Analysis
Where Korean chips actually go: the Hong Kong problem
Korea shipped $4.37bn of integrated circuits to Hong Kong in July 2026 and bought back $0.03bn. The destination column in customs data records where a chip last cleared, not where it is used, and the gap between those two questions is where most readings of Korean chip trade go wrong.
Korea shipped $4.37bn of integrated circuits to Hong Kong in July 2026 and bought back $0.03bn, a ratio of 0.8 per cent. On comparable export volumes, Taiwan returned 85.1 per cent. The two destinations sit more than a hundred times apart on the same measure, and that distance is the most useful thing in the whole dataset.
Hong Kong: $4.37bn out, $0.03bn back
Shipments of HS 8542 to Hong Kong rose from $1.50bn a month in July 2025 to $4.37bn in July 2026, peaking at $6.11bn in June. Traffic in the opposite direction never exceeded rounding error. That asymmetry is not a quirk of one month; it is the defining feature of the series.
It also tells you what Hong Kong is. The territory does not consume $4bn of integrated circuits a month. It forwards them. Monthly swings in the column are correspondingly violent, because they track shipping and clearance schedules rather than end demand.
Taiwan: two-way traffic on similar volumes
Taiwan’s column is built differently. Exports rose from $3.43bn to $3.90bn over the same period, while imports rose from $2.26bn to $3.32bn. Both directions are large because the relationship is a division of labour: Korean fabless designers send work to Taiwanese foundries, and system chips for servers and equipment come back.
| Destination | Exports | Imports | Imports ÷ exports |
|---|---|---|---|
| Hong Kong | $4.37bn | $0.03bn | 0.8% |
| China | $13.42bn | $3.55bn | 26.4% |
| Taiwan | $3.90bn | $3.32bn | 85.1% |
July 2026, HS 8542. Source: Korea Customs Service.
The obvious test fails
The standard explanation of a transit port implies something checkable. If Hong Kong is a waypoint into mainland China, the two columns should rise and fall together month by month. They do not.
| Pair | On levels | On growth rates |
|---|---|---|
| Hong Kong exports vs China exports | R² 73.5% | R² 0.0% |
| Taiwan exports vs Taiwan imports | R² 26.6% | R² 3.6% |
| Hong Kong exports vs imports (control) | R² 18.9% | R² 1.0% |
| China exports vs imports (control) | R² 91.3% | R² 14.3% |
The first row contradicts the story this article is telling. A level correlation of 73.5 per cent only says both series grew over the same period, which almost any pair of rising series would show. Strip out the common trend and nothing survives: on month-on-month growth rates the relationship is R² 0.0 per cent. A heavy month into Hong Kong does not imply a heavy month into China.
Taiwan fares no better. If the division of labour were tightly coupled, exports and imports would move together; on growth rates they reach R² 3.6 per cent, barely above the one-way control. Month to month, even a genuinely two-way port does not move in step.
What separates the two ports is the ratio, not the co-movement
The distinction survives in the fourth column of the first table and nowhere else. Hong Kong returns 0.8 per cent of what it receives; Taiwan returns 85.1 per cent. Monthly timing is scrambled by clearance and shipping schedules, but the ratio exposes the structure directly.
It also holds across the whole sample rather than in one convenient month, which is what makes it usable as a classifier. Over the thirteen months the return ratio stayed inside a band for each destination, and the three bands do not overlap at any point.
| Destination, Jul 2025 – Jul 2026 | Lowest month | Highest month | Character |
|---|---|---|---|
| Hong Kong | 0.7% | 3.4% | Transit |
| China | 26.4% | 48.2% | Mixed |
| Taiwan | 47.3% | 85.1% | Division of labour |
Imports as a percentage of exports, HS 8542, by month. Source: Korea Customs Service, own calculation. Taiwan is missing one month in the source data.
A single measure that never confuses a transit port with a manufacturing partner across thirteen consecutive months is worth more than a correlation that fails every detrending test. It is also directional in a useful way: China’s ratio has fallen steadily, from 48.2 per cent in July 2025 to 26.4 per cent in July 2026, meaning the mainland now returns a smaller share of what it takes in. That drift is a change in the structure of the relationship, and it is visible in no other column.
What to watch
- Do not read the Hong Kong figure as demand. It is a waypoint. A rise there says something about routing, not about consumption in the territory.
- Watch Taiwan’s two directions separately. If one side grows while the other stalls, the division of labour itself is changing, and that is a more consequential signal than the headline.
- Separate the pie from the slice. Destination shares move when the export total moves, and that total has risen 2.66 times in thirteen months. Establish which one changed before attributing anything to a single market; the daily tracker carries both.
- Read the mainland column with the transit column. Excluding this tariff line, Korea runs a deficit with China, and adding Hong Kong to the mainland raises the Greater China semiconductor surplus to roughly $14bn a month. Neither figure is visible in the headline bilateral balance.
Sources
- Korea Customs Service Trade statistics by commodity and partner country (open API) HS 8542, monthly, exports to and imports from China, Hong Kong and Taiwan, July 2025 to July 2026
- Dotori Economy Own calculation: monthly import-to-export ratio by destination, and its range over the sample Imports divided by exports for the same destination and month, HS 8542, July 2025 to July 2026. The bands quoted are the lowest and highest monthly values in that window; Taiwan is missing February 2026 in the source data and that month is excluded from its band
- Dotori Economy Own calculation: co-movement test for transit and division-of-labour ports Pearson correlations computed on both levels and month-on-month growth rates. The sample is 13 months (12 for growth rates) and is not seasonally adjusted, so the ranking and the size of the gaps are informative while the coefficients themselves are not precise
A Korean-language version of this article is available at 도토리경제.