Release Notes

Version 0.6

First-order perturbations

RECS now integrates a function that solves models using first-order perturbations (recsSolveLocal) based on complex generalized Schur decomposition. This solution is automatically calculated as first guess for problems without complementarity constraints.

Solve problems by fixed-point iteration (option explicit)

Undocumented function that allows to solve model by fixed-point iteration rather than time iteration.

Faster solution methods

• When response variables are approximated, only forward variables are interpolated to reduce the problem size, which makes solving for rational expectations faster.
• The construction of the sparse block tri-diagonal Jacobian for perfect-foresight problems is faster by pre-calculating the index of the sparse matrix.

Documentation

• Documentation on how to use MATLAB Parallel toolbox with RECS
• Documentation of the default solution method

Backwards compatibility breaks

• Model structures are no longer constructed using recsmodelinit but recsmodel, which builds an object rather than a structure.
• The option value resapprox-simple for the field funapprox is no longer supported. The default value for this field becomes resapprox and replace resapprox-complete.
• The option values iter and iter-newton for the field reemethod have been merged together. The choice is now automatic, based on the solver set by reesolver.
• Change the order of arguments in recsDecisionrRules.

Version 0.5.1

This version has been adjusted to restore compatibility with MATLAB R2013a. Previous RECS versions are not compatible with MATLAB R2013a.

Parallel computing

RECS is ready for parallel computing using MATLAB Parallel Computing Toolbox. The option field loop_over_s can take values other than 0 or 1. Any integer value above 1 implies that the equilibrium equations are solved by iterating over loop_over_s blocks, which allows to break down the problem in as many blocks as there are available processors.

New behavior of recsSimul

• It now simulates the model for nper periods and not nper+1
• Inversion of output fsim and stat to avoid excessive calculations of fsim.
• To calculate the asymptotic statistics, the first 20 observations are now discarded, not the first 19.

Documentation

• New demonstration model: stochastic growth model with recursive preferences and stochastic volatility (GRO3).
• Better display of equations using MathML.

Backwards compatibility breaks

• No longer compatible with MATLAB R2009b because this version is not able to parse properly files prepared for parallel computing.
• Because of the new behavior of recsSimul, programs have to be adjusted with respect to the position of stat in the output.

Previous versions

See RECS commit history for differences between the various releases.