Add to all of these that the return is uninvestable, so the market cannot eliminate the arbitrage.
1016x over 30 years is a gain of less than 0.001% per trading day. You would have spent much more than this through fees and market impact to buy and sell the stock each day, especially at any real volume.
The market is full of uninvestable or tiny capacity trades like this. A reliable pattern gets created by trading flows like the ones described in other replies, and it doesn't get arbed out because it is uninvestable or has tiny capacity.
Imagine how many HFT algorithms are out there trying to exploit every micro trend to the extreme. I am not in that game, but I can imagine all the things I would try, and I know there are really smart people and obviously a ton of money invested in it. Trying to find something simple that consistently makes money for a long time would be nearly impossible because the patterns would be found and exploited by many until they scrubbed out any potential gains. (But then again if someone did find it they wouldn't share). So you need to get novel and pour all sorts of external data into your models to have any edge. Something this simple just wouldn't be a thing. But it is still interesting to understand.
Algo trading teams are indeed research shops that have to continually replace their strategies. Often times success is less about the cleverness of a prediction than clever engineering (technical or contractual!) that makes a known pattern investable. Or else about being earlier to identify emerging, short lived patterns so that you can profit from more of their lifecycle. Teams make long term investments in technologies and strategies that help them do these things repeatably.
I think at this point the traditional stock market has so much algorithmic trading going on that making your own algorithm that can consistently win is basically impossible. But there are other less popular markets that don't receive the same attention.
It depends. Sure an obscure crypto or betting exchange does not have many competitors, but that's not because no one cared, but because even with perfect strategy it's going to be hard to make $$ if daily volume is 10k. How much can you reasonably extract?
Conversely, traditional stock markets is where relatively unsophisticated players do multi-million transactions all the time. This is where the money is. You might need to fight for it or it might be handed over on a silver plate by some redditors.
I don't think too many people would suggest algorithmic trading as a retirement strategy. Nor is investing in a single company a great plan, regardless of their dividend stability. I'm a confirmed Boglehead myself, but I do enjoy playing games with my lunch money. Options, crypto, or just individual stocks if I like the company. And yeah, it does help if they are a Dividend Aristocrat if I plan to hold that stock for any length of time.
I've also considered trying some algorithmic trading at lesser known markets, because it is interesting.
It seems like a HFT trader could arbitrage this pattern out by just taking it into account while planning other trades. If a HFT firm makes millions of trades a day, then they could just slightly bias their buys/sells by considering the .1% gain from holding overnight.
1016x over 30 years is a gain of less than 0.001% per trading day. You would have spent much more than this through fees and market impact to buy and sell the stock each day, especially at any real volume.
The market is full of uninvestable or tiny capacity trades like this. A reliable pattern gets created by trading flows like the ones described in other replies, and it doesn't get arbed out because it is uninvestable or has tiny capacity.