What even are these manufacturing PMIs?
By Wolf Richter for WOLF STREET.
Manufacturing in the US as seen by the two big US manufacturing PMIs:
The ISM Manufacturing PMI report this morning said: “Economic activity in the manufacturing sector contracted in November for the ninth consecutive month, following a two-month expansion preceded by 26 straight months of contraction.” The reading for November came in at 48.2%. Below 50% means decline.
Manufacturing activity has been declining for the past 37 months, with the exception of (barely) January and February, according to ISM.
But ISM’s measure would have to drop to 42.3% to ring the recession alarm bell, as a value above 42.3% “generally” indicates the economy is expanding overall. So ISM said today: “The overall economy continued in expansion for the 67th month after one month of contraction in April 2020.”
Four manufacturing industries reported growth in November, according to ISM: Computer & Electronic Products; Food, Beverage & Tobacco Products; Miscellaneous Manufacturing (medical equipment and supplies, jewelry, sporting goods, toys, and office supplies); and Machinery.
The S&P US Manufacturing PMI, also released this morning, told a different story, once again: “Operating conditions in the US manufacturing sector improved for a fourth successive month in November. A solid rise in production and a further increase in employment was reported as confidence in the outlook strengthened.”
According to S&P, manufacturing has been in growth mode all year, except for a small decline in July.
The S&P US Manufacturing PMI for November came in at 52.2 (above 50 = growth), “consistent with another solid, albeit slower, improvement in operating conditions.” In October, the reading had been 52.5. A key driver of the growth was “strong rise in factory production.” And it said, “further increase in employment was reported as confidence in the outlook strengthened.”
What even are these PMIs? Purchasing Managers Indices are diffusion indices based on surveys of a panel of executives of companies of all sizes spread across the major manufacturing industries. The surveys don’t ask for figures.
PMIs ask a series of questions, with three possible answers – increased, no change, decreased – in the current month versus the prior month.
For each respondent, increased gets a value of +1; no change gets a value of 0; decreased gets a value of -1. Then it’s averaged out.
In these PMIs here, the no-change line = 50, meaning the same number of respondents said “increased” as said “decreased.” A value over 50 means that more respondents said increased than said decreased. A value below 50 means that more respondents said decreased than said increased.
Actual figures are not given, neither in dollars nor in units. It’s just increased, no change, decreased; and always current month compared to prior month.
The S&P US Manufacturing PMI for November had some quibbles amid its growth reading (52.2). Rising costs due to inflationary pressures and tariffs were difficult to pass on due to competition, leaving “selling price inflation amongst the lowest of the year so far amid intense competition and weak demand.” And so, profit margins were “under pressure.”
More respondents said that sales in November decreased from October, than said sales increased. And more said that inventories increased from the prior month than said inventories decreased.
Again: based on three possible answers – increased, no change, decreased – not actual figures.
And it concluded: “Encouragingly, manufacturers have grown more optimistic about the year ahead, with the ending of the government shutdown helping lift confidence from the sharp drop suffered in October. Optimism is being fueled by hopes of improved policy support, including lower interest rates, as well as greater political stability, though it is clear that uncertainty remains elevated and a drag on business growth in many firms, holding confidence well below levels seen at the start of the year.”
The ISM Manufacturing PMI for November had lots of quibbles. But “production jumped into expansion” (51.4%) and prices rose (58.5%).
All other categories were below 50%, and therefore declining from the prior month:
- New orders (47.4%)
- Employment (44.0%)
- Supplier deliveries (48.9%, faster deliveries than in the prior month)
- Customer inventories (44.7% = “too low” which “is usually considered positive for future production)
- Backlog of orders (44.0%)
- Export orders (46.2%)
- Imports (48.9%)
Again, all based on increased, no change, or decreased – with no actual figures.
Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how:
I-N-F-L-A-T-I-O-N. The monster in the room.
ISM is closer to reality right now.
We’re running at 2009 levels of demand and output – off 50-60% from normal. Several suppliers have confirmed similar conditions with their other customers in unrelated industries. Our primary steel supplier recently said, “everything except metal going into data centers and grain storage is exceptionally slow”.
I heard through the grape vine that one of the largest names in our industry booked their worst month of sales on record in September 2025. In September, we logged our third-worst month on record. October and November were better, but not by much.
IMO This doesn’t mean the whole economy is weak… but I would argue that the “wealth-generating” segments of it are quite weak. The news is slowly picking up on this reality.
Generally, never extrapolate from one company to US manufacturing overall. It will just mislead you. It’s easy to do, and so it’s appealing, but company-specific stuff doesn’t apply to overall manufacturing. That’s precisely why we try to use data.
Companies go bankrupt all the time even in booming economies.
”ALL generalizations are false, including this one.” was one of the first aphorisms the CAL professor mentioned in the class, ”Lies, Damn Lies, and Statistics” back in the day before all the political correctness took over the education industry.
That class was the most fun of any class, with similar witticisms, etc., revealing the lack of rigor of statistics, but after the first few lectures, I had to take a job at the same time, and the prof allowed me to take only the final exam.
This reminds me of the issues with consumer sentiment surveys which are increasingly out of sync with hard data. Is there equivalent hard data for manufacturing? Tons of steel produced, # of widgets manufactured, etc?
The responses need to be weighted by the approximate size of the company in order to get a handle on the performance of the overall economy. Boeing is a lot more important than a balsa wood model airplane kit manufacturer.
Yes, that is one of the many drawbacks of these diffusion surveys.
Hey Wolf, are there any historical periods with similar “slowing, but not stopping” results? I’m mostly curious if things tend to get slower from here, leading to recession, or if there’s a historical basis to re-acceleration.
Here is the 10-year chart from the S&P Manufacturing PMI, from the S&P report:
Here is the 10-year chart from the ISM report, screenshot from YCharts.
“Soft patches” (a Fed chairman phrase from decades ago) happen all the time. 2015-2016 was a prior “borderline recessionary” period.
Wolf, Which measure is more reliable ISM or S&P. It is very confusing since both are conflicting each other.
I posted this mostly to show how the media picked the one that fit their narrative and ignored the other that didn’t fit their narrative. The media also routinely misrepresent the PMIs, and the PMI providers encourage that.
None of them are “reliable.” They’re something close to sentiment surveys, which are always “unreliable” in terms of economic indicators.
Also, they don’t tell you anything about a specific month, the moth-to-month changes are just squiggles, despite the headlines, unless something imploded, like in March/April 2020.
I look at them together for longer-term trends. I used to average them out when I still had access to the Markit PMI data, but when they got bought out by S&P, I lost access to the historic data. That was like making orapple juice. So now I just look at them together. And as long as they don’t deviate too much from the 50-line in either direction, I don’t worry about them. But when the manufacturing PMIs go over 55 or below 45, and stay there for several months, it’s time to pay attention.
Thank you once again for the clarity you bring,,, EXACTLY why I send you the C note every year!!!
Fantastic explanation
Say’s law has been denigrated.
This report seems accurate to me.
Our manufacturing levels and sales are up slightly in the past months (sporting goods.) Inventory levels are moving downward as I am getting more cautious about the future. Also doing a fraction more in the US so usually easier to control inventory. Our costs are still going up though.
We’ve always been fairly recession/downturn proof because when people are laid off they tend to do just as much or more recreationally.
Wolf,
Thanks for digging into the methodologies a bit and taking the time to explain it to the peanut gallery.
This is a fairly good example of why having an understanding of how the sausage is *really* made (how the “metric” is actually calculated, what its limitations are, what it appears to say versus what it definitively says, etc) is pretty important.
These kinds of sentiment-al surveys (impressionistic feelings relative to some arbitrary base date) always make my Spidey-sense tingle – they simply lack even the claimed objectivity of actual hard numbers that can be argued over.
To me, even a semi-reliable “metric” has to have more objective reality to it.
I not thrilled with the vast reliance on financial measures (knowing how easily, if transiently, they can be manipulated) but at least they are harder *numbers* – the sentiment -al -ish survey reports are…less so.
Post ZIRP, post inflation, I just wish that *output unit* measures were more easily available…for essentially everything.
Hard to really accurately judge the health of an economy without knowing actual output level trends – everything else (financials, sentiment) are just an overlay that can be easily distorted.
(And that’s not to say unit output measures are perfect – channel stuffing is a thing…)
“I just wish that *output unit* measures were more easily available…for essentially everything.”
Unit sales are available, and I discuss them here a lot, for big-ticket items such as homes and motor vehicles. But they also give you the wrong picture because the economy is far better off if 10 high-end vehicles are sold than 10 low-end vehicles. The first might add $2.5 million to GDP the second might add $250,000 to GDP. What matters for the economy is the amount, not the number units.
Counting smaller stuff — nails, socks, dishwashers, books, couches… — it makes zero sense as an economic measure. It’s nonsense to even suggest it but you keep clamoring for it, and I keep wasting my time shooting it down. That’s why we use “dollars” as a “unit of account.”
How much of this is being driven by the massive amount of capital being dumped into AI infrastructure build out? How healthy is it if the AI build out is the main driver keeping the numbers above 50? And what happens when build out finally catches up to the demand these AI companies are continually needing? The larger numbers indicate AI spending is largely keeping the market from entering a recession, unless I’ve heard/seen the wrong info.