On two sessions between July 1 and September 2, 2026, the most expensive slice of the secured overnight funding market traded at or above the ceiling of the Federal Reserve's policy corridor. On July 31 the 99th volume-weighted percentile of the Secured Overnight Financing Rate printed at 3.75 percent. On August 31 it printed at 3.77 percent. Those were the only two sessions in a 45-business-day window where that figure reached 3.75, and both of them fell on a month-end. The headline rate did nothing comparable. SOFR itself moved between 3.53 and 3.68 across the same window and stayed inside the target range on all 45 days. A reader watching only the published median would record two quiet months. The distribution underneath that median was not quiet, and the dates on which it stopped being quiet were calendar dates rather than economic events. The question this piece takes up is narrow. When the middle of a distribution sits still and its upper tail does not, which of the two ...
The standard explanation for why purchasing managers' index readings move currency and rates markets is that they arrive first. A survey published days after a month closes beats a national accounts release that takes weeks to assemble, so traders reprice on the survey and wait for the statistics office to confirm. That explanation is not wrong, but it treats PMI as a single global instrument with a uniform lead time and a stable translation into output. The releases of July and August 2026 do not support that reading.
In the same month, two established surveys of American manufacturing landed 1.7 points apart. Two surveys of Chinese manufacturing disagreed about the direction of activity itself, not its speed. In the euro area, the statistical office published a first GDP estimate close enough behind the survey that the survey's timing advantage was measured in weeks rather than months. And in Japan, a composite index that had been above 50 for sixteen consecutive months sat alongside a quarter of 0.3% growth. The lead a PMI buys, and what that lead is worth, is a jurisdiction-by-jurisdiction question.
A Diffusion Index Counts Firms, Not Output
The distinction most commentary elides is that a PMI is a diffusion index. It records the share of responding firms reporting an increase against the share reporting a decrease, with a half weight for no change. S&P Global states the convention in every release: the indices vary between 0 and 100, "with a reading above 50 indicating an overall increase compared to the previous month, and below 50 an overall decrease."
Nothing in that construction measures how large the increase was. A month in which every surveyed firm grows output by a tenth of a percent and a month in which every firm grows by five percent both produce a reading near 100. Breadth and magnitude are different quantities, and GDP measures the second one.
The composite construction differs between providers as well. The S&P Global manufacturing PMI uses fixed weights: New Orders at 30%, Output at 25%, Employment at 20%, Suppliers' Delivery Times at 15% and Stocks of Purchases at 10%. The delivery times component is inverted before entering the headline, so lengthening lead times push the index up. ISM applies equal weighting across its five components instead.
That inverted term deserves particular attention in 2026. The S&P Global US release for July described "one of the sharpest deteriorations in vendor performance over the past four years," and the ISM report put its Prices Index at 71.1%. Supply congestion and input cost pressure are, mechanically, headline-supportive in a diffusion framework, even when the underlying condition is a constraint on output rather than a sign of demand.
The GDP translation is re-estimated, and it moves
ISM publishes its own mapping from index level to output, which makes the instability of that mapping visible. The July 2026 report states that "a Manufacturing PMI above 47.5 percent, over a period of time, generally indicates an expansion of the overall economy." The equivalent sentence in the July 2025 report used 42.3 percent. The neutral level of the same index, against the same national accounts, was restated by more than five points inside a year. Any framework treating a fixed PMI level as a fixed growth rate is using a coefficient with a shelf life.
United States: Two Surveys, One Month, Two Sizes of Signal
On 3 August 2026, both American manufacturing surveys for July were published. ISM reported 55.6%, up from 53.3% in June, with New Orders at 56.7%, Production at 58.5% and Employment at 52.8%. S&P Global reported 53.9, unchanged from June, with production volumes rising at the weakest pace since March and new order growth easing for a third successive month.
Both signal expansion, but not the same expansion. ISM's own translation put 55.6% as corresponding to "a 2.8-percent increase in real gross domestic product on an annualized basis." The BEA advance estimate for the second quarter, released on 30 July 2026, showed real GDP increasing at an annual rate of 1.5%. The periods are not identical — the survey covers July, the accounts cover April through June — but a gap of that size between a survey's implied rate and the most recent measured rate is not a rounding difference.
Panel construction explains part of the spread. The S&P Global US panel is around 600 manufacturers, stratified by sector and workforce size against GDP contribution, with responses collected 9 to 28 July for that release. ISM stratifies by NAICS classification, weights each industry by its GDP contribution, and collects through month end.
Euro Area: The Statistics Office Nearly Caught Up
Europe is the case where the survey's timing advantage has been narrowed deliberately. The HCOB flash eurozone composite for July 2026 was published on 24 July at 51.9, against 50.0 in June and 48.5 in May. The final reading came in at 52.0 — a revision of one tenth. That precision is consistent with the flash methodology, which S&P Global describes as calculated from "around 80-90% of total responses." The eurozone survey draws on roughly 5,000 private sector companies, with manufacturers in eight nations and service providers in five.
Six days later, Eurostat published its preliminary flash GDP estimate for the second quarter, on 30 July 2026, showing euro area GDP up 0.4% quarter on quarter. The 14 August release confirmed 0.4% for the euro area and 0.5% for the EU. Eurostat's own documentation of the t+30 programme reports that across sixteen test estimates, the expected typical revision at t+45 for growth rates published at t+30 was "-0.1, 0.0 or +0.1 percentage points" for both aggregates.
That is a demanding benchmark to beat. The flash PMI leads the preliminary flash GDP estimate by under a week for the quarter just ended, and the official number arrives close to its final value. What the survey still adds in Europe is dispersion. The August release put Germany at 0.2%, France at 0.2%, Italy at 0.2%, Spain at 0.7% and the Netherlands at 0.4%, with Ireland at 3.9% flagged as frontier series data requiring careful interpretation. National PMI splits read that dispersion early, which is a narrower claim than "PMI leads GDP."
China: Two Panels Disagreed About Direction
China produced the sharpest illustration of panel composition mattering more than timing. The National Bureau of Statistics reported a July 2026 manufacturing PMI of 49.2%, down 1.1 points from 50.3% in June, with the non-manufacturing index at 49.0% and the composite output index at 49.3%. The size breakdown showed large enterprises at 49.5%, medium at 49.7% and small at 47.4%. The NBS survey covers 3,200 samples across 31 manufacturing divisions and 4,300 samples across 43 non-manufacturing divisions.
The private-sector series, published as the RatingDog China PMI since Caixin concluded its title sponsorship of the S&P Global-compiled index from July 2025, put July 2026 manufacturing at 50.9, down from 51.7 in June. One survey said contraction, the other expansion, over the same month.
The compositional explanation is documented rather than inferred: the official index has greater representation of larger companies and state-owned enterprises, while the S&P Global-compiled panel of around 650 manufacturers skews toward smaller, privately owned firms. When policy support and demand conditions hit those populations differently, the indices separate.
China is also the economy where a survey lead is worth least. The statistics bureau published first-half GDP on 15 July 2026, reporting 4.7% growth year on year for the half and 4.3% for the second quarter alone — roughly fifteen days after the quarter closed. A monthly survey published on the last day of the quarter leads the official number by about two weeks, not two months.
Japan and the United Kingdom: Long Expansions, Thin Growth
Japan is the clearest case of breadth diverging from magnitude. The au Jibun Bank composite PMI stood at 52.7 in July 2026 against 52.8 in June, a sixteenth consecutive month above the no-change mark. Cabinet Office first preliminary GDP for the second quarter, published on 17 August 2026, showed the economy growing 0.3% quarter on quarter, or 1.1% annualised — below the 0.5% quarterly consensus, with private consumption turning negative for the first time in eight quarters.
Sixteen months of majority-positive responses and a 1.1% annualised print are not contradictory. They are readings of different things. A diffusion index near 52.7 says slightly more firms grew than shrank. It does not say by how much, and where the marginal firm's growth increment is small, the same index level maps to far less output than it would elsewhere.
The United Kingdom sits between Japan and the euro area on timing. The ONS first quarterly estimate for January to March 2026 was published on 14 May 2026, showing GDP up 0.6% — around 44 days after the quarter closed. That leaves a wider survey-only window than the euro area's t+30, but the UK also publishes monthly GDP estimates, so the gap is populated by official data rather than survey data alone.
Ranking the Lead a PMI Actually Buys
| Economy | First official GDP estimate, most recent quarter | Approximate lag | What the survey lead is worth |
|---|---|---|---|
| China | 15 July 2026, Q2 at 4.3% year on year | t+15 | Narrowest. Roughly two weeks, and the two available surveys disagreed on direction |
| Euro area | 30 July 2026 preliminary flash, 0.4% quarter on quarter | t+30 | Narrow on the aggregate. Higher value on national dispersion ahead of the country split |
| United States | 30 July 2026 advance estimate, 1.5% annualised | t+30 | Moderate, but split across two surveys 1.7 points apart in July |
| United Kingdom | 14 May 2026 first estimate for Q1, 0.6% | t+44 | Wider window, partly filled by official monthly GDP |
| Japan | 17 August 2026 first preliminary, 0.3% quarter on quarter | t+48 | Widest window, loosest translation from index level to output |
What the Data Cannot Settle
Several parts of this frame do not hold. The honest version says which.
The direction of error is not established. A gap between an ISM-implied 2.8% and a measured 1.5% could be the survey overstating momentum, or the advance estimate understating a quarter that gets revised up. BEA advance estimates rest on incomplete source data, with components of the quarter's final month carried as agency projections. Treating the accounts as ground truth and the survey as noise is an assumption, not a finding.
Level comparisons across providers are not meaningful. ISM at 55.6% and S&P Global at 53.9 are not 1.7 points of disagreement in any interpretable sense, because the weighting schemes differ and the collection windows differ. Changes within a series are comparable. Levels across series are not.
Official GDP has its own distortions. The Irish figure of 3.9% quarter on quarter in the Q2 2026 European release is flagged by Eurostat as frontier series data requiring careful interpretation, a standing feature of an economy where multinational balance sheet activity dominates the accounts. Where GDP itself is a poor proxy for domestic activity, a survey that diverges from it is not necessarily the one that is wrong.
The arguments here are structural, not predictive. Nothing above establishes that the July 2026 divergences persist into the third quarter, and a single month of separation between two panels is a weak base for any inference about the next print. The compositional differences between the Chinese surveys are documented; the size of the gap they produce in any month is not stable.
What to Watch Next Week
- Flash composite releases around the 22nd to 24th. The relevant number is not the headline level but the gap between the flash and the eventual final. A revision inside 0.2 points, as in the eurozone July flash at 51.9 against a 52.0 final, indicates the 80-90% response base is behaving normally. A wider revision suggests late responses are carrying different information from early ones.
- Suppliers' delivery times, read separately from the headline. Because the component is inverted, a further deterioration in vendor performance lifts the manufacturing PMI while signalling constraint. Separating that contribution from new orders is the difference between reading demand and reading congestion.
- The NBS-versus-RatingDog spread for August. A single month of opposite-sign readings is noise. A second consecutive month, particularly with the official small-enterprise sub-index still below 48, would point at a genuine split between the state-linked and private manufacturing bases rather than a sampling artefact.
- Whether euro area national PMI splits track the Q2 country dispersion. Spain at 0.7% against Germany, France and Italy at 0.2% is a wide spread for one quarter. If national surveys do not reproduce that ordering, the case for using PMI as a dispersion tool weakens.
- Any restatement of the ISM neutral threshold. The move from 42.3% to 47.5% between the July 2025 and July 2026 reports is the kind of methodological update that quietly invalidates rules of thumb built on the older number.
Concrete Framework — How to Verify It Yourself
A checklist that treats PMI as one input rather than an early GDP number.
- Record the provider with every reading. "US manufacturing PMI, July 2026" is ambiguous between 53.9 and 55.6. Log the series name, not the country and month alone.
- Establish the local lead time once, then reuse it. China at roughly t+15, the euro area and United States at t+30, the United Kingdom at t+44, Japan at t+48. The survey-only window is the gap between survey date and that lag — the only period in which the survey is the sole read.
- Track changes within a series, never levels across series. The June-to-July move of +2.3 points at ISM and 0.0 points at S&P Global is the comparable pair. The 1.7-point level gap is not.
- Decompose before reacting. New orders and output carry demand information. Suppliers' delivery times, inverted into the headline, carries constraint information that moves the index the same way. In a month with a prices index at 71.1%, that separation matters more than usual.
- Hold a second survey against the first where one exists. China has two. The United States has two. Agreement on direction raises confidence; disagreement identifies which segment of the economy is driving each panel.
- Set a revision tolerance in advance. For euro area GDP, Eurostat's own test work puts the typical t+30 to t+45 revision at -0.1 to +0.1 percentage points. A position premised on a larger revision is premised on an outlier.
- Write down what would falsify the read. If a PMI is being used to argue for above-consensus growth, the falsifying observation is a specific GDP print at a specific date, not a general sense that the survey looked strong.
The survey's advantage is real, and in Japan or the United Kingdom it is measured in weeks. The error is assuming that advantage is the same everywhere, that the two surveys of an economy will agree, or that a count of firms reporting improvement translates into a rate of output growth by a coefficient that stays put. In 2026 none of those three held reliably.
This piece is macroeconomic analysis, not investment or financial advice. Figures cited are drawn from official statistical releases and published survey documentation as of mid-August 2026 and are subject to revision.
Comments
Post a Comment