Public procurement cartels: A large-sample testing of screens using machine learning – Dataset

This release provides a comprehensive public procurement dataset for replication and prediction of cartel-detection models in the "Public procurement cartels: A large-sample testing of screens using machine learning" publication. The dataset combines around 3 million contracts from seven European countries—Bulgaria, France, Hungary, Latvia, Portugal, Spain, and Sweden—covering the period from 2004 to 2021.The data were collected from official government publication portals and open data repositories, primarily opentender.eu, and harmonized into a consistent format to enable cross-country comparisons despite differing original data structures.

Each record includes detailed bid-level information, beginning with anonymized buyer and supplier identifiers and CPV-based product classifications. The dataset also records the total number of bids submitted for each contract and the corresponding tender publication and award dates. In addition, it reports bid prices and estimated contract values, along with binary indicators that capture key bidding characteristics such as whether the bid won, whether it involved subcontracting, and whether the supplier participated as part of a consortium. A detailed summary of availble variables is provided in the README.

A critical feature of the dataset is the integration of confirmed cartel case information, sourced from competition authorities’ court rulings and official reports. Cartel cases are linked to procurement contracts through company names and cartel activity periods. While exact identification of rigged contracts remains challenging, contracts awarded to cartel-involved firms during their documented collusion periods are labeled accordingly to facilitate analysis of cartel behavior.

The final dataset comprises 73 confirmed cartel cases and over 15,000 contracts awarded to cartel members. Additionaly, we also include the unlabelled contract level data to be used for prediction. It includes multiple cartel screen capturing pricing irregularities and bidding patterns consistent with collusion, aggregated at both market and company-year levels. These indicators support machine learning models that distinguish between collusive and competitive procurement activity.

This dataset is a valuable resource for researchers, policymakers, and competition authorities focused on detecting anti-competitive practices in public procurement. Its standardized structure and continuous data coverage allow for ongoing application in cartel screening, market monitoring, and supporting competition authorities.

For a comprehensive data description, please refer to the README or more details can be found in this publication (https://www.govtransparency.eu/wp-content/uploads/2023/04/Fazekas-et-al_PP-cartel-detection_GTI-WP_2023.pdf)

The work was generously supported by the Swedish Competition Authority (grant nr. KKV 72/2018-Ab-1) and the World Bank (Bulgaria - Country Economic Memorandum. 2023).