Algorithmic trading, market quality and information : a dual -process account
Gamzo, Rafael Alon
One of the primary challenges encountered when conducting theoretical research on the subject of algorithmic trading is the wide array of strategies employed by practitioners. Current theoretical models treat algorithmic traders as a homogenous trader group, resulting in a gap between theoretical discourse and empirical evidence on algorithmic trading practices. In order to address this, the current study introduces an organisational framework from which to conceptualise and synthesise the vast amount of algorithmic trading strategies. More precisely, using the principles of contemporary cognitive science, it is argued that the dual process paradigm - the most prevalent contemporary interpretation of the nature and function of human decision making - lends itself well to a novel taxonomy of algorithmic trading. This taxonomy serves primarily as a heuristic to inform a theoretical market microstructure model of algorithmic trading. Accordingly, this thesis presents the first unified, all-inclusive theoretical model of algorithmic trading; the overall aim of which is to determine the evolving nature of financial market quality as a consequence of this practice. In accordance with the literature on both cognitive science and algorithmic trading, this thesis espouses that there exists two distinct types of algorithmic trader; one (System 1) having fast processing characteristics, and the other (System 2) having slower, more analytic or reflective processing characteristics. Concomitantly, the current microstructure literature suggests that a trader can be superiorly informed as a result of either (1) their superior speed in accessing or exploiting information, or (2) their superior ability to more accurately forecast future variables. To date, microstructure models focus on either one aspect but not both. This common modelling assumption is also evident in theoretical models of algorithmic trading. Theoretical papers on the topic have coalesced around the idea that algorithmic traders possess a comparative advantage relative to their human counterparts. However, the literature is yet to reach consensus as to what this advantage entails, nor its subsequent effects on financial market quality. Notably, the key assumptions underlying the dual-process taxonomy of algorithmic trading suggest that two distinct informational advantages underlie algorithmic trading. The possibility then follows that System 1 algorithmic traders possess an inherent speed advantage and System 2 algorithmic traders, an inherent accuracy advantage. Inevitably, the various strategies associated with algorithmic trading correspond to their own respective system, and by implication, informational advantage. A model that incorporates both types of informational advantage is a challenging problem in the context of a microstructure model of trade. Models typically eschew this issue entirely by restricting themselves to the analysis of one type of information variable in isolation. This is done solely for the sake of tractability and simplicity (models can in theory include both variables). Thus, including both types of private information within a single microstructure model serves to enhance the novel contribution of this work. To prepare for the final theoretical model of this thesis, the present study will first conjecture and verify a benchmark model with only one type/system of algorithmic trader. More formally, iv a System 2 algorithmic trader will be introduced into Kyle’s (1985) static Bayesian Nash Equilibrium (BNE) model. The behavioral and informational characteristics of this agent emanate from the key assumptions reflected in the taxonomy. The final dual-process microstructure model, presented in the concluding chapter of this thesis, extends the benchmark model (which builds on Kyle (1985)) by introducing the System 1 algorithmic trader; thereby, incorporating both algorithmic trader systems. As said above: the benchmark model nests the Kyle (1985) model. In a limiting case of the benchmark model, where the System 2 algorithmic trader does not have access to this particular form of private information, the equilibrium reduces to the equilibrium of the static model of Kyle (1985). Likewise, in the final model, when the System 1 algorithmic trader’s information is negligible, the model collapses to the benchmark model. Interestingly, this thesis was able to determine how the strategic interplay between two differentially informed algorithmic traders impact market quality over time. The results indicate that a disparity exists between each distinctive algorithmic trading system and its relative impact on financial market quality. The unique findings of this thesis are addressed in the concluding chapter. Empirical implications of the final model will also be discussed.
Gamzo, Rafael Alon, (2017) Algorithmic trading, market quality and information: a dual-process account, University of the Witwatersrand, Johannesburg, https://hdl.handle.net/10539/26313