Algorithm Switching: Co-Adaptation in the Market Ecology

Abstract

Algorithm design for trading has concentrated on representing particular trading strategies that are familiar to, or at least understandable by, traders. A fundamental characteristic of markets however, is their evolvability, with traders continually adapting their strategies to current market conditions. In this sense, a market can be fruitfully viewed as an “ecology” of trading strategies. If trading algorithms are to be truly “intelligent” they must capture this capability of human traders to adapt, and “switch” strategy, in response to changing market conditions.

In this article, we first discuss why it is important to be able to switch algorithms. We then outline a framework within which such switching can be understood. This entails having a taxonomy for different trading strategies, and an understanding of how the transferal of information due to order flow between the market and a given strategy impacts the performance of an algorithm, both in terms of participation rate and market impact. We discuss the predictability of this information transferal showing how this can be used to build an “algorithmic switching engine” that chooses the algorithm most adapted to the current market niche. We discuss some of the challenges of constructing such a switching engine and analyze some results from a particular realization.