← All setups
ONGOING · EXPLORATIONresearch in progress

Stock cross-sectional momentum — exploration

stock-momentum-xsec
This is an ongoing exploration, not a frozen test. This page itself is the free in-sample sandbox — we iterate on the parameters to see where an edge might live. The counted, publishable tests are the separate frozen confirmations it fed. The 2023+ holdout has now been used once, for two frozen confirmations (D-005, D-006) — see the Lab discovery log. Headline: the momentum signal confirmed out-of-sample, and its beta-neutral packaging cleared the sealed holdout (a forward-paper candidate, not a live edge).
12,747 US common stocks · 5,411 (42%) delisted2004-12 → 2022-11 (216 months, in-sample)raw long-short 2.3%/yr · DD 68% · Sharpe 0.22risk-managed 8.4%/yr · DD 33% · Sharpe 0.53

Does cross-sectional momentum — buying recent winners, shorting recent losers — carry a tradeable signal among individual US stocks? On a survivorship-free universe (delisted names included), the signal is real and monotonic, and lives in the market-neutral long-short, not in beating the market long-only. Sweet spot: a ~9-month lookback (skipping the last month) held ~3 months.

The long/short effect — mean next-month return rises monotonically from the worst-momentum decile (D1, short) to the best (D10, long). This is the raw cross-sectional signal.
The long/short effect — mean next-month return rises monotonically from the worst-momentum decile (D1, short) to the best (D10, long). This is the raw cross-sectional signal.
The effect over time — growth of $1 in the market-neutral long-short (winners minus losers) vs the long-only top decile, the short leg, and SPY. Note the momentum-crash drawdowns.
The effect over time — growth of $1 in the market-neutral long-short (winners minus losers) vs the long-only top decile, the short leg, and SPY. Note the momentum-crash drawdowns.
Where the signal lives — decile-spread t-stat (top) and long-short net return per year (bottom) across lookback × holding, per liquidity tier. The sweet spot is a ~9-month lookback held ~3 months.
Where the signal lives — decile-spread t-stat (top) and long-short net return per year (bottom) across lookback × holding, per liquidity tier. The sweet spot is a ~9-month lookback held ~3 months.
How many stocks participated — the investable survivorship-free universe per month (delisted names included while listed), with the liquidity tiers marked.
How many stocks participated — the investable survivorship-free universe per month (delisted names included while listed), with the liquidity tiers marked.

Signal strength — most significant configurations

universelookbackholddecile spread /yrtIClong-short net /yr
top-30003-03mo6.3%3.30.013+3.1%
top-30003-13mo5.6%2.90.012+2.4%
top-30003-06mo4.3%2.80.016+2.7%
top-10003-03mo6.4%2.50.013+3.2%
top-5009-03mo7.7%2.30.022+4.5%
top-30006-03mo5.4%2.20.016+2.2%

Taming the momentum crash — variant comparison

constructionnet /yrmax DDSharpe
raw long-short+2.3%68%0.22
vol-target 3mo · cap1.5+7.6%45%0.43
short-guard (½ short in stress) ★+8.4%33%0.53
beta-neutral (matched pairs)+2.2%65%0.21
beta-neutral + vol-target 3mo+4.6%64%0.31

Survivorship-free US common-stock history extends back to 2004, so this in-sample window includes the 2008–09 momentum crash. The crash-guard scales the long-short inversely to its own trailing volatility; the best-Sharpe refinement shown is short-guard (½ short in stress). Holdout 2023+ remains sealed. The long-only top decile lags SPY after cost, so a tradeable version is market-neutral (long winners / short losers). The raw long-short is uninvestable (~68% drawdown in the 2009 crash), so we tame it: faster volatility-targeting (scale down when the strategy's own vol spikes) already cuts the drawdown to ~45%, but the biggest win is a short-leg guard — halving the short position in high-volatility months, because the crash is driven by beaten-down losers rocketing in the rebound. That lifts the Sharpe from 0.22 to 0.53 and cuts the worst drawdown from 68% to 33%. In-sample, beta-neutralising the long-short (the "matched-pairs" idea) looked to barely matter — the average net beta was only −0.10. The sealed holdout overturned that: out-of-sample the naive long-short carried market beta ≈+1.0 (2023-24 momentum ran into high-beta mega-cap winners), so the unhedged version is not market-neutral when it counts. The frozen beta-neutral confirmation (D-006) cut realised holdout beta to +0.14 while keeping +17.7%/yr (Sharpe 1.04) — it cleared the gate. Net read: cross-sectional momentum is real and confirmed out-of-sample; its beta-neutral packaging clears the holdout too, though the in-sample neutral edge is thin, so this is a forward-paper candidate, not a live edge. See the frozen confirmations D-005 / D-006 in the discovery log. Winsorised returns, $5 price floor, net of 40 bps round-trip.

Mechaniq provides information, not investment advice. We do not execute trades. Past results are no guarantee for future performance. You are solely responsible for your trading decisions.

Lab · Methodology · Greeks Lab · About · FAQ · Glossary · Privacy · Terms · mechaniq.trade © 2026

07 Jul 2026, 07:55