The Problem with Micron Technologies (and Similar)

[Chawasri]

My large-cap strategy has invested in Micro Technologies (MU) three times in the last year, recording returns of 79.5% from Nov. 2025 to Feb. 2026, then 184% in two months, from April to June 2026. However, the model bought it again a couple of weeks ago, and since then it has continued in a selloff that has only accelerated.

Micron is probably appearing on a lot of P123 screens and strategies these days because its fundamentals are phenomenal and it is considered undervalued by many measures. For example, it has:

• FQ3 (ended May 28) revenue of $41.4B, up 346% y/y. EPS of $24.67, up 1,368% y/y. Growth across all four segments, driven by HBM demand.

• Company guidance for the August quarter: about $50B revenue and $30.73 EPS.

• On TTM EPS of $44.23, that's a P/E near 18.6 versus roughly 32.6 for the Nasdaq-100. On the FY2027 consensus of $153.74, the forward P/E is about 5.3. That’s a forward PE of 5.3!!!

However, regardless of the incredible fundamentals of Micron and many other large-cap technology stocks (NVDA, LRCX, AMD, ARM, INTC, etc) fueled by the AI supercycle build-out, the market for that segment has been dramatically sinking based on sentiment alone since the late-June ATH. Mega-cap Tech is a verboten sector of the market, with individual losses of -30% to -70% since the end of June.

It was wonderful to score a 189% return in two months on MU, but now the chickens have come home for the entire AI-fueled segment of the market. There has been a significant rotation out of the semiconductor and memory suppliers selling to the incredible AI data-center buildout. Seems everyone is questioning the AI story now. Only time will tell…

I’ve tried adding some creative approaches in both the Buy and Sell rules to eliminate potential value-trap companies like Micron from getting in, but it resulted in lower CAGR and slightly higher MDD, accompanied by increased volatility. Apparently, these conditions don’t occur very often historically, so I’ve had poor luck trying to program out these value-trap stocks.

Curious whether anyone here handles semis with a sector-specific ranking system, or just accepts that generic value factors will periodically get run over by industry cycles. I've generally found trailing-earnings value needs a momentum or revision overlay to survive in this industry group, but this model has those factors built in from the start, to no avail.

Anyone have any favorite value-trap avoidance rules worth sharing? Thank you.

[regallow]

I filter out stocks that don’t pass a BeneishMScore test of being less than -1.78, and MU began failing my test on 3/21/26 using the P123 value of -1.52. It got much worse on 6/27/26, showing a high probability of earnings manipulation with a value of -0.32. There are eight factors in the BeneishMScore calculation. It would be insightful to analyze each to see if the score could be improved on or if the original assumptions all still hold.

[yuvaltaylor]

When I calculate the M-Score for Micron I don't get these numbers, or anywhere close. Portfolio123's formula uses TTM numbers, which Beneish never used. P123's MScoreSGI is 2.67, which pushes the company way up there. For Beneish's own calculation, see Beneish M-Score Calculator - Kelley School of Business. Also see a close discussion of the MScore at Detecting Financial Fraud: A Close Look at the Beneish M-Score - Portfolio123 Blog and its accompanying screen, https://www.portfolio123.com/app/screen/summary/284349?st=1&mt=1

[Chawasri]

@regallow

showing a high probability of earnings manipulation with a value of -0.32.

That would be a world-shaking event for such a high-profile company. All the other stocks I mentioned (NVDA, LRCX, AMD, ARM, INTC, etc), have very similar charts, with all-time highs at the end of June and an extreme deterioration of price ever since.

It will be very interesting to watch how this AI super-cycle build plays out with so many questions about whether or not they can ever even achieve profitability. It’s one thing for an industry to pour trillions of dollars into building data centers, and another thing to consider who is going to pay for it. I’m paying $20 a month each for a couple of AI services right now, and I would pay more - but not a whole lot more.

[parallaxblue]

In my opinion this really isn’t a problem; the models are picking up many of the semis as value plays because the fundamentals really have been and are so far very solid. Undoubtedly that will change in the future, and the models will then replace them with other companies.

The recent fall in prices of these companies IMO has less to do with the fundamentals so far than these companies being rampantly crowded trades, both in terms of US retail and institutional buying, as well as considerable speculation in South Korea via leveraged ETFs and FOMO.

Consequently, we’ve seen an incredible run up in the prices of these companies, likely beyond what the fundamentals would support, in large part due to speculation. And as usual with price momentum, it’s subject to sharp corrections as price eventually returns to the mean (classic mean reversion here after a run up).

tl;dr the models are functioning correctly, these are good companies based on their fundamentals, the recent price declines have much less to do with the company fundamentals themselves than the rampant speculation and money pouring into and out of the AI trade.

My 2 cents, could be wrong.

EDIT: one more thing

”I’ve tried adding some creative approaches in both the Buy and Sell rules to eliminate potential value-trap companies like Micron from getting in, but it resulted in lower CAGR and slightly higher MDD, accompanied by increased volatility.”

I don’t think these are value-trap companies at all, their fundamentals are solid, until the AI trade runs out. But, absolutely no one can predict exactly when that happens, which is why we rely on quantitative models to make these choices instead of gut feeling.

Basically you either trust the models or you don’t. If the model has a good OOS track record and appears robust, trust the model.

Again, just my 2 cents on this.

[Jrinne]

I honestly don't know what the story is, but it is not just a story about MU perhaps.

MU is cointegrated (moves with) SOXX (semi-conductor ETF) according to PortfolioVisualizer. And SOXX is it down, too:

Here is SMH (lower cointegration that does not match MU as well):

MU is not underperforming these benchmarks (SOXX and MU both down 16% over the last month):

Its not just that they are both down 16% but the cointegration proves (mathematically) that MU is not behaving abnormally here.

I asked Gemini's input on this (relative spread being what cointegration measures):

"The core point is that the relative spread between MU and SOXX hasn't broken down. This 16% dip is a systematic semiconductor factor drawdown, not an idiosyncratic failure in Micron's fundamentals."

BTW, my point here was that SOXX would have been a good hedge for MU over the last month,--in part because of the cointegration between the MU and SOXX: The market is approaching its peak; how can one achieve an effective and cheap hedge? - #20 by Jrinne

Shorting SOXX would have worked well as a hedge for MU over the last month. Totally market neutral over the last month (minus borrowing and transaction cost with equal weight).

Gemini again:

Yes, cointegration is a foundational technique in quantitative finance for designing ETF-based hedges, particularly within statistical arbitrage, market-neutral equity strategies, and factor-hedging frameworks.

While traditional portfolio management often relies on beta hedging (matching short-term return variance), quantitative traders use cointegration hedging to construct stationary, mean-reverting price spreads between individual stocks and sector ETFs.

Personally, if (or when) I hedged, I would use cointegration and correlation to help decide which ETFs or stocks to short. Presumably, more than one ETF or stock--hedging more than one holding. Maybe being selective on which stocks to hedge based on holding too many stocks in a sector, stocks at risk, volatile holdings etc.

[Chawasri]

Thank you, Jim. I was hoping for input from the community on approaches to avoid stocks like MU, NVDA, LRCX, AMD, ARM, INTC that have exceptional fundamentals, extraordinary growth stats, and deeply discounted valuation factors (PE of 5.8 times forward earnings for MU).

Today it’s the AI-supplier segment of the market, but next month or next year it may be other sectors. Is there any way to avoid these stocks that sink despite extraordinary fundamentals?

Paralaxblue suggested ignoring the selloff because the stocks are just exhibiting reversion to the mean after a big run-up. However, a loss of -36% on MU since July 1 is not something that I am willing to suffer. And it is still dropping every day.

“I don’t think these are value-trap companies at all, their fundamentals are solid, until the AI trade runs out.”

The question is, has the bloom already come off the rose in the AI trade? There are hundreds of articles out asking how AI is going to pay for the $5.7 trillion in expected build-out by 2030. There’s also fears the Chinese are going to take over the AI-supplier market with much lower-cost semis and HBC memory products.

Is the solution just to strap on a PctfromHi sell rule and forget about it? MU was already declining for two weeks when the strat bought it again. I would prefer not to enter positions like that. Maybe there’s a better Buy Rule to prevent purchasing decliners? I tried that with some simple Close(0>Close(5 or 10) rules and it deteriorated model performance.

A 5x forward multiple on record earnings isn't the market saying "bargain." It's the market saying "this E is not real." That's the trap mechanism worth engineering around. Suggestions to address this?

[Jrinne]

@Chawasri

sectors

I do think sectors can be timed to some extent but moving in and out of individual stocks has too much transaction costs for me. That is why i mentioned shorting a sector (or industry) using an ETF with a narrow bid/ask spread. Just me perhaps, but I don't think I could find a timing method to over-ride any of my present model's buys that I could make work and it would be difficult to backtest. I may have overemphasized hedging, but for me that would be the only way to keep transactions costs low and let my models pick stocks that should do well with most market regimes. My models are designed for that. That having been said I am not hedging at this time. Instead I diversify for now. I do go to cash in some sector ETFs based on hidden Markov Models.

Edit:

I think this article on CNBC may be on topic. I don't know your thoughts on Michael Burry (I don't really follow him), but at least someone in the news has the same concerns or observations you have expressed. And he remains short (mostly puts actually): Michael Burry Shorts both MU and SOXX

" In the face of the rally, Burry said he continues to hold short positions in the iShares Semiconductor ETF (SOXX), Micron, Nvidia, Caterpillar, Palantir, Tesla and Applied Materials.

The investor said he remains confident in his long-term outlook for those positions, though he added that he would cut his losses if the trades moved decisively against him. All of the positions remain profitable except for his bet against Nvidia, he said.

BTW, I looked into this further. As near as I can tell Burry has made a directional bet and is not using those shorts (puts or whatever) as hedges for long positions in the same industry (as I might consider at some point).:https://www.youtube.com/watch?v=0PuBQjn_dco

[plan_trader]

Scion Asset Management (Burry ) Performance - 1 yr 4.2%, 3Yr -6.3%

[Jrinne]

Thank you plan_trader. As I said I don't really follow Burry. and did not know that. I find it interesting. Not sure, but shorting (or buying puts) is hard and he may be a one-hit wonder.

So, Burry is not so good at a directional bet it would seem. To my broader point, it is possible that Scion Asset management COULD HAVE been a nice hedge depending on the pattern. Not necessarily but it could have eased the drawdown and reduced the beta of long srategy. And actually with great inverse correlation (which may not be the case), possibly keeping most or all of the alpha of a long strategy.

Whether it would have worked well as a hedge could be determined, in part, by the correlation (or inverse correlation), and cointegration of Scion's returns and any candidate long strategy.

[SZ]

It is important to distinguish between a stock that simply declined vs a value trap. There are many reasons a stock can decline significantly such as an overcrowded heavily leveraged momentum trade leading to margin calls

Was MU a value trap before the decline but a value buy right after simply because of the price change? Maybe, maybe not

[marco]

The recent MU low (what looks to be a low to me so I'm long now) was due to the Situational Awareness hedge fund blowing up. AI investor Leopold Aschenbrenner forced to unwind all public stock positions after steep losses, sources say

Here's an interesting MU chart. It shows SalesTTM in log scale with the price, and shares growth below. The sales trend clearly shows how cyclical the stock is, and how clearly the stock declines well before the cycle top.

But even more interesting is how MU starts buying back shares to support the price during the sales decline cycle.

How would you interpret this chart?

[SZ]

Seems like there were many similarly sized dips before large sales increases looking at your chart so the dip could easily be noise. Honestly for me it is hard not to see MU sales continuing to grow next year with all the cloud computing backlog (literally trillions in backlog) we have and still increasing llm demand. And this is with very low global penetration. Dram is not just a needed ingredient for new centers but it also increases the datacenter output for existing centers. Higher tokens per second from existing silicon. In a compute-constrained world it is hard to see how demand craters in the next year at the very least