Dynare 5.0 Released

Posted on 07 January 2022

We are pleased to announce the release of Dynare 5.0.

This major release adds new features and fixes various bugs.

The Windows, macOS and source packages are already available for download at the Dynare website.

All users are strongly encouraged to upgrade.

This release is compatible with MATLAB versions ranging from 8.3 (R2014a) to 9.11 (R2021b), and with GNU Octave version 6.4.0 (under Windows).

The new tools for semi-structural models and the improvements on the nonlinear solvers were funded by the European Central Bank. Special thanks to Nikola Bokan (ECB) for his contributions and numerous bug reports and fixes.

Major user-visible changes

  • New routines for simulating semi-structural (backward) models where some equations incorporate expectations based on future values of a VAR or trend component model. See the var_model, trend_component_model and var_expectation_model commands, and the var_expectation operator.

  • New routines for simulating semi-structural models where some equations are specified using the polynomial adjustment costs (PAC) approach, as in the FRB/US model (see Brayton et al., 2014 and Brayton et al., 2000) and the ECB-BASE model (see Angelini et al., 2019). The forward-looking terms of the PAC equations can be computed either using a satellite VAR model, or using full model-consistent expectations. See the pac_model command and the pac_expectation operator.

  • New Method of Moments toolbox that provides functionality to estimate parameters by (i) Generalized Method of Moments (GMM) up to 3rd-order pruned perturbation approximation or (ii) Simulated Method of Moments (SMM) up to any perturbation approximation order. The toolbox is inspired by replication codes accompanying Andreasen et al. (2018), Born and Pfeifer (2014), and Mutschler (2018). It is accessible via the new method_of_moments command and the new matched_moments block. Moreover, by default, a new non-linear least squares optimizer based on lsqnonlin is used for minimizing the method of moments objective function (available under mode_compute=13). GMM can further benefit from using gradient-based optimizers (using analytic_standard_errors option and/or passing 'Jacobian','on' to the optimization options) as the Jacobian of the moment conditions can be computed analytically.

  • Implementation of the Occbin algorithm by Guerrieri and Iacoviello (2015), together with the inversion filter of Cuba-Borda, Guerrieri, Iacoviello, and Zhong (2019) and the piecewise Kalman filter of Giovannini, Pfeiffer, and Ratto (2021). It is available via the new block occbin_constraints and the new commands occbin_setup, occbin_solver, occbin_graph, and occbin_write_regimes.

  • Stochastic simulations

    • stoch_simul now supports theoretical moments at order=3 with pruning.

    • stoch_simul now reports second moments based on the pruned state space if the pruning option is set (in previous Dynare releases it would report a second-order accurate result based on the linear solution).

  • Estimation

    • Performance optimization to pruned state space systems and Lyapunov solvers.

    • New option mh_posterior_mode_estimation to estimation to perform mode-finding by running the MCMC.

    • New heteroskedastic filter and smoother, where shock standard errors may unexpectedly change in every period. Triggered by the heteroskedastic_filter option of the estimation command, and configured via the heteroskedastic_shocks block.

    • New option mh_tune_guess for setting the initial value for mh_tune_jscale.

    • New option smoother_redux to estimation and calib_smoother to trigger computing the Kalman smoother on a restricted state space instead of the full one.

    • New block filter_initial_state for setting the initial condition of the Kalman filter/smoother.

    • New option mh_initialize_from_previous_mcmc to the estimation command that allows to pick initial values for a new MCMC from a previous one.

    • The xls_sheet option of the estimation command now takes a quoted string as value. The former unquoted syntax is still accepted, but no longer recommended.

    • New option particle_filter_options to set various particle filter options.

  • Perfect foresight and extended path

    • New specialized algorithm in perfect_foresight_solver to deal with purely static problems.

    • The debug option of perfect_foresight_solver provides debugging information if the Jacobian is singular.

    • In deterministic models (perfect foresight or extended path), exogenous variables with lead/lags are now replaced by auxiliary variables. This brings those models in line with the transformation done on stochastic models. However, note that the transformation is still not exactly the same between the two classes of models, because there is no need to take into account the Jensen inequality for the latter. In deterministic models, there is a one-to-one mapping between exogenous with lead/lags and auxiliaries, while in stochastic models, an auxiliary endogenous may correspond to a more complex nonlinear expression.

  • Optimal policy

    • Several improvements to evaluate_planner_objective:

      • it now applies a consistent approximation order when doing the computation;
      • in addition to the conditional welfare, it now also provides the unconditional welfare;
      • in a stochastic context, it now works with higher order approximation (only the conditional welfare is available for order ⩾ 3);
      • it now also works in a perfect foresight context.

    • discretionary_policy is now able to solve nonlinear models (it will then use their first-order approximation, and the analytical steady state must be provided).

  • Identification

    • New option schur_vec_tol to the identification command, for setting the tolerance level used to find nonstationary variables in the Schur decomposition of the transition matrix.

    • The identification command now supports optimal policy.

  • Shock decomposition

    • The fast_realtime option of the realtime_shock_decomposition command now accepts a vector of integers, which runs the smoother for all the specified data vintages.

  • Macro processor

    • Macroprocessor variables can be defined without a value (they are assigned integer 1).

  • LaTeX and JSON outputs

    • New nocommutativity option to the dynare command. This option tells the preprocessor not to use the commutativity of addition and multiplication when looking for common subexpressions. As a consequence, when using this option, equations in various outputs (LaTeX, JSON…) will appear as the user entered them (without terms or factors swapped). Note that using this option may have a performance impact on the preprocessing stage, though it is likely to be small.

    • Model-local variables are now substituted out as part of the various model transformations. This means that they will no longer appear in LaTeX or in JSON files (for the latter, they are still visible with json=parse or json=check).

  • Compilation of the model (use_dll option)

    • Block decomposition (option block of model) can now be used in conjunction with the use_dll option.

    • The use_dll option can now directly be given to the dynare command.

  • dseries classes

    • Routines for converting between time series frequencies (e.g. daily to monthly) have been added.

    • dseries now supports bi-annual and daily frequency data.

    • dseries can now import data from DBnomics, via the mdbnomics plugin. Note that this does not yet work under Octave. For the time being, the DBnomics plugin must be installed separately.

  • Misc improvements

    • The histval_file and initval_file commands have been made more flexible and now have functionalities similar to the datafile option of the estimation command.

    • When using the loglinear option, the output from Dynare now clearly shows that the results reported concern the log of the original variable.

    • Options block and bytecode of model can now be used in conjunction with model-local variables (variables declared with a pound-sign #).

    • The model_info command now prints the typology of endogenous variables for non-block decomposed models.

    • The total computing time of a run (in seconds) is now saved to oo_.time.

    • New notime option to the dynare command, to disable the printing and the saving of the total computing time.

    • New parallel_use_psexec command-line Windows-specific option for parallel local clusters: when true (the default), use psexec to spawn processes; when false, use start.

    • When compiling from source, it is no longer necessary to pass the MATLAB_VERSION version to the configure script; the version is now automatically detected.

Incompatible changes

  • Dynare will now generally save its output in the MODFILENAME/Output folder (or the DIRNAME/Output folder if the dirname option was specified) instead of the main directory. Most importantly, this concerns the _results.mat and the _mode.mat files.

  • The structure of the oo_.planner_objective field has been changed, in relation to the improvements to evaluate_planner_objective.

  • The preprocessor binary has been renamed to dynare-preprocessor, and is now located in a dedicated preprocessor subdirectory.

  • The dynare command no longer accepts output=dynamic and output=first (these options actually had no effect).

  • The minimal required MATLAB version is now R2014a (8.3).

  • The 32-bit support has been dropped for Windows.

Bugs that were present in 4.6.4 and that have been fixed in 5.0

  • Equations marked with static-tags were not detrended when a deflator was specified
  • Parallel execution of dsge_var estimation was broken
  • The preprocessor would incorrectly simplify forward-looking constant equations of the form x(+1)=0 to imply x=0
  • Under some circumstances, the use of the model_local_variable statement would lead to a crash of the preprocessor
  • When using the block-option without bytecode the residuals of the static model were incorrectly displayed
  • When using k_order_solver, the simult_ function ignored requested approximation orders that differed from the one used to compute the decision rules
  • Stochastic simulations of the k_order_solver without pruning iterated on the policy function with a zero shock vector for the first (non-endogenous) period
  • estimation would ignore the mean of non-zero observables if the mean was 0 for the initial parameter vector
  • mode_check would crash if a parameter was estimated to be exactly 0
  • load_mh_file would not be able to load the proposal density if the previous run was done in parallel
  • load_mh_file would not work with MCMC runs from Dynare versions before 4.6.2
  • ramsey_model would not correctly work with lmmcp
  • ramsey_model would crash if a non-scalar error code was encountered during steady state finding.
  • Using undefined objects in the planner_objective function would yield an erroneous error message about the objective containing exogenous variables
  • model_diagnostics did not correctly handle a previous loglinear option
  • solve_algo=3 (csolve) would ignore user-set maxit and tolf options
  • The planner_objective values were not based on the correct initialization of auxiliary variables (if any were present)
  • The nostrict command line option was not ignoring unused endogenous variables in initval, endval, and histval
  • prior_posterior_statistics_core could crash for models with eigenvalues very close to 1
  • The display of the equation numbers in debug mode related to issues in the Jacobian would not correctly take auxiliary equations into account
  • The resid command was not correctly taking auxiliary and missing equations related to optimal policy (ramsey_model, discretionary_policy) into account
  • bytecode would lock the dynamic.bin file upon encountering an exception, requiring a restart of MATLAB to be able to rerun the file
  • Estimation with the block model option would crash when calling the block Kalman filter
  • The block model option would crash if no initval statement was present
  • Having a variable with the same name as the mod-file present in the base workspace would result in a crash
  • oo_.FilteredVariablesKStepAheadVariances was wrongly computed in the Kalman smoother based on the previous period forecast error variance
  • Forecasts after estimation would not work if there were lagged exogenous variables present
  • Forecasts after estimation with MC would crash if measurement errors were present
  • Smoother results would be infinity for auxiliary variables associated with lagged exogenous variables
  • In rare cases, the posterior Kalman smoother could crash due to previously accepted draws violating the Blanchard-Kahn conditions when using an unrestricted state space
  • perfect_foresight_solver would crash for purely static problems
  • Monte Carlo sampling in identification would crash if the minimal state space for the Komunjer and Ng test could not be computed
  • Monte Carlo sampling in identification would skip the computation of identification statistics for all subsequent parameter draws if an error was triggered by one draw
  • The --steps-option of Dynare++ was broken
  • smoother2histval would crash if variable names were too similar
  • smoother2histval was not keeping track of whether previously stored results were generated with loglinear
  • The initval_file option was not supporting Dynare’s translation of a model into a one lead/lag-model via auxiliary variables

References

  • Andreasen et al. (2018): “The pruned state-space system for non-linear DSGE models: Theory and empirical applications,” Review of Economic Studies, 85(1), 1–49

  • Angelini, Bokan, Christoffel, Ciccarelli and Zimic (2019): “Introducing ECB-BASE: The blueprint the new ECB semi-structural model for the euro area,” ECB Working Paper no. 2315

  • Born and Pfeifer (2014): “Policy risk and the business cycle,” Journal of Monetary Economics, 68, 68–85

  • Brayton, Davis and Tulip (2000): “Polynomial adjustment costs in FRB/US,” Unpublished manuscript

  • Brayton, Laubach, and Reifschneider (2014): “The FRB/US Model: A tool for macroeconomic policy analysis,” FEDS Notes. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/2380-7172.0012

  • Cuba-Borda, Guerrieri, Iacoviello, and Zhong (2019): “Likelihood evaluation of models with occasionally binding constraints,” Journal of Applied Econometrics, 34(7), 1073–1085

  • Giovannini, Pfeiffer, and Ratto (2021): “Efficient and robust inference of models with occasionally binding constraints,” Working Paper 2021-03, Joint Research Centre, European Commission

  • Guerrieri and Iacoviello (2015): “OccBin: A toolkit for solving dynamic models with occasionally binding constraints easily,” Journal of Monetary Economics, 70, 22–38

  • Mutschler (2018): “Higher-order statistics for DSGE models,” Econometrics and Statistics, 6(C), 44–56