Introduction and motivation

The adverse effects of climate change on economic systems and human communities are becoming increasingly evident, manifesting through more frequent and severe extreme weather events (IPCC 2022). This has prompted a growing body of literature to explore how these impacts propagate through socio-economic networks (Barrot 2016, Carvalho 2021). However, while these studies typically consider how climate shocks propagate based on existing economic structures, they do not address how such shocks may influence the evolution of production networks themselves. This research addresses this gap by adopting a bottom-up approach to assess the dynamic impacts of extreme weather events on supply chain structures.

Methodology

We develop a multi-sector disequilibrium agent-based model with buyer-seller relations between agents located in different regions. Extreme weather events are represented as realizations of a stochastic process that directly impact the supply of labour and capital services in the affected regions, with propagation effects influencing the global economy. More specifically, extreme weather events are simulated with plant-level resolution, based on known frequencies for each simulation step. Using vulnerability functions derived from the literature, the direct economic impact on production facilities is calculated. The propagation of these impacts through supply chain disruptions is then tracked, enabling firms to seek alternative suppliers when their existing ones are affected by environmental hazards. The key idea behind the substitution mechanism is that when the production levels of suppliers of intermediate inputs decline, their clients experience shortages. If these shortages surpass a specified threshold—indicated by missed deliveries—clients will actively seek out alternative suppliers.

(Preliminary) results

Preliminary results from simulations highlight two key findings. First, there is no direct correlation between output behaviour and the trend of direct losses. Although direct losses stabilize over time, output continues to decline steadily throughout the simulation period. This discrepancy indicates that factors beyond direct losses—such as propagation mechanisms, market dynamics driven by myopic expectations, and feedback loops—play a significant role in influencing output.

Second, relying solely on average values for risk assessment can lead to underestimating the actual risks. Averages tend to smooth out extreme variations and fail to capture the full extent of potential catastrophic scenarios. Similarly, the distribution of direct losses per firm shows that while most firms experience small to moderate losses, some face disproportionately higher losses. These extreme cases, combined with the fat-tailed nature of extreme weather distributions (Weitzman 2011; Pindyck 2013), underscore the need for a more nuanced approach to risk assessment that accounts for the full range of possible outcomes, rather than relying on averages alone.

Expected contributions

This research makes two key contributions. First, it elucidates how substitution dynamics following environmental shocks influence the evolution of input-output networks, projecting how global value chains may spatially reconfigure in response to climate-related disruptions. Second, by aggregating the effects of disruptions across production nodes, the study develops a bottom-up metric to assess the impact of extreme weather events on global GDP and production, providing a more granular understanding of economic vulnerabilities within supply chains.

Specific mechanism regulating the spread of gossip on social networks remains an unsolved puzzle, but an early formulation of 18 undirected relational situations indicates, that many of those configurations have a freezing effect on talking about absent others. This paper turns to agent-based modeling to attempt to work out a directed version of this mechanism and test it against longitudinal data of ties changing in organizations. Gossip is a triadic mechanism (sender-receiver-target) that requires actors to consider both: the emotional charge of their message and the valences of ties within the triad. Because of that, it is reasonable to employ a signed and labeled triad census, yielding 486 configurations exhausting all possible situations within gossip triad. This census is then implemented to an agent-based model, where synthetic agents connected by an empirical social network talk to each other and modify their perception of others based on what they hear, which is then translated to their tie composition. Our goal is to find out which of those 486 candidates are best fitted to empirical data. To this end, we estimate statistical properties of empirical networks with SAOMs and compare this with the ABM output. Because of complexity of this approach, we need to control for typical network evolution effects, network balance, exogenous processes occurring simultaneously, as well as NP centralities of individual actors. This approach blends SNA methods with the inverse Generative Social Science to formulate predictions for empirical verification.

Over half of the world’s population now lives in urban areas, and while cities can foster outsized social collaboration, they can also be sites of social isolation. Housing and the built environment are factors influencing one’s social network and social capital (access to resources embedded in social networks), which spurs interest in how urban planning can affect these social outcomes. Residential density (dwelling units per unit of residential land) and social infrastructure—like public spaces —have emerged as relevant factors in urban social network formation. However, previous work finds varied—even conflicting or non-linear—effects of residential density on social networks and social capital measures.

I present an agent-based model to assess impacts of public space amenities and different housing densities on social networks and social capital. Agents move and interact within a landscape of residences, workplaces, and public/third spaces, with variation in population, housing density, population dispersal/urban sprawl, and public spaces. Agents have upper limits on network sizes, and they form strong, weak, and invisible ties depending on repeated proximity with each other, with homophilic tendencies (preferences toward people socially similar to oneself). Agents’ existing social networks also influence their movement in space and their future tie formation. I present results for how residential density and public spaces influence social connections, including agent-centered and network-level metrics like connection count distributions and small-world metrics, and I discuss implications for social capital formation. I also outline next steps to compare and contrast model findings with in-field survey and interview data.

This study introduces an agent-based SIS (Susceptible-Infected-Susceptible) model on a square lattice to investigate how voluntary quarantine strategies influence epidemic dynamics in a structured population. Each agent is categorized into four states—Susceptible-Quarantined (Sq), Susceptible-Non-Quarantined (Sn), Infected-Quarantined (Iq), and Infected-Non-Quarantined (In)—and updates both health status and behavioral strategy over discrete time steps. Evolutionary game theory underpins agents’ decisions, with each individual weighing the perceived costs of quarantine against the risk of infection. Disease transmission occurs through contacts with immediate neighbors, with infection and recovery rates modulated by quarantine adherence. Agents evaluate their payoffs—incorporating the burden of quarantine and health risks—before potentially switching to alternative strategies based on a Fermi function. Spatial clustering and correlations among compartmental states are quantified through Moran’s I, enabling a detailed exploration of how local interactions shape global outcomes. Simulations reveal that parameter configurations, notably recovery rate and infection rate, strongly affect the equilibrium distribution of quarantine adoption. For instance, when the recovery rate falls below 0.04, more than 99% of agents eventually choose to quarantine. Conversely, achieving over 99% uninfected individuals demands not just a low infection rate or high recovery rate but also collective cooperation. These findings highlight the critical interplay between adaptive behaviors and epidemiological factors, offering insight into how voluntary quarantine measures can mitigate disease spread in structured populations.

Social networks play a crucial role in creating links between individuals and in informal transmission of information across society. Although the brain computations engaged in social learning have started to be investigated in dyadic interactions and in very small groups1-6, little is known about the mechanisms used by the brain when individuals interact in social networks. First, I will present a taxonomy of different types of computations used by the brain for learning and inferences made during social interactions. I will illustrate how this taxonomy is useful to understand the computations underlying social interactions. In particular, I will present recent model-based functional MRI results showing how the human brain adapts to fluctuating intentions of others when the nature of the interactions (to cooperate or compete) is not explicitly and truthfully signaled.

Second, I will present a new study revealing the cognitive mechanisms underpinning the assessment of information veracity. The ambiguous nature of news contents fosters misinformation and makes news veracity judgments harder. Yet, the mechanisms by which individuals assess the veracity of ambiguous news and decide whether to acquire extra information to resolve uncertainty remain unclear. Using a controlled experiment, I will show that two characteristics of news ambiguity lure individuals into mistaking true news as false: the higher the news content imprecision and propensity to divide opinions, the greater the likelihood that news are assessed as false. Individuals’ accuracy in estimating veracity is independent from their confidence in their estimation, showing limited metacognitive ability when facing ambiguous news. Yet, the level of confidence in one’s judgment is what drives the demand for extra information about the news.

Third, I will present recent findings showing how the brain decides whether to share extra information with others, depending upon one’s own confidence about the reliability of information and upon our beliefs concerning the preferences of receivers.

Fourth, I will show that a variant of the classical DeGroot learning rule, captures transmission of information in social networks. This rule, which states that an agent updates beliefs by making weighted averages of neighbors' opinions as an integrated snapshot, accounts for information propagation in an experimental game played in network, better than a sequential error-driven process using successive weighted update of one’s neighbors’ opinions.

Finally, I will show how Agent-Based modeling can be used to account for the dynamic formation of a social network in a behavioral economic experiment called the linking game. In such game, self-interested agents aim to balance maximizing their connectivity with minimizing the number of links they maintain. Our model accounts for the temporal dynamics (frequency of actions) observed in this game better than other models (eg. best response model).

Together, these results pave the way to develop a mechanistic understanding of how the brain makes inferences in social networks and decide to spread information through them, providing a multilevel comprehension of information transmission, integrating the brain system-level and the levels of individual and collective behavior.