Periodic Reporting for period 1 - TCOU (The Cost of Uncertaincy - A Study on the Reform of the European Commission's Leniency Program)
Reporting period: 2023-06-01 to 2025-05-31
This project examines the effectiveness of the European Commission’s 2002 reform of its leniency program — a key tool used to detect and dismantle cartels. Cartels, which involve firms colluding to fix prices or restrict output, harm consumers and undermine fair competition. Leniency programs incentivize cartel members to self-report by offering varying degrees of fine reductions those who come forward early and provide meaningful cooperation with enforcement authorities.
In 2002, the European Commission reformed its leniency program to make it more transparent and generous, with the expectation that clearer and stronger incentives would improve cartel detection and deterrence. Assessing the reform’s impact is theoretically and empirically challenging. From a theoretical perspective, the effect of stronger leniency incentives on the overall cartel population is ambiguous. On one hand, stronger incentives to self-report can destabilize existing cartels and deter new ones from forming. On the other hand, a more predictable and generous leniency regime may give cartel members a credible tool to discipline internal defection, thereby increasing cartel stability. Empirically, the main difficulty lies in the fact that the total population of cartels — including those that are never discovered — is not observable. Researchers only have access to the subset of cartels that were investigated and made public.
This project set out to overcome this limitation by applying an alternative empirical strategy — one that draws inferences about the unobserved cartel population from the characteristics of the observed subset. The project aimed to evaluate whether the 2002 reform achieved its intended goals: increasing the rate at which cartels are discovered and discouraging the formation or persistence of collusive agreements. The research focuses on analyzing cartel-level data over a 25-year period to assess changes in cartel duration — a key indicator of deterrence — before and after the reform.
In 2002, the European Commission reformed its leniency program to make it more transparent and generous, with the expectation that clearer and stronger incentives would improve cartel detection and deterrence. Assessing the reform’s impact is theoretically and empirically challenging. From a theoretical perspective, the effect of stronger leniency incentives on the overall cartel population is ambiguous. On one hand, stronger incentives to self-report can destabilize existing cartels and deter new ones from forming. On the other hand, a more predictable and generous leniency regime may give cartel members a credible tool to discipline internal defection, thereby increasing cartel stability. Empirically, the main difficulty lies in the fact that the total population of cartels — including those that are never discovered — is not observable. Researchers only have access to the subset of cartels that were investigated and made public.
This project set out to overcome this limitation by applying an alternative empirical strategy — one that draws inferences about the unobserved cartel population from the characteristics of the observed subset. The project aimed to evaluate whether the 2002 reform achieved its intended goals: increasing the rate at which cartels are discovered and discouraging the formation or persistence of collusive agreements. The research focuses on analyzing cartel-level data over a 25-year period to assess changes in cartel duration — a key indicator of deterrence — before and after the reform.
The project began by constructing a novel dataset of 151 cartel cases investigated and convicted by the European Commission between 1996 and 2021. This required consolidating and harmonizing three sources: the widely used Private International Cartels (PIC) dataset, an updated version of the OECD cartel database (published in May 2025), and leniency-related information manually extracted from EC decisions. I used generative AI tools to assist in identifying and verifying key details from complex legal documents.
Although the original plan included more technical modeling, I determined that the best use of the data was to apply a reduced-form evaluation framework developed by Harrington and Chang (2009). This method analyzes changes in the characteristics of discovered cartels — such as their duration — to infer how reforms affect the broader population of cartels, including those that are never detected.
The analysis revealed that following the 2002 reform, the average duration of discovered cartels initially rose in the short term but declined in the long term. Under the Harrington and Chang framework, this pattern indicates that the reform likely led to a reduction in the number of active cartels over time while increasing the probability of detection. Additionally, I found that cartels formed after the reform were, on average, 65% shorter in duration than those formed before — strong evidence of a destabilizing effect.
Although the original plan included more technical modeling, I determined that the best use of the data was to apply a reduced-form evaluation framework developed by Harrington and Chang (2009). This method analyzes changes in the characteristics of discovered cartels — such as their duration — to infer how reforms affect the broader population of cartels, including those that are never detected.
The analysis revealed that following the 2002 reform, the average duration of discovered cartels initially rose in the short term but declined in the long term. Under the Harrington and Chang framework, this pattern indicates that the reform likely led to a reduction in the number of active cartels over time while increasing the probability of detection. Additionally, I found that cartels formed after the reform were, on average, 65% shorter in duration than those formed before — strong evidence of a destabilizing effect.
This is the first empirical study to apply the Harrington and Chang (2009) method and find results consistent with its predictions. Prior work by Brenner (2009), which focuses on the effect of the initial introduction of the EU leniency program in 1996, finds that the introduction was not followed by a significant variation in average cartel duration. The project also delivered a high-quality, reproducible dataset that brings together fragmented public data sources and enhances them with new, systematically extracted information from EC legal documents.
The use of generative AI in extracting and organizing policy-relevant information from formal decisions is another noteworthy innovation. This approach enabled the researcher to work more efficiently and to improve transparency and replicability in the construction of empirical legal datasets — a process with wider potential applications in regulatory and policy research.
The dataset and code will be published in an open-access repository to encourage further research in the field. The project’s methods and results may serve as a blueprint for evaluating other competition policy tools or reforms, particularly when full visibility of illegal conduct (like undetected cartels) is not possible.
The use of generative AI in extracting and organizing policy-relevant information from formal decisions is another noteworthy innovation. This approach enabled the researcher to work more efficiently and to improve transparency and replicability in the construction of empirical legal datasets — a process with wider potential applications in regulatory and policy research.
The dataset and code will be published in an open-access repository to encourage further research in the field. The project’s methods and results may serve as a blueprint for evaluating other competition policy tools or reforms, particularly when full visibility of illegal conduct (like undetected cartels) is not possible.