Academic Publications
Academic Publications
Introducing the Latin American Transnational Surveillance (LATS) Dataset
with Matias Spektor (FGV-IR), Lucas O. Paes (NUPI), João Dalla Pola (IPSOS), Vitor Sion (Independent scholar)
Journal of Peace Research, vol. 62, issue 4, pages 1262–1278, (2025)
Link: https://doi.org/10.1177/00223433241268837, [Draft] [Ungated] [Database: Harvard Dataverse],
Media Coverage: "O Globo" newspaper - Sunday front page, "A Hora Podcast - Ep. 33", approx. 21-min mark, "Xadrez Verbal Podcast - Ep. 411", approx. 2:50-min mark
Transnational surveillance is a key tool for autocrats but remains understudied due to limited data. This article introduces LATS, a micro-level dataset from declassified foreign surveillance reports by autocratic Brazil (1966-1986). LATS tracks 17,000 individuals' identities, locations, social ties, and activism, mainly across Latin America. Using this data, we empirically examine the motivations, methods, and effects of transnational surveillance and apply social network analysis to explore collective-action theories of political violence by autocrats and their opponents.
Confirmation Bias in Social Networks
Mathematical Social Sciences, vol. 123, pages 59-76, (2023)
Link: https://doi.org/10.1016/j.mathsocsci.2023.02.007, [Latest WP version] [arXiv] [SSRN]
This study develops a theoretical social learning model to examine how confirmation bias shapes opinions in a networked setting. Agents exchange views with friends and interpret ambiguous public signals through a confirmation-biased rule. The model shows that, regardless of ambiguity, only two biased opinion types emerge, with one being less biased depending on the true state. The extent of bias depends on ambiguity and the relative strength of state and confirmation biases, preventing long-run learning even with impartial interpretation. Simulations reveal three key findings: (i) certain network structures foster more efficient consensus, (ii) some partisanship enhances efficiency despite confirmation bias, and (iii) open-mindedness—where opposing partisans engage—can sometimes hinder efficiency.
Social Media Networks, Fake News and Polarization, with M. Azzimonti (Richmond FED)
European Journal of Political Economy, vol. 76, 102256, (2023)
Link: https://doi.org/10.1016/j.ejpoleco.2022.102256, [NBER WP No. 24462], Media Coverage: [Live Mint]
We analyze how social media network structure and fake news impact misinformation and polarization. Using a dynamic opinion exchange model with boundedly rational agents and imperfect information, we examine the role of internet bots that spread biased news. Simulating a Twitter-calibrated network, we show that even when only 15% of agents believe fake news, network externalities drive significant misinformation and polarization. Higher bot centrality increases polarization but reduces misinformation, while asymmetric influence lowers polarization but amplifies misinformation. Threshold rules mitigate both by limiting bots’ impact when unbiased sources are available.
Presented at: SAET Annual Meeting (Santiago, Chile), 2024 NSF Network Science and Economics (Minnesota, USA), International Conference on Game Theory (Stony Brook, USA), Concordia University VIEE Seminar (Montreal, Canada), LACEA-LAMES (Montevideo, Uruguay), SED Winter Meeting (Buenos Aires, Argentina), Coalition Theory Network (Paris, France), Rochester Economic Theory Seminar Series (Rochester, USA - scheduled) , 2025 Midwest Theory Conference (State College, USA - scheduled), University of Sao Paulo Seminar Series (Sao Paulo, Brazil - scheduled)
This study investigates the practice of experts aggregating forecasts before informing a decision-maker. The significance of this subject extends to various contexts where experts inform their assessments to a decision-maker following discussions with peers. In situations where experts are equally precise, and pair-wise correlation of forecasts is the same across all pairs of experts, the network structure plays a pivotal role in decision-making variance. For classical structures, I show that star networks exhibit the highest variance, contrasting with d-regular networks that achieve zero variance, emphasizing their efficiency. Additionally, by employing the Poisson random graph model under the assumptions of a large network size and a small connection probability, the results indicate that both the expected Network Bias and its variance converge to zero as the network size becomes sufficiently large.
Information Aggregation and Social Learning via the Delphi Method (2025) [draft], with JF. Cabral-Perez (Harvard University)
Presented at: 2025 NSF Network Science and Economics (Stanford, USA)
We study the Delphi Method, a process for aggregating expert opinions developed by the RAND Corporation in the 1950s. It consists of repeated rounds of (i) opinion elicitation, (ii) information aggregation using a summary statistic, and (iii) individual opinion updating. Using a Bayesian framework, we show that the method reaches efficient consensus as the number of experts increases. We characterize the expected time for approximate epsilon-consensus and examine the trade-off between the number of experts and rounds. When experts communicate outside the Delphi protocol, consensus is still reached under weak connectivity conditions, but efficiency is not guaranteed.
Incentivizing Accuracy: The Role of Public Rankings in Economic Forecasting (2025) [draft], with V. Sabbag (FGV-EESP)
Presented at: 25th Brazilian Finance Meeting (São Paulo, Brazil), LACEA 2025 (Recife, Brazil - scheduled)
Forecasters respond to various incentives, balancing accuracy with strategic considerations. This study examines how the competition among forecasters introduced by the Brazil's Central Bank (BCB) affects the accuracy of GDP growth projections. Using a counterfactual approach, we find that forecast errors would have been significantly smaller during crisis periods, such as 2013–2016 and 2020–2021, if the competition had always existed. To uncover the mechanisms behind this effect, we identify two distinct channels: (i) a copying channel, where forecasters mimic top performers, and (ii) an incentive channel, where competition drives greater effort for publicity and accuracy. Our findings indicate that during unexpected shocks (e.g., the COVID-19 pandemic), forecasters tend to rely on copying, whereas in structural crises (e.g., 2013–2016), competition primarily incentivizes individual accuracy improvements.
Attention Centrality (2025)
Masculinity Norms and Social Learning: Theory and Experimental Evidence (2024), with I. Matavelli (Univ. of New South Wales)
Polarization and Misinformation Cycles in Social Networks (2024), with T. Aguirre (Univ. of São Paulo)
Patent Licensing in Product-Variety Networks (2024), with C. Rubbini (Florida State University) [Slides]
Presented at: 2023 International Conference on Game Theory (Stony Brook, USA), 2025 Southern Economic Association (Tampa, USA, scheduled)
Debunking Extremism in WhatsApp: Evidence from a Natural Experiment (2024), with L. Meloni (Univ. of São Paulo), F. Bailez (Palver), MV. Cruz (Univ. of São Paulo), T. Aguirre (Univ. of São Paulo)
Targeting Transnational Dissident Networks (2024), with M. Spektor (FGV-IR), L. O. Paes (NUPI)