|
Activity Number:
|
212
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
| Abstract - #304963 |
|
Title:
|
Dynamic Factors in Periodic Time-Varying Regression Models
|
|
Author(s):
|
Marius Ooms*+ and Virginie Dordonnat and Siem Jan Koopman
|
|
Companies:
|
Vrije Universiteit Amsterdam and Electricité de France and Vrije Universiteit Amsterdam
|
|
Address:
|
De Boelelaan 1105, Amsterdam, 1081 HV, Netherlands
|
|
Keywords:
|
unobserved components ; seasonality ; state space models ; load forecasting ; electricity demand
|
|
Abstract:
|
We consider dynamic multivariate periodic regression modeling for high- frequency data. The dependent univariate time series is transformed to a lower frequency multivariate time series for periodic regression modeling. For hourly series we specify one equation per hour of the day. The regression coefficients differ across equations and vary stochastically over time. As the unrestricted model contains many unknown parameters, we develop a methodology within the state-space framework to model dynamic factors in the coefficients, with common coefficient dynamics across equations. We first present a small-scale simulation, comparing results with a univariate benchmark model. Our dynamic factor component estimates are more precise. We apply our method to French national hourly electricity loads with weather variables and calendar variables as regressors and analyze components and forecasts.
|
