Please use this identifier to cite or link to this item:
|Title:||Early warning signals for critical transitions in complex systems||Authors:||Wen, Haoyu||Keywords:||Science::Physics||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Wen, H. (2022). Early warning signals for critical transitions in complex systems. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/162348||Abstract:||Our societies are facing all kinds of extreme events that are hard to anticipate, yet can bring tremendous damages, such as earthquakes, financial crashes, and desertification. With recent development of complexity science starting from late 20th century, we are starting to be better equipped with theories that can help us foresee extreme events, or critical transitions. One famous approach is the Early Warning Signal (EWS) framework, which should theoretically be applicable across most empirical complex systems. However, empirical support for this framework is still lacking. There are also empirical studies that failed to identify EWSs. In this thesis, we first investigate why this might be so, by searching for EWSs in the foreign exchange market. We successfully identified EWSs and observed them to vanish when searched from down-sampled data. (Contribution 1.1) This finding suggests that EWSs can indeed be missed when data frequency is not high enough. (Contribution 1.2) Additionally, we provided the first successful demonstration of the EWS framework in the foreign exchange market. Despite this success, we expected that some might be unconvinced of our claim, due to the lack of independent and well-accepted standard for defining critical transitions in the foreign exchange market. To provide a more convincing empirical support to the EWS framework, we needed to identify EWSs in a system with well-defined critical transitions. Therefore, we proceeded to use Early Warning Indicators (EWIs) for earthquake forecasting in Taiwan, where EWIs are computed from high-frequency geoelectric data over a 7-year period. (Contribution 2.1) Through Hidden Markov Modelling, we have confidently shown that the EWIs computed from geoelectric data indeed have forecasting skills for earthquakes above magnitude 3. (Contribution 2.2) On the methodological aspect, we also contributed a successful case of applying Hidden Markov Models on EWIs, which can be valuable for future EWI studies in complex systems where we do not directly measure their states. Since EWSs cannot forecast the time of critical transitions, in our last project, we focused on a model that can provide such forecasts. We followed the recent progress of Soup-of-Group (SOG) forecasting formula. Despite the formula’s recent empirical success, it was limited by only considering the case of one single giant cluster and being incompatible with the case of simultaneous large clusters. (Contribution 3) Therefore, we proposed a new SOG forecasting formula that can work with simultaneous large clusters. We also demonstrated the new formula’s improved accuracy and reliability over the original one, as well as its significant out-of-sample forecasting skill.||URI:||https://hdl.handle.net/10356/162348||DOI:||10.32657/10356/162348||DOI (Related Dataset):||https://doi.org/10.21979/N9/JSUTCD
|Schools:||School of Physical and Mathematical Sciences||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 Dec 9, 2023
Updated on Dec 9, 2023
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.