Replicating Gertler & Trigari (2009)

[dkoh]

Hi,

I would like to ask any of your help on replicating results in Gertler & Trigari (2009)
"Unemployment Fluctuations with Staggered Nash Wage Bargaining" JPE.
In their paper, they presented log-linearized equations in the appendix
and hence it must be easily replicated by dynare.

But the simulated unconditional moments and IRFs are far from their results.
One free parameter that they haven't specified in their paper is H_ss (steady state worker's surplus).
So I played around with this parameter but no hope to even get closer to their results.
For your reference, I attached mod file.
I appreciate any of your help in advance.
Thank you.

[ilobayesian]

HI,
try to pin down H_ss by using the SS relationship between it and J_ss, as specified by the Nash bargaining focs

[dkoh]

Thanks.
I could pin down the steady state of H by using SS relationship of H and J.
But still the business cycle moments are way off,
although IRFs look quite close to the ones in the paper.
The following are the business cycle moments
(each simulated variable was hp-filtered with a smoothing parameter of 1600).
I am particularly interested in replicating movements of w, ls and n variables.

Rel.std Corr AutoCorr
y 1 1 0.7868
w 0.2797 0.4014 0.9672
ls 0.8576 -0.8394 0.7143
n 0.2897 0.801 0.9501
a 0.865 0.962 0.7028
i 2.962 0.9949 0.8226
c 0.3093 0.9911 0.7418

[dyndyn]

Hi, I am working on Gertler Sala Trigari (2008) and I find your advice about H_ss very useful. Are there any other equations that are not listed in the Appendix that you have to use? I ask that because I still have more variables than steady state relations.
Thanks

[dkoh]

I resolve the problem that the business cycle moments that they produced in the paper is a monthly cycle.
That is they hp-filtered the simulated variables with a monthly smoothing parameter.
I do not find which number they used for the monthly smoothing but either lambda=14400 or 129,600
yields the moments quite close to theirs.

To answer dyndyn's question,
they argue that a_ss (steady state of a) was determined outside the model.
This could be one way, but also setting the steady state of k/n could be another way.
The steady state k/n detemines a_ss and r_ss and other steady states
while the results won't change much.