Page 280 - Contributed Paper Session (CPS) - Volume 4

CPS2224 Habshah Midi et al.
where i indexes firms and t indexes years and variables
= Gross investment,
= Market value of the firm at the end of the previous year,
= Value of the stock of plant and equipment at the end of the previous
year.

All figures in the Grunfeld data are in millions of dollars. The variables and
reflect anticipated profit and the expected amount of replacement investment
required (Greene, 2016). The Grunfeld data heavily suffers heteroskedasticity
as indicated by the Figure 1 by which, the plot of residuals shows an apparent
funnel shape form.

Figure 1: Plot of residuals for Grunfeld Data

The performance of the newly proposed TSHO method is compared to the
existing RWGM-estimator under the robust data transformations of MM-
centering. Its performance is also compared to the classical method of FGLS.
Random block leverage contamination at 5% and 10% are presented in the
heteroskedastic Grunfeld data by adding HLPs into the data set. Once
contaminated, data are ready to be transformed by either the mean or MM-
centering. The transformed data are then regressed by each of the method
under study - either WG(OLS), FGLS or the newly proposed TSHO. It must be
noted that the non-robust WG(OLS) and FGLS will only be applied to the
mean-centered data. Otherwise, robust methods will be applied to the
robustly transformed data. Results are reported in Table 1 for the beta
estimates of the original and modified Grunfeld data. FGLS should provide
more efficient estimates than WG(OLS) for the original, uncontaminated
Grunfeld data.
Results show that both WG(OLS) and FGLS provide bias and wrong results
when contamination is introduced into the panel data set. At 5%
contamination level, for both WG(OLS) and FGLS bears a negative sign which
will give a completely different interpretation from the initial uncontaminated

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