Most of This Number Is Already Written

The question is whether U.S. consumer prices will be up 3.0% or more in the year to September 2026. Headline CPI was 3.4% in July. So the question is really about how fast a number can fall, not whether anything new pushes it up.

Two months is not very long, and much of the arithmetic is already fixed.

What 3.0% actually requires

Two figures decide this, both published by the Bureau of Labor Statistics.

September 2025 came in at an index level of 324.800. July 2026 came in at 333.918. For September 2026 to print 3.0%, the index has to reach 334.382.

That is a rise of 0.139% spread across August and September. Roughly seven hundredths of one percent a month.

For scale, prices have risen an average of 0.28% a month over the past year. The threshold asks for about a quarter of that pace.

Here is the part that surprised me. Because the bar is cumulative across two months, September CPI can print 3.0% even if prices fall outright in September. It depends entirely on what August does first, as the table above sets out.

So an outright monthly decline in consumer prices, the kind of number that would lead every news bulletin as inflation cooling, is compatible with the annual rate holding at 3.0%. Those are two different questions and they can have different answers in the same month.

Put it another way. At a soft 0.15% monthly core pace, the non-food-and-energy portion of the index would contribute enough that only a meaningful negative move in energy would be likely to pull the overall print below the threshold. So the interesting question is not whether inflation continues. It is whether energy falls hard enough to cancel the rest of the index out.

The base effect

Year-over-year inflation is partly mechanical. Each month, an old month drops out of the twelve-month window and a new one enters.

The months dropping out here are August and September 2025, which together added 0.542%. That is a firm base, not a soft one. September 2025 had gasoline up 4.1%. So the arithmetic is not doing the work for the YES side. If anything it makes September the harder of the two months.

What makes the bar low is not the base. It is that the starting point, 3.4%, sits so far above the threshold.

What is still inflating

Core inflation is 2.5% and has printed between 0.0% and 0.4% every month for well over a year, with 0.2% by far the most common. Shelter is 3.2% and slowly cooling, though the two recent soft readings were partly a lodging effect; rent and owners' equivalent rent both rose 0.3% in July. Food is 3.0% and steady. Goods are 0.8% and flat.

None of that is dramatic. All of it is enough.

What could push it over, or under

Energy is the swing factor, and it points up right now. Core sets the floor. Energy decides whether the floor holds.

Gasoline is 24.6% higher than a year ago because of disruption to shipments through the Strait of Hormuz. Brent fell to about $69 in early July after a US-Iran memorandum, jumped to roughly $105 on July 23 after renewed tanker attacks, and settled near $89 in mid-August.

At the pump, the EIA's weekly survey has August averaging about $4.06 against $3.93 in July. That is a gain of roughly 3% with one week left to report. In the past thirty-one years, the typical August has seen gasoline fall slightly. This one is in the top quarter of the distribution.

So August is not just neutral. It is contributing.

The strongest case the other way

Everything above leans one direction, which is when it is worth naming the case that beats it.

In August 2006, headline inflation was 3.8% and core looked stable. Gasoline then fell 13.4% in a single month and the annual rate dropped 1.7 points across two reports. Nothing in the core data gave any warning.

Under the central August and core assumptions, missing 3.0% likely requires September gasoline to fall roughly 8% or more. That has happened in two of the past thirty-one Septembers. Both times it was a supply story reversing.

There is a live version of that story now. If Hormuz transits normalise in early September and crude retraces to the low seventies, the pumps follow within a few weeks.

One caution about a tempting analogy. When the Ukraine premium unwound in 2022, gasoline fell almost 19% across August and September and headline inflation barely moved, from 8.5% to 8.2%. That is not reassurance. Core was running near half a percent a month then and absorbed the shock. Core is running near 0.2% now. The index is more exposed to energy today than it was in 2022, not less.

The forecast

PastBehavior puts the probability of September CPI printing 3.0% or higher at 86%. The central estimate is 3.2%, with a reasonable range of 2.9% to 3.5%.

Confidence is moderate rather than high, for a specific reason. One scheduled release moves this a lot.

What would change the view

The August CPI report on September 11. If August headline comes in at or above 0.25% unadjusted, the probability moves to about 92%. If it comes in under 0.20%, it drops to about 66%.

After that, weekly gasoline prices through the September collection window. Down 6% or more against August takes this below 60%. Down less than 2% takes it above 93%.

And a caveat on timing rather than outcome. The September report is due in mid-October, just after the federal fiscal year ends. In 2025, September data was collected before a lapse in appropriations and published nine days late; October data was never collected at all. The same shape is possible again. It would delay the answer without changing it.

Gasoline triggers are review ranges, not automatic point updates. Any revised probability is set on review and dated separately.

What this experiment is testing

PastBehavior Prediction Markets builds a probability from published data before looking at any market price, then records both and scores them later.

This study reached a PASS for a reason worth stating plainly: no matching September 2026 contract could be retrieved, so there is nothing to compare the forecast against. That is not a small disagreement. It is no comparison at all, and it gets logged that way.

The estimate stands on its own record regardless.

This is Prediction Markets No. 011.