Statistical Pairs Trading: Cointegration & Spread Mean-Reversion

The Market-Neutral Statistical Arbitrage Blueprint

Statistical Pairs Trading is the cornerstone of quantitative relative-value and market-neutral trading. Pioneered in the mid-1980s by quantitative researchers at Morgan Stanley (Gerry Bamberger and David Shaw), pairs trading eliminates broad market risk (beta) by simultaneously holding a long position in an undervalued asset and a short position in an overvalued, highly correlated peer.

Because both companies operate in the same sector and face identical macroeconomic headwinds, interest rate cycles, and regulatory environments, their shared systematic factor (Systematic Factor F) cancels out by mathematical construction. The trader profits purely from the mean-reversion of the idiosyncratic price spread back toward its historical equilibrium.

Correlation vs. Cointegration: The Vital Mathematical Distinction

Many retail traders fail at pairs trading because they confuse simple price correlation with statistical cointegration:

• Correlation (Short-Term Co-movement): Measures the linear co-movement of percentage returns over a fixed window. Two stocks can have high 90% correlation while their absolute price levels diverge permanently over time (e.g. one stock 10x while the other stagnates).
• Cointegration (Long-Term Stationary Equilibrium): Two non-stationary time series Price A(t) and Price B(t) are cointegrated if a linear combination of them forms a stationary series (Stationary Series I(0)) with a constant mean and finite variance: Spread = ln(Price A) - β · ln(Price B) - α. This guarantees mathematically that dislocations are temporary and will mean-revert.

Quantitative desks verify pairs using the Engle-Granger Two-Step Test and the Augmented Dickey-Fuller (ADF) unit root test before deploying capital.

Canonical Cointegrated Pairs on the National Stock Exchange (NSE)

Pairs trading requires high liquidity and low bid-ask spreads on both legs. On the NSE, prime cointegrated candidates exist within concentrated institutional sectors:

Z-Score Dislocation & Sizing Mathematics

To execute without visual guesswork, the price spread is normalized into a rolling statistical Z-Score (z-score) over a lookback window (typically 60 to 90 trading sessions):

Where Mean Spread is the rolling average spread and Spread Standard Deviation measures rolling spread volatility.

• Entry Trigger (|z| ≥ 2.0): When z ≥ +2.0, Stock A is statistically overvalued relative to Stock B. Simultaneously Sell Stock A and Buy Stock B. When z ≤ -2.0, Buy Stock A and Sell Stock B.
• Profit Target (z ≈ 0.0): Exit both legs when the spread reverts to the mean (|z| ≤ 0.2).
• Hard Stop-Loss (|z| ≥ 3.5): If the spread widens to 3.5 standard deviations, the cointegration thesis is considered statistically broken; flatten both legs immediately.

Worked Indian Trade Walkthrough: HDFC Bank vs. ICICI Bank

Consider a live statistical dislocation between HDFC Bank (Stock A, ₹1,700) and ICICI Bank (Stock B, ₹1,200) with a capital budget of ₹10 Lakh (₹5 Lakh per leg):

Execution in Indian F&O: Margin Optimization & Delivery Rules

In India, shorting equity shares for multi-day positional holding is restricted in cash equity (except via SLB - Securities Lending and Borrowing). Therefore, quantitative pairs trading is primarily executed using NSE Stock Futures contracts:

• SPAN Margin Offsetting: NSE Clearing offers substantial cross-margin relief for calendar and pair hedges within the same sector, reducing total required capital by up to 60%–70%.
• Lot Size Alignment: Because NSE stock futures trade in fixed lot sizes (e.g. 550 for HDFC Bank, 700 for ICICI Bank), share count sizing must be integer-rounded to the nearest lot ratio that maintains dollar neutrality.
• Physical Delivery Avoidance: Under SEBI physical delivery settlement rules, all in-the-money and open stock futures positions must be rolled over or squared off at least 3 days prior to monthly expiry to avoid massive delivery margin spikes.