Data Analysis

and results:

This

chapter explains the empirical results of the study including the:

·

Diagnostic

tests

·

Identification

of Multicolinearity

·

Descriptive

statistics

·

Selections

and explanation of the selected model results

Now

that the data under consideration in the study is panel data; where the same

entities as in this case Afghan/Pakistani banks, are observed in time, and for

the purpose of estimation there are several methods used in panel data. The

most general and frequently used models used in the panel data are as follow:

·

Random

effect model

·

Fixed

effect model

·

Pooled

effect model (rarely used)

Now

in-order to see which model will work as best estimator and give consistent

results and will be more appropriate for the study, different tests have been

taken into account as:

·

Diagnostic

test i.e. Chow test: If the P-value is high than 0.05 then this will indicate that

the null hypothesis (H0) should be rejected, where;

H0 in the Chow

test is that the Pooled OLS is preferred over the fixed effect due to the

higher efficiency and explanatory power.

·

Breusch-Pagan

test: If the P-value is high than 0.05 then this will indicate that the null

hypothesis (H0) should be rejected where

H0 in the Breusch-Pagan

test is that the Pooled OLS is preferred over the Random effect due to the

higher efficiency and explanatory power.

·

Haushman

test: If the P-value is high than 0.05 then this will indicate that the hull

hypothesis (H0) should be rejected where;

H0 in the

Hausman test is that the Random effect is preferred over the fixed effect due

to the higher efficiency and explanatory power.

Have been used in the study and

based on them appropriate model have been considered accordingly.

First the model selection and

results of the given model of Afghanistan banks will be discussed and then on

the same manner of Pakistani banks.

Afghanistan banks:

1. Diagnostic tests for

Model selection

Diagnostics test

Null hypothesis

P value

Recommended Model

Chow Test

Pooled is Better than Fixed Effects

0.0001

Fixed Regression Model

Breusch-Pagan

Pooled is Better than Random Effects

0.0592

Pooled Regression Model

Hausman Test

Random Effects is better than Fixed Effects

0.0018

Fixed Effects Model

Based on the results of the above

diagnostic tests, Breusch-Pagan tests fails to reject null hypothesis that

Pooled is better than Fixed Effect, so recommends Pooled OLS while two other tests,

namely Chow test and Hausman test recommend that Fixed regression model will be

more appropriate and give consistent result for the study, as they both rejected

their respective null hypothesis. Based on the results of the above diagnostic

test and model selection, the model under consideration and used is Fixed

effect model for the study of determinants of bank’s profitability in Afghanistan.

2.

Variance Inflating Factor (VIF) Test:

Table ***

VIF

Variable

VIF

1/VIF

LN_TA

1.621

0.617

EQ_TA

1.655

0.604

CD_TA

2.775

0.360

CCE_TA

1.809

0.553

Ln_TA

2.579

0.388

Multicollinearity; correlation

within linear combinations of independent variables, is a situation where a

number of independent variables in a multiple regression model are

closely/highly (but not perfectly) correlated with each other.

Effects of multicollinearity:

·

Multicollinearity

doesn’t affect the consistency of the model estimates and regression

coefficients, but makes them inaccurate and unreliable.

·

It

becomes difficult to isolate to isolate the impact of each independent variable

on the dependent variable

·

The

standard error for the regression coefficients are inflated which results in

t-stats becoming too small and less powerful in terms of their ability to reject

null hypothesis.

Multicollinearity can lead to skewed (positive

or negative) or misleading results when a researcher or analyst is attempting to

determine how well each one of the individual independent variable can most

effectively be utilized to predict or understand the dependent variable in a

statistical model. In general, multicollinearity can lead to

wider confidence intervals and less reliable probability values (P

values) for the independent variables.

So

in order to make sure that our variables don’t have the issue of

multicollinearity the calculated values of

VIF should be greater than 1 and that of 1/VIF be greater than 0.1. As it

can be observed in the VIF table the calculated values in the table above

values for LN_TA, EQ_TA, CD_TA, CCE_TA, Ln_TA are all higher than 1 and their

1/VIF values higher than 0.1 in the same manner, so we can conclude that there

is no close/high correlation among the independent variables used in the model

and the results are consistent, accurate and reliable.

3.

Results and discussion:

·

Model:

Fixed-effects, using 56 observations where the data taken into account is for 8

banks in Afghanistan, out of population of 12, over 7 years

·

Included

8 cross-sectional units, sample of 8 Commercial banks in Afghanistan

·

Time-series

length = 7 (2010-2016)

·

Dependent

variable: ROA_ where ROA is used as proxy for bank profitability in

Afghanistan.

Table

*** Fixed Effect Model

ROA

Coefficient

Robust

Std. Error

T-Value

P-Value

Const.

0.00934

0.01314

0.711

0.481

LN_TA

0.0297949

0.0253057

1.177

0.2455

EQ_TA

0.0636793

0.0173056

3.680

0.0006 **

CD_TA

0.0439047

0.0219900

2.997

0.0422 *

CCE_TA

0.00423444

0.0296916

0.1426

0.8873

Ln_TA

0.00251001

0.00126129

-.990

0.0430 *

Observations

F(

12, 43)

Prob

> F

R-squared

Adjusted

R-sq

56

5.403771

0.0000

0.6013

0.49001

Regression

equation:

The regression equation

obtained for the fixed effect model after the analyses of the above data is as

follows:

Y=

? + ?1 (LN_TA) + ?2

(EQ_TA) + ?3 (CD_TA)

+ ?4 (CCE_TA) + ?5 (Ln_TA)

+ ?

Where:

·

Y=

ROA (dependent variable) proxy used for bank profitability

·

LN_TA=

Loan/Total assets

·

EQ_TA=Equity

Capital/Total Assets

·

CD_TA=Customer

deposits/Total assets

·

CCE_TA=Cash

and Cash Equivalent/Total Assets

·

Ln_TA=

natural Log of total assets the proxy for bank size

Coefficients:

The

coefficients (the ?s) show the nature of relationship and the extent of change caused

in the dependent variable (in this case the ROA) by the independent variables

(in this case five independent variables used). In the coefficients table the ?

shows the amount of percentage change and the sign shows the positive or

negative impact on the dependent variable. The table shows that if all the

independent variables are zero still Afghan banks will be able to be somehow

profitable where profitability of a certain bank will be around 0.00934 units

for each unit of assets deployed in the given bank. Further it explains that

all the variables i.e. capital adequacy, size, deposits, liquidity and asset

composition, have positive impact on the overall profitability of Afghanistan

banks.

F-

Test (Test of over-all model significance):

F

test is a part of ANOVA table which shows the significance of the over-all

model. Now the decision rules based on F-test are as follow:

·

If the F-test calculated

value is more than the F test tabulated value i.e. Fcal >

Ftab, then we can conclude that the over-all model is significant

·

If the F-test

calculated value is lower than the F-test tabulated value i.e. Ftab >

Fcal, then we can conclude that the over-all model is insignificant

Now

based on the above ANOVA table, calculated value for the F-test is 5.40 which

is greater than 4 (tabulated value), therefore we can concluded that the model

as whole is significant. The p-value also approves this fact as its value is

less than 0.05 i.e. (0.0000) F

R-squared

Adjusted

R-sq

84

17.45435

0.0000

0.806509

0.760302

Regression

equation for Pakistan banks:

The regression equation

obtained for the fixed effect model after the analyses of the above data is as

follows:

Y=

? + ?1 (LN_TA) + ?2

(EQ_TA) + ?3 (CD_TA)

+ ?4 (CCE_TA) + ?5 (Ln_TA)

+ ?

Where:

·

Y=

ROA (dependent variable) proxy used for bank profitability

·

LN_TA=

Loan/Total assets

·

EQ_TA=Equity

Capital/Total Assets

·

CD_TA=Customer

deposits/Total assets

·

CCE_TA=Cash

and Cash Equivalent/Total Assets

·

Ln_TA=

natural Log of total assets (proxy for bank size)

Coefficients:

The

table shows that if all the independent variables are zero still Pakistani

banks will be able to be somehow profitable where profitability of a certain

bank will be around 0.00500134 units for each unit of assets deployed in the

given bank.

Further it explains that two variables namely

capital adequacy and deposits have positive impact while rest three variable;

liquidity, bank size and asset composition, have negative impact on the overall

profitability of Pakistani banks.

F-

Test (Test of over-all model significance):

Now

based on the above ANOVA table, calculated value for the F-test is 17.45435 which is greater than 4 (tabulated value),

therefore we can concluded that the model as whole is significant. The p-value

also approves this fact as its value is less than 0.05 i.e. (0.0000)