Singular matrix with diag BEKK estimation

[don quixote]

Hello!

This question regards estimations in Eviews 7.

Im currently trying to estimate a diagonal BEKK-model with data consisting of two different financial series. When I run the VAR(1) model and order it by variables i get the following

A = C(1)*A(-1) + C(2)*B(-1) + C(3)

B = C(4)*A(-1) + C(5)*B(-1) + C(6)

Then when I try to estimate the equation with diag-BEKK it gives me two different errors (only one displayed each time i try)

1. "No valid observations in equation A = C(1)*A(-1) + C(2)*B(-1) + C(3)"

2. Near singular matrix

I have trouble interpreting what the first error comes from since all the observations in A are valid.

The second error of course indicates that the series are linear combinations of each other, which is possible. However i have executed the same estimation with one of the series on nonsense random numbers and Eviews still tells me "near singular matrix".

Could someone please help me explain what I am doing wrong?

I have attached the workfile

Thanks in advance!

don quixote

[EViews Glenn]

I get...


System: E
Estimation Method: ARCH Maximum Likelihood (Marquardt)
Covariance specification: Diagonal BEKK
Date: 04/23/12 Time: 14:54
Sample: 4/01/2002 3/26/2012
Included observations: 2606
Total system (balanced) observations 5212
Presample covariance: backcast (parameter =0.7)
Convergence achieved after 78 iterations

Coefficient Std. Error z-Statistic Prob.

C(1) -0.345363 0.055393 -6.234815 0.0000
C(2) 0.341402 0.050790 6.721766 0.0000
C(3) 7.33E-05 9.35E-05 0.783353 0.4334
C(4) 0.053304 0.058779 0.906856 0.3645
C(5) -0.059335 0.055320 -1.072577 0.2835
C(6) 7.26E-05 9.69E-05 0.748752 0.4540

Variance Equation Coefficients

C(7) 1.83E-07 1.90E-08 9.636060 0.0000
C(8) 1.89E-07 1.65E-08 11.44972 0.0000
C(9) 2.01E-07 1.71E-08 11.71918 0.0000
C(10) 0.201087 0.004807 41.83607 0.0000
C(11) 0.201092 0.005068 39.67935 0.0000
C(12) 0.978879 0.000996 983.2859 0.0000
C(13) 0.978849 0.000940 1041.545 0.0000

Log likelihood 22961.03 Schwarz criterion -17.58243
Avg. log likelihood 4.405417 Hannan-Quinn criter. -17.60109
Akaike info criterion -17.61169


Equation: A = C(1)*A(-1) + C(2)*B(-1) + C(3)
R-squared 0.021354 Mean dependent var -4.25E-05
Adjusted R-squared 0.020602 S.D. dependent var 0.007236
S.E. of regression 0.007161 Sum squared resid 0.133481
Durbin-Watson stat 1.932770

Equation: B = C(4)*A(-1) + C(5)*B(-1) + C(6)
R-squared -0.003433 Mean dependent var -3.57E-05
Adjusted R-squared -0.004204 S.D. dependent var 0.007880
S.E. of regression 0.007896 Sum squared resid 0.162298
Durbin-Watson stat 1.822000


Covariance specification: Diagonal BEKK
GARCH = M + A1*RESID(-1)*RESID(-1)'*A1 + B1*GARCH(-1)*B1
M is an indefinite matrix
A1 is a diagonal matrix
B1 is a diagonal matrix

Transformed Variance Coefficients

Coefficient Std. Error z-Statistic Prob.

M(1,1) 1.83E-07 1.90E-08 9.636060 0.0000
M(1,2) 1.89E-07 1.65E-08 11.44972 0.0000
M(2,2) 2.01E-07 1.71E-08 11.71918 0.0000
A1(1,1) 0.201087 0.004807 41.83607 0.0000
A1(2,2) 0.201092 0.005068 39.67935 0.0000
B1(1,1) 0.978879 0.000996 983.2859 0.0000
B1(2,2) 0.978849 0.000940 1041.545 0.0000

[don quixote]

Thank you very much for your response!

When you estimated the BEKK model, did you go through the same steps as i did or am i doing something wrong?

don quixote

[EViews Glenn]

I just created the system by copying the equations you provided. Clicked on Estimate, then selected BEKK. I wasn't sure if you wanted any other option changes so I just used the defaults...

[don quixote]

Right! And when I create a system with the equations posted above and try to estimate a BEKK, I get "near singular matrix" or "no valid observations..".

Does this imply that something is wrong with my Eviews?

don quixote

[EViews Glenn]

Starting values?

Or perhaps version. What's the date on your copy of EViews?

[don quixote]

Since i'm using the exact same data as posted in the workfile i guess it must be something with the verision?

Im using Eviews 7.1

don quixote

[EViews Glenn]

Or perhaps your estimation settings. What's the build date of your copy of EViews 7.1 (Help/About EViews). I'd recommend updating and then trying again. If you're up-to-date we see if we can figure this out...

[don quixote]

I updated my verision of Eviews and now it works. Thank you very much for your answers!

don quixote