Cameron's Blog - Selection mechanism design

Selection mechanism design

Dewhurst, David Rushing, Michael Vincent Arnold, and Colin Michael Van Oort, 2018, Selection mechanism design affects volatility in a market of evolving zero-intelligence agents, arXiv:1812.05657 [cs, q-fin].

Here’s an interesting paper from Dewherst, Arnold, and Van Oort published to arXiv. Dewherst et al. use an evolving multi-agent landscape to model evolutionary selection mechanisms applicable to markets.

Lo, Andrew W, 2004, The Adaptive Markets Hypothesis, Journal of Portfolio Management, 15.

Choosing a representative selection mechanism is somewhat important if you follow Andrew Lo’s Adaptive Markets Hypothesis, where financial agents are weeded out in response to changing market conditions. Deherst et al. examine three selection mechanisms.

The authors use frequent batch auctions instread of continuous dual auctions. I’m a big fan of frequent batch auctions, and you can read more about them in the Budish, Cramton and Shim (2015) paper.

The first is a global quantile-based measure, where the bottom 10% of all agents are eliminated when a selection event occurs. The second is a localized variant, where agents in a subsample are kept in the environment according to the probability


where $π_i$ is the fitness of agent $i$. This can be understood as awarding higher retention probabilities to agents with a high share of the sampled agent’s fitness. The final metric uses the first metric with probability ½ and the second metric with probability ½.

The agents have risk aversion, and are allowed to “innovate” by drawing their parameters from distributions unaffected by existing agents. If you don’t innovate, you draw your parameters from a distribution represented by agents who were not removed.

The results show that the quantile metric creates agents with higher average profitability, but that the localized variant has a higher correlation between micro and macro volatility.

I’m not quite sold at the end of the paper as to their selection mechanism, but I think it’s an interesting vein of research.