30
research selects REM to analyse and examines the
correlation among variables in the model.
The correlation matrix 3 shows that there is a
high correlation between ROA and ROE which
the ratio stands at 89.4%. This implies the reasons
to remove one variable from the model. When
controlling NPL, the board of management and
boardofdirectorsareoften interested inshareholder
value (ROE). ROE is always the goal that investors
always focus on (Richardson, 2002). Therefore, ROE
variable is retained. The correlation between CAR
and SIZE variables is high (70.04%). The correlation
between CAR and LEVERAGE is 67.26%. For the
SIZE variable and the LEVERAGE variable, there
is a moderately correlative relationship, which is
above the mean (59.48%). Therefore, one of the two
variables can be keep within the model. The model
of this research after removing highly correlated
variables is as follows:
NPLR = β0 + β1*LLP + β2*LDR + β3*SIZE +
β4*ROE + β5*CAR + β6*LIQUIDITY + µ
After testing the correlation, the study continues
to test autocorrelation, heterosk edasticity and
functional error (Table 3, 4):
Since the p-value of the first and second order
are both less than 5%, the hypothesis H0”There
is no correlation” is rejected (Table 4, 5). Thus,
autocorrelation occurs in the model. This makes
the regression estimate no longer BLUE (Best linear
deviation estimate).
Because of the p-value = .3637> 0.05, the
hypothesis “H0: There is no heteroscedasticity”
cannot be rejected (Table 5). Hence, the model
estimation is reliable and does not deviate.
From the results of the wrong type assay, the
p-value was 0.8238> 0.05. The hypothesis “H0: the
model is unbiased and adequate.” cannot not be
rejected. It shows that the model is unbiased and
adequate. It can be seen that the model does not
have defects in the form of functions.
After performing model tests, the random
effect model is shown with a multiplication factor
of 5.45% and ensures that the model is adequate,
unbiased, and does not have defects in the form of
functions.
The estimation model shows that 5.45% of
changes in NPL of joint stock commercial banks
are explained by LLP, LDR, Bank Size (SIZE), ROE,
CAR, LIQUIDITY, and LEVERAGE.
In the model, only statistically significant ROE
with 95% confidence is obtained due to the p-value
value < 5%. The H3_2 hypothesis is rejected, which
means that ROE affects non-performing loan. At
the same time, the negative beta coefficient shows
the opposite effect of ROE on non-performing
loan. As ROE increases, non- performing loan
decreases and vice versa. This result is consistent
with the results of Anjom and Karim (2016). When
ROE increases, the bank is operating effectively
and there are mechanisms to limit and handle
non-performing loans in the process of granting
credits to customers. According to Karapetyan
(2016), ROE have elements of risky behaviors
which means higher returns lead to higher NPLR.
However, under the current Basel II framework,
all high-risk loans require banks to make more
loan loss provisions and this will lead to a decline
in profitability as well as ROE (Kupčinskas and
Paškevičius, 2017).
The remaining variables have a p-value of 5%,
so the variables are not statistically significant at
95% confidence interval and the null hypothesis
was rejected. For LLP, there is a negative impact on
NPL and has a positive sign which means that the
loan loss provision implies much NPL. The higher
TABLE 3: SERIAL CORRELATION LM TEST WITH LAG 1
Breusch-Godfrey Serial Correlation LM Test:
F-statistic
22.86368
Prob.
F(1,102)
0.0000
Obs*R-
squared
20.50822 Prob. Chi-
Square(1)
0.0000
Source: Results from Eviews
TABLE 4: SERIAL CORRELATION LM TEST WITH LAG 2
Breusch-Godfrey Serial Correlation LM Test:
F-statistic
11.82963
Prob.
F(2,101)
0.0000
Obs*R-
squared
21.25664 Prob. Chi-
Square(2)
0.0000
Source: Results from Eviews
TABLE 5: HETEROSKEDASTICITY TEST
Heteroskedasticity Test: Breusch-Pagan-Godfrey
F-statistic
1.091337
Prob.
F(8,103)
0.3753
Obs*R-
squared
8.751737 Prob. Chi-
Square(8)
0.3637
Scaled
explained SS
27.63674 Prob. Chi-
Square(8)
0.0005
Source: Results from Eviews
