ERC research project on Algorithmic Finance (AlgoFinance)
'Algorithmic Finance: Inquiring into the Reshaping of Financial Markets' is a four-year research project funded by the European Research Council (ERC) in the form of a Consolidator Grant. The project commenced on 1 May 2017. Project website: http://info.cbs.dk/algofinance.
Brief project synopsis: Present-day financial markets are turning algorithmic, as orders are increasingly being executed by fully automated computer algorithms, without any direct human intervention. Although algorithmic finance seems to fundamentally reshape the central dynamics in financial markets, and even though it prompts core sociological questions, it has not yet received any systematic attention. Against this background, and contributing to economic sociology and social studies of finance, AlgoFinance aims to understand how and with what consequences the turn to algorithms is changing financial markets. The overall concept and central contributions of AlgoFinance are the following: (1) on an intra-firm level, the project examines how the shift to algorithmic finance reshapes the ways in which trading firms operate, and does so by systematically and empirically investigating the reconfiguration of organizational structures and employee subjectivity; (2) on an inter-algorithmic level, it offers a novel methodology (agent-based modelling informed by qualitative data) to grasp how trading algorithms interact with one another in a fully digital space; and (3) on the level of market sociality, it proposes a new theorization of how intra-firm and inter-algorithmic dynamics can be conceived of as introducing a particular form of sociality that is characteristic to algorithmic finance: a form of sociality-as-association heuristically analyzed as imitation. None of these three levels have received systematic attention in the state-of-the-art literature. Addressing them will significantly advance the understanding of present-day algorithmic finance in economic sociology. By contributing novel empirical, methodological, and theoretical understandings of the functioning and consequences of algorithms, AlgoFinance will further open up new opportunities for future research into digital sociology and the algorithmization of society more broadly.
Objectives and research questions
During the past 10 years, financial markets have undergone significant transformations. Up until late in the twentieth century, the ‘machine room’ of financial markets was characterized by so-called open-outcry trading, where traders were physically co-present and would execute their orders by pushing and yelling. In the 1990s, open-outcry trading was gradually – sometimes rapidly – replaced by electronic trading. Working in large trading rooms, e.g. in investment banks, traders would now engage with the market via their computer screens and execute orders by clicking their mouse (Knorr Cetina and Bruegger, 2002; Zaloom, 2006). Referring to this technological set-up, traders operating in such environments are often labelled ‘click’ or ‘screen’ traders. In the past decade, however, the shift toward electronic trading has further intensified. Today, the vast majority of orders in financial markets are executed by neither pit traders nor click/screen traders, but by fully automated computer algorithms that are programmed to carry out trades, often in fractions of a second. While such algorithms are of course programmed by humans, they operate in digital, globally connected markets without any direct human intervention. To illustrate the significance of algorithms: in 2000, the US securities market witnessed ‘on average about 5 million trades and quotes per day; in the fall of 2012, at peak times there were up to 5 million trades and quotes per second’ (Malinova, Park, and Riordan, 2013: 1, italics in the original). This colossal growth in trading is due to algorithms being placed at the center of financial markets (as actual trading agents), rather than, e.g. serving as a tool for manual click/screen traders.
The academic debates on the algorithmization of financial markets are characterized by ambiguity. Some, especially financial economists (e.g. Hendershott, Jones, and Menkveld, 2011), welcome algorithmic trading as a more efficient and systematic type of trading that bids farewell to the alleged (emotional, greedy, etc.) biases of human traders. Others argue that the current centrality of fully automated algorithmic trading is likely to give rise to new types of devastating market crashes. Indeed, events such as the May 2010 so-called Flash Crash, in which inter-algorithmic dynamics contributed to a dramatic and rapid drop in the US markets, are widely seen as emblematic of how the algorithmization of financial markets may lead to unanticipated forms of negative cascading (e.g. Borch, 2016; Donefer, 2010; Sornette and von der Becke, 2011).
This academic ambiguity not only reflects the complexity of present-day financial markets, but is also due to a lack of systematic analysis of algorithmic finance, both within and beyond sociology, where only fragments of the shift have been studied. Indeed, and surprisingly, although algorithms are at the center of present-day financial markets, prompting a series of core sociological questions, there is still only sparse sociological interest in this algorithmic change to the financial markets, and so far no systematic academic attention has been paid to how the rise of algorithmic finance reshapes key dimensions of financial markets.
AlgoFinance seeks to fill this major gap in the scholarly literature, and thereby open up new opportunities for research. The objective of AlgoFinanceis to understand how and with what consequences algorithmic finance changes financial markets. In particular, AlgoFinance argues that algorithmic finance gives rise to profound reconfigurations of (a) how trading firms operate, as both organizational structures and employee subjectivity are reshaped to accommodate the new algorithmic reality; (b) the interactions of trading agents, as these are increasingly constituted by fully automated computer algorithms that interrelate with one another in a purely digital space; and (c) the forms of sociality characteristic of present-day financial markets. Reflecting these dimensions the project seeks to answer the following key research questions (RQ):
RQ 1: How and with what consequences does the use of algorithms change the ways in which trading firms operate in financial markets?
RQ 2: How and with what consequences do automated trading algorithms interact in financial markets?
RQ 3: What types of sociality characterize algorithmic finance?
AlgoFinance consists of three closely integrated work packages (WPs A, B, and C), each devoted to its respective RQ. The WPs are closely integrated in that WP A contributes data to WP B, and that WP C examines the sociality of algorithmic finance on the basis of WPs A and B.
Work package A: Intra-Firm Implications of Algorithmic Finance: This work package examines how and with what consequences the use of algorithms changes the ways in which trading firms operate in financial markets (RQ 1). This will entail attending to three dimensions: (a) the variety of algorithms being deployed by trading firms; (b) the organizational dynamics of algorithmic trading firms; and (c) the types of subjectivity that gain prominence with algorithmic trading.
The first layer will map the different types of algorithms that are being deployed for trading purposes by present-day financial firms across different markets. This mapping, which will provide an overview of the full spectrum of contemporary algorithmic trading as seen from the inside, i.e. by market professionals, will offer a highly valuable foundation for future research on this topic, across sociology and economics. It will be developed by examining the following key questions: What types of algorithms are being deployed (from high-frequency trading to advanced machine learning)? What is their purpose? What kinds of assumptions about markets go into the algorithms? What types of strategies do they purse? What types of data do they exploit? At what types of market(s) (equities, futures, etc.) are they aimed? And what speed levels do they rely upon or seek to benefit from?
The next layer focuses on the changes in the organizational dynamics that are triggered by the turn to algorithms. The work package conjectures that the shift to algorithmic finance leads to a series of important changes in the ways in which trading firms organize themselves. Preliminary findings from the high-frequency trading domain thus suggest (a) that algorithmic finance is associated with new types of collaboration between new types of staff, something that reconfigures established ways of organizing trading firms (e.g. by dismantling so-called ‘front office’ and ‘back office’ functions); and (b) that organizational design (hierarchical structures, information flows, etc.) is structured in ways that both reflect and co-shape the algorithmic trading strategies. The work package examines these and other organizational dynamics in algorithmic trading firms and demonstrates changes by comparing the findings to extant work on pre-algorithmic trading firms.
Finally, the third layer zeroes in on the subjectivity of the individual market participants who embody the shift to algorithmic finance. Examining subjectivity in algorithmic finance might seem counter-intuitive – after all, algorithmic finance is often lauded as a way of excising humans and their alleged biases from the trading process. However, AlgoFinance nonetheless argues that subjectivity remains of core importance for markets that turn algorithmic, because they are likely to fabricate their own forms of subjectivity that differ markedly from those of non-algorithmic markets (Borch and Lange, 2016). Previous research has demonstrated that pre-algorithmic types of financial markets indeed produced particular forms of trader identity and subjectivity (Abolafia, 1996; Knorr Cetina and Bruegger, 2002; Zaloom, 2006). By examining how algorithmic finance generates its own particular types of subjectivity – across the spectrum of algorithmic finance, WP A focuses on the idealized and normatively sanctioned skills, expertise, behaviors, concerns, and self-conduct characteristic of the market professionals (traders, software developers, etc.) who constitute the core of algorithmic finance, as well as how these individuals conceive of and relate to markets.
Work package B: Inter-Algorithmic Dynamics: This work package focuses on how and with what consequences automated trading algorithms interact in financial markets (RQ 2). Since, with algorithmic finance, the main trading activity is one that takes place between fully automated algorithms, as mediated via electronic order books, present-day financial markets can only be fully understood if this inter-algorithmic activity is comprehended. The interaction amongst algorithms is what differentiates algorithmic finance vis-à-vis previous financial market configurations, where the trading interaction was between, e.g. human traders or between human traders as mediated via their computers. Unfortunately, common methods used in economic sociology and social studies of finance, namely interviews with and ethnographic observations of traders, are inapt at accounting for this level of inter-algorithmic trading dynamics. The problem is, as MacKenzie (2015a: 19) has forcefully stressed, that traders may not be fully aware of the interaction effects of their algorithms. Therefore, WP B suggests an alternative and innovative methodological approach, namely agent-based modelling (ABM). The aim of ABM is to understand the interaction of agents (variously defined) in particular contexts. Such modelling can shed light upon emergent interaction dynamics, critical phenomena (crashes), and spontaneous pattern formations. In recent years, a growing number of scholars have successfully deployed ABM in sociological analysis (e.g. Cederman, 2005; Epstein, 2007; Gilbert and Abbott, 2005; Macy and Willer, 2002; Sawyer, 2003; Squazzoni, 2012). ABM is deployed in WP B for two purposes. The first is explorative: to understand the emergent interaction dynamics and spontaneous pattern formations that might arise from the interaction of trading algorithms. However, sociological applications of ABM are rarely explorative in nature (Squazzoni, 2012). The second purpose is therefore more in line with ABM’s standard sociological functions – reflecting a ‘generative approach’ (Epstein, 2007), which seeks to explain some macro phenomenon or pattern by building a model from the bottom up, WP B thus ABM to analyze the potential crash-like consequences of inter-algorithmic dynamics.
Work package C: The Sociality of Algorithmic Finance: This work package focuses on the types of sociality that characterize algorithmic finance (RQ 3). This will be examined by teasing out the dominant forms of sociality-as-association in algorithmic finance that can be identified on the basis of WPs A and B. Seeking to understand the sociality of financial markets continues, but also goes beyond, a long sociological tradition, as previous research has demonstrated that different market configurations give rise to particular forms of sociality (Abolafia, 1996; 1984; Knorr Cetina and Bruegger, 2002; MacKenzie, 2004). WP C argues for taking this sociological endeavor a step further and asks – even if the question appears counter-intuitive – whether algorithmic finance, despite its partial bracketing of humans, generates its own particular form of sociality. Similarly, MacKenzie (2015a) has suggested that Goffman’s (1983) ‘interaction order’ theory might help to understand how trading algorithms interact with one another without any direct human intervention. However, since the inter-algorithmic dynamics dispense with the kind of physical co-presence so key to Goffman, WP C argues for an alternative theoretical approach. Thus, WP C conjectures that an entirely new type of sociality might be emerging with algorithmic finance (across intra-firm and inter-algorithmic dynamics): one that revolves around imitation.
Imitation has been afforded some importance in economic sociology (e.g. Arnoldi and Borch, 2007; Beunza and Garud, 2007; Preda, 2009b; White, 1988), organization studies (DiMaggio and Powell, 1991), and financial economics (Orléan, 1989; Scharfstein and Stein, 1990; Shiller, 1984). However with few exceptions (Czarniawska, 2004; MacKenzie, 2003; 2004), imitation has been attributed a marginal role. AlgoFinance heuristically suggests that imitation is not a residual phenomenon in present-day financial markets, rather that it is fundamental to understanding the forms of sociality that characterize algorithmic finance. The emphasis on the increasingly central role of imitation is supported by preliminary findings from the high-frequency trading domain (Borch, 2016; Lange, 2015).
When analyzing imitation as a particular form of sociality, WP C mobilizes the work of the French sociologist Gabriel Tarde. According to Tarde, the social is constituted by imitation: when one person imitates another (or that person’s gestures, beliefs, etc.), an association is created whereby he or she implicitly pays respect to the person being imitated (Tarde, 1962). Since Tarde’s conception of imitative sociality is not tied to a constitutive subject or self, it is particularly well-suited to analyzing how imitative sociality may be prevalent both in settings in which humans play some role (as in WP A) and in settings where inter-human action is replaced by inter-algorithmic action (as in WP B). Further, Tarde’s conceptual apparatus invites analysis of how imitative phenomena might display intense forms of sociality, but may also lead to social collapse (Borch, 2012), something that will be important in discussions of algorithmic market crashes.
For questions relating to the AlgoFinance project, please contact Professor Christian Borch (email@example.com).
Last updated by: Christian Borch 23/11/2017