GDSGE: A Toolbox for Solving Global DSGE Models

🆕 v0.2.0 — GDSGE has been refactored

The toolbox has been rebuilt from the ground up and is now the main line. Highlights:

• Clearer diagnostics — actionable messages on solver errors and unsolved equations.
• SymPy symbolic-differentiation backend — analytic Jacobians, used automatically when a Python environment is present (the autodiff backend remains the default otherwise). One-time setup on any OS: run gdsge_setup_sympy in MATLAB (installs uv and the Python env for you).
• Better performance.
• Backward compatible (enforced) — every existing .gmod runs unchanged; the public API and IterRslt / SimuRslt result shapes are frozen and guarded by the test suite.

Overview: refactor_summary.md · full status: PROGRESS.md.

Looking for the original toolbox? It is archived on the legacy-0.1.x branch; the last release of the original line is v0.1.6.

The GDSGE Toolbox

GDSGE is a toolbox that solves nonlinear Dynamic Stochastic General Equilibrium (DSGE) models with a global method based on the Simultaneous Transition and Policy Function Iteration (STPFI) algorithm introduced in Cao, Luo and Nie (2023). It allows users to define economic models in compact and intuitive scripts, called gmod files (gmod stands for global model). It parses the scripts into dynamic libraries which implement the actual computations (policy function iterations and Monte Carlo simulations) efficiently in C++, and provides a convenient MATLAB interface to researchers.

The toolbox can be used to solve models in macroeconomics, international finance, asset pricing, and related fields.

See the toolbox website for examples and documentation.

Run on MATLAB Online

First, log into your MATLAB Online

Then, download the refactored toolbox and unzip in the MATLAB command window:

websave('gdsge.zip','https://github.com/gdsge/gdsge/archive/refs/heads/main.zip')
unzip gdsge.zip

Finally, change directory to the unzipped folder, set up the default mex compiler, and run tests in the MATLAB command window:

mex -setup c++
cd gdsge-main/tests
runtests

See docs/user-guide.md for authoring models and reading results.

Requirements for the local compiler

• Windows

  • MATLAB 2017b+.
  • MATLAB toolboxes: Symbolic Math
  • Compiler: MinGW64 C++ (installed via MATLAB Add-Ons); Visual Studio C++ Compiler 2019, 2022 (community version is fine); Intel C++ Compiler 2017 or newer

• macOS (silicon processor):

  • MATLAB silicon version (R2023b+).
  • MATLAB toolboxes: Symbolic Math
  • Compiler: g++ 15; see the instruction for how to setup the g++ compiler for Apple silicon processors

• macOS (Intel processor):

  • MATLAB 2017b+.
  • MATLAB toolboxes: Symbolic Math
  • Compiler: g++ 8.5; see the instruction for how to setup the g++8.5 compiler

Installation of the local compiler

First, Configure your mex C++ compiler by running in MATLAB

mex -setup c++

Then, acquire the source code by cloning the git repository (the local folder name should not contain spaces):

git clone https://github.com/gdsge/gdsge

Next, in MATLAB, change directory to gdsge/tests, run

runtests

which runs all the tests and produce all results in the companion paper Cao, Luo, and Nie (2023).

To compile a gmod file, add folder "src" to MATLAB search path and run gdsge_codegen after changing the working directory to the one that contains the gmod file. For example, suppose you have located tests/HeatonLucas1996 with HL1996.gmod in the working directory, then simply run

gdsge_codegen('HL1996')

which will generate all the source codes and call the C++ compiler to compile the mex files.

License

GDSGE is released under the Apache License, Version 2.0, which is available at http://www.apache.org/licenses/LICENSE-2.0. In short, this license allows you to use, compose and distribute the GDSGE compiler or generated codes freely. However, it is requested that the companion paper be cited:

Dan Cao, Wenlan Luo, and Guangyu Nie (2023). Global DSGE models. Review of Economic Dynamics, Volume 51, December 2023. Available at: https://www.sciencedirect.com/science/article/pii/S1094202523000017

GDSGE relies on the following external libraries, with their licenses described below and attached under folder licenses/:

• Adept: A combined automatic differentiation and array library for C++.

  Licensed under the Apache License, Version 2.0. Citation to the academic paper:

  • Hogan, R. J., 2014: Fast reverse-mode automatic differentiation using expression templates in C++. ACM Trans. Math. Softw., 40, 26:1-26:16.

• CoDoSol: a bound-constrained nonlinear equations solver.

  Citation to the academic paper:

  • Bellavia, S., M. Macconi, and S. Pieraccini (2012). Constrained dogleg methods for nonlinear systems with simple bounds. Computational Optimization and Applications 53(3), 771–794.

• myppual: Construct and Evaluate splines in ppform at flexible vector-valued dimensions, table look-up index, and spline dimension reduction in both vectorized pure MATLAB code and CMEX implementation.

  Copyright (c) 2014 Jinhui Bai (jinhui.bai@gmail.com) and Wenlan Luo (luowenlan@gmail.com)

• v2struct: Pack/Unpack Variables to/from a scalar structure.

  Copyright (c) 2014, Adi Navve, released under the MATLAB File Exchange License (BSD License)

• flat_hash_map: Copyright Malte Skarupke 2017.

  Distributed under the Boost Software License, Version 1.0 (http://www.boost.org/LICENSE_1_0.txt)

• Eigen: a C++ template library for linear algebra: matrices, vectors, numerical solvers, and related algorithms. Starting from the 3.1.1 version, it is licensed under the MPL2