Sherlock Holmes On The Jobs Report

A Quirky Report

Last week was not that exciting (except for the last bullet point).

The S&P 500 and Nasdaq 100 finished roughly where they started (while the Russell 2000 did well).
Credit spreads did very little (CDX tightened a smidge while bonds widened a touch).
Bond yields drifted a little higher (9 bps on 2s and 4 bps on 10s), while the 2s vs 10s inverted further to -86 bps (a topic discussed last weekend in Markets Heat Up as Global Relations Cool).
Oil sank 2%, but “Dr.” copper rose almost 2% as each marched to the beat of their own drum, as opposed to “recession on or off” trading.
Bitcoin slumped again, but like the proverbial tree falling in a forest when no one is watching, it really didn’t make a sound as so few look to bitcoin for any signal at all at this stage.
The VIX and MOVE (Treasury vol) indices continued to decline, though the MOVE index seems high relative to VIX. However, it could well be that VIX doesn’t do a particularly good job of assessing options interest in a world dominated by 0DTE options.
Academy had the opportunity to guest host Bloomberg Surveillance for an hour on Monday morning. While much like the rest of the week, it might be boring to you, but it was fun for us! Academy starts at the 0:49 mark and ends at 1:37. We get to interview experts on oil, rates, and travel and also get to discuss Academy’s unique capabilities including analyzing geopolitical risk for companies and markets.

This week has the potential to be a little more exciting with the Fed meeting taking place, but I wanted to focus on a few “quirky” things. However, with FOMC week coming up, we have to address the Fed before moving on to the “quirks” that I really want to write about.

What Can the Fed Do?

We get CPI on Tuesday which has the potential to move markets as expectations are for improvement across the board. However, I’m of the view that it will take weeks of consistent data to shift us away from this pause/skip narrative. Also, on Tuesday, Academy will be hosting drinks in Boston with General (ret.) Spider Marks. While not market moving, it should be fun and informative.

The only thing that I think the Fed could do to really move markets would be to adjust their dots.

They could add a hike or two to this year’s projection. However, I’m not sure that would do much for markets as one more hike is already largely priced in.
The median for the end of 2024 is for 4.25%, though Fed Fund futures are significantly lower. So maybe raising those dots to closer to 5% would do something, but since markets already doubt the existing dots, I’m not sure that will do much.
It is 2025 and beyond that get more interesting. The last dot plot had 3.125% for the “end of 2025” level, which markets actually think is too low. So, they could bump that up and “lock-in” expectations. The “longer-term” projection is 2.5%, which also seems low.
If I see the potential for a sell-off in stocks and bonds, it will likely result from a change in tone on the “terminal rate” as opposed to anything else.
Skip/pause is baked in and is likely the message that we will get (with a heavy dose of data dependency and warnings that inflation is still priority #1, which no one really believes).
They are unlikely to back down on balance sheet reduction (if they did, that would be a positive). They are also unlikely to discuss accepting a higher average level of inflation than the current 2% target (they probably are going to accept this, but it is just too early to admit it).
So, I’ll be keeping an eye out to see if they adjust the terminal rate.

In any case, I continue to believe (hope) that at least for a little while, we have a respite from the Fed being the main/only driver of markets! The meeting is important, but it should largely be a non-event.

What Are the Odds?

Let’s start with the one thing that is puzzling me more than almost anything else - how did the economists botch last month’s job numbers so badly?

Seriously, what are the odds of 76 estimates on the Bloomberg terminal ALL being lower than the actual published number?

This has been bothering me so much because the main argument against a recession is the strength of the job market. Job Market Strong = No Recession. I understand that, I’m just perplexed about how the industry (collectively) missed it so badly! These are very smart people. They have resources. They also have many different ways to calculate their estimates. Academy doesn’t have an official estimate of our own, I just look at consensus and what some people that I respect and trust have to say on the data.

So, if you have a number of animals typing away for long enough, one will randomly produce a work of Shakespeare. The odds of it happening are incredibly low, but with enough letters being typed, it will likely happen. Yet, with so many smart humans (some of which presumably use AI in their forecasts), not a single one was higher than the “official” data. Is that really likely?

The first question that I have is what are they predicting? At first blush, that seems like a stupid question because the obvious answer is that “they are predicting jobs data”, but it is actually far more nuanced than that.

Are they trying to predict the Establishment Survey jobs number or are they trying to predict the number of jobs created? Those are not quite the same thing.
ADP, for example, changed their methodology to try to produce a job number that would be more predictive of the NFP data. Why they would take their own unique payroll data (and manipulate it) to try to estimate the official government data is beyond me, but they did it. So, ADP isn’t really trying to analyze how many jobs were created, it is trying to produce data that helps people predict NFP (at least the Establishment Survey).
So, how many people in the Bloomberg consensus are trying to report the number of jobs created versus predicting what NFP will be (before the inevitably large revisions)? My working assumption is that people are trying to estimate the number of jobs (not predict NFP), but that could be a flaw in my logic. I feel incredibly sorry for those who apparently miss it on the initial report (which is how they will be remembered by traders) only to have been proven accurate once the final revisions are reported (which aren’t very market moving). “Hey, it turns out that you were right and we were wrong”, but that admission is only in the “fine print” and few will pay attention to it.
The Household Survey showed a loss of 310k! Why does no one seem to care about that? The Household Survey is used to calculate unemployment rates, so clearly someone at the BLS must think that it is useful. We largely refer to the NFP data in terms of jobs (Establishment Survey) when discussing the unemployment rate (Household Survey). These key numbers do not come from the same data series!
- The margin of error on the data (according to the BLS) is 130,000 for the Establishment and 600,000 for the Household! 600,000 for the Household! So real job creation in the Household would not be statistically significant if it were anywhere from -900,000 to +300,000! Seems insane. Even on the Establishment Survey, job creation of “only” 210,000 would be inside of the margin of error. This means that 20 estimates would have been spot on or a bit high (or the “real” number would have been an even higher print of 469k). It strikes me as so odd that we publish 339k as such an exact number, but a +/- 130k deviation would not be statistically significant.
By the way, response rates are slipping. BLS tracks response rates. In April 2013, the jobs data had a response rate of 64% (a C- even under a generous grading system). It is down to 42.7% (a level even Blutarsky might be embarrassed by). Why have survey response rates dropped so much? Is there any “selection” bias in those who respond to the survey, versus those who don’t? Is there anyone at the BLS asking what they are doing wrong since less than 50% of people respond? I would be shocked if there isn’t, yet here we are, publishing numbers down to 1,000 people. JOLTS is even worse - from 69% down to 31%.

The takeaway is that if we are predicting NFP (rather than jobs), that task has become more difficult as the survey response rate is reduced and more numbers must be “fudged” at the official level to produce a report. All of that “fudging” or “smoothing” or “seasonality” (which is always tricky) must be almost impossible to track in an economy that went through COVID, COVID shutdowns, and various stages of “re-openings” etc.

I’m back to where I started, what are the odds that so many smart people got it so wrong?

ADP versus NFP

ADP also came in higher than every single estimate, but only 31 people chose to submit estimates on ADP and at 278k versus a high estimate of 250k, at least some were “in the ballpark”.

So, let’s explore how the 31 who submitted estimates for both behaved.

The biggest differences were that one person had 55k more for NFP than ADP and one had a -30k differential. The average was 17k. So clearly, they are trying to pick up differences in how ADP calculates things versus NFP. If you were trying to estimate jobs created in May, wouldn’t you have the same submission for both? Apparently not. I’m not really sure why, unless they really are trying to understand the nuances of the two reports.

The other thing that I found surprising (though maybe this is indicative of how little people care about responding even to the Bloomberg surveys for estimates) was that 28 submissions for ADP and NFP were on the same date, and for the 3 that were different, ADP was actually the more recent of the submissions. Maybe it was because this month the data came out on Thursday (and then again on Friday) so no one had time to change their official estimates of NFP. However, I would have expected to see at least one person try to increase their estimate after the surprising beat by ADP.

On the NFP estimates, 4 firms submitted on June 2nd (presumably before the release) and they had an average estimate of 221k with a high of 241k (which was the 2nd highest estimate). Presumably they learned something from ADP, as they came in closer to the number (and some are within the margin of error).

I am not sure what to make of the difference in estimates between ADP and NFP. However, I was surprised by how few seemed to incorporate ADP results into revised forecasts (narrow window of time might explain that). In addition, I was surprised by indications that people tried to predict ADP separate from NFP. Is this a possible indication that they are trying to predict the specific data releases rather than predicting the number of jobs created?

Sherlock Holmes on Jobs

The fictional detective is credited with saying the following:

“When you have eliminated all which is impossible, then whatever remains, however improbable, must be the truth.”

What if the “impossible” is that many smart people, working independently with ample resources (including some harnessing AI), got it so wrong?

We know that response rates are low.
We know that “smoothing” data is difficult after the massive COVID disruptions.
We know that the acknowledged margin of error is large.

Then why is it so difficult to believe that the official data (which doesn’t even agree with itself) is wrong?

I am going to bet on the economists who can be fluid, rather than the rigid nature of official data.

I’m not all doom and gloom, but the assertion that jobs are extremely strong and therefore recession risk is low might be built on a shaky foundation (at least a foundation that Sherlock would question).

The Rule of 3

I might as well go “all-in” on quirky today.

In a world where we hear consensus this/consensus that, contrarian this/contrarian that, etc., I keep thinking about the Rule of 3.

The Rule of 3 is loosely related to the “by the time my mother knows” rule. By the time my mother is asking me about something in the financial world, the trade is played out.

The Rule of 3 is quirky, but perhaps useful.

Something happens that very few people pay attention to.
Then people start talking about this “thing” and it happens. The “thing” could be as simple as “stocks rise from 3:50 to 4:00” every day. Only some people are talking about it and few are betting on it
Then more and more people start talking about something that “always” happens. This time there is lots of “chatter” about “stocks go up from 3:50 to 4:00” and more people bet on it.
Then “everyone” is talking about how stocks have to go up from 3:50 to 4:00. All sorts of reasons are provided, and people bet on this. Then, guess what, it doesn’t happen.

Loosely the “Rule of 3” says that by the 3rd time something is supposed to happen (and “everyone” agrees that it will happen”), it doesn’t happen.

As quirky as it sounds, when trying to be contrarian, I find that thinking about this helps. What is “everyone” saying happens “after the FOMC” for example? Do we always sell-off on the announcement and rally hard after Powell is done with the press conference? I don’t know, but I think that this is an interesting way to think about positioning around things like this.

Though, does the “Rule of 3” apply to the “Rule of 3”? This to me is similar to asking, “if the consensus is always wrong” and that is consensus, isn’t it wrong too? If “the consensus is always wrong” is completely accurate, then every contrarian would be a zillionaire (and finding consensus should be easier than ever using AI). It would also violate its own rule of “something that is consensus cannot always be right”, so betting against the consensus will sometimes be wrong.

Or, in the “Rule of 3”, every third application of the “Rule of 3” will be wrong (so the 3rd time will work as well as the previous 2 times).

But, every 3rd time that this is applied, it too should be wrong

I guess that is a long and somewhat colorful way of saying that figuring out what is consensus is difficult and figuring out how markets will respond to consensus is trickier than it seems. This might be really relevant as we head into a week with more relevant data than last week, a long weekend coming up, and the FOMC. I do currently believe that “good news is good and bad news is bad” for markets, but that might be too consensus. However, that might not matter (and yes, my head hurts). As quirky as this might seem, I think that this is a useful exercise.