Made 2026-10-07 12:37 UTC, once a day
Forecast.
The next number of every index and the range it should land in 8 times in 10. Display only: a forecast never touches a settled number. How it works
Next prints
Plum: the next range · gray: past ranges · line: what printed
$LAYOFFSWeek to Oct 3157,824136,300 to 173,598The median of all models$YESNOOct 4$797.63M$683.66M to $1.02BThe same day a week before$TRENCHESOct 7$4.72M$4.05M to $5.77MThe median of all models$MINTSOct 754,01748,393 to 63,249The median of all models$GIGSOct 763,88547,416 to 95,819The 7-period median$WAGEOct 7$0.0093$0.0052 to $0.0180Exponential smoothing$JOBSOct 202647,211-115,812 to 163,321Exponential smoothing$UNEMPOct 20264.24.1 to 4.3The last number$CPISep 20260.40.2 to 0.5The median of all models$CORECPISep 20260.20.2 to 0.3The median of all models
Backtest
The whole daily procedure walked forward: at each past period it chose a model and forecast from earlier periods only. Inside: how often the number fell in the range (80% is right). The procedure was set after this backtest was seen, so the live record below is the clean test.
| Index | Model today | Periods | Inside | Against the last number |
|---|---|---|---|---|
| $LAYOFFS | The median of all models | 88 | 77% | 4% more error than the last number |
| $YESNO | The same day a week before | 116 | 81% | No better than the last number |
| $TRENCHES | The median of all models | 119 | 79% | 1% more error than the last number |
| $MINTS | The median of all models | 33 | 73% | No better than the last number |
| $GIGS | The 7-period median | 11 | 73% | 25% less error than the last number |
| $WAGE | Exponential smoothing | 11 | 64% | 11% less error than the last number |
| $JOBS | Exponential smoothing | 13 | 77% | 32% less error than the last number |
| $UNEMP | The last number | 12 | 67% | 34% more error than the last number |
| $CPI | The median of all models | 10 | 70% | 19% less error than the last number |
| $CORECPI | The median of all models | 10 | 70% | 29% less error than the last number |
Live record
First numbers land this week
Every forecast is written to a public log before its number lands and never rewritten; the first one made for a period is the one that counts. The log on GitHub
No forecast has met its number yet. The first daily ones are graded the day after they print.