Blog: Market Stability, AI, and Central Banks
I was considering the role that Central Banks play in the context of market stability and market efficiency. It is indeed true that Central Banks can help with both, for example, by matching tweaking interest rates in a way that matches the expectations of all market participants and thus reducing uncertainty.
However, it is also true that Central Banks could also hurt stability long-term too since very cheap interest rates and a plentiful supply of money can encourage risk taking. For example, if it costs you 0.01% annually to borrow, you’d want to borrow as much as you can even if you can get a low-ish return of 1% on it. But if it costs you 20% to borrow , you’re going to get eaten alive by the interest rate unless you have a very good reason to put that capital into use.
Anyways, there’s an entire field of study here with many complexities and nuances, each worth of their own thread.
What I was thinking about was the idea of trading strategies that are based on black box algorithms. Usually there’s an economic rationale (expectation of interest rate increase, over- or under-valuation of a stock, and so on) for various trading strategies, but what’s wild is that many of these quant hedge funds are starting to delve into AI and thereby relinquish control of many aspects of the model building that intakes and generates trading signals.
What’s crazy here is that the signals used for trading may actually have absolutely no economic or financial rationale. It’s an AI model built on previous data points in the market to build a model that can be used to generate other trading signals going forward from which to trade from. This is distinct from current black box algorithms where there are economic/financial rationales, just that there’s super complex algorithms acting using them as data points.
I read that a senior manager at the legendary quant hedge fund Renaissance Technologies (in the book Machine Trading by Ernest Chan) said that some of their strategies actually have no economic or financial rationale, and that’s why they cannot be arbitraged away.
This is extremely significant.
Contrast with the current world of trading, even fully black box and high frequency algorithmic trading. If you believe the stock of some company is overvalued, then you’d want to sell it. Likewise, if you believe it’s undervalued, then you’d want to buy it. So there’ll be market participants who will buy undervalued or short overvalued stock until the price goes to a more acceptable level. Now there’s a meta-optimization and arbitraging away of inefficiencies since you can do this in different markets involving different currencies if the stock is being traded on Tokyo (Yen), London (Pound), and NYC (Dollars).
All of these arbitrage opportunities may be complex, but are based fundamentally on economic/financial rationales that all market participants can identify. Furthermore, how to get an edge is straightforward even if you don’t have the means, for example, at faster microsecond scales or with higher capital scales.
The thing with AI-based trading strategies is that because they don’t have economic/financial rationales, no other market participants know what or how they’re optimizing. These other market participants include Central Bank traders and those related to them like the unelected Central Bank officials.
So the upshot here is that because these rationales are unknown, and even fundamentally unknowable, the role Central Banks play may neither be helpful nor hurtful, but uncertain.
What are the implications for market efficiency and market stability when there’s tens or hundreds of billions of dollars of capital flowing through the system driven by AI algorithms?