Are Bubbles Worth Predicting?
“Bubble” is underspecified. Whether AI should be labeled a bubble has raised the salience of the term recently, but it’s a broader issue.
As the word is used in casual conversation, “bubble” refers to an asset that has these qualities:
- The asset value rises and then drops dramatically
- The asset’s value at its peak is “irrational”
Both of these criteria have some problems.
Boom and Bust… Eventually
Lots of financial assets rise and fall in value, but we generally recognize this as a natural feature of asset valuation. Apple stock has bounced around, mostly rising, since 1980. On Tuesday, it dropped 2%. Was Apple a bubble all along?
No, basically no one would call modern Apple a bubble. It’s a company that’s repeatedly proven the ability to build products that people will pay for, and in the recent era it’s had more cash in the bank than it knows what to do with.
So the asset needs to have a dramatic fall in valuation, more than a few percent. But when?
Plenty of people have been calling Tesla a bubble, or at least bubble-adjacent, for a decade. But it’s trading at more than an order-of-magnitude higher valuation than in 2016, and has been for years now1.
For “bubble” to be a useful, falsifiable claim we need an upper bound on how soon it will pop. We should be able to make a statement like “Bubble assets drop at least 30% in value over a 3 year period”.
But there is no accepted timeline. To paraphrase Keynes, a bubble can remain unpopped longer than whatever timeline you assign to test it. People have been calling Bitcoin a bubble for ten years now and will probably be doing it for another ten.
Irrationality
Then let’s take irrationality. Even in retrospect, identifying which asset values were “irrational” is purely subjective.
Do we ask every investor? Every short seller? All market booms have naysayers. The naysayers of WeWork came out looking pretty smart. Meanwhile, some investors were vocally skeptical of Netflix’s valuation back in 2010/2011, and since then the stock has gone up something like 25x2. Whoops.
Even when there’s a crash, the valuations weren’t necessarily “wrong”. Take Peloton.
During the Covid era, home workouts were all the rage, and Peloton shares climbed to $160 or so in December 2020. By 2022, they were trading in $20-$30 range – roughly where they’d been before Covid. Was Peloton priced irrationally? It had an epic peak and then crashed, but it really did have exceptional revenue growth during this era; it wasn’t purely investors hyping themselves up.
Ultimately, having a definition so broad makes “bubble” more a narrative claim than an empirical one. There is no proving that prices are rational or that an asset isn’t going to crash someday.
This isn’t to say it’s a meaningless word. For better or worse, we’ve converged on using it to describe the 2007 housing market, the 1999 dotcom crash, and the 2021 Gamestop insanity. But this is just a vibes-y consensus, not a claim about the fundamental nature of what happened – these events didn’t qualify based on rigorous post hoc assessment, they just got labeled this way.
There’s no falsifying the claim that something is a bubble in real time or (in any substantial way) even years down the line. If you want an interesting discussion, that’s what empirical forecasts are for.