ADP private nonfarm payrolls unexpectedly rose by 497k, compared to consensus of +228k. For context, the m/m standard deviation of the log first difference is 0.0028, while the June reading is 0.0039.
Using the correlation between BLS non-farm payrolls and the log first difference of ADP, I have the following forecast or “instant forecast” for tomorrow’s BLS data release, literally 486,000 more payrolls (ignoring the corrections from previous months):
figure 1: BLS nonfarm payrolls series (black), forecast (light blue), +/- 1 standard error (gray +) and Bloomberg consensus 7/6 (red squares), all sa Source: BLS, ADP via FRED, Bloomberg and the author’s calculations.
My estimated specs for the period 2021M06-2023M05:
Δ pnfp_blst = 0.001 + 0.643 Δpnfp_adpt + yout
Adjustment.right2 = 0.66 SER = 0.0010, n = 21, DW = 2.24. (using HAC robust standard errors, coefficients significant at 5% msl bold italic.)
Note that simple autoregression of first-difference private NFP (ie ARIMA(1,1,0) for log NFP) yields a rather low adjusted R2 (0.30). If the lagged private NFP of the BLS is entered into the above equation, there is no significant effect.
That being said, ADP-based forecasts have outpaced actual changes from the Bureau of Labor Statistics over the past four months. GS argues that ADP seasonal estimates distort the number, which is estimated at 225k (Bloomberg consensus for private NFPs is 218k).
