Periodic Reporting for period 3 - AlgoFinance (Algorithmic Finance: Inquiring into the Reshaping of Financial Markets)
Reporting period: 2020-05-01 to 2021-10-31
Present-day financial markets are dominated by fully automated computer algorithms which can send orders to buy or sell securities at a fraction of a second. These algorithms are programmed by human beings, but once they are active in markets, they operate independently of humans. The objective of the AlgoFinance research project is to understand the consequences of such algorithmic trading. More specifically, the project pursued three aims. First, it analyzed how and with what consequences the use of algorithms changes the ways in which trading firms operate in financial markets. Among other things, the project has detailed the complexities involved in preventing that complex algorithmic systems engage in unwanted behaviors – including triggering financial crashes – and how the rise of opaque machine-learning-based systems enhance such complexities. Second, it analyzed how and with what consequences automated trading algorithms interact in financial markets. To this end, the project has simulated the interaction of trading algorithms which generated insights into the conditions under which algorithm-dominated markets become fragile. Third, the project studied the forms of “sociality” that might characterize algorithmic finance. For example, the project has demonstrated the advantages and limitations of applying notions originally developed to make sense of interhuman behaviour to inter-algorithmic behaviours. The project’s research and findings are important for society because algorithmic trading is at the center of present-day financial markets. Research on the use and consequences of algorithmic trading is crucially important if we want to design markets in which algorithmic crashes are less frequent. Furthermore, insights into algorithmic trading are important when seeking to understand other domains where automation is assuming an increasingly important role.
The AlgoFinance research project sought answers to three questions: (1) How and with what consequences does the use of algorithms change the ways in which trading firms operate in financial markets?; (2) How and with what consequences do automated trading algorithms interact in financial markets?; and (3) What forms of sociality characterize algorithmic finance?
To answer the first question, we conducted almost 200 interviews with market participants active in algorithmic trading. The majority of the informants were working in firms in Chicago, New York, London or Amsterdam, all central hubs for this form of trading. In addition to trading firms, we interviewed exchange officials, regulators, and non-algorithmic firms such as institutional investors who might be adversely affected by algorithmic trading. In addition to interviews, we did observational studies at two trading firms in Chicago and London, respectively. The interviews and the fieldwork gave us unique insights into the world of algorithmic finance. Specifically, we demonstrated that new types of expertise are getting traction in financial markets; that the advent of opaque machine-learning models has produced a rift among algorithmic traders, with some enthusiastically endorsing such models and with others being skeptical about their non-intuitive nature; and that the turn of machine-learning-based algorithmic trading produces new types of challenges for trading firms and therefore demand new forms of internal organization to meet these challenges.
We published 10 journal articles on this, and an additional five papers are currently in process. Furthermore, one full-length single-authored book manuscript is currently under consideration; another one is invited at proposal stage; and a further single-authored monograph is in progress.
To answer the second research question, we developed a semi-realistic simulation framework capable of simulating the interaction of thousands of algorithms with realistic properties. This simulation framework is the most complex and realistic one existing in the academic world. We consider having developed this simulation framework a core achievement of the project.
We are currently completing one journal paper based on the simulation work and have three more in the pipeline.
In answering the third research question, the project found that the interaction among fully automated trading algorithms can be understood through a comparison with the fleeting encounters among urban inhabitants. However, the project also demonstrated that, at least for machine-learning-based algorithmic trading systems, these can only be understood to a limited degree on the basis of sociological notions developed to understand inter-human behaviour. Examining this third research question led to one single-authored monograph and two journal articles. Three additional papers are in review/progress.
To answer the first question, we conducted almost 200 interviews with market participants active in algorithmic trading. The majority of the informants were working in firms in Chicago, New York, London or Amsterdam, all central hubs for this form of trading. In addition to trading firms, we interviewed exchange officials, regulators, and non-algorithmic firms such as institutional investors who might be adversely affected by algorithmic trading. In addition to interviews, we did observational studies at two trading firms in Chicago and London, respectively. The interviews and the fieldwork gave us unique insights into the world of algorithmic finance. Specifically, we demonstrated that new types of expertise are getting traction in financial markets; that the advent of opaque machine-learning models has produced a rift among algorithmic traders, with some enthusiastically endorsing such models and with others being skeptical about their non-intuitive nature; and that the turn of machine-learning-based algorithmic trading produces new types of challenges for trading firms and therefore demand new forms of internal organization to meet these challenges.
We published 10 journal articles on this, and an additional five papers are currently in process. Furthermore, one full-length single-authored book manuscript is currently under consideration; another one is invited at proposal stage; and a further single-authored monograph is in progress.
To answer the second research question, we developed a semi-realistic simulation framework capable of simulating the interaction of thousands of algorithms with realistic properties. This simulation framework is the most complex and realistic one existing in the academic world. We consider having developed this simulation framework a core achievement of the project.
We are currently completing one journal paper based on the simulation work and have three more in the pipeline.
In answering the third research question, the project found that the interaction among fully automated trading algorithms can be understood through a comparison with the fleeting encounters among urban inhabitants. However, the project also demonstrated that, at least for machine-learning-based algorithmic trading systems, these can only be understood to a limited degree on the basis of sociological notions developed to understand inter-human behaviour. Examining this third research question led to one single-authored monograph and two journal articles. Three additional papers are in review/progress.
The project has produced a simulation framework that can simulate the interaction among fully automated algorithms in financial markets in a way that goes significantly beyond the state of the art.
The project has generated insights into machine-learning-based automated trading that go significantly beyond the state of the art.
The project has generated insights into machine-learning-based automated trading that go significantly beyond the state of the art.