Revisiting the Forest This Weekend

The week's pattern was one day of relief, followed by two days of sharp falls. Monday rose on institutional net buying after the weekend release of October exports (semiconductors +259.4%), according to reports. Tuesday fell as reports of debt-funded AI investment coincided with caution ahead of the election. Wednesday, when the result was confirmed, recovered from a steep intraday fall to close higher. Thursday and Friday coincided with sharp declines in U.S. semiconductors.

• 11/2 (Mon)October exports USD 115.28 billion (+93.7%), semiconductors +259.4% — KOSPI +2.24%
• 11/3 (Tue)Reports of continued Big Tech bond issuance and Oracle raising USD 20 billion for AI infrastructure — KOSPI −3.59%
• 11/4 (Wed) 13:40AP's bylined article reporting its Senate majority call was published. AP's House call article followed at 15:27 — KOSPI closed +0.62% after an intraday −4.02%
• Early 11/5 (Thu)U.S. 11/4: S&P500 +1.58%, Dow +3.03%. Philadelphia Semiconductor Index −3.13%, Micron −7.00%
• 11/5 (Thu)KOSPI −5.28%. Sell sidecar (program-trading halt) at 10:16. Foreign net selling concentrated in the two semiconductor companies
• Early 11/6 (Fri)U.S. 11/5: Nasdaq 100 −6.36%, Philadelphia Semiconductor Index −12.25% (largest fall since March 2020)
• 11/6 (Fri) 09:03KOSPI circuit breaker (4,924.61, −8.52%) — closed −6.02%
• 11/6 (Fri) 22:30U.S. October payrolls +165,000 (expected 59,000–65,000)
• Early 11/7 (Sat)U.S. 11/6: S&P500 +0.31%, Philadelphia Semiconductor Index +4.21%. Weekly S&P500 −2.7% (worst week since April 2025)

Reading · Wednesday, when election results were confirmed, actually closed higher. The two steep down days were AI semiconductor days, centered on U.S. chips. Hong Kong and Shanghai rose that week. It is difficult to explain the week's decline through the election alone.

The election was closer to confirmation than to new information. On counting day, prediction-market prices had already tilted to one side before counting ended in the battleground states (#36 Boboo from Sky live record; prices are not probabilities). AP's bylined article reporting its Senate majority call was published at 13:40 KST on 11/4; the House call article appeared at 15:27. KOSPI fell as much as −4.02% intraday before closing +0.62%.

The first U.S. session after the result (11/4) brought relief. S&P500 +1.58%, Dow +3.03% and Russell 2000 +1.58%; the VIX fear index fell from 20.83 to 16.61. Yet semiconductors and data-center stocks fell separately that day, then tumbled the next.

The left column records price directions and interpretations reported that week. It does not establish causation.

Some races remain uncalled. In Alaska's Senate race, AP's 11/4 local-time count put the candidates 626 votes apart (more than 98% of precincts reporting, with mail ballots still arriving). A ranked-choice count takes place on 11/18 if no one wins a majority. Mississippi's Senate race had a 2,010-vote gap in AP's count and awaited the 11/13 county certification deadline (as of 11/6).

Reading · The election changed the stories used to price assets. Data-center regulation, oil and defense were central to those stories. Whether the stories become numbers—legislation, capital expenditure (capex) guidance and earnings—will be tested over the coming weeks.

In the week AI infrastructure baskets tumbled, platform and content baskets rose. Boboo from Sky's rotation ledger shows the largest falls in memory at −23.8%, semiconductor equipment at −14.6%, and AI power/data centers at −14.5%. Platforms rose +7.7%; content/IP rose +4.3%.

Memory costs were already showing up in buyers' earnings. Apple warned that rising memory costs could squeeze margins; its shares fell about 1.9% on 11/3 (AP). Qualcomm reported a 26% decline in smartphone chip revenue. Boboo from Sky's 11/5 observation read that day's Asian moves in Sony, Xiaomi and Nintendo as expectations of a turn lower in memory prices (interpretation). But the consumer-device basket was almost flat for the week. As specified in the 11/5 observation design, we will measure again after 5 and 20 trading days to see whether this path continues.

China and Hong Kong moved the other way. On 11/2, the People's Bank of China lowered the relending rate for share buybacks from 1.25% to 1.00%. Platform and device companies such as Tencent (+9.4%) and Xiaomi (+14.7%) led the indexes. Taiwan, Korea and Japan, with their concentration of AI semiconductor suppliers, fell most; markets with less of that supply chain were less shaken.

Reading · This week's divide was less about country than about exposure to AI infrastructure. Even within Asia, markets with more semiconductor equipment and memory fell sharply.

KOSPI's decline was similar in size to that of the U.S. semiconductor index. Over the week, Nasdaq 100 fell −4.91%, the Philadelphia Semiconductor Index −12.15%, and KOSPI −11.71%. The large weight of two companies in the index appears to be one reason. Samsung Electronics fell −16.13% and SK hynix −20.17%.

Friday morning unfolded as follows. At 09:03:33, KOSPI triggered a circuit breaker at 4,924.61, down 8.52%, halting trading for 20 minutes. A KOSDAQ sell sidecar followed at 09:06; after trading resumed, a KOSPI sell sidecar was triggered at 09:34 (the 30th this year). The intraday low was 4,911.28. Around midday (11:50), it recovered to 5,215.58 (Yahoo 5-minute bars), then closed at 5,059.69. The two-day decline was −10.98%.

Measured using the 1-minute bars from Boboo from Sky's live election-day record, KOSPI moved 1.51x as far as Nasdaq 100 futures during the decline and 2.26x as far during the rebound. On Friday, SK hynix (−10.44%) fell by a similar magnitude to the Philadelphia Semiconductor Index the night before (−12.25%). On days when AI semiconductors moved sharply, KOSPI again acted like their lever this week.

Reading · KOSPI's larger swings than the broad U.S. indexes are better explained by its AI semiconductor weight than by news about the Korean economy. Whether KOSPI outside those two companies can stand on its own (question Q6) remained unanswered this week.

Foreign investors were net sellers on all five days. Same-day reports sum to −KRW 11.0734 trillion. The buyers differed each day. Retail investors absorbed the most on Tuesday (+KRW 2.6038 trillion) and Thursday (+KRW 2.2722 trillion); institutions led on Friday (+KRW 1.7041 trillion). Pension funds were net buyers on all five days.

Foreign net selling concentrated in the two semiconductor companies. Over 11/5–6, it totaled −KRW 3.3551 trillion in SK hynix and −KRW 1.2772 trillion in Samsung Electronics. October's foreign net selling in KOSPI totaled −KRW 18.6756 trillion (EToday compilation). On the same basis, the daily average over 11/2–5 (−KRW 1.9435 trillion) was more than twice October's daily average (−KRW 933.8 billion).

The government moved to reduce forced liquidations. From 11/9, brokerages will be exempt for 3 months from the requirement to maintain a 140% margin-loan collateral ratio. They can lower collateral requirements or allow more time for top-ups, but this does not halt all forced liquidations. Taiwan tightened short selling on the same day. From 11/9, the daily borrowed-share short-selling limit falls from 30% to 3% of 30-day average volume, and the minimum margin rises from 90% to 130%.

Reading · Debt had already fallen substantially. Margin loans were a quarter below their June peak; deposits were down by more than a third. This decline came after deleveraging, rather than at the point of heaviest borrowing. Days when institutions replaced retail investors as the main buyers were a new feature of this week's flows.

How to read it · “Other corporations” means ordinary companies, including listed firms buying back their own shares. “Pension funds, etc.” is an exchange category combining pension funds such as the National Pension Service (NPS) with mutual aid associations and others; it is not the National Pension Service (NPS) alone.

Long-term yields fell on the day AI semiconductors tumbled. That did not translate into room for the Fed to cut. The U.S. 10-year yield fell 7bp over the week, from 4.93% to 4.86%, touching 4.84% on 11/5 as semiconductors collapsed (this week's answer to Q3: do yields fall when technology stocks fall?). But much stronger-than-expected payrolls on Friday pushed yields back up. The December hike probability implied by futures prices (CME FedWatch) rose to about 80%, from about 67% the previous day.

These columns group forces reported during the week. They do not predict which will prevail.

The dollar index rose +0.72% on Friday (Yahoo close), and USD/KRW rose KRW 14.5 on 11/6, its largest daily increase since 5/12. Gold rebounded from its 10/30 1-year low of USD 3,869.85 to rise about 5% for the week (reported). Bitcoin regained USD 80,000 just after the election before falling back to USD 75,463.

Reading · The path in which a cooling AI boom lowers yields and gives the Fed room to cut appeared for only one day this week. The Fed watches employment and inflation, and this week's jobs numbers pointed the other way. U.S. CPI on 11/10 is the next fork in the road.

Prices fell, but money already earned in the memory boom has a calendar. Boboo from Sky's ledger (measured on 11/3) estimates that higher memory prices brought USD 446–513 billion to memory companies over the latest 4 quarters, with 59% going to the two Korean companies. In Korea, about 3.4–3.6% of GDP is waiting in corporate coffers. From November 2026 to April 2027, an estimated KRW 248–311 trillion is due to flow out as taxes, dividends, bonuses and buybacks (Boboo from Sky estimate).

• 11/27 (Fri)Samsung Electronics pays about KRW 30 trillion in Q3 dividends (KRW 4,601 per share)
• 10/29~1/28SK hynix's additional KRW 20 trillion buyback period (DART filing) — KRW 10 trillion for cancellation, KRW 10 trillion for employee compensation (reported)
• DecemberSK hynix pays about KRW 14 trillion in Q3 dividends (KRW 20,000 per share; record date 11/30)
• January 2027Samsung Electronics board meeting — KRW 55–75 trillion remaining in 2026 shareholder returns (timing undecided)
• March–April 2027Remaining corporate taxes, bonuses and year-end dividends — about KRW 130 trillion (Boboo from Sky estimate)

The main recipients are the government (KRW 140–205 trillion in taxes) and shareholders (KRW 190–210 trillion promised for 2026). Cash reaching households is at most around KRW 8 trillion. As foreign investors sold the two companies this week, the future recipients of dividends and buyback proceeds changed too.

Reading · Share prices move 10% in a day, but payout dates for money already earned do not move with them. We will reconcile this ledger with the mid-November quarterly filings (deadline 11/16).

What does the Deep Value Index measure? Deep value describes a price that appears unusually low relative to a company's earnings, assets and shareholder returns. Boboo from Sky's index measures proximity to that state on a 0–100 scale. It combines cheapness against the stock's own history, a dip relative to its long-term trend, and the earnings and shareholder returns available at the current price.

The score is neither a discount rate nor a probability of future gains. A score of 75 does not mean a stock is 75% cheap. We compared 223 individual stocks with all three measurable pillars, from a watch universe of 393, at their 11/6 closes.

What goes into it? Prices are the 11/6 closes; assets use the latest disclosed book values; earnings use net income over the latest 4 quarters. P/B is price divided by book value per share. Earnings yield is net income divided by market capitalization. Dividend yield is dividends per share divided by price; shareholder yield is dividends plus share buybacks divided by market capitalization. Rather than add these unlike measures directly, we convert their positions within the relevant comparison set to 0–100 scores.

For example, Kia's 74.3 is the sum of contributions of 33.2 from Value, 13.2 from Trend and 27.9 from Absolute. The three bar colors show those contributions. Cheapness against a stock's own history can differ from substantial returns available now, so read each pillar alongside the total.

By sector, software and Big Tech had become cheapest against their own histories. Software faces fears that AI can replace it; Big Tech faces fears that capital expenditure (capex) will consume profits. Semiconductors and AI hardware had fallen sharply—a median −45.5% from their highs—but their average Value pillar score was a low 42. Prices had risen much faster than book value and earnings during the boom, then retraced only part of that rise.

The left column contains names that may prompt “Why deep value?” The right shows quality companies that rarely fall far, but have dipped by their own standards. Neither is a buy list.

Good companies do not fall far. So depth, too, is measured against each company's own history. Over 10 years, Apple's deepest 10% of days were −20.0% below its 2-year high; Berkshire's were −14.5%. Berkshire had no days below −30% over those 10 years. A uniform threshold such as “−30% from the high” would never catch these companies' dips. Recalculating the Trend pillar at each month-end using only data then available (2019–2026), Apple had 20 months with Trend scores of at least 60. The median price change 12 months later was +42.9%, versus +27.1% for all months. Berkshire's figures were +20.8% versus +15.1%. These quality companies were selected after the fact, creating substantial survivorship bias. Read this only as confirmation that the Trend pillar captures shallow dips, not as evidence of predictive power.

Reading · The closest to deep value now were not semiconductors, but software and Big Tech facing AI concerns, and Korean autos, brokerages, shipbuilders and entertainment companies that fell sharply after this year's frenzy. Within them, those offering substantial returns (Kia and brokerages) differ from those offering little (NAVER and Kakao). The three pillars measure different things; where only one pillar scores highly, examine why.

The floor band does not simply copy historical low prices. It applies the book-value multiples or dividend yields observed at past selloff lows to the current book values and dividends fixed in the 11/6 note. This puts past lows on a comparable basis after changes in a company's assets and dividends.

1. Select historical lows. Identify each stock's lowest close within historical stress episodes, counting a low only if it was at least 25% below its preceding 2-year high. The bands mainly use lows from 2015 onward (memory from 2011; SoftBank from 2020).
2. Translate into prices on the current basis. The P/B lens is current book value per share × historical floor P/B. The dividend lens is current annual dividends per share ÷ historical floor dividend yield. The median historical floor measure gives the typical floor; the measure giving the lowest price (minimum P/B or maximum dividend yield) gives the deep floor.
3. Set the current share price to 100. The band is floor price ÷ 11/6 close × 100. Samsung Electronics' deep floor of KRW 78,171 and typical floor of KRW 95,977, compared with its KRW 188,700 close, place its band at about 41.4–50.9. The dot is the current price of 100; the −49.1% label is the distance to the typical floor.

Calculation assumptions · Book values and dividends remain fixed at the 11/6 note's values; only distances were recalculated using 11/6 closes. Samsung Electronics and SK hynix book values per share are estimates incorporating dividend and buyback adjustments. Where floor multiples have risen over time, the band may appear excessively low. It is neither a price target nor a forecast of the next low.

These bands tell the same story as the three-pillar index. Even after large declines, memory and equipment stocks remain high on historical floor multiples. Companies with long dividend records (TXN and Qualcomm), and those with a history of crises (Intel and SoftBank), sit closer to their bands. Yet Korean memory tells a different story under another measure. Shareholder returns in 2026—about KRW 100 trillion for SK hynix and KRW 90–110 trillion for Samsung Electronics—exceed shareholder yields at any previous floor (11/6 note). They look far away on book value and close on shareholder returns. That difference boils down to one question: how many years can profits like 2026's continue?

Reading · This week, semiconductors became less expensive rather than cheap. Floor bands record past selloffs; they do not predict the future. In sectors where floor multiples rose each cycle, past bottoms occurred at progressively higher multiples.

The weekend follows a highly volatile week. Rather than guess the direction, here are five ideas to carry into next week.

1. Falling far and becoming cheap are different

   Samsung Electronics fell −16.13% this week, reaching its deepest drawdown from a high in 10 years. Yet it remains +61.3% above its 10-year trend line and far from historical floors measured by book value. The size of a fall does not tell us how expensive the starting point was.

2. Cheapness and substantial returns are different too

   NAVER sits near its historical cheap end (Value 92), but its returns available now are low (Absolute 34). “Why deep value?” requires separate answers about cheapness and returns.

3. Depth can be shallow for a good company

   Apple's “deep days” (the deepest 10%) were −20.0%; Berkshire's were −14.5%. Waiting by someone else's yardstick (−30%) can mean waiting for a level that never arrives. Depth is measured by each tree's own rings.

4. KOSPI acted as a lever on AI risk assets

   This week's KOSPI move (−11.7%) was closer in size to the Philadelphia Semiconductor Index (−12.1%) than to Nasdaq 100 (−4.9%). When looking at Korea's index, it helps to distinguish the Korean economy from AI semiconductors.

5. Keep stories and numbers separate

   Data-center regulation, an Iran ceasefire and tariff relief were stories moving prices this week. When dated numbers arrive—legislation, capital expenditure (capex) guidance, earnings and inflation—we can measure those stories again. Dates for money already earned (11/27 dividends; March–April taxes) are numbers of that kind too.

What not to read into this week · “It fell because of the election” (same time ≠ cause). · “A circuit breaker signals the bottom” (this was the 11th this year alone). · “Foreign investors have left” (this is one week's flow). · “An 80% chance of a December hike” (a number calculated from futures prices).

• 11/9 (Mon)Margin-loan collateral-ratio exemption begins · Taiwan's short-selling restrictions take effect 05
• 11/10 (Tue) 22:30U.S. October CPI — another test for December rate pricing 06
• 11/11 (Wed) 07:00CoreWeave Q3 earnings — numbers from AI data-center leasing 02·03
• 11/12 (Thu)Korean options expiry · MSCI November index review announcement (around 07:00) 05
• 11/13 (Fri)Mississippi Senate county certification deadline · U.S. October PPI (22:30) 02·06
• 11/16 (Mon)Q3 filing deadline — reconcile the two companies' cash, tax and shareholder-return ledgers 07
• 11/18 (Wed) 07:00NVIDIA Q3 earnings — the first new numerical test of AI demand 02·09

The forest is quiet after the storm. Boboo will pick up the telescope again next week, but comes down at the weekend to count the rings: to distinguish the trees that fell far from those that truly stand low.