Buy the Rumor, Sell the News: Why a Record Quarter Sank the Stock

"Buy the rumor, sell the news" names a specific, testable mechanism rather than a mood. A stock rallies in anticipation of a known, scheduled event — an earnings report, a product launch, a regulatory decision — and then falls when that event arrives, even when the outcome is good. Participants accumulate exposure into the anticipation (the rumor) and distribute it into the confirmation (the news). The variable that governs the reaction is the surprise relative to expectations, not the sign or size of the reported number.

The formal basis is Samuelson's (1965) result that properly anticipated prices fluctuate randomly: if an outcome is correctly forecast, the expected price change at the moment of announcement is zero, because the forecast is already impounded. Fama's (1970) semi-strong efficiency extends this to all public information — price adjusts as information becomes anticipated, so the announcement of an anticipated result is, in expectation, a non-event. The corollary is uncomfortable and central to this note: the better a market has forecast a good outcome, the less that outcome can lift the price when it lands, and the more the mechanical unwind of the anticipation trade can push it lower.

This distinguishes "sell the news" from its neighbors. Ball and Brown (1968) documented that most of the price adjustment to annual earnings occurs before the announcement, with the residual drift afterward. Post-earnings-announcement drift (Bernard and Thomas, 1989) describes underreaction to the surprise component. "Sell the news" is the opposite corner: when the anticipation has been so complete that a genuine beat carries no surprise, the residual is not drift but the reversal of a crowded position.

Three forces convert a strong, expected result into supply. The expectations bar comes first: when a stock has run hard into a print, the buy-side's internal forecast and the whisper number sit above published consensus, so a result that beats the official estimate can still fall short of what the marginal holder already owned the stock to capture. The run-up itself is the second: if the fortnight before the event delivered the move a good outcome would justify, the good outcome has little left to add. Positioning is the third, and the most mechanical: an anticipated binary attracts event-driven length and option structures, and once the binary resolves that positioning is released as supply unrelated to the fundamentals.

The earnings-announcement premium sharpens the intuition. Savor and Wilson (2016) show that firms scheduled to report earn an economically large abnormal return concentrated in the days around the release — an annualized premium near 10% — compensation for bearing the systematic risk of the announcement. That premium is earned by holding into the print. A stock that has already advanced 20% into its date has, in effect, front-loaded that compensation; the holder who bought the run-up has been paid for the risk before the risk resolved, and has every incentive to realize it once the uncertainty clears.

Attention amplifies the setup. Barber and Odean (2008) show that investors are net buyers of attention-grabbing stocks — those in the news, with abnormal volume, or with extreme recent returns — precisely the profile of a mega-cap running vertically into a marquee print. Attention-driven demand extrapolates the run and crowds the long side; when the event passes and the attention shifts, the marginal buyer has already bought, and the overextension left behind is what reverses. None of this requires bad news. It requires only that the news be expected and the positioning be crowded.

A pre-event run of this kind is not a smooth accumulation; it is a compounding of directional flow and dealer hedging. As call demand builds into an anticipated print, the options market-makers who sell that demand are left short gamma and must buy the underlying as it rises to stay delta-neutral — a positive-feedback loop that steepens the run beyond what the order flow alone would produce (Garleanu, Pedersen, and Poteshman, 2009). The same dealers accumulate exposure to vanna and charm: as the event nears and implied volatility stays elevated, their hedge is sensitive both to the spot move and to the passage of time, and both sensitivities lean the hedging flow toward buying into strength.

The unwind runs the loop backward, and does so on a schedule. The instant the binary resolves, the event uncertainty that supported elevated implied volatility is gone. Implied volatility collapses — the crush — and with it the dealers' gamma and vanna exposures reset abruptly. Hedges that required buying the underlying on the way up now require selling it, and the options that financed the run-up expire or are closed. Ni, Pearson, and Poteshman (2005) document that dealer delta-hedging is powerful enough to cluster stock prices around strikes on expiration dates, shifting aggregate capitalization by billions; the same hedging machinery that pins and lifts a name into an event releases it once the event clears.

The consequence is that the post-event decline is partly a plumbing event, not a verdict. The people who most needed to own NVIDIA into the number already did; the dealers who were mechanically long from hedging are mechanically flat afterward; and the option premium that levered the anticipation has decayed to zero. That structural de-risking is a source of supply with no informational content about the data-center business — which is exactly why a record quarter can coincide with a falling stock.

The claim that an outcome is "priced in" is only useful if it can be assessed ex ante. Four measurements, taken together, do that. The run-up magnitude is the first and bluntest: a near-vertical advance on heavy volume into a scheduled date — here +20% in seven sessions — is direct evidence that the good outcome is being discounted in advance rather than after the fact. The steeper and more crowded the run, the less surprise capacity remains.

The options-implied move is the second and most precise. The price of the at-the-money straddle expiring just after the event encodes the market's expected absolute move on the print; a large implied move confirms that a big reaction is already the base case, so only a result outside that implied cone can move the stock durably. Comparing the implied move to the eventual realized move is the cleanest test of whether the news surprised: when the realized move lands inside the implied cone, the event was, by construction, priced.

The implied-volatility term structure is the third. Ahead of a known catalyst, near-dated implied volatility sits above longer-dated — an event-driven kink — because the print concentrates uncertainty into a single session. That kink guarantees a volatility crush the moment the event clears, and the crush is itself a headwind for the stock, because the dealer hedges that the elevated volatility supported unwind as it collapses. Positioning proxies are the fourth: call-heavy open interest and skew, elevated short-dated call volume, and abnormal turnover all proxy for how crowded the long side has become. Read jointly, these four say the same thing the adage says — that the surprise, not the result, is what price trades on, and that the surprise had already been arbitraged toward zero before the report.

The two phrases are often used interchangeably; they describe different resolutions of the same setup. "Priced in" is the neutral case: the market has fully discounted the expected outcome, the result matches expectations, and the reaction is a non-event — price barely moves because the surprise is genuinely zero. This is the pure Samuelson (1965) case, and it is the modal outcome for a well-forecast release.

"Sell the news" is the active case: the result is confirmed as good, yet the stock declines, because the confirmation triggers the release of the positioning that built the run-up. The difference is the unwind. In the neutral case there is little crowded length to release; in the sell-the-news case the run-up was steep and financed by event-driven flow and options, so resolving the uncertainty forces mechanical de-risking that overwhelms the (absent) fundamental surprise. Distinguishing the two ex ante is a question of degree — how crowded, how vertical, how much of the reaction the options already imply — which is why the four measurements above are the operative diagnostic rather than the adage itself.

Both differ from post-earnings-announcement drift, in which the market underreacts to a genuine surprise and price drifts in its direction for weeks (Bernard and Thomas, 1989; Ball and Brown, 1968). Drift is a surprise slowly absorbed; sell-the-news is the absence of surprise plus the presence of a crowded position. The signs are opposite, and confusing them is how a fade of an anticipation trade gets mistaken for a bet against the business.

Coming into late May 2026, NVIDIA had staged a powerful, crowded advance. From a $196.50 close on 5 May the stock reached a $235.74 peak on 14 May — roughly +20% in seven sessions — on heavy volume, as the market positioned ahead of a fiscal-first-quarter print that had become one of the most anticipated events on the calendar. By the run-up measurement alone, the good outcome was being discounted in advance: a move of that magnitude into a scheduled date is the market pre-paying for a beat.

Momentum then rolled over before the date. The stock made its high on 14 May and drifted lower into the print — $225 on 15 May, $222 on 18 May, $221 on 19 May — lower highs into peak anticipation, the signature that the anticipation trade was complete and the marginal buyer had been exhausted. This is where the crowded long side and the dealer hedging that steepened the run become a liability: with the print imminent, the elevated implied volatility supporting the hedges was primed to crush, and the positioning was primed to release.

This note, written in June 2026, classifies the 20 May close of $223.47 retrospectively as a SELL — a sell-the-news fade of the pre-event run — over a five-to-eleven-session horizon, with a decline band of -3.1% to -8.6% and an invalidation at a volume-confirmed reclaim of the $235.74 run high. The thesis is that the reaction was governed by surprise, not by the absolute result, and that the surprise capacity had already been consumed by the run.

The quarter, when it was reported, was a record by any absolute measure: revenue of roughly $81.6 billion against a consensus near $78.9 billion — a beat of about $2.7 billion — with profit roughly tripled on continued data-center strength. Read as a fundamental event in isolation, it was superb, and it confirmed that the franchise was not in question.

The reaction was muted to negative, and the framework explains why without appeal to anything bad in the report. A beat of that size sat comfortably inside the move the run-up and the options had already discounted; measured as a surprise against a buy-side bar lifted by a 20% advance, the incremental information was close to zero (Samuelson, 1965; Fama, 1970). The earnings-announcement premium that a holder earns for carrying the print (Savor and Wilson, 2016) had been front-loaded into the run and was now being realized. And the resolution of the binary triggered the volatility crush and the mechanical de-risking of the crowded, dealer-hedged long side described above — supply that the beat, however large, could not absorb.

This is the direct answer to "why did NVIDIA stock drop after earnings": it dropped not despite the beat but because the beat had already been bought. A great result with no surprise, arriving on top of a crowded run, is precisely the configuration in which good news is sold.

Price worked lower through the window. NVIDIA slipped to $215 by 22 May, to $211 by 29 May on the heaviest volume of the stretch — consistent with the de-risking unwind rather than fresh information — and, after a brief bounce back toward the 20 May close, to a window trough of $205.10 on 5 June, down 8.2% from the $223.47 reference close at roughly session eleven. The realized decline sits in the lower half of the -3.1% to -8.6% band this classification sets.

Neither invalidation condition this classification sets was met on the realized path. The $235.74 run high was never reclaimed inside the window, so the volume-confirmed-reclaim boundary that would have signaled real incremental demand assuming control was not crossed. No distinct new catalyst arrived inside the horizon to re-rate the story. The path was not monotone — sell-the-news unwinds rarely are, because the underlying story is still good and dip-buyers keep testing it (an arbitrage bounded by capital and risk, per Shleifer and Vishny, 1997) — but the direction is the one the thesis describes: the anticipation trade unwound once the event that justified it had passed, and a record quarter was not enough to hold a price that had already discounted it.

The classification is a SELL referenced to the $223.47 close of 20 May, placed at the rollover rather than at the high — it takes the loss of upside momentum (the sequence of lower highs into the date) as confirmation that the anticipation trade was exhausted, rather than trying to pick the $235.74 top. Entering on the rollover trades a few points of give-back for confirmation that the crowded long side has stopped adding.

The horizon was deliberately short — five to eleven sessions — because the edge is the unwind of event positioning, and that unwind has a finite life. Once the volatility crush has completed and the event-driven length has been released, the mechanical source of supply is spent; extended past that horizon, the classification would decay into an open-ended fundamental short of a franchise it does not set out to short. The return band of -3.1% to -8.6% was scaled to the run-up it was fading, not invented as a round number.

The invalidation this classification sets is a volume-confirmed reclaim of the $235.74 run high — the level whose recapture on expanding volume would show that genuine incremental demand, not covering and hedging flow, had resumed control — or the arrival of a genuinely new catalyst distinct from the anticipated print. Either event would falsify the premise that the advance was a completed, positioning-driven anticipation trade. Defining the exit at entry is what keeps a live sell-the-news classification falsifiable rather than a permanently deferred bearish opinion.

Sizing follows from the invalidation, not from conviction. With the falsifying level fixed at the $235.74 high, the distance from the $223.47 reference close to that reclaim defines the risk per unit, and a position built on this classification is sized so that a volume-confirmed reclaim costs a bounded, pre-agreed amount. A short-horizon fade with a hard invalidation is sizeable precisely because the loss is capped by a defined boundary rather than by a discretionary stop.

The classification's horizon ends before the next catalyst by design. A separate event in early June — the Computex keynote and its product announcements — gave the stock a fresh, distinct reason to rally after the classification window, and momentum re-energized. That move sits outside the stated five-to-eleven-session horizon and reflects a new catalyst rather than the anticipation trade classified here, so the classification does not claim it; extending the classification into it would convert a spent statistical edge into an unhedged bet on a product cycle this note takes no view on. Attributing the June rally to the May fade would be the outcome-conditioned narrative the whole discipline is built to exclude.

The bounded contribution is this: a crowded, near-vertical run into a scheduled catalyst carries sell-the-news risk regardless of how good the news proves, because price trades on surprise relative to expectations (Samuelson, 1965; Fama, 1970), the announcement premium is earned in the run-up (Savor and Wilson, 2016), and resolving the binary forces a mechanical unwind of dealer-hedged, attention-driven length (Garleanu, Pedersen, and Poteshman, 2009; Ni, Pearson, and Poteshman, 2005; Barber and Odean, 2008). Recognizing the run's exhaustion and respecting the reclaim-the-high invalidation is the edge; "the company beat and the stock fell" is a question to be measured, not a verdict on the business.

Buy the rumor, sell the news — a stock rallies in anticipation of a scheduled event and falls when it arrives, because a properly anticipated outcome is already impounded in price before the announcement.

Priced in — the market has fully discounted an expected outcome, so the result produces little reaction; the neutral, non-event case, as opposed to the active decline of sell the news.

Options-implied move — the expected absolute move on the event encoded in the at-the-money straddle expiring just after it; comparing it to the realized move tests whether the result surprised.

Implied-volatility crush — the collapse of elevated, event-driven near-dated implied volatility once the binary resolves; it removes the support for dealer hedges and is itself a headwind for the stock.

Dealer gamma / vanna / charm hedging — the delta-hedging that option market-makers do as spot, volatility, and time move; short-gamma dealers buy strength into an event and sell weakness out of it, with no informational content.

Earnings-announcement premium — the abnormal return earned for holding a stock through its scheduled earnings release; a run-up front-loads this compensation and gives the run-up buyer a reason to sell once the print clears.

Sell-the-news exhaustion — the rollover in price and momentum (lower highs) into a known catalyst that signals the crowded anticipation trade is complete.

Invalidation level — here, a volume-confirmed reclaim of the run-up high, which would show real incremental demand and falsify the fade thesis.