On February 25, 2020, the PhD candidates András Borsos and Manran Zhu, from the Department of Network and Data Science, shared the most recent advances of their work in the DNDS Seminar Series.
András is using Agent Based Modeling to study shock propagation in the banking system of Hungary with real economy feedback. He uses a contagion model of coupled bank-firm networks. He explained that the financial system is very important for the economy from the point of view of propagating shocks. His project’s contribution is to consider interactions between the financial system and the real economy, i.e. production, consumption, etc. His model is structured as a network of banks, a network of firms, and links between these two networks in the form of loan contracts. There are many possible applications for this kind of model, such as assessing different regulatory tools, predicting dynamics and shock propagation, and understanding how we can influence the economy during crisis periods. András is working in collaboration with the Hungarian Central Bank. He is employing rather detailed data, such as the balance sheet of banks and firms, investment portfolios and how they overlap, profit and loss statements, information about the supply chain, and loan contracts between banks and firms - all with anonymous linkage. His work concentrates on the 9 more important banks in Hungary, which are the most significant for scenarios of crises. In the banking system part of the model, the simulations start with a stress event, after which banks react based on indicators of solvency and liquidity. If necessary, they proceed to make adjustments in order to comply with the original requirements, but they are not able to evaluate in advance what the system level consequences will be. If the adjustments are not enough, losses on the interbank market follow. Concerning the real economy feedback part of the model, he uses a probabilistic approach, and considers upstream, downstream and horizontal propagation. So far, he has been able to observe that the inclusion of contagion, and of real economy feedback, indeed aggravates the effect of crises. As a next step, András intends to move on to a full macroeconomic model.
Manran’s project is about human navigation on information networks. It is known that the human social network is a small world network, and that it is navigable (i.e. one can efficiently direct messages through their acquaintances to reach a distant target even without knowing the whole structure of the network), as shown by the famous Milgram experiment. Later on, Watts and colleagues (2002) have shown that networks with a hierarchical structure are usually navigable. Manran is interested in understanding how this applies to information networks, and how people navigate on them. For that, she put together an online game using Wikipedia content, similar to Wiki Game and wikiSpedia. In this game, players are presented with 2 Wikipedia articles, and they have to navigate from one to the other using only hyperlinks on the page. She hired 460 players using a platform called Prolific, similar to Amazon’s Mechanical Turk. Each player played 9 different games, where the start and target articles were about well known people of different profiles, e.g. politician, scientist or artist. At the end of the games, each player also answered a survey on the platform Qualtrics, including questions about prior experience with this type of game or computer games in general, spatial navigation skills, a personality test and demographic information. She prescreened the players to obtain a 50-50 gender ratio, of varied ethnic groups, all within the U.S. This online experiment was finished recently, and the data analysis phase is now starting. Some of her preliminary results seem to indicate that players with experience in the Wiki Game win more rounds, as well as those who use Wikipedia often. Players who are younger, of Asian ethnicity, who speak more than one native language and who are of a more liberal political view also seem to have achieved better scores. Spatial navigation skills, personality type, gender and years of study at first glance did not seem to have an effect. However, a more in depth evaluation of different variables while controlling for others is needed before definite results can be extracted, and we may expect exciting news in the near future. As another possible direction for future work, Manran hopes to investigate how navigation in concept space is related to spatial navigation, since the same region of the human brain (the hippocampus) is related to spatial search as well as memory.
Blog post by Juliana Pereira