Impact of Price Inflation on Algorithmic Collusion Through Reinforcement Learning Agents
Sebastián Tinoco, Andrés Abeliuk, Javier Ruiz del Solar · 2025
Read Impact of Price Inflation on Algorithmic Collusion Through Reinforcement Learning Agents by Sebastián Tinoco, Andrés Abeliuk, Javier Ruiz del Solar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-05335.txt.
Introduction
Sebastián Tinoco and co-authors test how price rise shapes tacit price fixing by learning agents. The work dates to 2025. Code-led price sets now shape many markets, and past work showed such agents can learn to fix prices in a tacit way. Big-scale price forces got less care in that work.
The team adds price rise shocks to a learn-based price model. Agents kept high gains as one, with no direct pact. Price rise cut how hard agents fought each other, the team finds. Yet the agents built no firm stick to punish price cuts. The work flags a need for watch in code-priced markets.
Catalog record
| CREATOR | Sebastián Tinoco, Andrés Abeliuk, Javier Ruiz del Solar |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-05335
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Impact of Price Inflation on Algorithmic Collusion Through Reinforcement Learning Agents Sebastián Tinoco, Andrés Abeliuk, Javier Ruiz del Solar Abstract Algorithmic pricing is increasingly shaping market competition, raising concerns about its potential to compromise competitive dynamics. While prior work has shown that reinforcement learning (RL)-based pricing algorithms can lead to tacit collusion, less attention has been given to the role of macroeconomic factors in shaping these dynamics. This study examines the role of inflation in influencing algorithmic collusion within competitive markets. By incorporating inflation shocks into a RL-based pricing model, we analyze whether agents adapt their strategies to sustain supra-competitive profits. Our findings indicate that inflation reduces market competitiveness by fostering implicit coordination among agents, even without direct collusion. However, despite achieving sustained higher profitability, agents fail to develop robust punishment mechanisms to deter deviations from equilibrium strategies. The results suggest that inflation amplifies non-competitive dynamics in algorithmic pricing, emphasizing the need for regulatory oversight in markets where AI-driven pricing is prevalent. K eywords Algorithmic Collusion, Reinforcement Learning, Inflation, Multi-Agent Systems, Market Regulation. 1 Introduction Inflation has reemerged as a central concern in economic policymaking and business strategy, with developed economies experiencing sustained increases in price levels over the past decade. This shift, partly driven by expansive monetary responses to the COVID-19 pandemic, has heightened interest in how inflation affects firm behavior, particularly with respect to pricing. As illustrated in Figure 1 , the annual inflation rate in the United States has markedly exceeded its long-term average in recent years, introducing greater volatility into markets traditionally assumed to be stable. Figure 1: Annual Inflation Rate in the United States, 2010–2024. Inflation introduces both cost-side and demand-side uncertainties that influence firms’ pricing decisions. Prior work has shown that inflation can lead to greater price dispersion [ 1 ] , reduce price informativeness, and amplify firms’ pricing power—particularly in environments with high search or switching costs [ 2 , 3 ] . These dynamics are well-recognized in classical economic theory [ 4 ] , yet their interaction with modern algorithmic pricing systems remains underexplored. At the same time, a growing body of literature has examined how reinforcement learning (RL) algorithms deployed in competitive markets can learn to coordinate on supra-competitive pricing without explicit communication. These findings have raised important questions about the susceptibility of algorithmic systems to tacit collusion, particularly in dynamic market environments. However, most existing models abstract away from macroeconomic fluctuations, assuming stationary cost and demand conditions that may not reflect the environments in which algorithmic pricing is increasingly deployed. This paper addresses this gap by examining how inflationary shocks affect the learning dynamics and strategic behavior of autonomous pricing agents. By incorporating macroeconomic volatility into a standard RL-based pricing framework, we seek to better understand how external economic forces shape the emergence and stability of algorithmic coordination in competitive markets. In doing so, we aim to contribute to a more comprehensive framework for analyzing algorithmic interactions under realistic economic conditions. Contributions This paper contributes to the literature on algorithmic pricing and macroeconomic dynamics by introducing and analyzing the role of inflation in environments governed by autonomous pricing agents. The key contributions are as follows: • We extend existing reinforcement learning pricing models by incorporating exogenous inflation shocks, enabling the study of algorithmic behavior under macroeconomic volatility. • We propose a novel performance metric that accounts for inflation-adjusted profitability, allowing for a more realistic evaluation of strategic pricing outcomes in dynamic economic settings. • We provide empirical simulation evidence on the behavior of deep reinforcement learning agents in inflationary contexts, contributing to the understanding of how non-stationary environments shape competitive dynamics. 2 Related work The possibility of tacit algorithmic collusion —where autonomous pricing algorithms learn to sustain supra-competitive prices without explicit communication—has attracted significant interest across economics and computer science. A foundational study by Calvano et al. [ 5 ] provides initial evidence that Q-learning [ 6 ] agents competing in a repeated differentiated Bertrand [ 7 ] duopoly can converge to supra-competitive pricing strategies, enforced through “punishment” phases after deviations. This finding has sparked a growing literature on the emergence of collusion in algorithmic markets. Subsequent work has expanded these results across various market conditions. Calvano et al. [ 8 ] demonstrate that even under imperfect monitoring—akin to the Green–Porter framework [ 9 ] —Q-learning agents sustain collusion via finite trigger strategies. Klein [ 10 ] shows that even simple learners can coordinate through Edgeworth price cycles, while Banchio and Mantegazza [ 11 ] propose a theoretical model in which spontaneous synchronization of experimentation schedules leads to implicit coordination. Hansen et al. [ 12 ] and Brown and MacKay [ 13 ] further show that correlated pricing behavior may arise when agents ignore or adapt to rivals’ actions, often without intentional coordination. Eschenbaum [ 14 ] extends these findings by exploring the robustness of learned collusion under varying noise and uncertainty levels. On the empirical side, Assad et al. [ 15 ] study the German retail gasoline market and find that algorithmic pricing adoption leads to higher margins when both firms in a local duopoly use pricing software. Musolff [ 16 ] reports similar patterns in e-commerce, where algorithmic repricing reduces price dispersion and aligns sellers’ behaviors. Parallel strands investigate interventions to limit algorithmic collusion. Johnson, Rhodes, and Wildenbeest [ 17 ] model how platform design—such as demand steering or ranking rules—can restore competitive pricing. Brero et al. [ 18 ] , using a Stackelberg reinforcement learning framework [ 19 ] , show that a meta-algorithm can shape agent incentives to prevent coordination by penalizing high prices. These works suggest that market design and platform policies can be effective tools to mitigate undesired emergent behavior. Recent contributions have also begun exploring the role of large language models (LLMs). Fish et al. [ 20 ] fine-tune LLMs on pricing tasks and observe that these models can autonomously generate natural-language policies that sustain collusive outcomes, highlighting new concerns about transparency and regulation in AI-driven markets. In contrast to prior work, which predominantly models algorithmic pricing in stationary environments, our study incorporates exogenous macroeconomic shocks—specifically inflation—into the learning environment. This departs from existing approaches that assume stable cost and demand conditions, allowing us to investigate how external economic volatility interacts with agents’ adaptive behavior. To our knowledge, this is the first study to examine the intersection of macroeconomic instability and reinforcement learning in pricing. 3 Proposed method Section 3.1 introduces our approach to incorporating inflation into price competition dynamics. In Section 3.2 , we formalize the Markov Decision Process (MDP) that governs agent decision-making in the pricing environment. Finally, Section 3.3 provides a detailed explanation of the reinforcement learning algorithm used by the agents to adapt their pricing strategies. 3.1 Adding Inflation to Price Competition We model market dynamics through a Bertrand competition framework with differentiated products [ 21 ] , employing its logit formulation. The quantity q_{i,t} sold by agent i at period t is given by the following equation: q_{i,t}=\frac{e^{\frac{\alpha_{i,t}-p_{i,t}}{\mu}}}{\sum_{j=1}^{n}e^{\frac{\alpha_{j,t}-p_{j,t}}{\mu}}+e^{\frac{\alpha_{0}}{\mu}}} (1) where p_{i,t} denotes the agent’s chosen price, \alpha_{i,t} represents the agent’s vertical differentiation index, \mu corresponds to the horizontal differentiation parameter, and \alpha_{0} indicates the inverse demand index. We assume that production costs c_{t} are identical across agents and