pyperfectforesight

A minimal Dynare-style perfect foresight solver in Python. pyperfectforesight lets you define dynamic economic models with Dynare’s familiar lag notation, then solve for the perfect foresight transition path using a sparse Newton BVP solver. It supports stock/jump variable models, homotopy continuation for large shocks, and Dynare’s expectation-errors (multiple surprise MIT shocks) protocol — all from pure Python with NumPy, SciPy, and SymPy.

Contents

• Installation
  • Requirements
  • From source (development)
  • With pip (when published)
  • Building the documentation
• Getting Started
  • Key concepts
  • Minimal RBC example
  • RBC model with exogenous TFP shock
• Solvers
  • solve_perfect_foresight
  • solve_perfect_foresight_homotopy
  • solve_perfect_foresight_expectation_errors
  • Terminal steady state
• Initial Guess
  • make_initial_guess
  • Methods
  • Usage examples
  • The decay parameter
• Auxiliary Variables
  • Overview
  • Declaring auxiliary variables
  • Methods
  • Complete example
  • Method comparison
  • Recommendations
  • Design philosophy
• API Reference
  • Utilities
  • Model processing
  • Steady state
  • High-level solvers
  • Initial guess
  • Auxiliary variables
  • Low-level solver components

Features

• Dynare lag notation — write v("k", -1) for \(k_{t-1}\) and v("c", 1) for \(c_{t+1}\), matching Dynare’s convention exactly

• Augmented-path BVP solver — pre-period-0 initial_state (k_{-1}) is pinned; all period-0 variables including jump variables are solved simultaneously

• Automatic stock-variable inference — lead-lag incidence detects predetermined variables; no manual classification needed

• Homotopy continuation — solve_perfect_foresight_homotopy for large shocks that defeat direct Newton

• Expectation-errors solver — replicates Dynare’s perfect_foresight_with_expectation_errors_solver for sequences of surprise shocks

• Auxiliary variable support — four methods (auto, analytical, dynamic, nested) to handle static/auxiliary variables

• Symbolic processing + automatic differentiation — models defined via SymPy; sparse Jacobians computed automatically