PCA residual stat-arb
This is a statistical arbitrage strategy that looks for stocks moving out of line with the market. It removes the main shared factors, found by principal component analysis, from the daily returns of 20 large caps, and watches what is left over for each stock. When that leftover drifts two standard deviations from its usual level, it buys the laggard's futures or sells the leader's, betting on a return to normal.How it trades
Results
PRE-BUILT · DEFAULTSWhere it made and lost money
Although the book holds longs and shorts, its results followed the NIFTY backdrop. Positions opened on days NIFTY closed above its open made +₹2,17,695 over 44 trades; those opened on days it closed below lost ₹1,30,355 over 35. Trades held while NIFTY rose over 1% earned ₹1,26,148 over 21, while trades held through a flat market (within ±1%) lost ₹60,755 over 30. Normal India VIX between 11.5 and 14.5 made ₹1,19,998 over 42 trades; calm evenings below 11.5 lost ₹57,698 over 22. The longs and shorts are sized per day's signals, so the book is often not balanced and keeps a market tilt; quiet markets also give residuals little room to snap back.
About this strategy
- Each evening, take the last 60 daily returns of the 20 stocks and find the 5 main factors that drive them together (principal components).
- Regress each stock on those factors; the running sum of what is left is the stock's residual.
- Fit a mean-reverting model to each residual. Stocks whose residual reverts slower than a 30-session half-life get no score; the rest get an s-score, how stretched the residual is.
- At 09:16 next session, buy the near-month future of any stock with an s-score of -2 or lower and sell any with +2 or higher. Each stock is its own position.
- Size each side's new names by inverse volatility against half of a ₹50 lakh book, capped at 15% of the book per stock, at least 1 lot.
- Close a long once its s-score is back to -0.5, a short once back to +0.75, or any position after 15 sessions.
- On monthly expiry, positions close at 15:15 and reopen in the next month at 09:16 the following session.
Why traders use it
Large-cap stocks share common drivers such as the market, sectors and interest rates. Removing those leaves the part of each move that is specific to the stock, which often fades when it was caused by short-lived flows rather than news. Running the idea across many stocks at once, with both longs and shorts, spreads the risk of any one call. It is a classic quantitative approach in global markets, adapted here to NSE stock futures.
The year in plain words
From 1 Oct 2025 to 30 Sep 2026 on default settings the book closed 80 trades, won 55% and made ₹53,487 before costs (profit factor 1.08). After ₹20 per order and bid-ask slippage only ₹12,472 was left, a profit factor of 1.02. Of the exits, 44 were the s-score coming back, 30 monthly rolls and 4 time stops; the median hold was 7 days. March made ₹82,546 and November ₹69,320, but January lost ₹87,915 and June ₹73,848, and the drawdown reached ₹2,03,095 before costs. Seven of 12 months were positive. Costs took most of a thin edge.
What the settings do
Risks
- There is no price stop; a stock whose residual keeps stretching on real news is held until the 15-session limit, and one trade lost ₹79,065.
- Longs and shorts are sized from each day's signals, so the book can be mostly long or mostly short and carries market risk.
- The edge before costs was thin, and ₹20 per order plus slippage removed about three-quarters of it in this test.
- A ₹50 lakh gross book across many stock futures needs substantial margin, and the drawdown exceeded ₹2 lakh.
Who it suits
Quantitatively minded traders with large margin who want a diversified, rules-based book of stock futures, can handle many orders and monthly rolls, and accept that the result after costs was close to flat in this test.
Questions traders ask
Is PCA statistical arbitrage profitable on NSE stock futures?
Only slightly, in this test. From 1 Oct 2025 to 30 Sep 2026 the default book made ₹53,487 before costs and ₹12,472 after ₹20 per order and bid-ask slippage, with a drawdown of ₹2,03,095 before costs. Statutory charges are not included, and they would weigh on it further. Past results do not guarantee future ones.
What is the s-score in stat-arb?
It measures how far a stock's residual, the part of its move not explained by common factors, sits from its estimated long-run level, in units of that residual's typical swing. An s-score of -2 means the stock has lagged unusually; +2 means it has run ahead. It works like a z-score on a mean-reverting model.
Is this a pair trade?
Not exactly. Each stock is its own futures position, long or short. The common factors play the role a partner stock plays in a pair, and longs and shorts are each sized from half the book, but they are not matched one to one, so the book is often unbalanced.
Why are there so many Wednesday trades?
Stock futures expire monthly, so every open position is closed at 15:15 on expiry day and reopened the next session. Each reopening counts as a new trade, and with the monthly expiry on a Tuesday many of them fall on Wednesday. Wednesday entries lost ₹81,762 over 36 trades in this test.