Using the correlation between log first differences for BLS nonfarm payroll employment and ADP, I get the following prediction or “nowcast” for the BLS number to be released tomorrow:

(Click on image to enlarge)

Figure 1: BLS nonfarm payroll employment sebries (black), and predicted (light blue), +/- 1 std error (gray lines), and Bloomberg consensus of 4/6, all s.a. Source: BEA, ADP via FRED, Bloomberg and author’s calculations.

I estimate this specification over the 2021M06-2023M02 period :

Δ pnfp_bls_t = 0.001 – 0.593 pnfp_adp_t + u_t

Adj. R² = 0.63 SER = 0.0010, n = 21, DW = 2.38. (Coefficients significant at 5% msl using HAC robust standard errors bold ital.)

Notice a simple autoregression of first differenced private NFP yields a considerably lower adjusted R2 (0.19). Lagged private NFP from BLS does not enter significantly if entered into the above equation.

In terms of numbers (rather than percentages), Bloomberg consensus is for a 215K net increase, while the model implies 281K. The Bloomberg consensus is the average of forecasters who are using a variety of approaches, some of which include disaggregating to individual categories, forecasting, and then aggregating (as in AS‘s approach for total NFP). I use a simpler approach (which I do not claim is necessarily better).

Below is the underlying data (in levels).

(Click on image to enlarge)

Figure 2: BLS nonfarm payroll employment series (black), and ADP series (light brown), both in 000’s, s.a. Source: BEA, ADP via FRED.

More By This Author: