Couldn't agree more on the quality and depth of the article.
I have only one (minor) issue with it. While it's true that the Citadels of the world have zero interest in using "a machine learning algorithm on your neighbor’s GM trades in their IRA to inform their thinking of the true value of GM as a business," I suspect there is value to them in being able to model and predict the aggregate behavior of a mass of retail investors, e.g., for use as an input to macro investing strategies. Retail investors have a reputation for being the last to join every party; there is value in knowing what they are doing in the aggregate.
You don’t need to be an internalizer to capture that, though. Just read the tape: the order for 3 shares that got the sub penny price improvement is the retail one.
(There’s a paper on this somewhere that I think I remember reading when doing level design for Starfighter.)
I have only one (minor) issue with it. While it's true that the Citadels of the world have zero interest in using "a machine learning algorithm on your neighbor’s GM trades in their IRA to inform their thinking of the true value of GM as a business," I suspect there is value to them in being able to model and predict the aggregate behavior of a mass of retail investors, e.g., for use as an input to macro investing strategies. Retail investors have a reputation for being the last to join every party; there is value in knowing what they are doing in the aggregate.