Financial Market Regime Detection

A regime detection system using Hidden Markov Models and machine learning to identify market states (bull, bear, high-volatility) and adapt investment strategies accordingly.

Problem Statement

Financial markets exhibit distinct behavioral regimes that require different trading strategies. A strategy optimized for bull markets often fails during bear markets or high-volatility periods. This system automatically identifies market regimes and enables regime-aware portfolio management.

Industry Applications

Quantitative trading firms
Asset management (regime-aware allocation)
Risk management (volatility forecasting)
Algorithmic trading systems
Portfolio optimization
Market timing strategies

Key Features

Hidden Markov Models (HMM): Unsupervised regime detection with Gaussian emissions
Gaussian Mixture Models (GMM): Alternative clustering-based approach
Change Point Detection: Structural break identification
Regime-Conditioned Trading: Adaptive strategies based on detected regime
Walk-Forward Validation: Proper backtesting without lookahead bias

Detected Regimes

| Regime | Characteristics | |--------|-----------------| | Bull/Calm | Low volatility, positive returns | | Bear/Crisis | High volatility, negative returns | | Transition | Medium volatility, mixed returns |

Technical Architecture

Market Data (Yahoo Finance)
        |
        v
+------------------+
|  Feature Eng     |
|  - Returns       |
|  - Volatility    |
|  - Technicals    |
+------------------+
        |
        v
+------------------+     +------------------+
| Regime Detection | --> | Strategy Engine  |
| - HMM            |     | - Position Sizing|
| - GMM            |     | - Risk Mgmt      |
| - Change Points  |     +------------------+
+------------------+              |
        |                         v
        v               +------------------+
+------------------+    |   Backtesting    |
| Visualization    |    | - Walk-Forward   |
| - Regime Plots   |    | - Metrics        |
| - Equity Curves  |    +------------------+
+------------------+

Project Structure

market-regime-detection/
├── data/                    # Cached market data
├── src/
│   ├── data_loader.py       # Yahoo Finance data fetching
│   ├── features.py          # Feature engineering
│   ├── hmm_regime.py        # HMM regime detection
│   ├── gmm_regime.py        # GMM clustering approach
│   ├── changepoint.py       # Change point detection
│   ├── strategy.py          # Regime-based trading strategies
│   ├── backtest.py          # Backtesting framework
│   ├── visualize.py         # Regime visualization
│   └── walk_forward.py      # Walk-forward validation
├── notebooks/
│   └── EDA.ipynb            # Exploratory data analysis
├── docs/
│   └── IMPLEMENTATION_PLAN.md
├── tests/
├── requirements.txt
└── README.md

Quick Start

# Clone repository
git clone https://github.com/Sakeeb91/market-regime-detection.git
cd market-regime-detection
Install dependencies
pip install -r requirements.txt
Download market data
python src/data_loader.py --ticker SPY --start 2000-01-01
Fit HMM and detect regimes
python src/hmm_regime.py --n-states 3
Run backtest
python src/backtest.py --strategy regime
Generate visualizations
python src/visualize.py

Expected Results

| Metric | Target | |--------|--------| | Regime detection | Correctly identify 2008, 2020 crises | | Strategy Sharpe | 0.5-1.0 (after costs) | | Max Drawdown | Reduced vs buy-and-hold | | Volatility | Lower than benchmark |