Reproduce “Deep learning for detecting bid rigging: Flagging cartel participants based on convolutional neural networks” on OpenResearch

About this paper

Martin Huber, David Imhof

Researchers from the University of Fribourg and the Swiss Competition Commission developed a deep learning approach using Convolutional Neural Networks (CNNs) to identify bid-rigging cartel participants by analyzing graphical representations of pairwise bidding interactions. This method consistently achieved over 90% accuracy in correctly classifying firms in within-country and mixed datasets, offering a proactive tool for competition agencies.

Key insights

• Convolutional Neural Networks can effectively learn subtle, complex collusive patterns directly from graphical representations of firm-level bidding interactions.
• The transformation of raw bid data into visual interaction graphs (bid rotation screen) provides a rich, interpretable input for deep learning models.
• Deep learning enables autonomous feature learning, overcoming the limitations of methods that rely on pre-defined statistical features or extensive feature engineering.

Summary by alphaXiv.