Blanchard Kahn conditions not met in estimation

[tamara b]

Hello,

I am trying to estimate a small open economy model with financial block and it works fine without the estimation block.
However once I add estimation (6 observables and just 1 parameter to estimate) it reports that suddenly Blanchard Kahn conditions are not met and the model fails to converge.
How is this possible?
Could it be that it has some other problem but reports this error in absence of the appropriate one?
I mean if the model is stable in itself, and I declared all parameter values for steady state calculations (guide says dynare will overwrite the value for the estimated parameter once it gets to the estimation part) how come the estimation block itself is making it unstable?

Thanks
t

[jpfeifer]

The error is triggered in estimation. Either your model has a unit root and you need to use the diffuse_filter option or you did not provide the calibration values as starting values for estimation, e.g. using an estimated_params_init-block. If you could post your mod-file and datafile, I can tell you more.

[tamara b]

thanks jpf

I attached the code,
it does not allow me to attach an xls file (I tried saving in 2003-7 version and 2010), so I am pasting the contents..
Could it be that it needs more observables, cause I did not include all that are relevant in equations using beta (the estimated parameter)?

p c q rfl rdd rdl rfd
7.67 101.72 114.26 8.46 9.10 16.15 2.99
10.26 101.46 115.82 8.26 11.36 17.30 3.64
14.14 99.91 125.63 8.15 12.57 18.40 3.52
12.66 100.31 131.50 8.87 12.34 19.38 3.78
9.31 99.51 127.45 8.24 10.98 18.45 3.71
7.00 99.70 124.39 8.25 9.76 17.17 3.92
3.18 101.46 116.57 8.06 9.13 18.02 3.67
5.46 99.98 113.79 7.77 10.12 17.33 3.08
10.30 100.34 117.67 7.70 10.35 17.85 2.75
12.18 99.31 123.78 8.06 10.60 17.40 2.73
11.19 97.21 126.10 8.01 10.26 19.25 2.71


thanks again
t

[jpfeifer]

The check-command says:

EIGENVALUES:
Modulus Real Imaginary

0 0 0
1.504e-16 1.504e-16 0
3.818e-16 3.818e-16 0
0.1 0.1 0
0.2363 -0.2363 0
0.3 0.3 0
0.5 0.5 0
0.5 0.5 0
0.518 0.518 0
0.5842 0.5842 0
0.7176 0.713 0.08111
0.7176 0.713 -0.08111
0.7534 0.7534 0
1 1 0
1.01 1.01 0
1.083 1.083 0
1.408 1.321 0.4895
1.408 1.321 -0.4895
10.1 10.1 0
2.983e+48 -2.983e+48 0

Thus, you have a unit root as one of the eigenvalues is 1. Try using the diffuse_filter-option. To post datafiles, put everything in a zip-file

[tamara b]

thanks jpf
it works now but I obviously have to go back to the basics and figure out how the dynare options work in the first place..
thanks again, spared me a lot of trial and error time waste

[jpfeifer]

If you are sure that the unit root belongs in the model, just use

Code: Select all

estimation(datafile=dataipa1qs,diffuse_filter);


[tamara b]

well that s the trouble..
I don t see a unit root in any of my equations, on the contrary all my rho-s in AR processes are well bellow 1..
however diffuse_filter option does override the Blanchard Kahn issue... and raises new issues, namely: the hessian matrix at the "mode" is not positive definite!

I figure I must go back and figure out how the estimation works so I could figure why it won't work in my case