The press blamed US rates and the yen. Measured against the ruler we run every day, both explanations were weak — and the move that matters started the day after the crash.
This was not a quiet week in Asia.
On September 1, the press reported Japan’s 10-year government bond yield above 3% for the first time since 1996 — in a country whose money cost close to nothing for thirty years. On September 2, Korea’s KOSPI fell 3.99%; foreign and institutional investors have sold a net 13.9 trillion won since August 20. On September 3 the index dropped 3% in ten minutes around 2 p.m., then closed up 0.26%. The yen strengthened to 156 per dollar. The US 30-year Treasury yield reached 5.27%, its highest since 2007.
All of that is press. Everything below is not.
When Korean stocks fall 4%, several reasons are mixed inside that 4%: the part caused by US rates, the part caused by Japanese rates, the part that is Korea’s own. The press lists the reasons. It does not say how many percent each one is. We do that arithmetic every day.
What the press said
This week’s story fits in one line: US long rates spiked, Japanese rates rose and the yen strengthened, the so-called yen carry trade began to unwind, and Asian stocks got sold.
One paragraph on what that means. When Japanese rates were zero, investors borrowed cheaply in Japan and put the money to work in Korea, the US, elsewhere. When Japanese rates rise, that borrowed money gets expensive, so the positions get closed and the money goes home — and Korean stocks get sold along the way. It is a plausible story. So we tested it in two pieces: first America, then Japan.
First, how much was America
Korean stocks tend to follow the US market. Foreign investors hold a large share, and Korea’s biggest industries — semiconductors above all — sell to American customers. We record that relationship every day: if New York moved this much yesterday, Seoul normally moves this much today.
On September 1 the US semiconductor index fell 2.1%. A bad day, but only the bottom 19% of the past year — not an extreme one. Put through the usual pattern, the fall Seoul “owed” to America was 0.6% to 1.1%. The range is that wide because the answer doubles depending on which window of the pattern you use, so we do not pin it to one number.
Seoul actually fell 3.99%. America explains a quarter at most. The remaining 2.9 to 3.4 percentage points came from somewhere else.
We were here two weeks ago. On August 19 the KOSPI fell 5.8% and the press blamed the US 30-year yield. That day the American signal really was large — bottom 4% of the year — and the same arithmetic gave America 23% to 45% of the fall. Side by side, this week’s crash was less American than August’s: a smaller US shock, a similar drop.
| Aug 19 | Sep 2 | |
|---|---|---|
| KOSPI close | −5.8% | −3.99% |
| US signal (rank in past year) | bottom 4% | bottom 19% |
| Share explained by the US | 23–45% | 14–28% |
| Sessions until back in the normal band | 4 | 1 |
The last row of that table is the point of this article, and we come back to it in the conclusion. First, the press’s second reason: Japan.
Then, was it Japan
The question is not whether Japanese rates are high. It is whether they jumped this week or had been rising for months. If they climbed slowly over months, the market already knew — and something the market already knew does not suddenly take 4% off Korean stocks on a Tuesday.
We have recorded the Japanese government bond curve, by maturity, every trading day since July. On that record, the 10-year’s rise has been under way for at least 124 trading days — about half a year. Across September 1–3, each day’s move sat inside the ordinary range. And on our record (Ministry of Finance closes) the 10-year crossed 3% on September 2, not September 1; the press used an intraday print a day earlier.
The chart says one thing: all three lines have been climbing since July, and nothing breaks in the first week of September. Which maturity moved first, our record cannot settle — but that question does not matter here. The question was “did this happen suddenly this week,” and the answer is no. Three percent made a headline. For the market it was one more number on a road it had been travelling for six months.
So: America explains a quarter at most, and Japan was old news. If both of the press’s reasons are weak, the remaining 3 points came from inside Korea, not outside. The way to check that is to look at what was sold that day — and what was not.
What was sold, and what was not
An outside shock and an inside move sell differently. Fear from outside — US rates, the yen — sells every sector at once. Money moving around inside a market sells one sector to buy another. So the pattern of who got sold on a given day tells you whether the shock came from outside or was made inside.
We rank each Korean sector every day by how unusually it moved once the US signal is accounted for, against the past year. Here the result was not what we expected.
On September 2, the day of the crash, Korean semiconductors ranked in the top 11% of their year — they held up. The press wrote “chip-led decline,” and in raw terms it was; but given that US chips had fallen 2.1%, Korean chips fell less than usual that day. That fits the 1.6 trillion won of buybacks Samsung and SK hynix put in that session. What got sold unusually hard was batteries (bottom 20%). Not every sector at once — not the shape of fear from outside.
Then the next day the two changed places. On September 3, semiconductors dropped to the bottom 4% of their year and batteries rose to the top 8%. The first day that looks like chips being sold to buy batteries is the same day the index fell 3% and bounced. The press has noted leveraged battery funds up 38–46% over a month; our record agrees on direction, but the order matters — this pattern appeared the day after the crash, not on it.
Read the two charts together and the week has an order. September 2 was not every sector sold at once — batteries fell hard, chips held. September 3 was the reverse: chips sold, batteries bought. Over two days investors did not sell the market; they moved from one sector to another. That is the shape of money changing seats inside Korea, not of fear arriving from outside.
Conclusion
America explains at most a quarter of this week’s 4% fall. The Japanese rates the press pointed at had been rising for half a year — not news to the market. And the selling on the crash day did not look like an outside shock either: batteries sold, chips held. So what were the remaining 2.9 to 3.4 points?
First, how long they lasted. On September 3 the KOSPI closed near where our US pattern said it should be. In other words, the extra fall that America could not explain on September 2 was gone a day later. On August 19 the same arithmetic took four sessions to close. A fall that disappears before it can be named does not deserve a large reason, American or Japanese. At the index level, this week’s 4% was a one-day shock.
But the index and the sectors tell different stories. The KOSPI — an average — is back where it belongs. The sectors that make up that average are not. Semiconductors are still far more sold than usual, batteries far more bought, both outside their normal band for a second session as of the September 3 close. A normal average does not mean a normal inside.
So the week in one sentence: the index’s crash was over in a day, but the switch from chips to batteries began the day after and is still running. From the next session, the thing to watch is not US rates or the yen. It is whether semiconductors and batteries come back inside their normal range.
This article is written on the record through the close of Thursday, September 3. Friday’s session is not yet in our record; it arrives on the next trading day, and the tests below count from there.
To see this you need two rulers running daily, from before the crash: how much the US moves Korea, and where each sector’s normal range is. Most desks hold one.
The index came back in a day. The sectors have not.
What would prove us wrong
Three numbers, each checkable on a single day.
- If the part not explained by the US leaves the normal band again within 5 sessions, “one-day” was wrong and the crash was not over.
- If semiconductors and batteries are both back inside the normal band within 5 sessions, the “switch” was two days of noise.
- If Japan’s 10-year falls back below 2.90% and Korean sector ranks normalise with it, we recompute the Japanese share we judged small.
Whichever fires goes into the log that day. Not rewritten later.
What this article cannot say
We can separate the American share of a Korean move. We cannot prove why the rest vanished in a day. The American share itself moves between 14% and 28% depending on the coefficient. Batteries have no sector index, so one representative stock stands in. That is not a footnote; it is the limit of what we measure.
Press figures: Seoul Economic Daily (Sep 3), Euronews (Sep 1). Everything else: Asia Pulse, measured daily; the tests were written down and sealed before the numbers were computed and were not changed to fit them. Model coefficients, daily series and single-name readings are not published here — they are what /chain and /japan return. Cite with attribution to Asia Pulse.