What Wall Street Screens for in a Crash: Three Traits, Three Names

The Tape Tonight

March was the month the megacap trade finally cracked. By the March 27 close the Nasdaq was in formal correction - down more than 10% from its high - with the S&P 500 off 7%, and four of the index's ten largest components sitting more than 20% below their 52-week highs. The post-mortems all read the same: roughly half the index's market cap in about twenty stocks, an AI capex payoff that keeps getting questioned, a consumer that is wobbling.

Then the selling stopped getting worse: the first two April sessions were strong across the names March hit hardest, the market took Good Friday off, and tonight closed flat. Nobody can tell you whether the correction just ended or just paused.

Which makes this the moment for the exercise institutional desks actually run mid-drawdown. Not "what fell the most." A quality screen: which businesses still clear a specific, checkable bar while the tape is ugly - so the list exists before prices decide the timing.

We used Barebone AI to pull the filings, index announcements, and brokerage metrics behind three names that keep clearing that screen this cycle: the photonics cluster of Lumentum, Coherent, and Tower Semiconductor; Interactive Brokers; and Alphabet. A 79% pretax margin. $132.2 billion of annual net income. Trade volumes up 25% in the teeth of the selloff. And $4 billion of NVIDIA's own money. Receipts below.

The Screen, in Three Traits

"Buy quality on discount" is the most repeated and least defined sentence in any correction. Defined operationally, the screen is three questions:

Trait	The question it answers	The name that shows it
Proven criticality	If every budget gets cut, is this the line item that survives?	Lumentum, Coherent, Tower
Countercyclical revenue	Does the chaos itself pay the company?	Interactive Brokers
Self-funding	If capital markets closed tomorrow, does growth continue?	Alphabet

The best evidence for each trait is never a narrative. It is a document: a purchase commitment, a monthly volume report, an income statement. Below: one trait, one name, the documents.

One thing the screen deliberately does not measure: price. Hold that thought - it is where the whole framework gets uncomfortable.

Trait One: The Spend That Cannot Be Cut

In an AI data center, the chips do the computing - and a different set of components moves the data between them. Lumentum and Coherent build the lasers and optical parts that carry that traffic as light; Tower Semiconductor runs one of the few foundries that manufactures silicon photonics, the technique that builds optical components directly into chips. As clusters scale past what copper wiring can physically connect, the optical layer stops being an accessory and becomes the constraint.

The customer said so on stage. At NVIDIA's GTC keynote in mid-March, Jensen Huang told suppliers he needs more capacity across copper, optical chips, and co-packaged optics - the design that moves the optics inside the switch package itself:

"For the first time, we will scale up with both copper and co-packaged optics."

But talk is cheap and the screen wants documents. It has three.

The customer's money. On March 2, NVIDIA announced $2 billion investments in each of Lumentum and Coherent - $4 billion total - layered with multi-billion-dollar purchase commitments and rights to future capacity. Companies do not pre-buy years of supply from vendors they consider optional.

The index's endorsement. Four days later, S&P Dow Jones Indices announced that Lumentum and Coherent would join the S&P 500 - alongside Vertiv and EchoStar, replacing Match Group and Molina Healthcare - effective March 23. Two optics suppliers entered the index the same week the index itself was breaking down.

The income statements. Lumentum's December quarter, reported February 3: revenue of $665.5 million, up more than 65% year over year. Coherent's datacenter segment booked more than four dollars of new orders for every dollar shipped - a book-to-bill above 4x, which is what demand looks like when the customer is panicking about supply, not price. Tower's silicon-photonics revenue went from $106 million in 2024 to $228 million in 2025, up 115%, and NVIDIA announced a collaboration with Tower on 1.6-terabit silicon photonics on February 5.

These stocks had already run before the crash - twelve-month moves as of mid-March were roughly +851% for Lumentum, +267% for Coherent, +253% for Tower. Then the correction tested them. Coherent fell 20.3% from its March 2 record close of $298.91 to $238.21 on March 31. Lumentum was down 17% from its record by mid-March.

And then the test reversed:

In five sessions off the March 31 low, Lumentum gained +16.1% and closed April 2 at a new record - the bid came back to the most supply-constrained layer first. Tower added +8.8%, Coherent +7.1%. Tonight's scoreboard against the March 2 deal-day marks:

	April 7 close	March 2 close (deal day)	vs deal day	NVIDIA entry price	vs NVIDIA entry
Lumentum (LITE)	$815.75	$783.25	+4.1%	$695.31	+17.3%
Coherent (COHR)	$255.10	$298.91	-14.7%	$256.80	-0.7%

Read the bottom row. The most informed buyer in the industry - the customer itself, with full visibility into its own order book - is still underwater on Coherent five weeks after writing the check, and it bought Lumentum well. The screen says the criticality is real. It does not say the entry was.

One structural note: as of tonight there is no US-listed pure-play photonics ETF. Owning the layer means owning the names directly - part of why the moves are violent in both directions.

Trait Two: The Business the Panic Pays

Interactive Brokers is a brokerage: it earns commissions when clients trade and interest on the cash and margin loans in their accounts. That first revenue line has a property almost nothing else in the market has - it is paid by volatility. Scared money trades more.

The fourth-quarter results, reported January 20, are the cleanest expression of the model. Pretax profit margin - the share of every revenue dollar left before tax - came in at 79%, both as reported and adjusted, up from 75% a year earlier. The script for almost any other company would treat that as a typo. Daily average revenue trades, or DARTs - the industry's count of commission-generating trades per day - rose 30% to 4.04 million. Customer accounts grew 32% to 4.40 million, customer equity 37% to $779.9 billion, margin loans 40% to $90.2 billion.

Then came the real-time test. In early April, mid-correction, the firm published its March monthly metrics: 4.329 million DARTs, up 25% from a year ago. The worst month for the indexes since this bull run began was, mechanically, one of the busiest months in the firm's history.

Notice what the stock did this week: almost nothing. Up +1.6% over the five sessions, the least of the five names on our screen - no crash, no V-bounce, just a business whose operating metrics are partially hedged against the very thing the market is afraid of.

The catch is the price of admission. At tonight's $68.11 close, the stock trades at roughly 31 times its 2025 earnings of $2.22 per share - a premium multiple that assumes the 25 - 40% growth rates keep compounding. The volatility dividend is real. It is also, visibly, not a secret.

Trait Three: The Balance Sheet That Funds Its Own War

The third trait is the bluntest: in a downturn, the companies that need capital markets are hostage to them, and the companies that fund themselves are not.

Alphabet's 2025, reported in early February: revenue of $402.8 billion, up 15%. Net income of $132.2 billion, up 32%. The script-level claim - that Google does over $400 billion a year and funds its AI buildout from its own profits - survives contact with the 10-K, which is rarer than it should be.

What makes it the cleanest large-cap expression of the self-funding trait is the stack it funds. Alphabet is the only company that simultaneously trains frontier models (Gemini 3 shipped in November), designs its own AI chips (the TPU line - which Anthropic agreed in October to use at up-to-one-million-chip scale), builds the data centers and signs the energy deals, and distributes the output through Search, YouTube, and Android. Compare the alternative funding model: OpenAI's race position required raising the largest private round in history - $40 billion led by SoftBank at a $300 billion valuation, announced a year ago - precisely because frontier AI consumes more cash than any startup ever has. Alphabet pays for the same race out of operations.

But put the bars side by side and the caveat draws itself. Management guided 2026 capital expenditures to $175 - 185 billion. At the midpoint, that is $1.36 of capex for every $1 of 2025 net income - self-funding now means spending every dollar of profit and then some, with operating cash flow covering the difference. The trait still holds; the margin of safety inside it is thinner than the slogan suggests.

The market spent this week deciding it cares more about the trait than the caveat: Alphabet closed tonight at $305.46, up +6.2% in five sessions, the strongest close of the stretch - at about 28 times 2025 earnings.

Where the Screen Fails

Run the screen honestly and it fails in four specific places.

Quality is not downside protection. Coherent - S&P 500 entrant, 4x book-to-bill, $2 billion of customer money - fell 20.3% peak-to-trough while the broad index fell 7%. The most critical supplier in the rack dropped roughly three times as hard as the market. High criticality usually comes wrapped in a high multiple, and high multiples are high beta. The screen tells you what belongs on the list, not what is immune.

Criticality is rented, not owned. NVIDIA deliberately funded two optics suppliers at once, non-exclusively, and is collaborating with Tower besides - the company writing the checks is engineering away any single vendor's pricing power. Optical components also carry one of the ugliest boom-bust histories in hardware; veterans of the telecom bust will recognize the order books. And the cleanest warning is already on the tape: NVIDIA's own Coherent entry is underwater tonight.

The volatility dividend cuts both ways. Interactive Brokers' trade-driven revenue fades whenever markets go quiet, and a meaningful share of its income is interest on client balances - set by the Federal Reserve, not by the VIX. A 31x multiple on a brokerage leaves little room for a calm, rate-cutting year.

Self-funding is necessary, not sufficient. Alphabet's capex guide exceeds its 2025 net income; the search-disruption-by-AI debate is dormant, not settled; and the company still operates under two antitrust dockets - it kept Chrome in the search-case remedies decided last September, but the ad-tech remedies remain open. A fortress balance sheet did not stop the stock trading at $287.56 a week ago.

And the meta-failure: the screen says nothing about when. The Nasdaq's 10% correction can become 20% - corrections that ride on a genuine capex cycle turning would do exactly that - and every trait above would still be true while the marks bled. Worse, the traits are correlated to the thing being doubted: if the next leg down is the market repricing AI capex itself, trait one converts from shield to epicenter. Hyperscaler spending guides are the tripwire to watch.

What This Means

The screen is a sequencing tool, not a timing tool. Its output is a short list and a set of falsifiable conditions - not an entry price.

The work happens before the tape forces a decision. Every document above existed by tonight's close: the March 2 investment terms, the March 6 index announcement, the January 20 margin print, the February capex guide, the early-April volume metrics. None of them required predicting the correction. They required reading.

Each trait has a live gauge. For criticality: hyperscaler capex language and the photonics order books - watch whether book-to-bill holds above 1 as budgets get re-cut. For the volatility dividend: the monthly brokerage metrics, published at the start of every month, the closest thing to a real-time P&L for fear. For self-funding: the gap between Alphabet's cash generation and that $175 - 185 billion bill.

A correction reprices everything; the question is what it actually changed. Five weeks of selling did not amend NVIDIA's purchase commitments, eject anyone from the S&P 500, slow the trade tape, or shrink Alphabet's profits. So far, the only line items that changed were the prices.

That is what the screen is for. Not knowing whether the dip continues - nobody does - but knowing exactly which falling prices you have already verified, before the market asks the question again.

---

Data: Barebone | Sources: NVIDIA - Lumentum and NVIDIA - Coherent investment announcements (March 2, 2026), S&P Dow Jones Indices index change announcement (March 6, 2026), Lumentum FQ2 2026 results, Coherent FQ2 2026 results, Tower Semiconductor Q4 2025 results, Interactive Brokers Q4 2025 results and March 2026 monthly metrics, Alphabet Q4 2025 results | Prices as of April 7, 2026 close