Abstract
Competition between traditional platforms is known to improve user utility by aligning the platform's actions with user preferences. But to what extent is alignment exhibited in data-driven marketplaces? To study this question from a theoretical perspective, we introduce a duopoly market where platform actions are bandit algorithms and the two platforms compete for user participation. A salient feature of this market is that the quality of recommendations depends on both the bandit algorithm and the amount of data provided by interactions from users. This interdependency between the algorithm performance and the actions of users complicates the structure of market equilibria and their quality in terms of user utility. Our main finding is that competition in this market does not perfectly align market outcomes with user utility. Interestingly, market outcomes exhibit misalignment not only when the platforms have separate data repositories, but also when the platforms have a shared data repository. Nonetheless, the data sharing assumptions impact what mechanism drives misalignment and also affect the specific form of misalignment (e.g. the quality of the best-case and worst-case market outcomes). More broadly, our work illustrates that competition in digital marketplaces has subtle consequences for user utility that merit further investigation.
Introduction
Conventional economic wisdom suggests that competition is the "gold standard" for maximizing user welfare. In traditional markets, if a company provides a poor product, users "vote with their feet" and move to a competitor, forcing improvement. However, digital platforms are powered by data: the quality of a recommendation system depends directly on the volume of data collected from its users. This creates a complex interdependency where a platform’s quality depends on its user base, and its user base depends on its quality.
This paper investigates whether the classical alignment between competition and user utility holds in data-driven environments. We specifically examine two regulatory scenarios:
- Separate Data Repositories: Platforms maintain proprietary data (the current status quo).
- Shared Data Repositories: Platforms share data access, a common regulatory proposal to curb big-tech dominance.
Methodology
We develop a theoretical framework using a duopoly market model based on multi-armed bandit (MAB) problems. The market consists of two platforms and a population of $N$ users. Platforms commit to a specific bandit algorithm (a strategy for exploration vs. exploitation), and users choose a platform based on the expected cumulative reward (utility).
We analyze the resulting Nash Equilibria, where no platform changes its algorithm and no user switches platforms. To quantify alignment, we introduce the User Quality Level, measuring how closely market outcomes approach the global social optimum—the utility achievable if all $N$ users cooperated perfectly. As a primary analytical tool, we utilize a "risky-safe arm" bandit problem where one arm has a known reward and another has an unknown value learnable only through user interaction.
Key Results and Theoretical Insights
Our analysis reveals that competition alone is insufficient to reach the global social optimum in data-driven marketplaces. While competition provides a safety net—ensuring users are no worse off than if they were alone—the "data advantage" of dominant platforms and "free-riding" in shared data environments prevent these markets from achieving the best possible service for everyone.
The "Data Buffer" and Market Tipping
In the separate data setting, a platform with a massive user base can afford to use a suboptimal algorithm. Because it possesses significantly more data than a small competitor, its "lazy" algorithm may still outperform a competitor's "perfect" algorithm that lacks data. This creates a "data buffer" and a barrier to entry, providing a mathematical foundation for "winner-take-all" dynamics.
The Free-Rider Problem in Shared Data
Sharing data does not necessarily resolve misalignment. In shared data environments, users and platforms have an incentive to "free-ride." No participant wants to undertake the "expensive" work of exploring new, risky options if they can wait for others to do so and then reap the data rewards. This leads to a "chilling effect" on exploration, ensuring the global social optimum is never achieved in a symmetric equilibrium.
Alignment Levels
The paper distinguishes between Idealized Alignment (achieving the absolute best service for $N$ users) and Weak Alignment (achieving service at least as good as a single user would get alone).
As summarized in the table above:
- Theorem 1: With a single user, competition works perfectly, forcing platforms to provide the optimal algorithm $\max_{A'} R_{A'}(1)$.
- Theorems 2 & 3: With multiple users and separate data, a multitude of equilibria exists, ranging from the single-user optimum to the global social optimum $\max_{A'} R_{A'}(N)$.
- Theorems 4 & 5: In shared data settings, the global social optimum is never achievable; the market always under-explores.
- Theorem 7: Regardless of the data regime, competition ensures user utility never falls below the single-user optimum.
Discussion and Future Work
Our findings suggest that competition in digital marketplaces has subtle consequences for user utility. While it prevents the worst outcomes, it fails to guarantee the best. Future research should explore dynamic choices where users can switch platforms over time, alternative platform objectives like ad revenue, and regulatory mechanisms to push markets toward "good" equilibria in separate data settings.
Summary updated