Market Signal

Verdict: this is a narrowness market backdrop wearing rotation's clothes. Over the past thirty days the S&P 500 rose 1.0% while the equal-weighted version of the same index fell 4.0%. Ten of eleven sectors are negative. The one that is not — Technology, up 7.1% — is carrying the entire index. That 5.00-point gap between the cap-weighted and equal-weighted tape sits at the 98.9th percentile of 5,036 sessions since 2006; only 52 days in twenty years have been narrower. There is very little rotation between sectors happening, because there is very little else working. What is rotating, quickly and measurably, is the thematic layer inside the part that works.

Verdict Evidence

---

Breadth 🔴 broken Index +1.0% over 30 days, average stock −4.0%. A 20-year extreme. Sector leadership 🟠 don't chase One sector above the benchmark. Ten below. This is not a rotation to position around. Theme leadership (AI hardware) 🟢 intact Memory, networking hardware and the semiconductor equipment layer are leading and broadening upstream. Theme durability 🟡 mixed Leadership at the one-month horizon reshuffles measurably harder than its twenty-year norm. Lane leaders decay. Everything cyclical/consumer 🔴 broken Travel, consumer, alt managers and real estate at the bottom of 122 lanes, several deeply oversold.

What Is Actually Rising

Ranked on thirty-day relative strength against the S&P 500 across 122 lanes, settled on the 2026-09-22 close:

Lane vs S&P, 30d | --- ---: | DRAM memory +13.4 | High-speed PCBs and server networking +13.2 | Cybersecurity +12.5 | Crypto +12.5 | NAND flash memory +12.4 | Data-centre switch and NIC vendors +9.7 | Memory (broad) +9.0 | Semiconductor test and burn-in equipment +8.9 | CPU shortage supply chain +8.7 | Neoclouds +8.6 | Semis (broad) +7.5 | Custom AI silicon and hyperscaler ASICs +7.5 |

Ten of the top twelve are the AI hardware stack. Three are memory. The two exceptions — cybersecurity and crypto — are the only leadership outside it.

The more interesting reading is one layer down, among lanes that are accelerating without yet ranking as leaders. Every one of the five strongest is semiconductor capital equipment: advanced packaging equipment (acceleration +13.2), wafer-fab test and metrology (+13.1), optical supply chain (+13.1), lithography equipment (+12.0), and package substrates and interposers (+11.5). Immediately behind them, turning up from below: etch (+12.9), deposition (+12.4), CMP slurries and consumables (+11.4), semiconductor bonding and interconnect materials (+9.7), and AI power (+8.9).

That is a coherent move, not a scatter. Money that was in the compute names has been stepping upstream into the tools that build the capacity — the picks-and-shovels of the picks-and-shovels. Two standing desk positions sit directly on it: the memory case, whose lane now ranks first of 122, and the optical case, whose lane is the third-strongest accelerator on the board.

What is falling is everything attached to the real economy: travel and leisure (−14.4 against the benchmark), alternative asset managers (−12.9), livestock disease (−12.5), consumer (−12.3), drone defence (−10.8) and lithium battery metals (−9.4). At the sector level the same story reads as financials at an RSI of 29.9, utilities at 27.7 and real estate at 31.4 — not rotation out of them so much as sustained bleed.

Sector 1D 30D vs S&P 30d RSI --- ---: ---: ---: ---: Technology +0.73% +7.07% +6.07 66.9 Communication services −1.06% +1.91% +0.91 53.7 Health care +0.52% −2.71% −3.71 54.4 Energy −1.09% −2.92% −3.92 42.7 Consumer staples +0.99% −3.79% −4.79 41.2 Financials −1.97% −4.66% −5.66 29.9 Consumer discretionary +0.09% −4.82% −5.82 43.6 Utilities −0.32% −5.24% −6.24 27.7 Industrials +0.17% −5.54% −6.54 36.5 Materials +1.65% −5.62% −6.62 42.3 Real estate −0.21% −5.72% −6.72 31.4

Is Leadership Turning Over Faster Than It Used To

The question deserves a measurement rather than an impression. Rank nine sector funds by trailing return on every trading day from 2006-08-17 through today, then compare that ranking against the one built on the immediately preceding window of equal length. A rank correlation of +1 means leadership carried over intact; 0 means the new ranking is unrelated to the old; negative means leadership actively reversed. Run the identical test on nine thematic funds over the identical window with the identical number of buckets — because both the average rank shift and the top-three holdover depend mechanically on bucket count, and an eleven-bucket panel churns more than a nine-bucket one with no change in the market at all.

The first version of this answer was wrong, and how it was wrong is worth stating. Measured on overlapping daily windows, the thematic panel produced the most reversal-prone year in the entire series at a one-month horizon. That result did not survive a change of window: at one quarter it reversed sign. Consecutive daily observations share every day of data but one, so a single year presents roughly 250 samples while containing about 250/window independent ones — four, at a sixty-three-day window. A metric that answers differently at twenty-one days than at sixty-three is describing the window, not the market.

Re-run on non-overlapping samples, each year's mean scored against the spread of the other nineteen:

Panel Horizon 2026 Other years z Reading | --- --- ---: ---: ---: --- | Themes, rank correlation 1 month −0.26 −0.02 −2.52 outside the normal range | Themes, dispersion 1 month 24.0% 15.6% +2.00 outside the normal range | Sectors, dispersion 1 quarter 28.6% 18.4% +1.79 elevated, not decisive | Sectors, rank correlation 1 quarter −0.36 +0.02 −1.84 elevated, not decisive | Sectors, dispersion 1 month 15.8% 10.6% +1.56 elevated, not decisive | Sectors, rank correlation 1 month +0.13 −0.03 +1.04 normal; if anything stickier | Themes, dispersion 1 quarter 37.8% 27.6% +1.42 normal | Themes, rank correlation 1 quarter +0.07 −0.01 +0.34 normal |

Dispersion is the gap between the best and worst performer in the panel.

One cell breaks out on two independent measures at once: thematic leadership at a one-month horizon, which both reshuffles harder than its twenty-year norm and does so across wider gaps. Nothing else clears the bar. Sector leadership proper is not turning over faster — month to month it is marginally more persistent than usual, which is the opposite of the popular impression.

So the felt experience is real and mislocated. A reader watching 122 thematic lanes is watching precisely the layer where the churn is measurable, and the standing rotation board excludes the sector funds entirely by design, so the calmer layer never appears next to the noisy one.

What It Means For A Position

Sector-level trend is as durable as it has been, which is the horizon a multi-month hold actually rides. The unreliable layer is the thematic one, and it is the layer that generates the daily narrative. A month-old lane leader is a weaker basis for an entry this year than its rank suggests — that is what a rank correlation of −0.26 means in practice.

The narrowness cuts the other way and matters more. An index carried by one sector is an index with a single point of failure. The breadth gap closing is not automatically bullish either: it resolves by the other ten sectors catching up, or by Technology coming back to them, and nothing in this measurement says which. What it does say is that positioning "rotation" trades between sectors right now is trading a phenomenon that is barely present.

Watch the breadth gap. The 21-day spread between the cap-weighted and equal-weighted index closing back below +1.0 point is the single cleanest sign this market backdrop is ending, and it is the first line of the invalidation list below.

Method and caveats

2026 is a partial year of roughly eight and a half months. Eight tests were run, and a z of about two is not decisive under that many comparisons; the thematic one-month reading is credited to its agreement across two independent measures rather than to either alone. Returns are price returns, not total returns, so higher-yielding defensive funds are understated by roughly two tenths of a point over a one-month window — that shifts the level of a rank slightly and consistently without manufacturing churn. The nine-sector panel runs from 2006-08-17, the first date all nine funds share; the eleven-sector panel would start in 2018 and was used only to size the bucket-count effect, which is a rise in average rank shift from 2.91 to 3.61 with no change in the underlying market.