‘Making’, ‘taking’ and the material political economy of algorithmic trading

Abstract

Drawing upon interviews with 72 practitioners of automated, ultrafast high-frequency trading (HFT), this paper identifies the most salient divide within HFT: between algorithms, trading groups and firms that specialize in 'making' (in adding bids to buy and offers to sell to exchanges' electronic order books) and those that specialize in 'taking' (in executing against existing bids and offers in those order books). The paper explores how 'making' and 'taking' algorithms interact, emphasizing the materiality of that interaction (including, e.g. the surprising effect on it of rain), and suggesting the importance of the 'material political economy' of algorithmic trading: the shaping of that trading, in economically consequential ways, by material orderings that could be different.

The virtue of simplicity: On machine learning models in algorithmic trading

“…The use of rule-based high-frequency trading (HFT) algorithms has been thoroughly studied in the social studies of finance (SSF). Studies cover topics such as the material political economy of HFT (MacKenzie, 2017, 2018a, 2018b, regulatory challenges associated with trading at speeds that exceed human perception (Coombs, 2016;Lenglet, 2011), interactions of algorithms (MacKenzie, 2019), imitation and herding behaviour (Borch, 2016;Lange, 2016), epistemic regimes (Seyfert, 2016), HF trader subjectivities (Borch and Lange, 2017), and market rhythms (Borch et al, 2015). Whilst rule-based trading algorithms have been studied extensively in SSF, little attention has been paid to non-rule-based, adaptive machine learning models and the people who develop and use them for trading and investment management.…”

Infrastructural power: discretion and the dynamics of infrastructure in action

“…As a result, their construction is often marked by contestation among multiple stakeholders. Technical components may be contested by engineers holding contrasting goals for a system (Pardo-Guerra 2019, 2015 or users seeking a material advantage over competitors (MacKenzie 2012, MacKenzie 2018b; managers and workers clash over infrastructures with an eye toward their impact on work routines and cultures Star 2000, Muniesa 2007); legislators and regulatory agencies fight to define infrastructures and their legitimate forms in ways that appease vested interests or broaden their own ambit (Millo 2007, MacKenzie 2018a. In short, market infrastructures are not neutral, efficient outcomes, but contested reflections of the perspectives and desires of those with the power to shape them.…”

Machine learning and social theory: Collective machine behaviour in algorithmic trading

“…While appreciating that applying the sociological face-toface approach par excellence to non-human systems might seem surprising, MacKenzie stresses that a similar analytical path has been trodden by scholars such as Knorr Cetina and Preda, both of whom productively extended Goffmanian insights to the study of (non-automated) electronic markets (Knorr Cetina, 2009;Knorr Cetina & Bruegger, 2002;Preda, 2009Preda, , 2017. Building on this research, MacKenzie argues that, in the context of algorithmic trading, a distinct Goffmanian interaction order may, for example, be identified when automated trading algorithms pursue particular forms of queuing practices (focusing on how they place orders in the electronic order book) as well as when they engage in dissimulation -such as when an algorithm seeks to manipulate markets by placing orders that give the impression of a larger market movement and then take advantage of how other market participants might be fooled by this strategy (see also MacKenzie, 2018a). For example, a manipulative algorithmic strategy seeking to sell an accumulated quantity of shares to the best possible price might consist in placing a large number of orders to buy that particular stock, hoping to ignite other algorithms to increase their sell price and then sell at the better price while quickly cancelling its orders to buy.…”

“…By contrast, automated markets are anonymous as a rule, meaning that no market participant who enters an order in today's electronic markets knows for sure who their counterparties are. Algorithms might detect that other particular algorithms follow specific patterns but which firms are behind these algorithms is generally not a piece of information they have access to (MacKenzie, 2018a). It follows that, although particular algorithms might be interacting with one another repeatedly, their lack of knowledge about counterparty identity renders Granovetterian social embeddedness impossible among them.…”