Version 3

Known bugs in Dynare version 3

Version 3.065

bug in moments reported by stoch_simul for model with nonstationary variables
bug in OSR corrected in snapshot as of 3/23/07
bug in histval (corrected)
option filtered_vars in estimation doesn't work in models with variables lagged on more than one period
bug in undocumented command model_comparison. Fixed in snapshot as of 09/18/07

there is a bug in option shocks_file for shocks command. See the forum for more details and a corrected file. The SVN version of dynare_v3 is now corrected.
in some cases, the CHECK command reports 'The rank condition is verified' when the number of eigenvalues larger than one is greater than the number of foward-looking variables. This is obviously absurd and there is no stable trajectory in such a case. This bug is corrected in the SVN version of Dynare and in the snapshot for version 3 dated July 13th or more recent.
one can't simulate series (use periods= option) with OLR.
stoch_simul may report finite variances for nonstationary variables in models with a unit root.
there was a bug in histval. This is corrected in the snapshot for dynare_v3 as of 10/10/06.
there is a bug in deterministic simulations when there are lags on exogenous variables
IRF were saved with the wrong name when only a subset of variables were selected. Corrected in the snapshot as of 02.05.07.
corrected bug in smoother with measurement error in some cases. Available in snapshot as of 02.14.07
corrected bug in stochastic simulation with deterministic shocks when there is more than one deterministic shock. Available in snpashot as of 02.14.07

the PDF version of the manual is obsolete. Please, use the HTML version.
in some cases, Metropolis iterations can wander in regions of the parameter space where the model has no solution. There is a new bug in the penalty for such cases.
setting a boundary on a correlation in estimated_params_bounds triggers a Matlab error
in metropolis.m (line 2496) MAX_nirfs should be MAX_nirfs_dsge
there is a bug in osr (optimal simple rules)

simulation of 2nd order approximations are replicated uselessly many times. Only the last one is made available.
one can't use a parameter called 'c' in deterministic models
one can't use the command 'check' before an 'estimation' command with options bayesian_irf, moment_varendo or forecast
in estimation with option moment_varendo the name of the exogenous variables are mixed up
in estimation the option tex doesn't work correctly

the doc directory is missing from the distribution. Use the documentation from the web site or from version 3.021. There was no change to the documentation in version 3.03
Since version 3.02, it isn't possible to calibrate the variance of a shock to zero. Instead, remove the exogenous variable from the model.

undocumented function for optimal policy (olr) is broken. Use version 3.02 if you want to use it.
still bugs with "check" on backward models (P. Hollinger 2/8/05)

In the file dynare_estimation (line 911) you mistakely wrote option._varobs (it should be options._varobs). The same on the line 918. (E. Pytlarczyck 2/3/05)
The command "check" doesn't work for backward models (P. Hollinger 2/4/05)
The specification of shocks in deterministic models is broken (M. Canzoneri 2/5/05)

There is a bug in the computation of the Kalman filter for models with measurement errors. Thanks to M. Ratto, E. Pytlarczyk and T. Husebo for reporting it. We are going to upload an update soon (01/24/2005)
Several bugs found by Marco Ratto in diffuse filter and smoother. Should only affect models presenting numerical difficulties in computing the Kalman filter.
In example F. Schorfheide (2000), a file is missing to be able to run fs2000a.mod that estimates the model in level of the variables.
Bug in specification of individual "jscale" for estimated parameters found by Peter Welz.