Where Resilience Comes From

Samsung and SK hynix share buybacks have begun shrinking. Below, we examine other-corporation flows, which include them. The measure is net buying: purchases minus sales. A positive value means more shares were absorbed from the market that day; a negative value means more were supplied.

KRX's other-corporation category includes general corporate trading as well as companies buying their own shares. It is not identical to company-level buyback amounts. Still, over 8/20–9/30, other corporations bought KRW 44.76 trillion on the KOSPI. Foreign investors sold KRW 28.78 trillion and individuals KRW 17.67 trillion. The difference shows who absorbed the selling then.

Samsung Electronics reportedly completed its KRW 15 trillion buyback on 10/6★. SK hynix continued buying but reportedly reduced daily purchases★. We can see how this difference appears in the chart below.

The green bars show other corporations' daily net purchases. The brown line averages the preceding 5 sessions, making changes in buying pace easier to see than daily swings. Lower bars mean companies absorbed less money's worth of shares from the market.

Purchases around KRW 1.64 trillion a day since late August declined to KRW 1.48 trillion on 10/2, KRW 1.01 trillion on 10/6 and KRW 0.71 trillion on 10/7. The latest observation, 10/8, is KRW 0.75 trillion. The brown line is falling too, but its 5-day average remains KRW 1.12 trillion. Less buying and buying ending altogether must be distinguished.

The right-hand dashed line marks SK hynix's expected completion. It does not extend observed values into the future. Reports differ: the week of 10/13, mid-October or around 10/20★. Aggregate other-corporation flows also cannot establish that all remaining buying is SK hynix.

Who fills the gap remains the question. Individuals' investor deposits available for trading were KRW 101.05 trillion; margin loans financing stock purchases were KRW 33.56 trillion (10/6). The balances differ in nature and must not be added as buying capacity. They need to be read alongside actual individual net buying.

Export receipts can support the won while foreigners sell Korean stocks. When a foreign investor sells shares and converts won proceeds into dollars to take out, that creates won selling. Exporters converting dollar receipts into won create won buying. Which is larger requires evidence of actual transactions.

This section introduces the current account and portfolio-investment flows. The current account is the difference between money earned abroad and paid abroad through goods, services, income and other items. Portfolio-investment flows include Korean residents' overseas investment as well as foreign investment in Korean stocks and bonds. They are therefore not the same as foreign KOSPI net selling.

August's current-account surplus was USD 46.11 billion (Bank of Korea release, 10/8). That indicates a large channel earning export dollars. It does not mean all those dollars immediately become won; companies can retain foreign-currency deposits.

The top of the chart shows flows month by month. Green bars are the current account; red bars are net portfolio-investment flows. The lower brown line sums both flows over the preceding 12 months through each month. The periods differ, so a single bar's height should not be compared directly with the line.

The trailing 12-month current-account surplus through August was USD 332.59 billion. Adding net portfolio flows over the same period leaves a balance of USD 143.10 billion (calculation). The observation is that the current-account surplus outweighed portfolio outflows.

This balance is neither actual dollar conversions nor the share explaining won appreciation. Direct investment and other financial transactions are excluded. The chart establishes only that stock selling alone misses an offsetting flow.

Read the exchange-rate direction too. Naver's comparable daily posted USD/KRW values fell from KRW 1,552.5 on 7/1 to KRW 1,341.0 on 10/7, a decline of KRW 211.5. Fewer won were needed to buy a dollar, so the won strengthened. These values differ in basis from the Seoul FX market's official 15:30 close.

The same semiconductor revenue can affect the won and stocks through different channels and timing. Exporters converting dollars into won act first in the FX market. Decisions to spend profits on share buybacks or dividends reach equities on a separate schedule.

These links are possible channels to examine. Arrows do not make every step automatic. Companies can hold dollars; higher profits can fund factories instead of buybacks. Strong semiconductor exports alone therefore cannot establish the direction of the won or KOSPI.

Korean semiconductor export prices were 181.9% higher year on year in August (saved series). Samsung Electronics' official 10/8 preliminary guidance put Q3 operating profit at about KRW 107.4 trillion. These are not yet final audited earnings. The specific total for the indicated roughly KRW 30 trillion cash dividend will also be set by the board in late October.

Samsung and SK hynix represented 28.3% and 23.1% of KOSPI common-share market capitalization on 9/30, without free-float adjustment. That is why their profits and shareholder-return changes deserve separate attention in the index.

This chart rebases two differently measured series, the KOSPI and won value, to 100 on 7/1. Green below 100 means the KOSPI is below that day's level. The brown line is not the exchange rate itself but Won value, so higher means a stronger won. Its vertical direction is opposite to a USD/KRW chart.

The widening gap suggests stocks and FX are difficult to explain as one money flow. It does not prove memory revenue caused the gap. The chart's lines stop at 10/7; the pale area to the right is the period preregistered for future observation.

We preregistered a single comparison of changes in the two prices on 10/30. That establishes whether prices diverged. The causal explanation in the registration below still needs additional evidence, such as company-level currency conversion.

To examine Korean semiconductor revenue, we also need the investment capacity of AI companies ordering the components. Companies expanding data centers must buy servers and memory. They fund this with operating earnings, investors or borrowing. Higher borrowing costs can reduce investment possible with the same money.

The 4 major US hyperscalers, or large cloud companies, have combined 2026 capex plans of USD 720–745 billion★. These are company-wide capital expenditures, not entirely AI investment. They are still large spending plans underpinning component orders.

The next chart presents borrowing costs in two ways. The left-hand spread shows compensation added above a reference rate for taking risk. The right-hand yield shows the total yield reflected in the bond or loan price. They are different measures.

The left-hand bp unit expresses small interest-rate differences. 100bp equals 1 percentage point; 60bp means demanding 0.60 percentage points above the reference rate. ICE's CCC index, a low-credit-quality corporate basket, was 1,211bp on 10/5. That means the spread alone was 12.11 percentage points. It must not be compared as equivalent to CoreWeave's 10.44% on the right★.

Even the left-hand bars are not identical contracts. Oracle measures the price of credit-default protection; other bars show spreads on corporate bonds and similar instruments. With different dates and products, this chart does not aim for a precise corporate ranking. It shows how widely borrowing terms differ across borrowers.

CoreWeave's official Q2 earnings recorded an operating loss of USD 49 million and interest expense of USD 640 million. It illustrates a company bearing interest costs before earning an operating profit. We do not generalize from this company to the whole AI industry.

Memory suppliers, by contrast, can benefit from price increases. TrendForce forecast Q4 DRAM contract prices up 10–15%; Micron's GAAP operating margin for the relevant quarter was 80.7%. Within the same capex cycle, component sellers and companies building facilities with debt can see different profit outcomes.

For that difference to last, buyers must keep ordering. Alongside high memory margins, ask whether customers can keep buying at those prices. The next section examines the base price of the money those customers borrow.

The recent rise in US long-term yields is not explained by policy-rate expectations alone. The companies in the previous section borrow at a reference rate plus compensation for their own risk. If the underlying US Treasury yield rises, their funding costs can rise even if credit quality is unchanged.

Think of the 10-year yield in two parts. One is the expected path of future short-term rates. The other is extra compensation for the uncertainty of tying money up for a long time: the term premium. The decomposition below is a New York Fed model estimate, not direct observations of two separately traded prices.

Each bar shows the yield change from its stated starting date through 10/6. Brown is the term-premium change; green is the risk-neutral yield change representing the expected short-rate path. Dots add the two components. A green piece below zero moved toward lower yields.

Since 9/16, the total yield rose 26.51bp. Brown rose 28.04bp; green fell 1.53bp. In this model, compensation demanded for holding long term outweighed higher expected future short rates in the recent rise.

The 12-month bar differs. Of the total 102.48bp rise, 72.26bp came from the expected short-rate path and 30.22bp from term premium. The explanation for recent weeks cannot be applied to the entire preceding 1 year.

The Fed minutes' market discussion also mentions heavy private debt issuance competing for capital. AI-related corporate bonds and Treasuries seeking the same investors' money could demand higher compensation. This alone does not establish AI as the cause of the latest yield rise.

Our hypothesis extends that far: more expensive long-term funding can burden AI investment, potentially reaching memory companies through orders. The size and lag at each step remain unknown. We therefore do not translate higher rates directly into lower Korean exports.

Evaluating today's price first requires stating the comparison. For stocks, we compare profits and Treasury yields; for crude oil, model values based on supply/demand assumptions and market prices. Neither is safely read as a calculation of one correct price.

For stocks, corporate profits relative to share prices can be expressed as a yield. Subtracting the Treasury yield gives a rough comparison of equity earnings versus government-bond interest. This report uses it as a proxy for the equity risk premium. Stock earnings still differ from guaranteed interest.

The first bar uses earnings already reported. Subtracting the 5.27% Treasury yield from 3.75% earnings/price gives −1.52 percentage points. Earnings are as of 10/7; yields, 10/6. Negative means the equity earnings yield is lower in this comparison, not that a stock decline is certain.

The second bar uses expected future earnings. Inverting FactSet's forward P/E of 19.0x gives an earnings yield around 5.26%. The gap versus the nominal Treasury yield is almost zero. The third separately uses an inflation-adjusted real yield, giving about +2.25–2.35 percentage points. We do not average the three into one signal.

The picture changes with the earnings period and yield benchmark. In the next earnings reports, we therefore need actual cash inflows as well as accounting profits. FactSet expects Q3 earnings growth of 29.5%. How this becomes reported profit and cash matters.

For crude oil, start by separating actual transaction prices from model calculations. December Brent traded at USD 100.58 per barrel on 10/6. Our model's median in the saved 10/7 version is USD 114.8. The model translates assumptions, such as the supply shortfall, into a price calculation. It is neither an actual price nor a correct future price.

In the next chart, A is the sequence of model adjustments, B compares the model with the market price. A's three numbers are not oil prices from different days. They show how calculations change when assumptions change under the same conditions.

① Before adjustment — USD 160.5. The starting value uses the supply-demand shortfall, translating the relationship between greater scarcity and higher prices into a calculation. It is not an actual oil price.

② Demand reduction — USD 155.7. Consumers can use less oil when it is expensive. Adding that extra demand reduction shrinks the modeled shortfall and lowers the calculation by about USD 4.8. It is a model assumption, not an observation of consumption falling that much.

③ Inventory benchmark — USD 114.7. If the initial shortfall persisted, stored oil should also decline rapidly. Saved global inventory data instead showed a smaller draw. Recalculating the shortfall against that benchmark lowers the result by another USD 41.0. This does not mean ample inventories justify a USD 41 discount.

B therefore shows more than one number. The ocher USD 100.1–129.4 band is a model range under varying assumptions. The vertical marker, USD 114.8, is that range calculation's median. The black dot below, USD 100.58, is an actual traded price. The market price is near the bottom of this range.

This does not establish that the market is wrong or oil will rise to USD 114.8. A shows how strongly changing the inventory benchmark alters the result. When fresh global inventory data arrive, first check whether the benchmark still fits. August observations used here exclude September's supply disruption.

Recovering crude supply does not immediately bring diesel prices down. Crude is refinery feedstock; diesel is a product made through refining. Even after feedstock arrives, constraints on refinery operations or product transport can slow diesel recovery.

The left of the chart examines how much volume has returned. With prewar volume at 100, crude flows were about 80 in September and diesel/gasoil net exports about 25 in August. Crude is a flow; diesel is exports minus imports, and dates differ. The gap of 55 cannot be calculated as volume blocked at refineries.

The right examines how much more diesel costs than crude. Subtract crude prices from diesel prices on the same per-barrel basis to get the diesel crack. If crude is USD 100 and diesel USD 125, the difference is USD 25. This illustrates the method, not current prices.

The first right-hand bar, USD 25, is our gauge comparison benchmark. The US Gulf value for 9/25 is USD 99.53. On that basis, diesel cost about USD 100 more per barrel than crude. It does not mean diesel itself cost USD 99.53.

The other bars also record different dates and bases. Our saved USD 95.94 lacks an actual diesel observation date; 10/7 is the file-generation date★. The 9/14 reported high of USD 118.62 awaits direct source checking★. Do not join the bars into a time-series rise or fall.

Even a large crack is not all refiners' net profit. Energy, labor and transport costs have not been subtracted, and facilities produce different product mixes. The question in this chart is whether product-supply constraints persist while crude supply returns. Which specific facility caused them needs separate evidence.

A similar question applies to AI. More computing facilities can still leave needed memory scarce and its price high. Rather than total supply alone, both require checking which stage constrains the next unit of production. This does not claim the two industries move at the same speed.

Expensive diesel does not produce the same outcome for every airline. United reportedly raised its profit outlook despite higher fuel costs★. Pricing power and other costs differ. Keep the bottleneck's location separate from the ultimate profit beneficiary is the principle for comparing these two industries.

Read the next observations in this order: buyback completion, customers' investment plans, then the price of money. A single day's stock reaction cannot decide whether the whole connection holds. Distinguish what company disclosures and actual transactions confirm at each stage.

First, look for other buyers after companies reduce buybacks. Second, check whether large AI companies buying memory maintain capex plans and actual cash spending. Third, examine changes in the rates and credit terms financing those investments.

Three scenarios, not verdicts. These conditional accounts organize what to check next. They assign neither probabilities nor trading signals.

• Investment and orders continue. Check whether the won holds up after SK hynix's buyback ends, customers maintain capex and term premium stays below 1%. Causes still require separate checks after each condition is observed.
• Customers' finances deteriorate first. Watch for worse corporate-bond terms or investment plans and a wider burden on low-credit-quality firms. Then check whether memory orders actually decline.
• The broad market's rate burden rises first. Watch for term premium above 1% or stronger signals of further hikes. Whether rates reach actual funding terms and investment spending requires the next stage of evidence.

These gauges continue checking the preceding questions. The first shows corporate buying, the second FX, and the third the won/KOSPI observation window. The rest show extra compensation for long-term funding, low-grade corporate spreads and the diesel/crude price gap.

No single color or figure judges the whole market. Read each item's date first; they may differ. Body/charts record the writing date; gauges read the latest available observations when the page opens.

We have confirmed that corporate buying fell and stocks and the won did not always move together. Export receipts supporting the won are a plausible channel to examine. Without company-level conversion evidence, however, we do not establish causation.

The next report first checks who filled the gap after corporate purchases ended, then retrieves the won/KOSPI outcome over the fixed window. Finally, it compares AI-company investment spending and funding terms to see whether money backing memory orders continues.

Boboo follows more than a single price: actual transactions among those earning, spending and lending money. Checking each connection makes the remaining source of resilience clearer.