This page is a companion analysis for Efficient Markets Through a Bayesian Lens. That essay used the Samsung announcement to discuss “already priced in” in Bayesian terms; this Lab narrows the question to closing prices, observation windows, and market-relative returns around the event. The result is not a single causal verdict. It is an experiment in matching a claim to the data that can actually test it.
On July 7, 2026, Samsung guided to second-quarter revenue of about ₩171 trillion and operating profit of about ₩89.4 trillion. The profit figure exceeded the ₩87.3 trillion LSEG SmartEstimate cited by Reuters by about ₩2.1 trillion, or 2.4%. Its shares nevertheless closed at ₩296,000, down from ₩318,000 the previous session: roughly −6.9%. The KOSPI fell about 4.9%, and SK hynix about 6.1% that day. Samsung guidance, Reuters reporting, historical-close API
“Profit beat the estimate” and “the shares fell” are both true. The 2.4% measures one quarter’s profit against one estimate. The 6.9% measures a one-day change in a price for future cash flows and risk. They are not two returns on the same thing with conflicting signs. The word because between them requires more evidence.
Explore the event
The chart loads historical daily closes for Samsung, SK hynix, and the KOSPI from the aikstockdata API, whose underlying price source is Korea’s Financial Services Commission open data portal. The key figures above and the argument below remain readable if the API is unavailable. Daily closes cannot show what happened in the minutes immediately after the announcement.
THE ARCHIVE / LAB 01
A profit beat, a falling share price
Change the comparison around July 7, 2026.
Previous close = 0%
Loading closes from the price API…
Every point is relative to the July 6 close. Only the July 7 point is also a one-day announcement return.
View daily returns table
| Date | Samsung | KOSPI | Difference |
|---|
Different numbers, different questions
How far above one LSEG estimate was this quarter’s profit?
How much did the price of future cash flows and risk move in one day?
How different was it from the KOSPI that day? This is not causal attribution.
Change one profit estimate
Samsung’s actual −6.9% daily return does not move. This slider excludes revenue, future outlook, risk, and positioning.
Closes: aikstockdata API (underlying source: Korea FSC open data portal). Announcement and consensus: Samsung and Reuters. Historical, not real-time data. API and terms
“It had already risen”—over which window?
In the 10 trading days before the announcement, Samsung’s close went from ₩353,500 on June 22 to ₩318,000 on July 6, a decline of about 10.0%. That hardly looks like a stock steadily climbing into the release. Widen the view to 60 trading days, though, and the close rose from ₩196,500 on April 7 to ₩318,000 on July 6, or about 61.8%. Both are real closes from the same source. Historical-close API
Reuters quoted an investor who described strong earnings as largely priced in after an earlier rally. The longer window helps explain that interpretation, but it does not establish that 61.8% was a measurable “pre-pricing effect” of this particular earnings release. Memory prices, AI spending expectations, interest rates, and risk appetite also changed along the way. Choose the start date after seeing the chart and you can tell opposite stories. Reuters report
A comparator does not decompose the cause
Subtracting the KOSPI’s roughly −4.9% from Samsung’s roughly −6.9% yields a simple −2.0 percentage-point difference on announcement day. It tells us that the whole market fell sharply. It does not mean earnings guidance caused precisely two percentage points of Samsung’s decline. Samsung itself is part of the KOSPI, and the stock does not normally move one-for-one with the index.
Switch to SK hynix and the shared semiconductor selloff becomes visible. SK hynix is not a perfect counterfactual for what Samsung would have done without the announcement: the firms differ, and both could react to the same industry and market news.
This chart is a descriptive starting point for an event study. Our simple difference is not a formal abnormal-return estimate or a causal effect. A fuller study would estimate normal market sensitivity from earlier data and check for other news in the event window. Reserve Bank of Australia event-study methodology
Changing an estimate does not change history
The final slider changes one assumed operating-profit estimate. It computes:
Surprise = (announced profit ÷ assumed estimate − 1) × 100
Using the ₩87.3 trillion LSEG estimate cited by Reuters gives about +2.4%. Assume instead ₩90 trillion, and the same announcement becomes a roughly −0.7% surprise. This is only a thought experiment about one benchmark. It does not make ₩90 trillion the actual historical consensus, or explain the share-price decline on its own. Reuters/LSEG
Reuters reported elevated prior expectations, worries about the durability of AI spending, and a wider chip selloff; one analyst also pointed to disappointment with revenue. A single profit estimate cannot summarize everything investors had priced in. Reuters reporting
The modest but useful conclusion: good earnings, a surprise relative to one estimate, and the day’s stock return are different measurements. Changing the window or comparator does not automatically decompose the cause. One episode cannot test how much was priced in or settle the efficient-market hypothesis. That would require a consistently assembled sample of announcement times, pre-announcement estimates, and prices across many quarters.
Discussion
Discussion
Have a question, correction, counterexample, or another interpretation?