Network Informed Methods for High Dimensional Time Series

点击翻译按钮获取中文摘要

Many real world data sets consist of data that changes over time, called a time series. For example, the daily average temperature in a particular region or the return on investment of some financial asset. Often,we are interested in multiple time series and the co-movement between them. If we can understand the co-movement, this can be used for improved predictions. For example, in environmental science, the concentration of various pollutants over time along with meteorological indicators can be used to predictair quality, which has a strong association with increased risk of cardiovascular and respiratory diseases. In this project we propose a new model and estimation technique for multiple time series prediction. The estimation procedure uses ideas from network science. We aim to understand the mathematical properties of the proposed model and investigate whether its predictions outperform other models on a variety ofapplication data sets. This project falls within the EPSRC MathematicalSciences research area