SW2007 Recursive Forecasts and Dynare 4.3.2


I’m trying to replicate Table 3 in SW2007 using the code provided by SW. The VAR portions of the table can be replicated exactly without much problem. If I rerun the recursive DSGE estimations at the bottom of the usmodel.mod file, however, the parameters from posterior maximization are somewhat different from the estimated parameters provided in SW’s package (i.e. in the usmodel_hist_dsge_f19_7_xx_xxx_mode.mat files).

I realize that SW did their estimations in a previous version of Dynare. I was wondering if there have been any changes in the new version that could make the model estimate so differently? Is this a known difference, or is it more likely I’m doing something wrong?

Just as an example, I’ve included some of the parameters from the first forecast period estimation (as in the usmodel_hist_dsge_f19_7_71_100_mode.mat file). The results from a new (Dynare 4.3.2) estimation and those in the SW package, along with the standard deviations are shown below. As you can see, some of the differences I’m getting are significant, and ultimately make it difficult to replicate Table 3. I’m not sure where I may be going wrong, since I’m just running the SW code as provided. Any help/thoughts/comments would be greatly appreciated.


Param = Dynare 4/ SW Pack/ Std Dev.
ctrend = 0.314/ 0.434/ 0.017
constepinf = 0.631/ 0.814/ 0.101
constelab = 0.850/ -0.048/ 1.177
csigl = 1.910/ 1.626/ 0.636
csadjcost = 3.955/ 4.577/ 0.959
cindp = 0.469/ 0.294/ 0.168
crhob = 0.625/ 0.230/ 0.122
crhopinf = 0.627/ 0.927/ 0.196
cmap = 0.559/ 0.772/ 0.177
eb = 0.191/ 0.274/ 0.036