USMacro

The Revisions Ledger

How much should you trust the number that was just published — and which way is it likely to move? Every first print here is set against the value three years later, from the archived record. The economy reads Expansion as of October 2026.

THE READ

The first version of an economic number is an estimate. This page shows those first estimates against what they turned out to be.

Every number here is compared with the value that stood three years later, read from the archive of what was actually published at the time — not from a later account of it.

How first prints like today's were later revised

The economy is in an Expansion right now (October 2026). When the economy has been in an Expansion before, here is how the first estimate of each number changed once the data settled.

  • Payrolls (the month's change in jobs, in thousands) was first reported as +165k for Oct. Of the 390 times it was first reported during an Expansion, the middle move was +22k, with half landing between −28k and +77k. It was revised up 62% of the time — 242 of the 389 that moved at all.
  • Personal income (the month's change, in per cent) was first reported as +0.20 for Sep. Of the 389 times it was first reported during an Expansion, the middle move was +0.06, with half landing between −0.09 and +0.21. It was revised up 65% of the time — 202 of the 313 that moved at all.
  • Retail sales (the month's change, in per cent) was first reported as +0.77 for Sep. Of the 203 times it was first reported during an Expansion, the middle move was +0.07, with half landing between −0.22 and +0.36. It was revised up 58% of the time — 106 of the 182 that moved at all.
  • Industrial production (the month's change, in per cent) was first reported as +0.66 for Sep. Of the 389 times it was first reported during an Expansion, the middle move was −0.02, with half landing between −0.24 and +0.24. It was revised up 48% of the time — 167 of the 351 that moved at all.
  • Core PCE prices (the month's change in prices, in per cent) was first reported as +0.32 for Sep. Of the 207 times it was first reported during an Expansion, the middle move was +0.02, with half landing between −0.02 and +0.06. It was revised up 80% of the time — 65 of the 81 that moved at all.
  • Real GDP growth (the quarter's growth, annualised, in per cent) was first reported as +3.37 for Q3. Of the 102 times it was first reported during an Expansion, the middle move was +0.04, with half landing between −0.61 and +0.82. It was revised up 51% of the time — 49 of the 96 that moved at all.
  • Corporate profits after tax (the quarter's change, in per cent) was first reported as +8.22 for Q2. Of the 128 times it was first reported during an Expansion, the middle move was −0.09, with half landing between −2.75 and +2.12. It was revised up 49% of the time — 62 of the 126 that moved at all.
  • Gross domestic income (the quarter's growth, annualised, in per cent) was first reported as +8.77 for Q2. Of the 40 times it was first reported during an Expansion, the middle move was +0.60, with half landing between −1.25 and +1.56. It was revised up 60% of the time — 24 of the 40 that moved at all.

These are counts of what happened before, not odds on what happens next. Nothing here says what this month's number will turn out to be.

What the record says about revisions

  • Which way a number gets revised depends on what the economy is doing. Payrolls has been revised upward 62% of the time when the economy was in an Expansion, and only 27% of the time in a Contraction. That is a large enough gap to be confident it is real. And it is useful while it matters, because you can see which stage the economy is in today.
  • Revisions are no bigger around turning points. We expected first estimates to miss by more when the economy changes direction, and tested it against the 5 recessions in the record. On 7 of 8 series, the typical revision near a turn was no larger than at any other time. What changes with the economy is which way numbers get revised, not how much.
  • A string of downward revisions is not a warning sign. If it were, it would be worth watching, because revisions are published the day they happen. We tried 4 versions of the idea and all 4 failed: the best one raised the alarm 63 times to catch 8 of the 10 turns in the record. A turn here is a month the site's own stage left Expansion — 10 of them on the record, against the 5 recessions, whose starts and ends are the turns of the test above.

Measured September 2026. The full record →

The dossiers

In each dossier, the stage table is sorted by what the economy was doing at the time — using the stage as it was read then, not as it was judged later — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much). Today's stage is highlighted.

Payrolls FRED: PAYEMS · vintages: ALFRED jobs · m/m Δk newest first estimate: 2026-10 settled up to: 2023-10 record starts 1955

How far it usually moves (the month's change in jobs, in thousands): the middle move is +19k, half of all moves fall between −38k and +82k, and the average size of a move, ignoring direction, is ±80k. The largest was 2020-03, which went −701k → −1,427k.

Which way it moves: of the 821 first estimates on record since 1955, 483 were later revised up, 336 down and 2 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion390+22k[−28k, +77k]62% of the time
Slowdown40−20k[−66k, +46k]42% of the time
Contraction52−50k[−131k, +8k]27% of the time
Recovery25−25k[−90k, +1k]24% of the time
Personal income FRED: PI · vintages: ALFRED nominal · m/m % newest first estimate: 2026-09 settled up to: 2023-09 record starts 1966

How far it usually moves (the month's change, in per cent): the middle move is +0.07, half of all moves fall between −0.10 and +0.21, and the average size of a move, ignoring direction, is ±0.23. The largest was 1993-01, which went +0.47 → −3.83.

Which way it moves: of the 692 first estimates on record since 1966, 368 were later revised up, 218 down and 106 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion389+0.06[−0.09, +0.21]65% of the time
Slowdown40+0.08[−0.12, +0.21]65% of the time
Contraction52−0.13[−0.31, +0.12]34% of the time
Recovery25−0.05[−0.27, +0.25]45% of the time
Retail sales FRED: RSAFS · vintages: ALFRED nominal · m/m % newest first estimate: 2026-09 settled up to: 2023-09 record starts 2001

How far it usually moves (the month's change, in per cent): the middle move is +0.06, half of all moves fall between −0.27 and +0.37, and the average size of a move, ignoring direction, is ±0.41. The largest was 2022-01, which went +3.75 → +1.82.

Which way it moves: of the 267 first estimates on record since 2001, 139 were later revised up, 105 down and 23 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion203+0.07[−0.22, +0.36]58% of the time
Slowdown17+0.06[−0.26, +0.28]56% of the time
Contraction31+0.07[−0.47, +0.57]53% of the time
Recovery16−0.02[−0.38, +0.43]50% of the time
Industrial production FRED: INDPRO · vintages: ALFRED index · m/m % newest first estimate: 2026-09 settled up to: 2023-09 record starts 1928

How far it usually moves (the month's change, in per cent): the middle move is +0.00, half of all moves fall between −0.27 and +0.34, and the average size of a move, ignoring direction, is ±0.48. The largest was 1952-06, which went −5.61 → −0.84.

Which way it moves: of the 1149 first estimates on record since 1928, 514 were later revised up, 475 down and 160 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion389−0.02[−0.24, +0.24]48% of the time
Slowdown40+0.10[−0.21, +0.32]59% of the time
Contraction52+0.01[−0.32, +0.27]53% of the time
Recovery25−0.01[−0.28, +0.34]48% of the time
Core PCE prices FRED: PCEPILFE · vintages: ALFRED prices · m/m % newest first estimate: 2026-09 settled up to: 2023-09 record starts 2000

How far it usually moves (the month's change in prices, in per cent): the middle move is +0.02, half of all moves fall between −0.02 and +0.06, and the average size of a move, ignoring direction, is ±0.05. The largest was 2001-09, which went −0.90 → −0.51.

Which way it moves: of the 277 first estimates on record since 2000, 90 were later revised up, 25 down and 162 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion207+0.02[−0.02, +0.06]80% of the time
Slowdown20+0.03[−0.01, +0.08]78% of the time
Contraction34+0.01[−0.03, +0.06]62% of the time
Recovery16+0.04[−0.01, +0.11]89% of the time
Real GDP growth FRED: GDPC1 · vintages: ALFRED real · SAAR newest first estimate: 2026-07 settled up to: 2023-07 record starts 1992

How far it usually moves (the quarter's growth, annualised, in per cent): the middle move is −0.03, half of all moves fall between −0.90 and +0.78, and the average size of a move, ignoring direction, is ±1.03. The largest was 2008-10, which went −3.80 → −8.89.

Which way it moves: of the 126 first estimates on record since 1992, 58 were later revised up, 62 down and 6 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion102+0.04[−0.61, +0.82]51% of the time
Slowdowntoo few to say (8)
Contractiontoo few to say (11)
Recoverytoo few to say (5)
Corporate profits after tax FRED: CPATAX · vintages: ALFRED nominal · % level q/q newest first estimate: 2026-04 settled up to: 2023-04 record starts 1981

How far it usually moves (the quarter's change, in per cent): the middle move is +0.01, half of all moves fall between −2.67 and +2.60, and the average size of a move, ignoring direction, is ±3.65. The largest was 2009-01, which went +1.11 → +29.39.

Which way it moves: of the 170 first estimates on record since 1981, 84 were later revised up, 83 down and 3 not at all.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion128−0.09[−2.75, +2.12]49% of the time
Slowdown15−0.33[−3.13, +1.03]43% of the time
Contraction17−0.82[−5.42, +3.08]47% of the time
Recoverytoo few to say (8)
Gross domestic income FRED: GDI · vintages: ALFRED nominal · SAAR newest first estimate: 2026-04 settled up to: 2023-04 record starts 2013

How far it usually moves (the quarter's growth, annualised, in per cent): the middle move is +0.66, half of all moves fall between −1.23 and +1.85, and the average size of a move, ignoring direction, is ±1.73. The largest was 2023-01, which went +1.70 → +7.05.

Which way it moves: of the 41 first estimates on record since 2013, 25 were later revised up and 16 down.

Sorted by the stage at the time
stagetimestypical movemiddle halfwent up
Expansion40+0.60[−1.25, +1.56]60% of the time
Slowdowntoo few to say (0)
Contractiontoo few to say (1)
Recoverytoo few to say (0)

The biggest revisions on record — all 184 of them, each one larger than 95% of the moves its own number has ever made — are listed in the methodology.

How long the data took to admit a recession

recessionreal GDPpayrolls
1990by 1992-07-30 (floor-bounded: GDPC1 archive begins 1991-12)1990-08-03 (Jul −219k)
20012002-07-31 (−0.60) — 16 months late; NOMINAL GDP never admitted it at any vintage (OBS-C1-10)2001-04-06 (Mar −86k)
20082009-07-31 (−0.72) — 18 months late2008-02-01 (Jan −17k)
20202020-04-29 (Q1 advance −4.78) — the only real-time admission2020-04-03 (Mar −701k)

Every first estimate here is compared with the value three years later, taken from the ALFRED archive. It is a record of what has happened, counted — not a forecast of what will. How it's measured →