From Rules to Regs: A Structural Topic Model of Collusion Research

TL;DR

This paper uses machine learning to analyze 20 years of research on collusion, revealing shifts from rule-based to empirical case studies and a decline in topic diversity, impacting future cartel detection.

Contribution

It introduces a novel application of structural topic modeling to analyze the evolution of collusion research over two decades.

Findings

Empirical case studies increased while rule-based research declined.

Topic diversity has decreased, indicating less pluralism in economic questions.

Research focus shifted from theoretical to empirical approaches.

Abstract

Collusive practices of firms continue to be a major threat to competition and consumer welfare. Academic research on this topic aims at understanding the economic drivers and behavioral patterns of cartels, among others, to guide competition authorities on how to tackle them. Utilizing topical machine learning techniques in the domain of natural language processing enables me to analyze the publications on this issue over more than 20 years in a novel way. Coming from a stylized oligopoly-game theory focus, researchers recently turned toward empirical case studies of bygone cartels. Uni- and multivariate time series analyses reveal that the latter did not supersede the former but filled a gap the decline in rule-based reasoning has left. Together with a tendency towards monocultures in topics covered and an endogenous constriction of the topic variety, the course of cartel research has changed notably: The variety of subjects included has grown, but the pluralism in economic questions addressed is in descent. It remains to be seen whether this will benefit or harm the cartel detection capabilities of authorities in the future.