Faculty Sponsor: Pavel Oleinikov
Lucas Holman
I’m a prospective Math and Physics double-major in the class of ’29! I’m spending my summer listening to 70s rock on my BlueTooth speaker in ALLB108, and keeping up with Love Island USA in the evenings. I’m a senator on the Wesleyan Student Assembly, as well as a theater techie, spending my free time either with Spike Tape Theater Company or the Wesleyan Sound Cooperative..
Abstract: This study examines modeling financial panic as a contagion using a discrete dynamical system. We construct a variant of the Susceptible-Infected-Recovered (SIR) disease model to capture the spread of financial instability native to pre-FDIC America. We webscrape the Historical Bank Runs Database, assembled by the New York Federal Reserve using LLMs, for historical data of American bank runs in the years 1929-1933. Using this information, we model the banks as an incomplete, undirected, weighted graph, with banks mutually linked by an edge weighted to the inverse-square of the banks’ physical distance from each other. Hence, we construct a set of difference equations according to the graph’s adjacency matrix, which correspond to the three SIR compartments and represent the probability of the banks’ appearances in each. Upon iterating the dynamical system over many time-steps, we find that our epidemiological model has a strong correlation coefficient (Pearson R = 0.76) with the historical data, though it explains little of the bank runs’ variability (R2 = 0.13). Rather, financial contagion embodies a “slow-burn” takeover of the banking system, and not the sudden spikes in failures characteristic to the Great Depression. We attribute this to the simple implementation of geographic position as a proxy for more sophisticated connections between banks, thus crediting a more complex model of financial contagion.
pdf-poster