Beat the Estimate, Lose the Week: What the July 2026 Earnings Season Actually Grades

Market Cycle Commentary

Through Thursday's close, more than 87% of the first forty S&P 500 companies to report this quarter had beaten Wall Street's estimates. The $SPX still lost 1.6% on the week. $NFLX beat on earnings and fell 7%. $TSM beat and got sold. And IBM, listed on the New York Stock Exchange since 1916, had the worst single day in its recorded history. Before you hold anything through a print this season, you should know what the reaction actually grades, because it was never the income statement.

An eight-cent miss and the worst day since records began

On Tuesday, July 14, the big banks reported blowout quarters, and $IBM published a letter nobody had on the calendar. The company wasn't due to report until the following week. Instead it pre-announced preliminary results more than a week early: revenue of $17.2 billion where the Street was modeling $17.86 billion, and adjusted earnings of $2.93 a share against the $3.01 analysts expected, per FactSet.

Read those numbers again. A 3.7% revenue shortfall. An eight-cent miss. The stock closed down 25%, its worst day on record, deeper than the roughly 23% it lost on Black Monday in October 1987. The daily records track back to 1968; the listing goes back to 1916. In more than half a century of recorded sessions, including Black Monday and the entire dot-com unwind, the stock had never lost this much in a day. An eight-cent miss did it.

CEO Arvind Krishna's letter explained that in the last weeks of June, clients shifted their quarterly capital spending toward servers, storage, and memory to lock up supply-constrained AI infrastructure before prices rose, and that IBM "did not anticipate the magnitude of the capex reprioritization." Whether that explanation satisfies you is beside the point for a trader. The arithmetic of the reaction is the point. The results missed the forecast by single-digit percentages and the repricing was 25%. What got graded was the error in the market's own model of the quarter, and the grading came on a curve.

You're not trading the company. You're trading the crowd's forecast error.

A stock price on the day before a report is a container holding every forecast the market has already made about the quarter, the guidance, and the decade after that. When the print lands, the market never asks whether the results were good. It asks whether they were different from what it had already paid for. John Maynard Keynes described this game in 1936, in chapter 12 of The General Theory: professional investing as a newspaper contest where the prize goes not to whoever picks the prettiest face, but to whoever best predicts the average pick. "We devote our intelligences to anticipating what average opinion expects the average opinion to be." Earnings season is that sentence with a countdown clock attached.

Last week's tape ran the demonstration in public. Netflix earned 80 cents a share against an expected 79, grew revenue 13% year over year to $12.56 billion, and made $3.4 billion in net income, up from $3.13 billion in the same quarter last year. Then it guided the current quarter to $12.86 billion in revenue, under the roughly $13 billion analysts were modeling, and narrowed its full-year revenue range to $51 billion to $51.4 billion, from an earlier $50.7 to $51.7 billion. Guiding under the model and trimming the top of a forecast are not losses on any accounting statement ever printed. The stock fell more than 7% on Friday anyway, because the market had already paid for the quarter Netflix delivered. The number it repriced was the ceiling Netflix implied.

Taiwan Semiconductor posted a better-than-expected quarter on Thursday and raised its capital spending plan for the year to a range of $60 to $64 billion, up from $52 to $56 billion. Good quarter, bigger bill. The stock lost more than 2%, and the semiconductor complex went down around it; the VanEck Semiconductor ETF dropped almost 4% that day. Two months ago this market was paying almost any price for aggressive AI spending. Last week it started penalizing it. Same companies, same theme, different average opinion. If that whipsaw feels familiar, it's the crowding mechanism we walked through in the June selloff post: when everyone owns the same idea, the grade depends on the crowd's mood, not the company's execution.

• $NFLX: beat on earnings, grew 13%, fell more than 7%.
• $TSM: beat the quarter, raised the spending plan, fell 2% and took the chip complex with it.
• $IBM: missed by eight cents, lost a quarter of its value.
• The index: an 87% beat rate, down 1.6% on the week.

One column of that scoreboard is results. The other is reactions. They don't reconcile, and they were never supposed to.

Holding through the print is a position without a stop

Here's where this stops being commentary and starts being about your account. Every risk tool you use during a normal session assumes a continuous market. The stop assumes there will be a price between here and disaster where someone takes the other side. An earnings print breaks that assumption. Netflix reported after Thursday's close; by the time any holder could act, the repricing was already in Friday's open. And when information hits fast enough, price doesn't walk to your stop. It teleports past it, and your fill, if you get one, comes from whatever is waiting on the other side of the air pocket.

That's why the only risk decision that survives an earnings print is the one you make before it: size. Not the stop, not the alert, not the plan to watch it closely. Size. If the position is small enough that the worst plausible gap is an annoying day, you're a trader holding through an event. If it's big enough that the worst plausible gap changes your month, you've bought a lottery ticket denominated in someone else's forecast error. We made the longer argument in position sizing is a psychology problem, and July 14 added the footnote that argument needed: the worst plausible single-day repricing for a hundred-year-old blue chip is now, verifiably, 25%.

One more assumption died that Tuesday: the calendar itself. IBM wasn't scheduled to report for another week. "I'll be flat before the print" is a fine plan right up until the print moves itself.

The two chairs the morning after

After a gap, there are two seats you can be sitting in, and each comes with a pre-installed mistake.

If you're in the position, the mistake is hope accounting. The gap goes against you and the mind instantly reclassifies the loss as temporary, because the loss aversion Kahneman and Tversky documented in 1979 makes realizing a loss hurt far more than an equivalent gain feels good. "It always comes back" is anesthesia dressed as analysis, and we've written about what it actually feels like at the desk. The trap that follows is averaging down into an information event that just resolved against you, which is doubling a refuted thesis at the exact moment the market finished grading it.

If you're flat, the mistake is the feeling that a violent print owes you a trade. A stock down 25% pulls the fade reflex. A stock up 30% pulls the chase, and we watched that version in slow motion when Dell gapped thirty percent on its own blowout quarter in May. Both urges are feelings about a price, not reads of structure. The professional move after a gap is boring: let the first half hour print, mark the levels the auction actually defends, and trade the second structure, never the first feeling.

What earnings season asks of a process

None of this requires a forecast. It requires a routine.

• Know the dates, and hold them loosely. Every position on your book has a report date. July 14 proved the date is an estimate too.
• Decide hold-or-flat in writing, while the market is closed. The calm version of you makes a defensible call. The 3:55 version negotiates.
• If you hold, size to the gap, not the chart. Ask what a 25% overnight repricing does to the account. That number stopped being hypothetical this month.
• After any gap, trade the second structure. The first move is the crowd repricing its own error. You don't have an edge inside someone else's margin call.
• Journal print trades separately. Ten of them will teach you more about your relationship with binary risk than a year of ordinary sessions.

The question worth sitting with

Pull up the last position you held through an earnings report. Ignore the outcome; the outcome is the least informative part. Look at the size. Then answer honestly: did that size say "I have priced in being wrong by 25%," or did it say "that can't happen to me"? As of July 14, 2026, one of those positions is a decision and the other is a donation, and the market has stopped being subtle about the difference.

If you want to know how you actually behave around risk before the market runs the experiment on your account, the TQ Assessment measures your Risk Management Mindset alongside five other dimensions in about fifteen minutes. The protocols that turn the audit into behavior live in The Complete Calm Trading Method, and Trade Calm covers the nervous system that has to hold a position while a forecast error decides what it's worth. The tape will do something absurd again soon. That part isn't a forecast either.

James