Employment (NFP, private NFP) slightly out of consensus As far as I know, a recession isn’t here yet. It also doesn’t seem to appear in H1.
First consider nonfarm payrolls measured in three ways:
figure 1: Reported nonfarm payrolls (black), Bloomberg consensus as of Sept. 1 (red squares), nonfarm payrolls with preliminary benchmark revisions (turquoise), and civilian employment adjusted for nonfarm payrolls concept (yellow-green) , both in the 000s, the Bloomberg consensus assumes no revisions to data from previous months. Source: BLS via FRED, BLScalculations by Bloomberg and the authors (see postal).
While the NFP series was revised downwards in the first two months, it has continued to rise in the months when some observers are claiming a recession. One could blame the agency investigation on which the NFP series was based, due to the birth/death model error. In fact, it is the accusation that the birth/death model underestimates firm growth (and therefore employment level estimates during the GW Bush period) that partly drives the development of household survey-based measures adjusted to the NFP concept.This issue is discussed in detail here.
Another point – benchmarking of employment levels involves using the quarterly census of employment and wages, which in turn should explain an almost accurate number of firms in the economy. The preliminary benchmark raises employment levels for March 2022, which I have included in the NFP series (as an implied benchmark) in the chart above.I added growth over the past 12 months and then assumed growth in NFP reported from April (see this postal).
Since the household survey measurement of employment does not involve the firm’s birth/death model, the series should be immune to this type of erroneous measurement (however, the measurement error will be larger due to the smaller sample size). On a month-on-month basis (ie, a 3-month change in monthly data), no series recorded negative growth. Only one series (household series) saw negative month-on-month growth (April), with an annualized growth rate of -0.5%. Taken together, the standard deviation of the annualized m/m change for the period Jun-22, 2021 is approximately 3.9 percentage points (all percentages are in logarithmic form).
What about a completely independent measure of employment growth?The newly revised ADP-Stanford Digital Economy Lab metric is growing at a slower rate than the BLS series (see Discussion here). However, it is still rising.
figure 2: Reported private nonfarm payrolls (black), Bloomberg consensus on BLS series (red squares), ADP-Stanford Digital Economy Lab nonfarm payrolls (light blue), all in the 000s, pre-Bloomberg consensus assumptions Data for several months has not been revised. Source: BLS via FRED, ADPcalculations by Bloomberg and the authors (see postal).
The annualized growth rate for the BLS series was 3.5% compared to 2.6% for ADP-Stanford.
Finally, what about the total working hours. Here are images from the series, which includes images adjusted for preliminary benchmarks.
image 3: Reported total weekly hours worked index (black), including preliminary benchmark revisions (turquoise), 2002=100, sa Source: BLS via FRED, and author’s calculations (see postal).
While total hours worked per week has not been increasing, it has generally increased over the past year. Interestingly, there was a big jump in January 2022, a period in which some observers believe a recession is starting.
Finally, does the Sahm rule indicate that a recession has begun (this is often interpreted as a coincidence indicator, not a leading indicator).
Figure 4: The three-month moving average unemployment rate minus the previous year’s minimum unemployment rate (3.5%) (black) and the trigger value of 0.5% (red dotted line), both in %. Source: BLS by FRED and author’s calculations.
Note that we are far from the 0.5% trigger value in the graph above. Neither do we.
