Dynamic-Regime-Allocation

A Fee-Aware, Regime-Switching Asset Allocation Strategy using Gaussian HMMs

📖 Overview

Dynamic-Regime-Allocation is a quantitative trading strategy designed for the Nifty 50 index. Unlike traditional forecasting models that attempt to predict exact prices, this system uses Unsupervised Learning (Gaussian Hidden Markov Models) to decode latent market regimes (Bull, Bear, and Sideways).

The primary objective is Risk-Adjusted Return. The strategy dynamically adjusts portfolio leverage based on the identified regime, aiming to capture market upside while preserving capital during structural downtrends (e.g., 2008, 2020).

Key Differentiator: This is not a theoretical model. It includes a rigorous Transaction Cost Analysis (TCA), factoring in brokerage fees, slippage, and the yield on idle cash (Liquid BeEs), making the results audit-ready and deployable.

---

Performance (2007–2025)

The strategy was backtested over an 18-year period covering multiple market cycles.

Metric	Market (Buy & Hold)	HMM Strategy
Net Return	319.27%	311.01%
Drawdown Risk	High (-60% in 2008)	Significantly Reduced
Total Trades	N/A	57 (Low Churn)
Fees Paid	0.0%	~5.6% of Capital

Visual Analysis: The strategy matches the market's long-term compounding but avoids the "emotional torture" of deep drawdowns. Note the flat-lining (protection) during the 2008 and 2020 crashes.

---

🛠️ Methodology

1. Feature Engineering

We feed the HMM three distinct features to separate market states:

• Volatility: 20-day annualized rolling standard deviation (Risk Proxy).
• Trend (Z-Score): Price distance from the 50-day Moving Average (Trend Proxy).
• Momentum: 14-day price difference (Velocity Proxy).

2. Unsupervised Learning (Gaussian HMM)

The model assumes the market operates in three hidden states. It calculates the probability of being in each state daily:

• 🟢 Bull Regime: Low Volatility, Positive Trend.
• 🟠 Sideways Regime: Mixed Volatility, Mean Reverting.
• 🔴 Bear Regime: High Volatility, Negative Trend.

3. Signal Smoothing (The "Churn Killer")

To prevent "whipsaw" losses (buying/selling every day), we apply a 10-day Moving Average to the regime probabilities. The model must sustain a signal for ~2 weeks before capital is committed. This reduced total trades from 441 to 57.

4. Fee-Aware Backtest Engine

• Transaction Costs: 0.1% per trade (Brokerage + STT + Slippage).
• Cash Yield: Idle cash (when not invested) earns a 6% annualized return, simulating parking capital in Liquid BeEs or Overnight Funds.

---

🚀 Strategy Logic

Market Regime	Signal Confidence	Action	Leverage
Strong Bull	P(Bull) > 80%	Aggressive Buy	1.5x
Bull	P(Bull) > 60%	Standard Buy	1.0x
Sideways	P(Side) > 60% & Dip	Buy the Dip	1.0x
Bear	P(Bear) > 60%	Sell / Cash	0x

---

💻 Installation & Usage

1. Clone the repository:

   git clone [https://github.com/SahilMotyar/Dynamic-Regime-Allocation.git](https://github.com/SahilMotyar/Dynamic-Regime-Allocation.git)
   cd Dynamic-Regime-Allocation

2. Install dependencies:

   pip install yfinance pandas numpy matplotlib hmmlearn scikit-learn

3. Run the strategy:

   python hmm.py

---

📡 Live Trade Signal

The script concludes by generating a Live "Trade Card" for the current day. It retrains the model on the most recent data to provide an actionable recommendation.

Example Output (2026-01-23):

📜 Disclaimer

This project is for educational and research purposes only. It does not constitute financial advice. Algorithmic trading involves significant risk. Past performance is not indicative of future results.