Study · re-run every night · updated Oct 15, 2026

Does the kimchi premium mark the top? Bitcoin's price in Korea against the world's, tested

Bitcoin closed at ₩110,338,000 on Upbit, Korea's biggest exchange, on Oct 14: 0.6% over its price elsewhere at the same close ($81,682 at 1,342 won to the dollar). The Korean premium - the 'kimchi premium' - is +0.6%: usual. On the 90 days since 2017 it was 10% or more (a mania), Bitcoin bought at that day's close did −28.4% over the next 90 days on average; at 3% to 10%, +5.6%; after every day, +14.8%. No band is clear of luck (a luck test of 3 with both halves of the days agreeing): the premium ran high in a few long stretches, and a few of them decide the averages. Past patterns, not financial advice.

Today

+0.6%

usual

Won per dollar

1,342

the Fed's noon rate

10% or more

−28.4%

Bitcoin, the next 90 days

Every day

+14.8%

Bitcoin, the next 90 days

First written Oct 15, 2026 · 1 revision

§ 01 · Today

Today's premium

as of Oct 15, 2026 · Upbit's won closes and ours, Sep 25, 2017 to Oct 14, 2026

When Korean buyers rush in, Bitcoin trades higher in Korea than anywhere else - the "kimchi premium" - because won cannot freely leave the country to buy it cheaper abroad. In past manias it ran far above 10%, and traders have watched it as a sign of the top since. Here it is measured every day: Bitcoin's close in won on Upbit, Korea's biggest exchange, against its dollar close at the same moment turned into won at the Fed's noon rate - discount under 0%, usual under 3%, hot from 3% to 10%, a mania at 10% or more - and every such day is a buy of Bitcoin at our close, held 30 or 90 days.

Oct 14, 2026: +0.6%, usual. Bitcoin closed at ₩110,338,000 on Upbit against $81,682 elsewhere at 1,342 won to the dollar. The highest on our record: +58.8% on Jan 8, 2018.

the Korean premiumhot, 3% to 10%a mania, 10% or more

Today's premium and every week's since 2017, free as JSON for your own charts and bots: /api/v1/kimchi.json (the free API).

§ 02 · After it

What came after it

as of Oct 15, 2026 · Bitcoin, from the day's close

Bitcoin's average next 90 daysevery day's averageunder each bar: its days

Premium	Days	Stretches	Next 30 days	Next 90 days
a discount under 0%	797	56	+6.8%	+30.8%
usual 0% to 3%	1,743	53	+5.5%	+13.4%
hot 3% to 10%	647	46	+3.3%	+5.6%
a mania 10% or more	90	8	−15.3%	−28.4%
every day	3,277	—	+4.8%	+14.8%

Days and stretches count the days with 30 days after them; a stretch ends after 7 days out of its band. Today's band is marked. Each band against all the other days over the same hold, with its luck test under it (✓ past the bar, both halves of the days agreeing; the halves on hover):

Premium	Next 30 days	Next 90 days
under 0%	+2.6% t 0.7	+21.0% t 1.7
0% to 3%	+1.5% t 0.5	−3.0% t −0.4
3% to 10%	−1.9% t −0.5	−11.5% t −1.1
10% or more	−20.7% t −1.8	too few

A hot premium (3% to 10%) came on 647 days in 46 separate stretches, the last on Jan 31, 2026. Bitcoin bought on one of those days rose over the next 90 days 48% of the time (−1.8% in the middle case, +5.6% on average): −11.5% against the other days, −29.4% in the older half of the days and +7.0% in the newer. A luck test of −1.1, not past the bar of 3.0: 16 holds that share no day, and a few big swings decide the average.

§ 03 · How it is tested

How it is tested

as of Oct 15, 2026 · methodology

A luck test (t) says how far a result sits from zero, in units of its own noise: about 2 is rarely luck for one test, but many tests at once need more, which is what the first cut is for.

Upbit's day ends at 00:00 UTC, as ours does, so its close is set against our close of the same day, turned into won at the newest noon rate the Fed had published for New York on or before that day (FRED's DEXKOUS, carried over weekends, a week at most). Tether on our side is taken as a dollar. Each day's premium is known by our close that day, the close the hold starts from. Neighbouring days share most of their 30 or 90 days, so the luck test does not count them as fresh evidence: each calendar day weighs in by how many of the band's holds, and of the other days', cover it, so a band counts as the holds that share no day. With 7 tests, a result needs a luck test of 3.0 or more and both halves of the days on its side.

The bands were set before any number here was read: 0%, 3% and 10%. One exchange's price can stray for reasons of its own - Upbit paused deposits and withdrawals after its November 2019 hack - and the premium's high stretches are few, so a mania's average rests on a handful of episodes. Moves are Bitcoin's own price, before trading costs.

§ 04 · Revisions1 dated

Revisions

as of Oct 15, 2026 · every change to its headline numbers, dated

This study is re-run every night on the whole history. It was first written on Oct 15, 2026; its numbers last moved on Oct 15, 2026. Each night its headline numbers moved is listed here, newest first.

Oct 15, 2026Bitcoin over the next 90 days against the other days: after a hot premium −12% · after a mania 0%

§ 05 · More

More research

as of Oct 15, 2026