CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0683
Title: Bootstrap-based bias correction for dynamic factor models Authors:  Andres M Alonso - Universidad Carlos III de Madrid (Spain)
Guadalupe Bastos - Universidad Carlos III de Madrid (Spain)
Carolina Garcia-Martos - Universidad Politecnica de Madrid (Spain) [presenting]
Abstract: The aim is to consider forecasts of a multivariate time series that follows a dynamic factor model. We propose to obtain interval forecasts for the common factors as well as the original time series by using bootstrap techniques. In particular, we consider the case when the factors are dominated by highly persistent AR processes. The factors' AR coefficients are estimated using small sample bias correction techniques. A Monte Carlo study points out that bias-correcting the AR coefficients of the factors allows us to obtain better results in terms of interval coverage both for the common factors and the time series. As expected, the simulation reveals that bias-correction is more successful for smaller samples. Results are gathered assuming the AR order and number of factors known as well as unknown.