Volatility Pairs Trading Strategy: Nifty & Bank Nifty

This repository contains a practical implementation of a volatility pairs trading strategy for Nifty and Bank Nifty index Futures . The approach is based on statistical arbitrage, aiming to profit from temporary divergences in implied volatility between these two highly correlated indices.

Data

• Source: data.parquet
• Fields: Nifty IV, Bank Nifty IV, Time To Expiry (TTE)
• Frequency: 1-minute bars

Strategy Models

Base Model

Core Logic:

• Calculates the spread: spread = banknifty_iv - nifty_iv
• Computes P&L: pnl = spread * (tte ** 0.7)
• Uses a rolling z-score of the spread to generate trading signals.
• Entry/Exit thresholds and lookback window are optimized by grid search.

Code Analysis:

• Data Loading: Reads parquet data, converts index to datetime, forward-fills missing values.
• Spread & PnL Calculation: Direct difference of IVs, scaled by time-to-expiry.
• Parameter Optimization: Grid search over entry/exit z-score thresholds and rolling window sizes.
• Signal Generation:
  • Enter long if z-score < -entry threshold
  • Enter short if z-score > entry threshold
  • Exit if z-score crosses exit threshold
  • Position states: Long (1), Short (-1), Flat (0)
• Performance Calculation:
  • Daily PnL aggregation
  • Annualized Sharpe Ratio
  • Drawdown and win rate metrics
  • Trade count and average trade duration (trading hours only)
• Risk Management:
  • Fixed position size
  • No stop-loss or dynamic sizing
  • No regime detection

Limitations:

• Assumes normal distribution of spread
• Static lookback window and thresholds
• No market microstructure or transaction cost modeling

---

Cumulative PnL - Base Model

Advanced Model

Core Logic:

• Uses a cointegrated spread: coint_spread = 38.8518 * nifty - 39.1766 * banknifty
• Applies volatility regime classification (low, medium, high) to adapt thresholds.
• Multiple rolling windows for feature engineering.
• Trades only during 10:00 to 15:00 IST.

Code Analysis:

• Cointegration:

  • Linear combination of Nifty and Bank Nifty IVs (coefficients from statistical analysis)
  • Produces a mean-reverting spread

• Feature Engineering:

  • Rolling mean and std for multiple windows (90, 120, 150, 180 min)
  • Z-score features for each window

• Regime-Based Trading:

  • Volatility regimes based on spread std percentiles
  • Entry/exit thresholds adapt to regime:
    • Low: (1.2, 0.5)
    • Medium: (1.5, 0.7)
    • High: (2.0, 0.6)

• Signal Generation:

  • Trades only during 10:00-15:00 (hour mask)
  • Filters trades by volatility regime
  • Dynamic position management (long, short, flat)

• Risk Management:

  • Regime-specific exposure limits
  • Volatility-based position filtering
  • No stop-loss, but more robust to changing market conditions

Cumulative PnL - Advanced Model

---

Key Formulas

• Spread:
  Spread = Bank Nifty IV - Nifty IV
• Cointegrated Spread:
  Coint_Spread = 38.8518 * Nifty IV - 39.1766 * Bank Nifty IV
• P&L:
  P&L = Spread × (Time To Expiry)^0.7

---

Project Structure

├── data.parquet          # Raw data
├── base-model.ipynb      # Base z-score model
├── advanced_model.ipynb  # Advanced regime/cointegration model
├── eda.ipynb             # Exploratory analysis
├── results/              # Plots and backtest results
└── grid_search_results.pkl  # Parameter optimization cache

Performance Summary

Metric	Base Model	Advanced Model
Absolute P&L	57,293.90	2,497,676.25
Sharpe Ratio	5.02	3.91
Max Drawdown	-6,770.12	-301,205.50
Max Drawdown %	-11.19%	-11.46%
Win Rate	65.26%	64.29%
Trade Count	4,858	3,926
Avg. Trade Duration (h)	0.71	0.87

---

Documentation

The full analysis is available in two formats:

• report.typ: Source file in Typst format
• report.pdf: Rendered PDF output

Requirements for editing documentation

• Typst (install via brew install typst)

Building the documentation

# Compile once
typst compile report.typ report.pdf

# Watch for changes
typst watch report.typ report.pdf

Assumptions & Limitations

• Assumes sufficient liquidity for execution
• Ignores transaction costs and slippage
• No market impact modeled

---