|
There are power laws-- or things that look like power laws-- in financial data. I did a quick analysis of the NASDAQ this morning with Claude Pro. (Amazing what you can do these days with LLMs.) If you look at the histogram of stock selloffs, it looks like the distribution is either power law or lognormal. Nobody seems to know why. For the sake of argument, let's go with power law, as that's what people have seemingly found. This was only five years of data, which may be why an analysis was inconclusive.
So why are selloffs power law, and what can we do about it? We should be scared that these selloffs are power law, meaning that a selloff of size x has a frequency that is power law in x. That means that a huge black swan selloff that crashes the economy is far more likely than it would be under another model. I have a proposal rooted in opinion dynamics research. We already know that opinion dynamics models can exhibit dynamical richness, including maybe transient chaos from a paper I wrote with Emily Dong. I'm sure there's more to come on this front. Some of the graphs I've seen in various opinion dynamics papers are suggestive of something more complicated than a fixed point, and the more cognitive biases you add to a Bayesian model, the more chaotic things should be. Chaos is kind of generic in complicated enough models, so why not for modified Bayesian-like models? Not that this selloff behavior is so much chaotic as it is suggestive of self-organized criticality. Given how complex systems with simple dynamical rules with nonlinearities can be, It may be possible that opinion dynamics models can explain the power law selloff distributions if you put in the right cognitive biases. And if that works to explain the power law selloff distributions, there is actually a constructive route to stabilizing the economy if you can just attack the cognitive bias that's causing the wreckage. You have to be careful, because the wrong way of attacking the cognitive bias can lead to even more instability if you're not careful. There are examples of interventions having backwards effects in complex systems everywhere because it's so hard to predict what's going to happen, as the system is literally too complex. Network-level effects can turn a well-meaning intervention on the psychology of financial traders into chaos, quite literally. But at least we'd know what to attack. Alas, the problem I foresee with this line of attack is that we really don't have a sense of what cognitive model to use. Bayesianism is well-formulated, but how do you model cognitive biases? Nobody knows. If someone says that they know, they may have a proposal that has been tested on a few data sets, but that's not enough to say the problem is solved. But maybe Bayesianism with cognitive biases and reasonable models for behavior based on beliefs can explain these power laws in selloffs.
0 Comments
|
AuthorWrite something about yourself. No need to be fancy, just an overview. Archives
November 2026
Categories |