Gooaye EP682: How to Read a Sell-Off With No Bad News

Magical-realist oil painting cover: a wide riverbed after a storm, the water gone but no wreckage left behind, a traveller sitting on a rock watching the clouds gather again, distant factory chimneys still steaming steadily

I walk to where the water ends,
and sit to watch the clouds rise.
—— Wang Wei, “My Retreat at Zhongnan” (Tang dynasty); translation mine

These are my personal notes after listening to Gooaye EP682 (released 2026-07-25). Not a transcript, not official show content. Go listen to the original if you want the full argument. What follows is what the episode sparked, plus my own reading.

What this episode is about

In one line: nobody can be blamed for this one.

Usually, after a bad week, you can name the culprit. A rate move, an earnings miss, something geopolitical. This time the host went looking through the macro and the industry data and came back without a suspect. His read is that this is a valuation reset after an overheated April and May, the same shape as the one last March. The move has been violent regardless. Some equipment and substrate names have been marked down to single-digit earnings multiples.

He compares this kind of tape to a Souls game. You don’t die from one clear mistake. You die from a hundred instances of “I can hold on a bit longer.” So the prescription isn’t to find the answer. It’s to trade less, raise cash, and accept an uncomfortable fact: when good results and good fundamentals are being sold indiscriminately, you should stop using fundamentals to explain the price.

The main points

This is a valuation reset, not a break in fundamentals. No macro or industry trigger has shown up. Don’t invent one. Things ran too far in the spring and are giving it back.
AI software has a different cost structure than software used to have. Write once, sell a million times, marginal cost near zero — that was the old model. Models burn compute on every inference. Until compute gets meaningfully cheaper, enterprise margins carry that weight, and it will show up in the numbers of companies sold as AI software stories.
Hyperscaler capex is survival, not a discretionary line item. The host’s analogy is buying property in Monopoly. Skip the early spend and you have no position later. Judging cloud capex by “that’s a lot of money” may be asking the wrong question.
Memory pricing for Q4 is expected to flatten. The speculation is that margins at this level (the high end of the range, around 80 percent) attracted enough government attention to slow things down. The warning that matters: the market may well punish these names for prices “not going up,” while the suppliers remain extremely profitable. (Post-publication check: this is a minority read. Sell-side consensus over the same period still expected Q4 contract prices to rise, just at a decelerating rate, with one major bank in late June looking for another 30 to 40 percent quarter on quarter. “Flat” and “still rising but slower” are very different outcomes, so this is worth tracking rather than treating a single source’s call as settled.)
Component fundamentals have not changed. Power devices, passives, substrates — market pricing is still rising across those chains. The recent damage came from the tape, not from the businesses.
Companies mentioned in passing: Google’s results and cloud numbers as evidence that AI can be profitable; Apple’s end-user stickiness putting it in a position where it doesn’t have to immediately fight over foundation models; Tesla’s physical AI and robotics as a long-horizon bet that requires belief; Texas Instruments reporting a broad data-centre recovery and starting to raise prices; Palantir with almost no competition in high-barrier compliance work. All of these are context from the show, nothing more.
Two lines on execution. First, when you can’t read the tape, do less. Cash is a position, and rotating frantically through an indiscriminate sell-off gets you hit on both sides. Second, set a technical fail-safe stop and wait for the shape to reclaim its major moving averages.

Extensions

One: markets price the rate of change, not the level.

The memory section is the part worth chewing on. Pricing going from “rising fast” to “flat” sounds like bad news. But the suppliers’ margins haven’t fallen off a cliff; they’ve simply stopped accelerating. Market pricing is brutally sensitive to the second derivative. Not how much you earn, but how much more than last quarter.

That suggests a practical habit. When a headline makes you uneasy, ask whether the level got worse or whether the rate of improvement slowed. Both can hit the share price equally hard. They mean completely different things for whether you should still own it. The first calls for re-examining your thesis. The second only calls for re-examining your holding period.

Two: when a verifiable signal and the price narrative disagree, I start from the verifiable side.

The episode contains a clean natural experiment. Component pricing is still rising. Texas Instruments said in its results that data-centre demand has broadly recovered and that it is raising prices. Those are checkable, traceable facts with third-party records attached. Meanwhile the stocks fell.

Something I’ve been working on for the last six months is keeping “numbers I can check” in a separate bucket from “my impression of the tape.” Impressions get contaminated by the last three sessions. Quoted prices, monthly revenue, the exact wording in a filing — none of that changes because you lost money yesterday. This doesn’t mean the verifiable stuff tells you when something will go up. It can’t. But it tells you whether the business actually changed, which is the judgement you most need and most often get wrong in a drawdown.

Once they’re in separate buckets, the decision gets more honest. If the numbers held and only the price moved, you’re facing a question about tolerating volatility, not a question about being wrong. Those deserve very different responses.

Three: a moving average used as a stop is not the same instrument as a moving average used as a signal.

The episode suggests using moving averages as a fail-safe stop in this kind of tape. I want to be careful here, because I’ve run the opposite test and the conclusion looks like it contradicts the show. It doesn’t.

I tested moving-average crossovers as entry and exit signals, and found no tradeable edge (the full write-up is here). That test asked whether a moving average predicts future returns. The answer was no. But a fail-safe stop isn’t asking for prediction. It’s asking for a line you agreed to in advance. Its whole value is removing “let me hold on a bit longer” from the decision path, which has nothing to do with predictive power.

A tool without predictive power can still have disciplinary value. Confusing the two produces opposite errors: using moving averages to pick stocks because you think they forecast, or abandoning stop discipline entirely because you heard moving averages don’t work.

One honest boundary

The advice here is to trade less, hold cash, and wait for a reclaim. Advice of this shape has a structural property: it always looks right afterwards. If the market keeps falling, you’re glad you were flat. If it bounces, you say you were waiting for the reclaim. It’s close to unfalsifiable.

I’m not saying it’s bad advice. Doing less in a tape you can’t read is probably the highest-win-rate move available. I’m reminding myself of something else: advice that can’t be falsified shouldn’t be treated as evidence that you made a decision. A real decision needs a criterion written down beforehand that can embarrass you later. Waiting patiently is excellent execution discipline. It is not, by itself, a judgement.

What this episode is about

The main points

Extensions

One honest boundary

Further reading