← all research
REJECTED

"Equity Stat-Arb — dollar-neutral L/S over 446-name walk-forward scores"

2026-08-23

Long top-N / short bottom-N by combined (OOS walk-forward scores, ml_active only). Costs: 0.100%/side slippage on turnover, 0.50%/yr borrow. Gross 1x, dollar-neutral, weekly/fortnightly rebalance, fills at next-day close-to-close.

NrebaltotalCAGRvolSharpeMaxDDPFbeta(SPY)corr(SPY)
205-33.0%-2.00%6.30%-0.29-52.5%0.940.060.17
2010-34.8%-2.10%5.80%-0.34-47.0%0.940.040.14
505-51.4%-3.60%4.00%-0.88-57.0%0.850.030.12
5010-39.5%-2.50%3.70%-0.66-45.7%0.890.010.06

Year by year (base cell N=20, weekly)

yearreturn
2006+0.00%
2007+0.00%
2008+0.00%
2009+16.93%
2010+10.23%
2011-6.80%
2012+0.48%
2013-9.95%
2014-2.43%
2015-8.15%
2016+0.86%
2017-10.39%
2018-0.65%
2019-3.36%
2020-3.19%
2021-3.07%
2022-12.67%
2023+0.44%
2024-1.30%
2025-9.92%
2026+5.34%

ALL DONE

Cost decomposition (N=20, weekly)

variantCAGRSharpePF
gross of costs (0 slip, no borrow)+2.67%0.451.09
net of 10bps/side + 50bps/yr borrow-2.00%-0.290.95

Verdict

REJECTED as stat arb. A faint raw cross-sectional spread exists (gross Sharpe ~0.45) but it is an order of magnitude below realistic trading costs at weekly rebalance turnover. The book IS market-neutral (beta ~0.03-0.06 across cells; random-rank control confirms machinery). Consistent with the 446-name expansion result: the walk-forward scores carry little broad rank discrimination. Note: an earlier version of this study reported +467% total - that was a weight-persistence bug (replace(0,nan)+ffill resurrecting stale positions, ~8.7x accidental gross leverage); fixed and re-run. Do not deploy.