Cross-Sectional Momentum (XSMOM, 截面动量)

Buy past winners, sell past losers, relative to the cross-section. The "Jegadeesh-Titman" momentum. Distinct from Time-Series Momentum.

Definition

At each rebalance, rank all assets by their past return over a lookback window (typically 12 months, skipping the most recent month), go long the top decile and short the bottom decile. The position in asset i is:

w_i^{XSMOM} = \frac{1}{N} \left(\text{rank}(R_i^{past}) > 1 - q\right) - \frac{1}{N} \left(\text{rank}(R_i^{past}) < q\right)

for top/bottom deciles (q = 0.1). Many variants: quintiles, terciles, residual momentum (against a factor model), industry-neutral momentum.

Canonical reference

Jegadeesh & Titman (1993), "Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency." Journal of Finance, 48(1), 65–91. Documented in US equities 1965–1989.

TSMOM vs. XSMOM

Moskowitz et al. (2025) compare the two head-to-head on the same futures universe:

Strategy 12M lookback, monthly Sharpe
XSMOM (cross-sectional) 0.48
Linear TSMOM higher
Nonlinear TSMOM (FS or empirical) ~0.65

Key differences:

  • TSMOM conditions on the asset's own past — long the asset if it went up. Works on a single-asset basis.
  • XSMOM conditions on the asset's rank relative to peers — long the winner relative to losers. Requires a universe.
  • TSMOM has convex payoff in equity-market returns (extreme up or down markets); XSMOM does not have the same tail-hedging property.
  • TSMOM is more robust in 1M and 3M lookbacks; XSMOM is statistically significant mostly at the 12M horizon.

Relationship to Ferson-Siegel Optimal Weighting

The 2025 Moskowitz et al. paper notes that whether the FS S-curve weighting also improves XSMOM is left for future research. The theory applies to any signal-to-position mapping, so it should transfer, but the empirical test is open.

容量与交易成本:等权是最容易崩的(2026-10-21 补记)

Korajczyk & Sadka (2004, JF 用日内数据估了比例型(佣金、价差)与非比例型(价格冲击)交易成本,算出动量策略的盈亏平衡资金规模:

口径 金额(1999 年 12 月市场规模基准)
收益被成本驱动到零 $4.5 – 50 亿以上
收益统计上不再显著 $1.1 – 20 亿

最反直觉也最重要的一条:等权策略在扣成本前表现最好,扣成本后表现最差。

机制:等权意味着同等资金分配到每只股票,因此在流动性最差的小盘股上仓位最重——冲击成本最高。

论文构造的解法:流动性加权动量组合(按流动性排序以降低交易成本),其盈亏平衡资金规模最大(≥50 亿美元),另有混合流动性/价值加权版本。

对 A 股的直接含义:A 股小盘股占比更高、T+1 额外加了持仓约束、涨跌停板让建仓/平仓在极端行情下做不成——这三重成本论文完全没考虑。因此 A 股动量的真实容量应当比论文的估计更紧。这是一个可以直接落地的实证课题:把论文的「流动性加权」构造改成 A 股版(用换手率或 Amihud ILLIQ 排序),检验 T+1 + 涨跌停下还剩多少容量。

论文保留的结论:即使按最保守的成本假设,交易成本也不足以完全解释动量的持续性——「This anomaly remains an important puzzle.」

崩盘风险:这是 XSMOM 尚未被处理的问题(2026-10-20 补记)

Daniel & Moskowitz (2016 的分析针对 WML(赢家减输家)零成本组合,而 XSMOM 的主用法是只做多赢家这一半。两者的崩盘风险不同:

WML(多空) XSMOM(只做多赢家)
崩盘时谁在亏 做空输家的一侧被反噬——输家在熊市后急弹 赢家组合自身的下跌
崩盘幅度 极端(1932 年输家 +232%、赢家仅 +32%) 本文没有单独给出
机制解释 输家的高 beta 被做空 本文未覆盖

论文对只做多组合的适用性没有明确说明——这是本 wiki 记录的一个空白。

可以确定的一点:论文指出市场大跌之后,过去输家十分位的 beta 可以升到 3 以上,赢家十分位降到 0.5 以下。若只做多赢家,则持有的是低 beta 那一半——市场反弹时这一半受益有限,市场继续跌时跌幅也较小。这个推断需要实证检验,本文未做。

实践上更直接的风险是:只用波动率管理(如 波动率管理组合(Volatility-Managed Portfolio))对动量崩盘是不完整的——崩盘的驱动是条件均值的骤降,需要同时预测动量溢价的条件均值。

See also

References