No. 010 · Economy / Labor · Open · August 26, 2026
Unemployment Is Falling for a Reason That Has Nothing to Do With Hiring
Will the U.S. unemployment rate be 4.5% or higher in September 2026?
PastBehavior Action
PASS
Reason: Market Unavailable
No Comparison
No market price for the target September contract was observed at study time, so no PastBehavior-versus-market gap was calculated.
This PASS means no actionable comparison was available. It does not mean PastBehavior and the market were measured and found to be within five points.
Original forecast snapshot
The mechanism
The unemployment rate can fall even when hiring is weak. That happens whenever the labor force shrinks faster than employment does.
Unemployment rate = unemployed people ÷ labor force
If people leave the labor force faster than unemployment rises, the published rate can fall even while employment weakens. That is the situation the September contract sits inside.
The contract does not ask whether hiring is weak. It asks whether enough people remain in or return to the labor force while employment weakens enough to push the published rate to 4.5%.
How far 4.5% is
That single crossing began in August 1957, the start of the 1957 recession.
How thresholds actually get crossed
We looked at every historical month sitting four tenths below a round threshold. The pattern was consistent.
- No case crossed in a single month.
- Crossings generally required both months.
- They were preceded by visible deterioration.
- Payrolls weakened.
- Claims rose.
- The labor force held steady or grew while employment fell.
Labor-force contraction usually works in the opposite direction.
September unemployment distribution
Sum100%
The question
Will the U.S. unemployment rate come in at 4.5% or higher for September 2026? The number gets published on Friday, October 2. Our answer is 5%.
How far 4.5 is
In July the rate was 4.1%. Reaching 4.5% means rising four tenths in two months. That sounds small. It is not.
Since 1948 the published rate has moved four tenths or more over a two-month window in about one month out of nine. Almost all of those are recessions. Strip out recession months and the figure falls to 2.4%. Narrow further to months where the rate had declined in each of the two prior months, which is where we are now, and it falls to 2.2%.
The tightest comparison is months that printed exactly 4.1%. There have been sixteen since 1948, excluding the pandemic. One of them reached 4.5% two months later: August 1957, the month the 1957 recession began. The other fifteen stayed within two tenths in either direction.
How unemployment actually crosses thresholds
We looked at every month sitting four tenths below a round number, whether that was 4.0, 4.5, 5.0, 5.5, or 6.0. Not one of them crossed the threshold in a single month. When crossings happened, they took both months, and something always showed up first. Jobless claims rose for several weeks running. Payrolls turned negative and stayed negative. Crucially, the labor force held steady or grew while employment fell, so people who lost jobs stayed in the count as unemployed.
That last part matters more than anything else in this forecast.
What the labor market is doing now
Two things are happening at once and they point in opposite directions.
Payrolls are weak. July lost 23,000 jobs. May and June were revised down by 103,000 combined. Over the past year the economy added an average of 34,000 jobs a month, which is very slow.
Payroll weakness
Weak hiring is real. The question is whether it translates into higher measured unemployment before October 2.
And yet the unemployment rate keeps falling. It has dropped from 4.4% in February to 4.1% in July. The reason is that people are leaving the labor force faster than jobs are disappearing.
Labor force vs employment
The labor force is shrinking faster than employment, which mechanically suppresses the unemployment rate. This is why payroll weakness and a falling unemployment rate can coexist. It is not a sign of a healthy labor market.
When people stop looking for work, they stop being counted as unemployed. That is why the rate can fall in a month when payrolls shrink. The consequence is that the number of jobs the economy needs to add just to keep unemployment flat is currently negative. Employment can decline and the rate can still go down.
Meanwhile the layoff data is about as quiet as it gets. Weekly claims for unemployment benefits are running near 205,000, and in mid-July the four-week average touched a level not seen in roughly sixty years.
Layoff signals
Current weakness is showing up more through low hiring than through rising layoffs.
Labor deterioration check
3-month unemployment average: 4.20% · 12-month low: 4.1% · Gap: +0.10
The unemployment rate is currently near its twelve-month low, not accelerating away from it.
The strongest case the other way
Here is the argument we take seriously. The mechanism is participation reversal, not job destruction.
Roughly two million people have left the labor force in the past year. If some of them come back and do not immediately find work, the unemployment rate rises without a single additional layoff.
The 1957 parallel differs in one important way. Claims and payroll deterioration were already visible before the rate moved. Today they are not. But the mechanism is real, and it is why the forecast is 5% rather than zero.
There is a second version of the worry. Low claims might be measuring a smaller pool of covered workers rather than a healthier market. People may be exhausting benefits before they find work, which would make continuing claims look better than the underlying situation. Both are fair points.
The forecast
No market comparison was possible at study time.
The argument against a higher number is not that the labor market is strong. It clearly is not. It is that the specific thing the contract asks about requires roughly 600,000 more unemployed people in two months, and that has effectively only happened at the start of recessions. The current leading labor indicators do not resemble the recession-onset cases that produced most historical moves of this size. What they show instead is a labor market that is barely growing while its workforce shrinks, and that combination produces low unemployment rates, not high ones.
What would change the view
August jobs report · September 4
4.4% or higherForecast rises to roughly 35%
4.3%Forecast rises to roughly 15%
4.1% or lowerForecast falls to roughly 2%
Labor-force rebound
A monthly increase of about 400,000 or more triggers a review regardless of what the headline unemployment rate prints. This is the participation-reversal mechanism firing.
Initial claims
A four-week average sustained above roughly 245,000 would materially raise concern.
Continuing claims
A move above roughly 1.95 million triggers a review.
JOLTS layoffs
A meaningful rise in the layoffs and discharges rate triggers a review.
What this experiment is testing
PastBehavior Prediction Markets builds an independent probability from historical base rates and current data, locks it before looking at any market price, and then compares.
This month the comparison did not happen. The September contract was not listed with a live quote at study time, so there is no gap to report and no action to take.
A forecast with no market comparison can still be valid and scoreable. This one stands and gets scored against the October 2 release. Recording that a price could not be found is part of the discipline, not a gap in it.
This is Prediction Markets No. 010.
Contract validity
The settlement variable itself is clear.
- U-3 unemployment rate
- Seasonally adjusted
- September 2026 reference month
- BLS Employment Situation
- Initial published value
- Later revisions excluded
- 4.5% exactly resolves YES
The exact September strike was not verified as listed at study time. That affects market availability, not forecast validity. The two are tracked separately.
Market status: unavailable at study
No live September 2026 strike equivalent to 4.5% or higher was observed during the post-lock comparison stage.
Market quality is graded NOT ASSESSABLE because no target-contract spread, depth, volume, or open interest was observed. No price was inferred from sibling contracts.
Footnote: other near-dated strikes in the same unemployment series have carried meaningful volume and open interest. That observation does not create a grade for this market.
No-comparison guard
This study should not be grouped with normal PASS forecasts where PastBehavior and the market were measured and found to be within five points.
It belongs to NO_COMPARISON until a target contract becomes available and the market-availability trigger fires.
Independence
The September target market was not observed before the forecast was locked. Accidental exposures during contract verification involved different months, different strikes, or different unemployment-market structures, and were logged and quarantined.