Why Is Only Tech Going Up? What 13 Similar Rallies Actually Predicted

Research note. Computed from daily ETF closes in our maintained TradeStation price database. XLE−XLK and XLV−XLK begin in 1999; RSP−QQQ begins in 2003. The historical test is frozen through August 7, 2026; the latest reading and charts continue to update. See How we checked it.

It wasn’t only tech. Tech was the accelerator.

The question contains a useful feeling and a false premise. SPY rose +3.5% over the five sessions and finished at a record. RSP, the equal-weight S&P 500 ETF, gained +2.4%. Materials, communication services, financials, industrials, consumer staples, health care, and consumer discretionary all rose too. Only energy, utilities, and real estate finished the window lower.

That matters because relative weakness is not the same as absolute deterioration. The average S&P stock was participating; it simply was not keeping pace with the Nasdaq’s leaders. That is consistent with the equal-weight confirmation in our recent breadth work and with the live Hidden Bear Index. The honest description is a broad advance powered disproportionately by tech. It is too early to call it market-wide abandonment.

The fresh-high list makes the same point. Eaton, Parker Hannifin, Snowflake, CrowdStrike, and Airbnb all reached 252-session highs, a mix of industrial, software, cybersecurity, and consumer names. AI spending may explain part of the leadership, but the list does not support calling every winner one AI-capex trade. Earnings and company-specific news were also moving individual stocks.

Extreme rotation, not market-wide selling

The simplest gauge is each ETF’s five-session price return minus tech’s return. On August 7, energy’s spread was −10.6 percentage points. Only 1.1% of the 6,908 rolling five-session windows since 1999 were lower. The August 5 reading reached −13.9 points before energy recovered some ground into Friday.

This is not a sample of 6,908 independent weeks. The windows overlap, so four of every five observations share most of their prices. Percentiles tell us how unusual the reading was; episode clustering is what prevents the forward test from pretending consecutive signals are separate events. Health care versus tech finished in its bottom 3.1%, and RSP versus QQQ in its bottom 4.5%. Three of the eight structural pairs watched by the scanner were simultaneously in their bottom twentieth; four pairs were flagged overall.

The first backtest gave a bearish answer for the wrong reason

A naïve test looks ominous. Cluster every bottom-2% energy-versus-tech reading into episodes, with no regard for the surrounding market, and the sample contains 41 episodes,40 with a complete next quarter. Their median SPY return was -1.9%, and only 18 were positive.

But that sample mixes unlike setups. It includes late-stage dot-com rotations, the 2008 crisis, the post-pandemic market, and ordinary rallies near records. Energy can lose badly to tech because tech is surging, because energy is collapsing, or because both are falling at different speeds. Pooling those regimes answers a different question from the one investors faced on August 7, when SPY had risen 3.5% in five sessions and closed at a high.

Match the setup, and the bearish signal disappears

I kept the same bottom-2% threshold and 21-session clustering rule, then added two facts that define the current tape: SPY had to be up over the same five sessions and within 5% of its trailing 252-session closing high on the episode’s deepest day. That leaves 13comparable episodes, 12 old enough to grade.

SPY’s median next-quarter return was +2.7%, with 8 of 12 positive. The matched baseline, every rising-market day within 5% of a trailing high, was +2.8%with 69% positive. Those are effectively the same result. The rotation was rare; the index outcome was ordinary.

The more repeatable result sits inside the pair. Energy beat tech over the next 63 sessions in 8 of 12 completed episodes, by a median of +4.2pp. Even that is not automatic. After the April 14, 2026 episode, SPY gained 8.7% and tech kept beating energy by another 21.8 points. The current episode, anchored on August 5, remains open.

What the signal actually says

The evidence supports a narrower conclusion than the headline intuition. A violent five-day leadership gap has often been followed by some reversal between the two sectors. It has not, in comparable near-high rallies, reliably predicted a weak quarter for SPY. The practical read is “expect the leadership gap to become harder to extend,” not “sell the market.”

The funding side still deserves attention. WEC Energy and W.P. Carey had each fallen ten consecutive sessions by August 7; CMS Energy had fallen eight. Those streaks show persistent pressure in rate-sensitive defensives while the index was making records. They do not prove that the entire market was deteriorating. For that, equal-weight participation would need to break in absolute terms, not merely trail QQQ for one explosive week.

What would change my mind

I am treating this as an exhausted rotation; the evidence does not support a bearish index signal. Three observable outcomes would challenge that view. First, if XLE has recovered none of its relative loss against XLK by September 8, the historical mean-reversion tendency is failing early. Second, if RSP closes more than 5% below its August 7 record while SPY remains within 2% of its own high, relative concentration has become genuine breadth erosion; the Hidden Bear Index will show it. Third, a new bottom-2% XLE−XLK episode after at least 21 quiet sessions would make this a sequence rather than a one-week overshoot. All three can be monitored on the sector board and our breadth tools.

How we checked it

Each spread is ETF A’s five-session price return minus ETF B’s, calculated on shared trading dates. Blank or nonnumeric closes are discarded before calculation. Percentiles rank a reading against every overlapping five-session window available through August 7, 2026. Because adjacent windows overlap, the study merges bottom-2% readings separated by 21 sessions or fewer and anchors each episode on its deepest day.

The comparison then retains only anchors where SPY was up over five sessions and no more than 5% below its trailing 252-session closing high. Forward returns cover 63 trading sessions and are graded only when the complete window exists. The baseline applies those same SPY context rules to every eligible day since the XLE−XLK history begins. All figures are price returns, not total returns. The episode set and narrative are frozen through August 7, 2026; only explicitly labeled latest-reading exhibits update.

Frequently asked questions