Stock Pairs Trading

Pairs trading requires holding two positions simultaneously, so capital requirements are higher than single-stock trading. If you trade cash equities, you need close to full capital for both legs, plus the short leg must be borrowed via SBL (paying a borrow fee). A basic bank pair such as DBS-OCBC with roughly S$50,000 per leg needs about S$100,000 of buying power for the cash version. If you use CFDs for one or both legs, the broker's margin is only a fraction of notional - so the same pair might need S$10,000-20,000 of margin - but you pay daily financing and the leverage cuts both ways. A practical starting capital of S$20,000-50,000 (CFD route) or S$100,000+ (cash route) allows proper position sizing across one or two pairs. Singapore's uniform 100-share board lot makes sizing precise.

No, pairs trading is NOT risk-free. While it's market-neutral (immune to broad market direction), it has specific risks: (1) Spread blowout - the divergence continues instead of reverting, (2) Correlation breakdown - the historical relationship stops working, (3) Company-specific events - results, an S-REIT distribution cut, or a rights issue affecting one name permanently, (4) Execution and carry risk - slippage when entering/exiting both legs, plus accruing SBL borrow fees or CFD financing. You can lose 10-20% on a pairs trade if the spread moves against you significantly.

Most major Singapore brokers can support pairs trading, but look for: (1) The ability to short - either SBL borrow on cash shares or CFDs (Phillip Securities/POEMS, UOB Kay Hian, Maybank/KE Trade, DBS Vickers, CGS International all offer cash and SBL; IG, Saxo, CMC and the cash brokers' CFD arms offer CFDs), (2) Basket or multi-leg orders for near-simultaneous execution, (3) Spread or pairs analysis tools, (4) API access if you want to automate (Interactive Brokers API, Saxo OpenAPI, and Tiger/moomoo APIs are popular). Compare commissions, the minimum fee, GST on fees, and the SBL borrow rate or CFD financing rate, since carry costs erode narrow-convergence profits. AlgoKing integrates with multiple brokers and provides pairs analysis tools for simulation practice.

In Singapore markets, intraday pairs trades can work for the highly liquid bank pairs. Most pairs trades last 2-10 days. Bank pairs typically revert in 3-7 days, S-REIT pairs in 7-12 days. Set maximum holding periods based on the pair's historical half-life. If a trade hasn't begun reverting within about 1.5x the expected half-life, exit regardless of P&L - the original thesis may no longer apply, and borrow/financing cost keeps accruing. Be cautious holding pairs over weekends and around major US events (FOMC, US data), because Singapore rates and sentiment are heavily influenced by the US given the open economy and the MAS exchange-rate policy regime.

On SGX this is far simpler than in markets with large variable lots. The standard board lot is just 100 shares for most counters, and you can even trade single units on the Unit Share Market, so you can match SGD exposure very precisely. The key is matching SGD notional, not share count. For example, a DBS lot of 100 shares at S$45.00 is S$4,500 of notional, while an OCBC lot of 100 shares at S$16.50 is S$1,650. To equalise exposure you'd hold roughly 3 OCBC lots for every 1 DBS lot (about S$4,950 vs S$4,500), then fine-tune with odd lots. Because lots are small, you rarely need awkward ratios - you just scale each leg to the target SGD amount.

Distributions and dividends create an artificial spread change. When a stock goes ex-dividend (or an S-REIT goes ex-distribution) its price drops by roughly the payout, widening or narrowing the spread depending on which leg goes ex. This matters more in Singapore than in many markets because S-REITs distribute frequently (quarterly or semi-annual). Two approaches: (1) Avoid pairs with an upcoming ex-date within the holding period (check each name's corporate-action calendar), or (2) adjust your spread calculation for the expected ex-date drop. Note Singapore's one-tier system means dividends are tax-exempt in the shareholder's hands, so there's no dividend-tax complication - but if you are short the stock over the ex-date you must pay the distribution to the lender, which is a real cost to model.

The distance method simply tracks the normalized price difference (spread) and trades when it exceeds historical thresholds - it's simpler but makes no assumption about mean reversion. The cointegration method uses statistical tests to confirm the spread is actually mean-reverting, providing a stronger theoretical foundation. In practice the cointegration method has a higher win rate but fewer signals. The distance method generates more trades but with lower conviction. Most professionals use cointegration as the primary filter, then distance/Z-score for entry timing.

Hedge ratios aren't static - recalculate at least weekly. The relationship between stocks changes due to results, market-cap shifts, and evolving fundamentals. Use rolling regression (60 days for active trading) to update hedge ratios. If a hedge ratio changes dramatically (>20% shift) the pair may be entering an unstable period - consider reducing the position or avoiding the pair temporarily. For existing positions you can adjust by adding/reducing shares in one leg to maintain neutrality, but frequent adjustments increase transaction costs (commission, GST on fees, spread).

Yes, index vs constituent pairs trading (basis trading) is a valid strategy. Going long an index instrument and short an individual constituent (or vice versa) exploits deviations in the index vs component relationship. In Singapore the three local banks together are roughly half the Straits Times Index by weight, with DBS alone around a fifth, so a basis trade against DBS is effectively a large bet on the banking sector vs the rest of the index. Key considerations: (1) calculate the proper weight so the constituent leg matches its index weight, (2) you can express the index leg via an SGX index future (e.g., MSCI Singapore Index Futures) or an STI ETF, and (3) track the futures basis separately from the pair spread. This works well around index rebalancing events when weights change.

For active individual traders, 3-5 pairs is optimal. This provides diversification without overwhelming monitoring capacity. Each pair requires attention - watching both legs, monitoring correlation, tracking the spread, and watching borrow availability. More than 5 pairs splits focus too much and increases execution complexity. Capital efficiency also matters: in Singapore, where shorts are funded via SBL borrow or CFD margin, 3-5 pairs typically uses a manageable share of capital while keeping borrow costs visible. Systematic/algorithmic traders can handle 20-30 pairs with automation, but discretionary traders should stay focused - and even automated SG systems are capped by how many names have a reliable, affordable short borrow.

A Kalman filter provides an adaptive hedge ratio that responds to recent price action while smoothing noise. Implementation: (1) State equation: the hedge ratio follows a random walk plus noise, (2) Observation equation: Stock A return = hedge ratio × Stock B return + noise, (3) update the hedge ratio each period using the Kalman gain. Advantage over rolling regression: the Kalman filter adapts faster to regime changes while being more stable than a short-lookback regression. Python's pykalman or a manual implementation works. The key parameter is the process variance - higher values make the filter more responsive but noisier.

Joint hypothesis problem: when backtesting pairs you're simultaneously testing whether (1) your pair selection criteria work, (2) your entry/exit rules work, and (3) your position sizing works. If results are bad you don't know which component failed. Solution: test components separately. First, test whether selected pairs actually mean-revert out-of-sample (regardless of trading rules). Then, given mean-reverting pairs, test different entry thresholds. Finally, test sizing rules. Use walk-forward optimization with separate in-sample selection and out-of-sample validation periods to avoid overfitting.

PCA identifies common factors driving returns across stocks. In stat arb, decompose universe returns into principal components (PC1 might be the market factor, PC2 sector rotation, etc.). Calculate each stock's loadings on the PCs. A stock's residual return (after removing PC influences) represents its idiosyncratic movement. Pairs trade on residual spreads rather than raw price spreads - this isolates true relative value from common-factor moves. In a market like Singapore, where the three banks and a handful of S-REITs dominate the index, PCA is especially useful for stripping out the heavy bank/rate factor before judging a pair's idiosyncratic spread. PCA-based stat arb typically shows lower volatility and more consistent returns.

Microstructure matters significantly for pairs profitability. Key effects: (1) Bid-ask spread - you pay the spread on both legs, eating into narrow convergence profits; calculate round-trip cost as a % of expected profit. (2) Market impact - larger orders move prices; stagger entry for large positions, and note that beyond the big index names SGX liquidity thins quickly. (3) Lead-lag effects - one stock may lead another by seconds/minutes; the leader provides signals, so trade the laggard. (4) Order-book imbalance - check depth on both sides before entering. Solutions: use limit orders, break up large trades, and avoid illiquid times (the open and the late afternoon). Microstructure costs can consume 30-50% of gross profit if not managed.

Cost-robust design principles: (1) Widen entry thresholds beyond the theoretical optimal - enter at Z-score ±2.5 instead of ±2.0 so the move is significant enough to cover costs, (2) Include realistic Singapore costs in the backtest - SGX clearing fee (about 0.0325%), the SGX trading/access fee, brokerage (varies by broker, often around 0.08-0.28% or a flat online fee), 9% GST on commissions and fees, plus the SBL borrow fee on the short or CFD financing on both legs, (3) Filter by expected holding period - quick reversion means more turnover and more cost; prefer pairs with a moderate half-life (5-10 days), (4) Scale position with spread extremity - larger positions at Z-score ±3.0 where expected return is higher, (5) Partial exits reduce impact cost vs a full close, (6) Avoid pairs with wide bid-ask spreads or hard-to-borrow shorts. Net Sharpe after costs should exceed 1.0 for a viable strategy.