Macro Economic and Market Regime Allocator

A machine learning framework that forecasts economic and market regimes — such as Expansion and Recession — and applies dynamic asset allocation strategies based on regime signals.

📈 Project Overview

This project integrates macroeconomic indicators and market-based data to:

• Predict long-term economic regimes (e.g., Expansion, Recession)
• Detect short-term market stress (e.g., Normal)
• Map regime forecasts into portfolio weights
• Backtest performance using dynamic regime-based allocation

---

🧠 Regime Prediction Models

The framework includes two main models:

🟢 Economic Regime Model (econ)

• Objective: Predict long-term macroeconomic conditions — Expansion or Recession
• Data: Based on macroeconomic indicators from the FRED-MD dataset
• Algorithm: XGBoost with rolling-window training
• Output: 1M_econ_preds.csv — Monthly recession probabilities from 3 models (DT, RF, XGB)

🔴 Market Regime Model (market)

• Objective: Detect short-term market stress — Normal or Crash
• Data: Market-based indicators and lagged macroeconomic features
• Algorithm: XGBoost with rolling-window training
• Output: 1M_mkt_preds.csv — Monthly crash probabilities from 3 models (DT, RF, XGB)

---

📈 Equity Curve Comparison

📉 Drawdown Curve Comparison

🔍 Feature Importance - Economic Regime

🔍 Feature Importance - Market Regime

---

⚙️ Pipeline Structure

src/
├── allocator.py                 # Map regimes to portfolio weights
├── modeling-evaluation-econ.py # Economic regime training, evaluation, feature importance
├── modeling-evaluation-market.py # Market regime training, evaluation, feature importance
├── model_evaluation.py         # Metrics, visualizations
├── data_understanding.py       # Plot regimes timeline
data/
├── raw_data/                   # Original FRED/Moody's data
├── datasets/                   # Prepared datasets with labels
├── predictions/                # Model outputs (.csv)
├── report/                     # Evaluation plots, metrics (.png, .csv)

---

🛠️ How to Run

1. Install dependencies:

pip install -r requirements.txt

2. Run economic model training and evaluation:

python src/modeling-evaluation-econ.py

3. Run market model training and evaluation:

python src/modeling-evaluation-market.py

Outputs:

• CSV: 1M_econ_preds.csv, 1M_mkt_preds.csv
• Reports: report/DT_recession_probs.png, report/XGB_confusion_matrix.png, etc.

---

💼 Portfolio Allocation

The allocator.py script maps predicted regimes to asset weights:

Regime	SPY	TLT	GLD	BIL
Expansion	70%	20%	10%	0%
Recession	0%	50%	30%	20%

You can integrate these weights with your backtesting engine.

---

📊 Evaluation Metrics

• Accuracy, Precision, Recall, F1, ROC-AUC
• Confusion matrix
• Time-series plots of regime prediction probabilities
• Feature importance (XGBoost)

---

📁 Sample Outputs

• ./data/predictions/1M_econ_preds.csv
• ./report/DT_confusion_matrix.png
• ./report/XGB_recession_probs.png

---

📌 Notes

• Economic labels are based on NBER Business Cycle Dating
• Models use rolling-window training for out-of-sample robustness
• Feature engineering relies on FRED-MD macroeconomic dataset

---

🔒 License

MIT License — © 2023–2025 Ali Rahimi, with modifications by contributors.

---

🙏 Acknowledgements

This project is inspired by and adapted from the excellent work by ARahimiQuant.
Special thanks to their comprehensive framework for forecasting economic and market regimes, which provided valuable guidance and reference for this implementation.