Document Type
Conference Item
Publication Date
1-1-2008
Abstract
The problem of detecting an outlier and then identifying its type for bilinear time series data is studied. The e®ects of temporary change type of outlier on the observations and residuals for general bilinear processes are considered and the corresponding least-squares measure of the decision threshold is proposed. Due to the complexity of the statistics, we use a bootstrapping method to estimate the mean and standard deviation of the threshold statistics. We look at the ability of the proposed procedure to correctly detect temporary change type of outlier when compared to additive outlier and innovational outlier procedures developed in previous studies. The performances of three bootstrap-based procedures are investigated through simulation studies and shown to be good.
Keywords
Bilinear, Outlier, Least squares method, Bootstrapping, Rainfall data
Divisions
MathematicalSciences
Event Title
Conference of the Asian Regional Section of the IASC on Computational Statistics and Data Analysis
Event Location
Yokohama, Japan
Event Dates
5-8 Dec 2008
Event Type
conference