Hi Johannes !
I am reading your replication code of JQ_2012. I want to compute the mode myself. so I changed mode_compute to 9 and delete the mode_file option. however errors reports:
Error using print_info (line 45)
Blanchard Kahn conditions are not satisfied:
indeterminacy
why is that?
So I want to ask how you find the mode at the beginning, how can I change the code to find the mode?
thanks very much !
What did you change in the code?
thanks for your reply!
I only changed some option in estimation command. first, I delete the mode_file=JQ_2012_NK_mode, and then I changed the mode_compute to 9 instead of 0 to start the mode finding.
I guess the problem is about the prior mean as the initial value of estimation,right?
Yes, that may be the problem. Try providing different starting values for the parameters.
thanks!
I also want to know how do you set the starting value or how you get the right mode when you do this research. can you tell me some about it? that would be very helpful, thanks again!
I add the code with
check;
steady;
the dynare reports
There are 14 eigenvalue(s) larger than 1 in modulus
for 14 forward-looking variable(s)
The rank condition is verified.
why the estimation still have the BK condition problem?
Try adding
estimated_params_init(use_calibration);
end;
thanks for your reply
I notice that the calibrated values are the wrong mode computed by JQ . the result seems not right
prior mean mode s.d. prior pstdev
siggma 1.500 1.2655 0.2397 norm 0.3700
epsilon 2.000 2.4167 0.6661 norm 0.7500
h 0.500 0.4561 0.0814 beta 0.3000
omega 0.500 0.8237 0.0532 beta 0.3000
phi 0.100 18.0471 6.9525 invg 0.3000
varrho 0.100 0.2170 0.1295 invg 0.3000
psi 0.500 0.7392 0.1090 beta 0.1500
kappa 0.200 0.6222 0.2160 invg 0.1000
eta_bar 1.200 1.7748 0.0564 beta 0.1000
upsilon_bar 1.200 1.1556 0.0998 beta 0.1000
rho_z 0.500 0.9045 0.0290 beta 0.2000
rho_zeta 0.500 0.8965 0.0771 beta 0.2000
rho_gamma 0.500 0.9399 0.0225 beta 0.2000
rho_eta 0.500 0.8744 0.0473 beta 0.2000
rho_upsilon 0.500 0.1642 0.0922 beta 0.2000
rho_G 0.500 0.9762 0.0120 beta 0.2000
rho_varsigma 0.500 0.2597 0.0752 beta 0.2000
rho_xi 0.500 0.9893 0.0069 beta 0.2000
rho_gz 0.500 0.8582 0.1049 beta 0.2000
rho_R 0.750 0.8294 0.0303 beta 0.1000
nu_1 1.500 2.1431 0.1643 norm 0.2500
nu_2 0.120 0.0255 0.0432 norm 0.0500
nu_3 0.120 0.2171 0.0287 norm 0.0500
standard deviation of shocks
prior mean mode s.d. prior pstdev
eps_z 0.001 0.0047 0.0003 invg 0.0500
eps_zeta 0.001 0.0126 0.0036 invg 0.0500
eps_gamma 0.001 0.0193 0.0039 invg 0.0500
eps_eta 0.001 0.0179 0.0041 invg 0.0500
eps_upsilon 0.001 0.7067 0.3552 invg 0.0500
eps_G 0.001 0.0160 0.0012 invg 0.0500
eps_varsigma 0.001 0.0013 0.0001 invg 0.0500
eps_xi 0.001 0.0149 0.0014 invg 0.0500
so can you tell me what the initial value you used to compute the mode, thanks!
I used the ones of JQ, but run a long mode-finding with mode_compute=9, getting the mode-file I uploaded
1 Like
Hi thanks for your patience!
I do as you said, using the JQ mode as initial value.and mode_compute=9, after a long time finding but get different results
what do you think might be the reasons? if I try a few times, can I get better results?
Mode-finding is tricky. For that model, I ran mode-finding from starting several randomly drawn points and picked the point with the highest posterior density.
1 Like