Recession Risk Auditor

Recession-Audit

An auditing framework for evaluating synthetic data generators (SB-GAN, ctdGAN) in highly imbalanced macroeconomic recession prediction.

Key Features

• Time-aware splits (no leakage)
• Leave-One-Country-Out (LOCO) evaluation
• Threshold tuning inside training only
• Synthetic data auditing (fidelity, purity, stability)
• Cluster bootstrap confidence intervals
• Nowcasting and forecasting-ready design

Experiments

• Experiment A: Nowcasting (t → y_t)
• Experiment B: Forecasting (t → y_{t+h})

This repository contains the full experimental pipeline. Trained models are excluded by design.

Structure

• src/: Core Python scripts and model logic.
• data/: Datasets used for the 7-country audit.