Revising the Revision

I can always tell when people are bothered by something because I get a lot of calls, emails and messages on certain topics. Lately, the topic seems to be the employment report, and more importantly the revisions to the employment report. The tinfoil hat brigade seems to be out in full force lately with this one. There is a looming feeling building that somehow the employment report is being manipulated to further an interest rate agenda. Ironically, I have never seen so many mortgage agents and realtors that are suddenly experts on employment reports, revisions, and labour statistics. I also find that the closer your paycheck aligns with lower interest rates, the more upset you are at the jobs report revisions. Well, let’s take a cursory look and see.

For this article I am going to us US numbers. Not because I am based out of the US, but because the US side gives us a lot better picture of how things work due to size and population. Also, the US has an entire division dedicated to the labour report, whereas Canada uses Statistics Canada which also is responsible for all government numbers.

In the US the employment report ( officially called the NFP or non-farm payrolls ) report. As the name suggests it is a monthly estimate of employment excluding the farming, private household, non-profit and military. Now, just based on this description we already have some problems. Firstly, you will notice it is an estimate of jobs. Estimate. Not definitive number. Second of all, it eliminates farm jobs, private household jobs, non profit jobs and military. Later on in the blog you will see where this can cause some confusion. The ‘ jobs report ‘ as it is often called is published the 1st friday of the month ( unless there is a holiday ). Most of this we already kind of knew.

However, most people don’t dig beyond that, so let’s get our shovels out and dig. Screw that, lets get a massive excavator and dig a trench, shall we?

Contrary to popular belief, the Bureau of Labor Statistics ( BLS ) does not go around and count every person, and ask them if they have a job or not. The BLS uses a survey sample to determine the unemployment, job, employment rate, etc. The BLS uses data collected from 560,000 business throughout the US. Every month these employers are responsible to submit their info to the BLS on things like number of employees, part time vs. full time, hours worked, hiring intentions etc. So, herein lies one of the massive problems – the data the BLS uses is dependant on timely and ACCURATE submissions from businesses. Now, I have run businesses, and I can promise you that reporting numbers to a government agency falls way down the list of things to do. But, for a minute, let’s assume business do at least report timely numbers. In Fact, the BLS says that for the initial report due the first Friday of every month, the receive timely collection of data from about 74% of the businesses. Much like in an election, when you have 74% of polls reporting, they can sort of extrapolate the data to get a final number.

Now of course, there could be outlier data in the remaining 26% of the reports that come in late – and their usually is. When this happens, the BLS is forced to state a revision to the numbers to ‘ true up ‘ the actual data received against the initial extrapolation or estimates that they made. This always happens 30 days after the initial report. This is referred to as the first revision. According to the BLS by the deadline for the first revision, they have received 92% of the data needed. So, they take the 74% of the data they had on time, the then remove their estimates for the data between 74% and 92%, and replace it will the actual data they received, and they release the 1 revision data that should be 92% accurate. Then or course by the time they release the 2nd revision 60 days after the first release ( 3rd actual release ) the number now includes data from 100% of the actual businesses.

However, the issue isn’t always with the timeline of the data but the accuracy. Let me explain. Let’s use a business like Wal Mart for a minute. Wal Mart turns over employees very fast. Wal Mart runs a good business, but the person who is preparing these reports for the BLS is based in Bentonville Arkansas, and has never met the workers working in the store in Rochester New York. Wal Mart on its own employs 2.1 million people. So, this random person sitting in Bentonville Arkansas has to tell the BLS how many employees were on the payroll on, let’s use May of 2024, so May 12 of 2024. May 12 2024 was a Sunday. So, how many people quit their job, or how many people did Wal Mart fire on the Friday, Saturday and Sunday? For a little context, Wal Mart has an employee turnover rate of about 22%. So that means on any given day, Wal Mart could gain, lose, or remain the same number of employees by about 1,265 people. So, in our example, the person at Wal Mart reporting to the BLS on Sundays data could be off by 1,265 people for Friday, 1,265 people for Saturday, and 1,265 people by Sunday. The reason I say they could be off, is that the person at HR in Bentonville Arkansas may not be updated on the hirings, firings, payroll etc of Friday and Saturday and Sunday until the following week after they have submitted the number to the BLS. So, Wal Mart is 1 of 560,000 business chosen to report, and just their numbers could be off by 2,795 jobs. This is just natural normal turnover, nevermind Wal Mart fires someone and chooses not to replace them. If Wal Mart decided to shrink their workforce over time by 5%, that could amount to a little over 100,000 jobs that may or may not be reported correctly in the first report, but would eventually be corrected for the 1st and 2nd revision.

Now let’s move on to the initial sentence part about what is included in the numbers. One of the aspects that is specifically NOT included in the employment report is military. However, this is a lot harder than it sounds. Quite often US employers will hold positions for people who are on military leave. They still need to hire someone to do the job of someone on military leave, but they can easily count the person who is on military leave but not working – because they are still on the payroll. The employer eventually catches that and revises the report to the BLS 30 days later. Now, a lot of people reading are going to argue that the US has not been at war in a while, but I assure you there are still hundreds of thousands of people who are active military and engaged. My little franchise location in the US with only 70 stores has 2 people at our head office operations alone that are National Guard. Both of these people have been ‘activated’ to work the Texas border with the migrant crisis. We only employ around 20 people at our head office location, so that is 10% of the corporate staff. So, do we count them as employed, or military? Well, we count them as employed, even though they are activated, because we know they are coming back. However, it is really simple to do that when there is only 20 people. Wal Mart cannot possibly know the active duty status of 2.1 million people, nor keep a timely, accurate count on whether the people that replaced them are full time, part time, etc. It is really hard to get accurate numbers on how many people this situation applies to, but it would not be hard for it to easily eclipse 100,000 people a month across the entire United States.

So, we know that employers can be late or inaccurate with the data, we know that military service can throw a curveball at the numbers, but let’s also give honourable mention to the household survey. This survey is how the BLS calculates the labor force participation rate, and the unemployment rate. This report is a lot to unpack, and I won’t do it here, because it is a very long, tedious process, but it is also part of the data the BLS needs to release alongside the jobs report. The unemployment rate and the labour force participation rate take up just as much time for the BLS as the non farms payroll report.

Now I want to look at the estimates that lead up to the release of the numbers on Friday. Whenever you see the report, you will see a headline along the lines of “jobs number blows past expectations” or something to that effect. Here is the actual one from the May report:

U.S. adds a much-better-than-expected 272,000 jobs in May

We have all seen these, but where exactly do we get the ‘expected’ part from? Where are the ‘estimates’ coming from? Every month, all kinds of economists, labour companies, statitions, payroll processors like ADP, etc all put out what they think the jobs number will be. Now, personally, a company like ADP who process hundreds of millions of paychecks in REAL TIME should have a pretty good advantage on this little guessing game, but they are usually one of the worst “guessers” of all. However, for May 2024, the estimates of the number ranged from 120,000 new jobs to 258,000 new jobs. So, basically they take the hundreds of estimates and guesses, add them all together and divide to get the expectation number. So, in May the expectation was 190,000, and the US, at least on the initial reporting on Friday reported 272,000.

The reason I bring this up, is I have seen a lot of people absolutely hammer the BLS for being wrong, and revising jobs down or up later on, yet even private companies who have up to the minute data cannot come anywhere close to the actual number. People are almost out in the street with pitchforks when they see the BLS do a revision of jobs by 100K. Now, let’s add some context to the revision.

As of the March report ( I used March because it has already had the initial report, the 1st revision, and the 2nd revision done ) there was a total of about 161,045,000.00 employed Americans. March’s initial jobs number was 303,000 new jobs created. In the 1st revision to March’s jobs numbers, the BLS revised the number UP to 315,000, and then in the May report the BLS revised the number down to 310,000. So, that gives us a total difference ( from the high to the low ) of about 12,000 jobs. So, the BLS was off by 12,000 on a working population of 161 million. Statistically we could all be so lucky to be as accurate on our lives. For those that are numbers people, it means they were off by .000074%.

Now, I can already hear the chorus of people that will claim I picked March because it was the month with the lowest revisions. They will tell me to pick January of 2024 which is the worst month. Okay, let’s run January 2024 then. Initial jobs numbers in January were 353,000, the 1st revision took the numbers down to 229,000, and then the final revision lifted the jobs number to 256,000. So, people will argue they were ‘ off ‘ by 97,000 jobs for January. Okay, well, divided by about 161 million employed people, they were off by about .00060%. January is also a month where employers hire a lot of help for Christmas, fire a lot of people at year end, and have a bunch of part time workers that must be calculated and reported to the BLS. Add in the holidays that get taken by the people who report the numbers, the busyness of year end, and a whole host of other factors , I am actually surprised the initial January employment number isn’t off by a hell of a lot more.

To all those that complain about the numbers being off, I suggest you strive for such ‘ offness’ in your life. I supremely hope I am only wrong by .0006% in my daily life.

However, there is a trend to revisions that you can use. First we must keep in mind that the employment report is a lagging economic indicator, it is not a leading economic indicator. Ragging on and on about the January jobs numbers in June is a huge waste of breath. The data is certainly used by economists and policy makers, but the data is stale and out of date, but it can show you trends, which are important. If, for example, Tiff and Co saw 25,000 job losses month after month after month consistently, then they would use that to help their argument of lowering rates. Central bankers use this data all the time, which is why central bankers are usually late to the party. They like to see 3 employment reports of data to support their thesis, but 3 months of employment data ( with appropriate revisions ) is already 6 months old. It tells us how the economy was reacting 6 months ago. There is one trend though that is pretty consistent with revisions: revisions up or down, will give you a pretty good idea of the economy. Revisions tend to revise downward heading into a recession, and they tend to revise to the upside heading out of a recession. That is not to say there cannot be false positives, but generally the trend is your friend here. The larger the revision, the harder the recession, or the better the economic boom, and the smaller the revision, the softer landing, or the softer economic boom.

Yes, there can be big revisions to the numbers. All the moving parts that go into the numbers by default make it a big project. The jobs numbers are one component of the data that gives us a look into the economy. There are dozens and hundreds of other numbers, reports, and releases people use to set monetary policy. I have been critical of the numbers myself from time to time. The problem is that if you are going to be critical of the numbers, you need to be consistent. When Canada was consistently revising jobs numbers up a couple of years ago, that should have meant that Tiff and Co. should have been raising rates. Of course no one in our space wanted that. Now though, that the revisions are below the initial reports, everyone in our industry is using that as an argument to give out lower rates. That makes our industry look bad. If you are going to be critical, be critical, but be consistent.

I find too many people in our business are using data like a drunk uses a lamp post – for support rather than for illumination.