At Risk Transformation

A forecasting package that implements the methodology from the paper "At Risk Transformation for U.S. Recession Prediction".

Working paper link: https://www.philadelphiafed.org/-/media/frbp/assets/working-papers/2025/wp25-34.pdf

Project Structure

at_risk/                # Core Package
  ├── forecasting.py    # Recursive OOS Loop
  ├── models.py         # ModelRegistry & Templates
  ├── evaluation.py     # Metrics
  ├── data.py           # Data Prep
  ├── config.py         # Configuration
  └── cli.py            # Command Line Interface
data/                   # Data Directory
  ├── raw/              # Original CSVs
  └── processed/        # Pickle files
results/                # Outputs
  ├── predictions/      # Forecast .pkl files
  └── figures/          # Exported Plots
notebooks/              # Visualization
  └── Analysis.ipynb    # Figures and tables for Publication

Quick Start

1. Installation

Install the package in editable mode:

pip install -e .

2. Run an Experiment

Use the CLI to generate OOS forecasts:

python -m at_risk --horizons 3 6 12 --lags 3 6 12 --specific-sets Deter_States

Included flags:
--horizons: Set forecasting horizon.
--lags: Add specific lags.
--rerun-all: Regenerate all results from scratch and overwrite existing results files.
--specific-sets: Rerun the forecasting loop for a specific predictor set.
--use-subset: Run the predictor set specifications with the parsimonious set identified in the paper.

3. Generate Figures

Open notebooks/Analysis.ipynb to view performance tables and generate plots.