Korea · Investor flows · Industry research
These selected flows point to research candidates in semiconductor equipment, testing, packaging and AI-server components. The evidence is stock-specific, and foreign investors and domestic institutions often sit on opposite sides.
Electrical equipment has a narrower foreign-buying signal. Banks show persistent reported institutional support, but block trades and share repurchases weaken the case for calling it fresh growth conviction. Public data cannot trace the same won from a sold index constituent into a new asset.
KOSPI foreign net purchases
June 23–October 12 · vendor trade-date display
KOSDAQ foreign net purchases from its own peak
April 28–October 12 · a different window
Weeks of positive institutional bank-basket net buying
June 29–October 8 · unadjusted for buybacks
The critical distinction: KOSDAQ foreign investors were still slightly net buyers over the full interval after its April peak, even though they were net sellers over the later period shared with KOSPI. A single “after the peak” total would conceal that reversal.
1. Define the peak before following the money
On a narrow screen, swipe tables horizontally. Keyboard users can Tab to each table region and use the arrow keys.
| Reference | Reported peak | Flow measurement |
|---|---|---|
| KOSPI closing peak | June 22 · 9,114.55 | June 23–October 12 · 74 sessions |
| KOSPI intraday sensitivity | June 19 · 9,385.59 | June 22–October 12 · 75 sessions |
| KOSDAQ closing peak | April 27 · 1,226.18 | April 28–October 12 · 110 sessions |
| KOSDAQ intraday peak | April 27 · 1,229.42 | Same next-session start: April 28 |
Chosen reference peaks are reported market records, not a fresh reconstruction of the full KRX index history. Sources: KOSPI closing record, KOSPI intraday record, KOSDAQ closing reference and KOSDAQ intraday reference.
| Window | KOSDAQ foreign | KOSDAQ institutions |
|---|---|---|
| Own peak: April 28–October 12 | +0.4262 | −1.3504 |
| Common: June 23–October 12 | −2.9446 | +1.2481 |
| Latest 20 sessions | −1.7458 | +1.4418 |
On the common June 23–October 12 window, KOSPI institutions bought ₩4.0041tn net while individuals bought ₩21.1073tn and other corporations ₩50.7763tn. These are counterpart investor categories in the same displayed market accounting. They do not identify the later destination of foreign sellers’ proceeds.
Starting instead after KOSPI’s intraday peak adds June 22: foreign net purchases become −₩78.4344tn and institutional net purchases +₩4.3075tn. The reference choice is economically material; the rest of this article keeps the closing-peak common window fixed unless a different window is named.
Own sums of NAVER Pay Securities daily investor tables, KRX venue selected. The display is a licensed vendor source, not an official KRX CSV. October 13 live rows are excluded; NXT is not separately added.
2. “Institutions” contains opposing decisions
KOSPI’s common-window institutional total of +₩4.0041tn includes financial-investment firms at +₩9.5822tn, investment trusts/private funds at −₩7.0241tn and pensions/others at −₩0.7074tn. Excluding financial investment leaves −₩5.5781tn. That residual is not a pure long-only fund category.
Common post-peak window: June 23–October 12
74 trading sessions. Same dates and scale for both markets in this view. Institutional subcategories are included in the total, not added to it.
Read all investor categories and exact values
| Investor category | KOSPI | KOSDAQ |
|---|---|---|
| Individuals | +21.1073 | +1.4948 |
| Foreign investors | −75.8878 | −2.9446 |
| All domestic institutions | +4.0041 | +1.2481 |
| Financial investment | +9.5822 | +1.3576 |
| Insurance | +1.0281 | +0.0160 |
| Investment trusts / private funds | −7.0241 | +0.2322 |
| Banks | −0.1608 | −0.1196 |
| Other financial institutions | +1.2861 | −0.4359 |
| Pensions and others | −0.7074 | +0.1976 |
| Other corporations | +50.7763 | +0.2014 |
Previous 20 sessions: August 11–September 8
20 trading sessions. Same dates and scale for both markets in this view. Institutional subcategories are included in the total, not added to it.
Read all investor categories and exact values
| Investor category | KOSPI | KOSDAQ |
|---|---|---|
| Individuals | −18.3414 | +2.9706 |
| Foreign investors | −1.4814 | −0.5102 |
| All domestic institutions | −0.1127 | −2.4654 |
| Financial investment | +0.6394 | −1.6009 |
| Insurance | +0.0492 | −0.0251 |
| Investment trusts / private funds | −0.3592 | −0.4506 |
| Banks | −0.0302 | −0.0497 |
| Other financial institutions | +0.3593 | −0.2658 |
| Pensions and others | −0.7715 | −0.0733 |
| Other corporations | +19.9354 | +0.0046 |
Latest 20 sessions: September 9–October 12
20 trading sessions. Same dates and scale for both markets in this view. Institutional subcategories are included in the total, not added to it.
Read all investor categories and exact values
| Investor category | KOSPI | KOSDAQ |
|---|---|---|
| Individuals | +2.8480 | +0.3148 |
| Foreign investors | −31.5945 | −1.7458 |
| All domestic institutions | −0.6364 | +1.4418 |
| Financial investment | +1.6688 | +0.4229 |
| Insurance | +0.1093 | +0.0331 |
| Investment trusts / private funds | −2.3554 | +0.9311 |
| Banks | +0.0126 | −0.0672 |
| Other financial institutions | +0.3324 | −0.1098 |
| Pensions and others | −0.4040 | +0.2322 |
| Other corporations | +29.3826 | −0.0106 |
Financial-investment firms include trading and intermediation activity. ETF liquidity provision and creation can generate stock purchases without expressing the same view as a pension manager’s portfolio allocation. Samsung Securities explains this mechanism; its 2025 discussion does not quantify the mechanical share of these 2026 totals.
The recent KOSDAQ improvement is unusually concentrated
Across the latest 20 sessions, KOSDAQ institutions bought +₩1.4418tn, compared with −₩2.4654tn in the previous 20. Investment trusts/private funds bought on 14 of those latest 20 days; pensions/others bought on 16. But net amount and daily breadth tell different stories.
| Latest 20-session category | Reported total | Largest absolute day removed | Remaining total |
|---|---|---|---|
| Institutions | +1.4418 | September 10: +1.3996 | +0.0422 |
| Trusts / private funds | +0.9311 | September 10: +1.2724 | −0.3413 |
| Pensions / others | +0.2322 | October 1: +0.0731 | +0.1591 |
Removing each category’s largest absolute day is a descriptive stress test chosen after observing the sample. It is not an allegation of an abnormal trade, a causal explanation, or a reason to erase a real transaction. It shows why the headline inflow alone is insufficient evidence of broad, durable accumulation.
3. Where net buying actually appeared
Completed-period stock tables sharpen the picture. The latest window is September 9–October 12; the longer check is July 13–October 12. Both end on the same day, but neither is the entire interval since each index peak. The source publishes only the 20 largest net purchases and sales, with ETFs mixed among stocks.
| Selected security | Foreign investors | All institutions |
|---|---|---|
| Samsung Electro-Mechanics KOSPI | +271.5 | +373.7 |
| LS ELECTRIC KOSPI | +70.6 | Not shown |
| Hana Micron KOSDAQ | +48.8 | +68.4 |
| LEENO Industrial KOSDAQ | −373.1 | +300.8 |
| HPSP KOSDAQ | Not shown | +222.1 |
| Techwing KOSDAQ | +86.6 | −65.1 |
| SFA Semicon KOSDAQ | +160.1 | −76.8 |
| SK Innovation KOSPI | +291.1 | +144.9 |
| Hyosung Heavy Industries KOSPI | −274.0 | −171.4 |
Source: NAVER foreign-investor lists and institutional lists, KRX selected, one-month completed-period view. “Not shown” means absent from these top/bottom-20 lists, not zero or a known sign. Values above are rounded to ₩0.1bn.
Semiconductor exposure is being redistributed within the supply chain. Institutions bought LEENO, HPSP and other equipment/test-related names. LEENO is a clear example of opposite signs: institutional buying met foreign selling. HPSP’s recent foreign sign is not established by the truncated list. Foreigners bought Techwing and SFA Semicon while institutions sold them. Hana Micron is a narrower recent overlap: both groups appear as net buyers. This does not establish a uniform foreign-and-institutional semiconductor-sector inflow. Business classifications are supported by LEENO’s industry-exhibition profile and Techwing’s test-handler product page, alongside the company sources below.
Samsung Electro-Mechanics offers a particularly useful time-window check. Foreign investors bought ₩271.500bn in the latest 20 sessions but sold ₩1,715.453bn across the longer July 13–October 12 window. Institutions were positive in both, at ₩373.718bn and ₩445.087bn respectively. Recent convergence is not three months of shared conviction.
Adjusting for trading activity without confusing volume and value
For a limited liquidity comparison, divide displayed net shares by total traded shares. LEENO’s latest-window institutional net purchases were approximately +17.96% of its reported traded share volume; foreign net purchases were −22.56%. Hana Micron’s corresponding ratios were +4.63% and +1.52%. The numerator is rounded to thousands of shares.
This signed net-volume ratio measures net buying relative to share turnover. It is not gross participation, ownership change, or a net-value/turnover-value ratio. Full-market trading-value denominators and a complete sector constituent panel were not collected, so no market-wide flow-intensity or industry-allocation-share ranking is claimed.
Large memory stocks and overseas-linked ETFs require separate interpretation
In the same latest stock window, foreigners sold ₩17,483.820bn of SK hynix and ₩7,973.721bn of Samsung Electronics common shares. Institutions sold ₩512.937bn of SK hynix but bought ₩169.147bn of Samsung common shares. Selling a memory producer does not prove investors abandoned every semiconductor activity.
The longer-window lists also show foreign net purchases of ₩635.984bn in the Korea-listed TIGER NASDAQ 100 ETF, versus institutional net sales of ₩1,312.810bn. That is observable trading in an overseas-linked product. It is not proof that the same proceeds left Korea, and secondary-market ETF trading is not equivalent to net fund subscriptions or a measured cross-border transfer. These product figures are not added to the market totals or underlying US-stock exposures.
4. Banks: persistent buying, but inspect the transaction
Fifteen consecutive Bank Weekly tables from Hana Securities provide a separate, deeper sector case. Summing the reported eight-bank basket from June 29 through October 8 gives institutional net purchases of ₩2,829.8bn, including a pension subset of ₩147.3bn, and foreign net purchases of −₩708.5bn. Institutions were net buyers in 13 of 15 weekly observations and positive cumulatively in all eight names.
| Bank | Institutions | Pensions (subset) | Foreign |
|---|---|---|---|
| KB Financial | +996.4 | −62.2 | −799.1 |
| Shinhan Financial | +767.7 | −39.3 | −478.7 |
| Woori Financial | +357.1 | +187.6 | +1,052.0 |
| Hana Financial | +373.8 | −45.7 | −211.8 |
| Industrial Bank of Korea | +27.9 | +7.1 | −30.3 |
| BNK Financial | +99.0 | +26.1 | −107.2 |
| iM Financial | +78.1 | +14.9 | −45.0 |
| JB Financial | +130.2 | +58.5 | −87.8 |
Foreign buying was much narrower: Woori was the only one of the eight with positive cumulative foreign net purchases, at +₩1,052.0bn. The basket excluding Woori was approximately −₩1,760.5bn. A positive headline for one bank is not evidence of broad foreign rotation into all banks.
Transaction context changes the meaning. For the week ending October 2, Hana reported bank-basket foreign net buying of ₩389.9bn, with Woori alone at ₩599.6bn. The analyst linked the week’s pattern to a ₩676.5bn Woori block trade largely taken by foreign investors. That is a one-week explanation, not an adjustment to the whole post-peak period.
In the October 12 report, the latest bank week showed institutions buying just ₩8.1bn net, while the analyst estimated about ₩90bn of share-repurchase buying; its commentary described the result excluding those repurchases as net selling. These estimates demonstrate why reported institutional purchases cannot all be treated as new active-manager conviction. No complete 15-week buyback-adjusted series is available here.
Sources: Hana Securities October 6 Bank Weekly, October 12 Bank Weekly and the preceding weekly tables from July 6 onward. Pension amounts are already inside institutions. This broker basket has its own inclusion conventions and is not merged with NAVER market totals. The report labeled September 25 refers to a holiday-shortened week whose last trading session was September 23. The June 29 publication’s earlier week is excluded because it includes the June 22 peak session.
Open the 15 original bank-weekly source reports
2026-07-06 · 2026-07-13 · 2026-07-20 · 2026-07-27 · 2026-08-03 · 2026-08-10 · 2026-08-18 · 2026-08-24 · 2026-08-31 · 2026-09-07 · 2026-09-14 · 2026-09-21 · 2026-09-28 · 2026-10-06 · 2026-10-12
5. Did the money move into bonds or leave Korea?
Official foreign-investment releases do not support a simple “equities out, bonds in” story across the whole period.
| FSS settlement month | Foreign stock net purchases | Foreign bond net investment |
|---|---|---|
| July 2026 | −31.664 | +2.388 |
| August 2026 | +0.344 | −4.736 |
July shows equity selling and bond investment at the same time. August shows a small positive stock figure and a negative bond figure. Within August bonds, the five-years-and-longer remaining-maturity bucket gained ₩2.130tn even as the overall bond total fell. These are coexisting aggregates, not a transaction-level trail connecting a particular stock sale with a particular bond purchase. FSS July release; FSS August release.
Why official and market-screen numbers can point in opposite directions
FSS reports settlement-based stock investment and excludes ETFs, ELWs and ETNs from its stock net-purchase measure. Its August market breakdown is KOSPI +₩1.888tn and KOSDAQ −₩1.545tn, versus the separate NAVER daily-display sums of −₩10.1656tn and −₩1.8394tn. These series must not be silently substituted or spliced.
The Bank of Korea’s September 17 release explicitly illustrates the date-basis issue: August equity flows were +$0.04bn on its settlement basis, while its trade-date footnote reported −$8.66bn. BOK also includes on- and off-exchange equity transactions and IPOs, with its own definitions. The sign difference is documented; a complete numerical bridge between FSS and the NAVER screen remains unreconciled. Timing alone has not been demonstrated to explain the entire won-denominated gap.
Printed official rounding also matters: FSS’s displayed market components sum to +₩0.343tn while its published total is +₩0.344tn. BOK reports a total −$4.50bn even though its displayed stock and bond components sum to −$4.49bn. Preserve the official totals rather than “correcting” them from rounded components.
For September, the Korea Center for International Finance’s October 1 public summary reports continued foreign outflows from both stocks and bonds. Only that public summary was available for this analysis; its full methodology is not treated as inspected evidence.
Neither a securities-account cash balance, an FX conversion nor the ultimate beneficiary destination is observed here. A rise in foreign holdings’ market value is also not an inflow: FSS’s August stock-holdings value rose by ₩25.066tn, far more than its +₩0.344tn net purchases. Valuation and other stock/flow adjustments matter.
6. What might investors be positioning for next?
These are research priorities inferred from the observed selections and company evidence, not a complete sector ranking or a return forecast. Different investors can buy for valuation, hedging, index changes or liquidity needs without sharing the same growth thesis.
Semiconductor equipment, testing and packaging
Evidence: selective stock buying; material disagreement between investor groups
HPSP’s institutional buying fits a business making semiconductor process equipment; Hana Micron combines recent foreign and institutional buying with packaging/test exposure. The company reported Q2 consolidated revenue of ₩683.2bn and operating profit of ₩142.3bn, and described capacity and product-mix expansion. SK hynix’s Q2 release described HBM4 shipments and customer commitments, providing an upstream demand context rather than proof of orders for every supplier. HPSP business description; Hana Micron’s August 28 release; SK hynix Q2 release.
What must hold: customer qualification converts into shipped tools, test utilization and cash-generating orders; expansion does not outrun demand. What weakens the thesis: order deferrals, lower utilization, margin erosion or buying that disappears outside isolated rebalancing days. Supplier-level revenue exposure and valuations still need individual underwriting.
AI-server components and package substrates
Evidence: recent two-sided investor buying; longer-window foreign selling remains
Samsung Electro-Mechanics reported Q2 consolidated sales of ₩3.4572tn and operating profit of ₩440.4bn. On September 1 it announced an approximately ₩1.0722tn AI-server MLCC supply contract for calendar 2027. That provides a concrete demand link alongside its substrate business. A contract amount is not current recognized revenue or cash collected. Q2 company results; 2027 supply agreement.
What must hold: delivery schedules, product mix and margins translate contracts into earnings and cash flow. What weakens the thesis: cancellations, customer concentration, capacity bottlenecks or an entry valuation already discounting more growth than contracts support.
Power distribution and electrical equipment
Evidence: a narrower foreign-stock signal, not broad sector agreement
LS ELECTRIC appears as a foreign net purchase in both stock windows: +₩70.569bn recently and +₩206.874bn over July 13–October 12. Its Q2 company release reported approximately ₩2.1tn of new orders and ₩7tn of backlog, with data-center distribution and transformer demand. Yet Hyosung Heavy Industries was sold by both investor groups in the latest table. Company-level selection is more defensible than a blanket electrical-equipment rotation. LS ELECTRIC Q2 release.
What must hold: backlog converts into profitable deliveries, manageable working capital and cash. What weakens the thesis: project delays, input-cost pressure, slower order replenishment or capacity spending that reduces returns.
Financials and shareholder returns
Evidence: persistent reported bank buying, heavily qualified by transaction structure
Bank institutional breadth is observable, but the foreign result is Woori-specific and block-trade-sensitive. Woori reported H1 net profit of ₩1.609tn and announced a ₩350bn annual share-repurchase/cancellation program. Announcement size is not assumed to be fully executed. This is primarily a profitability, capital-allocation and shareholder-return hypothesis, not automatically a high-growth industry thesis. Woori H1 results.
What must hold: sustainable return on equity, adequate capital, contained credit losses and actual cash distributions. What weakens the thesis: margin compression, higher credit costs or purchases explained mainly by repurchase execution rather than new portfolio demand.
Method, coverage and limits
- Window discipline. Exclude the reference peak session. Use each market’s own post-peak window for its headline history, a shared 74-session period for cross-market comparisons, and two nonoverlapping 20-session windows for recent change. The 1- and 3-month stock lists overlap and are never added.
- Units and checks. Daily NAVER amounts are displayed in ₩100 million; divide their sum by 10,000 for trillions. Stock-list amounts are ₩million; divide by 1,000 for billions. Hana weekly amounts are already ₩billion. Rounded daily category sums can differ by a few ₩100-million units. No derivatives notional is added to cash equities.
- Persistence and concentration. Count positive days/weeks and examine investor subcategories, not just one net total. The largest-day exclusion is sensitivity analysis, not a forecast or statistical anomaly label. No causal attribution or probability of future industry leadership is estimated.
- Size adjustment. Selected stocks use the signed net-volume ratio: rounded net shares ÷ traded shares × 100. It is not a portfolio weight. Comprehensive market-value weights, value-turnover normalization and full-sector breadth are unavailable; top-20 tables cannot supply them.
- Source hierarchy. FSS and BOK are primary official cross-asset sources. NAVER is the public vendor display used for reproducible daily and stock-level comparisons. Hana is the original broker source for the bank case. Company releases support business conditions, not investor intent. These panels have different definitions and are kept separate.
- Remaining gaps. The exact official/vendor reconciliation, account-level cash destination, complete post-peak sector totals and stock-level institutional subcategory flows are not established. Public trading labels do not prove the motivation of an investor or the identity of cash across transactions.
Questions to ask before treating a flow as a growth signal
Does the sign survive changing from a headline total to its investor subcategories? Does it survive removing the largest day? Are several companies involved, or one block trade? Do orders become revenue, margins and free cash flow? Is the valuation already demanding? A stronger conclusion needs these checks to align, not merely an attractive industry name.
Does a net sale prove capital left Korea? No. Its destination requires separate account, settlement or cross-border evidence.
Does institutional buying predict a rally? Not from these observations alone. Institutions can disagree, hedge, rebalance and execute transactions unrelated to a long-term growth view.
