US labor market
Payrolls Lab first-print scoreboard
How the RoboMacro Payrolls Forecasting Lab scores 10 models against the BLS first print, not the revision. Core ensemble vs street consensus. Replication notes. Free.
What is scored
Ten independent models forecast the BLS Employment Situation. Scoring uses the first print, not later revisions. Street consensus is the Dow Jones / survey benchmark. “We beat street” means the core ensemble was closer to the first print than consensus.
The ten models
Econometric: naive AR, SARIMAX, VAR, unobserved components, MIDAS. Machine learning: ElasticNet, random forest, XGBoost, LightGBM, neural net. Core ensemble excludes VAR and LightGBM (documented corrections, August 2026).
What we do not claim
We do not replace the BLS print or FRED. The lab is a public scoreboard of our forecasts vs the print and vs street. Live: /models/payrolls. Indicator dashboard: /payrolls.
Replication: each model’s Python module is listed on the lab page. Cache JSON at /payrolls_forecasting_cache/.
