CGE-Core

A Pyomo-based Computable General Equilibrium framework faithful to the textbook by Hosoe, Gasawa & Hashimoto (2010). Named to align with the Policy Simulation Library convention (cf. OG-Core).

Note. This is an independent project. It is not affiliated with or endorsed by the Policy Simulation Library; it merely follows the *-Core naming pattern.

---

Provenance and license

CGE-Core is a corrected fork of PyCGE by Juan Fung and Charley Burtwistle (U.S. National Institute of Standards and Technology). The original PyCGE is a work of the U.S. federal government and is in the public domain under 17 U.S.C. 105; the original NIST notice is preserved in LICENSE_NIST.txt.

Modifications in this fork — the Walras'-law degree-of-freedom fix, bug fixes, the engine API, and the test suite — are released under the MIT License (LICENSE.txt).

Citing

If you use CGE-Core, please cite both the original PyCGE and the Hosoe textbook:

@software{fung2017pycge,
  author      = {Juan Fung and Charley Burtwistle},
  title       = {{PyCGE}: A Python Interface for Solving {CGE} Models},
  year        = {2017},
  url         = {https://github.com/juanfung/pycge},
  institution = {National Institute of Standards and Technology}
}

@book{hosoe2010textbook,
  author    = {Hosoe, Nobuhiro and Gasawa, Kenji and Hashimoto, Hideo},
  title     = {Textbook of Computable General Equilibrium Modelling:
               Programming and Simulations},
  year      = {2010},
  publisher = {Palgrave Macmillan},
  doi       = {10.1057/9780230281653}
}

---

What this is

CGE-Core separates model definition (the algebraic structure) from model workflow (calibration, simulation, comparison). The model equations are a verified 1:1 port of the GAMS Model Library files splcge.gms (SEQ=275) and stdcge.gms (SEQ=276). All 24 constraints of the standard model have been checked equation-by-equation against the GAMS source.

Model	Hosoe ch.	Description
splcge	3–4	Simple closed economy: 2 goods, 2 factors
stdcge	5–6	Open economy: Armington, CET, government, investment

---

Why a CGE needs one equation dropped (important)

A CGE is a square system: after fixing one price as numeraire, the number of independent equilibrium conditions equals the number of free variables. But Walras' law makes one market-clearing equation redundant — once every other market clears, the last clears automatically. If all market-clearing equations are kept, the assembled system is over-determined by exactly one equation, and a gradient-based NLP solver such as IPOPT aborts with "too few degrees of freedom" (return code -10).

CGE-Core handles this explicitly with model_drop_redundant, which deactivates one market-clearing equation so the system is square (DOF = 0). The dropped market then clears automatically at the solution — a built-in consistency check on Walras' law.

The original PyCGE avoided this only by using the NEOS-hosted CONOPT/MINOS solvers, which absorb the redundancy internally. A local IPOPT workflow does not, so the step is required here.

---

Installation

CGE-Core needs Pyomo and one local NLP solver. Two options:

Option A — IPOPT executable (simplest if you use conda):

conda install -c conda-forge ipopt
git clone https://github.com/jamesmiraflor/CGE-Core.git
cd CGE-Core
pip install -e .

Then use solver name 'ipopt'.

Option B — cyipopt (pip-only, no conda):

# system IPOPT library + headers (Debian/Ubuntu)
sudo apt-get install -y coinor-libipopt-dev
git clone https://github.com/jamesmiraflor/CGE-Core.git
cd CGE-Core
pip install -e ".[solver,test]"
# build PyNumero's ASL bridge (needs cmake + a C++ compiler)
python -m pyomo.contrib.pynumero.build

Then use solver name 'cyipopt'.

---

Quick start

from pyomo.environ import value
from cge_core.examples.stdcge_model_def import StdModelDef
from cge_core.engine import PyCGE

cge = PyCGE(StdModelDef())
cge.model_data('cge_core/data/stdcge_data_dir')

cge.model_instance('pf', 'LAB')          # fix numeraire (Hosoe: pf_LAB = 1)
cge.model_drop_redundant('eqpf', 'LAB')  # Walras' law -> square system
cge.model_calibrate('cyipopt')           # solve base (reproduces the SAM)

cge.model_sim()                          # clone calibrated base -> sim
cge.model_modify_sim('taum', 'BRD', 0)   # abolish bread tariff
cge.model_modify_sim('taum', 'MLK', 0)   # abolish milk tariff
cge.model_solve('cyipopt')               # solve counterfactual

cge.model_postprocess('compare', 'print')  # base vs sim, % changes

Run the bundled experiments directly:

python -m cge_core.examples.stdcge   # tariff & production-tax abolition
python -m cge_core.examples.splcge   # closed-economy base calibration

---

Workflow

ModelDef ──▶ PyCGE ──▶ model_data()
                          │
                    model_instance()        ← fix numeraire
                          │
                  model_drop_redundant()    ← Walras' law (DOF 0)
                          │
                    model_calibrate()       ← solve base
                          │
                      model_sim()           ← clone base ▶ sim
                          │
                   model_modify_sim()       ← apply shocks
                          │
                     model_solve()          ← solve counterfactual
                          │
                  model_postprocess()       ← compare / export

---

Tests

python -m pytest tests/ -v

Seven tests check structure (build, DOF before/after the drop), correctness (base reproduces the SAM, solver recovers from a perturbed start, the dropped market clears), and economics (tariff abolition raises welfare). Solver- dependent tests auto-skip if no local NLP solver is present.

---

References

• Hosoe, N., Gasawa, K. & Hashimoto, H. (2010). Textbook of Computable General Equilibrium Modelling. Palgrave Macmillan.
• GAMS Model Library: splcge.gms, stdcge.gms
• Original PyCGE: github.com/juanfung/pycge
• Naming convention: github.com/PSLmodels/OG-Core