Dynare 4.6.0 Released
Posted on 20 February 2020We are pleased to announce the release of Dynare 4.6.0.
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 7.9 (R2009b) to 9.7 (R2019b), and with GNU Octave versions 5.2.0 (under Windows) and 4.4.1 (under macOS).
Major uservisible changes

Stochastic simulations

The perturbation method is now available at an arbitrary approximation order. In other words, the
order
option ofstoch_simul
accepts an arbitrary positive integer (of course, up to some modelspecific computational limit). 
New option
filtered_theoretical_moments_grid
ofstoch_simul
, that supersedeshp_ngrid
.


Estimation

Nonlinear estimation is now also available at an arbitrary approximation order. In other words, the
order
option ofestimation
accepts an arbitrary positive integer (of course, up to some modelspecific computational limit). 
Various improvements to particle filters.

It is now possible to estimate models under optimal policy (see below).

Variance decomposition of observables now accounts for measurement error.

New option
mh_tune_jscale
ofestimation
command for tuning the scale parameter of the proposal distribution of the Random Walk Metropolis Hastings. 
Added debugging info when parameters take a
NaN
orInf
value. 
Option
mode_compute=1
is now available under Octave.


Perfect foresight and extended path

A significant speed improvement should be noted on large models (when neither
bytecode
norblock
option is used). The stacked problem is now constructed using a dedicated machinecompiled library that greatly speeds up the process (in particular, the time spent in that step can become negligible when theuse_dll
option is used). 
New options
print
andnoprint
ofperfect_foresight_solver
command. 
Option
stack_solve_algo=2
is now available under Octave.


Steady state
 Option
solve_algo=7
is now available under Octave.
 Option

Optimal policy

The
ramsey_policy
command is now deprecated. It is superseded by successive calls toramsey_model
,stoch_simul
, andevaluate_planner_objective
(in this order). 
It is now possible to estimate a model under optimal policy (either Ramsey or discretionary) by running the
estimation
command after eitherramsey_model
ordiscretionary_policy
. It is however not yet possible to estimate parameters that appear in the discount factor of the social planner. 
Discretionary policy returns a more informative error message when the objective has nonzero derivatives with respect to some variables.


Identification

Added minimal system identification check of Komunjer and Ng (2011).

Added spectrum identification check of Qu and Tkachenko (2012).

Identification is now also available for approximation orders 2 and 3 with either analytical or numerical parameter derivatives. The relevant moments and spectrum are computed from the pruned state space system as in Mutschler (2015).

All tests (moments, spectrum, minimal system, strength) can be turned off.

More numerical options can be changed by the user.

Improved printing and storage (same folder) of results.


Sensitivity analysis

New
diffuse_filter
option to thedynare_sensitivity
command. 
Arbitrary expressions can now be passed for the interval boundaries in
irf_calibration
andmoment_calibration
. ⚠ This breaks the previous syntax, requiring that the lower/upper bounds be separated by commas.


Forecasting and smoothing

In
conditional_forecast_paths
, it is no longer required that all constrained paths be of the same length. There may now be a different number of controlled variables at each period. In that case, the order of declaration of endogenous controlled variables and ofcontrolled_varexo
matters: if the second endogenous variable is controlled for less periods than the first one, the secondcontrolled_varexo
isn’t set for the last periods. 
New option
parameter_set
to thecalib_smoother
command. 
⚠ The results of
conditional_forecast
command is now saved inoo_
(used to be in a file)


Shock decomposition

Added
fast_realtime
option to real time shock decomposition (deactivated by default, runs the smoother only twice: once for the last insample and once for the last outofsample data point). 
New
diff
,flip
,max_nrows
,plot_init_date
andplot_end_date
options toplot_shock_decomposition
. 
New
initial_decomposition_decomposition
command, for computing and plotting the decomposition of the effect of smoothed initial conditions of state variables. 
New
squeeze_shock_decomposition
command, for removing decompositions of variables that are not of interest. 
New
with_epilogue
option (common toshock_decomposition
,realtime_shock_decomposition
andinitial_condition_decomposition
). 
New
init2shocks
block to attribute initial conditions to shocks.


Macro processor

New object types: real (supersedes integers), boolean (distinct from integers), tuple, userdefined function.

New operators: various mathematical functions, set operations on arrays (union, intersection, difference, cartesian power and product), type checking and conversion.

Added support for comprehensions (e.g. the set containing the squares of all even numbers between 1 and 5 can be constructed with
[ i^2 for i in 1:5 when mod(i,2) == 0]
). 
Userdefined functions can be declared using the
@#define
operator (e.g.@#define f(x) = 2*x^2+3*x+5
). 
@#elseif
clauses are now supported in conditional statements. 
@#for
loops can iterate over several variables at the same time (e.g.@#for (i,j) in X
, whereX
is an array containing tuples of size 2). 
Added the possibility to exclude some elements when iterating over
@#for
loops (e.g.@#for i in 1:5 when mod(i,2) == 0
iterates over all even numbers between 1 and 5). 
A
defined()
function allows testing whether macro variables have been defined. 
Empty arrays (with the
[]
syntax) are now possible. 
Arrays of arrays are now supported.

New macro directives
@#echomacrovars
and@#echomacrovars(save)
for displaying or saving the values of all macrovariables. 
Inline comments are now supported.

⚠ All division operations are now done with doubles (as opposed to integers). To achieve the old functionality, use the new
floor
operator. 
⚠ Colon syntax used to require braces around it to create an array (e.g.
[1:3]
would create[1,2,3]
). Now this is not necessary (1:3
creates[1,2,3]
while[1:3]
would create[[1,2,3]]
). 
⚠ Previously, printing a boolean would print
1
or0
. Now, it printstrue
orfalse
. To achieve the old functionality, you must cast it to a real, e.g.@{(real)(1!=0)}
.


LaTeX output

New command
write_latex_steady_state_model
. 
New option
planner_discount_latex_name
oframsey_model
anddiscretionary_policy
. 
New command
model_local_variable
command for assigning a LaTeX name to modellocal variables. 
The
write_latex_static_model
andwrite_latex_original_model
commands now support thewrite_equation_tags
option.


Compilation of the model (
use_dll
option) made easier and faster
Under Windows, it is no longer necessary to manually install the compiler, since the latter is now shipped by the Dynare installer.

Under macOS, the Dynare installer now automatically downloads and installs the compiler.

It is no longer necessary to configure MATLAB to let it know where the compiler is, since the compilation is now done by the preprocessor.

The compilation phase is now faster on large models (this has been achieved by disabling a few timeconsuming and notsouseful optimization passes otherwise done by the compiler).

New
compilation_setup
block for specifying a custom compiler or custom compilation flags.


Model, variables and parameters declaration

New syntax to declare model variables and parameters onthefly in the
model
block. To do this, simply follow the symbol name with a vertical line (
, pipe character) and either ane
, anx
, or ap
. For example, to declare a parameter namedalpha
in the model block, you could writealphap
directly in an equation where it appears. Similarly, to declare an endogenous variablec
in the model block you could writece
. 
New syntax to declare model variable and parameters onthefly in equation tags. In the tag, simply state the type of variable to be declared (
endogenous
,exogenous
, orparameter
followed by an equal sign and the variable name in single quotes. Hence, to declare a variablec
as endogenous in an equation tag, you can type[endogenous='c']
. 
New
epilogue
block for computing output variables of interest that may not be necessarily defined in the model (e.g. various kinds of real/nominal shares or relative prices, or annualized variables out of a quarterly model).


Commandline options

Added the possibility to declare Dynare commandline options in the
.mod
file. 
New option
nopreprocessoroutput
to disable printing of messages from the preprocessor. 
It is now possible to assign an arbitrary macroexpression to a macrovariable defined on the commandline, using the
D
syntax. 
New option
linemacro
to revert to the old format of the macroprocessed file (see below).


Preprocessor outputs and inputs

Added JSON output to the preprocessor. A representation of the model file and the whole content of the
.mod
file is saved in.json
files. These JSON files can be easily parsed from any language (C++, Fortran, Python, Julia, MATLAB, Octave…). This new feature opens the possibility to develop alternative backends for the Dynare language. 
⚠ Most files generated by the preprocessor are now grouped under two subdirectories. Assuming your file is
FILENAME.mod
, then Mfiles and MEXfiles will be under+FILENAME/
, while other output (JSON, LaTeX, source code for the MEX files) will be underFILENAME/
. 
The macrogenerated output is now more readable (no more line numbers and empty lines). The old behaviour can be restored using the
linemacro
option (see above). 
Ability to call the preprocessor by passing the
.mod
file as a string argument from the macOS or GNU/Linux command line.


dseries classes

New functionalities and efficiency improvements.

Complete rewrite using the new
classdef
syntax and exploiting in place modifications when possible. 
Integration of the
dates
classes withindseries
.


Reporting classes

Automatically create titlepage with page numbers/page titles.

Allow for the removal of headers and footers from a given page.

Allow user to set page number.

Split up report output. Create new files for the preamble, the body of the report, and each individual page of the report.

The classes have been converted to the new
classdef
syntax.


Misc

External functions can be located in MATLAB/Octave namespaces.

Improvements to the balanced growth path test that is performed after Dynare has detrended the model (given the trends on variables declared by the user): the default tolerance has been raised, and a different value can be set with new option
balanced_growth_test_tol
to themodel
block; as a consequence, failing the test is now an error again. 
New collection of MATLAB/Octave utilities to retrieve and alter objects:
get_irf
,get_mean
,get_shock_stderr_by_name
,get_smooth
,get_update
,set_shock_stderr_value
. 
⚠ Previously, when some MEX files were missing, Dynare would automatically fall back to slower Mfile functional alternative; this is no longer the case. It is however still possible to manually add these alternatives in the MATLAB/Octave path (they are located under
matlab/missing/mex
; this only applies to themjdgges
,gensylv
,A_times_B_kronecker_C
,sparse_hessian_times_B_kronecker_C
andlocal_state_space_iteration_2
DLLs).

Since there are a few backwardincompatible changes in this release, users may want to have a look at the upgrade guide to adapt their existing codes.
Bugs that were present in 4.5.7 and that are fixed in 4.6.0
 Estimation: the check for stochastic singularity erroneously would only take estimated measurement error into account.
 Estimation: if the Hessian at the mode was not positive definite, the Laplace approximation returned a complex number, but only displayed the realvalued part.
 Conditional Forecasting: using one period only would result in a crash.
 Firstorder approximation was not working with purely forwardlooking models.
 The preprocessor would not allow for inline comments including macro statements.
 Using the
STEADY_STATE()
operator on exogenous variables would lead to crashes in stochastic simulations. moment_calibration
: for autocorrelation functions, the xaxis labeling had the wrong order.plot_identification
: placement of white dots indicating infinite values was incorrect Automatic detrending would sometime refuse to detrend model despite the user having given correct trends.
 Using
use_dll
+fast
options would not always recompile the model when the equations were changed.  Under certain circumstances, the combination of
bytecode
andstack_solve_algo=1
options could lead to crashes or wrong results.
References

Komunjer, I. and S. Ng (2011), “Dynamic Identification of Dynamic Stochastic General Equilibrium Models,” Econometrica, 79(6), 1995–2032

Qu, Z. and D. Tkachenko (2012), “Identification and frequency domain quasi‐maximum likelihood estimation of linearized dynamic stochastic general equilibrium models,” Quantitative Economics, 3(1), 95–132

Mutschler, W. (2015), “Identification of DSGE models—The effect of higherorder approximation and pruning,” Journal of Economic Dynamics and Control, 56, 34–54