the attached page; with m=3, K=5 and h=1.
My code to estimate this in eviews is:

ls gdpg c(3) c(5)*gdpg(-1) c(4)*((1/(exp(c(1)+c(2))+exp((2*c(1))+(4*c(2)))+exp((3*c(1))+(9*c(2)))+exp((4*c(1))+(16*c(2)))+exp((5*c(1))+(25*c(2)))))*exp(c(1)+c(2))*(permitg(-1)-c(5)*permitg(-2))+(1/(exp(c(1)+c(2))+exp((2*c(1))+(4*c(2)))+exp((3*c(1))+(9*c(2)))+exp((4*c(1))+(16*c(2)))+exp((5*c(1))+(25*c(2)))))*exp((2*c(1))+(4*c(2)))*(permitg1(-1)-c(5)*permitg1(-2))+(1/(exp(c(1)+c(2))+exp((2*c(1))+(4*c(2)))+exp((3*c(1))+(9*c(2)))+exp((4*c(1))+(16*c(2)))+exp((5*c(1))+(25*c(2)))))*exp((3*c(1))+(9*c(2)))*(permitg2(-1)-c(5)*permitg2(-2))+(1/(exp(c(1)+c(2))+exp((2*c(1))+(4*c(2)))+exp((3*c(1))+(9*c(2)))+exp((4*c(1))+(16*c(2)))+exp((5*c(1))+(25*c(2)))))*exp((4*c(1))+(16*c(2)))*(permitg(-2)-c(5)*permitg(-3))+(1/(exp(c(1)+c(2))+exp((2*c(1))+(4*c(2)))+exp((3*c(1))+(9*c(2)))+exp((4*c(1))+(16*c(2)))+exp((5*c(1))+(25*c(2)))))*exp((5*c(1))+(25*c(2)))*(permitg1(-2)-c(5)*permitg1(-3)))

where gdpg is measured, permitg, permitg1 and permitg2 are measured quarterly. permitg1 is the the month prior to permitg and permitg2 is 2 months prior to permitg.

My problem is that when I try to run this regression i am greeted by the "Near singular matrix" error message.

Any help on this would be much appreciated.

Thanks

[EViews Gareth]

You've probably got perfectly collinear regressors. Obviously it is pretty difficult to diagnose what is causing the collinearity with such a large set of regressors.

[startz]

It is also possible that multicollinearity is a consequence of the coefficient starting values.

[EViews Gareth]

that's a good point too.