The future of algorithmic competition: a network-based perspective
Lead Research Organisation:
University of Oxford
Department Name: Mathematical Institute
Abstract
Digital platforms increasingly dominate the digital and physical economy. Their rise and subsequent impact on markets is being scrutinized by politicians and regulators, concerned about the use of algorithms on these platforms, on how this might change the nature of competition online. The potential of algorithms to facilitate collusion has been emphasized in competition law, in economics, and by regulators. Despite the relevance of the topic to society, there is a lack of mathematical research on how to prevent algorithms from hindering competition at the expense of consumers.
This project initiates a broad interdisciplinary research approach at the intersection of network science, economics, and machine learning. We plan to model the behavior of algorithms in a market using tools from network science and graph theory, in order to determine the conditions that lead to strong competition with non-collusive algorithms. We hope to better understand and ultimately address algorithmic concerns online, to foster testing procedures and standards for algorithms in the digital economy.
This project initiates a broad interdisciplinary research approach at the intersection of network science, economics, and machine learning. We plan to model the behavior of algorithms in a market using tools from network science and graph theory, in order to determine the conditions that lead to strong competition with non-collusive algorithms. We hope to better understand and ultimately address algorithmic concerns online, to foster testing procedures and standards for algorithms in the digital economy.
Organisations
Publications
Meenatchi Sundaram Muthu Selva Annamalai
(2023)
A Linear Reconstruction Approach for Attribute Inference Attacks against Synthetic Data
in arXiv
Rocher L
(2023)
Adversarial competition and collusion in algorithmic markets
in Nature Machine Intelligence
| Description | This project piloted the use of network science methods to model the dynamics of pricing algorithms in online markets. We showed that algorithmic strategies can be fully represented using graphs, and dynamics learned by setting and observing prices. This allows adversaries to learn the strategies of other pricing algorithms and manipulate them. We developed a new mathematical approach to adversarial attacks and tested it empirically in 2 and 3-firm environments. The key objective, predicting the likelihood for one algorithm to unilaterally take advantage of its competitors, has been fully met. We showed theoretically and empirically how to develop adversarial attacks against weaker algorithms in online markets. Our method is however limited to 2 and 3-firm environments. Capturing the dynamics of markets with many firms is likely too computationally intensive, requiring approximation methods or machine learning-based methods. |
| Exploitation Route | We expect that the findings will be primarily of interest for economists to understand the impact of new forms of algorithmic collusion and network scientists. We believe insights will be of interest to policymakers and regulatory bodies. |
| Sectors | Creative Economy,Financial Services, and Management Consultancy |
| Description | Despite the short length of the project, positive research progresses have resulted in: (a) a successful new network science approach to model algorithmic markets, that was presented at seminars in Oxford, (b) a working paper jointly written with colleagues at Imperial College London, currently in peer review under revision (c) a software toolbox to simulate adversarial attacks in algorithmic markets, that we plan to release in open source on Github upon publication. |
| First Year Of Impact | 2023 |
| Sector | Creative Economy,Financial Services, and Management Consultancy |
| Description | Partitipation in expert workshop and discussion with regulators |
| Geographic Reach | National |
| Policy Influence Type | Contribution to a national consultation/review |
| Description | A talk or presentation - Network science approaches to model algorithmic markets |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Postgraduate students |
| Results and Impact | 10-15 researchers attended a local seminar presentation, where early research findings were presented and discussed. Suggestions were made to improve the mathematical models and translate research to other fields. |
| Year(s) Of Engagement Activity | 2022 |
| Description | Presentation of Non Markovian paths and cycles in NFT trades |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Industry/Business |
| Results and Impact | Our work "Non Markovian paths and cycles in NFT trades" has been presented at the Complex Data Blocks satellite of #CCS2022. |
| Year(s) Of Engagement Activity | 2022 |
| URL | https://www.raphtory.com/papers/Complex_Data_Blocks.pdf |