The GS desk called Sep 14 a de-grossing of the AI-infrastructure right tail — electrical, power, cooling, optical and data-center names fell 500–800bps beyond expectation, "the baby went out with the bathwater." The tempting trade is to fade the winners-vs-infra ratio, but that fails the forward test (AIWIN/AIRISK is a momentum ratio, not a reversion one — shorting it at these extremes has lost, badly, in the current regime). The edge is one level down: in the individual names. A single infra name that washes out in one session has reliably bounced — buying the NAME the day after a bottom-decile drop, across this group, wins 68% of the time for a +5.9% median at 21 sessions. That is where the "oversold" thesis actually pays.
Across the 21 de-grossed infra names, this ranks each one's own washout-reversion history: buy the close the day AFTER a bottom-decile down day, hold ~3 weeks. The leaders — DELL, ANET, GEV, LITE, CIEN, APH — carry 69–71% win rates with +4.7% to +7.0% medians. Click any row to see that name's forward-path distribution. Data through 2026-09-15 (the daily EOD bar lags a session, so the live 1-day moves land shortly).
The top six as an equal-weight washout basket: +5.9% median / 68% win at 21 sessions, +3.4% / 65% at 10 — a genuine reversion drift, not one fat tail. The green rows clear a 55% / +3% bar. Samples vary (names like GEV/CRDO/ALAB have shorter public histories); treat the deep-history names (CIEN/ANET/HUBB/ETN, n>400) as the most robust and the young ones as directional.
| Name | Px | Washouts | +10d | +21d | |
|---|---|---|---|---|---|
| DELL | 549.38 | 144 | 68%+3.82% | 69%+7.05% | chart › |
| ANET | 191.98 | 146 | 64%+2.86% | 68%+6.55% | chart › |
| GEV | 886.72 | 59 | 66%+4.40% | 71%+6.20% | chart › |
| LITE | 847.76 | 142 | 63%+4.44% | 65%+6.10% | chart › |
| CIEN | 324.59 | 143 | 62%+2.44% | 62%+4.87% | chart › |
| APH | 78.59 | 146 | 65%+2.21% | 71%+4.71% | chart › |
| MOD | 183.49 | 147 | 58%+2.73% | 62%+4.65% | chart › |
| PWR | 624.69 | 145 | 68%+4.21% | 68%+4.43% | chart › |
| CRDO | 151.96 | 112 | 56%+1.79% | 57%+3.70% | chart › |
| HUBB | 444.5 | 146 | 64%+1.29% | 64%+3.38% | chart › |
| NVT | 150.21 | 146 | 71%+3.63% | 63%+3.34% | chart › |
| VRT | 242.21 | 145 | 60%+2.70% | 56%+3.25% | chart › |
| TLN | 289.0 | 81 | 64%+4.31% | 63%+2.64% | chart › |
| ETN | 399.77 | 145 | 66%+2.17% | 63%+2.43% | chart › |
| VST | 142.25 | 148 | 55%+1.50% | 55%+2.25% | chart › |
| CEG | 265.78 | 115 | 61%+2.08% | 57%+2.18% | chart › |
| SMCI | 37.22 | 149 | 56%+3.65% | 52%+1.98% | chart › |
| NRG | 108.93 | 143 | 59%+1.58% | 59%+1.84% | chart › |
| COHR | 277.2 | 144 | 58%+2.75% | 54%+1.60% | chart › |
| POWL | 172.23 | 145 | 50%-0.07% | 52%+1.15% | chart › |
| ALAB | 261.74 | 60 | 58%+3.81% | 50%+0.19% | chart › |
Structure — long the washout basket, equal-weight the top names. Rather than pick one, buy the 6-name basket of the best reversion candidates (DELL, ANET, GEV, LITE, CIEN, APH) the day after they wash out, sized equally. Hold ≈ 2–3 weeks (the edge peaks near the 21-session mark). Expressed in shares — it is a reversion, not a convex bet. Size at ¼-Kelly, capped at the house 0.75% NAV rule → ≈ $3,750 at risk on the 500k book, split across the names. Buy the deepest de-grossers within the leaders as they print.
TEF cross-check. Probability-first: a 68% / +5.9% basket edge with per-name evidence clears the bar. Stress-reversal: we are buying exactly the side that got obliterated — the de-grossed infra — which is the fund's stance. Exit-before-entry: invalidation is a name making a fresh lower low and holding it (the p25 adverse path runs to roughly -3.0% on the leaders); cut a name that keeps making new lows two sessions after entry. Risk budget: one ¼-Kelly clip total, $3,750, split across the basket. Macro caveat (GS): rates and oil must stabilize before valuation reversions truly work — this is a tactical washout clip, not a risk-on bet, and GS himself says nothing needs to bounce immediately.