US ISM and PMI business surveys

What a diffusion index actually is 

A purchasing managers' index does not measure how much output changed. It measures how many respondents said it changed direction. Each month a panel of supply and purchasing executives is asked, for each of several activities, whether the level this month was better, the same, or worse than last month. The answers are counted, not weighted by the size of the change, and the result is scaled onto an index where the neutral point separates a majority reporting improvement from a majority reporting deterioration.

This is a measure of breadth, and its properties follow from that. It has a floor and a ceiling, because the share of respondents answering in one direction cannot exceed everyone. It is insensitive to magnitude, because a firm reporting a small improvement and a firm reporting a very large one contribute identically. And its reference point is the previous month, not a baseline level, so a sustained sequence of readings above neutral describes continued expansion rather than accelerating expansion.

Two competing families, published days apart 

The United States has two distinct PMI families and confusing them is the most common error in reading them. The Institute for Supply Management, a professional body of supply management practitioners, publishes a manufacturing index and a services index. A commercial data provider publishes a separate set of manufacturing, services and composite indices for the United States as part of a global family compiled on one harmonised methodology across dozens of countries.

They differ in panel composition, in question wording, in the weighting of sub-components and in publication timing. The commercial family publishes a flash estimate late in the reference month, based on a majority of the final responses, followed by a final reading in the first days of the next month. The institute's indices publish only once, at the start of the following month. The two families can and regularly do disagree about the same month, and neither is a revision of the other.

The components inside a headline 

A headline PMI is a weighted combination of several separately published sub-indices, and the sub-indices carry most of the information practitioners take from the release.

• New orders. Demand arriving now for output delivered later, and conventionally treated as the most forward looking component of the survey.
• Production or business activity. What was actually made or delivered in the reference month.
• Employment. Whether headcount rose or fell, which is a breadth measure of hiring and not a count of jobs.
• Supplier deliveries, sometimes called delivery times. This component is inverted: slower deliveries raise the index, on the reasoning that suppliers slow down when demand is heavy. It is the component most likely to mislead, because deliveries also slow during supply disruption, when demand is weak.
• Inventories, both the respondents' own and their customers'. Read together with new orders as an indication of whether stock is accumulating against or ahead of demand.
• Prices paid. Input cost pressure, again as breadth rather than magnitude, and one of the timeliest monthly readings available on cost conditions.

The prices paid component and the inverted supplier deliveries component are the two most frequently misread. Prices paid records how many respondents saw costs rise, not by how much, so it can remain well above neutral while the rate of cost increase is slowing. Supplier deliveries can rise for opposite economic reasons, and distinguishing them requires reading it against new orders rather than alone.

Cadence and coverage 

Both families are monthly. The commercial flash estimates arrive in the last week of the reference month, which makes them among the earliest reads on a month that has not yet finished. The institute's manufacturing index is published on the first business day of the following month and its services index a couple of business days later, in the morning Eastern Time.

Coverage is worth keeping in proportion. Manufacturing is a small share of United States output and employment and a much larger share of its cyclical variation, which is why a manufacturing survey attracts attention out of proportion to the sector's size. The services indices cover the far larger part of the economy and have a shorter history, which is one reason the manufacturing series retains the longer analytical track record.

How they are conventionally read 

Three conventions dominate. The first reads the index against its neutral level, treating a crossing as a change in the sign of the majority view rather than as a change in the growth rate. The second reads the direction of travel over several months, on the reasoning that a single monthly survey of a panel carries sampling noise. The third reads new orders against inventories, since orders accumulating faster than stock and stock accumulating faster than orders describe different situations.

A fourth practice, converting a PMI level into an implied rate of GDP growth, is a statistical mapping fitted to history rather than a property of the survey. The institute publishes such a correspondence for its own indices, and the mapping is an empirical regression that has held better in some periods than others. Treating it as arithmetic rather than as a fitted relationship overstates what a breadth measure can say about magnitude.