On August 28, 2024, NVIDIA announced its financial results for the second quarter of FY2025. And they were blown away. Quarterly revenue came in at a record $30.04 billion against the expected $28.86 billion, a 4.1 beat. Earnings per share (EPS) posted a higher beat of 6.3% and the guidance also exceeded expectations by 2.3%.
And yet, the NVIDIA stock fell by about 6% in after-hours trading, and a further 6% the next trading day. The stock dropped by approximately 12% in two days after the company released exceptionally strong earnings!
The opposite happened four months earlier on April 23, 2024, when Tesla’s Q1 2024 earnings came out. Quarterly EPS missed expectations by 12% and revenue fell short by 4%. In fact, the revenue declined by 9% year-over-year, which was the worst performance since 2012.
Despite all the misses and declines, Tesla’s stock jumped 13% in after-hours trading. And it rose by 12.1% the next trading day.
These two cases present one of the most interesting paradoxes in investing. That markets do not reward good data or punish bad data, they react to expectations. Why is that?
We will use two arguments to answer this question.
The “Whisper Number” Effect
In November 1999, Mark Bagnoli and colleagues published a seminal study, Whisper Forecasts of Quarterly Earnings per Share, in the Journal of Accounting and Economics. They defined whisper numbers as “informal forecasts derived from private information or market rumors” and proved they were more accurate than official consensus estimates.
Per Afrell, a former UBS Warburg analyst, told the audience at a Riksbank conference in 1999 that analysts typically have complex spreadsheets with evolving calculations. However, they only publish official estimates periodically. And when clients call them asking for the real number, the analysts share their updated models, a practice that creates a disconnect between published (official) and whispered estimates.
This happens to have been the case for NVIDIA. According to Dan Morgan from Synovus, the whisper number for NVIDIA’s Q3 FY2025 guidance was between $33 billion and $34 billion. As you can see, analysts had loftier “whispered” expectations.
Ryan Detrick from Carson Group explained that “death, taxes, and [NVIDIA] beats on earnings are three things you can bank on.” But the issue after Q2 FY2025 results was that the size of the beat was “much smaller than we’ve been seeing.”
So, this tells you that when a stock consistently outperforms, expectations inflate beyond what’s officially projected. The bar keeps rising.
Markets Are Forward-Looking Machines
Elyse Ausenbaugh, CFA, head of investment strategy at JPMorgan Wealth Management, stated in an article in 2022 that “stock and bond markets are forward-looking machines, with prices that reflect what investors think may happen to the economy in the future.” She explained that markets digest risks quickly and recalibrate because they focus on what lies ahead rather than reacting to the latest headline.
This is a view that Baird Wealth Management shares. They explained in a 2021 market insight that the stock market is a “forward‑looking mechanism that discounts the probability of future events.” Contrariwise, economic data are often backward‑looking. And because prices embed expected earnings and growth, the market may rise even when current economic indicators look gloomy.
Plainly speaking, markets discount the future. That is, current data is only valuable as a signal for what’s coming. So, when Elon Musk announced accelerated production of affordable EV models, moving the timeline from H2 2025 to early 2025 or late 2024, Tesla’s terrible Q1 numbers became irrelevant. To investors, Tesla offered hope, and NVIDIA offered merely strong performance.
What This Means for Investors
The short answer to the question in the title is that markets are not report cards for quarterly performance. Rather, they are prediction markets about the future.
As Baird notes, markets often rally during recessions or crises because investors anticipate recovery. This wouldn’t happen if they looked at the data. So, the question isn’t “Was the data good?” but “Was it better than what was already expected?”
