Inflation Forecasting: Econometrics + Machine Learning

Production-style inflation-forecasting pipeline that combines classical ARIMA diagnostics in Stata with modern ML/DL models in Python. It covers data preparation, exploratory analysis, model selection, residual checks, dynamic multi-step forecasts, and a clean benchmarking suite with MSE, MAE, RMSE and R^2.

Table of Contents

• Project Overview
• Highlights
• Repository Structure
• Models Included
• Quickstart
• CLI Usage
• Docker
• Results and Reporting
• Dependencies
• Contributing
• License

Project Overview

This repository formalizes a full inflation-forecasting workflow:

1. Stata econometrics for diagnostics and classical model selection (ADF, ACF/PACF, ARIMA candidates).
2. Python modeling for LSTM/GRU tuning, baselines, and reproducible evaluation.
3. CLI and artifacts for repeatable training, inference, and metrics export.

The goal is to provide a professional, reproducible project structure that is easy to extend and present.

Highlights

• Dual-stack approach: Econometric rigor + ML/DL performance.
• Multi-model benchmarks: ARIMA/ARMA/SARIMA/ARIMAX, ARCH/GARCH, LSTM/GRU, Prophet, and ML baselines.
• Dynamic forecasts: Multi-step out-of-sample predictions for policy or investment use cases.
• Reproducible outputs: All metrics and predictions are saved to outputs/.
• Clean modular code: Reusable Python package with unit tests.

Repository Structure

Models Included

Econometrics

• ARIMA, ARMA, SARIMA, ARIMAX
• ARCH, GARCH
• HP filter and seasonal decomposition

Machine Learning / Deep Learning

• LSTM, GRU (with hyperparameter tuning)
• Random Forest, Gradient Boosting, Linear Regression
• XGBoost
• Prophet

Quickstart

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pip install -e .

CLI Usage

# Descriptive stats and decomposition
inflation-forecast describe --state Maryland
inflation-forecast hp-filter --state Maryland --lamb 1600
inflation-forecast decompose --state Maryland --period 4

# ARIMA (fixed order or grid search)
inflation-forecast arima --order 1,0,3
inflation-forecast arima --grid --p-min 0 --p-max 3 --d-min 0 --d-max 1 --q-min 0 --q-max 3

# Econometric extensions
inflation-forecast arma --order 1,1
inflation-forecast sarima --order 1,0,1 --auto-seasonal
inflation-forecast arimax --order 1,0,1 --exog-cols pi_nt,pi_t
inflation-forecast arch --p 1
inflation-forecast garch --p 1 --q 1

# LSTM / GRU
inflation-forecast lstm-train --look-back 4 --epochs 80 --forecast-steps 4 --save-model
inflation-forecast lstm-tune --look-back 4 --max-trials 10
inflation-forecast gru-train --look-back 4 --epochs 80

# LSTM inference from saved artifacts
inflation-forecast lstm-forecast --model-path outputs/lstm_model_*.keras --scaler-path outputs/lstm_scaler_*.joblib --steps 4

# Prophet and ML baselines
inflation-forecast prophet
inflation-forecast ml-train --model random_forest --lags 4
inflation-forecast ml-train --model xgboost --lags 4

Docker

docker build -t inflation-forecast .
docker run --rm -v "$PWD:/app" inflation-forecast describe --state Maryland

Results and Reporting

All runs export metrics and predictions into outputs/ as CSV/JSON. This makes it easy to build reports, dashboards or client-ready visualizations.

Dependencies

Optional dependencies by model

• ARIMA/HP filter/decomposition: statsmodels
• ARCH/GARCH: arch
• LSTM/GRU + tuning: tensorflow, keras-tuner
• Prophet: prophet
• XGBoost: xgboost

Contributing

Contributions are welcome. Please open an issue or submit a pull request at:
https://github.com/pablo-reyes8

License

Released under the MIT License - free for personal or commercial use.