The Cracks Beneath the Surface: July's US data

There is a strange disconnect running through American markets this summer. The S&P 500 has spent August printing record after record, carried almost entirely by the AI infrastructure trade, while nearly every hard data release covering the real economy has come in soft. Payrolls contracted. Housing starts collapsed. Builder confidence has now spent longer below the neutral line than at any point since 2012. And yet the index keeps climbing, and the long end of the Treasury curve keeps selling off, which is precisely the opposite of what a weakening economy is supposed to produce. In this article I want to work through what the July data actually said, and what it means for a Federal Reserve that entered August with three of its own members voting to raise rates.

Context matters here. At the meeting ending 29th July, the FOMC voted 9-3 to hold the funds rate at 3.50-3.75%. The three dissents, Hammack, Kashkari and Logan, wanted an immediate hike. The energy shock from the Iran conflict had pushed headline inflation to a four-year high of 4.2% in May, and a bloc of the committee had decided that the risk of second-round effects outweighed anything happening in the labour market. Going into the 7^th August payrolls report, Fed funds futures were pricing a 57% probability of a September hike. What followed dismantled that in a single morning.

Non-Farm Payrolls: The headline and the revisions

The July employment report showed non-farm payrolls falling by 23,000 against a consensus expecting a gain of roughly 80,000. A negative print alone would have been notable. What made it worse was the revision history: June was cut from +57,000 to +20,000, and the two-month net revision came to negative 103,000. That drags the three-month average down to just 20,000, from 77,000 as of the June report. For scale, the average monthly gain over the prior twelve months was 34,000. The US economy is now creating roughly a fifth of the jobs it needs to absorb normal labour force growth, and the trend is deteriorating rather than stabilising.

The composition was ugly in a specific way. Government payrolls fell 53,000, driven overwhelmingly by local government, which is a genuine drag but also an idiosyncratic one. Private payrolls rose 30,000, so the headline overstates the private-sector weakness. The problem is where those private gains came from. Healthcare and social assistance added 22,600 and construction added 22,000 and construction is a number that looks increasingly borrowed against the future given what happened to housing starts a fortnight later. Against that, leisure and hospitality shed 40,000 (with food services alone down 26,100), retail trade lost 19,400 and financial services lost 14,000. Strip out healthcare and you have a private economy that is not hiring. Manufacturing was the one genuine upside surprise at +30,000 against expectations of +4,000.

The Household Survey: A falling unemployment rate that means nothing good

The unemployment rate fell to 4.1% from 4.2%, beating expectations for it to hold steady. On any normal reading that is a good number. It is not.

The rate fell because the labour force shrank by 264,000. Total employment declined by 87,000 to 162.18 million. The participation rate slipped to 61.4%, the lowest since early 2021, and the employment-population ratio fell to 58.9%, the lowest since September 2021. Since May, the US labour force has contracted by 984,000 people. An unemployment rate that falls because people stop looking for work is a symptom, not a recovery.

The arithmetic is worth walking through slowly, because it is the whole point. The unemployment rate is simply the number of unemployed people divided by the size of the labour force. In July the numerator fell with 178,000 fewer people counted as unemployed. Ordinarily that would mean they found jobs. But employment fell too, by 87,000. If both the unemployed and the employed decline in the same month, the people concerned have not moved between those two categories; they have left the measured labour force altogether, which is exactly what the 264,000 contraction shows. The rate improved because the denominator shrank faster than the numerator.\

Then there is the question of who those people were. Of the 178,000 fall in total unemployment, 167,000, which was around 94%, came from teenagers, whose ranks of unemployed dropped from 907,000 to 740,000 and whose jobless rate fell from 14.6% to 12.1%. Among workers aged 20 and over, the group that drives household income and consumption, the unemployment rate did not move at all, holding at 3.8%. The broader U6 underemployment measure was likewise unchanged at 7.9%. So the headline improvement came almost entirely from one demographic, in a month when the total number of people working fell. That is not a labour market tightening. It is one that is quietly shedding participants.

Wages: The number that killed the hike trade

For a committee worried about second-round inflation effects, average hourly earnings were the decisive line in the report. Wages rose 0.1% on the month which was two cents against expectations of 0.3%. The annual rate slipped to 3.2% versus a 3.5% forecast, the slowest pace since May 2021.

This is the single most important number in the release, because it directly rebuts the case the hawks were making. Second-round effects require workers to win compensation for higher energy prices. At 3.2% nominal wage growth against 3.4% headline CPI, American workers are losing real income, not extracting a wage-price spiral. Whatever the risk from oil, it is not currently transmitting through the labour market.

The 12th August CPI report confirmed the picture. Headline CPI rose 0.1% on the month, taking the annual rate to 3.4% from 3.5%. Core rose 0.2%, with the annual rate easing to 2.5% which was the lowest since March 2021 and back to its pre-conflict February pace. The momentum measures were better still: three-month annualised core slowed to 1.6% from 2.3%, six-month to 2.4%. Shelter, roughly a third of the basket and the stubborn component for three years, rose just 0.1% and still accounted for around two-thirds of the monthly headline increase, which tells you how little else moved. Energy fell 1.5% on the month, though it remains 14.7% higher year-on-year. September hike odds were cut to 42% on the release and drifted lower from there.

Housing: Where 6.7% money actually bites

The clearest evidence that policy is already restrictive came from the housing data, and it is worth taking together rather than release by release.

Existing home sales fell 1.7% in July to a seasonally adjusted annual rate of 4.06 million which is still up 0.7% year-on-year, but stability at a depressed level. The median existing-home price hit a record $434,100, up 2.0% and marking the 37th consecutive month of annual gains, while inventory fell 1.9% to 1.54 million units, or 4.6 months of supply against roughly six for a balanced market. Prices are being held up by the absence of sellers, not the presence of buyers.

Construction is where the strain shows. Housing starts fell 12.4% in July to a 1.239 million annualised pace, 13.5% below July 2025 and well short of the 1.35 million expected. Single-family starts dropped 9.9% to 808,000, down 15.7% on the year, with the three-month moving average at 865,000. Multifamily fell 16.8% to 431,000 and completions dropped 5.8% to 878,000. Only the Northeast rose; the South, where roughly half of all US construction happens, fell 12.6%.

The counterpoint, and it is a real one, is permits. Building permits rose 5.0% to a 1.443 million rate, up 3.1% year-on-year, with single-family authorisations up 2.5% to 894,000 and multifamily up 9.4% to 549,000. Builders are still preparing ground even as they refuse to break it. That gap between permits and starts is the cleanest expression of the current bind: developers believe in the medium-term demand and cannot make the maths work at today's financing cost.

That cost is the whole story. The average 30-year fixed mortgage rate in July was 6.54%, up from 6.49% in June, and Freddie Mac's survey had it at 6.69% for the week ending 6th August which was a fifth consecutive weekly increase and the highest level of 2026. The NAHB/Wells Fargo Housing Market Index sat at 34 in July and ticked up one point to 35 in August, its sixteenth consecutive month below 40 and the longest such run since 2012. New single-family home supply stood at 9.3 months in June against a balanced market of six. Pending sales fell 2.3% across all four regions.

Overview

The July data describes an economy that is slowing in construction, retail, hospitality, discretionary hiring while inflation cools back toward target and workers lose real income. On a conventional reading, that is a cutting cycle, not a hiking one, and the market has repriced accordingly.

What makes the current setup genuinely unusual is that the long end refused to play along. Three separate releases in August argued for lower yields, and the 30-year Treasury still pushed to 5.31% on 17th August, the highest since 2007. The message is that the front end is now a cyclical instrument responding to data, while the long end has become a fiscal instrument responding to supply. It is worth separating the two numbers that get conflated here. The annual deficit, which is the gap between what Washington spends and what it collects in a single year, is running at roughly $2tn for FY2026, after a July shortfall of $432.3bn, the widest single month since March 2021. The debt stock, which is the accumulated total of every past deficit, has reached around $40tn, with the publicly held portion approaching 100% of GDP. The deficit is what determines how much new paper the Treasury must sell this year; the debt stock is what determines how much of the budget goes to servicing it, and financing costs have already run to $1.12tn through July. Add a wave of AI-related corporate issuance competing for the same pool of capital, and you have a long end that is being priced by supply rather than by the economic cycle. Weak payrolls fix the first problem and do nothing for the second which is why mortgage rates kept rising through a month of soft data.

The dates that matter from here are Jackson Hole on 28th August, where Warsh gives his first symposium address and where the market will be listening for whether the September hike is genuinely dead; the August payrolls report on 4th September; CPI on 11th September; and the FOMC decision on 16th September. If August payrolls print negative again, the argument stops being about whether the labour market is cooling and starts being about whether it has already turned.