After rest, ADP released today A new series (co-authored with the Stanford Digital Economy Lab) that attempts to independently measure private nonfarm payrolls rather than forecast the BLS series. How will it compare to the old series (and the BLS series) over time?
First, we have at least two measures for each series of samples:
figure 1: Private nonfarm payrolls from BLS (blue), ADP (tan), ADP, and Stanford Digital Economy Lab (green) are all in the 000s, peak-to-trough dates for recessions as defined by the National Bureau of Economic Research are in gray shadow. Source: BLS, ADP via FRED, ADPand NBER.
Note that the new ADP series and the BLS series are quite different in level, reflecting that the new ADP series is not for the BLS series (technical description here). In one sentence:
Based on anonymized and aggregated wage data for more than 25 million U.S. employees, this independent measure will detail nonfarm private sector employment changes for the month and provide weekly employment data for the previous month. Data will be disaggregated by industry, business size, and U.S. Census tract. Historical data on monthly and weekly frequencies for the past 12 years will be benchmarked and available at launch.
How does each series change look like? It is difficult to discern the pattern of the chain change, so I show the chain change (in logarithmic form) in Figure 2.
figure 2: Three-month change (not annualized) in log private nonfarm payrolls from BLS (blue), ADP (tan), ADP and Stanford Digital Economy Lab (green), NBER-defined recession date from peak to Valleys are shaded in gray. Source: BLS, ADP via FRED, ADPNBER, and the authors’ calculations.
During 2010-22M05, the adjusted R2 The regression between the old ADP series and the BLS series is about 0.98, while the regression between the new ADP series and the BLS series is about 0.78. This high correlation is largely due to the recession in 2020. During 2020M03-2020M12 omitted, the corresponding R2 are 0.42 and 0.13, respectively. During 2021M01-22M05, the adjusted R2 is 0.01 and -0.03, respectively. In other words, it would be unwise to infer what happened to BLS measurements from ADP measurements, even in previous series.
Therefore, Goldman Sachs kept its forecast (+325K) unchanged for the relatively small increase (132K).
Below is the recent history of the series, as well as the Bloomberg consensus as of today.
image 3: Private nonfarm payrolls from BLS (blue), from ADP (tan), from ADP and Stanford Digital Economy Lab (green), Bloomberg consensus as of Aug. 30, assuming July numbers (sky blue Squares) unrevised, all 000, sa NBER-defined peak-to-valley decline dates are shaded in grey. Source: BLS, ADP via FRED, ADPBloomberg (accessed 8/30), NBER, and author’s calculations.
While the revised ADP series may not tell us what the BLS series on private nonfarm payrolls will be, it still tells us that job growth has slowed for companies covered by the ADP. ADP data can also tell us which companies are increasing their wages and how wages are changing (see release).
