Cross-Sectional Alpha Signals: Ranking, Neutralization, and Long-Short Formation

A raw alpha signal — a momentum score, a valuation ratio, an S-score — cannot be directly plugged into a portfolio. Applied directly, it exposes the portfolio to unintended sector bets (e.g. growth stocks happening to be the high-momentum names at a given time), market-cap biases (large caps screened differently than small caps), and numerical inconsistencies across different signal scales. The cross-sectional pipeline normalises, neutralises, and ranks raw signals into a clean long-short score that bets on relative differences within a peer group while keeping systematic exposures flat.

This note covers the mechanics of that pipeline: from the taxonomy of alpha signal types to the z-scoring and neutralisation steps, to the final long-short formation. The result is a per-stock forecast $\alpha^i$ that is directly comparable across the universe and ready for IC-weighted combination.

• classify alpha signals by their underlying phenomenon (trend, reversion, value/yield, growth, quality);
• compute a cross-sectional z-score from raw signal values;
• neutralise a signal for sector exposure and market-cap bias;
• form a dollar-neutral long-short portfolio from the neutralised signal;
• explain why neutralisation is not optional for production stat-arb.

1. 1. A taxonomy of alpha signals
2. 1. From raw signal to z-score
3. 1. Neutralisation
4. 1. Long-short portfolio formation
5. 1. Signal taxonomy in production systems