Not a dinosaur yet: how playing with AI talked me out of a trade I'd believed for years

My birthday was this month (I turned 38), and I’m definitely feeling more and more middle aged. I’m more sore the day after workouts, a bunch of greys are starting to bloom through my beard, and I had to quit drinking ~four years ago because the hangovers were getting unbearable.

All of those are inevitable signs of aging. Unfortunately, we can’t fight father time.

But in the past few months, I’ve caught myself seeing things online and saying “Oh man, if I was 20 I would have been all over that.”

Sometimes, it relates to something silly and that I’m honestly glad I missed out on. If I was in my early 20s right now, I have high confidence that my friends and I would have gotten way too into betting on niche prediction markets and live streaming our reactions as our bets played out. I cannot imagine how embarrassing those videos would have been with hindsight!

But I’ll frequently find myself thinking that about something with AI. “Oh man, I wish I was younger, because I’d be all over using AI for that or figuring out how to apply AI to that.”

And I’m positive I’m not alone in that feeling. These days, most conversations I have with investors inevitably come back to AI at some point. Yes, we will bemoan not being YOLO long in memory stocks, or we might talk about how AI will impact XYZ company…. but the main way it turns back to AI is that we talk about how we’re using AI to improve our research process. With a lot of those conversations, I can sense a little hesitation or fear from the person I’m talking to. They’re scared of AI; not in a “will AI take our jobs” way but in the very human way that we’re all scared to try out something new. The person doesn’t want to “try” AI and be “wrong” or do something silly. They’ll use a chat interface as a form of “super google”, but they don’t want to have AI start summarizing earnings calls for them or build out unique tools or do anything else. They’re worried about failing, or getting hallucinations, or stepping out of their comfort zone.

So I’m writing this article both for that person and myself: we’re not dinosaurs. There’s nothing to be scared of. Just start throwing stuff at AI and seeing what it can and can’t do. There’s no harm in trying, and sometimes you’ll find really interesting new stuff or AI will help you find a flaw in your thinking.

Let me give you a personal example. For years, I’ve thought that shorting double levered ETFs was “free alpha” given volatility drag¹. The basic thought here is that, in order to maintain 2x leverage, levered ETFs are forced to buy more stock as it rises and sell stock as it falls. Thus, by their very nature, levered ETFs buy high and sell low, and my thought was that if you could short a double levered ETF, you could thus capture this volatility drag for yourself and profit from a fatally flawed product design.

That thought was largely the table below (this one from MUU’s prospectus), which is contained in the prospectus for basically every 2x levered ETF. It shows that if the underlier goes up 50%, the 2x levered ETF can have a negative return if the volatility is high enough:

You could even see these results in practice; there’s something crazy about seeing TSLA be basically flat over the past 18 months but the levered ETF being down 50%:

So my simple thesis was that you should short 2x levered ETFs on highly volatile names (and perhaps hedge yourself by going long the underlying) to take advantage of the volatility drag. I know I’m not the only one to have that thought; Einhorn wrote about it in his 2024 annual letter.

I’ll pause here to remind you that shorting is crazy risky; I’m not making investing recommendations (in fact, as you’re about to see, this particular trade carries massive tail risk and does not work!). And I’ll also note that the borrow on these double levered ETFs is so thin that, in practice, these are very hard to do (a major reason I had stopped trying to do these)!

This year, we’ve seen an explosion in 2x and 3x levered ETFs. These have mainly targeted “AI winners” like memory stocks and semis. Memory stocks and semis are insanely volatile names, so I thought it was time to dust off a “levered ETF structural decay” basket.

Fortunately for me, I started talking to Claude about different ways to structure the trade. And Claude noted something to me: the volatility drag “free alpha” was something of a mirage. You were kind of getting paid for a tail risk, because if the stocks went up enough the 2x levered ETFs would outperform the underlying even with very high volatility. So, for example, if the stock went up 30%, the 2x ETF would be down 37% with 75% vol…. but if you added a zero and the stock went up 300%, the 2x levered ETFs would be up >800%. The 2x levered ETF would have substantially outperformed the underlying, and if you were running a “buy 2x the underlying, short the 2x levered ETF” vol drag trade, you’d have your face ripped off!

By the way, I did not choose that 300% return number by accident. MU is up ~300% so far this year. That’s an insane result, and it’s driven their 2x levered ETF to go up an even more insane ~976%:

That’s right: MU’s parabolic move has caused the 2x levered ETF to substantially outperform 2x the underlying. In fact, it’s up more than 3x the underlying!!!!

Anyway, turning back to AI: it sounds silly given how much I’ve used it and some of the cool tools I’ve built with AI, but this might be the most useful thing AI has done for me all year (particularly if you’re judging on a dollars saved/made basis!). Because I was just fooling around with AI and forcing myself to use it to stress test different ideas and trades, it quickly broke an idea (2x levered ETF underperformance = “free alpha”) that I have been carrying for years (and, to judge from public letters and my private talks with others, that I am certainly not alone in having carried!).

I think there are two lessons here.

First, the whole saga is a useful reminder that markets are really competitive. "Free alpha" usually isn't, and it's better to check your ego and let a tool like AI stress test a setup for tail risks than to put the trade on, call it free money, and get blown out (as I would have on my hypothetical MU / MUU trade).

Second, I may be middle aged, but I’m not a dinosaur (and neither are you). Don’t sit there saying “I wish I was younger and could use AI to do XYZ.” Just get out there and start playing with it. You never know what you’ll find (or what mistakes AI will stop you from making!).

PS- might as well plug myself while I’m here! If you’re interested in improving in AI, I’m doing an AlphaSense webinar focused on using AI as an investor; you can sign up here to access the webinar when it launches later this week

1. Disclaimer: shorting is risky and I was being very cavalier with my terms; go check out our legal disclaimer here!