Korea’s monthly numbers have a moving denominator
Every monthly figure out of Korea — output, shipments, hours, payroll — is a count of what happened across however many days the workplace was actually open. That count is itself published, and it moves. Across 185 months from 2011-01 to 2026-05, an average February ran 18.5 days against 21.2 days in July — a 14.4% gap before anything economic has happened at all.
The seasonal shape
Averaged over every year we hold, the calendar alone moves the denominator by 14.4%. A month-on-month comparison that ignores it is reading the calendar and calling it the economy.
| Month | Mean operating days | Years counted | Relative |
|---|---|---|---|
| January | 20.2 days | 16 | |
| February | 18.5 days | 16 | |
| March | 20.6 days | 16 | |
| April | 20.6 days | 16 | |
| May | 19.6 days | 16 | |
| June | 20.0 days | 15 | |
| July | 21.2 days | 15 | |
| August | 20.1 days | 15 | |
| September | 19.2 days | 15 | |
| October | 19.5 days | 15 | |
| November | 20.7 days | 15 | |
| December | 20.6 days | 15 |
Bars are scaled from 16 to 22 days so the differences are visible; they do not start at zero, and they are not meant to be read as magnitudes.
The long drift
The annual mean fell from 21.2 days in 2011 to 19.1 days in 2025 — -9.9% over the span. Any Korean monthly series compared across those years is being divided by a number that moved by roughly a tenth. We do not say why it moved. Working-hour law, substitute public holidays and survey redesign are all plausible; none of them is something we measured, so none of them appears here as a cause.
| Year | Mean operating days | Months | Complete |
|---|---|---|---|
| 2011 | 21.2 days | 12 | Yes |
| 2012 | 21.0 days | 12 | Yes |
| 2013 | 20.8 days | 12 | Yes |
| 2014 | 20.6 days | 12 | Yes |
| 2015 | 20.6 days | 12 | Yes |
| 2016 | 20.4 days | 12 | Yes |
| 2017 | 20.1 days | 12 | Yes |
| 2018 | 20.0 days | 12 | Yes |
| 2019 | 20.0 days | 12 | Yes |
| 2020 | 19.7 days | 12 | Yes |
| 2021 | 19.8 days | 12 | Yes |
| 2022 | 19.6 days | 12 | Yes |
| 2023 | 19.4 days | 12 | Yes |
| 2024 | 19.2 days | 12 | Yes |
| 2025 | 19.1 days | 12 | Yes |
| 2026 | 18.8 days | 5 | Partial year |
1 of these years hold fewer than twelve months. They are marked rather than averaged in as though complete — a partial year is a different object from a full one, and putting them in one column without saying so is how a table starts lying.
The series breaks — and the break has two sizes
The statistical office renumbers this survey periodically, and at each changeover the published figure steps. The source states a change rate at the joint. That rate compares December with January — two months that are never alike, because January is structurally one of the thinnest months of the year. So the stated step carries the season inside it.
Below, each break is measured twice: once as the source states it, and once as the mean of the twelve months either side, which removes the season. Both are true. They answer different questions, and publishing only the larger one would make the break look worse than it is — publishing only the smaller one would hide what a reader of the raw file actually sees.
| Break | As stated (month on month) | Twelve-month means | Months either side |
|---|---|---|---|
| 2019-12 → 2020-01 | -4.5% | -1.2% | 12 / 12 |
| 2025-12 → 2026-01 | -3.0% | -1.7% | 12 / 5 |
Where fewer than twelve months have accumulated after a break, we average what exists and print the count rather than waiting or padding. The most recent break has only a few months on the far side, and that figure will move as more arrive.
What this page does not tell you
- Why the count fell. We measured the movement, not its cause. A page that counted days is not entitled to an explanation it did not test.
- A spliced series. We could publish one number running cleanly from 2011-01 to 2026-05 by applying a correction factor at each break. We do not, because that number would be ours rather than the statistical office's, and a reader could not check it against the source.
- How to adjust your own series. Dividing by operating days is one choice among several, and the right one depends on what you are measuring. We draw the terrain; the needle is yours to set.
- Anything about one firm or one sector. This is a survey-wide figure. A single workplace's calendar is not in here.
Source: 국가데이터처 KOSIS · 고용노동부 「사업체노동력조사」. Collected 2026-08-29. Every figure on this page is computed at build time from workday-denominator.json by scripts/measure-workday-denominator.mjs; none is transcribed by hand. Months we could not read are dropped rather than filled with zero, and the counts above say how many went into each average.