Please use this identifier to cite or link to this item:
|Title:||Complexity science approach to study decentralized financial systems using tools from statistical physics and machine learning||Authors:||Aspembitova, Ayana T.||Keywords:||Science::Physics||Issue Date:||2020||Publisher:||Nanyang Technological University||Source:||Aspembitova, A. T. (2020). Complexity science approach to study decentralized financial systems using tools from statistical physics and machine learning. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/148631||Abstract:||Decentralized Finance is the new socioeconomic system growing with an extremely fast pace and changing the way financial interactions are being conducted. In comparison with the growing importance of digital assets and blockchain technology, there is still little understanding of Decentralized Finance as a system. In this thesis we analyze transaction datasets from Bitcoin and Ethereum blockchains to obtain a comprehensive understanding of digital assets - from studying the behaviour of each part to investigating the whole structure and deriving the relations between micro and macro properties of the cryptocurrency systems. Using the Complex Networks approach we explained the system's overall structure and dynamics, and uncovered the mechanism behind network formation. It was found that there is fitness preferential attachment among nodes in the bitcoin network that leads the system to scale-free behaviour. We proposed the quantifiable definition of fitness and supported our finding by simulating a synthetic network and reproducing the main properties of the bitcoin network. After having a good understanding about the structure of the system, we zoom in into its parts by studying the behavioral patterns among the system's users (people). We develop the methodology based on Machine Learning models to define distinct behavioral types in the cryptocurrency systems and find that despite differences between the bitcoin and ethereum systems, there are four common strategies that users follow in both markets. Based on our finding, we model the dynamics of people's behaviour in market as an Absorbing Markov Chain. This approach allowed us to present the behavioral switches in a comprehensive and intuitive way. Moreover, we were able to obtain the predictions on the longevity of users in the system according to their behaviour. Finally, we use the Granger causality test to derive the relations between all system characteristics. We attempt to explain the effect of behavioral switches on the structural properties and price; we find that indeed, switches of users from certain behavioral groups causes a change in price which affects the size of the network as well. We hope that the work and results presented in this thesis will advance the understanding of the new field of Decentralized Finance and expect that the research approach and methodologies developed for this study will be helpful to investigate various complex systems as well.||URI:||https://hdl.handle.net/10356/148631||DOI:||10.32657/10356/148631||Rights:||This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SPMS Theses|
Updated on Jun 22, 2022
Updated on Jun 22, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.