DynareOBC v3.21.52.1942
This is a minor release with some slight improvements to the previous one, along with fixes for a few small issues from previous releases. Changes in this version include:
- a new example, Adam & Billi (2006) to demonstrate the new cubature options, as well as how to recover policy rules (based on code provided by Shifu Jiang);
- fixing
InitialStateFile
which should never have required the user to provide values for variables other than the original endogenous variables; - fixing parallelization of simulations which was being inadvertently disabled in some situations;
- a new option to disable restoring the original path,
NoRestorePath
; - not clearing the workers after a DynareOBC run with
NoPoolClose
, - a new option to control clustering effort
CubatureClusteringEffort
, with a faster default option.
The release notes from the last release are below.
This release contains major changes to the cubature algorithm.
DynareOBC now always starts by drawing large numbers of points via QMC. Of these, many will not hit the bound, and can be discarded. Of the remaining, DynareOBC optionally performs an initial clustering step to divide them into regions, controlled by the CubatureRegions
option. Then for each region, it either integrates with a rule exact for linear functions (if CubatureCATCHDegree=0
) or with a rule exact for all monomials in q
and min{0,q}
with degree less or equal to the given CubatureCATCHDegree
.
Running DynareOBC with the Cubature
option without changing the defaults for CubatureRegions
or CubatureCATCHDegree
produces a one-point integration rule which should be almost as fast as using no-cubature at all.
Please see the read-me for the full set of new cubature options.
This release also contains a few other minor changes. (New option NoPoolClose
. Using 2 threads for Gurobi and CLP even when MultiThreadBoundsProblem
or MultiThreadCubatureRuleCreation
are not on. Improvements to progress messages.)