On machine learning methods for time series with financial applications

Thesis

This doctoral project investigates machine learning methods for time series that are motivated by challenges found in financial market time series data. In this thesis, three research projects are described. The first project, which is entitled “Lead–lag detection and network clustering for multivariate time series with an application to the US equity market”, proposes a method for the extraction of clusters of leading and lagging time series in multivariate time series systems using directed network clustering. The second project, which is entitled “Time Series Prediction under Distribution Shift using Differentiable Forgetting”, proposes a bi-level optimisation framework for updating time series prediction models in response to distribution shift. The third project, “Rethinking Neural Relational Inference for Granger Causal Discovery”, studies the limitations of Neural Relational Inference, which is a graph-based variational auto-encoder model, in recovering the Granger Causal structure of multivariate time series. While a unifying theme of the thesis is that the methods developed were motivated by the characteristics of financial time series, the methods themselves can also be applied to non-financial data.

Authors: Bennett, S

• DOI: 10.5287/ora-4rg1mggrd
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English