As the S&P 500 keeps setting new all-time highs, something entirely different is unfolding beneath the surface. Hedge funds are selling technology stocks on one hand while pushing their "Mag 7" holdings to an all-time peak on the other; confidence indicators have fallen to extreme lows, yet positioning data across asset classes continues to set record after record. Goldman Sachs' latest weekly report reveals that beneath the calm surface, the market's structural tensions have never been greater.
Last week, the S&P 500 rose 1.2% to 7,811 points, setting another all-time high. However, according to Goldman Sachs data, hedge fund net leverage stands at just 46.9%, in the 2nd percentile over the past year; the long-short ratio has fallen further to 1.565, in the 1st percentile on a five-year lookback. At the same time, technology stocks saw consecutive net selling, with the worst-performing sector being information technology, which had been the most favored sector four times in the previous four weeks.
The market's structural divergence is also reflected in derivatives pricing. Goldman Sachs' derivatives team notes that VIX upside convexity is extremely in demand, while S&P 500 downside protection is "extremely cheap" — in the words of its colleagues, "the probability of VIX rising above 25 is almost higher than the probability of the S&P 500 falling 2%." This anomalous signal indicates that investors' pricing of tail risk has shifted direction, from worrying about a stock market decline to guarding against a jump in volatility itself.
Rotation Direction: From Chips to Aircraft and Missiles
Fund flows are the clearest narrative thread this week. Goldman Sachs prime book data shows that after five consecutive weeks of net selling industrials, hedge funds this week net bought the industrials sector at the fastest pace since the start of the year (+1.8 standard deviations), almost entirely driven by long buying, with leading sub-sectors including aerospace and defense, construction and engineering, airlines, and industrial conglomerates.
Clear divergence is also emerging within the technology sector. Semiconductors and tech hardware were the most sold sub-sectors, while software was the only direction seeing net buying. Goldman Sachs TMT specialist Pete Callahan noted that over the past week, software outperformed semiconductors by about 800 basis points, with Goldman Sachs' software/semiconductor hedge pair gaining 11.8% in a single week.
The trigger for this rotation can be traced. Last Thursday, the Financial Times reported that OpenAI's revenue data was disappointing, and the Nasdaq fell sharply that day. Goldman Sachs data shows that Thursday was precisely the trading day with the most concentrated net selling of tech stocks, combining long position reductions with increased short positions, "pointing to a simultaneous rise in profit-taking and hedging demand."
Mag 7 Holdings Hit Record High While Confidence Plummets to Rock Bottom
The contradiction lies here: hedge funds are massively reducing technology stock holdings, yet their exposure to the largest technology stocks has never been higher. Goldman Sachs' equities division data shows that the Mag 7's share of total U.S. equity exposure has rapidly climbed from a year-to-date low of 14% in June to 22%, setting an all-time record for the prime book.
Morgan Stanley prime brokerage data confirms this assessment from another angle: information technology currently accounts for 32% of hedge funds' North American net exposure, in the 98th percentile since 2010. In other words, selling at the sector level does not mean underweighting — no one has truly reduced their exposure to this theme.
Meanwhile, fundamental long-short funds continue to face pressure. According to Goldman Sachs Prime Services weekly estimates, during October 2 to 8, the average fundamental long-short fund fell 0.70%, while the MSCI World Index rose 0.78%, with losses "almost entirely coming from alpha-side losses on both the long and short sides." Year to date, fundamental long-short strategies have returned 11.82%, trailing the MSCI's 12.42% and the S&P 500's 14.45%. Systematic strategies rose 21.40% over the same period, with quantitative continuing to outperform human judgment.
Earnings Season Opens: High Standards, Low Rewards
Earnings season officially kicks off this week, with major banks leading the way.
Goldman Sachs' equities division directly points out the challenge: consensus expects S&P 500 earnings per share to grow 27% year-over-year, the highest since 2021, while the median stock's expected growth is only 9% (versus 14% last quarter). Behind this gap, Micron and NVIDIA alone account for about one-third of this quarter's earnings growth.
More noteworthy is the "beat but no reward" phenomenon.
Goldman Sachs data shows that last quarter, stocks that beat earnings expectations outperformed the S&P 500 by only 35 basis points, far below the historical average of 95 basis points. Micron posted 1,003% EPS growth, beating expectations across the board, yet the market reaction was quite restrained. Goldman Sachs asks: has this "high-standards dilemma" already spread across the entire AI sector?
Bank stocks face a particularly peculiar situation. Goldman Sachs' derivatives team notes that the financial sector's pullback from its high is about five times the option-implied single-day move, yet the market is pricing its earnings day extremely mildly. Morgan Stanley has fallen 20%, with options implying a move of only 3.7%; Bank of America has fallen 19%, with corresponding implied volatility of 3.4%.
Goldman Sachs financial sector specialist Christian DeGrasse notes that market sentiment toward banks has "clearly turned cautious," with the debate centering on whether the third-quarter activity slowdown is a "temporary pause" or a "signal of sustained decline."
Morgan Stanley's Nick Savone goes further: U.S. bank stocks' oversold level relative to the S&P 500 has reached the highest since 1993, with the 14-day RSI at about 27, and the BKX forward P/E has fallen below its ten-year median.
OpenAI Incident's Chain Reaction: Bond Market Unexpectedly Benefits
Last Thursday, the market volatility triggered by OpenAI's revenue data not only affected stocks but also unexpectedly drove a rebound in long-duration bond prices. Goldman Sachs' futures team notes that long bonds "may have finally bottomed this week," with the trigger for the rebound being OpenAI-related headlines.
The logic: once expectations for AI capital expenditure expansion show signs of loosening, duration buying in Treasuries that had been continuously squeezed by AI financing demand may flow back. Previously, asset managers' short positions in long bonds had risen sharply since mid-September; if Thursday's move triggers a short squeeze, the bond market may receive further support early next week — though whether this rebound can be sustained still depends on Tuesday's CPI data.
On gold, prices rose 1.3% this week, "holding above $4,100 per ounce." Goldman Sachs' futures team notes that systematic strategy positioning has fallen to the 0th percentile on a one-year lookback, and CFTC managed money net longs have returned to normal levels, arguing that "the appeal of adding gold long positions is rising."
Power: What Goldman Sachs Sees as the "Most Underestimated AI Bottleneck"
Goldman Sachs' thematic basket team calls its "Power Up America" basket (ticker GSENEPOW) the "most underestimated AI bottleneck," basing this judgment on three dimensions: positioning at a near-two-year low, valuations having returned to normal, and fundamentals remaining solid. The basket rose 4.9% this week, with the independent power producers sub-sector gaining 12.7%.
The team believes that market concerns about midterm elections and the politicization of power policy are "fading," and expects future stimulus for U.S. power supply through grid construction and "behind-the-meter" solutions. Goldman Sachs forecasts that by 2030, "behind-the-meter" power will meet 25% of U.S. data center electricity demand. Near-term key catalysts include: FERC's ruling on the PJM large load framework on Monday, October 12, as well as PJM-related documents and ERCOT data center interconnection audit results.
Comments