Markets are becoming less efficient, not more

(Published by JP Bouchaud | January 2024)

In my last post, I have argued that faced with complex problems, rational agents cannot compute the optimal solution nor learn the optimal strategy, but go for "satisficing" solutions. But maybe when averaging over a large crowd of agents, individual errors average out and markets are efficient, as if we were rational?

But individual errors are in fact correlated, not only because we humans share the same behavioural biases, but more importantly because we interact strongly with one another, both through social contacts (physical or virtual), and through price changes themselves, which we interpret as information even when they carry none (like past trends).

At a deeper level, this is related to price impact -- see my previous posts and this paper, but this is another story.

Now, we know that beyond a certain level of interaction, collective behaviour totally decouples from the one that would prevail if individuals acted independently. This is arguably why history is strewn with mass panics, fads and trends, bubbles and crashes, booms and busts, crises and upheavals of all sorts. In a large fraction of cases, it is hard to fathom why these events occur: there is no exogenous shocks that would explain their severity. But interactions and feedback loops can easily explain why anecdotal news may self-amplify into full blown crises (see for example this paper).

Missing such ubiquitous - in space and across time - social effects is one of the main flaw of classical models, DSGE, efficient markets and the likes. Quite frankly, these models are blind to what makes us human, for the better and for the worse, i.e. social animals.

This is probably even more relevant with modern social media, and perhaps why Cliff Asness argues that financial markets are less efficient than 30 years ago, cf. this.