EViews.com

Hello!
I used a k-class method and noticed that it is a difference between k-class with k=1 and two-stages least squares methods. In example that i attached to this topic are 2 factors and constant and only 1 instrumental variable. When i run TSLS it is an error "Order condition violated - Insufficient instruments". But when in this example i run k-class with k=1 there are no errors (if i set CoefficientsCovarianceMatrix = K-Class based). I read in Help that k-class becomes TSLS when k=1. Why in my example k-class doesn't need as much instrumental variables as TSLS needs?
The file with example is attached.
Thanks in advance.

k-class with k=1 is equal to TSLS only when the instruments are of sufficient rank (i.e. you have at least as many instruments as regressors). The documentation should be more specific about this.

I use EViews7 and i hadn't found in Help that k-class may have less instruments than TSLS. In section "Limited Information Maximum Likelihood and K-class Estimation" i had found only "There must be at least as many instrumental variables as there are independent variables". What section of Help should i see to find features of k-class instruments needed count?

It isn't in the help/documentation.

k-class with k=1 is equal to TSLS only when the instruments are of sufficient rank (i.e. you have at least as many instruments as regressors). The documentation should be more specific about this.

Can you run a k-class estimate if the instruments are not of sufficient rank?

Can you run a k-class estimate if the instruments are not of sufficient rank?

Yes. There are no errors after calculating. All coefficients are estimated, but one t-staistic is NA and one is 8.44E-05. T-statistic values looks strange. In this example there are 3 factors and only 1 instrument.

I'm pretty sure that k=1 estimation without sufficient instruments makes no sense, but it may not be easy for EViews to catch because of how the calculations are done. (Note also that TSLS is only a large sample estimator, so doing it with 15 observations is probably not a good idea anyhow.)

And does anybody know why there are no errors when used Bekker or k-class based or HHN coefficient covariance matrix and error "Near Singular Matrix" occure when IV Based coeff. covariance matrix used? Is this OK?

Iv based requires full rank instruments