I am estimating Optimal Policy using the example file included in Dynare (version 4.5.6). It works fine with a first order approximation but when I increase it to second (or higher) I get the error:
“Caught (not yet fatal) Kord exception: At ./decision_rule.hweb:679:(253):NaN or Inf asserted in DRFixPoint::calcFixPoint”
Any thoughts on what’s gone wrong?
Files attached below.
NKbaselinepp_o2.mod (10.6 KB)
NKbaselinepp_o1.mod (10.6 KB)
Actually, as stated in the message you get, the error is not fatal. Dynare++ continued to do computations, as you can see from the journal (the file with the
.jnl extension) that is created.
The problem seems to be that Dynare++ is not able to compute the so-called centralized version of the decision rule, i.e. the decision rule expressed as a deviation around the stochastic steady state (but note that it is not approximated around the stochastic steady state, see the Dynare++ manual for more details).
You can eliminate this message by adding the
--no-centralize option to Dynare++.
In any case, with or without this option, the computation goes ahead and the results are stored in the generated
.mat file, including the decision rules.
However all IRFs seem to be explosive or degenerate, so Dynare++ is unable to simulate a single IRF.