MACROGAUGE
OK: Self-test 35/35

Forecast Scoreboard graded in public

Vintage-true MAE
0.29pp
110 BT observations · 3-month-average benchmark, not the live model
Naive MAE
0.25pp
Last known monthly print
Live grades
3
1 pending

Live grades — real-time calls, receipts included

Export data
↓ JSON
PrintBadgeForecast MoMActual MoMErrorCalled onGraded on
2026-08LIVE0.23%0.32%−0.09pp2026-08-272026-09-11
2026-07LIVE0.03%-0.01%+0.04pp2026-08-112026-08-12
2026-06LIVE-0.16%-0.35%+0.19pp2026-07-112026-07-14
2026-09pending0.43%——2026-10-02—

Signed error = forecast − actual (positive = ran hot). Calls freeze at their as-of date and grade automatically when the print lands — nothing is revised after the fact.

Head to head — Macrogauge vs Cleveland Fed vs Kalshi

SA, first release; last value each forecaster published before the release, over the last 12 prints. Ensemble weights are equal until every forecaster has 6 graded prints.
ForecasterPrintsMAE (pp)Bias (pp)
Macrogauge30.1050.035
Kalshi30.1140.040
Cleveland Fed30.1380.114
MonthActual (SA)MacrogaugeCleveland FedKalshi
2026-080.40%0.29%*0.36%0.32%
2026-070.07%0.11%*0.09%0.04%
2026-06-0.42%-0.25%*-0.06%-0.19%

* Macrogauge calls recorded before 2026-09-26 were not seasonally adjusted; shown converted with the same BLS seasonal factors the live model now uses.

Walk-forward backtest — vintage-true history

Export data
↓ JSON
MonthBadgeVintage cutoffForecastNaive (carry-fwd)ActualErrorvs naive
2026-08BT2026-09-100.09%-0.01%0.32%-0.23ppbeat
2026-07BT2026-08-110.38%-0.35%-0.01%0.39pplost
2026-06BT2026-07-130.84%0.63%-0.35%1.19pplost
2026-05BT2026-06-090.79%0.85%0.63%0.16ppbeat
2026-04BT2026-05-110.63%1.05%0.85%-0.22pplost
2026-03BT2026-04-090.27%0.47%1.05%-0.78pplost
2026-02BT2026-03-100.20%0.37%0.47%-0.27pplost
2026-01BT2026-02-120.17%-0.02%0.37%-0.20ppbeat
2025-12BT2026-01-120.23%0.25%-0.02%0.25ppbeat
2025-09BT2025-10-230.26%0.29%0.25%0.01ppbeat
2025-08BT2025-09-100.23%0.15%0.29%-0.05ppbeat
2025-07BT2025-08-110.29%0.34%0.15%0.14ppbeat
2025-06BT2025-07-140.25%0.21%0.34%-0.09ppbeat
2025-05BT2025-06-100.33%0.31%0.21%0.12pplost
2025-04BT2025-05-120.44%0.22%0.31%0.13pplost
2025-03BT2025-04-090.38%0.44%0.22%0.15ppbeat
2025-02BT2025-03-110.21%0.65%0.44%-0.23pplost
2025-01BT2025-02-110.03%0.04%0.65%-0.62pplost
2024-12BT2025-01-140.07%-0.05%0.04%0.04ppbeat
2024-11BT2024-12-100.12%0.12%-0.05%0.17pplost
2024-10BT2024-11-120.12%0.16%0.12%0.00ppbeat
2024-09BT2024-10-090.08%0.08%0.16%-0.08pplost
2024-08BT2024-09-100.11%0.12%0.08%0.02ppbeat
2024-07BT2024-08-130.20%0.03%0.12%0.08ppbeat

BT rows are vintage-true walk-forward values frozen the day before each release — the model never sees data it wouldn't have had. The backtested model is a three-month average of previously known official prints — a long-history benchmark, not the live bottom-up nowcast graded in the table above (which is too young to backtest vintage-true).

Also graded — PCE

PrintBadgeForecast MoMActual MoMErrorCalled onGraded on
2026-08LIVE0.34%0.31%+0.03pp2026-09-292026-09-30
2026-07LIVE0.11%0.16%−0.05pp2026-08-122026-08-26
2026-06LIVE0.08%-0.11%+0.19pp2026-07-142026-07-30
2026-09pending0.35%——2026-10-02—

Same freeze-and-grade rules as CPI above: forecast is MoM % on the PCE price index, graded against the first print when it lands.

Also graded — NFP

PrintBadgeForecast (k jobs)Actual (k jobs)ErrorCalled onGraded on
2026-09LIVE+91k+29k+62k2026-10-012026-10-02
2026-08LIVE+31k+162k−131k2026-09-032026-09-04
2026-07LIVE+137k−23k+160k2026-07-302026-08-07
2026-10pending+69k——2026-10-02—

NFP calls are monthly payroll changes in thousands of jobs, not percentages; signed error = forecast − actual, also in thousands. Same freeze rules — nothing is revised after the fact.