SK Hynix Just Posted Its Best Quarter Ever — So Why Did the Stock Fall 9.61% the Same Day? Four Answers You Only See If You Do the Math

Data as of the 2026-07-29 close on the Korea Exchange (KRX). Earnings figures are from SK Hynix's same-day disclosure (K-IFRS consolidated, pre-external-audit). Prices are first-hand KRX closing prices, not a delayed feed. Every ratio below can be re-derived from the raw figures given in the text.

The one-line conclusion

This day delivered the numerator of the valuation equation, not the denominator. Record profit is earnings power. What the market priced that day was capital allocation behavior and shipment visibility — one of which was not announced, the other not disclosed. Treating record profit as a reason for the stock to rise conflates two different things.

1. The facts first: this genuinely was the best quarter ever

First-half revenue reached KRW 131.895tn (first half-year above KRW 100tn) with cumulative operating profit of KRW 98.1529tn.

Cross-check: Q1 2026 (revenue 52.5763 / operating profit 37.6103) + Q2 2026 = the reported half-year totals. The books reconcile.

Same-day KRX close: KRW 1,401,000, -9.61%. Intraday high KRW 1,619,000 (+4.45%), intraday low KRW 1,246,000 (-19.61%), a 24.1% full-day range, on 12.17m shares versus 8.12m the prior session (2.1x the recent average).

A 14-percentage-point intraday reversal on record-earnings day. That is the thing requiring explanation.

2. Answer one: the miss was entirely in revenue; margin took zero damage

Most commentary stopped at the word "miss" without asking which line missed.

The two misses are nearly identical in size (-4.9% vs -4.7%), while margin did not compress at all — it expanded +4pp QoQ.

That combination admits only one reading: the miss came from volume, not price, and not cost.

• A price war would necessarily compress margin — it cannot coexist with +4pp
• Cost overruns, likewise
• Revenue and profit missing by the same magnitude while margin improves → constrained bit shipments with better unit economics

This also splits a popular narrative in half. The claim "a rising HBM mix suppresses ASP" decomposes into two testable parts:
- "HBM crowds out capacity → revenue misses": supported
- "HBM suppresses ASP → earnings quality deteriorates": not supported. Were it true, operating margin could not have risen +4pp to 76%. Company commentary likewise described continued price increases in both DRAM and NAND.

3. Answer two: net profit is a trap — it contains a KRW 62.17tn one-off gain

Net profit of KRW 93.9226tn exceeds revenue of KRW 79.3187tn — a 118% net margin. Any quarter with a net margin above 100% is unusable for valuation as reported.

How to decompose it correctly (there is a common arithmetic trap here)

Many people back out non-operating income as "net profit − operating profit" and get KRW 33.38tn. That is wrong — it silently nets out income tax. You must work from pre-tax:

Non-operating P&L = pre-tax 122.708 − operating 60.543 = KRW 62.166tn
Income tax        = pre-tax 122.708 − net       93.923 = KRW 28.785tn
Effective tax rate = 28.785 / 122.708 = 23.46%   ← plausible, which validates the split

Disclosed composition: other non-operating income KRW 60.889tn (of which investment-asset-related gains of KRW 63.27tn, non-recurring) plus FX gains of KRW 1.147tn. Note that 63.27 > 60.889, implying offsetting negative items within other non-operating lines. The specific assets were not named, so this cannot be attributed to any single holding.

What this does to the valuation

Using the 7/29 KRX close of KRW 1,401,000 and an implied 709.9m shares (back-solved from the prior session's market cap of KRW 1,100.3tn ÷ KRW 1,550,000):

Same closing price, 78% difference in valuation. "It trades at 3.7x, absurdly cheap" is factually true and decision-theoretically useless — it annualizes a one-time investment gain.

Methodological takeaway: before concluding a company is cheap, check whether net margin sits inside the bounds common sense allows given gross margin. An abnormally high net margin points to one-off items roughly 99% of the time, not to operating efficiency.

4. Answer three: no buyback was announced — but the cash statement turns it from rumor into arithmetic

The release contained no buyback and no incremental shareholder-return program. However, the company has publicly set its own precondition: net cash reaching KRW 100tn (confirmed twice — by the CEO at the shareholder meeting and by the CFO on the Q1 2026 call, the latter phrased as an additional shareholder-return plan to be formulated within the year).

That converts the question from "guess management's intent" into "run the cash flow":

End of Q2 2026: cash & equivalents KRW 88tn − borrowings KRW 18.6tn = net cash KRW 69.4tn

Net cash trajectory:
  End 2025      KRW 12.7tn
  End Q1 2026   KRW 35.1tn
  End Q2 2026   KRW 69.4tn      single-quarter generation +KRW 34.3tn

Gap to the KRW 100tn threshold = KRW 30.6tn  <  quarterly generation of KRW 34.3tn
→ Absent a marked slowdown in cash generation, the self-imposed threshold is clearable by the end of Q3 2026

There are two implications, pointing in opposite directions, and both must be stated:
1. The threshold is reachable → the key to the shareholder-return path was not lost, only pushed out
2. But nothing lands before the Q3 print → for the next quarter, this catalyst is simply absent

That is why record profit failed to support the stock: the market was waiting on the timetable for (2); the release delivered evidence for (1).

5. Answer four: why a 2x leveraged ETF still cannot break even after the underlying fully recovers

This is a structural math problem independent of any view on the stock, and it applies to every leveraged product. Using this week's real data:

2026-07-28, single session:
  Underlying (SK Hynix)   -14.65%
  A 2x long ETF           -30.24%     realized leverage = 30.24 / 14.65 = 2.06x ✓ structure behaving normally

The asymmetry lives in the gain required to get back to even:

Now the thought experiment: assume the underlying fully recovers to its 7/27 price (i.e. +17.16%). Tracking at exactly 2x linearly, the ETF gains +34.32%:

0.6976 × (1 + 0.3432) = 0.9371   →   still 6.3% below where it started

The underlying is whole; the ETF is down 6.3%. That gap comes from neither fees nor premium/discount. It is pure compounding asymmetry.

The general form

For an underlying and a k-times leveraged product, whenever the underlying falls and then returns to its starting point, the leveraged product's net result is strictly negative:

Fall by d, then recover to the start:  underlying NAV = 1
Leveraged NAV = (1 − k·d) × (1 + k·d/(1−d))
              = 1 − k(k−1)·d² / (1−d)      ← always < 1 for k>1, d>0

With k=2, d=14.65%:  loss = 2×1×0.1465² / 0.8535 = 5.03%

Note the worked example above produced -6.3% while the formula gives -5.03%. That is not an inconsistency: the example uses the actual -30.24% decline (realized leverage 2.06x), whereas the formula assumes exactly 2.00x. Realized leverage overshooting by just 0.06x widens the break-even gap by roughly 1.3pp — itself a real cost of holding leveraged products.

This is not a bull-versus-bear question; it is a holding-period question. Leveraged ETFs are engineered to replicate k times the daily return, not k times the return over an interval. The greater the volatility and the longer the holding period, the wider the divergence. In a session with a 24% range, holding a 2x product can lose money even when the directional call is right.

6. A correction, stated in public

Seventeen minutes into the Korean session (SK Hynix +2.65%, Samsung Electronics +4.32%), this author's read was that the market was trading the print as "bad news exhausted." That read was fully falsified by the close at -9.61%.

It is recorded here for a practical reason: the first twenty minutes of a session is noise, not information. The real verdict on an earnings day forms in the closing session (14:00–15:30 KST in Korea) in the net buy/sell direction of institutions and foreign investors — not in the opening emotional impulse. Any method that concludes from the opening half hour, including this author's on the day, should be rebutted by this data point.

From the 2026-06-22 peak of KRW 2,919,000, the 7/29 close is -52.0%, with the intraday low at -57.3%.

7. Conditions under which this analysis fails

Honest analysis must state its own falsification conditions:

• If the Q3 2026 call provides explicit HBM revenue-mix and shipment guidance showing the bit-shipment bottleneck clearing in Q3 → the "volume problem" conclusion in section 2 still holds, but the affected window shortens and the valuation discount loses its rationale
• If net cash generation slows materially in Q3 (below KRW 30.6tn for the quarter) → the "threshold cleared by end-Q3" inference in section 4 is void and shareholder returns slip further
• If the KRW 63.27tn investment gain proves repeatable (assets unnamed; treated here as non-recurring) → the 6.6x P/E in section 3 must be recalculated
• If a buyback is announced ahead of the Q3 print → the premise of section 4's "catalyst absent for a quarter" is directly overturned

Sources and definitions

• Earnings figures: SK Hynix disclosure dated 2026-07-29 (K-IFRS consolidated, pre-external-audit); cross-verified across the official release and two Korean financial media wires
• Prices: first-hand KRX closing prices (not yfinance or other delayed/adjusted feeds); percentage changes cross-checked against the prior session's close for basis consistency
• Consensus: sell-side consensus as compiled by major Korean financial data providers
• Implied share count: back-solved from the prior session's market cap ÷ closing price — an approximation, not the company's disclosed shares outstanding
• Not disclosed: quantitative Q3 2026 guidance, HBM revenue mix, and the specific assets behind the one-off investment gain. None are public; this piece does not speculate on them

This is public-layer market analysis containing opinion, public data, and reproducible derivations. It contains no position data, no trade instructions, and no investment advice. All workings are shown above and are open to replication and rebuttal.