QZ decomposition failed/Blanchard-Kahn condistions failed

[yc1122]

Hi all,

I am a beginner on dynare. When I was trying to estimate my model, I found this error message:

Error using print_info (line 36)
The generalized Schur (QZ) decomposition failed. For more information, see the documentation for Lapack function dgges: info=25,
n=25

Error in check (line 76)
print_info(info, options.noprint);

Error in basenew (line 313)
oo_.dr.eigval = check(M_,options_,oo_);

Error in dynare (line 120)
evalin('base',fname) ;


Dynare also gives me an error msg that "Blanchard-Kahn condition is not satisfied: indeterminacy". I've checked the related posts on this forum and still get no clues. Just learn from other posts that it may derive from the timing issue of endogenous variables or the parameterization. However, I'm sure my model is good with the timing. As for the parameterization, cuz the initial values of the parameters are chosen from the literature and Taylor-rule is well satisfied, so I guess it should be fine as well.

Any help would be greatly appreciated. Thanks in advance.

[jpfeifer]

Please post the mod file

[yc1122]

Please post the mod file

I've attached the mod file, thanks a lot.

[jpfeifer]

Despite your model being linear, the steady state is not 0 (see u, un, etc):

STEADY-STATE RESULTS:

R -1.24345e-16
c 1.84424e-16
a 0
pi -1.24497e-16
eps_b 0
w -3.03159e-18
F 8.88178e-16
N 1.85047e-16
L 5.78902e-17
chi 0
mc -2.46529e-18
G 8.88178e-16
M 0
eps_p 0
wn -3.03159e-18
xi 0
Y 1.85047e-16
H 1.85047e-16
u 0.0410959
m 0
Ul -7.03132e-16
V 1.07323e-15
x 1.77636e-15
Gn -3.12542e-17
Yn -1.23266e-16
Nn -1.23266e-16
cn -1.23244e-16
Fn -3.12542e-17
LLn -1.30502e-16
Hn -1.23266e-16
Uln -9.20115e-17
un 0.0410959
Yg 3.08312e-16
ug -1.20774e-16
Ng 3.08312e-16
Fg 9.19433e-16
mrs -5.56952e-18
mrsn 2.33015e-17
r_obs 11.9412
inf_obs -4.97988e-14
gy_obs 7.92105
gw_obs 7.92105
gc_obs 7.92105
u_obs 0.0410959
x_obs -5.52146

There must still be something wrong with your model.

[yc1122]

yep, but I design this. Because u is actually unemployment rate, I want the level instead of percent deviation from the steady state. And I think as long as each variable has a steady state, the model is justified. Though u and un has no zero steady state, they do have steady states. Do I miss anything?

[jpfeifer]

But then you need to make sure that the steady states are the correct ones. Having "a" steady state is not sufficient. Apart from that, I don't get your error message, but rather

There are 11 eigenvalue(s) larger than 1 in modulus
for 13 forward-looking variable(s)

The rank conditions ISN'T verified!

indicating that your calibrated model has a problem. After that, I get a crash, because you did not provide the data-file

[yc1122]

Many thanks for your reply. The non-zero steady states are derived from the FOCs based on the initial values of structural parameters, so I think they are exact. In fact, I forgot to initialize parameters at the beginning which led me to the error msg of "QZ decomposition failure". Now, I got the same error msg as you did

There are 11 eigenvalue(s) larger than 1 in modulus
for 13 forward-looking variable(s)

The rank conditions ISN'T verified!

Does that mean it is the initial values of the parameters that cause this error? Actually, most of the initial values are picked from the literature. I am testing on different initial values but still get the same error msg. Any idea for fixing this problem? Many thanks in advance.

[jpfeifer]

I doubt that it is a parameter issue. It rather looks like a mistake in the model or a timing problem.

[yc1122]

Hi jpfeifer,

I've checked my model setup, equilibrium derivations and Dynare codes, but the problem remains even if I log-linearize the whole dynamic system including the unemployment rate (and steady states of endogenous variables are zeros now).

No matter how I change the initial values of parameters, it always has two less eigenvalues (larger than 1 in modulus) than # of non-predetermined variables. That reminds me of the potential timing issue about the two stock variables, employment and natural employment. In my model, employment Nt at the end of time t is composed of (1-delta)*N_{t-1} inherited from last period and the new hires (flow) Ht. But different from the case of capital and investment, the new hires become productive in the current period, so Nt is not predetermined. Does this setup conflict with the timing design in Dynare? Many thanks.

cheers,
yc

[jpfeifer]

If you are entering the timing in a consistent manner then there should be no problem. If a variable is not predetermined, you should not enter it as predetermined. That being said, I would guess that there is still a mistake here and that this is the source of your problem.

[ptr]

Dear All,

Am stuck with a code (attached) am running after reviewing closely related comments in the forum.

Basically, my current error is:

Blanchard Kahn conditions are not satisfied: indeterminacy due to rank failure

There are 9 eigenvalue(s) larger than 1 in modulus for 9 forward-looking variable(s)

Of which I have extensively searched for solution methods in the forum. I know, the problem could be of timing of the predetermined variables or mis- parameterization.

Thus, to try and resolved the error, I’ve again and again looked at my model and all the equations and parameters are fine.

Further, I run model_diagnostics(M_,options_,oo_) which yields:

model_diagnostic: the Jacobian of the static model is singular there is 2 colinear relationships between the variables and the equations

Relation 1
Colinear variables:
b
b_star_tilde
b_tilde
Relation 2 …

For this error I’ve also re-looked at the equations with b, b_star_tilde and b_tilde and they are also alright. This leaves me stuck and needs some help given that all my equations are fine.

Your help is much appreciated. Thanks in advance.

Kind Regards.

[ptr]

Could someone help with my code posted above?