Conference Agenda

Our research delves into the optimal pricing strategy within a duopolistic media platform market, considering the interplay of cross-network effects among viewers, content creators, and advertisers. Employing a game-theoretical linear city model, we analyze the dynamics among these interconnected stakeholders across three sides. Our investigation examines how platform pricing adjusts with the presence of multi-homing on one or more market sides. The insights obtained from our study underscore strategic opportunities for competitive platforms to leverage pricing, differentiation, and cross-network effects effectively to optimize their profits.

We evaluate the relevance of existing literature findings within our context, identifying instances of both alignment and deviation across various scenarios. Our findings suggest that under specific conditions, multi-homing can yield benefits for both market agents and platforms. We explore scenarios wherein multiple sides of the market engage in multi-homing concurrently, proposing that platforms should incentivize one side to multi-home only when other sides also participate in multi-homing. Furthermore, we outline criteria for offering free access or implementing subscription fees for market agents, offering practical guidance for platforms operating within a competitive landscape in the presence of cross-network effects and multi-homing.

Over the past two decades, a significant trend in the business-to-consumer (B2C) sector has been the migration to online platforms like Amazon and Alibaba, relying on Artificial Intelligence (AI) and big data for pricing strategies.

This has sparked debate on whether pricing algorithms may tacitly collude to set supra-competitive prices without explicit programming. Our study addresses these concerns by examining the risk of collusive behavior using Reinforcement Learning (RL) algorithms to decide on pricing strategies in competitive markets.

Prior research in this field focused exclusively on Tabular Q-learning (TQL) and led to opposing views on whether learning-based algorithms can lead to supra-competitive prices. Against this background, our work contributes to this ongoing discussion by providing a more nuanced numerical study that goes beyond TQL by additionally capturing off- and on-policy based deep RL algorithms:

Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO). We study multiple Bertrand oligopoly variants and show that algorithmic collusion indeed depends on the algorithm used. In our experiments, TQL exhibits higher collusion and price dispersion phenomena, while DQN and PPO show lower collusion tendencies. With PPO, in particular, achieving competitive outcomes in a significant share of scenarios. Moreover, our study suggests that continuous action spaces may foster competition by enabling firms to respond more effectively to each other compared to discrete action spaces. Our study also reveals that algorithm dynamics are sensitive to the market environment and general algorithm design decisions, highlighting the complexity of implementing AI in competitive markets.

Over the last decades, dynamic pricing has become increasingly popular and challenging. Pricing problems become even more complex if the customer’s and competitor’s behavior are strategic and unknown. Reinforcement Learning (RL) are a promising approach for solving such dynamic problems with incomplete knowledge. RL algorithms have shown to outperform rule-based heuristics if the underlying Markov decision process used to model market dynamics are kept as simple. Note, this was necessary to calculate optimal pricing policies with traditional solution methods. In this context, in the literature, the customer’s behavior is mostly assumed to be myopic. However, the myopic assumption is becoming increasingly unrealistic since technology like price trackers allows customers to act more strategically. In this work, we introduce several types of strategic customers. Specifically, we consider strategic customers who base their purchase decision on past prices and those who anticipate future price development. Further, we investigate whether RL agents are able to cope with strategic consumer behavior. We consider monopoly as well as duopoly markets.