SymbolicDSGE 1.4.1

A diagnostic suite for linear DSGE models with a custom backend supporting symbolic manipulation.

by Güney Kıymaç

---

   

 

Documentation   |   PyPI

Installation

pip install SymbolicDSGE
pip install "SymbolicDSGE[fred]"  # FRED API utilities
pip install "SymbolicDSGE[sr]"    # Symbolic Regression Deps
pip install "SymbolicDSGE[ui]"    # Web based GUI

Useful Links

• Installation
• Quick Start Guide

Overview

SymbolicDSGE is a Python DSGE engine with a JIT compiled backend for linear and linearized DSGE models, supporting symbolic manipulation features for in-place model modification. It also provides measurement-equation augmentation tools, including symbolic regression for complete or restricted free-form function discovery and OLS, Ridge, Lasso, and Elastic Net for structured linear coefficient estimation. The library supports a wide set of features beyond augmentation:

• DSGE model specification, symbolic manipulation, and linearization
• Bayesian and maximum-likelihood estimation
• Simulation and impulse-response-function utilities
• Gaussian and extended Kalman filtering
• Automatic data retrieval from FRED
• Custom shock distributions via SciPy or user-defined samplers
• Monte Carlo pipelines for comparing models or model-generated samples, including:
  • Statistical tests
  • Regression setups
  • Output transformations
  • Filtering
  • Data generation

Web-UI

SymbolicDSGE ships a [ui] dependency with the goal of making DSGE basic experiments accessible to non-programmers. While model estimation is currently not implemented, given a configuration file GUI users can:

• Adjust parameters
• Solve
• Simulate
• Create and run Monte Carlo experiments
• Apply data transformations to outputs
• Produce figures
• Access raw output data

There are two intended ways to access the GUI:

# Option 1: Serve the GUI from the command line
#           This launhes an empty GUI and the configuration should be done through the Builder tab.
sdsge-ui

# Option 2: Serve the GUI with a pre-loaded SolvedModel object

# Read a config and solve the model as usual
from SymbolicDSGE import ModelParser, DSGESolver

model, kalman = ModelParser(...).get_all()
solver = DSGESolver(model, kalman)
compiled = solver.compile(...)
sol = solver.solve(compiled, ...)

# Serve the GUI with the solved model pre-loaded in the reference slot
sol.serve(open_browser=True)

The UI does not support all capabilities of the library; however it can significantly aid the demonstration of findings to a non-technical audience by allowing them to interact with a given model. The UI is in active development and supporting all built-in features is a long-term goal.

Read the Docs

Alongside API references and implementation conventions, the documentation includes guides covering model setup, estimation, simulation, and filtering. The documentation is kept up to date and aims to clarify conventions, workflows, and implementation choices throughout SymbolicDSGE. Suggestions for improving or extending the documentation are welcome as issues.

AI Parsability

The documentation is a static Material MkDocs site, so web-enabled AI tools should be able to search, parse, and summarize it reliably.

Minimal Example

from SymbolicDSGE import ModelParser
from SymbolicDSGE import DSGESolver
from numpy import float64, array

# Read the YAML config (Equations, Measurements, Parameters, Optional Filter Spec)
parsed = ModelParser("<path-to-config>.yaml").get_all()
model, kalman = parsed

# Compile the model
solver = DSGESolver(model, kalman)
compiled = solver.compile(
    variable_order=None,
    params_order=None,
    n_state=3,
    n_exog=2,
    linearize=False,
)
print("Equations with symbols removed: \n", "\n".join(map(str, compiled.objective_eqs)), "\n")
print("Equations as passed to the solver: \n", compiled.equations)

>>> Equations with symbols removed:
 -beta*fwd_Pi + cur_Pi - cur_x*kappa - cur_z
-cur_g + cur_x - fwd_x + tau_inv*(cur_r - fwd_Pi)
-cur_r*rho_r - e_R + fwd_r + (rho_r - 1)*(fwd_Pi*psi_pi + fwd_x*psi_x)
-cur_g*rho_g - e_g + fwd_g
-cur_z*rho_z - e_z + fwd_z


Equations as passed to the solver:
 <function DSGESolver.compile.<locals>.equations at 0x0000012D16AB5B20>

# Solve the compiled model
sol = solver.solve(
    compiled,
    parameters=None,
    steady_state=array([0.0, 0.0, 0.0, 0.0, 0.0], dtype=float64),
)
print("Is stable: ", sol.policy.stab == 0)
print("Eigenvalues: ", sol.policy.eig)

>>> Is stable:  True
Eigenvalues:  [0.27920118+0.j 0.83000003+0.j 0.84999992+0.j 2.56517116+0.j 1.18470582+0.j]

# Plot IRFs (single or multi shock)
sol.transition_plot(
    T=25,
    shocks=["g", "z"],
    scale=1.0,
    observables=True,
)