EViews.com

Hello!

I tried to estimate a GMM equation with monthly data from 1999M01 to 2009M03. I put lagged exogenoeus variables as instruments as usually practiced. But I get an error "Near Singular Matrix".

Equation: stir_eu=c(1)*stir_eu(-1)+(1-c(1))*(c(2)+c(3)*outputgap+c(4)*d4log_hcpi)

Instruments: stir_eu(-1) stir_eu(-2) stir_eu(-3) stir_eu(-4) stir_eu(-5) stir_eu(-6) stir_eu(-7) stir_eu(-8) stir_eu(-9) stir_eu(-10) stir_eu(-11) stir_eu(-12) outputgap(-1) outputgap(-2) outputgap(-3) outputgap(-4) outputgap(-5) outputgap(-6) outputgap(-7) outputgap(-8) outputgap(-9) outputgap(-10) outputgap(-11) outputgap(-12) d4log_hcpi(-1) d4log_hcpi(-2) d4log_hcpi(-3) d4log_hcpi(-4) d4log_hcpi(-5) d4log_hcpi(-6) d4log_hcpi(-7) d4log_hcpi(-8) d4log_hcpi(-9) d4log_hcpi(-10) d4log_hcpi(-11) d4log_hcpi(-12)

Include lagged regressors is ticked.

Bandwith selection: Fixed: 12

Weighting matrix: HAC


I tried to reduce the instruments etc. but nothing helps. Do you have any idea here? What should I do?

If you need any information just ask.

Thank you !!!

Hello!

I tried to estimate a GMM equation with monthly data from 1999M01 to 2009M03. I put lagged exogenoeus variables as instruments as usually practiced. But I get an error "Near Singular Matrix".

Equation: stir_eu=c(1)*stir_eu(-1)+(1-c(1))*(c(2)+c(3)*outputgap+c(4)*d4log_hcpi)

Instruments: stir_eu(-1) stir_eu(-2) stir_eu(-3) stir_eu(-4) stir_eu(-5) stir_eu(-6) stir_eu(-7) stir_eu(-8) stir_eu(-9) stir_eu(-10) stir_eu(-11) stir_eu(-12) outputgap(-1) outputgap(-2) outputgap(-3) outputgap(-4) outputgap(-5) outputgap(-6) outputgap(-7) outputgap(-8) outputgap(-9) outputgap(-10) outputgap(-11) outputgap(-12) d4log_hcpi(-1) d4log_hcpi(-2) d4log_hcpi(-3) d4log_hcpi(-4) d4log_hcpi(-5) d4log_hcpi(-6) d4log_hcpi(-7) d4log_hcpi(-8) d4log_hcpi(-9) d4log_hcpi(-10) d4log_hcpi(-11) d4log_hcpi(-12)

Include lagged regressors is ticked.

Bandwith selection: Fixed: 12

Weighting matrix: HAC


I tried to reduce the instruments etc. but nothing helps. Do you have any idea here? What should I do?

If you need any information just ask.

Thank you !!!

Several ideas, perhaps one of these will help.

(1) Be sure you are using the most recent EViews update.
(2) Try unchecking "include lagged regressors." If that fixes the problem, then report this as a bug and add the lagged regressors in manually.
(3) As an experiment, use tsls instead of GMM and see what happens.

Thank You for your answer.. I converted my equation and now its working fine... :)

But I've another Problem... my equation seems to be highly autocorrelated... typically GMM should be robust against autocorrelation and heteroskedasticity, but it's not.

do you have any idea how to fix that ? look at the uploaded correlogram test.. the q-stat and the probability values are highly significant..

Thank you for your help!

Thank You for your answer.. I converted my equation and now its working fine... :)

But I've another Problem... my equation seems to be highly autocorrelated... typically GMM should be robust against autocorrelation and heteroskedasticity, but it's not.

do you have any idea how to fix that ? look at the uploaded correlogram test.. the q-stat and the probability values are highly significant..

Thank you for your help!

GMM using Newey-West adjusts the standard errors to account for autocorrelation and heteroskedasticity. It doesn't eliminate either one and doesn't adjust coefficient estimates. So there is nothing surprising in finding serial correlation after GMM.

Ok, good to know.

What would be the best option to eliminate the autocorrelation from this GMM equation? Where I could find the problem?

As you see I tried to estimate the short-term interest rate, using series data for the consumer price index and the industrial production. further, I applied the first difference of order 12 for both log series, because its monthly data and then used this values in this equation.

If you need more information, just tell me.

Thank you in advance!

Ok, good to know.

What would be the best option to eliminate the autocorrelation from this GMM equation? Where I could find the problem?

As you see I tried to estimate the short-term interest rate, using series data for the consumer price index and the industrial production. further, I applied the first difference of order 12 for both log series, because its monthly data and then used this values in this equation.

If you need more information, just tell me.

Thank you in advance!

I'm not sure why you want to eliminate the autocorrelation.

As I know autocorrelation lead to inefficiency of equations. They are biased and do not show the real coherence. Isn't it?

As I know autocorrelation lead to inefficiency of equations. They are biased and do not show the real coherence. Isn't it?

Autocorrelation does not necessarily lead to bias. You are right it leads to inefficiency.

You might want to include a lagged dependent variable, but if you do then you may also need more lagged instruments.

Hello!

I have been faced with some similar problem hence I am writing here. I have been trying construct correlogram for ARMA(2,1) equation
x=1+0.9*x(-1)+0.5*x(-2)+e-1.2*e(-1) but EViews has been getting "Near singular matrix". May be that it is connected with exponential growth of x? Last value is 1.202130296508553e+275 (its number is 2500). Or it is result of hight correlation? What can I do to construct the correlogram?

You might want to show us more specifically the steps you've taken.

The steps are next:

1. Constructing the time series ARMA(2,1):

smpl @all
series e=nrnd

smpl @first 2
series x=1

smpl @first+2 @last
series x=1+0.9*x(-1)+0.5*x(-2)+e-1.2*e(-1)

2. View --> Correlogram

http://forums.eviews.com/viewtopic.php?f=3&t=858 - here the similar problem was solved by EViews upgrading but I use version 6.

There is a February 7 update to version 6 at the EViews website. If you haven't already done so, you might install that patch and see what happens.

Hello!

I have been faced with some similar problem hence I am writing here. I have been trying construct correlogram for ARMA(2,1) equation
x=1+0.9*x(-1)+0.5*x(-2)+e-1.2*e(-1) but EViews has been getting "Near singular matrix". May be that it is connected with exponential growth of x? Last value is 1.202130296508553e+275 (its number is 2500). Or it is result of hight correlation? What can I do to construct the correlogram?

Your model is not stationary and you simulate it for 2500 observations. The system inevitably produces very large numbers and does not allow you to make further calculations like computing variances, since it will ultimately cause overflow. You can, however, simulate for 1000 observations and view the correlogram...