What a diffusion index is, and what happens when you count every negative month

“Below fifty means contraction” is the whole of what most readers know about purchasing managers’ indices, and it can be tested. The best-known American PMI could not be read at all on the day of writing — its report page serves a CAPTCHA form and nothing else — so this page uses the three Federal Reserve business surveys that are free, published back to 1968, 2001 and 2004, and counts every month each spent below its own contraction line against the recessions the NBER actually dated. The answer is not close.

The source that would not open, and what that means for the page

The obvious source for this subject is the American national manufacturing PMI. On 13 August 2026 its report page returned HTTP 200 and a body of 922 bytes consisting of a Google reCAPTCHA form and the script that submits it. There is no index value in that response, no methodology and no threshold. A second address for the same report returned 404.

This is the most dangerous shape a source can take, and it is worth naming precisely: a successful HTTP status with none of the content. Nothing about a 200 tells a script that it has failed. A page built by reading a status code rather than the body would silently publish an empty measurement.

The response taken here is the one this site takes everywhere: no figure from that publisher appears anywhere on this page— not an index level, not a threshold, not a description of how it is built — and the gap is stated instead of being filled from a site that reprints it.

What a diffusion index is, in the publisher’s own words

The Federal Reserve Bank of Philadelphia has run a manufacturing survey since May 1968 and describes it plainly. It is a monthly qualitative survey of manufacturing firms in the Third District that captures information about current business activity and firms’ expectations of future activity, and it asks participants to indicate the direction of change in activity and other measures at their firms over the past month as well as their expected direction of change six months from now. Its published data include diffusion indexes and components (percentage of firms reporting increases, no change, decreases), both seasonally adjusted and nonseasonally adjusted, and the series begin in May 1968.

Three consequences follow from that description and they apply to every index of this family, whatever its name or its centre line.

• It counts firms, not output. A firm reporting a 0.1% rise counts the same as one reporting a 40% rise. An index can improve while total production falls, if more firms improve slightly than deteriorate severely.
• It measures direction, not level. A reading above the centre line says more firms saw an increase than a decrease last month. It says nothing about whether activity is high or low, only about which way it moved.
• It is an opinion about the recent past. Respondents report the direction of change at their own firm, and the survey separately asks what they expect six months out. The two are different series and are routinely quoted as though they were one.

These three indices are centred on zero rather than fifty, because they are computed as the percentage reporting increases minus the percentage reporting decreases. Zero here means what fifty means elsewhere: as many firms up as down. The arithmetic is the same and only the offset differs.

All three reached their minimum in the same month, April 2020, at −58.7, −80.0 and −73.6. The Dallas figure is worth pausing on: its all-time highof +47.6 is the very first observation in the series, June 2004, so the entire published history of that index is a descent from its own starting point — which is a reminder that a series beginning at a peak has no history of anything higher.

Counting every negative month against the recession dates

Now the test. If a reading below the contraction line indicated a contraction of the economy, then months below the line should coincide with months the NBER dates as recession. Here is that comparison run over the complete history of each series, not a selection.

Between 63% and 84% of all the below-the-line months in these three surveys fell in months the NBER does not date as a recession at all. The Philadelphia survey, which has the longest history and covers eight recessions, spent 121 monthsbelow zero outside any dated contraction — more than ten years’ worth.

Two honest qualifications belong beside that result and neither rescues the popular claim. First, these are manufacturingsurveys, and manufacturing can contract while the economy grows; a negative reading may be perfectly accurate about the sector it measures. That is the point: it is a statement about factories in one district, and it is quoted as a statement about the economy. Second, the two shorter series cover only two recessions each, so their percentages rest on a thin denominator of recession months — which the table makes visible rather than hiding, since the observation counts are printed beside them.

The conclusion that survives both qualifications is narrow and firm. A reading below the contraction line is common: it happened in 27.5%, 33.2% and 47.7% of all months in these three series. An event that occurs in between a quarter and a half of all months is not a warning.

When these surveys appear, and how far ahead

The reason these indices are quoted before the official statistics is that they arrive first. The Philadelphia Fed publishes its release calendar with the date and hour: the August 2026 survey was scheduled for 20 August 2026 at 8:30 a.m. and the September survey for 17 September 2026. A survey covering the current month, published within the same month, has no equivalent among the hard statistics: industrial production for June 2026 was the latest available on 13 August, and manufacturing and trade sales were still at May 2026.

That is a real advantage and it has a real price, which the count above measures. Speed is bought with a qualitative question to a regional sample, and the resulting series spends between a quarter and a half of its life below the line that is supposed to mean trouble. Both facts are true at once, and a page that reports the first without the second has sold the reader half the instrument.

The Philadelphia Fed also publishes the components behind the index — the percentage of firms reporting increases, no change and decreases, seasonally adjusted and not — which is the single most useful thing to look at and the one nobody quotes. A move from +5 to +15 could be more firms improving, or fewer deteriorating, or a shift out of “no change”; the headline cannot distinguish them and the components can.

What these sources do not say

• No national PMI value appears here.The publisher’s report page served a CAPTCHA form and 922 bytes on 13 August 2026, and nothing from it was taken from elsewhere.
• These three surveys are regional and sectoral. None of them measures the United States, and manufacturing can contract while the wider economy grows.
• A diffusion index does not measure quantity. It counts firms reporting a direction, so a small rise and a large one count the same.
• None of the publishers states a recession threshold. The centre line means equal numbers up and down, and nothing was read that assigns it any further meaning.
• The two shorter series cover only two recessions each. Their percentages rest on a thin denominator, and the observation counts are printed so the reader can see it.
• No claim about markets. No price series was read for this page.

How to verify this yourself

1. Request the ISM report page and look at the size of the body. It should be under a kilobyte and contain a CAPTCHA form.
2. Open the Philadelphia Fed survey page and read the paragraph describing what the survey asks. The word “qualitative” is in it.
3. Download fredgraph.csv?id=GACDFSA066MSFRBPHI and count the rows. There should be 699, beginning May 1968.
4. Count how many of those rows are below zero. You should get 192.
5. Cross those 192 dates against the NBER peak and trough months. 121 fall outside every dated recession.
6. Find the minimum of each of the three series. All three fall in April 2020.

Where to go next on this site

• Leading and lagging indicators — eight series timed against two peaks, and the order that reshuffled.
• Recession indicators — the same counting method applied to the yield curve and the Sahm rule.
• The economic calendar — when the hard statistics arrive, and how old they are when they do.
• All macroeconomics guides — the area index, one click away.

Sources opened for this page

Each source below was opened and read on 13 August 2026. Verified on 13 August 2026.

• Federal Reserve Bank of Philadelphia — Manufacturing Business Outlook SurveyUsed for: the description of the survey as a monthly qualitative survey of manufacturing firms in the Third District asking participants to indicate the direction of change over the past month and six months ahead, the statement that published data include diffusion indexes and the components of increases, no change and decreases both seasonally adjusted and not, the start date of May 1968, and the release calendar entries of 20 August 2026 and 17 September 2026 at 8:30 a.m.. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series GACDFSA066MSFRBPHI, GACDISA066MSFRBNY and BACTSAMFRBDALUsed for: all 699, 301 and 266 monthly values with their start dates of May 1968, July 2001 and June 2004, the July 2026 readings of 41.4, 15.6 and 1.3, the minima of minus 58.7, minus 80.0 and minus 73.6 all in April 2020, the maxima of 58.5, 39.0 and 47.6, and the counts of 192, 100 and 127 months below zero from which the recession comparison was computed. Read 13 August 2026.
• National Bureau of Economic Research — US Business Cycle Expansions and ContractionsUsed for: every peak and trough month used to classify each below-zero survey month as inside or outside a dated recession, and the note that the business cycle data were last updated on 14 March 2023. Read 13 August 2026.
• Institute for Supply Management — ISM Report On Business, two addressesUsed for: the record that on 13 August 2026 one address returned HTTP 200 with a 922-byte body containing only a Google reCAPTCHA form and its submitting script, and a second returned 404, so no index value, methodology or threshold from this publisher is reported anywhere on this page. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series INDPRO and CMRMTSPLUsed for: the latest reference periods of June 2026 and May 2026 on 13 August 2026, used to establish how far ahead of the hard statistics these surveys are published. Read 13 August 2026.

Change log

• 13 August 2026— page rewritten from scratch against the Philadelphia Fed survey page, three FRED regional survey series, the NBER chronology and two attempts at the ISM report page, all opened that day. Removed from the previous version: national PMI values and a fifty-point threshold taken from a publisher whose page serves only a CAPTCHA; the claim that a reading below the line signals economic contraction, which fails in 63% to 84% of the below-the-line months in three complete series; sub-index values quoted with no source; a description of PMI methodology that no document read supports; and regional surveys presented as measures of the United States.

Vextor Capital is not authorised under MiFID II as an investment firm, is not registered with the SEC, FINRA, the CFTC or the NFA, is not an investment adviser, holds no position in any instrument named on this page, and receives no payment from any bank, broker, exchange, index provider or data vendor. Every outbound link on this page points to a statistical agency, a central bank or a public data series.