A few weeks ago I wrote here that the U.S. economy added 57,000 jobs in June, and that this was a weak figure. Last Friday it turned out that it wasn't only weak, it also wasn't right. The real June number is 20,000 jobs. The Bureau of Labor Statistics, the very same body that published the figure at the time, went back and revised it down by 37,000, and revised May down by another 66,000, from 129,000 jobs to 63,000. Together, 103,000 jobs that were on the books were erased. Nothing surprising in that, by the way: this figure is always revised after the fact, up to two months back. What matters to us is the trend.
The July report itself, published the same day, strengthened the direction even further. The U.S. economy didn't add jobs in July at all, it lost 23,000. Economists had expected a gain of roughly 83,000, so the gap between what was expected and what happened is about a hundred thousand jobs in a single month. It's the first monthly decline in some time, and it comes after an average of 34,000 new jobs a month over the preceding 12 months. A good part of it came from the public sector, which shed 53,000 jobs in July. The official unemployment rate, despite all of this, actually fell from 4.2% to 4.1%. We already took that strange mechanism apart here in Letter #16: the number goes down even when people give up and stop looking for work, because the moment they stop looking they are erased from the count. The participation rate, the share of adults who are either working or looking for work, fell this month to 61.4%, the lowest in more than five years. The same story exactly, only deeper.
So why does it actually work this way, with revisions running backwards every time? Well, the U.S. jobs report isn't counted, it's sampled. Every month the Bureau contacts about 120,000 businesses and government offices and asks how many people they employed that month. Some answer on time, some answer late, and some don't answer at all. The report comes out on a date fixed in advance, whether the answers arrived or not, so the first release rests only on those who managed to reply. Over the next two months the answers keep trickling in, the sample fills out, and the number is updated twice before it's locked. Every employment number you see in a headline on publication day is an early estimate: the best one available at that moment, but not final, and everyone in the field knows it.
The process will be familiar to you from somewhere else entirely. On election night, the networks call states hours before the counting is anywhere close to done, working off partial returns and exit polls, and every so often a call has to be walked back as the mail ballots and the late precincts come in. The official canvass takes days, sometimes weeks, and only when each state certifies do we know what actually happened. Nobody feels cheated when the numbers move overnight, because everyone understands that what they watched at 11 p.m. was an estimate. The jobs report works exactly the same way, except that nobody goes back on air to tell the viewers the number they were shown two months ago was 66,000 too high.
There's another layer here, and it explains why the revisions of recent months keep running in the same direction. A business that is growing and hiring is usually an organized business, with somebody to fill in the survey on time. A business that is shrinking, laying people off or closing is exactly the one that doesn't rush to fill in forms, and if it has closed it never will. So when the economy loses speed, the answers that arrive late are the bad ones, and the first release runs systematically too optimistic. This is not malice and not manipulation, it's a property of the method. But its meaning is uncomfortable: exactly at turning points, when it matters most to know where the economy is heading, the first number is the least reliable one you'll get.
This is where it stops being a statistical curiosity and becomes money. Nine days before this report came out, on July 29, the rate-setting committee of the Federal Reserve, the central bank of the U.S., left interest rates at 3.5% to 3.75%. The vote was nine to three, and the three dissenters didn't want to cut rates, they wanted to raise them. The official statement said economic activity is expanding at a solid pace. Nine days later it emerged that this solid pace had been measured on a May and a June that were, together, 103,000 jobs smaller than what the committee was shown when it voted. I'm not claiming they would have decided differently. I am saying the people holding the interest rate all of us pay decided on the basis of numbers that changed nine days later, and that explains a good deal of the market's nervousness around every release.
The market's reaction was immediate and strong. The index of the 500 largest U.S. companies rose about 3.6% over the week and closed Friday at an all-time high, the technology-heavy Nasdaq rose about 5%, and it was the best week in the U.S. markets since April. In Letter #16 we explained why the market rises precisely on bad news, and the principle hasn't changed: in the short term the market doesn't react to the economy itself, it reacts to what it believes the Fed will do because of the economy. What did change is the direction. Back then investors priced in a rate cut moving closer; this time they priced in a rate hike moving further away. The market still assumes there will be a hike this year, only now it has been pushed out to December. Alongside that, the yield on the 10-year U.S. government bond, the annual interest the government pays whoever lends it money for ten years, fell to 4.64%. That's part of the same story.
One number in the report was clean of all these revisions, and in my view it's the most troubling of them: wages. Average hourly earnings in the U.S. private sector rose by just 2 cents in July, to $37.62, and their annual rate of increase, meaning how much wages rose over the past 12 months, fell to 3.2%. That's the lowest since May 2021. An employer who struggles to fill positions raises pay; an employer who knows someone will replace tomorrow whoever quits today does not. Wage growth losing momentum is usually the most honest evidence there is about the state of the labor market, not least because there's nothing in it to revise two months later.
What does this mean for your pocket?
The first lesson is a reading habit, and it's worth real money. The next time you see a headline with an economic number in it, ask first whether it's a first release or a figure that has already been revised. A first release is a good estimate, not a fact, and anyone moving money on the difference between 57,000 and 83,000 is moving it on noise. You can simply wait for the corrected version, and at the very least know that it's on the way.
The second point concerns Israelis no less than Americans. Most Israelis hold a substantial slice of the U.S. market without ever having bought a single American stock, through their pension fund, their study fund, or a fund that tracks the index of the 500 largest U.S. companies. This week, in which that index closed at a record, went into your savings as well, and not because the American economy got stronger this week, but because the fear of a rate hike was pushed off by a few months. That's a difference worth remembering on the day this assumption flips, because then it will work in precisely the opposite direction.
Anyone waiting for a rate cut in the U.S., whether because of a dollar loan or because of a portfolio, would do well to stay clear-eyed. The labor market there really is weakening, but Kevin Warsh, the Fed chair, said explicitly that his priority is bringing inflation down to target rather than supporting employment, and we wrote about that in both Letter #16 and Letter #17. To understand where rates are heading, you have to see that two forces are pulling them in exactly opposite directions. On one side there's a weakening labor market: fewer people working, pay rising more slowly, and anyone worried about their job spending less. That's a force pushing the Fed to cut rates, so money gets cheap and people and companies go back to spending. On the other side there's inflation still above target: prices in the shops are still climbing too fast, and that's a force pushing him in exactly the opposite direction, to keep rates high and maybe even raise them, so people buy less and prices settle down. The Fed can't satisfy both forces at once, and it has to choose which one worries it more. Warsh has already told us plainly what his answer is, so as long as prices don't cool, he probably won't be in a hurry to change anything, and a market that is mainly pricing the odds of a hike treats no change at all as good news.
Before we part, a word about a stock we left open here two weeks ago. In Letter #18 we took apart the collapse of IBM, which lost a quarter of its value in a single trading day. Since then the stock has been crawling back up: from its post-earnings low of about $211 it reached about $237 at Friday's close, roughly 12% higher in three weeks. In my view, that drop was an overreaction. The market today is nervous, and it doesn't really know how to price the question of how much artificial intelligence will eat into a company's profitability five years from now, so when bad news arrives it would rather sell now and think later. That is exactly the mechanism we took apart in Letter #19: uncertainty about future profits pulls down the multiple the market is willing to pay and widens the swings in the short term. So nothing dramatic was needed for the stock to start coming back, no new announcement and no good quarter. All that was needed was for the nervousness to ease a little, and the stock began crawling its way back up. Let's keep proportion: on the eve of the collapse it traded around $290, so even after this crawl it is still about 18% lower. The next report, in October, will tell us whether the crawl is justified.
We shall see.
See you in next week's letter, or during the week — if I can't help myself until then...