Three Dissents Are Not a Majority

The mechanism, in one line: a hike requires the chair, and in the modern Fed, rate decisions do not normally move against the chair. This chair has spent four months explaining why he thinks the current inflation is the kind you look through.

The question

On September 16 the Fed announces what it is doing with interest rates. It has held them at 3.50% to 3.75% for five straight meetings. The question is whether it holds again, raises, or cuts. Prediction markets price a hold at 66%. Our estimate is 80%.

What the Fed usually does from this position

Long flat stretches are the normal shape of Fed policy. Once the Committee is four or five meetings into holding and has not signalled a change, it usually holds again. Modern plateaus have run anywhere from four meetings to fifteen. That is the starting point, somewhere around 85%.

Then you adjust for what makes this one unusual, and quite a lot does. Three voting members dissented in July and wanted to raise. That is the first time in a decade three people dissented in the same direction. Half the committee has a rate rise pencilled in for this year. The minutes released on August 19 said many participants thought tightening would likely be necessary if inflation did not come down.

So there is real hawkish pressure here. It is just not a majority, and it lost nine to three with better data behind it than it has now.

What causes a pause to become another hike

The interesting question is what actually flips the Fed from waiting to moving. Looking at 1994, 1997, 1999, 2004, 2016 and 2023, three conditions keep showing up, and in every case at least two were present. Inflation stops falling. Financial conditions turn out to be looser than the Fed thought. And the labour market is tight, which makes moving cheap.

I could not find a modern Fed case where tightening restarted on easier financial conditions alone while the labour market was weakening. That absence matters, because that is exactly where we are.

The strongest counter-analogue is 1999. The strongest disanalogy is that 1999 had a tight labour market and an announced tightening bias. August 2026 has neither.

These moves also get telegraphed. In 1999 the Fed announced a tightening bias a month ahead. In 2023 the dot plot said two more hikes were coming. In 2016 there were speeches. There is nothing like that now.

Except there is a catch. Kevin Warsh took over as chair in May and deliberately abolished the telegraph. He removed forward guidance, shortened the statement, and has been evasive at both of his press conferences. So the absence of a signal means less than it normally would. Every meeting is genuinely live. I have adjusted for that, and it is the main reason the hike probability is 17% rather than 8%.

Where inflation is now

Headline inflation is 3.4% and falling. Core CPI, which strips out food and energy, is 2.5%. Wage growth is 3.2%, also a five-year low.

The headline number is doing a lot of work. Energy is up almost 15% on the year because of the war, and it has been falling month over month since June as that shock unwinds. Strip food and energy out and you get 2.5%.

What the labour market changes

In July the economy lost 23,000 jobs when forecasters expected 83,000 gains. The previous two months were revised down by 103,000 between them.

That does two things, and only one of them is obvious. It materially weakens the case for raising rates. Outside the Volcker years, the Fed does not start tightening the month after payrolls go negative, and Volcker was fighting 12% inflation.

What it does not do is build a case for cutting. Unemployment actually fell, to 4.1%. Layoffs are low. Almost the entire miss was government jobs, mostly local school districts, which looks like a seasonal adjustment problem. Private payrolls were positive. Every cut the Fed has made in the modern record was preceded by a rising unemployment rate, and this one is falling.

What still arrives before the meeting

Warsh gives his first Jackson Hole speech on August 28. July PCE lands August 26. The August jobs report is September 4. August CPI is September 11, five days before the decision.

That last one carries the most weight. If August core CPI comes in at 0.4% or higher, reversing the recent improvement, the original 80% maintain forecast would need to be revised materially downward.

That is a different statement from saying the forecast was wrong. New information can change the current probability without rewriting the August 25 forecast. PastBehavior scores the original estimate against what was knowable at the study date, and tracks updates separately.

Where PastBehavior and the market disagree

The difference is +14 points on maintain, and it is concentrated almost entirely in the hike. Action: watch. Signal state: potential signal. Direction: yes.

I think the market is pricing the June dot plot, the three dissents and the July minutes as if they were one coherent hawkish bloc. They describe the losing side of a vote taken before the two reports that undercut it. And they leave out the thing that actually decides this, which is that a rate rise requires the chair.

Warsh has said inflation is a choice, which sounds hawkish. He has also said one-time price shocks from energy or AI demand are not automatically inflationary, and that the Fed should stop believing inflation comes from strong growth and rising wages. Those are the words of someone building a case for patience. He has five internal task forces on inflation and AI due to report late this year. Raising rates in September concedes their conclusions before they are written.

There is one more piece. The political pressure on this Fed runs toward cuts, not hikes. That flips the usual credibility argument. Holding is already the independent choice.

The forecast

Kalshi has drifted from about 73.5% to 66% over the past week, going into Jackson Hole. Warsh has said the speech will frame big questions rather than give guidance. If he does what he has done every other time, some of that move is nerves, and nerves decay. That is why I would not average the estimate down toward the market. It is not a reason to treat 14 points as 15.

What would change the view

New information can change the current probability without rewriting the August 25 forecast.

What this experiment is testing

Whether you can price a central bank decision better by asking who has to agree to it than by counting who is arguing for it. Three people are arguing loudly. Nine are not, and one of the nine holds the gavel.