Is this fundamentally any different from the claim that with perfect competition profits are driven to zero? This is a similar sort of paradox: the profit incentive drives production, and yet when the market is “perfect,” this incentive no longer exists.
Though I haven’t looked at the details, I get the feeling that this kind of analysis is based on a scaling problem. Mathematically, if you want to see a macroscopic structure emerge asymptotically, you generally have to choose your scaling parameters correctly. For example, Brownian motion is only the scaling limit of a random walk when dt ~ (dx)². Surely you have to do something similar when talking about “perfectly efficient markets.” It’s very easy for people to get an extremely wrong idea here, thinking that imposing an artificial friction on the market must be a good thing because, hey, the perfectly efficient case erases the whole incentive to merit. That can’t possibly be right.
Yes, I think that’s roughly correct. The issue, of course, is that perfectly efficient markets are an abstract concept where the rate of “excess” profit (that is, over and above the risk adjusted profit you are “due”) are driven to zero. The point is that if that logic applies and if AI makes markets more and more efficient (ie, driving alpha to zero) then the outcomes will become perfectly random. Obviously an idealization, but one worth thinking about. That’s why you get new-style economic models that say that what businesses actually try to do is to capture “temporary” monopoly profits while they are “ahead” of their competition, and the job of a businessman is to arrange that this keeps happening.
Two thoughts:
Is this fundamentally any different from the claim that with perfect competition profits are driven to zero? This is a similar sort of paradox: the profit incentive drives production, and yet when the market is “perfect,” this incentive no longer exists.
Though I haven’t looked at the details, I get the feeling that this kind of analysis is based on a scaling problem. Mathematically, if you want to see a macroscopic structure emerge asymptotically, you generally have to choose your scaling parameters correctly. For example, Brownian motion is only the scaling limit of a random walk when dt ~ (dx)². Surely you have to do something similar when talking about “perfectly efficient markets.” It’s very easy for people to get an extremely wrong idea here, thinking that imposing an artificial friction on the market must be a good thing because, hey, the perfectly efficient case erases the whole incentive to merit. That can’t possibly be right.
Yes, I think that’s roughly correct. The issue, of course, is that perfectly efficient markets are an abstract concept where the rate of “excess” profit (that is, over and above the risk adjusted profit you are “due”) are driven to zero. The point is that if that logic applies and if AI makes markets more and more efficient (ie, driving alpha to zero) then the outcomes will become perfectly random. Obviously an idealization, but one worth thinking about. That’s why you get new-style economic models that say that what businesses actually try to do is to capture “temporary” monopoly profits while they are “ahead” of their competition, and the job of a businessman is to arrange that this keeps happening.