IRFs incompatible w/ decision rule trajectory


I am trying to solve, using Dynare 4.0.3, a DSGE with 260 variables (26 forward and 81 state ones). The 1st order approximation was running apparently ok, without errors. But, when I was trying to verify the irf, carefully, I found completely incompatible behaviors for only few variables (around 6). For instance, when comparing the capital stock with the investment behavior, as a reaction to a moderate monetary policy tightening (+100bp in the nominal rate), I found that the investment reduced a little (-3%), but, at the same time, and surprisingly, the capital stock increased a lot (15%), just after the shock.

Since that behavior is totally incompatible with the capital accumulation equation, I decided to verify if all the equations were balanced at all points of the trajectory generated by de irf’s reported by Dynare.

After the verification, I found that the capital accumulation equation and some other few equations (around 6) were really not appropriately balanced. Then, I decide to ignore the irfs, and to reconstruct the trajectories through a Matlab’s routine, using the only the Dynare’s matrices of the 1st order approximation of the decision rule (the ones reported together the irfs).

Applying the same level of shock, I obtained the same trajectory except for, exactly, the same variables that seemed to be incompatible (the ones that generated that unbalanced equations). And, at this time, all the equations became balanced at all the points of this decision rule generated trajectory. In particular, the capital accumulation became balanced, when the capital stock became decreasing, as expected, and perfectly compatible with the decreasing investment.

Did anyone experience this type of problems with the Dynare’s irfs? Since the matrices seem to be ok, could it be the consequence of a bug at the irfs routine?

Best regards,
Marcos Valli.

Dear Marcos,

could you please, post or send me the *.mod file, the Matlab file that you used for checking and all the details making it possible to spot quickly the problem?
Thank you in advance,