Dear Colleagues,
Applications are now open for the Dynare Masterclass 2026, the next
evolution of our annual Dynare Summer School.
This week-long (Mo-Fr), hands-on program covers the full DSGE modeling
workflow using Dynare 7 — from simulating policy models through
estimating New Keynesian models (GMM, IRF matching, Bayesian) to
heterogeneous agent (HANK) models.
Participants get daily open lab time with direct feedback from Dynare
developers on their own work.
The Masterclass welcomes
- Master/PhD students
- postdoctoral researchers
- professors
- junior as well as senior economists and professionals
at central banks, ministries, universities, and research institutes —
whether new to Dynare or looking to deepen their skills.
Event Details
- Dates: 6–10 July 2026
- Location: École Normale Supérieure, Paris
- Application deadline: 20 April 2026
- Acceptance notifications: 2 May 2026
- Full announcement:
https://www.dynare.org/events/dynare-masterclass-2026/
- Preliminary program:
https://git.dynare.org/Dynare/dynare/-/wikis/Dynare-Masterclass-2026
Fees
- Academic: 500 EUR
- Institutional/Professional: 1,800 EUR
- Fees include all meals, refreshments, and the social dinner on
Thursday.
How to Apply
Please fill out the form and upload your CV and a recent research paper
or work-in-progress (extended abstract, presentation) here:
https://framaforms.org/dynare-masterclass-2026-1774368113
We look forward to seeing you in Paris.
Best regards,
The Dynare Team
Dear Dynare friends,
We are pleased to announce the release of Dynare 7.0. This major
release adds new features and fixes various bugs.
The Windows, macOS, MATLAB Online and source packages are already
available for download at the Dynare website [1]. This release is
compatible with MATLAB versions ranging from 9.8 (R2020a) to 25.2
(R2025b), and with GNU Octave versions ranging from 8.4.0 to 11.1.0
(NB: the Windows package requires version 11.1.0 specifically).
Major user-visible changes *
New framework for solving and simulating models featuring rich
heterogeneity
(which includes HANK models)
The solution technique for the dynamics combines elements from
Bhandari,
Bourany, Evans, and Golosov (2023) and Auclert, Bardóczy, Rognlie,
and
Straub (2021).
The steady state can be computed through a time iteration method
within
Dynare, or computed outside of Dynare and then loaded therein.
New statements heterogeneity_dimension, var(heterogeneity=NAME),
varexo(heterogeneity=NAME), model(heterogeneity=NAME),
shocks(heterogeneity=NAME), and the SUM() aggregation operator for
declaring heterogenous features.
New heterogeneity_compute_steady_state command to compute the steady
state numerically.
New heterogeneity_load_steady_state command to load a pre-computed
steady
state (policy functions, discretized shocks, stationary
distribution) from
a MAT file.
New heterogeneity_solve command to compute a first-order linearized
solution.
New heterogeneity_simulate command to compute impulse response
functions
and stochastic simulations, with support for both unanticipated
shocks and
anticipated news shock sequences.
Example .mod files include: Krusell-Smith (krusell_smith.mod),
a one-asset HANK (hank_one_asset.mod), a two-asset HANK
(hank_two_assets.mod), with steady state computation variants. The
models
are the same as in the Sequence-Space-Jacobian package
(shade-econ/sequence-jacobian on GitHub).
*
Skew normal shocks and closed skew normal (CSN) distribution support
New skew keyword in shocks and estimated_params blocks to declare
skewness.
Added support for skew normal shocks in stoch_simul command.
Added support for skew normal shocks in estimation command by means
of
the Pruned Skewed Kalman Filter and Smoother of Guljanov, Mutschler,
and
Trede (forthcoming).
*
Posterior samplers
New posterior samplers (available via the posterior_sampling_method
option of the estimation command):
The Dynamic Striated Metropolis Hastings sampler proposed by
Waggoner,
Wu, and Zha (2016) is now available under the value 'dsmh';
The Differential-Independence Mixture Ensemble (“DIME”) MCMC sampler
of
Boehl (2022) is now available under the value 'dime_mcmc'.
New posterior_sampler_option options for
posterior_sampling_method='slice':
save_iter_info_file to shut off saving of iteration information to a
file;
fast_likelihood_evaluation_for_rejection to reject poor draws in
OccBin
more efficiently;
use_prior_draws to use the prior as the proposal within slice
(Gibbs)
sampling for up to 10 draws.
Sequential Monte Carlo (SMC) sampler:
Trace plots now allow plotting the particles;
Now supports the bayesian_irf, moments_varendo, and smoother
options.
*
New functionality for leveraging MATLAB’s Parallel Computing Toolbox
(PCT)
for accelerating computations
Computations that can be thus parallelized:
running multiple Metropolis-Hasting chains in parallel (as set by
the
mh_nblocks option);SMC sampler initialization;parallel likelihood
evaluations in SMC samplers.
Controlled by a new use_pct option of the estimation command.
If Dynare detects MATLAB R2024b or later with the PCT and a valid
license,
these supported tasks are parallelized by default (unless the user
sets
use_pct to false); random seeds are precomputed so results are
reproducible between serial and parallel execution (thanks to Eduard
Benet
Cerda from MathWorks).
*
New options to the estimation command:
estimate_initial_states_endogenous_prior, that estimates initial
states
along with model parameters, for stationary models;
frequentist_smoother that allows shutting off the smoother in cases
the
Bayesian smoother is not applicable;
use_univariate_smoother_if_singularity_is_detected to use the
univariate
Kalman smoother in case stochastic singularity is detected.
*
OccBin
New piecewise linear particle filter (PPF).
New posterior importance sampling feature: allows to bootstrap
posterior
draws obtained from e.g. linear KF estimation, using PKF or PPF and
check
effective sample size and estimator bias.
*
Improvements to the the method_of_moments command:
It now supports estimation with discretionary optimal policy;
With mom_method=SMM it now supports the bandpass_filter,
one_sided_hp_filter, and hp_filter options to match filtered
moments;
It now allows matching first moments even when the prefilter option
is
used to obtain centered higher order moments.
*
Perfect foresight
Improvements to perfect foresight solver with iterative linear
solvers,
i.e., GMRES (stack_solve_algo=2) and BiCGStab (stack_solve_algo=3):
New preconditioner option to the perfect_foresight_solver and
perfect_foresight_with_expectation_errors_solver commands, which
controls the preconditioner used with GMRES or BiCGStab;
The new default, preconditioner=first_iter_lu, does a LU
decomposition
with complete pivoting at the first Newton iteration, and uses it as
a
preconditioner in further iterations;
The value preconditioner=block_diagonal_lu computes a block diagonal
preconditioner based on a LU decomposition for only a few periods of
the
stacked system;
The old default, using an incomplete LU decomposition, remains
available
under precondition=incomplete_lu;
The first_iter_lu and block_diagonal_lu preconditioners deliver a
significant performance improvement over the old default
(incomplete_lu) and make this algorithm competitive and in some
cases
faster than other perfect foresight solvers;
Other new options that control the behaviour of the iterative
solvers:
iter_tol, iter_maxit, gmres_restart, block_diagonal_lu_maxlu,
block_diagonal_lu_nperiods, block_diagonal_lu_nlu,
block_diagonal_lu_relu.
New perfect foresight solvers:
ParU from SuiteSparse, a direct sparse LU solver, available under
the
stack_solve_algo=8 option;
Panua PARDISO direct sparse LU solver, available under the
stack_solve_algo=9 option (license file has to be provided by the
user);
Panua PARDISO direct-iterative solver using CGS, available under the
stack_solve_algo=10 option (license file has to be provided by the
user).
New perfect_foresight_controlled_paths block that gives the
possibility
of controlling the value of some endogenous variables for some
simulation
periods (in which case some exogenous variables must be left free
and are
thus solved for by Dynare, to avoid over-determination of the
problem). Can
be used either in a pure perfect foresight context, or in perfect
foresight
with expectation errors.
Perfect foresight integration with dates/dseries:
New options first_simulation_period and last_simulation_period to
perfect_foresight_setup and
perfect_foresight_with_expectation_errors_setup commands, to specify
the real dates for the simulation (e.g. 2024Q1);
The periods keyword of the shocks and mshocks blocks now accept
real dates;
The learnt_in option of the shocks, mshocks and endval blocks now
accept real dates;
When real dates are used, the perfect_foresight_solver and
perfect_foresight_with_expectation_errors_solver commands return the
simulation as a dseries object in the workspace variable
Simulated_time_series (which contains both endogenous and exogenous
variables).
The perfect_foresight_with_expectation_errors_setup command can now
be
combined with a histval block or a histval_file command.
New endval_steady_nocheck option to the perfect_foresight_solver and
perfect_foresight_with_expectation_errors_solver commands, to skip
checking the terminal steady state values when they are provided
explicitly
either by a steady state file or a steady_state_model block (in
combination with the endval_steady option for the
perfect_foresight_solver case). This is useful for models with unit
roots
as, in this case, the steady state is not unique or doesn’t exist.
Using several endval blocks in the same file is now supported, in
which
case they are concatenated.
Option values stack_solve_algo={2,3} are now available without block
nor bytecode.
New shock_paths block for describing a perfect foresight scenario in
a
more compact and flexible way, which supersedes the shocks, mshocks
and
endval blocks. Additionally, a new database command has been added
to
refer to external databases that can be used from within
shock_paths.
*
Steady state
The steady command now accepts the noprint option.
The steady command now accepts the non_zero option (with the same
meaning as in the resid command).
Option values solve_algo={6,7,8} are now available without block nor
bytecode.
*
Decision rules (reduced form solution)
New dr=aim option to the stoch_simul, estimation and
method_of_moments commands (equivalent to the former aim_solver
option,
which is now deprecated).
New dr_cycle_reduction_maxiter option to the stoch_simul, estimation
and method_of_moments commands, to control the maximum number of
iterations when the cycle reduction algorithm is used.
*
Extended path
New option use_first_order_solution to the the extended_path
command.
It is now possible to have zero-variance shocks.
*
Forecasting
The forecast command now also works at higher order and supports the
pruning option.
Option forecast_type to shock_decomposition to allow shock
decomposing
conditional and unconditional forecasts.
*
Syntax
More flexible syntax for specifying complementarity conditions:
The complementarity condition is now specified between the
associated
equation and its closing semicolon, using the perpendicular symbol
as a
separator (the latter can be input either in UTF-8, as ⟂,
corresponding
to Unicode codepoint U+27C2; or alternatively as pure ASCII, as _|_,
i.e. a vertical bar enclosed within two underscores);
Both the lower and the upper bounds can be specified at the same
time in
a given complementarity condition;
Arbitrary functions of parameters can appear in the bounds;
The old syntax using the mcp equation tag is still supported but
deprecated;
Here is an example of an equation with an associated complementarity
condition, using the new syntax:
mu = 0 ⟂ 0 < i < 1+2*alpha;
Model-local variables (i.e. defined with a # sign) can now appear
with
a lead or a lag in the model block (the expression by which they
will be
replaced will be shifted accordingly).
The heteroskedastic_shocks block now supports dates syntax.
*
The osr command now supports using a planner_objective function to
conduct welfare analysis at higher order.
*
Significant performance improvements for:
Higher-order perturbation solution under Windows;
Estimation with the analytic_derivation option.
*
New debugging information
The multivariate Kalman filter smoother now allows using the debug
option
to display information on linear combinations causing stochastic
singularity.
dynare_minimize_objective now returns runtime, iterations, function
evaluations, exitflag, and messages from all optimizers.
The model_diagnostics command now:
checks for redundant equations based on the dynamic Jacobian;
displays equation tags of affected equations;
displays its information in a more readable way.
New debugging options to the perfect_foresight_solver command:
check_jacobian_singularity to check the singularity of the dynamic
Jacobian for stack_solve_algo option equal to 0, 2 or 3;
allow_nonfinite_values to prevent the automatic removal of nonfinite
values during Newton iterations, which may be useful for debugging
purposes.
*
dseries
New routines for converting annual data into quarterly data, and
quarterly
data into monthly data.
New --print-table option to the dplot command, for displaying the
results as a table instead of a plot.
The dseries constructor based on a table is now available under
Octave
(requires the datatypes package).
The dates and dseries classes now accept the string syntax (double
quotes) in their constructor and methods.
*
The calib_smoother command now plots the smoother results.
*
New preprocessor option output=first that instructs the preprocessor
to
only compute first-order dynamic derivatives if no other command in
the
.mod file requires higher-order ones.
*
New irfs(distribution) option to the prior command from the command-
line
interface, for computing the prior distribution of the impulse
response
functions of the endogenous variables.
*
New structure M_.state_var storing information on the state
variables in
both declaration and decision-rule order.
*
References:
Auclert, A., B. Bardóczy, M. Rognlie, and L. Straub (2021): “Using
the
Sequence‐Space Jacobian to solve and estimate heterogeneous‐agent
models,”
Econometrica, 89(5), 2375–2408.
Bhandari, A., T. Bourany, D. Evans, and M. Golosov (2023): “A
perturbational approach for approximating heterogeneous agent
models,”
NBER Working Papers, 31744.
Boehl, G. (2022): “DIME MCMC: A Swiss Army Knife for Bayesian
Inference”,
SSRN Electronic Journal, https://dx.doi.org/10.2139/ssrn.4250395
Guljanov, G., W. Mutschler and M. Trede (forthcoming): “Pruned
skewed
Kalman filter and smoother with application to DSGE models,” Journal
of
Economic Dynamics and Control
Waggoner, D. F., H. Wu and T. Zha (2016): “Striated
Metropolis–Hastings
sampler for high-dimensional models,” Journal of Econometrics,
192(2),
406–420.
Incompatible changes *
With the osr command, the legacy linear-quadratic approach with an
optim_weights block now requires explicitly setting order=1.
*
The default filename produced by the savemacro option of the dynare
command has changed. It now contains an underscore (instead of a
dash), for
better compatibility with MATLAB/Octave syntax.
*
Prior truncation is now shut off by default via setting
prior_trunc=0.
*
The infile option of the smoother2histval command has been removed.
An
equivalent functionality can be obtained with the outfile option of
the
smoother2histval command and the histval_file command.
*
The det_cond_forecast, init_plan, basic_plan and flip_plan commands
have been removed. A similar functionality is provided by the new
perfect_foresight_controlled_paths block (or alternatively by the
shock_paths block with and endogenize stanza).
*
The unit_root_vars command has been removed. It has been superseded
by the
diffuse_filter of the estimation command, and the nocheck option of
the
steady command.
*
Estimated standard errors internally now get a prefix stderr instead
of
just the name of the exogenous variable.
*
The deprecated dynare_sensitivity command has been removed, use the
sensitivity command instead.
*
The hp_ngrid option was removed, use
filtered_theoretical_moments_grid
instead.
*
Declaring dsge_prior_weight as a parameter has been removed. It will
be
automatically added when using the dsge_var option of the estimation
command.
Bugs that were present in 6.5 and that have been fixed in 7.0 *
The smoother2histval(outfile = …) command would store incorrect
values for
variables appearing with a lag of 2 or more.
*
Using several endval blocks in the same file could lead to incorrect
results (this setup was not supported, but there was unfortunately
no error
message preveting users from doing this). Proper support for
multiple
endval blocks has been added (they are now concatenated).
*
The prior moments and prior optimize commands would crash.
*
The matched_irfs block would not accept a single entry following the
weights keyword when there were multiple entries following the
periods
keyword, contrary to what the manual says.
*
The mode_compute=5 option of the estimation command would crash in
case
of an invalid initial parameter draw.
*
Uniform priors did not correctly check for invalid hyperparameter
specifications.
*
The method_of_moments command would crash upon trying to randomize
the
starting value for steady state finding.
*
Filtered theoretical autocorrelation matrices at lag>0 in
oo_.autocorr and
oo_.gamma_y were erroneously transposed.
*
The mh_recover and load_mh_file options of the estimation command
would
not correctly work if the value of the mh_nblocks option was
different from
what was originally run.
* The extended_path command would:
not return the initial condition in oo_.exo_simul;crash if NaN was
encountered in the solution;crash for purely forward or purely
backward models.
*
The ramsey_model command would crash in the presence of external
functions.
* The model_info command would:
not display the dynamic block structure if the block_static option
was
specified;sometimes crash when displaying the incidence matrix when
passed the
incidence option.
* When running the estimation command with measurement errors:
the logic for computing calibrated measurement error entries was
broken;the analytic_derivation option would crash.
*
Using Weibull priors in estimation could trigger infinite loops.
*
The generalized Weibull prior distribution did not correctly take
the third
hyperparameter shifting the lower bound into account.
*
The display of problematic variables in debug mode for the
perfect_foresight_solver command was broken.
*
OccBin posterior moments would crash when the smoother encountered
an
unsuccessful draw.
*
OccBin: when multiple solutions were found, the relaxation algorithm
would
crash.
*
The diffuse Kalman smoother would return incorrect smoothed state
uncertainty.
*
The block decomposition in a perfect foresight context would
occasionally
crash Octave.
*
Using the forecast command with a model solved at order=2 with
varexo_det, Dynare would return uncertainty bands based on a first-
order
approximation.
*
OccBin would not correctly sum the likelihood of detected multiple
solution
paths.
* The identification command would ignore the kalman_algo option.
On behalf of the Dynare Team,
--
Sébastien Villemot
Economist at CEPREMAP
Co-leader of the Dynare project
https://sebastien.villemot.name
[1] the Dynare website https://www.dynare.org/download/
Dear colleagues,
Due to several requests, we have decided to extend the submission
deadline for the 20th Annual Dynare Conference to
31 March 2026
The conference will be held in person at the Bank of Canada in Ottawa,
Canada, on 2-3 June 2026. It is organized by the Bank of Canada in
collaboration with DSGE-net and the Dynare project at CEPREMAP.
Keynote Speakers:
Juan F. Rubio-Ramírez (Emory University)
Serguei Maliar (Santa Clara University)
We invite submissions of theoretical, empirical, and quantitative
research on a wide range of macroeconomic topics. While the conference
is hosted by the Dynare project, we welcome all contributions regardless
of whether they use Dynare.
Important Dates:
New Submission Deadline: 31 March 2026
Notification of Acceptance: 10 April 2026 (expected)
Workshop on HANK models: 1 June 2026 (free for conference participants)
Conference Dates: 2-3 June 2026
Submission Procedure: Papers should be submitted via the online form
available on the conference
website:https://www.dynare.org/events/dynare-conference-2026/
There is no registration fee. Lunch and a conference dinner will be
provided for all participants. Please note that authors of accepted
papers are expected to cover their own travel and accommodation
expenses.
For any further inquiries, please contact: conference(a)dynare.org
We look forward to receiving your submissions and seeing you in Ottawa!
Best regards,
The Scientific Committee
Stéphane Adjemian, Anastasiia Antonova, Willi Mutschler, Chris Naubert,
Johannes Pfeifer, Sébastien Villemot, and Yang Zhang
