New Statistical Learning Methods for Large Curve Time Series with Applications
Dr. Xinghao Qiao
Associate Professor
Department of Statistics
London School of Economics and Political Science
With the rapid development in technology, datasets containing a large collection of functional data objects that are observed consecutively over time are becoming increasingly common in many economic and scientific applications. Examples include cumulative intraday return trajectories for a large collection of stocks, age-specific mortality or annual temperature curves at different locations and daily energy loading curves for a large number of households, to list a few. Such new type of large curve time series data requires the development of new statistical models and learning tools to extract useful information and predict future values. In this talk, I will present three new statistical learning methods to analyze large curve time series, including sparse functional vector autoregressions, functional factor models and segmentation-based functional prediction. The sample performances of the proposed methods are examined through simulations and applications to a number of real datasets.
