Tastes Like Chicken: The Messy Truth About Monthly Jobs Data

The truth about employment data: estimates, adjustments, and a whole lot of math.

KEY TAKEAWAYS

• Employment drives consumption and GDP growth

• Fed policy depends heavily on labor data

• BLS monthly Nonfarm Payrolls are estimates based on samples and models

• Benchmarking against QCEW may reveal a 700k overcount

• AI is reshaping hiring patterns, making old models less reliable

MY HOT TAKES

• Markets cheering weak jobs data is upside-down but real

• The BLS process is more nugget than tender–processed, not pure

• Benchmark revisions could shake confidence in past data

• The Fed is flying half-blind with estimates instead of hard data

• AI will make traditional labor statistics less accurate over time

• You can quote me: “Nonfarm Payrolls are the Chicken McNuggets of economic data–processed and tasty, but not the real thing.”

Reconciliation. It’s all about the jobs, my friends. How do you pay your bills? Unless you are a trust fund baby or you happen to discover a cache of rare earth elements under your yard, you, like the rest of us, rely on your income to pay your bills, save money, and buy those things that your kids think make you look cool on Instagram. That’s right, our jobs are important to us. If those jobs are lost or they are imperiled, bad things happen, and not just for us, but for the economy as a whole.

I have said this what feels like a million times, but for a reason: strong labor market ➡ ️ consumer confidence ➡ ️ consumption    ➡ ️ GDP growth. Consumption makes up more than ⅔ of GDP, so it is important that consumption remains healthy. Money is necessary for consumption, and money is acquired from employment. Look, I know that YOU know this, but I have to make it crystal clear, so you understand why everyone is so obsessed with employment releases, starting with the Fed.

The Fed has good reason to focus on the labor market because it literally represents 50% of its raison d’être. That’s a fancy French term for: reason for existence. Posh! The reason for that… er, reason is because of its importance to economic growth. Got it? We want a strong labor market because it’s good for the economy. Great, now let’s move on. 

Here is where it gets a bit weird. Markets have been on the hunt for something to fuel the equity rally further, and rate cuts have become that thing. If the Fed perceives that the labor market is imperiled, it is more likely to cut rates. That is why–except last Friday–weak labor numbers have been somewhat bullish for the market. Twisted, but that is the way it has been working, at least recently.

I highlighted the stream of weak monthly labor prints in yesterday’s newsletter/blogpost, and it would be hard to argue that equities have not been buoyed by those weak numbers and the increasing prospect for rate hikes. Those numbers are published by the Bureau of Labor Statistics (BLS), whose boss was recently unceremoniously fired by the President. If you have been following me so far, you would know that the numbers crunched by BLS are extremely important, so the top job at the Bureau comes with a lot of responsibility.

It is obviously not an easy job–getting those numbers right. Last Friday, BLS reported that only 22,000 new jobs were added to the US economy which was far less than economists were expecting and lower than the amount added in the prior month. Not good for the economy but possibly good for the folks rooting for rate cuts. Did you ever wonder how BLS gets that exact number of new hires?

This is where things start to get muddy. You see, these monthly Nonfarm numbers–that are all the rage on the first Friday of each month–are only an estimate. “Wait what,” you exclaim! 😳 It’s true. BLS conducts a monthly survey called Current Employment Statistics (CES). Clever name, eh? 😉 It surveys around 122,000 businesses covering approximately 666,000 work sites and asks them how many employees have been hired or fired in the past month. Now, if you were astute, you would probably be thinking that the US economy has far more business, and you are correct. The Census Bureau estimates (more estimates, lol) that there are roughly 6.1 million employer firms (firms with at least 1 employee). If you count non-employer firms (sole proprietors, freelancers, gig workers), the number jumps to 33 million. So, the CES survey only represents, what we refer to in statistics, a sample. In order to get the total monthly Nonfarms number, BLS must put that sample through a mathematical meat grinder.

First, BLS must weight expand the sample, by industry, to represent the total population. For example, if only 10% of construction businesses were surveyed, the sample number would be multiplied by 10 to estimate the total in the economy (an oversimplified example, but pretty close). This is done across all industries to get a total. That total is then seasonally adjusted based on historical seasonality. It does this by using a time series ARIMA model (called, affectionately X-13ARIMA-SEATS). ARIMA stands for Autoregressive Integrated Moving Average, a statistical model that predicts future values in a time series using its own past levels and past forecast errors.

BLS then applies a “birth/death” model which generates assumptions of new firms and business closures. It uses another ARIMA model for this exercise–basically, it uses existing data to guess how many new jobs might have not been reported because new businesses have sprung up, and conversely, how many have been lost because of closures.

After all this processing, BLS releases its monthly estimate of how many new jobs were created in the US in the past month. Knowing this, you are probably now questioning just how accurate that monthly number is, and you are correct to do so. You might be thinking, “wow, so we have no clue how many real jobs have been created or lost, just educated guesses.” Well, not exactly.

You see, there is something called the Quarterly Census of Employment and Wages (QCEW). The QCEW is a near-census of U.S. payroll jobs based on unemployment insurance tax records that cover about 97% of wage and salary workers. In other words, it is based on actual administrative records–not voluntary reporting. Because it covers about 97% of ALL wage and salary employment, it is the “gold standard” of employment.

So, why don’t we just use that number instead of the CES estimate? Well, its name should give that away. It is only quarterly and it lags. The market, and the Fed (we hope), wants more current data, so we rely on the estimate to get a monthly read on the employment situation. But don’t fret, everything gets reconciled in the end.

Once a year in September, BLS conducts a benchmarking process where it compares its CES monthly estimates to the gold standard QCEW to determine over- or under-estimation for the year that ended last March. That benchmarking process will be announced today at 10:00 AM Wall Street Time. Blue-chip economists are expecting the number to show the CES over-estimate employment adds by 700k jobs. That would mean that the labor market was in worse condition than the past monthly numbers have suggested. Oh, and those 700k jobs will be added to next January’s Nonfarm Payrolls number reported on the first Friday of February 2026! Wow!

So, there are a few takeaways. Employment numbers are extremely important in assessing the health of the US Economy. Those numbers are a primary input in the Fed’s process of determining monetary policy. The accuracy of those numbers is therefore, paramount in getting policy right. We will get an idea of just how accurate those numbers were today. Everyone will be watching–trust me!

Now if you are still awake and you have even roughly followed me through this morning’s journey, you would have noted that there are many places where the estimates could pick up inaccuracies, the least of which is the methodologies and models utilized. The models themselves are constantly being modified–hopefully for the better. That process along with the data collection is why BLS is under such scrutiny.

Now, I am just going to present you with an interesting twist. Remember how I told you about those ARIMA models? They rely on past data to predict the future. That makes sense if the future is similar to the past. Have you noticed that something has changed in the economy recently? Ok, I am going to stop beating about the bush. ARTIFICIAL INTELLIGENCE! That’s right, AI is making workers more efficient, ALREADY. That simply means that hiring patterns today and in the future are going to be quite different than they were in the past, even last year. You don’t need to have a doctorate in finance 😉 to know that that difference is likely to make statistical prediction less accurate. Just saying! And this needs to be explored much more.

Have you ever had a Chicken McNugget from McDonalds? It’s been a while for me–my kids are grown, but I remember them tasting great. They are technically chicken, though it is pretty clear that they are highly processed. They do a great job at filling little tummies, though they may not be the most accurate representation of an actual chicken tender. Look, it is true that the monthly employment numbers’ accuracy can be easily scrutinized and should be questioned. Despite that, the direction and pattern of recent months cannot be questioned. Whether it was 22k or 20k or 30k matters less than the fact that those numbers are trending in the wrong direction. Tastes like chicken to me.

YESTERDAY’S MARKETS

Stocks gained yesterday as investors raised their hopes of more rate cuts sooner in the wake of last Friday’s weak employment figures. Treasury Note yields slipped across the curve as rate-cut expectations resonated with bond traders as well.

NEXT UP

• Preliminary Benchmark Payrolls Reconciliation (2025) is expected to show an over-estimate by 700k jobs after last year's estimate was off by 818k. Weird and laggy, but notable.

• Important earnings today: SailPoint, Core & Main, Rubrik, Synopsis, GameStop, and Oracle.

DOWNLOAD MY DAILY CHARTBOOK HERE 📈