Every first Friday of the month markets gather like sinful parishioners awaiting Pavlov’s bell, the gospel according to the Bureau of Labor Statistics (BLS). The numbers descend from the mount, etched in holy government letterhead. Jobs added, jobs lost, unemployment rate up, unemployment rate down, labor force swelling or shrinking. And the market screams, up or down, as if the deity of statistics has spoken to the profits of capitalism.

Expectations unmet? The elevator drops. Expectations exceeded? Euphoria ascends. Every stinking first Friday the ritual repeats. Buyers and sellers beware.

The markets, seasoned by countless cycles of god-awful truth and revision, know the numbers are suspect. Model-based, massaged and provisional, destined for downward revisions not once, not twice, but likely thrice. And everyone knows this. Logic says to remain circumspect. Religion says yell hallelujah.

But they react like Charlie Brown charging towards Lucy’s planted football, full of conviction and hope that this time it’s real. No, it’s a con of smoke and mirrors, a ritual sealed with a wink and nod from Lucy. A trust not earned.

For Charlie Brown, it’s not just a kick, but a ritual of belief. A sprint to a promise of truth and redemption. This first Friday, the numbers will be true. This time the markets aren’t the chump. But then the revisions put you flat on your back in the muddy turf.

This isn’t market ignorance, just the willful need to believe. To believe in a clear signal in a noisy world. Maybe this Lucy keeps her word. Maybe this time the BLS can count timecards.

The U.S. labor market is a statistical mirage and the BLS is the magician. A clumsy one at that. Given the scale of the U.S. labor force the BLS sample is statistical noise superimposed on a trendline, and the trend itself shaped by the unreliable Birth-Death Model.

The Birth-Death Model estimates job creation from new businesses (births) and closed businesses (deaths) that aren’t captured in their monthly survey sample. The BLS’s Current Employment Statistics (CES) survey covers about 122,000 businesses each month, but it can’t track firms that just opened or shut down. Since new firms don’t immediately appear in the sample, and closed firms may linger as ghosts in the data, the BLS uses a statistical model to estimate their net effect on employment.

The model has two main components. Imputed deaths are estimated from trends for similar firms. For example, if 3 out 4 newly created restaurants fail annually, the model adjusts employment accordingly. The other component forecasts net job changes using historical data from the Quarterly Census Employment and Wages (QCEW).

The model’s assumptions presume that today’s labor market mirrors yesterday’s. It somewhat works in stable times but thoroughly breaks down when the sharp kinks of recessions and recoveries are introduced into the system. Also, business owners fail to report their failures, quickly—go figure. BLS just finds it is too much work to keep track of them all. A brief detour, if you’ll indulge it, is worth mentioning here. I was standing in line at the local US Post Office and a customer in front of me, after much back-and-forth discussion, showed a photo of a misdelivered package to the postal clerk. The clerk lit up in cheerful vindication. “Oh that wasn’t us. We don’t have time to take pictures of our deliveries.” The BLS operates with similar blind spots. Ghosts in the data, and no time to chase them.

Now back to our regularly scheduled harangue. All the BLS assumptions in their Birth-Death Model leads to inevitable revisions. The model is the altar. The CES survey, is the smoke, wafting over a ritualized trend.

In practice, this means employment is overstated during downturns and understated during booms with the actual numbers taking another 3-12 months to correct. By then, the market had moved on. The altar cleared. The smoke dispersed. And the ghosts remain.

But there is hope, but only for the patient.

ADP, Automatic Data Processing, offers a monthly employment snapshot that often outpaces the BLS. Released each Wednesday before the government’s numbers, the ADP National Employment Report draws from payroll data covering 26 million workers. No government jobs. No statistical smoke. Just raw payrolls. It consistently lands closer to the QCEW gold standard, though still misses the mark, just not as wildly as the BLS.

Between March 2024 and March 2025, QCEW, the altar of actual payroll filings, reported a net gain of 675,000 jobs. BLS, guided by surveys and the Birth-Death Model, claimed 1.79 million. ADP, closer but still adrift, reported 1.69 million.

Together, ADP and BLS conjure the headlines that move markets. But their numbers are unreliable. Why bother? A million-job error in a labor force of 167 million is less than 1%. A rounding error. A statistical ghost.

Ghosts in the data. Smoke in the temple. Floating through the firmament. By the time the truth arrives the markets have moved on.