Time series data (observations indexed by time or other meaningful orderings) are encountered in many areas such as finance, economics, medicine, engineering, natural and social sciences. A recent challenge which many areas in modern statistics and data science commonly face is that, due to technological advances, observed datasets are high-dimensional as well as massive in volume. Factor modelling has been adopted as a popular dimension reduction technique for high-dimensional time series, particularly in finance and econometric data analysis, but the prevailing approaches to modelling and estimation focus on the ‘mean’ behaviour of the data, which may be inadequate for identifying more interesting features of the data. For instance, in climate data collected at a large number of sites, there may exist latent factors that drive the co-movements of extreme as well as the mean temperatures, the knowledge of which is highly relevant to forecasting and planning against heat or cold waves. We aim to propose a high-dimensional time series factor model that allows for a low-dimensional structure to govern stochastic properties other than the mean, and to develop an accompanying estimation technique that is efficient both theoretically and computationally. The proposed research will facilitate progress across a wide range of disciplines where high-dimensional time series data is routinely collected and analysed.
|Effective start/end date||22/01/19 → 8/07/19|