Short-Term Momentum Just Had Its Worst Run Since 2020

• Short-Term Momentum has completed the weakest 63-day stretch of any window since 2020, the 63-day z=-3.7 — three months in which each week's leaders were faded in turn.
• Marvell is -7.93% pre-open despite record quarterly revenue, erasing $17.2B of market value on third-quarter gross-margin guidance of 57.5%–58.5%.
• Liquidity is the largest signal in the pre-open cross-section at -0.31%, z=-1.1, a return in the 4.5th percentile since 2020, with the highest-turnover AI and semiconductor names on the losing end.
• SMH rose +3.10% Thursday and is -0.99% pre-open while KWEB has gone from -0.68% to +1.36% — Thursday's leaders are the ones giving back.

August 28, 2026 · 08:17 ET · pre-open

The number that matters this morning printed at last night's close. Short-Term Momentum's trailing 60-day return is -7.25%, and the completed 63-day window through Thursday is the weakest of all 1,538 overlapping windows since 2020 (z=-3.7). Last week's leaders were faded, and the week before that, and the week before that. Composition is the part that gets misread: the basket re-sorts every week, so after a rebound week its winners are largely the names that crashed the week before. What the record describes is a recurring pattern, each leg faded in turn, not one position being ground lower.

The fast end has decayed, and that is the limit on the record. Today's print is +0.04% (z=+0.1), no reversion at all. The five-day z is -0.9 and the 20-day z=-1.5 against the 63-day's -3.7, and the trailing five-day cumulative return is only -0.58%: most of the damage is more than a month old. Implied single-stock volatility (Cboe VIXEQ) at 36.52 against implied index volatility (Cboe VIX) at 14.48 is the environment this pattern feeds on — single names churning under a quiet index.

The mechanism behind a run this deep is documented. Goldman Sachs Prime Brokerage disclosed that 16% of its total prime-brokerage exposure sat in AI memory-related names as of mid-2026, with margin calls following July's drop in AI stocks; a three-month reversion run of this depth is indicative of the deleveraging and releveraging churn that follows an episode like that.

Where the Pre-Open Selling Is Landing

Liquidity carries this morning's cross-section: -0.31% at z=-1.1, a return in the 4.5th percentile since 2020, and 3.1% of cross-sectional variance — above its 90th percentile of the past year, so the axis is doing unusual work for a session that has not opened. The direction is specific. The lowest-turnover names outperformed the highest-turnover ones, and the losing end is the top-turnover bucket at -0.76%. That bucket is where the heaviest options activity sits, and this morning it is AI, semiconductor and data-center names — SNDK, IREN, BBAI and LUNR among them, -2.37% weighted across the five.

These are not company stories. Sandisk is -2.29% with the factor complex accounting for -3.58% and the stock's own move a positive +1.20%; AXT is -2.87% against -3.80% from factors. The top 50 high-beta momentum names are -1.19% while everything outside that screen is +0.16%.

Marvell's Margin Guide Sets the Chip Tone

Marvell is -7.93% pre-open after Thursday's post-close report, erasing $17.2B of market value on record quarterly revenue of $2.74 billion, up 37% year over year. The mix is the problem: management guided third-quarter non-GAAP gross margin to 57.5%–58.5% from 58.9%, roughly 90 basis points lower sequentially as lower-margin custom AI silicon ramps, with the bulk of the Google custom-chip revenue landing in fiscal 2029, according to Benzinga's and Yahoo Finance's coverage of the call. Marvell's Liquidity exposure of +2.9σ and Beta of +2.9σ put it at the extreme end of the very axis that led this morning. Autodesk is -4.37% after its own post-close printBloomberg.

The complex is handing back Thursday's gains rather than breaking down, and the ETF grid shows it cleanly in the one-day-ago column.

Information Technology's median stock is -0.45%, against +0.36% to +0.48% for Materials, Consumer Discretionary, Staples and Communication Services — the pressure is one sector deep, not market-wide.

The curve is a non-event so far: the 10-Year sits at 4.69% and the 30-Year at 5.21%, each up 2 basis pointsCNBC, and the Treasury factor is flat at z=-0.1 even though duration-sensitive names have lagged for a week (five-day z=-1.4). Stocks and bonds are wavering into Warsh's Jackson Hole speechBloomberg. Precious metals are holding: SLV +1.28% with spot silver above $70Economic Times, and the Gold factor's completed 20-day return in the 99.7th percentile since 2020 despite a flat print today.

The Cross-Section Belongs to the Reporters

Away from chips, the pre-open board is earnings gaps. Elastic is +22.98%, adding $2.11B of market value after its fiscal first-quarter report, following a Thursday software session in which Okta gained 29% and CrowdStrike posted its best day everCNBC. Gap +16.19% and Affirm +12.96% report into macroeconomic data showing the US consumer under strainBloomberg, while PayPal is -15.52% and $8.12B lighter after suitors reportedly walked away from a takeoverBloomberg. KE Holdings is +11.29% on a second-quarter beat alongside Beijing's new housing-presale circular and S&P Global's affirmation of China's A+ sovereign rating, part of the International factor's +0.15% (z=+0.5) with UMC +9.94%. One mover to discard: Vistance Networks at -41.81% is the ex-dividend adjustment for a $5.00 special distributionBloomberg.

Every one of those gaps is a fresh input into next week's fast-horizon sort — which, on the evidence of the past three months, is where they get worked against.

Economic Context

The Michigan Consumer data.sca.isr.umich.edu Survey (Final) is due at 10:00 AM ET, with consensus centered on 51.0–52.0 after the preliminary slump to 51.0 snapped two months of improvement; one-year inflation expectations rose to 4.3% in the preliminary report, the key upside risk in the release. The New York Fed Staff Nowcast follows at 12:45 PM ET, currently at 2.3% for Q3 2026 and well below the Atlanta Fed's 4.6% GDPNow estimate.

Data compiled by FactorPulse AI; edited and verified by Jeff Klein. For informational purposes only. Does not constitute financial advice, an investment recommendation, or an offer to buy or sell any securities. Always consult a qualified financial professional before making investment decisions.

For more on factor construction methodology, see www.factorpulse.com/glossary.

← FactorPulse  ·  All Digests