Wang, Yuanrong;
(2024)
Network Representations for Multivariate Time-series with Applications in Portfolio Optimization and Deep Learning.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
Wang_10190440_thesis.pdf Download (4MB) | Preview |
Abstract
This dissertation probes the role of complex network theory in modelling multivariate time-series systems, a vital aspect in a wide spectrum of contemporary science. Our methodology counters many practical limitations by exploring the sparse topology of time-series data. Financial time series are marked by persistent discontinuities and low signalto-noise ratios. In this work, we propose two methodologies predicated on information filtering networks—a noise filtering technique—to address these complexities. These methodologies are subsequently extended to portfolio optimization problems. Inverse Covariance Clustering, a multivariate temporal clustering method, is integrated with contemporary portfolio optimization strategies with the aim of mitigating the impact of time-series discontinuities, colloquially termed as regime shifts in finance. Statistically Robust Information Filtering Network represents a novel framework designed to augment noise filtering in information filtering networks and enhance the signal-to-noise ratio in processed financial time-series data, thereby bolstering the diversification of portfolio construction. Moreover, we explore the utilization of information filtering networks within the domain of deep learning for modelling multivariate time series. We exhibit the benefits of deploying a filtered, sparse graph predicated on the input time-series network topology, as opposed to a fully connected graph in GNN. Further inspired by this concept, we propose an innovative MLP-like sparse architecture that also leverages network topology, and explicitly considers higher-order interactions. The incorporation of this network topology into both proposed architectures has demonstrated notable efficacy and efficiency in managing multivariate time-series data.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Network Representations for Multivariate Time-series with Applications in Portfolio Optimization and Deep Learning |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Computer Science |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10190440 |
Archive Staff Only
 |
View Item |
