fred-forecaster 0.1.0

Time series forecasting for FRED economic data

fred_forecaster

A Python package for time series forecasting of Federal Reserve Economic Data (FRED).

Features

• Flexible FRED data retrieval: Fetch any FRED series using its ID (e.g., GFDEBTN for total public debt).
• Multiple forecasting approaches:
  • Classical SARIMAX modeling: Fit SARIMAX(1,1,1)x(0,1,0)[4] to quarterly data.
  • Bayesian structural time series: Decompose data into level, trend, and seasonal components using PyMC.
• Simulation: Generate multiple (e.g., 1,000) random simulations for future quarters.
• Calibration: Reweight simulations to match external targets (e.g., CBO forecasts).
• Probability analysis: Calculate probabilities of quarter-over-quarter declines.
• Visualization: Create plots of forecasts, probability distributions, and model components.

Installation

From PyPI

pip install fred-forecaster

From source

git clone https://github.com/maxghenis/fred-forecaster.git
cd fred-forecaster
pip install -e .  # Install core package
pip install -e ".[dev]"  # Include development dependencies
pip install -e ".[app]"  # Include Streamlit app dependencies

Usage

Basic usage

import os
from fred_forecaster import (
    fetch_fred_data,
    fit_sarimax_model,
    generate_simulations,
    plot_forecasts
)

# Set your FRED API key
os.environ["FRED_API_KEY"] = "your_api_key_here"

# Fetch FRED data
data = fetch_fred_data("GFDEBTN")  # Federal debt

# Fit SARIMAX model
model = fit_sarimax_model(data["Debt"])

# Generate simulations
simulations, forecast_index = generate_simulations(model, data, end="2028Q4", N=1000)

# Plot results
fig = plot_forecasts(data, simulations, forecast_index)
fig.savefig("forecast.png")

Bayesian forecasting

from fred_forecaster import (
    fetch_fred_data,
    fit_bayesian_model,
    generate_bayesian_simulations,
    plot_forecasts
)

# Fetch data
data = fetch_fred_data("GFDEBTN")

# Fit Bayesian model
model, idata = fit_bayesian_model(data["Debt"])

# Generate simulations
simulations, forecast_index = generate_bayesian_simulations(
    model, idata, data, end="2028Q4", N=1000
)

# Plot results
fig = plot_forecasts(data, simulations, forecast_index)

Calibration to external targets

from fred_forecaster import calibrate_simulations

# Define your own targets (or use the default CBO targets)
targets = {2024: 35.230, 2025: 37.209}

# Calibrate simulations
weights = calibrate_simulations(simulations, forecast_index, targets)

# Plot calibrated results
fig = plot_forecasts(data, simulations, forecast_index, weights)

Demo App

The package includes a Streamlit demo app that showcases its functionality:

cd demo
streamlit run app.py

This will open a web browser with an interactive interface where you can:

• Select a FRED series ID
• Choose between SARIMAX and Bayesian models
• Toggle calibration to CBO targets
• Adjust simulation parameters
• View forecasts and probability analyses

Development

Setup

pip install -e ".[dev]"

Running tests

pytest  # Run regular tests
pytest --run-slow  # Include slow tests (e.g., Bayesian model tests)
pytest --cov=fred_forecaster  # Run with coverage

License

MIT

Acknowledgments

• Federal Reserve Economic Data (FRED)
• PyMC
• statsmodels
• Streamlit