"Only in darkness, we can see the stars" - Thomas Carlyle
Primary fields: financial stability & banking, networks analysis, dynamic systems
Secondary fields: behavioral economics, computational modeling, data analysis
Shock diffusion in large regular networks: the role of transitive cycles
(with Noemí Navarro)
This paper studies how the presence of transitive cycles in the network affects the extent of financial contagion. In a regular networks where the same pattern of links repeats for each node, we allow an external shock to propagate losses through the system of linkages. The extent of contagion (contagiousness) of the network is measured by the limit of the losses when the initial shock is diffused into an infinitely large network. This measure indicates how a network structure may or may not facilitate shock diffusion, independently to other external factors. Our analysis highlights two main results. First, contagiousness decreases as the length of the minimal transitive cycle increases, keeping the degree of connectivity constant. Second, the extent of contagion is non-monotonic as degrees of connectivity increases. Our results provide new insights to better understand systemic risk and could be used to build complementary indicators for financial regulation.
Bank run, fast and slow: from behaviors to dynamics (JMP)
This paper studies the dynamic occurrence of bank runs. Existing models mainly consider bank run as miscoordination that occurs instantly in simultaneous games. In this model, bank runs can arise as continuous cascades of withdrawals. Depositors make decisions based on (i) their types, (ii) their information on total withdrawal and (iii) the observed actions of others within a network. By both analytical and numerical methods, this paper explores two novel aspects of bank runs. First, the model is able to characterize the speed and abruptness of runs. Particularly, two distinct dynamics emerge: slow runs that build up progressively vs. sudden runs that occur abruptly “out of nowhere”. Second, regarding the behavioral aspect, increase herding generates a trade-off between activation and speed of runs, bank runs are more frequent but slower to build up. By contrast, increase heterogeneity amplifies both activation and speed of runs, strictly increase bank failures.
This paper models bank runs as dynamic cascades of withdrawals, providing a complementary view to the coordination-game framework. The aim is to better understand how bank runs emerge and develop in continuous time, without imposing an exogenous sequence of actions. Agents employ a switching strategy that combines strategic actions and heuristics to make decisions. When a fraction of random agents withdraw, under the right conditions, some depositors withdraw preemptively in response, increasing the probability that other depositors will withdraw subsequently. There are two main contributions. First, the model is able to analytically characterize patterns of runs, particularly sudden runs that occur “out of nowhere”, with massive withdrawals concentrate in a very short duration, after a period of apparent inactivity. Second, we provide explicit computations of the tipping point, i.e. when the panic bursts out, to determine the time windows for interventions.
Works in progress
Systemic banking crisis
This paper investigates how small shocks trigger systemic crises, where both banks and depositors are connected among themselves. Existing literature on systemic risk mostly takes on one dimension at a time: direct contagion through defaults of payment, and amplification by runs (or fire-sales). This paper aims to endogenize both channels to study how dynamic contagion interacts with amplification in the activation, development, and tipping phase of the crisis. Potential results might have important implications for macroprudential and intervention policies.
This paper studies the dynamic diffusion of contagious diseases, where agents are arranged in a network, and their observations influence their vaccination behavior. While existing literature mostly considers vaccination as a parameter, this paper aims to endogenize vaccination behavior to determine the timing needed to dissolve the epidemic disease.
This paper studies how mobility affects collaboration and performance of inventors. We collect available data of inventors from the internet, then use machine learning and data analysis to match, disambiguate, and reconstruct the network of collaborations, in conjunction with other large datasets. Matching public datasets and individual data will allow more detailed results and better measures of performance.
Microeconomics (5 years)
Macroeconomics (3 years)
Programming (3 years)
Computational modeling (2 years)
Machine Learning (current)
GREThA-CNRS (National Center for Scientific Research), University of Bordeaux
Observatoire des Sciences et Techniques (now part of the HCERES)
Project "Ranking algorithms in networks by eigenvector centralities"
Ph.D. in Economics
University of Bordeaux, GREThA-CNRS
Thesis title "Financial fragility by network analysis and behavioral approach"