This paper explores decision-making processes in experience goods markets, emphasizing how agents learn to choose in situations where value is uncertain until after consumption. The study examines agents employing individual and social learning strategies within a multi-dimensional reinforcement learning framework, particularly in scenarios of repeated choices. Agents in our model gather insights from personal experiences and through recommendations within their social networks. The simulation results highlight the benefits of combining individual and social learning. While social learning yields consistent outcomes across agents, individual learning offers the possibility of higher rewards but also greater risks. A notable finding of this research is the development of asymmetrical influence patterns in social networks. This phenomenon refers to a tendency where certain agents become disproportionately influential in guiding others' choices, leading to a centralization in how advice is sought and followed within the network. This aspect of the model sheds light on the nuances of social dynamics in decision-making processes. The study enhances our understanding of consumer behavior in markets for experience goods, providing insights into the complex interplay of individual experiences and social influences in shaping economic decisions.
Opinions Polarization, Influencers, and Endogenous Online Social Networks:
We propose a model to improve our understanding of the escalating polarization of opinions in recent Western societies. This phenomenon is often attributed to the role of online social network platforms' algorithms and the influence of key individuals in creating echo chambers of opinions. Our model situates the population within a network, enabling interactions both among individuals and with these influencers. Distinctly, influencers, as a separate class of agents, learn to shape their opinions to maximize their utility by expanding their follower base. Concurrently, the platform can implement policies to control content visibility based on users' existing beliefs. Computer simulations of our model reveal several crucial insights. Contrary to popular belief, influencers tend to moderate public opinion, while platform policies favoring content alignment with users' existing views contribute to reducing polarization. Furthermore, our findings explain the Pareto distribution of influencers' network degrees, attributing it to a combination of word-of-mouth dynamics within the population and a limit on the number of influencers an individual is likely to follow.
Network Structure and the Dynamics of Social Platform Migration:
Why do users remain on a social platform even when a large share prefer to leave? We study this question through an agent-based model of platform migration in which agents adopt a rival platform whenever the fraction of their network neighbours who have already migrated exceeds a personal threshold. We simulate the model on four stylised topologies --- small-world, Erdős--Rényi, scale-free, and a preferential-attachment graph calibrated to observed Twitter/X degree statistics --- and show that network structure is a first-order determinant of migration outcomes. On homogeneous networks, equilibria are continuous and well predicted by the mean preference distribution. On the heterogeneous Twitter/X topology, outcomes are bimodal: the same preference parameters produce either near-complete migration or near-complete stagnation depending on whether early adopters reach the high-degree hub nodes.
We then examine three departures from the baseline. First, spatial clustering of platform loyalists raises the effective threshold barrier on small-world networks but accelerates collapse on Twitter/X by concentrating loyalists around hubs. Second, symmetric homophily --- in which both loyalists and pro-migration agents cluster with like-minded peers --- enables substantial migration even when average preferences are unfavourable. Third, targeted retention of hubs is highly efficient: converting half of the top-decile nodes by in-degree into loyalists reduces the mean migration rate on the X topology by more than eighty percentage points, while an equivalent number of randomly selected loyalists leaves migration largely unaffected. Together the results show that the failure of the 2022--2024 Twitter exodus can be understood as a structural coordination failure rooted in the degree heterogeneity of the underlying follow graph, and point to hub retention as the dominant mechanism sustaining platform lock-in.