Can Data Science Uncover Cartels? A Classroom Game Reveals All
"Explore how a unique economics game uses data science to detect hidden market collusion, offering a practical approach to understanding economic dynamics."
Cartels, secret agreements between businesses to control prices or limit production, have long been a subject of intense scrutiny by economists and policymakers. These illicit arrangements not only distort markets but also harm consumers by artificially inflating prices and reducing choice. As the global economy becomes increasingly data-driven, new techniques leveraging data science are emerging as powerful tools in the fight against cartels.
Traditionally, cartel detection relied heavily on theoretical models and qualitative analysis. However, the rise of machine learning and big data analytics has opened up new possibilities for identifying suspicious patterns and behaviors that might indicate collusion. A recent study introduces a classroom game designed to integrate economics and data science, providing students with a hands-on experience in cartel detection.
This innovative game simulates a procurement market where participants act as firms, making decisions about competitive behavior and potential collusion. By combining economic principles with data analysis techniques, the game offers a practical and engaging way to understand the complexities of market dynamics and the role of data science in uncovering illegal activities.
Cartels: A Persistent Economic Threat
Cartels—agreements between competitors to restrict competition—represent one of the most damaging forms of anti-competitive behavior in modern economies. They inflate prices for consumers, distort public procurement outcomes, and undermine the integrity of markets. While exact global figures on cartel activity remain difficult to pin down due to the covert nature of collusion, enforcement agencies across jurisdictions have reported that cartels remain a pervasive and persistent challenge, costing economies billions annually.
Traditional Detection and Its Constraints
Competition authorities have historically relied on leniency programs, whistleblower tips, and direct investigations to uncover cartel activity. While these methods have proven effective—particularly leniency programs, which incentivize cartel members to self-report—they are inherently reactive, depending on insiders to come forward. Traditional approaches also struggle with cartels that leave minimal documentary evidence, and they may miss collusive arrangements entirely if no participant cooperates with authorities.
The Rise of Data-Driven Cartel Detection
Over the past 30 years, cartel detection has evolved significantly with the introduction of leniency programs, whistleblower rewards, and data screening methods. A newer approach—text screening—has joined the detection toolkit alongside these established methods. At an OECD roundtable on data screening tools held in November 2022, most of the 16 competition authorities that submitted written contributions reported that they already had or aimed to develop cartel screens, with a focus on public procurement markets. Historically, cartel detection research relied on labeled data, where competition authorities had already collected hard evidence about collusion and revealed information on cartel members' strategies. A landmark 2023 study by Silveira et al. shifted this paradigm by proposing an unsupervised machine learning method using unlabeled data, applying statistical screens to flag potential bid-rigging cartels in the Brazilian road maintenance sector.
How the Game Works: An Overview
The classroom game is structured in three distinct parts, each designed to build upon the previous one and provide a comprehensive learning experience:
- Part 1: Competitive Bidding: Firms operate independently, without communication, and must submit bids based on their cost estimates and market knowledge.
- Part 2: Collusion Allowed: Firms are now allowed to communicate and coordinate their bidding strategies, simulating a cartel environment. However, they must be wary of detection.
- Part 3: Cartel Detection: Participants transition to the role of competition authority members, tasked with analyzing bidding data to identify patterns indicative of collusion. They use data science techniques to flag suspicious tenders and advise their board on which cases merit further investigation.
Machine Learning Enters the Cartel Detection Arena
Recent scholarship has brought machine learning into cartel detection research. Silveira et al. (2023) introduced an unsupervised ML approach that applies statistical screens to new datasets for ex-ante screening—anticipating cartel movements before hard evidence is collected. Building on this literature, Wallimann et al. (2024) developed a classroom game in which students analyze bidding data for patterns indicative of collusive behavior, with success depending on their data-science skills. These pedagogical implementations use documented cartel data and machine learning classifiers to reinforce the link between economic theory and data science.
The Limits of Current Detection Methods
Not all cartels are equally detectable. Designing methods capable of uncovering all types of cartels, including incomplete and local arrangements, remains a significant challenge, as these often produce only weak or localized distortions in the distributional pattern of bids. Two main strategies have been proposed to address this problem, but none has proven fully comprehensive. Meanwhile, as enforcement agencies enhance detection through data analytics and whistleblower programs, companies face growing pressure to reassess whether their compliance infrastructure can identify risks before regulators do.
Testing Cartel Screens Across Borders
A 2026 study tested the accuracy of multiple cartel screens using large-scale public procurement data from seven European countries—Bulgaria, France, Hungary, Latvia, Portugal, Spain, and Sweden—spanning 2004 to 2021. The researchers combined this administrative data with judicial records on proven cartels from official government sources, comparing screen performance against confirmed cases. Alternative detection approaches, including proactive measures using market data and behavioral analysis of major companies, have also been explored. In pedagogical settings, machine learning classifiers such as random forest have been used to demonstrate the trade-off between false positives and detection efficiency.
The Future of Cartel Detection
The integration of data science techniques is poised to revolutionize cartel detection, making it more efficient and effective. As technology advances, new methods for analyzing market data and identifying suspicious behaviors will continue to emerge. By equipping students with the skills and knowledge to apply these techniques, educators can play a vital role in shaping the next generation of economists and competition authorities.
A Data-Driven Framework for Flagging Cartels
Research on cartel detection has proposed a data-driven machine learning approach that operates in two stages. First, a clustering algorithm groups tenders based on their attributes; second, the labels created by clustering are used as a target variable for a classifier to predict them. Predictors are then ranked according to their relevance to decrease false positives and improve detection accuracy. This approach, tested on Brazilian road maintenance data, demonstrates how unsupervised and supervised methods can be combined to create a practical screening tool for competition authorities.
AI and Global Cooperation Reshape Enforcement
The detection of cartels is becoming increasingly sophisticated with advancements in technology and growing international cooperation. AI, big data analytics, and digital forensics are emerging as key tools in the global enforcement landscape. As authorities across jurisdictions embrace these new tools, companies face a dynamic and complex environment that requires proactive compliance and strategic adaptation. Emerging technologies are reshaping both cartel operations and the methods used to combat them, creating a continually evolving arms race between colluders and enforcers.
Big Data, Digital Cartels, and Enforcement Gaps
The International Competition Network has identified significant challenges in using big data to detect digital cartels, including the apprehension of large volumes of data and the localization of digital information across jurisdictions. Cartel screening at scale requires overcoming these technical hurdles, and the fight against bid rigging in digital markets remains an evolving frontier. These systemic challenges suggest that while data-driven tools hold promise, their effectiveness depends on addressing fundamental infrastructure and coordination gaps among enforcement agencies worldwide.
Parallel Pricing and Wholesale Fuel Markets
The detection and deterrence of collusive agreements, such as price-fixing cartels, remain pivotal in maintaining market competition. A 2024 study investigated the application of machine learning methodologies in behavioral screening for detecting collusion, with a specific focus on parallel pricing behaviors in the wholesale fuel market. By employing unsupervised learning techniques, the research demonstrated how data-driven methods can complement traditional enforcement by identifying suspicious patterns in real-world market data where human analysis alone may be insufficient.