Quant Momentum Strategy

Historically, momentum strategies in US markets have generated 3-5% annual alpha over broad market indices like the S&P 500. However, returns vary significantly by year - some years +15% alpha, others -5%. Expect higher volatility than index investing. Long-term (5+ years) investors benefit most from momentum's compounding alpha.

25-35 stocks is the sweet spot. Fewer than 20 creates concentration risk and high tracking error. More than 50 dilutes momentum exposure and approaches index returns. 30 stocks provides good diversification while maintaining meaningful momentum tilt.

Research shows 12-month (excluding recent month) is optimal. 1-month has too much noise and reversal effects. 6-month works but misses some momentum. 12-month captures the full momentum cycle. The '12-1' formulation (skipping recent month) further improves by avoiding short-term reversal.

Monthly is standard - it captures momentum shifts while keeping turnover manageable (~200% annually). Quarterly rebalancing reduces costs but may miss momentum changes. Weekly is too frequent and costly. Monthly with buffer zone (exit at rank 40, not 30) optimizes the trade-off.

Yes, momentum can and does lose money, especially during 'momentum crashes' when past winners sharply underperform. Crashes of -20% or more happen every few years (2009, 2020, 2022). However, over full market cycles (5+ years), momentum has historically delivered positive alpha. Risk management (volatility scaling, stops) helps manage drawdowns.

Multiple approaches: (1) Volatility scaling - reduce exposure when VIX > 25. (2) Portfolio stop-loss at -15% drawdown. (3) Factor monitoring - reduce if momentum factor underperforms 2+ months. (4) Diversify with other factors (quality, low volatility). (5) Hedging with index puts during elevated risk. Combining approaches provides layered protection.

Equal weight is simpler and more diversified - each stock gets same allocation. Momentum-weighted concentrates on strongest stocks for potentially higher returns but more risk. Risk-parity (inverse volatility) weighting is another option that balances risk. For most investors, equal weight with sector constraints works well.

Several approaches: (1) Sequential screening - first filter for momentum, then quality. (2) Composite scoring - weighted sum of momentum, quality, value scores. (3) Separate portfolios - allocate 60% to momentum, 40% to quality, rebalance between them. (4) Intersections - only buy stocks in top quartile of BOTH factors. Start simple (sequential) and add complexity as needed.

For liquid S&P 500 stocks: commission ~$0 at retail brokers, SEC fee + FINRA TAF negligible (~0.003% on sells), impact cost 0.05-0.15%. Total round-trip: ~0.10-0.20% (impact-dominated; no STT or stamp duty - a key difference from India). With 200% annual turnover, expect ~0.15-0.20% annual cost - materially lower than India. Reduce via: buffer zone (-20% turnover), TWAP/VWAP execution (-30% impact), and index futures (E-mini) for beta adjustments, which also get Section 1256 60/40 tax treatment. The larger cost lever in the US is tax: monthly rebalancing produces mostly short-term gains (ordinary income, up to ~40.8%), so prefer > 1-year holds where possible and mind the 30-day wash sale rule when harvesting losses.

Key requirements: (1) 10+ years data, survivorship-bias free (include delisted stocks). (2) Point-in-time universe (S&P 500 as of each date, not current constituents). (3) Walk-forward testing (train on 70%, test on 30%, roll forward). (4) Include realistic transaction costs. (5) Out-of-sample validation - expect 20-30% performance decay from in-sample. Use out-of-sample results for planning, not in-sample optimized results.

Feature engineering: 12-1 return, risk-adjusted momentum, relative strength, earnings momentum, RSI, sector momentum, VIX. Target: Top tercile classification or return regression. Model: XGBoost or LightGBM (best for tabular data). Training: Time-series split, walk-forward retraining monthly. Integration: ML probability as 40% weight combined with 60% traditional momentum. Expect 2-3% additional alpha from ML enhancement.

Long leg: Buy top 40 stocks by momentum (50% of capital). Short leg: Short bottom 40 by borrowing and selling the actual shares (another 50%). Gross exposure 200%, net exposure ~0%. In the US, single-stock futures aren't actively traded, but the securities lending market is deep - you can short virtually any liquid S&P 500 name. Considerations: Reg SHO locate requirement, borrow costs (low for large caps, high for hard-to-borrow names), and the alternative uptick rule (Rule 201) that restricts shorting a stock only after a 10% intraday drop. The broad shortable universe is a structural advantage over India. Rebalance monthly, monitor shorts more frequently (squeeze risk). Higher Sharpe but more complex execution.

Risk parity weights each stock inversely to its volatility so each contributes equal risk. Formula: Weight_i = (1/Vol_i) / Sum(1/Vol_j). Low-vol stocks get more capital, high-vol less. Combined with momentum: select by momentum, weight by inverse vol. Add volatility targeting: scale total exposure to maintain target portfolio vol (e.g., 15%). Result: Momentum exposure with balanced, consistent risk.

Base allocation 100%. Modifiers: VIX > 25 → -25%; VIX > 35 → -50%. Momentum factor negative 2 months → -25%. Bear market regime → -25% to -50%. Ranging market → -25%. Calculate weekly, but smooth transitions over 2-4 weeks. Example: VIX 30 (-25%), factor weak (-25%) = 50% allocation. When multiple signals negative, compound reductions but maintain 25% minimum.

Data pipeline: Automated daily price/volume updates, corporate action adjustments, quality checks. Execution: Broker API integration, order generation from targets, TWAP/VWAP algorithms. Risk: Real-time position monitoring, drawdown alerts, factor exposure tracking. Reporting: Daily P&L, weekly risk, monthly attribution. Governance: Change management process, backtest before modifications. Start simple, add sophistication as AUM grows.