Our Board Scores Semiconductors at 98. Our Own Backtest Says 82% of These Entries Fail.
Eleven of the fourteen semiconductor names we track are scoring in Buy territory or better right now. Every one of those eleven is also down between 10% and 38% in the last month. Those two facts describe the same stocks, on the same day, and the second one is the more important.
The Numbers
Measured from each name's own peak inside our 24-session record to Thursday's close:
- Marvell — $297.89 on June 30 to $183.30. Down 38.5%. Score: 98.
- Intel — $139.63 to $91.13. Down 34.7%. Score: 98.
- ARM — $354.57 to $241.54. Down 31.9%. Score: 98.
- Lam Research — $433.33 to $297.72. Down 31.3%. Score: 97.
- Applied Materials — $723.00 to $501.77. Down 30.6%. Score: 98.
- Micron — $1,154.29 to $874.66. Down 24.2%. Score: 95.
98 is the maximum the model can output. Scores are clamped there. So on Marvell, a stock that has lost more than a third of its value in a month, our model is registering the strongest possible reading it is capable of producing.
Why The Model Does This
The score is, at its core, a mean-reversion measure: how far below its own 20-day average is the price, expressed in units of that asset's own volatility. When a stock falls hard, the 20-day average follows it down more slowly. The gap widens. The score rises.
That is the model working exactly as designed. It is not broken and the arithmetic is not wrong. It simply has no opinion about why the price fell, and it never has.
Correction — Added August 1
The first version of this piece drew the wrong conclusion from the right numbers, and the correction inverts most of it.
The drawdowns above are measured from each name's peak inside our 24-session record. Our crowding backtest sorts on something different: the trailing 60-session return. Those are not the same measure, and we treated them as if they were.
Now that the full 60-day figure is recorded, here is the actual picture:
- Micron +23.5%, Applied Materials +18.4% — hottest quartile, historically our best cohort at 60.8% stop-first.
- AMD +13.0%, Marvell +9.0%, ASML +5.4% — third quartile, 65.3%.
- ARM, Lam Research, TSMC, Texas Instruments, Nvidia, Broadcom — second quartile, 70.5%.
- Only Intel (−20.2%), Qualcomm (−23.4%) and Super Micro (−18.1%) are genuinely in the coldest quartile at 81.6%.
Three of fourteen, not eleven. Most of these chips fell hard from a peak that was itself the top of a large run — which places them in the better-performing buckets, not the worst one.
The honest version of the warning is narrower: Intel, Qualcomm and Super Micro carry the historically weak setup. For the rest, the high score reflects a pullback inside a run, which our data treats far more kindly.
What The Backtest Actually Says
Sorting 1,988 historical Buy Zone entries by trailing 60-session return:
- Coldest quartile — entries after a 60-day decline of 10.6% or worse: 81.6% hit the stop before the first target. 17.2% reached target one.
- Hottest quartile — after a 60-day gain of 14.8% or more: 60.8% stopped first, 39% reached target one.
A high score reached through a genuine 60-day decline is our worst setup by a twenty-point margin. Measuring that correctly requires the 60-day return, which until this week we computed daily and discarded — which is precisely how the original version of this article went wrong.
The Board Barely Registered The Week
Broader context makes it starker. The Nasdaq entered a correction on Wednesday. The Dow had its worst session since April 2025. Over that stretch, the median score across our whole 180-asset board went from 83 to 81. On the day of the correction itself, the count of names scoring 90 or better went up, from 49 to 54.
To a mean-reversion model, a market breaking down and a market going on sale are the same event.
What's Actually Actionable Here
None of this says the chips are uninvestable, and none of it is a forecast. What it says is narrower and more useful: the number on the card is measuring one thing, and right now that thing is diverging sharply from the thing most people read it as.
A 98 means the price is far below its own recent average. It has never meant the price is likely to rise. For Intel, Qualcomm and Super Micro our own data says the opposite is more common; for the rest of the group it does not, and the original version of this piece said otherwise. If you act on these scores this week, the entry is only half the decision — where the position is wrong, and how much of it you own, is the other half, and it is doing more work than usual.
Numbers here come from our own daily record; the correction and index figures are from external reporting. Nothing here is financial advice.
Scores in this piece come from the ZonedInOut model. How the score works →
Get the Sunday Brief
One email a week, before the open: what the board did, which calls worked, which didn't, and what changed. Free. No spam, unsubscribe anytime.