In August 2020 Singapore's Ministry of Trade and Industry reported that the economy had shrunk by 41.2 per cent in the second quarter. In October, the next release described the same quarter as a 13.2 per cent contraction.
Both numbers are correct. The basis changed between the two releases, and both bases are still published today — updated on the same day, by the same statistical system, roughly four times apart.
What changed, and when
MTI documented the switch in a footnote to the October release:
"With effect from the advance GDP estimates for the third quarter of 2020, MTI will be reporting non-annualised quarter-on-quarter seasonally-adjusted growth rates rather than annualised quarter-on-quarter seasonally-adjusted growth rates. The annualised quarter-on-quarter seasonally-adjusted growth rates will continue to be made available on DOS' website."
Annualising asks what the year would look like if this quarter's pace continued for four quarters. It is the quarterly rate compounded: a 1.4% quarter becomes (1.014)⁴ − 1, or 5.7% a year. In an ordinary quarter it flatters or deepens a number by a factor of about four. In the second quarter of 2020 it turned a severe contraction into a spectacular one.
We located the break by counting the word "annualis" across MTI's own release pages, using "seasonally-adjusted" as a control that should never be zero on a live page:
| Release | "annualis" | control |
|---|---|---|
| 2Q2020 (August) | 4 | 4 |
| 3Q2020 (October) | 0 | 4 |
| every release since | 0 | 2–4 |
The control is never zero, so the zeros are real.
The gap is not all annualisation, and saying it is would be wrong
It is tempting to present −41.2% and −13.2% as one number twice. They differ for two reasons, and only one of them is the basis change.
advance 2Q20, annualised, as published = −41.2%
implied non-annualised = −12.43%
check: (1 − 0.1243)^4 = 0.5881 → −41.2% ✓
revised 2Q20, non-annualised, published = −13.2%⚠️ Annualisation accounts for most of the gap, turning −12.4% into −41.2%, a factor of 3.31. The rest, from −12.4% to −13.2%, is an ordinary data revision between the advance estimate and the revised one. Attributing the whole gap to the basis change would overstate it by about a fifth.
The clean test is to compare the three quarters that appear in both releases. Each reconciles under the (1+r)⁴ − 1 transform to within 0.22 percentage points, confirming the change is purely arithmetic.
Both series are live, today
This is not a historical curiosity about 2020. SingStat publishes both series now, and we fetched them on the same day with the same last-updated stamp:
| Quarter | Annualised | Non-annualised | compounded |
|---|---|---|---|
| 2026 Q2 | 5.7 | 1.4 | 5.72 |
| 2026 Q1 | 5.0 | 1.2 | 4.89 |
| 2025 Q4 | 5.2 | 1.3 | 5.30 |
| 2025 Q3 | 7.8 | 1.9 | 7.82 |
| 2024 Q3 | 12.3 | 2.9 | 12.11 |
Nine quarters, reconciling to within 0.19 percentage points throughout. MTI's own current commentary uses the lower basis: "the economy expanded by 1.4 per cent, extending the 1.2 per cent growth in the first quarter."
So a reader seeing headlines of "Singapore grew 5.7%" and "Singapore grew 1.4%" for the same three months may be looking at two current, official and correctly-computed figures.
The explanatory note is gone
MTI's footnote points readers to an explanatory note for the details. That URL now returns 404.
The returned page is a true 404 rather than a failed fetch: it contains the string "not found" nine times. The terms "annualis" and "seasonally" appear zero times, confirming the explanatory content is gone.
⚠️ That establishes the note is gone, not that it was removed. Singapore government sites migrated platforms during this period and link rot is the ordinary consequence. We are reporting a dead link, not a decision.
What this does not mean
Neither basis is wrong, and neither agency has erred. The non-annualised figure is the more conservative presentation and arguably the better default. It avoids extrapolating a single volatile quarter across a full year, which was the core problem in 2020. MTI moved toward the more cautious number, not away from it.
Publishing both is also right. Annualised rates are what international comparisons often use, and withdrawing the series would have broken every model built on it. Keeping both and labelling them precisely — which SingStat does, in full, in the series names — is the correct answer to a real trade-off.
⚠️ We are not claiming anyone quotes the wrong figure. This piece makes no assertion about how the two numbers are used in the press or in research, because we did not survey it.
What to do with it
When you see a Singapore quarter-on-quarter GDP figure, check which basis it is before comparing it to anything — including to the same country's own past. A 2019 headline and a 2021 headline are not on the same scale, and nothing in either number says so.
The tell is the ratio. If a quarterly growth figure is roughly four times another for the same period, you are almost certainly looking at the two bases rather than a disagreement. Convert with (1+r)⁴ − 1 and see whether they land within a couple of tenths.
If you are pulling the series programmatically, take the table identifier rather than the label: the annualised and non-annualised series have distinct identifiers and nearly identical names, differing by one word.
Where this comes from, and what will date it
The two contradictory descriptions of 2Q2020 are quoted from MTI's own advance-estimate PDFs for the second and third quarters of 2020, at 145,597 and 148,772 bytes. The footnote announcing the change is quoted from the second of those. The break was located by term counts across twelve MTI release pages, each with a control. The current figures are from SingStat's table service, both series fetched the same day and both carrying the same last-updated date. Every compounding check above is arithmetic on those published figures.
⚠️ Two limits. We did not read the explanatory note, because it is gone — so MTI's full reasoning for the change is described here only from the footnote that announced it. And the nine-quarter reconciliation is our own arithmetic on two published series; neither agency publishes a crosswalk asserting that the series are related this way, and we are inferring the relationship from the numbers rather than citing it.
This finding will not date. Both series are maintained, the 2020 releases are permanent, and the transform is simple arithmetic. The only thing that would change it is a public announcement that one series is being discontinued.
View author profile → · Editorial policy
