Six series measured the same way as the Taiwan card: customs chip exports (KCS 10-day, 20-day and final prints), the trade ministry's monthly memory and system-chip exports, SK Hynix and Samsung against SMH (in dollars), and the won. Korea items use Seoul dates; 09:00 Seoul is 20:00 ET the previous day (19:00 in winter), before the US open. System-chip exports to Dec 2025 come from UN Comtrade.
Study notes from the Korea study of Oct 3, 2026 (2021-now first, 2013-20 for consistency). The numbers in these notes do not update; the readings below them do.
How early, how good
The 20-day print is public about 9 days before month end, 10 days before the full-month first print and 25 days before the final. In 2021-now it matched the final at r 0.98 (median gap 6 points, same sign in 97% of months) and called whether the month would beat the previous one 73% of the time (79% when the final moved by more than 5 points); 2016-20 was similar (r 0.96, median gap 5 points, same sign 89%, direction 84%). The 10-day print is 10 days earlier again but noisier (2021-now r 0.95, median gap 10 points), because holidays shift the working days.
How far behind price
SMH's 12-month return lines up best with the 20-day print 3 months later, in 2021-now and in 2016-20 alike; SK Hynix's 3–4 months, Micron's 1 month. Month by month, past SMH moves explain later changes in the print (r 0.37) better than the print's acceleration explains SMH's next 3 months (r 0.18). Memory exports sit about 3 months behind SMH and 1 behind Micron; system chips about 8 behind SMH.
At the lows
2022: the 20-day print read +3% at the low (Oct 1–10: −21%, out 3 days before it), bottomed at −45% in March 2023 (+5 months); its trough-turn signal fired a year after the low, with SMH already +67%. 2025: it had slowed to +12% and bottomed in the low month (+11% on Apr 21); its acceleration turn-up came Jul 21 with SMH +61%. 2026: +164% and accelerating at the low, so nothing to turn. Memory exports read −18%, +5% and +147% at the three lows. 2013-20: in Dec 2018 the print first turned negative (−10%) one session before the low and kept falling to −29% in Jan 2019; in Feb 2020 it read +15% and rising, with SMH still to fall 30% to its Covid low.
The June 2026 top and the July drop
The June 20-day print came out the morning of SMH's Jun 22 top: +188%, a little under May's +202%. Through the −17.6% July the prints read +193% (Jul 13) and +181% (Jul 21), and memory exports kept accelerating (3-month +250% to +271%). SK Hynix peaked with SMH and fell 52% in dollars to its Jul 30 low (SMH −25%, Micron −39%). No Korean series warned.
Turn signals vs a fair test
Its acceleration turn-ups did not beat the fair benchmark in 2021-now (1 of 3; −8 points after about 3 months) but did in 2016-20 (2 of 2). The trough-turn (Oct 2023, SMH +67% off the low) and cross-above-zero (Dec 2023, +104%) are single 2021+ episodes; the Feb 2020 cross-above-zero came before a further 30% fall. Nothing passes in both eras, and the print's level and acceleration had no forecasting value out of sample. Memory exports: one 2021+ cross-above-zero win (Jan 2024); acceleration signals were false in 3 of 6 fires in 2013-20.
Release days
SMH against SPY did not move with the surprise on KCS release days (r −0.003 over 138 releases since 2021), and SK Hynix's own move on those days was no larger than on other days (2.3% on average, both). The print has not been news to prices.
SK Hynix, Samsung, KOSPI and the won
SK Hynix and Samsung move with SMH and Micron within the same week (no weekly lead in either era; SMH's Friday trades into Korea's Monday, a time-zone overlap, not a lead). Their strength against SMH did not forecast SMH over the next 3 months or so. In dollars, SK Hynix bottomed within 2 sessions of SMH at 3 of 4 lows in 2021-now and 2 of 3 in 2016-20, and 2–3 weeks earlier in 2016 and 2022 (Micron too in 2022); Samsung at 3 of 4 and 1 of 3. The March 2026 low was a dip inside a rise for both (+43% and +49% over the 3 months before it). Too few and too irregular to use as a lead. KOSPI's move is confirmation, and memory-vs-logic baskets did not pass the race's bar.
A late-cycle reading, not a timing signal
Very high Korea memory export growth has been followed by weaker SMH 12-month returns in both eras (r −0.46 in 2021-now, −0.57 in 2014-20; about 5–7 independent years each). The +311% reading of Oct 1 is far outside that sample. Read it as caution about the cycle's stage, not as a date.
Not shown
Korea's DRAM and flash price indexes (Bank of Korea ECOS) need a free API key, which has not been set up. Company filing times (OpenDART) also need a key; note only. The Korea Exchange data portal's terms restrict automated use, so Korean foreign-investor flows are not shown.
