|
Activity Number:
|
425
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Business and Economics Statistics Section
|
| Abstract - #305784 |
|
Title:
|
Forecasting and Dynamic Updating of Time Series of Curves
|
|
Author(s):
|
Haipeng Shen*+ and Jianhua Z. Huang
|
|
Companies:
|
The University of North Carolina at Chapel Hill and Texas A&M University
|
|
Address:
|
304 Smith Building, Chapel Hill, NC, 27599,
|
|
Keywords:
|
functional data analysis ; principal component analysis ; regularization ; shrinkage ; call center
|
|
Abstract:
|
We extend functional data analysis ideas to the case of a time series of curves, and develop time series models of functional data and new methods for forecasting and dynamic updating of curves. Our approach starts with dimension reduction through functional Principal Component Analysis, which is achieved via a regularized low rank approximation technique. Curve forecasting is then obtained using the principal components and time series forecasts of their coefficient series. For dynamic updating within a curve, a shrinkage approach is proposed to combine information from the previous curves and the early part of the current curve. A data-driven mechanism for selecting the shrinkage parameter is also developed, and appears to work well empirically. The methods are illustrated via a call center application, where both inter-day forecasting and intra-day updating of call volumes are needed.
|
