Page 281 - Contributed Paper Session (CPS) - Volume 4

CPS2224 Habshah Midi et al.
data. The results become worse as the intensity of contamination increases at
10% whereby negative signs are observed in for FGLS and for WG(OLS). The
results show that both WG(OLS) and FGLS are highly influenced by the block
HLPs which is due to their least square basis. Moreover, the mean-centering
procedure may introduce more outlying values or HLPs into the data set and
hence, the results become more distorted as the intensity of contamination
increases. On the other hand, the robust estimators – RWGM and TSHO are
able to give resistant and efficient beta estimates even though more HLPs are
introduced into the data set. By being efficient means that they are able to
provide similar results to the estimation by WG(OLS) in the original,
uncontaminated data. It is also observed that the newly proposed TSHO is
able to provide better results than RWGM as more HLPs are added into the
data at 10% contamination.
Table 1: Beta estimates of the original and modified Grunfeld data
with standard error in parentheses

Conta- WG(OLS) FGLS RWGM TSHO
mination Estimate Mean Centering MM Centering

ˆ
 3.541e-15 -0.5573 -0.4677 1.9147
0 (3.6450) (3.4372) (0.8036) (1.3450)
0% ˆ 0.1101 0.1000 0.0689 0.0636
(Original Data)  1 (0.0115) (0.0105) (0.0054) (0.0080)
0.1123
0.1212
0.3101
0.2797
ˆ
 2 (0.0170) (0.0172) (0.0071) (0.0098)
2.5580
-0.8041
2.2533
ˆ
 -2.219e-14 (6.8737) (0.7898) (1.3819)
0
(7.4440)
5%
-0.0604
-0.0519
0.0605
0.0600
ˆ
(Modified  1 (0.0096) (0.0107) (0.0051) (0.0061)
Data)
0.1175
0.1167
0.1498
0.1242
ˆ
 2 (0.0225) (0.0255) (0.0067) (0.0092)
ˆ
 5.265e-15 2.0435 -0.5590 0.4358
0 (8.7250) (8.3197) (0.9403) (1.4389)
10%
0.0663
0.0787
0.0007
-0.0128
ˆ
(Modified  1 (0.0072) (0.0087) (0.0063) (0.0076)
Data)
0.1358
-0.0086
0.0269
0.1081
ˆ
 2 (0.0160) (0.0199) (0.0082) (0.0109)

270 | I S I W S C 2 0 1 9