Wading through noisy economic data

OK, we know economic data is messy and noisy. Methodologies, samples, etc: so many things can go wrong. Japan is a perfect example: whatever one thinks about whether GDP growth is really an appropriate target for an economy with a shrinking population (maybe, just maybe, per capita GDP?), I wish more folks would do a deeper dive into Japanese statistics to discover how seriously flawed the consumption data is. But I digress.

So the US payrolls data came as something of a jolt. Not that we weren’t due for a jolt given how close the data has hewed to median forecasts for so long now. But there were few reasons to expect such a hiring slowdown (well, one or two), especially coming a month after US consumer spending seemed to suddenly come back to life. What’s going on? Let’s take a look at both May payrolls, household and establishment surveys, and the PCE spending data so far this year.

Of course it’s easy to blame poor seasonal adjustments. Yet clearly some of the adjustments being applied to data are creating more noise than reducing it — and this matters when the global slowdown means there’s even more obsession about how the US is faring to help carry the world along.

Guillermo Roditi captured the suspicious feeling well yesterday many hours after the May payrolls data came out.

So in simplistic terms, there are two surveys that compile the payrolls data: 1) the household survey that is made up of a broad survey of what people are saying about their employment/unemployment situation, and; 2) a regular survey of many businesses about what they are doing on the hiring/firing front. The two surveys have their quirks, and the household survey in particularly contains a significant amount of volatility. Just look at the historic one-year standard deviations of the monthly changes in the household survey on civilian employment relative to that of non-farm payrolls.

As Guillermo alludes to, there have been such differences between the two surveys lately that something is clearly amiss: we may be seeing some convergence to make up for some of the weird discrepancies of recent months and huge volatility in the guts of the reports that results in either this being a one-off outlier or there is a big revision later on that drags this outcome of just +38k non-farm hiring closer to recent trends. Thus, it assumes this is out-of-whack rather than an indicator of a sudden deterioration in the economy. And as we’ll get to in a bit, the PCE report gives more reason to be a little more careful about jumping to conclusions on some of this data.

Since October, the household survey had mostly showed much bigger employment gains than the establishment survey did, resulting in a rise in people re-entering the workforce and thus helping keep the unemployment rate near 5.0% for months even as the establishment survey showed pretty steady job gains, if to a much lesser extent than the household survey. For example, between October and March the household survey showed an increase of 2.38 million jobs vs 1.43 in the establishment survey — a nearly 1 million difference. Some of these trends started to reverse in April with what was a nominally soft payrolls report: the difference between the household and establishment survey shrank by almost 500k. The picture was more complete with May: payroll growth dropped off suddenly and the workforce shrank much more dramatically, thus the unemployment rate dropping to just 4.7% on such a poor hiring report.

There were some other weirdly large changes going on beneath some of the headline figures: the number of long-term unemployed fell 178k, so a positive; those unemployed less than 5 weeks plunged 338k; yet the number of involuntary part-time workers that would prefer a full-time job shot up 468k. More reasons to be cautious about reading too much into this result.

Before the May report, it seemed clear either the household survey was overstating hiring or the establishment survey was understating hiring and some convergence was bound to happen. Now it seems both were overstating things at a time when many economists and Fed officials had expected demographics (retiring baby boomers) to exert more influence on headline payrolls via a shrinking labor force, thus requiring fewer payroll gains each month to just keep the unemployment rate steady or even still keep dragging it lower. So perhaps just +80–120k payroll gains rather than the +200–250k ones the market has long been used to. Thus what a near full employment economy looks like at a time of big demographic changes.

The payrolls revisions to previous months give reason for caution as well. The trend in this very long recovery had long been for upward revisions. That tailed off in 2015 when revisions turned more net flat vs the first report as seen below. But since February those revisions have turned negative, if not massively so. The momentum does not look good.

One could say the May report was similar to the March surprise last year (just +84k, a third of that the previous month) that helped defer a Fed rate hike for many months (along with a strong USD causing so many global problems and the tightening financial conditions of Aug/Sept). Ultimately that was an outlier and ultimately almost as many jobs were added in April-Aug 2015 (1 mln) as had been added in Nov-Feb 2014/15 (1.1 mln). So a lot depends on the next payrolls report to show that this was an outlier than a new trend.

Let’s keep in mind the confusing data goes beyond payrolls, way beyond payrolls. I have narrowed my focus because manufacturing PMIs still contain a lot of oil drilling/energy related noise that doesn’t reflect the broader service sector and, most importantly, consumer spending. Consumption is everything in the US economy and even the global economy — which is why it’s a pity consumer spending is so poorly measured in so many places. Even services PMIs can confuse rather than elucidate.

So when looking at payrolls, the key figure really is aggregate payrolls: totaling up payrolls, hours worked and wages to get a sense of how the employment picture is supporting household/consumer spending. That suffered a setback in May but not nearly to the same extent, with aggregate payrolls slipping to 4.5% yr/yr growth from 4.7% and still comfortably above measured consumer spending in the PCE data. From a pure consumption point of view, the weaker employment data is not so much a concern just yet.

And let’s talk about my bete noire this year: PCE-measured consumer spending. As George Pearkes and others have pointed out, monthly retail sales only captures a fraction of real spending and can be quite misleading. PCE is a much more complete picture that feeds directly into GDP. Thus it’s the gauge you really want to watch, away from the PCE price indices that are the Fed’s main inflation target.

What a mess it’s been this year. Even as payroll growth was chugging along and most other indicators suggested the consumer was doing just fine, we had very weak readings on PCE-measured spending. The chart below showing the 2011–15 post-crisis trend tell the story: Jan and Dec tend to be unusually depressed, while the middle of the year sees a bounce back. Weather has clearly played a role in wreaking havoc with seasonal adjustments, while new patterns of consumption tied to Amazon and other factors may mean that spending is happening in different ways over the course of the year than the seasonals assume.

This year has been extra bizarre. Spending stayed weak throughout the first quarter (some auto impact there). The March reading really started make me question my relatively upbeat view on the state of the US consumer given employment/housing/leverage trends. And then spending suddenly exploded higher in April in what was a true outlier effect. Indeed, when averaging out Jan-April this year now you get +0.34% monthly consumption vs +0.32% during the same months of 2011–15. Nothing has really changed!

Just as some pessimists were talking about a US recession, revisions pushed Q1 growth back above 1% and Q2 is still tracking around 2.5%, which means the economy really hasn’t changed its steady if sluggish 2.0–2.5% growth trajectory that it has maintained over much of the post-crisis period.

And I kick myself because a simple look at year-over-year changes in retail sales removing seasonal adjustments would have told you back in Feb that consumer spending was doing just fine and the data was normalizing after a string of unusually subdued readings. The steadiness of the overall PCE spending yr/yr tells you so, even tracking back near the highs of the past few years.

Which is all a long-winded way of saying: try to see through the noise as best as possible, and assume noise is the primary reason for some of the data volatility rather than changing narratives so quickly. Especially right now when so many want to paint a negative story on everything by avoiding any serious investigation of the data.