Why I Was Wrong to Be Bearish on U.S. Stocks
A year ago I wrote the blog post Why I'm Bearish on U.S. Stocks (for the Second Time Since 2017). Since then U.S. stocks are up 16% (total return) and my bearish prediction turned out to be misguided.
So where did I go wrong? And how can you prevent yourself from making the same kind of error in the future?
As you can see, the P/S metric peaked at 3.41 in 1999. After seeing the run-up in the ratio in 1999, in 2021, and again in 2025, I concluded that we were in bubble territory.
Unfortunately, this data series was inaccurate, but I didn't know it at the time. I'm not sure exactly what changed in how the data was constructed. What I can say is that the revised series matches other sources I've cross-checked.
If you go to look at the price-to-sales ratio of the S&P 500 today on DQYDJ.com, the 1999 peak is now around 2.09 (instead of 3.41):
This chart tells a very different story. When the P/S ratio passed 2.09 (the 1999 level) in 2017, market conditions weren't anything like those during the DotCom bubble. No one thought there was the same kind of froth in 2017 as in 1999. And the ratio has kept climbing since, to around 3.5 today, without any accompanying DotCom-style implosion. Therefore, the only correct interpretation of the P/S ratio over the last decade is that it doesn't signal what it used to.
I've made this argument before when discussing the problem with valuation metrics, especially the P/E ratio. It looks like the same argument can be made about the P/S ratio too. Since many of today's companies have higher margins, the aggregate P/S ratio can rise even as companies remain fairly valued.
I don't blame DQYDJ.com for the data mishap. Corrections get made all the time. I get it. It's no one's fault. But if I had the correct chart above (instead of the incorrect one), it would have given me pause. Of course, it's possible that I would've overlooked this and found another piece of evidence to confirm my theory, but it's difficult to say.
Either way, I learned a valuable lesson from my failed bearish prediction a year ago—it's easy to overfit the data. If you want to find parallels between one period (2025) and another (2021), you'll be able to. You can find opinion pieces, charts, and investor behavior that are similar across any two periods if you look hard enough.
My issue wasn't that the parallels didn't exist—they did. My issue was that much of the speculative stuff of 2021 ended up failing (NFTs, DeFi, etc.) while much of the speculative stuff of 2025 (AI) seems to be succeeding. I pattern matched the frothy behavior and concluded that the result would be the same. But I was wrong.
The best example of this is Anthropic, which had a revenue run rate of around $5 billion in July 2025 (when I first got bearish). This was up from $1 billion in January 2025. Its growth was incredible, but it couldn't continue, right? Wrong. As of late July 2026, Anthropic's run rate was estimated at $74 billion, or about 15x higher than a year prior.
I know Anthropic is an outlier and run rate isn't the same as profit, but the company's meteoric rise illustrates why this time is actually different. The growth is happening, the models are getting better, and more people are using AI. Note that this is evidence about AI adoption, not market-wide valuation. A private company's run rate tells you little about whether the S&P 500 was fairly priced or whether future stock returns will be positive. But it does highlight why I got the direction wrong a year ago.
Of course, this isn't the only reason I was mistaken. Because I also forgot one of the oldest and most consistent lessons about markets and life.
This Time is Different (When Overfitting Fails)
A year ago I saw a few signs that reminded me of the 2021 market exuberance:- Chamath Palihapitiya was filing for a new SPAC
- Meta was paying $250M+ to hire individual AI researchers
- The S&P 500's Price-to-Sales ratio was back near an all-time high
As you can see, the P/S metric peaked at 3.41 in 1999. After seeing the run-up in the ratio in 1999, in 2021, and again in 2025, I concluded that we were in bubble territory.
Unfortunately, this data series was inaccurate, but I didn't know it at the time. I'm not sure exactly what changed in how the data was constructed. What I can say is that the revised series matches other sources I've cross-checked.
If you go to look at the price-to-sales ratio of the S&P 500 today on DQYDJ.com, the 1999 peak is now around 2.09 (instead of 3.41):
This chart tells a very different story. When the P/S ratio passed 2.09 (the 1999 level) in 2017, market conditions weren't anything like those during the DotCom bubble. No one thought there was the same kind of froth in 2017 as in 1999. And the ratio has kept climbing since, to around 3.5 today, without any accompanying DotCom-style implosion. Therefore, the only correct interpretation of the P/S ratio over the last decade is that it doesn't signal what it used to.
I've made this argument before when discussing the problem with valuation metrics, especially the P/E ratio. It looks like the same argument can be made about the P/S ratio too. Since many of today's companies have higher margins, the aggregate P/S ratio can rise even as companies remain fairly valued.
I don't blame DQYDJ.com for the data mishap. Corrections get made all the time. I get it. It's no one's fault. But if I had the correct chart above (instead of the incorrect one), it would have given me pause. Of course, it's possible that I would've overlooked this and found another piece of evidence to confirm my theory, but it's difficult to say.
Either way, I learned a valuable lesson from my failed bearish prediction a year ago—it's easy to overfit the data. If you want to find parallels between one period (2025) and another (2021), you'll be able to. You can find opinion pieces, charts, and investor behavior that are similar across any two periods if you look hard enough.
My issue wasn't that the parallels didn't exist—they did. My issue was that much of the speculative stuff of 2021 ended up failing (NFTs, DeFi, etc.) while much of the speculative stuff of 2025 (AI) seems to be succeeding. I pattern matched the frothy behavior and concluded that the result would be the same. But I was wrong.
The best example of this is Anthropic, which had a revenue run rate of around $5 billion in July 2025 (when I first got bearish). This was up from $1 billion in January 2025. Its growth was incredible, but it couldn't continue, right? Wrong. As of late July 2026, Anthropic's run rate was estimated at $74 billion, or about 15x higher than a year prior.
I know Anthropic is an outlier and run rate isn't the same as profit, but the company's meteoric rise illustrates why this time is actually different. The growth is happening, the models are getting better, and more people are using AI. Note that this is evidence about AI adoption, not market-wide valuation. A private company's run rate tells you little about whether the S&P 500 was fairly priced or whether future stock returns will be positive. But it does highlight why I got the direction wrong a year ago.
Of course, this isn't the only reason I was mistaken. Because I also forgot one of the oldest and most consistent lessons about markets and life.