Journal of Advanced Research in Applied Mechanics
ISSN (online): 2289-7895 | Vol. 8, No. 1. Pages 13-31, 2015
13
Penerbit
Akademia Baru
Performance of Robust Wild Bootstrap Estimation
of Linear Model in the Presence of Outliers and
Heteroscedasticity Errors
R. Adnan
*,1,a
, B. A. Rasheed
1,b
, S. E. Saffari
2,c
and K. D. Pati
1,d
1
Department of mathematics, Faculty of Science, UTM, 81310 UTM Skudai, Johor, Malaysia
2
Centre of Education, Sabzevar University of Medical Sciences, Sabzevar, Iran.
a,*
robiah@utm.my,
b
arasheedbello@yahoo.com,
c
ehsanreiki@yahoo.com,
d
kafi_dano@yahoo.com
Abstract- Bootstrap techniques are widely used today in many other fields such as economics,
Business Administration, Physics, Engineering, Chemistry, Meteorological, Biological Sciences and
Medicine. This paper is concerned with the estimation of linear regression model parameters in the
presence of heteroscedasticity using wild bootstrap approaches of Wu and Liu. The empirical
evidence has shown that these techniques are effective in the presence of heteroscedasticity. However,
when there are outliers in the data, this method is no longer effective. To overcome this situation, this
paper proposed robust wild bootstrap estimation methods where heteroscedasticity and outliers occur
simultaneously. The proposed method is based on the Tukey-redesceding M-estimator which
incorporate the LTS and LMS estimator, robust scale and location, and the wild bootstrap sampling
procedures of Liu and Wu. Its performance is compared with other existing robust wild bootstrap
estimator of MM-estimator using real data and simulation study. The results obtained from this study
disclosed that the proposed methods offer a substantial improvement over the existing techniques and
proved to be a good alternative estimator. Copyright © 2015 Penerbit Akademia Baru - All rights
reserved.
Keywords: Robust Estimation, Wild Bootstrap, Bias, Standard error and RMS.
1.0 INTRODUCTION:
Wild bootstrap method was first proposed by [1] which gives a better performance for
homoscedastic and heteroscedastic models. However, a better alternative estimation method
is introduced by [2-3] following the idea of [1] to estimate the regression model parameters.
The most common bootstrap methods are the residuals bootstrap and the paired bootstrap
which are defined in [4], and some of their asymptotic properties can be found in [5-7]
among others. For bootstrap method, [8] proposed a bootstrap procedure based on random
weight on the loss functions, [9] established a modified form of the residuals bootstrap, and
[10] considered the validity of paired bootstrap techniques. [11] proposed a modified
weighted bootstrap estimation method based on LTS. To account for heteroscedasticity [1-3]
proposed the wild bootstrap techniques by randomly weighting the residuals. However,
different attempts have been made to use the procedure of [1-2] wild bootstrap techniques to
remedy the problem of heteroscedasticity error variance. Others including [12], [13-15] have
considered the properties of wild bootstrap, but the existing theories of wild bootstrap are all
based on ordinary least squares (OLS) method they can be seriously affected in the presence