Utilizing the correlation between log first variations for BLS nonfarm payroll employment and ADP, I get the next prediction or “nowcast” for the BLS quantity to be launched tomorrow:
Determine 1: BLS nonfarm payroll employment sebries (black), and predicted (gentle blue), +/- 1 std error (grey traces), and Bloomberg consensus of 4/6, all s.a. Supply: BLS, ADP by way of FRED, Bloomberg and writer’s calculations.
I estimate this specification over the 2021M06-2023M02 interval :
Δ pnfp_blst = 0.001 + 0.593 pnfp_adpt + ut
Adj. R2 = 0.63 SER = 0.0010, n = 21, DW = 2.38. (Coefficients important at 5% msl utilizing HAC strong customary errors daring ital.)
Discover a easy autoregression of first differenced non-public NFP yields a significantly decrease adjusted R2 (0.19). Lagged non-public NFP from BLS doesn’t enter considerably if entered into the above equation.
By way of numbers (slightly than percentages), Bloomberg consensus is for a 215K web improve, whereas the mannequin implies 281K. The Bloomberg consensus is the typical of forecasters who’re utilizing a wide range of approaches, a few of which embody disaggregating to particular person classes, forecasting, after which aggregating (as in AS‘s strategy for whole NFP). I take advantage of a less complicated strategy (which I don’t declare is essentially higher).
Under is the underlying knowledge (in ranges).
Determine 2: BLS nonfarm payroll employment sequence (black), and ADP sequence (gentle brown), each in 000’s, s.a. Supply: BEA, ADP by way of FRED.
