Congratulations to the team for this major release full of new features!

See you tomorrow for those who will be at the Summer School,

Le dimanche 11 juin 2017 à 00:58 +0200, Stéphane Adjemian a écrit :
Dear Dynare users and friends,

We are pleased to announce the release of Dynare 4.5.0.

This major release adds new features and fixes various bugs.

The Windows packages are already available for download at:

 http://www.dynare.org/download/dynare-stable

The Mac and Debian/Ubuntu packages should follow soon.

All users are strongly encouraged to upgrade.

This release is compatible with MATLAB versions ranging from 7.3
(R2006b) to 9.2 (R2017a) and with GNU Octave version 4.2.1.


Here is the list of major user-visible changes:


 - Ramsey policy

   + Added command `ramsey_model` that builds the expanded model with
     FOC conditions for the planner's problem but doesn't perform any
     computation. Usefull to compute Ramsey policy in a perfect
     foresight model,

   + `ramsey_policy` accepts multipliers in its variable list and
     displays results for them.


 - Perfect foresight models

   + New commands `perfect_foresight_setup` (for preparing the
     simulation) and `perfect_foresight_solver` (for computing it). The
     old `simul` command still exist and is now an alias for
     `perfect_foresight_setup` + `perfect_foresight_solver`. It is no
     longer possible to manipulate by hand the contents of
     `oo_.exo_simul` when using `simul`. People who want to do
     it must first call `perfect_foresight_setup`, then do the
     manipulations, then call `perfect_foresight_solver`,

   + By default, the perfect foresight solver will try a homotopy
     method if it fails to converge at the first try. The old behavior
     can be restored with the `no_homotopy` option,

   + New option `stack_solve_algo=7` that allows specifying a
     `solve_algo` solver for solving the model,

   + New option `solve_algo` that allows specifying a solver for
     solving the model when using `stack_solve_algo=7`,

   + New option `lmmcp` that solves the model via a Levenberg-Marquardt
     mixed complementarity problem (LMMCP) solver,

   + New option `robust_lin_solve` that triggers the use of a robust
     linear solver for the default `solve_algo=4`,

   + New options `tolf` and `tolx` to control termination criteria of
     solvers,

   + New option `endogenous_terminal_period` to `simul`,

   + Added the possibility to set the initial condition of the
     (stochastic) extended path simulations with the histval block.


 - Optimal simple rules

   + Saves the optimal value of parameters to `oo_.osr.optim_params`,

   + New block `osr_params_bounds` allows specifying bounds for the
     estimated parameters,

   + New option `opt_algo` allows selecting different optimizers while
     the new option `optim` allows specifying the optimizer options,

   + The `osr` command now saves the names, bounds, and indices for the
     estimated parameters as well as the indices and weights of the
     variables entering the objective function into `M_.osr`.


 - Forecasts and Smoothing

   + The smoother and forecasts take uncertainty about trends and means
     into account,

   + Forecasts accounting for measurement error are now saved in fields
     of the form `HPDinf_ME` and `HPDsup_ME`,

   + New fields `oo_.Smoother.Trend` and `oo_.Smoother.Constant` that
     save the trend and constant parts of the smoothed variables,

   + new field `oo_.Smoother.TrendCoeffs` that stores the trend
     coefficients.

   + Rolling window forecasts allowed in `estimation` command by
     passing a vector to `first_obs`,

   + The `calib_smoother` command now accepts the `loglinear`,
     `prefilter`, `first_obs` and `filter_decomposition` options.


 - Estimation

   + New options: `logdata`, `consider_all_endogenous`,
     `consider_only_observed`, `posterior_max_subsample_draws`,
     `mh_conf_sig`, `diffuse_kalman_tol`, `dirname`, `nodecomposition`

   + `load_mh_file` and `mh_recover` now try to load chain's proposal
density,

   + New option `load_results_after_load_mh` that allows loading some
     posterior results from a previous run if no new MCMC draws are
     added,

   + New option `posterior_nograph` that suppresses the generation of
     graphs associated with Bayesian IRFs, posterior smoothed objects,
     and posterior forecasts,

   + Saves the posterior density at the mode in
     `oo_.posterior.optimization.log_density`,

   + The `filter_covariance` option now also works with posterior
     sampling like Metropolis-Hastings,

   + New option `no_posterior_kernel_density` to suppress computation
     of kernel density of posterior objects,

   + Recursive estimation and forecasting now provides the individual
     `oo_` structures for each sample in `oo_recursive_`,

   + The `trace_plot` command can now plot the posterior density,

   + New command `generate_trace_plots` allows generating all trace
     plots for one chain,

   + New commands `prior_function` and `posterior_function` that
     execute a user-defined function on parameter draws from the
     prior/posterior distribution,

   + New option `huge_number` for replacement of infinite bounds with
     large number during `mode_compute`,

   + New option `posterior_sampling_method` allows selecting the new
     posterior sampling options:
     `tailored_random_block_metropolis_hastings` (Tailored randomized
     block (TaRB) Metropolis-Hastings), `slice` (Slice sampler),
     `independent_metropolis_hastings` (Independent
     Metropolis-Hastings),

   + New option `posterior_sampler_options` that allow controlling the
     options of the `posterior_sampling_method`, its `scale_file`-
option
     pair allows loading the `_mh_scale.mat`-file storing the tuned
     scale factor from a previous run of `mode_compute=6`,

   + New option `raftery_lewis_diagnostics` that computes Raftery/Lewis
     (1992) convergence diagnostics,

   + New option `fast_kalman_filter` that provides fast Kalman filter
     using Chandrasekhar recursions as described in Ed Herbst (2015),

   + The `dsge_var` option now saves results at the posterior mode into
     `oo_.dsge_var`,

   + New option `smoothed_state_uncertainty` to provide the uncertainty
     estimate for the smoothed state estimate from the Kalman smoother,

   + New prior density: generalized Weibull distribution,

   + Option `mh_recover` now allows continuing a crashed chain at the
     last save mh-file,

   + New option `nonlinear_filter_initialization` for the
     `estimation` command. Controls the initial covariance matrix
     of the state variables in nonlinear filters.

   + The `conditional_variance_decomposition` option now displays
     output and stores it as a LaTeX-table when the `TeX` option is
     invoked,

   + The `use_calibration` to `estimated_params_init` now also works
     with ML,

   + Improved initial estimation checks.


 - Steady state

   + The default solver for finding the steady state is now a
     trust-region solver (can be triggered explicitly with option
     `solve_algo=4`),

   + New options `tolf` and `tolx` to control termination criteria of
     solver,

   + The debugging mode now provides the termination values in steady
     state finding.


 - Stochastic simulations

   + New options `nodecomposition`,

   + New option `bandpass_filter` to compute bandpass-filtered
     theoretical and simulated moments,

   + New option `one_sided_hp_filter` to compute one-sided HP-filtered
     simulated moments,

   + `stoch_simul` displays a simulated variance decomposition when
     simulated moments are requested,

   + `stoch_simul` saves skewness and kurtosis into respective fields
     of `oo_` when simulated moments have been requested,

   + `stoch_simul` saves the unconditional variance decomposition in
     `oo_.variance_decomposition`,

   + New option `dr_display_tol` that governs omission of small terms
     in display of decision rules,

   + The `stoch_simul` command now prints the displayed tables as LaTeX
     code when the new `TeX` option is enabled,

   + The `loglinear` option now works with lagged and leaded exogenous
     variables like news shocks,

   + New option `spectral_density` that allows displaying the spectral
     density of (filtered) endogenous variables,

   + New option `contemporaneous_correlation` that allows saving
     contemporaneous correlations in addition to the covariances.


 - Identification

   + New options `diffuse_filter` and `prior_trunc`,

   + The `identification` command now supports correlations via
     simulated moments,


 - Sensitivity analysis

   + New blocks `irf_calibration` and `moment_calibration`,

   + Outputs LaTeX tables if the new `TeX` option is used,

   + New option `relative_irf` to `irf_calibration` block.


 - Conditional forecast

   + Command `conditional_forecast` now takes into account `histval`
     block if present.


 - Shock decomposition

   + New option `colormap` to `shocks_decomposition` for controlling
     the color map used in the shocks decomposition graphs,

   + `shocks_decomposition` now accepts the `nograph` option,

   + New command `realtime_shock_decomposition` that for each period
`T= [presample,...,nobs]`
     allows computing the:

     * realtime historical shock decomposition `Y(t|T)`, i.e. without
observing data in `[T+1,...,nobs]`

     * forecast shock decomposition `Y(T+k|T)`

     * realtime conditional shock decomposition `Y(T+k|T+k)-Y(T+k|T)`

   + New block `shock_groups` that allows grouping shocks for the
     `shock_decomposition` and `realtime_shock_decomposition` commands,

   + New command `plot_shock_decomposition` that allows plotting the
     results from `shock_decomposition` and
     `realtime_shock_decomposition` for different vintages and shock
     groupings.


 - Macroprocessor

   + Can now pass a macro-variable to the `@#include` macro directive,

   + New preprocessor flag `-I`, macro directive `@#includepath`, and
     dynare config file block `[paths]` to pass a search path to the
     macroprocessor to be used for file inclusion via `@#include`.


 - Command line

   + New option `onlyclearglobals` (do not clear JIT compiled functions
     with recent versions of Matlab),

   + New option `minimal_workspace` to use fewer variables in the
     current workspace,

   + New option `params_derivs_order` allows limiting the order of the
     derivatives with respect to the parameters that are calculated by
     the preprocessor,

   + New command line option `mingw` to support the MinGW-w64 C/C++
     Compiler from TDM-GCC for `use_dll`.


 - dates/dseries/reporting classes

   + New methods `abs`, `cumprod` and `chain`,

   + New option `tableRowIndent` to `addTable`,

   + Reporting system revamped and made more efficient, dependency on
     matlab2tikz has been dropped.


 - Optimization algorithms

   + `mode_compute=2` Uses the simulated annealing as described by
     Corana et al. (1987),

   + `mode_compute=101` Uses SOLVEOPT as described by Kuntsevich and
     Kappel (1997),

   + `mode_compute=102` Uses `simulannealbnd` from Matlab's Global
     Optimization Toolbox (if available),

   + New option `silent_optimizer` to shut off output from mode
     computing/optimization,

   + New options `verbosity` and `SaveFiles` to control output and
     saving of files during mode computing/optimization.


 - LaTeX output

   + New command `write_latex_original_model`,

   + New option `write_equation_tags` to `write_latex_dynamic_model`
     that allows printing the specified equation tags to the generate
     LaTeX code,

   + New command `write_latex_parameter_table` that writes the names
and
     values of model parameters to a LaTeX table,

   + New command `write_latex_prior_table` that writes the descriptive
     statistics about the prior distribution to a LaTeX table,

   + New command `collect_latex_files` that creates one compilable
LaTeX
     file containing all TeX-output.


 - Misc.

   + Provides 64bit preprocessor,

   + Introduces new path management to avoid conflicts with other
     toolboxes,

   + Full compatibility with Matlab 2014b's new graphic interface,

   + When using `model(linear)`, Dynare automatically checks
     whether the model is truly linear,

   + `usedll`, the `msvc` option now supports `normcdf`, `acosh`,
     `asinh`, and `atanh`,

   + New parallel option `NumberOfThreadsPerJob` for Windows nodes that
     sets the number of threads assigned to each remote MATLAB/Octave
     run,

   + Improved numerical performance of
     `schur_statespace_transformation` for very large models,

   + The `all_values_required` option now also works with `histval`,

   + Add missing `horizon` option to `ms_forecast`,

   + BVAR now saves the marginal data density in
     `oo_.bvar.log_marginal_data_density` and stores prior and
     posterior information in `oo_.bvar.prior` and
     `oo_.bvar.posterior`.



* Bugs and problems identified in version 4.4.3 and that have been
fixed in version 4.5.0:


 - BVAR models

   + `bvar_irf` could display IRFs in an unreadable way when they moved
from
     negative to positive values,

   + In contrast to what is stated in the documentation, the confidence
interval
     size `conf_sig` was 0.6 by default instead of 0.9.


 - Conditional forecasts

   + The `conditional_forecast` command produced wrong results in
calibrated
     models when used at initial values outside of the steady state
(given with
     `initval`),

   + The `plot_conditional_forecast` option could produce unreadable
figures if
     the areas overlap,

   + The `conditional_forecast` command after MLE crashed,

   + In contrast to what is stated in the manual, the confidence
interval size
     `conf_sig` was 0.6 by default instead of 0.8.

   + Conditional forecasts were wrong when the declaration of
endogenous
     variables was not preceeding the declaration of the exogenous
     variables and parameters.


 - Discretionary policy

   + Dynare allowed running models where the number of instruments did
not match
     the number of omitted equations,

   + Dynare could crash in some cases when trying to display the
solution,

   + Parameter dependence embedded via a `steady_state` was not taken
into
     account, typically resulting in crashes.

 - dseries class

   + When subtracting a dseries object from a number, the number was
instead
     subtracted from the dseries object.


 - DSGE-VAR models

   + Dynare crashed when estimation encountered non-finite values in
the Jacobian
     at the steady state,

   + The presence of a constant was not considered for degrees of
freedom
     computation of the Gamma function used during the posterior
computation; due
     to only affecting the constant term, results should be be
unaffected, except
     for model_comparison when comparing models with and without.


 - Estimation command

   + In contrast to what was stated in the manual, the confidence
interval size
     `conf_sig` for `forecast` without MCMC was 0.6 by default instead
of 0.9,

   + Calling estimation after identification could lead to crashes,

   + When using recursive estimation/forecasting and setting some
elements of
     `nobs` to be larger than the number of observations T in the data,
     `oo_recursive_` contained additional cell entries that simply
repeated the
     results obtained for `oo_recursive_T`,

   + Computation of Bayesian smoother could crash for larger models
when
     requesting `forecast` or `filtered_variables`,

   + Geweke convergence diagnostics were not computed on the full MCMC
chain when
     the `load_mh_file` option was used,

   + The Geweke convergence diagnostics always used the default
`taper_steps` and
   `geweke_interval`,

   + Bayesian IRFs (`bayesian_irfs` option) could be displayed in an
unreadable
     way when they move from negative to positive values,

   + If `bayesian_irfs` was requested when `mh_replic` was too low to
compute
     HPDIs, plotting was crashing,

   + The x-axis value in `oo_.prior_density` for the standard deviation
and
     correlation of measurement errors was written into a field
     `mearsurement_errors_*` instead of `measurement_errors_*`,

   + Using a user-defined `mode_compute` crashed estimation,

   + Option `mode_compute=10` did not work with infinite prior bounds,

   + The posterior variances and covariances computed by
`moments_varendo` were
     wrong for very large models due to a matrix erroneously being
filled up with
     zeros,

   + Using the `forecast` option with `loglinear` erroneously added the
unlogged
     steady state,

   + When using the `loglinear` option the check for the presence of a
constant
     was erroneously based on the unlogged steady state,

   + Estimation of `observation_trends` was broken as the trends
specified as a
     function of deep parameters were not correctly updated during
estimation,

   + When using `analytic_derivation`, the parameter values were not
set before
     testing whether the steady state file changes parameter values,
leading to
     subsequent crashes,

   + If the steady state of an initial parameterization did not solve,
the
     observation equation could erroneously feature no constant when
the
     `use_calibration` option was used,

   + When computing posterior moments, Dynare falsely displayed that
moment
     computations are skipped, although the computation was performed
correctly,

   + If `conditional_variance_decomposition` was requested, although
all
     variables contain unit roots, Dynare crashed instead of providing
an error
     message,

   + Computation of the posterior parameter distribution was
erroneously based
     on more draws than specified (there was one additional draw for
every Markov
     chain),

   + The estimation option `lyapunov=fixed_point` was broken,

   + Computation of `filtered_vars` with only one requested step
crashed Dynare,

   + Option `kalman_algo=3` was broken with non-diagonal measurement
error,

   + When using the diffuse Kalman filter with missing observations, an
additive
     factor log(2*pi) was missing in the last iteration step,

   + Passing of the `MaxFunEvals` and `InitialSimplexSize` options to
     `mode_compute=8` was broken,

   + Bayesian forecasts contained initial conditions and had the wrong
length in
     both plots and stored variables,

   + Filtered variables obtained with `mh_replic=0`, ML, or
     `calibrated_smoother` were padded with zeros at the beginning and
end and
     had the wrong length in stored variables,

   + Computation of smoothed measurement errors in Bayesian estimation
was broken,

   + The `selected_variables_only` option (`mh_replic=0`, ML, or
     `calibrated_smoother`) returned wrong results for smoothed,
updated, and
     filtered variables,

   + Combining the `selected_variables_only` option with forecasts
obtained
     using `mh_replic=0`, ML, or `calibrated_smoother` leaded to
crashes,

   + `oo_.UpdatedVariables` was only filled when the `filtered_vars`
option was specified,

   + When using Bayesian estimation with `filtered_vars`, but without
     `smoother`, then `oo_.FilteredVariables` erroneously also
contained filtered
     variables at the posterior mean as with `mh_replic=0`,

   + Running an MCMC a second time in the same folder with a different
number of
     iterations could result in crashes due to the loading of stale
files,

   + Results displayed after Bayesian estimation when not specifying
     the `smoother` option were based on the parameters at the mode
     from mode finding instead of the mean parameters from the
     posterior draws. This affected the smoother results displayed, but
     also calls to subsequent command relying on the parameters stored
     in `M_.params` like `stoch_simul`,

   + The content of `oo_.posterior_std` after Bayesian estimation was
based on
     the standard deviation at the posterior mode, not the one from the
MCMC, this
     was not consistent with the reference manual,

   + When the initialization of an MCMC run failed, the metropolis.log
file was
     locked, requiring a restart of Matlab to restart estimation,

   + If the posterior mode was right at the corner of the prior bounds,
the
     initialization of the MCMC erroneously crashed,

   + If the number of dropped draws via `mh_drop` coincided with the
number of
     draws in a `_mh'-file`, `oo_.posterior.metropolis.mean` and
     `oo_.posterior.metropolis.Variance` were NaN.


 - Estimation and calibrated smoother

   + When using `observation_trends` with the `prefilter` option, the
mean shift
     due to the trend was not accounted for,

   + When using `first_obs`>1, the higher trend starting point of
     `observation_trends` was not taken into account, leading, among
other things,
     to problems in recursive forecasting,

   + The diffuse Kalman smoother was crashing if the forecast error
variance
     matrix becomes singular,

   + The multivariate Kalman smoother provided incorrect state
estimates when
     all data for one observation are missing,

   + The multivariate diffuse Kalman smoother provided incorrect state
estimates
     when the `Finf` matrix becomes singular,

   + The univariate diffuse Kalman filter was crashing if the initial
covariance
     matrix of the nonstationary state vector is singular,


 - Forecats

   + In contrast to what is stated in the manual, the confidence
interval size
     `conf_sig` was 0.6 by default instead of 0.9.

   + Forecasting with exogenous deterministic variables provided wrong
decision
     rules, yielding wrong forecasts.

   + Forecasting with exogenous deterministic variables crashed when
the
     `periods` option was not explicitly specified,

   + Option `forecast` when used with `initval` was using the initial
values in
     the `initval` block and not the steady state computed from these
initial
     values as the starting point of forecasts.


 - Global Sensitivity Analysis

   + Sensitivity with ML estimation could result in crashes,

   + Option `mc` must be forced if `neighborhood_width` is used,

   + Fixed dimension of `stock_logpo` and `stock_ys`,

   + Incomplete variable initialization could lead to crashes with
`prior_range=1`.


 - Indentification

   + Identification did not correctly pass the `lik_init` option,
     requiring the manual setting of `options_.diffuse_filter=1` in
     case of unit roots,

   + Testing identification of standard deviations as the only
     parameters to be estimated with ML leaded to crashes,

   + Automatic increase of the lag number for autocovariances when the
     number of parameters is bigger than the number of non-zero moments
     was broken,

   + When using ML, the asymptotic Hessian was not computed,

   + Checking for singular values when the eigenvectors contained only
     one column did not work correctly,


 - Model comparison

   + Selection of the `modifiedharmonicmean` estimator was broken,


 - Optimal Simple Rules

   + When covariances were specified, variables that only entered with
     their variance and no covariance term obtained a wrong weight,
     resulting in wrong results,

   + Results reported for stochastic simulations after `osr` were based
     on the last parameter vector encountered during optimization,
     which does not necessarily coincide with the optimal parameter
     vector,

   + Using only one (co)variance in the objective function resulted in
crashes,

   + For models with non-stationary variables the objective function
was computed wrongly.


 - Ramsey policy

   + If a Lagrange multiplier appeared in the model with a lead or a
lag
     of more than one period, the steady state could be wrong.

   + When using an external steady state file, incorrect steady states
     could be accepted,

   + When using an external steady state file with more than one
     instrument, Dynare crashed,

   + When using an external steady state file and running `stoch_simul`
     after `ramsey_planner`, an incorrect steady state was used,

   + When the number of instruments was not equal to the number of
     omitted equations, Dynare crashed with a cryptic message,

   + The `planner_objective` accepted `varexo`, but ignored them for
computations,


 - Shock decomposition

   + Did not work with the `parameter_set=calibration` option if an
     `estimated_params` block is present,

   + Crashed after MLE.


 - Perfect foresight models

   + The perfect foresight solver could accept a complex solution
     instead of continuing to look for a real-valued one,

   + The `initval_file` command only accepted column and not row
vectors,

   + The `initval_file` command did not work with Excel files,

   + Deterministic simulations with one boundary condition crashed in
     `solve_one_boundary` due to a missing underscore when passing
     `options_.simul.maxit`,

   + Deterministic simulation with exogenous variables lagged by more
     than one period crashed,

   + Termination criterion `maxit` was hard-coded for `solve_algo=0`
     and could no be changed,

   + When using `block`/`bytecode`, relational operators could not be
enforced,

   + When using `block` some exceptions were not properly handled,
     leading to code crashes,

   + Using `periods=1` crashed the solver (bug only partially fixed).


 - Smoothing

   + The univariate Kalman smoother returned wrong results when used
     with correlated measurement error,

   + The diffuse smoother sometimes returned linear combinations of the
     smoothed stochastic trend estimates instead of the original trend
     estimates.

 - Perturbation reduced form

   + In contrast to what is stated in the manual, the results of the
     unconditional variance decomposition were only stored in
     `oo_.gamma_y(nar+2)`, not in `oo_.variance_decomposition`,

   + Dynare could crash when the steady state could not be computed
     when using the `loglinear` option,

   + Using `bytcode` when declared exogenous variables were not
     used in the model leaded to crashes in stochastic simulations,

   + Displaying decision rules involving lags of auxiliary variables of
     type 0 (leads>1) crashed.

   + The `relative_irf` option resulted in wrong output at `order>1` as
     it implicitly relies on linearity.


 - Displaying of the MH-history with the `internals` command crashed
   if parameter names did not have same length.

 - Dynare crashed when the user-defined steady state file returned an
   error code, but not an conformable-sized steady state vector.

 - Due to a bug in `mjdgges.mex` unstable parameter draws with
   eigenvalues up to 1+1e-6 could be accepted as stable for the
   purpose of the Blanchard-Kahn conditions, even if `qz_criterium<1`.

 - The `use_dll` option on Octave for Windows required to pass a
   compiler flag at the command line, despite the manual stating this
   was not necessary.

 - Dynare crashed for models with `block` option if the Blanchard-Kahn
   conditions were not satisfied instead of generating an error
   message.

 - The `verbose` option did not work with `model(block)`.

 - When falsely specifying the `model(linear)` for nonlinear models,
   incorrect steady states were accepted instead of aborting.

 - The `STEADY_STATE` operator called on model local variables
   (so-called pound variables) did not work as expected.

 - The substring operator in macro-processor was broken. The
   characters of the substring could be mixed with random characters
   from the memory space.

 - Block decomposition could sometimes cause the preprocessor to crash.

 - A bug when external functions were used in model local variables
   that were contained in equations that required auxiliary
   variable/equations led to crashes of Matlab.

 - Sampling from the prior distribution for an inverse gamma II
   distribution when `prior_trunc>0` could result in incorrect
   sampling.

 - Sampling from the prior distribution for a uniform distribution
   when `prior_trunc>0` was ignoring the prior truncation.

 - Conditional forecasts were wrong when the declaration of endogenous
   variables was not preceeding the declaration of the exogenous
   variables and parameters.


On behalf of the Dynare Team,
Stéphane.


-- 
Sébastien Villemot
Économiste
Sciences Po, OFCE