What a base effect is — why half of '68.7% export growth' is a fact about last year
A base effect is what happens to a growth rate when the period it is measured against was itself abnormal. Every year-on-year growth rate is a fraction that divides by last year, so if last year was unusually low the same increase produces a much larger percentage, and if last year was unusually high a normal year prints as a decline. South Korea's August 2026 trade figures contain both cases in a single release. Total exports of 98.25 billion dollars rose 68.7 percent and semiconductor exports of 46.65 billion rose 209.0 percent, because August 2025 fell during a semiconductor downturn — working backwards puts that month at roughly 58.2 billion and 15.1 billion respectively. In the same release car exports fell, not because sales weakened but because the main manufacturers' summer shutdown shifted from late July 2025 into early August 2026, so the comparison month was unusually productive. The way past this is to read the absolute level rather than the rate, compare with the previous month, and know when the base resets
The three lines
- Definition — a growth rate distorted because the comparison period, the denominator, was abnormally low or high
- Both directions — in one release, chips at +209% came from a low base and cars fell on a high base
- How to read — use absolute levels and month-on-month changes, and know the month the base resets
Key questions
- What does base effect mean
- **It means the comparison period is abnormal, so the growth rate is not describing what it appears to describe.** Start with the formula: **year-on-year growth = (this period − same period last year) ÷ same period last year.** Last year is the denominator. **A small denominator turns an ordinary increase into a large percentage.** Concretely: going from 100 to 200 is +100 percent, while going from 400 to 500 is +25 percent. **Both rose by exactly 100.** The growth rates differ fourfold because of where each started, not because of what happened. That is a base effect. It is not a lie — the arithmetic is correct — but roughly half the number is a statement about the past rather than the present. It also runs in both directions, and the high-base case causes more confusion: if last year was exceptional, a perfectly normal year prints as a decline, and the headline reads like deterioration when nothing deteriorated.
- How inflated is Korea's 68.7 percent export growth
- **Not inflated so much as half-owned by 2025.** Reverse the arithmetic and the starting line appears. Dividing 98.25 billion dollars by 1.687 puts **August 2025 exports at about 58.2 billion.** Dividing semiconductors' 46.65 billion by 3.09 gives roughly **15.1 billion.** August 2025 sat inside a semiconductor downturn, and that low position is what produced a 209 percent multiple. **The result is still real.** The 46.65 billion dollar semiconductor figure is a **record monthly total in absolute terms**, and the category has now cleared 40 billion for three consecutive months — neither claim depends on what last year looked like. The practical rule is to separate the two sentences in the release. *Record monthly total* is a fact about now. *209 percent growth* is half a fact about then. They come from the same document but they do not carry the same weight.
- When does a base effect end
- **When the comparison period returns to normal — and the growth rate drops sharply at that moment.** If Korean semiconductor exports began recovering in the second half of 2025, then from the second half of 2026 onward the comparison is no longer low, and **the same sales volume prints a far smaller growth rate.** The headline that follows is 'export growth slows,' which is ambiguous on its face. Two very different situations produce it: **exports actually fell** (this month's absolute figure is below last month's) or **last year caught up** (this month's absolute figure is flat or higher). **The only way to tell them apart is the absolute level.** The same trap recurs in inflation data. Once the calendar passes a month when oil spiked, the inflation rate falls on its own even if oil has not moved at all, because the denominator rose. Reading that as 'inflation has been beaten' is a mistake with real consequences for rate expectations.
- How is a base effect different from seasonal adjustment
- **Both are about making comparisons fair, but they correct different distortions and only one can be corrected automatically.** **Seasonality** is a regular, repeating pattern — holiday retail in December, summer shutdowns in manufacturing, festival spending. Because it repeats, statistical agencies can model and remove it, which is why US payrolls and retail sales are published seasonally adjusted. **A base effect comes from a one-off event** — a strike, a natural disaster, an industry downturn, a schedule change. One-off events have no pattern, so no model can subtract them. **The reader has to know and correct for them manually.** Korea's August 2026 car export decline sits exactly on this boundary. A summer shutdown is a seasonal factor, but the shutdown **moving** from late July one year to early August the next is not a repeating pattern — it is a one-off shift. That is why it did not wash out in seasonal adjustment and instead appeared as a base effect.
A growth rate is always a fraction. And the bottom of that fraction is always last year.
Year-on-year growth = (this period − same period last year) ÷ same period last year
A base effect is what happens when that bottom half — the comparison period — was itself abnormal, so the resulting percentage stops describing what changed.
1. Same increase, four times the growth rate
One table settles the concept.
| Case | Last year | This year | Increase | Growth |
|---|---|---|---|---|
| A | 100 | 200 | +100 | +100% |
| B | 400 | 500 | +100 | +25% |
Both grew by exactly 100. The growth rates differ fourfold, and the difference was made by last year, not this year.
Call A "explosive growth" and B "a slowdown" and neither sentence is technically false — but the reader ends up with a picture that does not match reality.
The reason base effects cause so much trouble is that they run both ways. A low base inflates; a high base makes a normal year read as decline.
2. One release, both directions
Korea's August 2026 trade data is a textbook case, because a low base and a high base appear in the same document.
| Category | August 2026 | Growth | August 2025 (derived) | Base |
|---|---|---|---|---|
| Total exports | $98.25bn | +68.7% | about $58.2bn | Low |
| Semiconductors | $46.65bn | +209.0% | about $15.1bn | Low |
| Cars | (amount not published) | decline | shutdown had already ended | High |
The chip side (low base) — August 2025 fell inside a semiconductor downturn. Measured from there, an ordinary recovery reads as 209 percent.
The car side (high base) — demand did not weaken. The main manufacturers' summer shutdown moved from late July 2025 to early August 2026. Last August the plants were running; this August they were partly closed. The comparison month was unusually strong. A partial strike added lost output on top.
Same release, same month, opposite directions. This is why a growth rate alone is never enough.
3. That does not make the number fake
The opposite error is to distrust every growth rate once you know about base effects.
What matters is knowing which sentences are independent of the base.
| Statement | Base sensitivity | Verdict |
|---|---|---|
| "Semiconductor exports up 209%" | High | Half of it belongs to last year |
| "Semiconductor exports of $46.65bn" | None | An absolute fact |
| "A record monthly total" | None | Compared against every previous month |
| "Third consecutive month above $40bn" | None | An absolute threshold |
Absolute levels and record claims survive the base effect intact. Only the growth rate is half-owned by the past.
The working rule: use growth rates for direction, use absolute levels for magnitude.
4. Base effects always expire
The most important property of this distortion is that it is temporary. It disappears by itself the month the comparison period returns to normal.
And the moment it disappears, the growth rate falls sharply. The headline is "growth slows," and that sentence cannot distinguish between two very different things:
- Genuine deterioration — this month's absolute figure is below last month's
- The base catching up — this month's absolute figure is flat or higher
The absolute level is the only test.
The same problem recurs in inflation. Once the calendar clears a month when oil spiked, the inflation rate falls on its own even with oil unchanged, because the denominator rose. Reading that as inflation being beaten has direct consequences for rate expectations.
5. Not the same thing as seasonal adjustment
| Seasonality | Base effect | |
|---|---|---|
| Cause | Regular, repeating variation | One-off event |
| Examples | Holiday retail, summer shutdowns, festivals | Strikes, disasters, downturns, schedule changes |
| Correction | Removable by seasonal adjustment | Not removable — the reader must adjust |
US payrolls and retail sales are published seasonally adjusted because the pattern repeats and can be modelled. A base effect has no pattern to model.
Korea's August 2026 car exports sit precisely on that boundary. A summer shutdown is a seasonal factor, but a shutdown that moves from late July one year to early August the next is a one-off shift, not a repeating pattern. That is why it survived seasonal adjustment and surfaced as a base effect instead.
6. What is still open
- The 2025 figures here are derived. 58.2 billion and 15.1 billion come from reversing the growth rates and carry rounding error, because published rates are already rounded to one decimal place.
- The car decline has no published amount. The direction and the reasons were confirmed; the magnitude was not.
- The reset date is unknown. The timing of any semiconductor recovery in the second half of 2025 was not checked against monthly data. Establishing it would identify in advance the month when growth rates will drop.
Sources
- Ministry of Trade, Industry and Energy — August 2026 export and import trends (press release)
- MoneyToday — August exports up 68.7% to $98.25bn, semiconductor exports up 209%
- Industry Journal — Chips lead while cars and ships lag: August exports of $98.25bn
- Asiae — Record semiconductor exports again: 'strong chance of $1 trillion in exports this year'
- Businesspost — Semiconductor exports of $46.65bn in August set a record; total exports $98.25bn, up 68%