Q1 2024
+0.5pp
2.5% first release → 3.0% later vintage.
Research Release 003 · 17 August 2026
The Revision Economy
Economic history is often read from the latest database. Release 003 instead compares what was first published with what later official vintages say, and asks whether the superseded evidence remains recoverable.
10
verified revision pairs across Morocco, Jordan and Saudi Arabia, covering headline GDP, sector composition and seasonally adjusted quarterly growth.
Finding 01 · Morocco’s 2024 growth profile moved upward
First quarterly releases compared with the values later reported for the same quarters.
Q2 2024
+0.6pp
2.4% → 3.0%.
Q3 2024
+0.7pp
4.3% → 5.0%.
Q4 2024
+0.5pp
3.7% → 4.2%.
Annual 2024
+0.6pp
3.8% provisional annual growth → 4.4% in the next annual accounts vintage.
Interpretation
The narrative changes.
A researcher using only the latest vintage would reconstruct a stronger 2024 than the one visible in the releases available at the time.
Finding 02 · A stable headline can hide a revision
Saudi Arabia, Q3 2024: flash estimate versus full national-accounts publication.
| Metric | Flash | Full | Revision | What changes? |
|---|---|---|---|---|
| Headline real GDP y/y | 2.8% | 2.8% | 0.0pp | Headline unchanged. |
| Oil activities y/y | 0.3% | 0.05% | -0.25pp | Oil contribution weaker than flash suggested. |
| Non-oil activities y/y | 4.2% | 4.3% | +0.1pp | Non-oil contribution stronger. |
| Seasonally adjusted GDP q/q | 0.8% | 0.9% | +0.1pp | Quarterly momentum stronger. |
The lesson is not that flash estimates are “wrong.” It is that revision analysis must inspect the composition as well as the headline.
Finding 03 · Jordan exposes a comprehensive revision break
Original Q3 2024 release
2.6%
Published growth at constant prices versus Q3 2023.
Later comparator
2.5%
The Q3 2025 release cites Q3 2024 at 2.5%.
Revision architecture
2008→
DoS explicitly states that a new comprehensive revision was applied to quarterly GDP estimates from 2008 through Q2 2025.
Vintage availability audit
Can a researcher recover both the old statement and the later statement?
| Country | Old releases | Later releases | Revision notice | Current structured series | Frozen machine-readable old vintage |
|---|---|---|---|---|---|
| Morocco | Verified | Verified | Partial | Verified | Unverified |
| Jordan | Verified | Verified | Verified | Verified | Unverified |
| Saudi Arabia | Verified | Verified | Partial | Verified | Unverified |
“Unverified” is not coded as “absent.” The audit only records what was directly established in the source perimeter.
Why this matters
Backtests can cheat accidentally.
A model trained on the latest revised GDP series may use information unavailable when a historical forecast or policy choice was actually made.
Policy narratives can drift.
“Growth was weak” or “growth was accelerating” can change when historical quarters are revised after the event.
Archives are part of data quality.
If first releases disappear or are overwritten, researchers cannot reconstruct real-time information sets even when the current series is high quality.
Research standard
Each revision pair requires two recoverable official-source statements referring to the same statistical object. Revision is calculated as later-vintage value minus first-release value in percentage points. The release does not infer why a number changed unless the publishing institution documents the cause.
Suggested citation: MENA Open Data & Evidence Lab (2026). Research Release 003: The Revision Economy. Version 1.0, 17 August 2026.