Time-Varying risk

[rteconomics]

Hi guys! I'm working to an SOE model with time-varying risk of the form:

Code: Select all

a=rho*a(-1)+sigma_a*e;
sigma_a=(1-rho_z)*sigma_ss+rho_z*sigma_a(-1)+kappa_a*u;


After specifying the dll option in the 'model' block, I was asked to choose a compiler:

Code: Select all

Select a compiler:
[1] Lcc-win32 C 2.4.1 in I:\w5\MATLAB~3\sys\lcc
 
[0] None

I chose 1 (btw, what is the difference?). As a result, I get this error (and it seems I cannot change to 'None' anymore):

Code: Select all

Error time_var_dynamic.c: 1058  expression too complicated
Error time_var_dynamic.c: 1058  expression too complicated
Error time_var_dynamic.c: 1058  expression too complicated
Error time_var_dynamic.c: 1058  expression too complicated
Error time_var_dynamic.c: 1058  expression too complicated
5 errors, 0 warnings
 
  I:\W5\MATLAB~3\BIN\MEX.PL: Error: Compile of 'time_var_dynamic.c' failed.
 
??? Error using ==> mex at 221
Unable to complete successfully.

Error in ==> time_var at 71
    eval('mex -O PRELINK_CMDS1="echo EXPORTS > mex.def & echo
    mexFunction >> mex.def & echo Dynamic >> mex.def"
    time_var_dynamic.c')

Error in ==> dynare at 132
evalin('base',fname) ;

The same happens with the code posted here (viewtopic.php?f=1&t=2829) by MichelJuillard.

I attach my code.
Could anyone please help? Thanks!
Ciao,
RT

[rteconomics]

I forgot to tell you that I'm using the 4.1.3 version. Does this matter? Thanks. RT

[jpfeifer]

Please upgrade to a newer version. There have been some bugs discovered since then. Moreover, in more recent versions, the use_dll option is not required anymore.

[rteconomics]

Hello!
Thank you for your reply. I did it this morning and the code seems to work fine now. However, I still have a little issue (this is an issue that other might have faced).
I specified the TFP process as:

Code: Select all

sigma_e = rho_sigma_e*sigma_e(-1) + (1-rho_sigma_e)*mean_sigma_e + sigma_eta*eta_e;
a = rho_a*a(-1) + sigma_e*e;


and the 'shocks' block as (I took this from a code MJ posted, see my original post above):

Code: Select all

shocks;
var e = 1;
var eta_e = 1;
end;
stoch_simul(order=3, periods=1000, irf=10);

ex_ = zeros(40,2);
ex_(1,2) = 1;
irf2 = simult_(oo_.steady_state,oo_.dr,ex_,2);

figure;
plot(irf2(2,:)-oo_.steady_state(3));
title('Consumption');

The output are the IRFs relative to e (standard TFP level shock):
[img]

IRFs(lev).pdf

[/img]

to eta (shock to the second moment):
[img]

IRFs(var).pdf

[/img]

and the IRF of consumption:
[img]

consumption.pdf

[/img]

I have 2 questions:
1) Is there a way to smooth the IRFs coming from shocks to eta?
2) What is exactly the difference between the IRF of consumption and the IRF of c in the 'eta shock' case (basically in the second and third figure)? Why are they different and which one is the true one?

[rteconomics]

Hi guys!
Any news on this issue?
Thank you!
RT

[jpfeifer]

1. If you don't do the IRFs in the way the authors of the AER paper proposed, the only way would be to run really many replications, see e.g. http://www.imf.org/external/pubs/ft/wp/2010/wp10249.pdf. Note that pruning also helps but is not implemented for higher orders than 2

2. There is no right or wrong. The ways to compute them are different. stoch_simul generates IRFs as described in http://www.dynare.org/DynareWiki/IrFs. During the whole simulation period there are random shocks hitting the system. The difference between the baseline and the shocked series is a shock at time t=101. IRFs are plotted as the percentage difference between the shocked series and the baseline simulation.

In contrast, by using

Code: Select all

ex_ = zeros(40,2);
ex_(1,2) = 1;


you have 40 periods without any shock except for the first shocks at period 1. Then you compare this shocked series to the deterministic steady state.