When the Last Buyer Leaves

We remeasured #17's needle readings under the same definitions: 2026 KOSPI+KOSDAQ cumulative totals. Our series reproduces #17's 9/2 values to one decimal place.

Cumulative flows combine KOSPI+KOSDAQ. Institutions include pension funds and others. Changes use unrounded source data. ①, ③ and ④ are #17's needle numbers.

The mobilization gauge barely moved, from 76% to 75%. Open it up, however, and individuals reduced exposure by KRW 15.8 trillion while other corporations added KRW 29.7 trillion. A needle combining the two hands hid the changeover. #17 wrote, "Do not combine the three axes; doing so hides which one moved", yet left this needle as a composite. From this report, we split it in two in the gauge panel.

#17 linked other-corporation buying with buybacks, but said that before stock-level separation it could establish only "a consistent alignment". Other corporations include general trading alongside buybacks. Stock-level investor data remain unavailable through a free channel. The two companies have instead disclosed progress.

From August 20 to September 28, other corporations net bought KRW 41.47 trillion on the KOSPI. The two companies' purchases over that period were KRW 40.76 trillion: Samsung Electronics KRW 12.25 trillion (KRW 15 trillion minus KRW 2.75 trillion remaining), SK hynix KRW 28.51 trillion (KRW 40 trillion minus KRW 11.49 trillion remaining), based on 9/28 reports of remaining budgets. That matches 98%. These purchases were back-calculated from 9/28 reports of remaining budgets★. This strongly suggests the two companies' buybacks explain most other-corporation net buying. It compares gross purchases inferred from reported remaining budgets with market-wide net purchases; individual trade counterparties remain unidentified.

Buying kept a steady beat. From August 24 to September 30, other corporations bought an average KRW 1.64 trillion per day. The minimum was KRW 1.49 trillion, maximum KRW 1.78 trillion and standard deviation KRW 0.06 trillion. Even on four days when foreigners sold more than KRW 3 trillion (8/24, 8/25, 9/14 and 9/28), it was KRW 1.49–1.64 trillion. Similar-sized buying continued daily across the market. It was closer to a metronome, repeating a fixed beat.

Individuals absorbed the daily swings. Across 28 trading sessions from 8/20 to 9/30, foreign/other-corporation correlation was −0.24; foreign/individual correlation was −0.90. The daily average above starts on 8/24, when Samsung began buying, so the sample starts differ. While corporate buying stayed steady, individual trading moved strongly opposite foreigners. Outside buyback periods in 2026, normal other-corporation buying was KRW 0.05 trillion a day. September's wall was more than 30x normal.

Two stocks accounted for 76.6% of September foreign net selling: SK hynix KRW 11.58 trillion and Samsung Electronics KRW 4.89 trillion (reports citing KRX totals, The Seoul Economic Daily, 9/30). The stocks foreigners sold most were the companies buying back shares. Individual trade counterparties are unconfirmed. Prices are compared from around the August buyback start through 9/30. SK hynix rose about 18%, from KRW 1.5 million (8/19, our record★) to KRW 1.776 million (9/30). Samsung Electronics began at KRW 257,000 after an 8.7% fall on its first buyback day, 8/24, and closed 9/30 at KRW 268,500.

Neither stock rose straight up. The dataset's 8/24–9/30 daily closes show Samsung Electronics falling to KRW 248,500 on 9/14, then rising to KRW 286,500 on 9/23. SK hynix's period low was KRW 1,596,000 on 9/3; its high was KRW 1,879,000 on 9/21. The 8/19 starting value for SK hynix comes from a separate record★; this intervening path uses regular-session data from 8/24 onward.

The index responded differently with and without the wall. A simple regression compares daily KOSPI returns with foreign net buying: 34 days in the earlier period, 28 in the later one.

• 7/1–8/19: KOSPI −1.72% per KRW 1 trillion of foreign net selling.
• 8/20~9/30: −1.01%.
• The regression's return at 0 foreign net buying rose from about 0.0% to +1.3%. This is not the average return on days when net buying actually equaled 0.

For the same foreign selling amount, the regression's index decline became about 40% smaller. The buyback wall may have absorbed the shock (hypothesis). Rates, earnings and sentiment also changed; this regression does not isolate the effect of buybacks.

Our interpretation: buybacks return value to remaining shareholders. For departing shareholders, they also offered an exit without a large price concession. Corporate buying and foreign selling in the same stocks shaped September. Once the exit closes, those wanting to leave must sell less or sell cheaper.

#16 and #17 described Samsung Electronics' KRW 15 trillion buyback acquisition period as "through November 21". The disclosure says so. Actual buying ran much faster.

• A 9/30 report said 49.8 million shares had been bought by 9/29, 93.5% of the 53.29 million-share limit.
• The average purchase price was below the KRW 281,500 reference price (reported KRW 261,951★), so the share-count limit fills before the money runs out.
• Completion was therefore reported for early October. With 3.49 million shares remaining on 9/29 and about 2 million bought daily at the reported pace, 10/1 was likely the last day.
• On 10/1, KOSPI other corporations still bought KRW 1.64 trillion. The beat remained steady. Check the first step in flows from 10/2. Actual completion will be checked through the results disclosure.

November 21 was a deadline, not a timetable.

SK hynix's KRW 40 trillion buyback is for full share cancellation. By 9/29 it had bought 17.15 million shares (71.3%); reports put the final total around 22.58 million because the monetary limit would fill first. Buying the remaining roughly KRW 9 trillion at KRW 1.1 trillion daily takes 8–9 sessions. October 3 is Saturday and the 9th Hangeul Day, putting completion around October 13–16. The wall drops in two steps: KRW 1.6 trillion/day → about KRW 1.1 trillion (estimated early October) → the usual KRW 0.05 trillion (estimated mid-October). The chart's 10/5 and 10/15 are illustrative dataset boundaries, not confirmed end dates.

There is another point. Samsung Electronics' purchased shares will not be canceled. They are for employee performance compensation, transferred over time; if employees sell, the shares return to the market (payment schedule unconfirmed). Today's buying also gathers tomorrow's possible supply in advance.

What happened over 20 trading sessions after buybacks ended? These are four Samsung Electronics cases available in our series, which starts in 2024, alongside the current program.

The first three were smaller, with different subsequent foreign flows and index directions. This table alone cannot separate buyback completion from the memory cycle. April is the closest precedent. Over 20 days after the comparison period, foreigners sold KRW 41.1 trillion and individuals bought KRW 34.8 trillion. The KOSPI instead rose 15.9%. Individuals had accumulated fuel around then. Deposits reached KRW 139.7 trillion on June 4, the highest deposit observation in this dataset. It is after April's completion, so it does not directly show buying capacity on the completion date.

Today's wall, by daily average, is 4.3x April's and about 20x the three 2024–25 programs. Our series has no precedent for purchases of this scale ending. Previous outcomes are not this report's forecast.

Individuals sold KRW 14.7 trillion on the KOSPI in September. Separately, money waiting in brokerage accounts grew. Investor deposits rose KRW 11.1 trillion, from KRW 93.6 trillion on September 4 to KRW 104.7 trillion on September 30 (reaching KRW 107.7 trillion on 9/29). Margin loans stayed around KRW 33.4 trillion; margin loans/deposits fell from 34.1% on 9/1 to 31.9%. Individual net selling alongside rising deposits looks like selling and waiting (an interpretation of the alignment). It does not mean the same people deposited their sale proceeds unchanged.

There is fuel, but the borrowing gate is tightening. Coverage and timing follow the 9/29 report.

• October 1 (effective): 10 comprehensive financial investment firms voluntarily manage credit balances within 90% of equity. This does not uniformly change the statutory 100% cap (reported).
• October 19: manage single-stock shares within 15% of brokers' total margin loans. Applies to new investments (reported).
• By year-end: raise the collateral ratio from 45% to 50% (reported).
• Individuals in April bought with added borrowing. This time, capacity to borrow more differs by broker. Some firms have room, so margin-financed buying is not blocked for all individuals.

Pension funds were quiet in September. Pension funds and others sold only another KRW 0.3 trillion over the 4 weeks since the September 2 sell-off (#17). The overhang feared in #17 also shrank. The National Pension Service (NPS)'s domestic-equity weight fell from 29.1% at end-June (NPS release) to 24.9% at end-July (reported★). July's stock decline may have lowered the weight; actual trading and other asset-price changes must also be examined. It remains above the 20.8% target, so this weight alone offers little basis to expect sustained pension net buying.

① Samsung Electronics' next round. A 5/30 report estimated after-tax treasury shares to be distributed over the next 3 years at about KRW 80 trillion, expecting several buyback rounds. It highlighted July, October and next January as likely purchase periods (reported★). No disclosure yet. A new October round would keep the wall in place. This is the report's largest unknown.

② Earnings and exports.

• Exports over September 1–20 were USD 71.41 billion, up 78.3%. Semiconductors were USD 34.13 billion, up 259.4%, or 47.8% of the total (Korea Customs Service, 9/21).
• Full-September exports were USD 120.94 billion, up 83.5%. A record, passing USD 120 billion for the first time. Semiconductors were USD 60.3 billion, up 262.8%, exceeding USD 60 billion for the first time and accounting for about half, 49.9%, of total exports. The trade surplus also hit a record USD 49.85 billion (Ministry of Trade, Industry and Resources original, 10/1).
• Prerelease forecasts were USD 100–108 billion (reports/broker estimates). The remaining ten days after 1–20 added USD 49.5 billion, including USD 26.2 billion of semiconductors, despite the Chuseok holiday.
• Converting the monthly trade surplus at the 10/1 rate of KRW 1,358.83 gives about KRW 68 trillion, roughly twice the two companies' buybacks of about KRW 33 trillion that month. The trade surplus is neither domestic stock-buying funds nor those companies' cash flow. This is a scale comparison only. Foreign investors net sold about KRW 21 trillion on the KOSPI that month.
• Q3 operating-profit forecasts are Samsung Electronics KRW 110.3 trillion and SK hynix KRW 78.1 trillion (reported). These are single-quarter figures.
• If foreigners sold because of rates and profit-taking rather than earnings, these figures could be a reason to return.

③ KRW 30 trillion dividend.

• Samsung Electronics' 8/21 disclosure indicated a KRW 30 trillion Q3 cash dividend, to be finalized by the board in late October. Payment is estimated around late November★.
• About half belongs to domestic shareholders. If it returns to equities, the gap between SK hynix's mid-October completion and late-November dividend receipts is about 5 weeks.

The last buyer sets the price. September's last buyers were companies with departure dates. The next could be a new corporate round, individuals waiting in accounts or foreigners returning on earnings. We do not know yet. We can watch the needles at the dates when the steps descend.

Korea reopened on 9/28.

• On the first session after the holidays, the KOSPI fell 2.70% to 6,889.74.
• Foreigners sold KRW 3.22 trillion; individuals bought KRW 2.60 trillion and other corporations KRW 1.64 trillion. Corporate buying kept its usual beat that day too.
• The next report will examine refining stocks' responses as a series.

East–West Pipeline.

• A 9/29 report put Yanbu loadings back at about 2 million barrels/day★. Kpler estimates pipeline throughput at 2.65 million barrels/day (Pipeline Technology Journal).
• Restoring normal flows around 5.5 million barrels/day reportedly takes about a month.
• December Brent fell from USD 99.42 on 9/29 to USD 96.28 on 9/30.

Our model correction.

• Our expected oil fair-value estimate rose from USD 109.1 on the night of 9/29 to USD 113.0, without new facts. Temporarily reduced exit flows in the week the pipeline stopped had entered the route-allocation calculation.
• From now on, retain the last reliable value until a new reliable value arrives (Kay decision). That rule gives about USD 111 (linear approximation, to be confirmed by rerun).

1. If Samsung Electronics announces a new October buyback, the wall stays. This report's question changes from "if it leaves" to "when it leaves".
2. If foreigners turn buyers on earnings and exports, the exit is no longer needed. Section 03's "door" matters only while they want to leave.
3. If the steps do not appear in the data, meaning other corporations still buy nearly KRW 1 trillion daily after mid-October, a large buyer beyond the two companies must exist. That conflicts with Section 02's 98% match.
4. If deposits turn into individual net buying, individuals absorb the sales as in April. Section 06's "selling and waiting" turns into "buying again".
5. If the index still responds by around 1% per KRW 1 trillion of foreign selling after the wall disappears, Section 03's "the wall absorbed the shock" would be wrong.

This lineup carries forward #17's needles for Korea and splits combined gauge ① into two. It adds the wall and exit.

There is no composite score. ① and ③ view the same category over different windows: ① measures the wall's beat; ③ its accumulated amount. #17's ⑤–⑧ (won/memory, real yield–BEI, ERP and yen switch) sit outside this report's axes and are removed here; existing observations continue collecting their values.

10/1 (Thu) September trade — confirmed. USD 120.94 billion (+83.5%); semiconductors USD 60.3 billion (+262.8%); surplus USD 49.85 billion. Growth over 1–20 continued through month-end. KOSPI foreign net selling eased to −KRW 0.55 trillion that day (Section 07 box). Next confirmation: does easing foreign net selling last more than three days? Section 09, item 2.

10/2 (Fri): first-step observation begins — did Samsung Electronics complete buying? Confirmation: do daily KOSPI other-corporation purchases fall from KRW 1.6 trillion to near KRW 1.1 trillion? If not, move Section 04's date forward one day at a time.

Effective 10/1 — lower credit-provision limits at large brokers. Confirmation: direction of margin loan balances and investor deposits.

Around 10/7 (customary timing, unconfirmed★): Samsung Electronics Q3 preliminary earnings and possible new buyback disclosure. Confirmation: a new round means Section 09, item 1; otherwise, keep observing the steps.

10/2 (Fri) 21:30: US September employment. US Bureau of Labor Statistics calendar converts 08:30 Eastern to Korea time here. Confirmation: direction of US long-term yields, one reason for foreign selling.

10/13–16: estimated SK hynix buyback completion★. Confirmation: do other corporations fall toward the usual KRW 0.05 trillion daily, the second step? Fill Section 05's empty cells with foreign/individual flows and the index over the next 20 sessions.

10/19: 15% single-stock margin-credit concentration limit. Confirmation: margin-credit weight of the two semiconductor stocks.

Late October: Samsung Electronics board meeting. Confirmation: final KRW 30 trillion dividend and the first cancellation marker, carried over from #16 and #17.

Around late November: dividend payment★. Confirmation: individual net buying and deposits after payment.