Page 97 - CITN 2017 Journal

Diagnostic Test on the Variables

The Sargan test of over-identifying restrictions, suggesting whether the instrumental
variables and residuals are independent was conducted. This is to check for the validity of
the specification of the instrumental variable used in the GMM estimation; the Sargan test
will be implemented. Sargan Test as developed by Sargan (1958) was employed to test for
over-identifying restrictions for one-step and two-step GMM respectively. Hence, the first
and second order autocorrelation disturbance was also confirmed using the p-values of AR
(1) and AR (2) using Arellano-Bond Serial Correlation Test.

The probability value of J-statistics of 0.548362 is insignificant indicating that the
instruments are valid and the GMM estimates are reliable. To estimate Sargan test of over-
identifying restriction, the formula “Scalar pval = @chisq (J-statistic, instrument rank-
number of parameters estimated” must gives the same probability value. For the stated
model, the Scalar P-Value was computed using ScalarPval = @chisq (34.32538, 45-9).
The resultant value is the same P-Value as reported for J-statstics shown in Table 4. This
indicates the validity of the instrument used.In order to confirm the efficiency of the first
difference GMM estimator was computed and the soundness of the result with AR (1) with
p-value of 0.0001 which is significant and AR (2) with a p-value of 0.2455 which is not
significant, hence, there is no serial correlation in residual.

Test of Stationarity
Before estimating the model, the dependent and independent variables are separately
subjected to some stationary tests using unit root test since the assumption for the classical
regression model require that both variables be stationary and that errors have a zero mean
and finite variance. The formulated model was tested for stationarity using the using Levin,
Lin & Chut, Im, Pesaran & Shin W-stat, Augmented Dickey-Fuller- Fisher Chi-Square and
Phillips-Perron (PP) - Fisher Chi-Square Unit root test to be sure that one is not analysing
inconsistent and spurious relationship. A series that exhibit a stochastic trend, or even
simply wanders around at random will not be stationary and cannot be forecast far into the
future. A stationary series will constantly return to a given value and no matter the starting
point, in the long-run, it is expected to attain that value (Hall, 1994). The Table 4 shows that
all the variables were at level

Table 4: Summary of Panel Unit Root Test Results

Levin, Lin Im, Pesaran& ADF - Fisher PP - Fisher
Variables Status
&Chut Shin W-stat Chi-Sq Chi-Sq
-7.58257*** -2.75222*** 141.270*** 174.627***
TOBINQ 1(0)
(0.0000) (0.0030) (0.0042) (0.0000)
-1.95795** -1.29503* 123.257** 212.267***
ETR 1(0)
(0.0251) (0.0977) (0.0573) (0.0000)
-5.25813*** 1.33844* 121.210* 163.248***
LEV 1(0)
(0.0000) (0.0904) (0.0733) 0.0001
-52.2878*** -9.42506*** 174.519*** 177.961***
ROA 1(0)
(0.0000) (0.0000) (0.0000) (0.0000)
-18.4810*** -6.21517*** 56.020*** 214.614***
SIZE 1(0)
(0.0000) (0.0000) (0.0003) (0.0000)
-28.9996*** -7.02961*** 180.088*** 180.677***
LIQ 1(0)
(0.0000) (0.0000) (0.0000) (0.0000)

90