Arnoud Victor Den Boer
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Arnoud Victor Den Boer has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Nuri Bora Keskin
Fuqua School of Business, Duke University, Durham, North Carolina 27708
3 shared publications - Janusz Martin Meylahn
Department of Applied Mathematics, University of Twente, 7522 NB Enschede, Netherlands
2 shared publications - Benedikt Valentin Meylahn
Korteweg-de Vries Institute for Mathematics, University of Amsterdam, Amsterdam, The Netherlands
1 shared publication Boxiao Chen
University of Illinois at Chicago - College of Business Administration
1 shared publication- Boxiao Chen
College of Business Administration, University of Illinois Chicago, Chicago, Illinois 60607
1 shared publication Michel Mandjes
Amsterdam Business School, University of Amsterdam, 1018 TV Amsterdam, Netherlands
1 shared publication
Latest Publications
- Open Access
- 28 July 2026
- Published by Elsevier in SSRN Electronic Journal
A (mathematical) definition of algorithmic collusion
Legal scholars and competition experts have expressed concerns regarding the ability of algorithms to collude without breaking competition law. This concern about algorithmic collusion has initiated a stream of literature that aims to construct algorithms that collude, but also a stream of literature that criticizes claims of algorithmic collusion. The debate partly originates from the fact that there exists no common definition of what it means for an algorithm to collude. In this paper, we contribute to the debate by proposing a mathematical definition of algorithmic collusion. Our definition allows for the systematic comparison of claims regarding algorithmic collusion and thus paves the way for identifying algorithm designs conducive to collusion. This is in turn necessary for the regulation and detection of algorithmic collusion in the future.
- 28 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA (mathematical) definition of algorithmic collusion
Legal scholars and competition experts have expressed concerns regarding the ability of algorithms to collude without breaking competition law. This concern about algorithmic collusion has initiated a stream of literature that aims to construct algorithms that collude, but also a stream of literature that criticizes claims of algorithmic collusion. The debate partly originates from the fact that there exists no common definition of what it means for an algorithm to collude. In this paper, we contribute to the debate by proposing a mathematical definition of algorithmic collusion. Our definition allows for the systematic comparison of claims regarding algorithmic collusion and thus paves the way for identifying algorithm designs conducive to collusion. This is in turn necessary for the regulation and detection of algorithmic collusion in the future.
- 1 November 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 72 (6) , 2446-2466
Pricing and Positioning of Horizontally Differentiated Products with Incomplete Demand Information
How to Manage Horizontally Differentiated Products When Customer Preferences and Demand Distributions Are Unknown In the paper “Pricing and Positioning of Horizontally Differentiated Products with Incomplete Demand Information,” we consider the problem of determining the optimal prices and product configurations of horizontally differentiated products when customers purchase according to a locational choice model and where the problem parameters are initially unknown to the decision maker. We propose a data-driven algorithm that learns the optimal prices and product configurations from accumulating sales data, and we show that their regret—the expected cumulative loss caused by not using optimal decisions—after T time periods is [ ]. We accompany this result by proving an almost-matching lower bound of regret, implying that our algorithms are asymptotically near optimal. In an extension, we show how our algorithm can be adapted for the case of fixed locations.
- 1 November 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 72 (6) , 2446-2466
Pricing and Positioning of Horizontally Differentiated Products with Incomplete Demand Information
How to Manage Horizontally Differentiated Products When Customer Preferences and Demand Distributions Are Unknown In the paper “Pricing and Positioning of Horizontally Differentiated Products with Incomplete Demand Information,” we consider the problem of determining the optimal prices and product configurations of horizontally differentiated products when customers purchase according to a locational choice model and where the problem parameters are initially unknown to the decision maker. We propose a data-driven algorithm that learns the optimal prices and product configurations from accumulating sales data, and we show that their regret—the expected cumulative loss caused by not using optimal decisions—after T time periods is [ ]. We accompany this result by proving an almost-matching lower bound of regret, implying that our algorithms are asymptotically near optimal. In an extension, we show how our algorithm can be adapted for the case of fixed locations.
- Open Access
- 1 June 2024
- Published by AIP Publishing in Chaos: An Interdisciplinary Journal of Nonlinear Science
- Vol. 34 (6)
Interpersonal trust: Asymptotic analysis of a stochastic coordination game with multi-agent learning
We study the interpersonal trust of a population of agents, asking whether chance may decide if a population ends up with high trust or low trust. We model this by a discrete time, stochastic coordination game with pairwise interactions occurring at random in a finite population. Agents learn about the behavior of the population using a weighted average of what they have observed in past interactions. This learning rule, called an “exponential moving average,” has one parameter that determines the weight of the most recent observation and may, thus, be interpreted as the agent’s memory. We prove analytically that in the long run, the whole population always either trusts or doubts with the probability one. This remains true when the expectation of the dynamics would indicate otherwise. By simulation, we study the impact of the distribution of the payoff matrix and of the memory of the agents. We find that as the agent memory increases (i.e., the most recent observation weighs less), the actual dynamics increasingly resemble the expectation of the process. We conclude that it is possible that a population may converge upon high or low trust between its citizens simply by chance, though the game parameters (context of the society) may be quite telling.
- 1 June 2024
- Published by AIP Publishing in Chaos: An Interdisciplinary Journal of Nonlinear Science
- Vol. 34 (6)
Open AccessInterpersonal trust: Asymptotic analysis of a stochastic coordination game with multi-agent learning
We study the interpersonal trust of a population of agents, asking whether chance may decide if a population ends up with high trust or low trust. We model this by a discrete time, stochastic coordination game with pairwise interactions occurring at random in a finite population. Agents learn about the behavior of the population using a weighted average of what they have observed in past interactions. This learning rule, called an “exponential moving average,” has one parameter that determines the weight of the most recent observation and may, thus, be interpreted as the agent’s memory. We prove analytically that in the long run, the whole population always either trusts or doubts with the probability one. This remains true when the expectation of the dynamics would indicate otherwise. By simulation, we study the impact of the distribution of the payoff matrix and of the memory of the agents. We find that as the agent memory increases (i.e., the most recent observation weighs less), the actual dynamics increasingly resemble the expectation of the process. We conclude that it is possible that a population may converge upon high or low trust between its citizens simply by chance, though the game parameters (context of the society) may be quite telling.
- Open Access
- 1 April 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (4) , 1169-1186
Data‐driven collusion and competition in a pricing duopoly with multinomial logit demand
We consider dynamic pricing and demand learning in a duopoly with multinomial logit demand, both from the perspective where firms compete against each other and from the perspective where firms aim to collude to increase revenues. We show that joint‐revenue maximization is not always beneficial to both firms compared to the Nash equilibrium, and show that several other axiomatic notions of collusion can be constructed that are always beneficial to both firms and a threat to consumer welfare. Next, we construct a price algorithm and prove that it learns to charge supra‐competitive prices if deployed by both firms, and learns to respond optimally against a class of competitive algorithms. Our algorithm includes a mechanism to infer demand observations from the competitor's price path, so that our algorithm can operate in a setting where prices are public but demand is private information. Our work contributes to the understanding of well‐performing price policies in a competitive multi‐agent setting, and shows that collusion by algorithms is possible and deserves the attention of lawmakers and competition policy regulators.
- 1 April 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (4) , 1169-1186
Open AccessData‐driven collusion and competition in a pricing duopoly with multinomial logit demand
We consider dynamic pricing and demand learning in a duopoly with multinomial logit demand, both from the perspective where firms compete against each other and from the perspective where firms aim to collude to increase revenues. We show that joint‐revenue maximization is not always beneficial to both firms compared to the Nash equilibrium, and show that several other axiomatic notions of collusion can be constructed that are always beneficial to both firms and a threat to consumer welfare. Next, we construct a price algorithm and prove that it learns to charge supra‐competitive prices if deployed by both firms, and learns to respond optimally against a class of competitive algorithms. Our algorithm includes a mechanism to infer demand observations from the competitor's price path, so that our algorithm can operate in a setting where prices are public but demand is private information. Our work contributes to the understanding of well‐performing price policies in a competitive multi‐agent setting, and shows that collusion by algorithms is possible and deserves the attention of lawmakers and competition policy regulators.
- Top Cited Papers
- 1 October 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 68 (10) , 7112-7130
Dynamic Pricing with Demand Learning and Reference Effects
We consider a seller’s dynamic pricing problem with demand learning and reference effects. We first study the case in which customers are loss-averse: they have a reference price that can vary over time, and the demand reduction when the selling price exceeds the reference price dominates the demand increase when the selling price falls behind the reference price by the same amount. Thus, the expected demand as a function of price has a time-varying “kink” and is not differentiable everywhere. The seller neither knows the underlying demand function nor observes the time-varying reference prices. In this setting, we design and analyze a policy that (i) changes the selling price very slowly to control the evolution of the reference price and (ii) gradually accumulates sales data to balance the trade-off between learning and earning. We prove that, under a variety of reference-price updating mechanisms, our policy is asymptotically optimal; that is, its T-period revenue loss relative to a clairvoyant who knows the demand function and the reference-price updating mechanism grows at the smallest possible rate in T. We also extend our analysis to the case of a fixed reference price and show how reference effects increase the complexity of dynamic pricing with demand learning in this case. Moreover, we study the case in which customers are gain-seeking and design asymptotically optimal policies for this case. Finally, we design and analyze an asymptotically optimal statistical test for detecting whether customers are loss-averse or gain-seeking. This paper was accepted by Omar Besbes, revenue management and market analytics.
- 1 October 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 68 (10) , 7112-7130
Top Cited PapersDynamic Pricing with Demand Learning and Reference Effects
We consider a seller’s dynamic pricing problem with demand learning and reference effects. We first study the case in which customers are loss-averse: they have a reference price that can vary over time, and the demand reduction when the selling price exceeds the reference price dominates the demand increase when the selling price falls behind the reference price by the same amount. Thus, the expected demand as a function of price has a time-varying “kink” and is not differentiable everywhere. The seller neither knows the underlying demand function nor observes the time-varying reference prices. In this setting, we design and analyze a policy that (i) changes the selling price very slowly to control the evolution of the reference price and (ii) gradually accumulates sales data to balance the trade-off between learning and earning. We prove that, under a variety of reference-price updating mechanisms, our policy is asymptotically optimal; that is, its T-period revenue loss relative to a clairvoyant who knows the demand function and the reference-price updating mechanism grows at the smallest possible rate in T. We also extend our analysis to the case of a fixed reference price and show how reference effects increase the complexity of dynamic pricing with demand learning in this case. Moreover, we study the case in which customers are gain-seeking and design asymptotically optimal policies for this case. Finally, we design and analyze an asymptotically optimal statistical test for detecting whether customers are loss-averse or gain-seeking. This paper was accepted by Omar Besbes, revenue management and market analytics.
- 1 October 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 68 (10) , 7480-7496
Price Optimization Under the Finite-Mixture Logit Model
We consider price optimization under the finite-mixture logit model. This model assumes that customers belong to one of a number of customer segments, where each customer segment chooses according to a multinomial logit model with segment-specific parameters. We reformulate the corresponding price optimization problem and develop a novel characterization. Leveraging this new characterization, we construct an algorithm that obtains prices at which the revenue is guaranteed to be at least [ ] times the maximum attainable revenue for any prespecified [ ]. Existing global optimization methods require exponential time in the number of products to obtain such a result, which practically means that the prices of only a handful of products can be optimized. The running time of our algorithm, however, is exponential in the number of customer segments and only polynomial in the number of products. This is of great practical value, because in applications, the number of products can be very large, whereas it has been found in various contexts that a low number of segments is sufficient to capture customer heterogeneity appropriately. The results of our numerical study show that (i) ignoring customer segmentation can be detrimental for the obtained revenue, (ii) heuristics for optimization can get stuck in local optima, and (iii) our algorithm runs fast on a broad range of problem instances. This paper was accepted by Omar Besbes, revenue management and market analytics.
- 1 October 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 68 (10) , 7480-7496
Price Optimization Under the Finite-Mixture Logit Model
We consider price optimization under the finite-mixture logit model. This model assumes that customers belong to one of a number of customer segments, where each customer segment chooses according to a multinomial logit model with segment-specific parameters. We reformulate the corresponding price optimization problem and develop a novel characterization. Leveraging this new characterization, we construct an algorithm that obtains prices at which the revenue is guaranteed to be at least [ ] times the maximum attainable revenue for any prespecified [ ]. Existing global optimization methods require exponential time in the number of products to obtain such a result, which practically means that the prices of only a handful of products can be optimized. The running time of our algorithm, however, is exponential in the number of customer segments and only polynomial in the number of products. This is of great practical value, because in applications, the number of products can be very large, whereas it has been found in various contexts that a low number of segments is sufficient to capture customer heterogeneity appropriately. The results of our numerical study show that (i) ignoring customer segmentation can be detrimental for the obtained revenue, (ii) heuristics for optimization can get stuck in local optima, and (iii) our algorithm runs fast on a broad range of problem instances. This paper was accepted by Omar Besbes, revenue management and market analytics.
- 1 September 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 24 (5) , 2577-2594
Learning to Collude in a Pricing Duopoly
Problem definition: This paper addresses the question whether or not self-learning algorithms can learn to collude instead of compete against each other, without violating existing competition law. Academic/practical relevance: This question is practically relevant (and hotly debated) for competition regulators, and academically relevant in the area of analysis of multi-agent data-driven algorithms. Methodology: We construct a price algorithm based on simultaneous-perturbation Kiefer–Wolfowitz recursions. We derive theoretical bounds on its limiting behavior of prices and revenues, in the case that both sellers in a duopoly independently use the algorithm, and in the case that one seller uses the algorithm and the other seller sets prices competitively. Results: We mathematically prove that, if implemented independently by two price-setting firms in a duopoly, prices will converge to those that maximize the firms’ joint revenue in case this is profitable for both firms, and to a competitive equilibrium otherwise. We prove this latter convergence result under the assumption that the firms use a misspecified monopolist demand model, thereby providing evidence for the so-called market-response hypothesis that both firms’ pricing as a monopolist may result in convergence to a competitive equilibrium. If the competitor is not willing to collaborate but prices according to a strategy from a certain class of strategies, we prove that the prices generated by our algorithm converge to a best-response to the competitor’s limit price. Managerial implications: Our algorithm can learn to collude under self-play while simultaneously learn to price competitively against a ‘regular’ competitor, in a setting where the price-demand relation is unknown and within the boundaries of competition law. This demonstrates that algorithmic collusion is a genuine threat in realistic market scenarios. Moreover, our work exemplifies how algorithms can be explicitly designed to learn to collude, and demonstrates that algorithmic collusion is facilitated (a) by the empirically observed practice of (explicitly or implicitly) sharing demand information, and (b) by allowing different firms in a market to use the same price algorithm. These are important and concrete insights for lawmakers and competition policy professionals struggling with how to respond to algorithmic collusion.
- 1 September 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 24 (5) , 2577-2594
Learning to Collude in a Pricing Duopoly
Problem definition: This paper addresses the question whether or not self-learning algorithms can learn to collude instead of compete against each other, without violating existing competition law. Academic/practical relevance: This question is practically relevant (and hotly debated) for competition regulators, and academically relevant in the area of analysis of multi-agent data-driven algorithms. Methodology: We construct a price algorithm based on simultaneous-perturbation Kiefer–Wolfowitz recursions. We derive theoretical bounds on its limiting behavior of prices and revenues, in the case that both sellers in a duopoly independently use the algorithm, and in the case that one seller uses the algorithm and the other seller sets prices competitively. Results: We mathematically prove that, if implemented independently by two price-setting firms in a duopoly, prices will converge to those that maximize the firms’ joint revenue in case this is profitable for both firms, and to a competitive equilibrium otherwise. We prove this latter convergence result under the assumption that the firms use a misspecified monopolist demand model, thereby providing evidence for the so-called market-response hypothesis that both firms’ pricing as a monopolist may result in convergence to a competitive equilibrium. If the competitor is not willing to collaborate but prices according to a strategy from a certain class of strategies, we prove that the prices generated by our algorithm converge to a best-response to the competitor’s limit price. Managerial implications: Our algorithm can learn to collude under self-play while simultaneously learn to price competitively against a ‘regular’ competitor, in a setting where the price-demand relation is unknown and within the boundaries of competition law. This demonstrates that algorithmic collusion is a genuine threat in realistic market scenarios. Moreover, our work exemplifies how algorithms can be explicitly designed to learn to collude, and demonstrates that algorithmic collusion is facilitated (a) by the empirically observed practice of (explicitly or implicitly) sharing demand information, and (b) by allowing different firms in a market to use the same price algorithm. These are important and concrete insights for lawmakers and competition policy professionals struggling with how to respond to algorithmic collusion.
- 1 May 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 70 (3) , 1613-1628
Continuous Assortment Optimization with Logit Choice Probabilities and Incomplete Information
This paper considers a novel formulation of the classical assortment optimization problem with multinomial logit demand and unknown model parameters. The novelty lies in the fact that the set of products is not finite but a continuum, motivated by the desire to understand the problem characteristics for many products, as well as by applications where products are characterized by a continuous quality variable. For settings with and without capacity constraints, the authors design data-driven decision policies and prove upper and lower bounds on the regret, which imply that these policies are asymptotically optimal.
- 1 May 2022
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Operations Research
- Vol. 70 (3) , 1613-1628
Continuous Assortment Optimization with Logit Choice Probabilities and Incomplete Information
This paper considers a novel formulation of the classical assortment optimization problem with multinomial logit demand and unknown model parameters. The novelty lies in the fact that the set of products is not finite but a continuum, motivated by the desire to understand the problem characteristics for many products, as well as by applications where products are characterized by a continuous quality variable. For settings with and without capacity constraints, the authors design data-driven decision policies and prove upper and lower bounds on the regret, which imply that these policies are asymptotically optimal.
- 12 April 2022
- Published by Springer Nature
- p. 79-101
Dynamic Pricing with Demand Learning: Emerging Topics and State of the Art
- 12 April 2022
- Published by Springer Nature
- p. 79-101
Dynamic Pricing with Demand Learning: Emerging Topics and State of the Art
- 12 April 2022
- Published by Springer Nature
- p. 137-150
Dynamic Pricing and Demand Learning in Nonstationary Environments
- 12 April 2022
- Published by Springer Nature
- p. 137-150
Dynamic Pricing and Demand Learning in Nonstationary Environments