How Morgan Stanley Picks Stocks: A Conversation About Which Losses You Can Wait Out

A high-rise office at dusk, one lamp at the end of a long desk, cranes and distant data-center lights beyond the window stretching to the horizon

Not delighted by external gains, not saddened by personal losses.

—— Fan Zhongyan, “Record of the Yueyang Tower” (Northern Song, 1046; translated by the author)

What This Episode Is About

The September 4, 2026 episode of The Compound and Friends features Dan Skelly, a portfolio manager at Morgan Stanley Wealth Management. He runs a long-term book of forty to fifty stocks and also sits on the firm’s asset allocation committee. The conversation starts with the question of why stocks keep rising while the news stays bad, moves through why this quarter’s earnings look so unusual, then into how the AI capex cycle differs from 2000 — and ends on the part I listened to most closely: when to admit you were wrong, and when to give a position more time.

What stayed with me was not his market view. It was the way he splits selling into two entirely different questions. I still get this wrong after several years of trying, so I wanted to write it down.

The Main Points

The earnings strength is not only the AI names. Second-quarter growth was 28% year over year at the index level, and 14% for the median company. He says the median number gets too little airtime — if only AI were working, the median would not look like that.

A histogram of company earnings growth: most bars cluster left of center with a low long tail to the right; the median dashed line sits in the middle of the distribution while the index-weighted dashed line is pulled further right by a few bars

His explanation for the median is tariffs. Tariffs were the headline through the first half of last year. Most companies were forced to take price and passed through sixty to seventy percent of the cost. Then the tariff scare faded and the prices never came back down. He tells it through his own landscaper, who raised prices fifteen to twenty percent after covid and never reversed it. A year later, that forced pricing shows up as operating leverage.

Two lines over time: the cost line rises then falls back, the price line rises and stays flat, and a shaded gap opens between them on the right

Productivity is landing on top of the existing workforce; the labor lever has not been pulled. The software apocalypse and the labor apocalypse everyone feared in January have not arrived. What shows up in the data is the same headcount producing more. He thinks the actual reduction in labor is a twelve-to-twenty-four month story, and that it carries a second leg of margin. Revenue up 15% with no headcount growth is something the older, goods-heavy economy could not do.

The biggest difference from 2000 is who the buyers are. In 1999, Cisco hit its targets by selling to customers that had gone public the week before. Today the orders come from platforms buying on behalf of thirty million cloud customers — hospitals, insurers, manufacturers, government. And those buyers spent the last decade with net cash balance sheets. The fiber went dark for ten years until someone invented YouTube; GPUs, if the buildings run late, can be absorbed back into cloud demand.

Two paths, one above the other: the upper one runs to a dead end with a small buyer block and only three customer dots below it; the lower has a large buyer block with a dense field of customer dots below, plus an extra downward exit at the end

He locates the risk in deployment, not demand. Forty to fifty gigawatts of data center capacity are planned over the next two to three years, and chips, servers and electrical components are ordered against that plan. Do all those buildings get built on time? Is anyone double-ordering through the supply chain? Not a risk today, he says. Possibly a risk a year from now.

Small caps carry two tails. One is the left tail of businesses AI disintermediates. The other gets discussed less: unprofitable small companies lack the capital to invest in AI, so they miss the right tail of the adoption wave too. Add that roughly eighty percent of US companies with over $100 million in revenue now stay private, and the public small-cap index carries a selection problem before you pick anything.

Both ends of a bell curve are grayed out, the left tail labeled displaced and the right tail labeled missing out; below, eighty percent of a bar is taken by gray, leaving only the right fifth as public markets

Gross margin misses are being executed. The show put up a chart: companies with year-over-year gross margin contraction above 1% badly underperform over the following twelve months. Broadcom printed strong numbers that day, gross margin down two hundred basis points, stock down. His read is that margin now encodes two things at once — whether you captured the pricing wave, and whether your peer group is getting an AI productivity benefit that you are not.

Three bars against a zero line: the contracting gross margin group falls deep, the flat group dips slightly, the expanding group rises

Going Further

”The news is bad every day. Do I need to care?”

This is where I get stuck most often. Bad news always sounds reasoned, and reading it produces the feeling of having done work — so you adjust the position, and then you miss the next stretch.

Dan offers a framing I like: the policy shocks and inflation scares of recent years behave like pop-up ads. Something jumps into the frame, holds it for a few seconds, you close it, and the main content is still there. The main content is earnings.

I do not want to read that as “ignore all bad news.” He says as much himself — Liberation Day mattered for a moment, the Middle East mattered in March and April, and he will not claim the midterms cannot matter. So what separates them? The test I took away: has the news walked into the income statement? Tariffs did, and they reshaped pricing and margin for four quarters afterward. If a political event leaves no trace on any company’s financials four weeks later, it was a pop-up ad.

Since this episode I read news in a different order. Not “how serious is this,” but “which line, on whose statement, does this show up on.” If I cannot answer, I do not touch the position. That gross margin chart is the counterexample — a piece of bad news that walked straight into the income statement, and the market executed without hesitation.

An earnings curve with two pop-up windows carrying close buttons floating above it, touching nothing; a third piece of news points into the curve with an arrow and bends the later stretch steeper

“I got the direction right and still made no money”

This has happened to me more than once, so I laughed at the relevant stretch of the episode.

Dan wrote a note in February arguing software was oversold. New AI tools had just landed and the whole sector was being priced for obsolescence. His reasoning: those private labs pitching future Wall Street investors cannot build a business plan around putting corporate America out of business — that is a futile deck. So he expected interdependence rather than obsolescence, with wide dispersion of outcomes inside the sector.

The direction was right. Then he says himself that translating it into holdings did not produce a hundred percent hit rate — one name versus another produced very different results. Right race, wrong horse.

That stretch is worth sitting with, because it separates two things that get merged: your judgment about an industry, and your judgment about a company. The first you can improve by reading. The second depends on execution by management, which you cannot observe on the day you buy.

A long row of forty-five thin bars of equal height, and after a gap on the right, five much taller thick bars

Their answer is structural. Around forty-five positions are core holdings sized near two percent, with five or six slots reserved for genuinely big ideas. What that structure says is: I accept that I will be wrong about individual companies, so no single error decides the year.

My own habit ran the other way — the clearer the thesis, the bigger the position. After this episode I think the clearer the thesis, the more I owe myself the question of which horse, and why that one.

”I’m underwater. Do I cut it or hold it?”

This was the part I most wanted to keep. They start on stop-losses: for a forty-to-fifty stock book with tax-sensitive clients and thirty to forty percent annual turnover, a hard stop is too rigid. So what replaces it?

His dividing line runs back to the original thesis, then asks two questions. Is the earnings power temporarily impaired or permanently impaired? Is the competitive moat temporarily in question or permanently dismantled?

Then he applies that line differently to two kinds of stocks, which is the most useful sentence in the episode: cut cyclical losers faster, give secular winners room. Cyclical downside momentum almost always begets more cyclical downside momentum, and when cyclical earnings break, they do not break by ten or twenty percent — so waiting there for a V-shaped recovery usually fails. For a genuine long-term winner, he points to the 2010 payments regulation: two networks down thirty percent in six months, ten-baggers afterward.

Two small panels: the left line drops one big step and then runs flat at the low, the right line falls then recovers and runs off the top of the frame

He adds a distinction I had not made: government and regulatory risk is almost always overdone and discounted quickly, while competitive risk works the opposite way. A brand losing relevance, share slipping away — that is the death knell for many businesses, and it is nebulous and hard to see early. You can wait out the first. You cannot wait out the second.

Two paths relative to the same baseline: on the left a narrow, deep, clearly bounded dip that returns quickly; on the right a line sloping gradually down and fading out as a dashed line

There is also a sell reason that has nothing to do with losses. When a high-tracking-error position works and the thesis has played out, it becomes his fortieth idea while a better forty-first sits outside the portfolio. His question is blunt: if you were allocating fresh cash today, would you buy this one? If not, what is it doing in there?

Worth Reading Next

The episode itself: The Compound and Friends, September 4, 2026, with Josh Brown and Michael Batnick, guest Dan Skelly.
To understand how the 2000 capex cycle resolved, look at public material on the fiber buildout, especially the customer mix of the equipment vendors — that is the benchmark this episode argues against.
The gross margin research cited on the show comes from Adam Parker; his public writing and interviews are easy to find.
On sizing big ideas inside a concentrated portfolio, Byron Wien’s investment writing is the source Dan names.

帶得走的一件事

One idea: separating temporary impairment from permanent impairment is two different questions that need two different answers.

Most of us ask only one question when we are losing money — will it come back? That question has no answer, because it merges two situations. A price that fell while the thing itself is unchanged, and a thing whose substance has been dismantled, are not the same case. The first needs time. The second needs an exit. Treating one as the other produces the two most familiar kinds of pain: holding what should have been let go, and abandoning what only needed waiting.

A two-by-two grid where the top-left and bottom-right cells are outlined solid as the right calls, while the other diagonal is labeled as two kinds of pain

Here is something I have tried that has nothing to do with stocks, and that you can do today. Pick one thing you have recently wanted to give up on — a friendship gone quiet, a skill three months in with nothing to show, a job that is not working. Take a sheet of paper and make two columns. On the left: if this is temporary, which three signals will I see in the next three months? On the right: if this has permanently changed, which three signals will I see?

The signals have to be visible to someone else. “It feels better” does not count. “He reached out twice on his own” does.

What you notice when you finish is that the hard part was never the judgment. It is that you had never written the signals down, so every time you were re-deciding from scratch on the mood of the day.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.