Korea market structure · October 12, 2026
The near-term risk is a weaker buyer base as repurchases run down. Foreign accounts sold both chip leaders heavily, but the direction changes by investor and time window. Samsung’s latest buying reportedly ended October 6; SK hynix was still approaching completion. That creates a conditional downside risk for the KOSPI, not a mathematically inevitable fall.
20 sessions through October 8
Last five sessions
October 6–8, exchange notices
1. How much selling has actually occurred?
Over 20 sessions, foreign* net selling reached ₩6.507tn in Samsung and ₩16.298tn in SK hynix. Yet Samsung institutions bought a net ₩1.267tn over that window. Their position changed to net selling over the last five sessions. Retail investors bought both stocks on October 8 and over the last five sessions, but were net sellers over the full 20-session window.
This is therefore neither a uniform institutional exit nor a continuous retail rescue. The most persistent signal in this sample is foreign selling; the shorter window shows an additional shift in institutional demand. One-day flows should not replace the longer comparison.
Samsung Electronics · 1 session
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -0.348 | -1.307 |
| Institutions | -1.081 | -4.088 |
| Retail | +1.393 | +5.259 |
SK hynix · 1 session
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -0.866 | -0.513 |
| Institutions | -0.046 | -0.023 |
| Retail | +0.211 | +0.130 |
Samsung Electronics · 5 sessions
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -1.237 | -4.599 |
| Institutions | -1.167 | -4.440 |
| Retail | +1.019 | +3.980 |
SK hynix · 5 sessions
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -4.080 | -2.325 |
| Institutions | -0.304 | -0.176 |
| Retail | +0.299 | +0.212 |
Samsung Electronics · 20 sessions
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -6.507 | -25.202 |
| Institutions | +1.267 | +4.256 |
| Retail | -3.936 | -13.392 |
SK hynix · 20 sessions
| Investor | KRW tn | Million shares |
|---|---|---|
| Foreign* | -16.298 | -9.225 |
| Institutions | -0.739 | -0.454 |
| Retail | -3.689 | -1.842 |
Definitions, freshness and reconciliation
Net value is purchases minus sales in won; net quantity is purchased shares minus sold shares. These are separate measures. Net value divided by net shares is not a valid investor execution VWAP. The source’s money fields have KRW1 million resolution even though the field unit is won.
October 12 flow fields were still null when checked; null is not zero. The provider’s general metadata was provisional and carried a newer market date, so this article uses the last populated flow date rather than silently treating the metadata date as complete. It is an independent provider’s processed series, not a direct, independently certified KRX download.
For October 1, 2, 6, 7 and 8, all 20 institutional/retail stock-day rows matched Pinpoint News’s later-day KRX-attributed tables within published rounding. Earlier 15:30 snapshots can differ. Foreign figures are not directly comparable: the provider explicitly includes “other foreign”, while the comparison tables do not specify that inclusion; the differences were not fully reconciled and the series are not blended. Other corporations and other categories are not shown, so these three groups need not sum to zero.
2. Buyback completion is a sequence, not one date
Samsung: buying reportedly finished
The August 21 decision authorized about ₩15tn of common-stock purchases from August 24 to November 21. Its stated purpose was employee/executive share compensation, with the acquired shares held in treasury.
The October 6 exchange execution notice records 1,056,305 shares filled, worth ₩292.226bn, and cumulative fills of 56,856,305 shares. SBS Biz reported on October 12 that the program finished October 6. Samsung is absent from the complete October 7 and 8 execution notices.
SK hynix: still winding down
The August 19 filing authorized about ₩40.004tn, with a planned August 20–November 19 window. The company says shares acquired under this program will be cancelled after acquisition completion.
Actual fills were 400,000 shares on each of October 6, 7 and 8, worth ₩717.283bn, ₩697.989bn and ₩691.536bn respectively. October 8 cumulative quantity was 20,050,000 shares. An October 12 order for 400,000 shares and contemporaneous coverage still described the program as active or nearing completion.
Both initial share quantities were calculated from reference prices and could vary with execution prices. Exceeding an indicative quantity, reaching an originally planned deadline, filing a purchase result and cancelling treasury shares are different events. Nor is an announced budget the same as the budget still available today.
3. Put the flow in scale before predicting a price
At the October 8 reference close, Samsung was ₩262,000 with common-equity market capitalization of ₩1,531.725tn; SK hynix was ₩1,681,000 with ₩1,227.958tn. Their reported KRX-day turnover was approximately ₩5.617tn and ₩5.457tn respectively; the provider lists NXT separately. These are dated reference figures, not today’s live prices.
Twenty-session foreign* plus institutional selling equals approximately 0.342% of Samsung’s October 8 capitalization and 1.387% of SK hynix’s. This is a scale comparison, not an estimate of price impact: capitalization is not the available order book or free float. A small percentage of capitalization can still matter if marginal liquidity is thin.
| Stock | 1 session | 5 sessions | 20 sessions |
|---|---|---|---|
| Samsung | 6.19% | 6.03% | 7.20% |
| SK hynix | 15.88% | 17.92% | 14.32% |
Matching the dates and venue changes the interpretation: SK hynix’s foreign selling was roughly twice Samsung’s share of turnover over 20 sessions. Samsung’s 20-session traded value was ₩90.348tn, or ₩4.517tn per session; Hynix’s was ₩113.844tn, or ₩5.692tn per session. These are value-based liquidity measures, not share-volume ADV or an estimated price-impact coefficient.
Turnover is summed from the provider’s dated KRX market-cap snapshots for the same sessions, with the October 8 close, capitalization and turnover cross-checked against FnGuide. NXT is excluded. “Other corporations” net buying cannot be relabelled as the issuer’s repurchases; use the execution notices for that leg. A causal flow model would also need order-book liquidity and whether trades supply or demand immediacy.
4. The most defensible quantitative framework
Signed flow and participation
Sum daily net values and shares across actual sessions. Track 1/5/20-session direction and normalize by matching turnover. Check whether net selling is slowing before calling a reversal.
Issuer demand withdrawal
Use executed repurchases, not orders or announced budgets. Compare the buying period with the post-completion period while tracking price, depth and spread changes.
Index-weight arithmetic
Use beginning-period constituent weights. The stocks’ contribution is weight × return; the rest of the market can amplify or offset it. A combined market-cap headline is not automatically an official index weight.
Conditional abnormal returns
Compare each stock with a market/sector counterfactual and separate earnings, FX, rates and index-rebalancing news. Flow-return correlation alone is not causation.
Why a square-root impact equation is useful but insufficient
A common execution heuristic is impact ≈ Y × daily volatility × square root of (order quantity / daily volume), with the order’s direction supplying the sign. It concerns large executed parent orders, not a daily investor category’s net flow. Y and the permanent-versus-temporary component must be estimated for the market and execution style.
Research also finds that duration and participation rate matter and that the square-root approximation has a limited range. Subtracting a buyback budget from market capitalization, or inserting aggregate foreign selling into this equation with an arbitrary coefficient, does not yield a defensible KOSPI target. This article therefore does not manufacture such a target.
What a proper event study would require
Estimate a pre-event model such as stock return = intercept + market factor + semiconductor factor + FX factor + residual. Prefer a benchmark excluding the two stocks, or explicitly address their heavy overlap with the KOSPI. Abnormal return is the observed return minus that model’s prediction; cumulative abnormal return sums the residuals through the event window.
Pre-specify the event as the last executed purchase date; distinguish it from the announcement and disclosure dates. Test alternative windows, robust errors, pre-trends and placebo dates. A panel of comparable completed programs would help separate common shocks from issuer-specific events. October 8 also coincided with Samsung’s earnings release, options expiry and reported ETF rebalancing, so attributing the whole move to buyback completion would be unsound.
The available 20-session flow sample and very short post-Samsung-completion window do not support a calibrated causal coefficient or probability forecast. No fitted regression, statistically significant effect or backtest result is claimed here. A forecasting model would need a longer, point-in-time price/flow history, out-of-sample testing and trading-cost assumptions.
5. How the two stocks can move the KOSPI
For a single period with fixed beginning weights and no special index adjustments, approximate index return (%) = Samsung weight × Samsung return + SK hynix weight × SK hynix return + remaining weight × remaining-market return. Weights enter as fractions. This is an accounting sensitivity, not a forecast of any component’s return.
Illustration A: both chip stocks fall 5%, the rest is flat
Assume Samsung starts at 25% of the index and SK hynix at 20%. Their combined contribution is (0.25 × −5%) + (0.20 × −5%) = −2.25 percentage points. If the remaining 55% is flat, the illustrative index return is −2.25%.
Illustration B: both chip stocks fall 5%, the rest rises 2%
The chip contribution remains −2.25 points. The remaining 55% contributes 0.55 × 2% = +1.10 points, making the illustrative index return −1.15%. A diversified rally offsets some of the chip drag.
Illustration C: Samsung falls 5%, SK hynix falls 10%
With the same assumed weights and the rest of the market flat, (0.25 × −5%) + (0.20 × −10%) = −3.25%. Different stock returns matter even when combined weight is unchanged. None of these scenarios has an assigned probability.
6. Which direction is most reasonable to expect?
Judgment: the evidence supports near-term fragility and a downside bias conditional on persistent selling after issuer demand disappears. It does not establish that a further fall is the most likely outcome with a measurable probability. The size of the observed foreign selling and the recent institutional shift justify watching the transition closely; a price target would require evidence this dataset does not provide.
Continued pressure
Foreign* and institutional five-session flows stay negative, actual issuer fills stop, and prices make lower lows with broad market weakness. In that combination, both chip-stock contributions and the rest-of-market term can be negative. This is the clearest continuation case.
Absorption and stabilization
Selling slows or remains high while prices stop falling, spreads normalize and non-chip breadth improves. That would suggest replacement demand is absorbing supply. Buyback completion alone would then be a poor bearish timing signal.
Relief and renewed demand
Foreign flows turn positive across several sessions, institutions confirm and earnings expectations or global semiconductor conditions improve. A rebound could occur despite completed buybacks. A single positive session is insufficient to establish that regime change.
The next checks should be actual Hynix fills and completion evidence, rolling flow changes, price response per unit of selling, market breadth and the behavior of peer semiconductor markets. National-market confidence is a separate hypothesis: foreign selling alone is not evidence of a sovereign rating downgrade. Testing a Korea-specific risk premium would require currency, sovereign credit spreads and sector-adjusted relative valuation alongside these stock flows.
Sources and calculation notes
- maedong.kr public API definitions, Samsung series, SK hynix series and source limitations. This article publishes a small derived aggregate summary, not a raw-data mirror.
- Evening cross-checks: Pinpoint News for October 1, October 2, October 6, October 7 and October 8.
- Samsung’s August 21 buyback decision; SK hynix’s August 19 KIND filing; SK hynix’s official explanation.
- Exchange execution notices reproduced by Maeil Business: October 6, October 7, October 8.
- Completion-status reporting: SBS Biz, October 12; Newsway, October 12 at 15:17 KST; October 12 buyback orders.
- October 8 close, capitalization and separate KRX/NXT turnover: FnGuide’s IBK company monitor for Samsung and SK hynix.
- KRX on market-cap weighting; MacKinlay (1997), event-study methodology; Zarinelli and co-authors, market-impact limits. These sources support the methods, not a fitted forecast for these stocks.
- October 8 earnings, expiry and rebalancing context. Multiple simultaneous events complicate causal attribution.
General market research and education. No personal investment recommendation, guaranteed outcome or calibrated trading signal is offered. Facts, derived calculations and conditional judgments are identified separately; newer prices and filings can change the analysis.
