Algorithmic Pricing and Algorithmic Collusion

Martin Bichler, Julius Durmann, Matthias Oberlechner

TL;DR

This paper explores how autonomous pricing algorithms in online retail can lead to "algorithmic collusion," where prices rise above competitive levels without explicit agreements. It bridges computer science (online learning) and economics (game theory) to understand this phenomenon and identify crucial research opportunities for the Business & Information Systems Engineering community.

What Problem Does This Paper Address?

The increasing use of AI and algorithms for pricing in online retail presents a complex challenge: how do these autonomous agents interact, and what are the market outcomes? While traditional economic models predict competitive pricing (e.g., Bertrand competition), learning algorithms can adapt and potentially lead to supra-competitive pricing, a phenomenon termed "algorithmic collusion." This isn't explicit cartels but rather emergent behavior where algorithms, through repeated interactions and learning, might implicitly coordinate to keep prices high. Existing research has shown this can happen in simulations, but a robust theoretical understanding of when and why it occurs, especially across different learning algorithms and market structures, is still limited. This paper aims to bridge this gap by connecting online learning algorithms with game-theoretical concepts and highlighting the interdisciplinary nature of this problem for the BISE community.

Methods

The paper doesn't present new empirical results or algorithms but rather synthesizes existing literature and frames the problem. It discusses:

  • Online Learning Algorithms: Focuses on algorithms like multi-armed bandits and Q-learning, which are used by pricing agents to learn optimal prices through trial and error and feedback. These algorithms balance exploration (trying new prices) and exploitation (using known profitable prices).
  • Game Theory Models: Primarily uses the Bertrand competition model as a framework. This model involves multiple firms simultaneously setting prices for homogeneous products, with demand depending on all prices. The paper also touches upon extensions like repeated games and the concept of Nash Equilibrium.
  • Algorithmic Collusion Analysis: Reviews existing experimental and theoretical work that uses simulations of these learning algorithms within game-theoretic models to observe emergent collusive behavior.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Emergent Collusion: Algorithmic collusion is a form of tacit collusion, arising from the learning process of algorithms without explicit communication or programming for collusion.
  • Interdisciplinary Challenge: Understanding algorithmic collusion requires a deep integration of computer science (online learning, reinforcement learning) and economics (game theory, industrial organization).
  • Feedback Matters: The type of feedback an algorithm receives (e.g., bandit feedback vs. more comprehensive information) significantly influences its propensity for collusion.
  • Beyond Simple Models: While much research focuses on simplified Bertrand models, algorithmic collusion might occur in more complex, real-world markets with fluctuating demand and supply.
  • BISE's Role: The Business & Information Systems Engineering community is uniquely positioned to address this challenge due to its expertise in both technical systems and market dynamics.

Key Results

The paper synthesizes findings from existing literature, indicating that:

  • Experimental research has demonstrated that certain reinforcement learning algorithms (like Q-learning) can lead to supra-competitive prices higher than the Nash equilibrium in simulated Bertrand competition environments.
  • However, the literature also shows that algorithmic collusion is not guaranteed; factors like the specific Q-learning algorithm variant (e.g., exploration strategy) and market structure can prevent it.
  • Some studies suggest that collusion can break down when algorithms are trained offline and then deployed in real-world market conditions.
  • The paper highlights that while the phenomenon is observed, a comprehensive theoretical framework to predict when and why it occurs is still lacking.

Limitations and Future Work

The authors acknowledge several limitations and point to future research directions:

  • Limited Theoretical Understanding: Despite experimental evidence, a robust theoretical understanding of the conditions under which algorithmic collusion emerges and persists is still limited.
  • Model Abstraction: Much of the current research relies on simplified models (e.g., static Bertrand competition with specific demand functions). Real-world markets are more complex, with fluctuating demand, supply, and product differentiation.
  • Algorithm Specificity: While Q-learning is frequently studied, its unique importance for algorithmic collusion is not definitively established, and other algorithms might also play a role.
  • BISE Research Opportunities: The paper explicitly calls for research in BISE across several areas:
  • Regulations & Monitoring: Developing frameworks to detect and regulate tacit algorithmic collusion.
  • Detection: Designing algorithms to identify pricing patterns indicative of collusion.
  • Algorithms: Analyzing how specific algorithmic features influence collusive tendencies and exploring mechanisms to ensure algorithmic accountability and fairness.
  • Beyond Oligopoly: Investigating algorithmic collusion in broader market contexts like platform competition and two-sided markets.

Summary updated

Abstract

The rise of algorithmic pricing in online retail platforms has attracted significant interest in how autonomous software agents interact under competition. This article explores the potential emergence of algorithmic collusion - supra-competitive pricing outcomes that arise without explicit agreements - as a consequence of repeated interactions between learning agents. Most of the literature focuses on oligopoly pricing environments modeled as repeated Bertrand competitions, where firms use online learning algorithms to adapt prices over time. While experimental research has demonstrated that specific reinforcement learning algorithms can learn to maintain prices above competitive equilibrium levels in simulated environments, theoretical understanding of when and why such outcomes occur remains limited. This work highlights the interdisciplinary nature of this challenge, which connects computer science concepts of online learning with game-theoretical literature on equilibrium learning. We examine implications for the Business & Information Systems Engineering (BISE) community and identify specific research opportunities to address challenges of algorithmic competition in digital marketplaces.
BibTeX
@article{bichler2025algorithmic,
  author = {Martin Bichler and Julius Durmann and Matthias Oberlechner},
  title = {{Algorithmic Pricing and Algorithmic Collusion}},
  journal = {Bus. Inf. Syst. eng},
  year = {2025},
  doi = {10.1007/s12599-025-00965-z},
}
Figure