Is the economy still K-shaped?

The claim, verbatim: “I can say here definitively, the K-shaped economy is over” — Treasury Secretary Scott Bessent, CNBC, August 4, 2026 (PBS NewsHour, Aug 12, which defines the term as “high-income and low-income households moving in different directions for several metrics”). Two of those metrics resolve in public monthly and quarterly data, and they are different instruments: a flow — whose wages are growing faster — and a stock — who holds the wealth. This run tests both and pools neither: the wage series is the Atlanta Fed Wage Growth Tracker’s 12-month moving average of median hourly wage growth for the top quartile (76th–100th wage percentile) minus the bottom quartile (1st–25th), monthly since 1997; the wealth series is the Fed’s Distributional Financial Accounts shares of net worth, quarterly since 1989. Both pulled fresh 2026-09-12, keyless, from FRED. Positive gap = top growing faster = the K direction.

The era the claim inherits was running the K backwards. From 2019-01 to 2024-12 the gap’s mean is -1.49 points — bottom-quartile wages growing a point and a half per year faster than top-quartile — 95% CI [-2.19, -0.79] under NW-HAC(13), classical [-1.72, -1.25], p=7.0×10⁻⁵, n=72 months. The extreme is in the table: July 2022, −3.7 — bottom-quartile median hourly growth of 7.6% against the top’s 3.9%, the deepest reading in all 355 months of the series — the tight post-pandemic labor market compressing wages from below. On the flow metric, the “K-shaped economy” of 2021–24 commentary was already inverted before anyone declared it over.

Then 2025, and the step is real by every method this desk owns. The gap crossed zero in October 2024, averaged +0.92 across 2025 (peak +1.2 in June and August; top quartile 4.8% against 3.6%), and the era contrast — 2025–26 mean minus 2019–24 mean — is +2.11 points, NW-HAC(13) CI [1.39, 2.83], classical [1.65, 2.57], and 100.0% of 4,000 stratified year-block bootstrap draws positive (CI [1.27, 2.94]). The formal interaction test at 2025-01 puts the level jump at +2.19 points, HAC CI [0.73, 3.65]; the step model net of trend says +1.62 [0.13, 3.11] and still excludes zero with the split moved three months in either direction; and collapsing to the honest annual unit — each calendar year one observation, 2025–26 (mean +0.57) vs 2019–24 (mean −1.49) — the Welch interval is [+0.58, +3.53], p=0.019, excluding zero at the power floor. Four ways, one answer: the level of the wage-growth gap stepped up by about two points at 2025.

What is not certified: the trend. The interaction’s slope-change coefficient is −0.79 pts/yr, and the three error models disagree about it — NW-HAC(13) excludes zero ([−1.09, −0.50], p=1.4×10⁻⁷); classical OLS contains it ([−1.61, +0.02], p=0.055); the year-block bootstrap contains it ([−1.32, +0.17], 5.6% of draws positive). Reported as an unresolved disagreement, not rounded to the HAC answer — on a 19-month hump-shaped post era the HAC long-run variance is the fragile one. The certified fact is a level step, not a new slope; the pre-2025 slope is itself indistinguishable from zero (+0.14 [−0.14, +0.43]) because the pre era is U-shaped, and a straight line was never the right tool for it.

And in 29 years of context, the 2025 ‘K’ was ordinary. The same annual-unit Welch test run against all 28 prior years instead of just 2019–24: 2025–26 (+0.57) minus 1997–2024 (−0.57) is +1.14, CI [−0.88, +3.16], p=0.12 — contains zero. The gap’s all-time high is +1.4 (February 2012); 2011–2013 averaged +0.99; 34.1% of all months since 1997 are positive. The two-point step is certified relative to 2019–24 — and 2019–24, not 2025, is the outlier: three extraordinary reverse-K years, followed by a reversion to the series’ ordinary small positive tilt.

The closing is a fact about levels, and only that. From the +1.2 peak to August 2026: +0.1 — top quartile 3.8% against 3.7%, one-tenth of a point from equal growth, in the very month Bessent declared the K over. The 2026 year-to-date mean is +0.23 (nominal CI [0.11, 0.35] — but 8 months of a 12-month moving average carry less than one independent year of information, and the desk does not certify intervals that thin in either direction). The within-2026 slope is −0.54 pts/yr on classical errors only ([−0.86, −0.23], p=0.006); no honest HAC exists at n=8. So on the claim’s own metric — growth rates — the August 2026 reading is equal-growth within a tenth of a point, and the desk can say that much. Whether the divergence has closed or is still closing is not distinguishable on the data that exist; and converging rates, wherever they settle, stop a gap widening without returning any of the ground already lost or gained. The rate series says nothing about levels, by construction.

The wealth side never turned at all. Over the identical window, 2019Q1 to 2026Q1 (the DFA’s latest published quarter): the top 1% share of net worth went 30.6% → 31.6% — up 1.0 point, slope +0.147 pts/yr, 95% CI [+0.015, +0.279], excluding zero. The bottom 50% went 1.6% → 2.5% (+0.9, CI [+0.055, +0.190], excluding zero — a real rise off a base of 1.6, landing 2020–21 and holding; still a share 12.6× smaller than the top 1%’s). Shares sum to one total, so someone lost: the omitted 91st–99th percentile band, down 3.3 points (39.6 → 36.3) — the window’s redistribution ran from the upper-middle to both ends, not from the top to the bottom. The long arc is unchanged: 22.8% in 1989Q3 → 31.6% in 2026Q1, +8.8 points, +0.210 pts/yr, CI [+0.174, +0.245], p=6×10⁻³¹ across 147 quarters — a 37-year climb that first lifted the top 1% past the entire next-40% band in 2014Q1. “Over” as an economy-wide claim would need this line to bend the other way. It did not; if it moved at all, it moved up.

What survives contact with the table. Four statements are certified: (1) 2019–24, bottom-quartile wage growth beat top-quartile by 1.49 pts/yr on average — the reverse of K; (2) the gap stepped up about 2.1 points at 2025 — four methods agree; (3) it reads +0.1 as of August 2026 — equal growth within a tenth of a point, though the closing itself cannot be certified as a trend on 8 months of a moving average; (4) the top 1%’s wealth share is 1.0 point higher than 2019 and 8.8 points higher than 1989 — no reversal, anywhere, in the stock the “K” was originally drawn around. The Secretary’s sentence is defensible on the flow, on the narrow reading his argument invokes, for the single latest month. As a claim about the economy’s shape — rates converged, wealth structure untouched — it is half a claim. The K he declared over was, on this desk’s arithmetic, mostly an upside-down K for three years, briefly a right-side-up one for a year at a magnitude last seen in 2012, and — where it was always most visibly a K, in the wealth shares — still exactly as K-shaped as the quarter before he said it.

The quotation was longer than the sentence this run tested. PBS NewsHour’s Aug. 12 write-up carries the fuller passage from the same Aug. 4 CNBC interview. “I got sick of hearing about this K-shaped economy,” Bessent said, and then: “I can say here definitively, the K-shaped economy is over.” On what replaced it: “We’re seeing more of a ‘C economy,’” he said. “Where the lower end of wage earners are finally calling it back, just like they did in President Trump’s first term.” (The desk also holds an automated-speech-recognition receipt of the CNBC audio, “Bessent: A Deal with Iran,” timestamped 00:36:53–00:37:07, reading the same words — a receipt, not a citation; every word quoted here is PBS’s published text.) The fuller sentence upgrades the claim’s specificity. “Over” could mean the two arms converged; “C” means the lower arm now rises faster. That is a reversal claim, and it is checkable against two series: this run’s own, and the one the Secretary cited.

On this run’s series, the reversal count is 0 of 18. Since the 2025 turn — 2025-01 through 2026-08, the 18 published months of the fitted post era, October 2025 still a hole — not one month shows bottom-quartile median wage growth above top-quartile. The gap’s floor for the entire post era is +0.1 (July and August 2026); its last negative month was September 2024 (−0.3), before the zero-crossing certified above. On matched-worker wage growth, “the lower end of wage earners calling it back” is, as of the latest published month, a tenth of a point from a tie, on the top’s side.

The 5.5 percent is real, and it is a different instrument. The same PBS piece attributes the number to the Secretary directly: “Citing the Bureau of Labor Statistics, Bessent said data shows ‘the bottom quartile of wage earners have had, year over year, 5.5% wage gains. And that’s three times more than the top quartile.’” Before adopting or dismissing a figure that does not match this run’s series, the desk went looking for the series that produces it. It exists, to the decimal: usual weekly nominal earnings of full-time wage and salary workers (16+), first quartile, $806 in 2025Q2 → $850 in 2026Q2 = +5.46% — the 5.5. Third quartile over the identical two quarters: $1,887 → $1,915 = +1.48%, so “three times more” is, at 3.7×, an understatement of his own series. PBS’s text prints no top-quartile figure at all, only the “three times” ratio; the series itself puts the top quartile at +1.5. Nothing was mis-paired: the number is the BLS quartile-of-levels cut, quoted correctly.

Both instruments are built on the same Current Population Survey; they are not the same measurement. The BLS quartile compares the 25th and 75th percentiles of the earnings-level distribution — everyone employed this quarter against everyone employed a year ago, whoever they turn out to be. The Atlanta Fed tracker follows individual workers matched twelve months apart and reports each bin’s median growth. The first moves when the workforce’s composition moves; the second is built not to. Limit 3 above anticipated exactly this divergence. And the BLS cut’s own history is why one quarter of it cannot carry a shape-of-the-economy claim: first-quartile growth exceeded third-quartile in 45 of 101 quarters since 2001 — 45 percent, near a coin flip — with 2026Q2’s +3.98-point gap the second-widest of those 101, behind 2021Q4 (+4.04) and just ahead of 2017Q3 (+3.89), both squarely inside the years the K-shaped framing describes. On this cut the economy read “C” in 2017 and 2021, and in all three published quarters of 2025 — the very year the matched-worker series certifies as the top-ahead year. One quarter earlier, the same BLS series printed +2.95 against +0.84; quarterly CPS cross-sections move like that.

Corroboration, none of it this desk’s. The same PBS piece carries three independent versions of this page’s finding. Peter Orszag, CEO of Lazard, on CNBC: “I think that declaring the death of a K-shaped economy is a little bit premature.” Deon Strickland, financial services professor at Wake Forest, to PBS: “I would say the probability is higher for the continuation of a K than the ascent of a C.” Mark Mathews, chief economist for the National Retail Federation, writing in July: “The K-shape persists, but lower-income consumers have increased their spending versus last year.”

What changes, and what does not. The verdict above stands as published. The update adds three facts to the record. The claim is sharper than the sentence originally tested — a reversal, and on matched-worker growth the reversal count is 0 of 18 months, minimum gap +0.1. The Secretary’s 5.5 percent is a real, correctly cited reading of the BLS quartile-of-levels cut — an instrument whose own history has read “C” nearly half the time since 2001, including the K’s loudest years. And nothing in the new information touches the wealth half, which was never in dispute: the top 1% share of net worth closed 2026Q1 at 31.6 percent, one point above 2019. If the economy’s shape changed, the stock has not registered it.