The Institute for Supply Management publishes the manufacturing PMI on the first business day of every month at 10 ET. It is the earliest hard read the strip gets on the prior month, ten calendar days ahead of the payroll print. It is also the noisiest, because it is built from a diffusion index, and diffusion indices have a specific math structure that produces bigger tails than a level series does. This piece walks through what a diffusion index actually is, why the 50 line is the recession-versus-expansion knife-edge in the underlying math, and why a single sub-index surprise inside the ISM release routinely moves the front-end reprice further than an in-line payroll number does.
What a diffusion index is
The ISM survey goes out to purchasing managers at roughly 400 US manufacturing firms. For each of ten survey questions (new orders, production, employment, supplier deliveries, inventories, customer inventories, prices, backlog of orders, new export orders, imports), each respondent picks one of three answers about the prior month relative to the month before: better, the same, or worse. The ISM then computes the diffusion index for each series using a fixed formula.
The formula is: percentage answering “better” plus one half of the percentage answering “the same.”
That is the entire construction. There is no dollar weighting, no size weighting, no sector weighting inside a given series. Each of the 400 respondents contributes one vote per question. The resulting number always lies between zero and one hundred. Fifty is the level that obtains when exactly zero respondents say better, zero say worse, and one hundred percent say the same. It also obtains when the shares saying better and worse are exactly equal, regardless of how many say the same. That is why 50 is the neutral line: it is the mathematical center point where “better” and “worse” answers balance.
The headline manufacturing PMI is a weighted composite of five of the ten diffusion indices: new orders (weight 0.20), production (0.20), employment (0.20), supplier deliveries (0.20), and inventories (0.20). The other five (customer inventories, prices, backlog, new export orders, imports) publish alongside as diagnostic sub-indices but do not roll up into the headline.
Why the 50 line is a knife-edge
The 50 line is not a soft threshold. It is a hard mathematical boundary between two behaviorally different regimes of respondent answers. A print at 49 means, in the average sub-index, more respondents said worse than said better. A print at 51 means the reverse. There is no way to sit on 50 with a normal distribution of answers: any nonzero skew shows up as a print off the neutral line.
That is why the recession-flag literature keys off the 50 line so mechanically. The NBER business-cycle dating committee does not use PMI as a formal input, but the market’s recession-timing playbook has treated a sustained sub-50 headline as a leading indicator of contraction with about a six-to-nine-month lead going back to the 1948 series origin. The historical hit rate on that signal, using the convention of three consecutive sub-50 headline prints, sits at roughly 70 percent for the pre-2008 sample and roughly 55 percent for the post-2008 sample. The false positive rate has climbed as manufacturing’s share of GDP has fallen (down to roughly 10 percent of nominal GDP by 2025 from roughly 20 percent in the early 1980s), which is why the desk treats a sub-50 headline print as a warning bell for the goods complex more than a recession call for the full economy.
The 49 to 51 band is the range where the strip prices the reading as neutral. Anything inside that band is treated as a null result on the recession question. Anything outside it forces a reprice, and the reprice is bigger on a per-point basis at the edges of the band than at the center. That asymmetry falls out of the diffusion index math: at 50 the respondent distribution is symmetric, and small changes in the answer mix produce small headline moves. At 55, the distribution is skewed enough that additional skew in the same direction runs into the ceiling of “everyone says better,” and each additional point of headline reads as a bigger regime shift.
Sub-index versus headline: where the read lives
The desk reads the headline first because that is the number the wires publish at 10:00:01 ET. The reprice on the release, however, gets driven by two sub-indices more often than by the headline itself: new orders and employment.
New orders is the leading sub-index. Purchasing managers place orders one to three months ahead of the production those orders drive, so the new orders diffusion index leads the production sub-index by roughly one quarter. A new orders print that runs three to five points above the headline usually means the next-month headline will follow the new orders line up, and the strip prices that persistence within about six basis points at the front end.
Employment is the sub-index that maps most directly to the payroll print ten days later. The ISM manufacturing employment sub-index has a rolling six-month correlation with the manufacturing NFP diffusion index of roughly 0.70, and a same-month correlation of about 0.55. That is why a sharp move in the ISM employment sub-index, particularly a print that crosses the 50 line in either direction, forces a reprice on the front-end fed-funds strip that only partially reverses on the actual payroll number a week and a half later. When the ISM employment print crosses 50 to the upside from a sub-50 start, it is often the first hard data confirmation that the manufacturing labor complex has stopped shedding jobs, and the strip prices that as a hawkish signal into the next FOMC meeting.
The prices paid sub-index is the third sub-index the rates desk carries, but it lives in a different framework. The prices sub-index reads goods-side inflation about six weeks ahead of the goods-CPI print, and the reprice on prices paid runs through the two-year on-the-run more than through the front of the fed-funds strip. A prices paid print above 70 has historically preceded a goods-CPI acceleration in the following three-month window about 60 percent of the time, which is why the desk treats a prices paid print in the 70s as a hawkish overlay on whatever the headline did.
Why the miss on Monday walked the strip further than a payroll miss usually does
Monday’s July ISM manufacturing headline at 55.6 landed 5.8 points above the June 49.8 print. The pre-release consensus band centered at 49.5, with the highest sell-side call inside the survey window at 52.0. The 55.6 print sat outside every published forecast in the Bloomberg survey and inside the plus 55 to plus 56 tail the options market had priced at four percent probability going in.
Two features of the print explain why the December fed-funds probability walked from 72 percent Friday to 58 percent Monday close, a 14-percentage-point reprice on a single 10 AM data release.
The first is the size of the sub-index moves. Production ran at 58.5 (up from June’s 51.4). Employment ran at 52.8 (up from June’s 44.2), the first expansion print in fifteen months. Both are moves of 7 to 9 points on their respective sub-indices, and because diffusion indices carry more information at the tails than at the center, a 9-point employment sub-index move from 44 to 53 is a bigger signal than a 9-point move from 47 to 56 would be. The 44 to 53 move crosses the 50 line, which the strip reads as regime change, not drift.
The second is the position the strip carried into the print. The consensus had priced a continuation of the June sub-50 headline, and the December second cut had been priced at 72 percent probability on that path. When the print landed 5.8 points above the June headline and 6.1 points above consensus, the reprice ran through two channels simultaneously: the level of the strip repriced the goods-CPI trajectory hotter through the prices paid overlay, and the shape of the strip repriced the labor-market read tighter through the employment sub-index crossing 50. Both channels push in the same direction on cut probabilities.
The 14-point walk on the December contract was the largest single-day reprice on a manufacturing PMI release since the September 2021 print, and only the fourth double-digit reprice on the manufacturing PMI since the FOMC started publishing the SEP dot plot in 2012. Every one of those double-digit reprice events has been a print that crossed the 50 line in either direction while producing a same-direction move in the employment sub-index.
What the desk does next
The 10 ET JOLTS release Tuesday morning is the first cross-check. The JOLTS quits rate has been the labor-tightness line the FOMC prices against for the last two years, and a soft JOLTS print at 10 AM would take some of the reprice back through the front end. The 8:15 ET ADP print Wednesday and the 10 ET ISM services print Wednesday are the next two cross-checks, and the 8:30 ET July nonfarm payroll print Friday is the terminal read that either confirms the ISM manufacturing signal or reverses the reprice.
The strip walked. The mechanics of the diffusion index are why the walk was as large as it was on a single release. The next four data prints inside the same week are what the desk carries the walk into.