Pair Trading Strategy

Random stock pairs lack the statistical relationship needed for mean reversion. You need stocks that are fundamentally related (same sector, similar business) and statistically proven to have a cointegrated relationship. Without this, the spread can diverge permanently and never revert, causing losses.

Because shorting requires a margin account, and active trading triggers the Pattern Day Trader rule, you generally need at least $25,000 in equity. Practically, $50,000-100,000 lets you run 2-3 pairs with proper position sizing and diversification while keeping each pair's risk near 2% of capital. A Reg T margin account gives roughly 2:1 buying power on the combined long and short legs.

This is the main risk in pair trading. You have a stop loss at Z = 3.5 - if the spread widens beyond this, exit the trade accepting the loss. Also investigate why the spread diverged - it could be a structural break (M&A, a spin-off, a guidance cut) that means the pair is no longer valid.

Yes. The long leg is a normal stock purchase, and the short leg is a margin short - your broker borrows the shares for you to sell (a Reg SHO 'locate' is required). You pay a borrow fee on hard-to-borrow names and owe any dividend if you're short over an ex-date. Many traders use options for one leg to define risk, but plain stock legs work well for liquid, easy-to-borrow names.

Daily monitoring of Z-score and correlation is recommended. Weekly deep review of cointegration status. Set alerts for Z-score approaching exit targets (0.5) or stop levels (3.5). Also watch borrow cost on the short leg. Pair trades are not set-and-forget - active monitoring is essential.

Avoid new entries 3-5 days before either stock's earnings. For existing positions, consider reducing size or tightening stops. Earnings can cause temporary divergence that eventually reverts, but can also trigger stop losses. Some traders close all pairs during the peak of earnings season.

Stocks: No expiry, can hold indefinitely, but shorting needs a margin account, a borrow, and you owe dividends-in-lieu on the short. Options: Defined risk, leverage, and no borrow needed (a synthetic short is long put + short call), but theta decay and Greeks complexity, plus 100-share contract granularity. Note that U.S. single-stock futures are not practically available, so futures only enter at the index/ETF level. Choose based on holding period, capital, and comfort with derivatives.

Investigate the cause: Stock-specific news (FDA action, a guidance cut, management change) may be permanent for that stock. Market-wide fear causing divergence is usually temporary. If correlation drops below 0.60 for >2 weeks without an obvious temporary cause, assume the relationship may have broken and exit.

Ratio spread is generally better - it's normalized for different price levels and more stable over time. Price spread works for similar-priced stocks. Beta-adjusted spread is most accurate for proper hedging. Start with ratio spread, graduate to beta-adjusted for precision - this matters a lot for pairs like JPM/BAC that trade at very different dollar prices.

Optimal is 5-8 pairs for diversification without over-complexity. Too few (1-2) = concentrated risk. Too many (>10) = difficult to monitor and manage, with diminishing marginal benefit. Ensure sector diversification - maximum 2 pairs from the same sector.

Generate all sector pairs from liquid, optionable, easy-to-borrow stocks. Filter sequentially: correlation > 0.70, cointegration p < 0.10, half-life 3-25 days, stable relationship. Rank remaining pairs by composite score (cointegration strength, half-life, stability, liquidity). Walk-forward validate. Rerun quarterly.

For pair selection: XGBoost/Random Forest classification (profitable vs not). For entry timing: Gradient boosting with Z-score features + momentum + regime. For exit: Reinforcement learning or supervised regression. Ensemble ML with traditional rules - use ML to filter, not replace, fundamental logic.

Calculate volatility (std dev of daily P&L) for each pair. Weight = 1/Volatility, normalized to sum to 100%. Lower volatility pairs (banking, payments) get more capital, higher volatility (steel, materials) get less. Rebalance monthly. This equalizes risk contribution across pairs.

Track: (1) Average correlation among S&P 500 constituents (high = trending), (2) VIX level (>25 = elevated risk), (3) Market breadth (extreme = trending, mixed = mean-reverting). Combine into a regime score. Adapt allocation and thresholds based on regime.

Execute the less liquid leg first. Use algorithmic execution (TWAP/VWAP) for large sizes. Place limit orders to capture the bid-ask (orders route to the NBBO under Reg NMS). Track slippage and execution cost - it should be <20% of expected profit. For very large positions, spread execution over multiple days or use options to avoid borrow constraints.