estimated coefficients of the b matrix are all 0.1

[Jasonjjh]

Im attempting to create a SVAR model of my 5 factors AUGDP, AUGE, CASH, USGDP, RTWI. My EViews file is getting this odd result that the estimated coefficients of the B matrix are all 0.1 which seem to be the starting values since it doesn't actually go through any iterations (this with the default Choleski restrictions in EViews). So I re-ran it with all u's and no e's on the right-hand side (also for the Choleski) and get sensible results. My supervisor states he doesn't know what's going on with the failure to converge and believes it may be a bug in the program. In my work file i have attempted to use the Hodrick–Prescott filter but this has not helped.

[EViews Gareth]

Your workfile doesn't include a VAR object.

[Jasonjjh]

ok i have taken lags of my data and run them through the hodrick prescott filter and have estimate the var with the 5 varibales lhusgdp lhaugdp lhcash lhauge lhrtwi. If i use the sample formual given than the coeffecients are all 0.1.

the svar formula i want to impose is
@e1 = c(1)*@u1
@e2 = -c(2)*@e1 - c(3)*@e3 - c(4)*@e4 - c(5)*@e5 + c(6)*@u2
@e3 = -c(7)*@e2 - c(8)*@e5 + c(9)*@u3
@e4 = -c(10)*@e2 + c(11)*@u4
@e5= -c(12)*@e2 + c(13)*@u5

this also gives 01 as the results.

[EViews Gareth]

You have your estimation set to use 0.1 as starting values, and the VAR is reporting it could not converge (Failure to improve after 1 interation). Thus the final estimates are all 0.1.

[Jasonjjh]

what do i need to do to change the starting value to not be 0.1.

[EViews Gareth]

The Optimization Control tab of the Structural Factorization dialog has options for starting values.

[donihue]

If you draw the starting values from a Uniform(0,1) distribution, your SVAR converges; however, the LR test indicates that the (over-)identification restrictions are strongly rejected, thus indicating that the proposed SVAR is rejected by the data, implying the need to change the underlying restrictions.
Regards
Donihue