第37回東工大経済物理学セミナー(37th Workshop on Econophysics+Statistical Physics)
|A Bayesian reconstruction approach to the history of banking in the United States (14:00-15:00)|
|Eduardo Viegas (Imperial College London)|
Abstract:There has been a remarkably rapid acceleration in the development of financial and business datasets since the latter years of the last century, in a similar manner to other fields of research. Such phenomena, enabled the establishment of new and advanced analytical methods, popularly named ``Big Data". However, one of the few downsides of such evolution is that most research switched its focus to analysis of business dynamics for the period that the data exists, without necessarily considering all potential dynamics that lead to the initial condition of the dataset being analysed. This is particularly important given that businesses and economies are non-stationary systems whereby the present dynamics are directly associated with long term evolutionary processes. Therefore, we argue that current data analytics research should be complemented by scientific methods that can provide a better understanding of the dynamics before the dataset window, to provide a more complete understanding of potential long term evolutionary process and refined data information. Here, we develop a framework, supported by Bayesian network and evolutionary dynamics methods, to estimate the ancestral distribution of US banks over long time horizons, and to adjust the existing granular level data with pre-collection period information. By making use of the Shannon--Wiener Index for the enhanced data, we find that bank crises are followed by significant increases in diversity. Moreover, we show that establishment and subsequent repeal of the Glass-Steagall Act had a profound impact to the diversity of the US banking system.
|Recent topics in social temporal network analysis (15:15-16:15)|
|Taro Takaguchi (National Institute of Information and Communications Technology)|
Abstract:Some empirical data of social interactions between individuals over time, “who interact with whom at when”, have become available for research purposes (with the consent of subjects) and keep drawing the attention of research community to the study of temporal networks, an extension of complex networks including time. Firstly, research interests were mostly in the similarity of statistical properties in different social settings, such as distributions of inter-contact times. Recently, one of main research focuses is moving to higher-order properties such as temporal correlations of activities. In this talk, we review this path of previous researches to point out three representative features of social temporal networks observed in common. In addition, as an example of analysis of higher-order properties, we introduce our recent work on application of spike train analysis methods to social communication data. Our analysis suggests difference in the response behavior of individuals when using different communications tools, such as cell phone calls, short messages, or emails.
|Statistical mechanics of exploding phase spaces (16:30-17:30)|
|Henrik Jeldtoft Jensen (Imperial College London)|
Abstract:Real complex systems, as encountered in biology or neuroscience, typically involve components that interact and create new emergent states leading to phase spaces whose volumes grow super- exponentially (“exploding”) with the number of degrees of freedom. We argue that the standard ensemble theory could break down for such cases and illustrate this phenomenon using simple models. We present a rigorously defined entropy which is extensive in the micro canonical, equal probability, ensemble for super-exponentially growing phase spaces. We suggest that this entropy may be useful in determining probability measures for such systems through appropriately constrained maximum entropy procedures. Work done in collaboration with Roozbeh Pazuki, Gunnar Pruessner and Piergiulio Tempesta