Two people looked at the same speech this week and reached opposite conclusions about whether it contained new information.
Goldman's Jan Hatzius read Warsh's Jackson Hole remarks as his most hawkish yet, then kept his call for a September hold. Citigroup's Andrew Hollenhorst read the same words as "relatively uncontroversial" and "restated by Warsh each time he has spoken."
Both are reportedly right. That is the problem.
Nothing in the data changed between Thursday and Monday. No jobs print, no CPI, no PCE revision. Yet the implied probability of a September hike reportedly jumped to 66.1% on CME FedWatch, nearly double where it sat before Warsh opened his mouth.
The market repriced an entire policy path off tone. That, not the tightening itself, is the story.
This week the ambiguity resolved, but only halfway. Markets now agree a hike is more likely than not; the banks pricing that hike cannot agree on how many.
Goldman sees zero. Citi reportedly sees rate hikes as unlikely this year, with no consensus to hike in September. Barclays sees two — September and December. Société Générale sees three, reportedly running through March 2027. CITIC holds at zero, reportedly allowing at most a "symbolic hike" if long-end rates force one.
Zero versus two versus three, off one speech, off one PCE print. That dispersion is the actual signal, and it is worth being precise about where it comes from.
It does not come from disagreement about the data. These desks are reportedly looking at the same July PCE: headline 3.7%, core 3.3%, and the Dallas trimmed-mean measure — which strips the extremes off both tails — at 2.3%, reportedly "much closer to the Fed's goal."
Read that number and you hold. Read the headline and you hike. Warsh has told the market which number he refuses to read.
He acknowledged the soft prints directly and then dismissed them: they do "not tell me that underlying trends have meaningfully improved." He wants confidence that inflation is moving to target "clearly and at sufficient speed. Otherwise, we have work to do."
Notice what that sentence does. It removes the data point as the decision criterion and replaces it with "underlying trends."
A data point is falsifiable in real time. August core PCE either prints at 0.2% or it doesn't; you can check on the release date. An "underlying trend" is a judgment call that can only be adjudicated in hindsight, and by then the meeting is over.
So the desks are no longer forecasting inflation. They are forecasting Warsh's tolerance for it — guessing at the threshold in his head that no release will ever cross or clear on its own.
The desks are no longer forecasting inflation. They are forecasting Warsh's tolerance for it — guessing at a threshold no release will ever cross or clear on its own.
That is why identical inputs produce a four-way split. SocGen's Yan Gruen reads the trimmed-mean series as stuck near 3% with non-housing core services — reportedly nearly 60% of the core PCE basket — showing persistent momentum since the pandemic, and concludes Warsh will act. Goldman reads a base case of 0.2% August core, reportedly per its note, and concludes the bar won't be cleared.
Both are legitimate reads of the price data. Neither is a read of the reaction function, because there is no reaction function to read.
Warsh, reportedly by CITIC's account, "offered no significant new explanations and did not commit to a clear 'reaction function'" while abandoning forward guidance. He said short-term rates are "the primary tool" and that "higher inflation will require a higher federal funds rate" — but higher by whose measure, past which threshold, he did not say.
This is the substitution at the center of the week: a legible rule replaced by a rhetorical posture. Under Powell, you could argue about whether the Fed would follow its stated function. Under Warsh, there is no stated function to follow — only a tone to interpret.
And a tone has a trap built into it, which Bank of America's Aditya Bhave names with unusual precision.
Warsh, Bhave writes, "has raised the bar for standing pat by arguing that the Fed should focus on trends rather than 'isolated data points'." Having said the soft data doesn't count, he has pre-committed to discounting the exact evidence that would normally justify a hold.
"Absent a material downside surprise, the onus is now on Warsh to deliver a [September] hike," Bhave concludes. "Otherwise, he risks undermining some of the credibility he gained on Friday."
Read that carefully. The decision has been detached from the inflation number and attached to Warsh's own words.
If August CPI comes in soft — Goldman's 0.2%, Citi's cooler trend — the pre-Warsh logic says hold. The post-Warsh logic says a soft print is an "isolated data point" that doesn't move the "underlying trend," so holding on it looks like flinching.
The chair has boxed himself into a position where hiking defends his credibility and holding forfeits it, regardless of what the data does. That is a decision about a rhetorical posture, not about prices.
Now add the second master. The moonlight framing from earlier this year still hangs over the seat: Warsh arrived under the "sock puppet" label, hand-picked by a president who wanted cuts, reportedly accused by Elizabeth Warren of being installed to deliver them.
Trump's position reportedly hasn't moved — he continues to want lower rates. Which means a hold in September now reads, to at least part of the market, not as data-dependence but as capitulation to the White House.
So the hawkish posture is doing double duty. It proves Warsh takes inflation seriously and proves the hand inside the puppet isn't the president's — and both proofs now require the same action, a hike, largely independent of what August's releases show.
This is the illegibility trap closing. The committee is authoring a decision the market can't map to any rule, and the market is pricing a mood it can't verify against any release.
Bessent, reportedly from the G20 in Asheville, offered the cleanest version of the case against acting: "we've seen a supply shock, and traditionally you don't raise into a supply shock unless you see second- or third-order effects," with "core inflation has remained very, very restrained." That is a reaction-function argument — a rule about when tightening is appropriate — and it is precisely the kind of legible criterion Warsh has stopped offering.
Here is what makes the 60%-plus number false as a consensus. A probability that high implies the market has converged on a shared model of the decision. It hasn't.
It has converged on a shared guess about a man's mood, while the desks pricing that guess disagree by three full hikes on what the mood implies. The 66% is not a forecast of inflation policy; it is a forecast of Warsh's need to be seen as hawkish.
Which sets up the real event. If the hike comes in September against soft August data — and Goldman, Citi, and Barclays are reported to expect the monthly prints to run below the trend Warsh says he cares about — it will not be a data-driven move.
It will be a credibility move: a chair validating his own rhetoric and demonstrating independence from a fuming president, using the funds rate to settle a question about himself rather than about prices.
The tell will be simple. Watch whether a soft CPI in the week before the meeting lowers the hike odds or leaves them stuck.
If good inflation news no longer moves the price of a hike, the market has stopped pricing inflation entirely. It will be pricing a man who has decided that standing pat costs more than being wrong.