EconoMetron

Appearance

Econometron ​

Econometron Logo

Econometron

Overview ​

Econometron is a compact Python toolkit for econometric modeling, multivariate time‑series analysis, and dynamic stochastic general equilibrium (DSGE) research. It combines classical econometric methods (state‑space models, VAR/VARMA, MLE/Bayesian inference) with Deep Learning Models (N‑BEATS) and projection‑based solvers to support estimation, identification, simulation, and forecasting workflows in applied research and policy analysis.

Core Objectives ​

  • Comprehensive Model Coverage: VAR, VARMA, VARIMA, SVAR, N-BEATS models, state-space, and Linear/non-linear DSGE.
  • Rigorous estimation & inference — likelihood‑based estimation (MLE), Bayesian options, and uncertainty quantification for parameters and forecasts.
  • Structural identification — VARMA supports echelon‑form identification using Kronecker indices to obtain parsimonious, identifiable parameterizations.
  • Impulse response & causal analysis — IRFs via local projection.
  • State dynamics & forecasting — Kalman filter / smoother and state‑space forecasting for latent dynamics and prediction.

Installation ​

Install via pip:

bash
pip install econometron

Architecture ​

bash
Econometron
│
├── filters/          # Kalman filtering, smoothing, preprocessing
├── Models/           # VAR, VARMA, VARIMA, SVAR, State-Space, N-BEATS, DSGE
│   ├── StateSpace.py # Linear and non-linear dynamic systems
│   ├── VectorAutoReg # VAR, VARMA, VARIMA, SVAR
│   ├── Neuralnets    # N-BEATS and RevIN-enhanced models
│   └── dynamicsge    # Linear and non-linear DSGE models
├── utils/            # Data preparation, estimation, solvers, projection, optimizers
└── Tutorials/         # Sample workflows, notebooks, tutorials
  • Filters Layer: Kalman filtering, smoothed state estimation.
  • Models Layer: Econometric, neural, state-space, and DSGE modeling capabilities.
  • Utils Layer: Data preparation, Estimation , solvers, projection, optimizers .
    • Estimation Layer: MLE, Bayesian inference.
    • Projection & Solver Layer: DSGE projections, collocation methods, Local Projection.
    • Optimizers: Genetic Algorithms, Simulated Annealing.

Models ​

State-Space Models ​

Formalizes linear dynamic systems, supporting latent state estimation, filtering, and simulation of underlying processes. Integrates seamlessly with Linear DSGE econometron.Models.dynamicsge.linear_dsge for detailed study of system dynamics.

Vector Autoregressive Models ​

  • VAR: Multivariate dependencies with autoregressive structure.
  • VARMA: Extends VAR with moving average terms; supports echelon form identification.
  • VARIMA: Integrated VARIMA for non-stationary series.
  • SVAR: Encodes structural economic theory constraints.

Neural Network Forecasting ​

  • N-BEATS: Interpretable deep learning for univariate and multivariate series.
  • N-BEATS + RevIN: Enhances generalization with reversible instance normalization.

Non Linear DSGE Solvers ​

  • Projection Methods: Galerkin, collocation, and least squares approximations for non-linear DSGE policy functions.

Estimation and Inference ​

  • MLE: Likelihood-based parameter estimation.
  • Bayesian Estimation: Posterior distributions with prior integration.
  • Regression: Ordinary Least Square.
  • Impulse Response Functions: Local projection for robust dynamic response analysis.
  • Optimization Algorithms: Genetic Algorithms, Simulated Annealing for complex likelihood surfaces.

Econometron provides a rigorous, flexible, and efficient platform for classical econometric analysis, modern forecasting, and complex dynamic system investigation.

Resources ​