The comparison of the board impacts on the FP of two banks’ groups remains a restricted task to be generalized for three main reasons. First, the lack of comprehensive international data related to governance and performance at the same time prevents analysts from deepening their theoretical propositions of assumptions and their empirical interpretations. Therefore, the various difficulties are transformed into several consequences generating many problems. The causes of the birth and the evolution of the problems come back mainly to the decrease of the governance mechanism quality or the decline of the banking performance. Then, the generalization of its results is difficult because of the proportional impact of vulnerable economic events. Finally, phenomena related to the banking governance quality and FP that have occurred on the financial market are not planned by the same techniques. Moreover, the corrective actions of their results and their consequences are not addressed with the same methods. Empirical methodology in the research sphere is a very complicated approach. It is based on the theoretical justification of the most appropriate and effective systematic methods. However, the theoretical proofs are sometimes non-existent with the topic to be discussed or they may be unavailable in reality. There are different types of paradigms and research approaches that are possible. Among them we have chosen the demonstrative approach, considering as it is the most appropriate to our current study.
3.1. Methodological aspects
The methodology applied in our exploratory study is a demonstrative comparison by resorting to modeling. The data analysis for this study focused on associations between mechanisms, relationships between shareholders, and individual behaviors to explain correlations between the different stakeholders. This helps to identify the impact factors affecting the relationships between basic FP measures and the influences due to the BOD. The research plan to be followed to answer the questions already mentioned began with the clarification of the data sources, then we quoted the variables to be modeled, finally, we exposed our object models.
3.1.1. Data collection
Two samples were taken from two reference populations. The choice of banks is limited to countries whose banking systems incorporate both Islamic and conventional banks over the period 2010–2018 regardless of the proportion of each model in each country’s banking market. These populations are made up of 2,974 conventional financial institutions and 683 Islamic financial institutions. The countries part of our study are USA, France, Singapore, Algeria, Thailand, India, Egypt, Bangladesh, Indonesia, Pakistan, Tunisia, Malaysia, Canada, Sudan, Turkey, United Kingdom, Luxembourg, Bahrain, Jordan, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates, South Africa, Senegal, Nigeria, Sri Lanka, Kazakhstan, and Lebanon. However, we have excluded all specific financial institutions subject to particular regulations. The tested samples include only purely conventional or Islamic banks. Besides, due to difficulties in collecting information on FP and BOD, we excluded banks marked by some missing observations, variables or data. We also removed the multi-type mutated banks (Islamic-conventional window banks and conventional-Islamic window banks). These three conditions led us to eliminate 571 conventional financial institutions and 2862 Islamic financial institutions. Subsequently, we have reduced the banks’ number remaining for each bank type based on qualitative and quantitative filtering criteria (samples equality, activity type, similarity of origin country, bank width), each CB has its Islamic equivalence in terms of capital and size taken from the same country. This restriction reduced the size of our samples to 112 banks each. Finally, after several elimination and deletion steps, we obtained two pairs of equal samples (n1 = n2).
3.1.2. The measurement of the variables to be tested
3.1.2.1. Endogenous variables
In this sub-section, we presented FP measures. The main variable to explain was represented by four dependent variables: profitability, efficiency, liquidity, and solvency. Table 1 shows the parameters we worked on, the symbols and the relative reports.
Table 1
Description of variables to explain
FP measurement | Rating for CB | Rating for IB | Measurement | Previous studies |
Profitability ratio | Rtc | Rti | Marginal Profit / Total Revenues | Sujan et al. (2013);Atyeh et al. (2015); Ogbeide and Akanji (2017); Haddad et al. (2019b). |
Liquidity ratio | Ltc | Lti | Net Loans / Total Assets | Olson and Zoubi (2008); Onakoya and Onakoya (2013); Osama et al. (2013); Haddad et al. (2020). |
Efficiency ratio | Etc | Eti | Operating result / Average Total Assets | Moin (2008); Emilia and Judit (2012); Rashid and Khaleequzzaman (2015); Haddad et al. (2019a). |
Solvency ratio | Stc | Sti | Total Loans / Total Deposits | Olson and Zoubi (2008); Onakoya and Onakoya (2013); Ola and Suzanna (2015); Haddad et al. (2019c). |
3.1.2.2. Exogenous variables
Throughout the remaining part of this work, banks’ FP is explained by four determinants of BOD. Referring to the review of the previous literature, the predominantly independent variables have been described in Table 2 as follows:
Table 2
Description of the explanatory variables
The internal governance mechanism | Rating for CB | Rating for IB | Measurement | Previous studies |
Board of directors | Board size of CB (TCONSc) | Board size of IB (TCONSi) | The number of directors in the BOD | Uwuigbe and Fakile (2012); Aebi et al. (2012); Fanta et al. (2013). |
Rooting of the board chairman or accumulation of the post of CEO and theboard chairman (ENRADIRc) | Rooting of the board chairman or accumulation of the post of CEO and the board chairman (ENRADIRi) | Binary variable: 1: if the CEO also holds the post of board chairman of the bank or the board chairman exceeded the mandate 0: if not | Coleman and Biekpe (2007); Al-Hawary (2011); Hoque and Muradoglu (2015). |
Board independence: presence of external directors in the BOD (INDCONSc) | Board independence: presence of external directors in the BOD (INDCONSi) | Number of directors who are not related to any professional/ family relationship, nor the bank nor the executives, except for their roles in the board on the total number of the BOD | Rachdi and Ameur (2011); Pan (2014); Hoque and Muradoglu (2015). |
Number of meetings held by the CB’s board (REUCONSc) | Number of meetings held by the IB’s board (REUCONSi) | Number of meetings held by the BOD in a year. | Chen et al. (2006); Choi and Lai (2014); Thu et al. (2016). |
3.1.2.3. Measurements of control variables
Table 3 displays the list of control variables supported by some previous studies that employed the same variables and their measures.
Table 3
Description of control variables
Control variable | Rating for CB | Rating for IB | Measurement | Previous studies |
BankType | TYc | TYi | A qualitative variable takes three forms: 1: if the bank is a commercial bank 2: if the bank is an investment bank 3: if the bank is a universal bank | Cornett et al. (2009); Kim and Rasiah (2010); Farazi et al (2011); Charles et al. (2015). |
Bank Age | AGc | AGi | Age of conventional / Islamic bank for each year | Jemric and Vujcic (2002); Jeff et al. (2010a); Filip et al. (2013); Arif et al. (2017). |
Bank Size | TAc | TAi | Logarithm of book value of total assets of conventional / Islamic bank at the end of each year | Rachdi and Ameur (2011); Berger et al. (2014); Saha et al. (2015); Rashid and Jabeen (2016) |
Inflation | INFc | INFi | The rate of inflation in the country of origin of the conventional / Islamic bank object of study | Gul et al. (2011); Rashwan and Ehab (2016); Nahar and Sarker (2016); Tugba et al. (2017). |
3.1.3. Presentation of models to estimate
Before proceeding to the estimations, it is necessary to present the typical models to reassess several times the FP and each time the dependent variable will be changed according to the FP measures and the bank type.
Conventional models of multiple regressions are of the following form:
Islamic models of multiple regressions are of the following form:
3.2. Econometric validation of models
The FP of conventional and Islamic banks depends on the systematic use of explanatory variables. Interdependence between the board determinants forced us to test also the correlation between the explanatory variables one by one.
3.2.1. Error Heteroscedasticity Test
Heteroscedasticity qualifies data do not have constant variances; the objective of this test is to know if the errors’ variance for each individual is constant. This situation is frequently encountered in the case of panel data. It is therefore important to know how to detect heteroscedasticity before correcting it. Heteroscedasticity does not generally bias the coefficient estimate, but it remains uncommitted because of the poor quality of the standard deviations.
To identify heteroscedasticity in linear regression, there is a range of specific tests designed mainly for this reason, among which we mention the most common name the Cook-Weisberg test, Breusch-Pagan test, White test, Modified Wald test, Goldfeld test, Gleisjer test, etc. In our study, we chose the Breusch-Pagan test and the Modified Wald test to check for this type of problem.
If the probability associated with the test is below the tolerance threshold (5%), we reject the hypothesis of heteroscedasticity (H0). Nevertheless, if the probability is greater than (5%), the null hypothesis will be retained, it is possible in this case to confirm the heteroscedasticity of the residues.
Before starting the analysis of the results, first, we referred to the agreement of choice between the applied tests. Depending on whether it is a fixed or random-effect, for models that are confirmed fixed, such as those affiliated with the efficiency and solvency of IBs (Table 5), we compared these models to the Modified Wald test. Nevertheless, in the case of models relating to the profitability, the efficiency, the liquidity and the solvency of CBs (Table 4) and the models specific to the profitability and liquidity of IBs (Table 5), the Breusch-Pagan test proves more appropriate to detect heteroscedasticity.
Table 4 illustrates the heteroscedasticity test results of CBs for random-effect models (LnRtc, Etc,Ltc, and LnStc). For this effect type, the random-effects models have highlighted the need for the Breusch-Pagan test. We speak of heteroscedasticity when the magnitude of the error risk is constant over time. The χ2 statistics of these models showed risks associated with the rejection of the heteroscedasticity hypothesis well below the desired threshold of 5%. The probabilities of χ2 test of banks' profitability, efficiency, liquidity and solvency models are lower than the minimum risk acceptance rate of the null hypothesis, leading to the rejection of the proposal of heteroscedasticity problems at the level of these models. Therefore, we concluded that the CBs’ models are not heteroscedastic.
Table 4
Heteroscedasticity tests of the CBs’ sample
Model type | Modified Wald Test | Breusch- Pagan Test | χ2 | Prob > χ2 | Heteroscedasticity |
LnRtc | - | Test of Breusch- Pagan | 11.69 | 0.0003 < 5% | Absence of the heteroscedasticity problem |
Etc | - | Test of Breusch-Pagan | 172.06 | 0.0000 < 5% | Absence of the heteroscedasticity problem |
Ltc | - | Test of Breusch-Pagan | 224.67 | 0.0000 < 5% | Absence of the heteroscedasticity problem |
LnStc | - | Test of Breusch-Pagan | 49.80 | 0.0000 < 5% | Absence of the heteroscedasticity problem |
Similarly, Table 5 showed the heteroscedasticity tests’ results of the IBs sample for fixed-effects models (Eti and LnSti) as well as for random-effects models (LnRti and Lti). Furthermore, the Modified Wald test gives a probability of the χ2 lower than the predetermined threshold of 5%, the fixed-effects Eti and LnSti models recorded values of χ2 equal to (0.0000). Therefore, we rejected the heteroscedasticity problems hypothesis, the variance of errors in these two models is not the same for all IBs.
Moreover, results of the Breusch-Pagan test revealed χ2statistics below 5% for the models assumed until now random, more specifically, the probabilities of the models relating to the profitability and the liquidity of IBs equal respectively to (0.0006) and (0.0000). Therefore, we have been able to reject the null hypothesis, LnRti and Lti models are not heteroscedastic and the variance of their errors is not the same for all IBs. As well as we have deduced that our data do not have a heteroscedastic structure.
Table 5
Heteroscedasticity tests of the IBs’ sample
Model type | Modified Wald Test | Breusch-Pagan Test | χ2 | Prob > χ2 | Heteroscedasticity |
LnRti | - | Breusch- Pagan Test | 10.45 | 0.0006 < 5% | Absence of the heteroscedasticity problem |
Eti | Modified Wald Test | - | | 0.0000 < 5% | Absence of the heteroscedasticity problem |
Lti | - | Breusch- Pagan Test | 194.48 | 0.000 < 5%0 | Absence of the heteroscedasticity problem |
LnSti | Modified Wald Test | - | | 0.0000 < 5% | Absence of the heteroscedasticity problem |
In the case of fixed-effects models, whatever the model is heteroscedastic or not, if it does not contain individual effects, reasoning procedure requires the direct navigation to analyze the correlation. However, with a random-effect model, we must check before, if the square of the residuals can be explained by the model variables. In this case, we confirm that there is a problem of heteroscedasticity. The choice of one test or another depends on the need, the type of variables and the econometric effect of the models.
3.2.2. Residual autocorrelation test
The autocorrelation of residues test is used to detect for each individual in the sample whether the errors of a period are influenced by the errors of the previous period. The autocorrelation hypothesis of the residues is a necessary condition before the validation of the estimated results. It also facilitates the choice of the best modeling method that lends itself to our data after the classification of existing effects.
There are several autocorrelation errors’ tests, among which we have chosen the Wooldridge test and the Durbin-Watson test. Each of them is employed in a well-defined econometric situation. In our research, we have insisted on the most often known: the Wooldridge test is generally the most suitable when the model is subjected to a fixed-effect, however, the Durbin-Watson test is the most applicable in the case of random-effects models because of its ability to detect particular forms of autocorrelation.
The hypotheses of the autocorrelation test are as follows:
H0: The errors are not autocorrelated.
H1: The errors are autocorrelated.
The decision rule of this test is as follows: if the risk obtained is greater than the critical value (Prob > F)<(5%), we reject the null hypothesis, in this case, the errors of the individuals (conventional or Islamic bank) are considered autocorrelated and vice versa.
The results of the autocorrelation tests for residues are shown in Tables 6 and 7 below:
Table 6
Autocorrelation tests of the CBs’ sample
Model type | Wooldridge Test | Durbin Watson Test | F of Fisher | Prob > F | Autocorrelation | Decision |
LnRtc | - | Durbin Watson Test | 3.172 | 0.0841 > 5% | Absence of autocorrelation | Random-effect |
Etc | - | Durbin Watson Test | 16.518 | 0.0003 < 5% | Presence of autocorrelation | Fixed-effect |
Ltc | - | Durbin Watson Test | 56.355 | 0.0000 < 5% | Presence of autocorrelation | Fixed-effect |
LnStc | - | Durbin Watson Test | 0.032 | 0.8583 > 5% | Absence of autocorrelation | Random-effect |
For the models provisionally classified with fixed-effects, the (Wooldridge, 2002) test results referring to the efficiency and the solvency of IBs (Table 7) have shown two values greater than 5%. They are illustrated with respective risks equal to (0.000) for Eti and LnStc. Since the risk of rejecting the null hypothesis is not high, therefore, it was rejected to confirm that the errors are autocorrelated. Based on these results, we concluded that their errors are autocorrelated, hence, we have rectified the class of these two models to random-effects.
After the Durbin Watson test, Fisher statistics on the profitability and the solvency models of CBs (Table 6) and the liquidity model of IBs (Table 7) generated p-values strictly greater than 5%, the risks being respectively equal to (0.0841) for LnRtc, (0.8583) for LnStc and (0.5246) for Lti. For this reason, we confirmed the absence of autocorrelation of errors, which indicates that in these models, the errors are dependent on each other. Therefore, these models have purely random-effects. Furthermore, for the specific model to the efficiency and the liquidity of CBs and the profitability of IBs recorded risks of less than 5%, equal to (0.0003) for Etc, (0.0000) for Ltc, and (0.0338) for LnRti respectively. The Durbin Watson test displayed a rate of risk null, which indicates the presence of autocorrelation between the errors. Hence, we have rectified the class of these two models to fixed-effects.
Table 7
Autocorrelation tests of the IBs’ sample
Model type | Wooldridge Test | Durbin Watson Test | F of Fisher | Prob > F | Autocorrelation | Decision |
LnRti | - | Durbin Watson Test | 4.889 | 0.0338 < 5% | Presence of autocorrelation | Fixed-effect |
Eti | Wooldridge Test | - | 74.399 | 0.0000 < 5% | Presence of autocorrelation | Random-effect |
Lti | | Durbin Watson Test | 0.412 | 0.5246 > 5% | Absence of autocorrelation | Random-effect |
LnSti | Wooldridge Test | - | 70.944 | 0.0000 < 5% | Presence of autocorrelation | Random-effect |
Given these results, 3 final models took the form of heterogeneous panels with fixed-effects that vary from one individual to another (Etc, Ltc, and LnRti) and 5 models took the final form of heterogeneous panels with random-effects (Etc, LnStc, Eti, Lti, and LnSti). The second models’ types assume that the relationships between the FP measurement of conventional or Islamic banks and the ownership structure determinants in each sample are not identical for all individuals of the same sample.
Apart from the distinguishing features and differences between the organizational structures of Islamic and conventional banks, there are financial peculiarities that separate each banking model from the other and prevent the merger and proximity of the two models (Nganga, 2013). Each CB in our first sample is governed by its own effect regardless of individual temperaments, conflicts of individual interests, internal or external environmental factors of the bank, location of governance mechanisms, regions, etc. Similarly, identifying the effects associated with IBs implies that each of them is governed by a varying effect among banks, but it is unchangeable over time.
3.2.3. Verification of multicollinearity problems
The multicollinearity test is performed to prevent the instability risk of the coefficients estimated by the OLS method. It also makes it possible to see if the matrix of the exogenous variables is regular. Any linear regression calls for the presence of collinearity and multicollinearity problems will be integrated into the same model, between the exogenous variables.
To verify the degree of correlation of the independent variables between them, when the majority of the explanatory variables did not satisfy the normality condition, we examined Spearman’s correlation coefficients, the non-parametric version of Pearson (Ricco, 2015). The solution consists in eliminating the collinear variable or the block of exogenous variables containing the same types of information, affecting the quality of the regression. The correlation that deviates from the significance level of the Spearman test meaning is considered useless in the model.
Tables 8 and 9 illustrate the matrices of Spearman's correlation coefficients measuring the degree of linear connection, on the one hand, between the variables to be explained and the explanatory variables, and on the other hand, between the explanatory variables between them. Table 8 displayed the CBs’ Spearman matrix, whereas Table 9 displayed the IBs’ matrix. Each coefficient is between − 1 and + 1, the sign takes into account the meaning of the relationship.
There is a collinearity between two independent variables when the correlation between them is greater than 0.8 (Gujarati, 2004). The appearance of collinearity makes the signs and values of the estimated coefficients seem contradictory; in this case, their variances are unstable. Since collinearity leads to redundant estimates and misleading significance, it must be detected by performing the necessary corrections and treatments before completing the analysis of the results.
Correlation measures the intensity of the relationship between variables. The link strength between the variables is classified into three types, there is a strong, medium or weak correlation. It is also positive if the correlation coefficient is positive and negative if this coefficient is negative. From Table 8, analysis of the Spearman correlation matrix of the CBs’ sample revealed that the signs of the explanatory variables on the FP vary from one model to another. As a result, the separate interpretation of each model gave us a clear idea of the individual effect of each variable.
Beginning with the CBs’ profitability (LnRtc), the matrix conceded that (LnAGc) and (LnTAc) have acted positively on the CBs’ profitability. This leads us to argue that the more the effects of the cited variables develop, the better the CBs’ profitability will progressively improve. However, (LnTCONSc), (LnREUCONSc), and (LnINFc) are negatively affected the CBs’ profitability. This led us to conclude that the greater the impacts of these variables are, the more the CBs’ profitability deteriorates.
The shift to efficiency analysis (Etc) revealed that only variables related to (TYc), (LnAGc), and (LnINFc) are positively correlated with the CBs’ efficiency. This has indicated that as the value of these variables grows, more CBs will become more efficient. Nevertheless, (LnTCONSc), (ENRADIRc), (LnINDCONSc), (LnREUCONSc), and (LnTAc) reflected the negative and destructive impacts of the bank efficiency.
Concerning the liquidity (Ltc), the correlation matrix arrangement found that (LnREUCONSc) and (LnAGc) are positively correlated with the CBs’ availability. This summarizes that the more the tendency of these variables is evaluated, the more CBs increase their wealth. However, other variables demonstrated the opposite, an inverse impact was generated by (ENRADIRc), (TYc), (LnTAc), and (LnINFc), they generally recorded negative effects on bank performance and specifically their liquidity powers. This explains why any improvements in these variables lead to the weakening of bank liquidity.
Similarly, after an inventory of solvency (LnStc), we attributed to the conclusion that (LnREUCONSc) supported the CBs’ solvency. The more this variable has protected the CBs’ solvency, the more its effects are propagated within the decision-making nodes, the more they guarantee its solvency capabilities. However, (LnTCONSc), (TYc), (LnAGc), (LnTAc), and (LnINFc) harmed the sustainability of the bank solvency, therefore, the evolution of the impact associated with these variables contributes, no doubt, to the deterioration of the CBs’ solvency.
Finally, the analysis of the correlation between the explanatory variables did not reveal any inter-variable correlation coefficient greater than 0.8. For this, we admitted the absence of collinearity problems in all the models reasoning the board quality’s impacts on the FP parameters.
Table 8
Spearman correlation matrix of the CBs’ sample
| LnRtc | Etc | Ltc | LnStc | LnTCONSc | ENRADIRc | LnINDCONSc | LnREUCONSc | TYc | LnAGc | LnTAc | LnINFc |
LnRtc | 1.0000 | | | | | | | | | | | |
Etc | 0.2982* 0.0000 | 1.0000 | | | | | | | | | | |
Ltc | -0.0578 0.3344 | -0.1086* 0.0742 | 1.0000 | | | | | | | | | |
LnStc | -0.0434 0.4683 | -0.0862 0.1572 | 0.3861* 0.0000 | 1.0000 | | | | | | | | |
LnTCONSc | -0.2904* 0.0005 | -0.3677* 0.0069 | -0.0305 0.5900 | -0.1165* 0.0021 | 1.0000 | | | | | | | |
ENRADIRc | 0.0536 0.3704 | -0.2231* 0.0428 | -0.1106* 0.0298 | -0.0243 0.6672 | -0.2200* 0.0001 | 1.0000 | | | | | | |
LnINDCONSc | 0.0113 0.8780 | -0.3746* 0.0000 | -0.0295 0.6652 | 0.0261 0.7025 | 0.4087* 0.0000 | 0.1060 0.1195 | 1.0000 | | | | | |
LnREUCONSc | -0.3737* 0.0000 | -0.2919* 0.0013 | 0.1729* 0.0021 | 0.1058* 0.0066 | 0.0895 0.1128 | -0.2472* 0.0000 | -0.0974 0.1525 | 1.0000 | | | | |
TYc | -0.0761 0.2034 | 0.1070* 0.0008 | -0.1132* 0.0012 | -0.1924* 0.0034 | -0.2336* 0.0000 | -0.0454 0.4221 | 0.0722 0.2894 | -0.1770* 0.0016 | 1.0000 | | | |
LnAGc | 0.1576* 0.0081 | 0.2050* 0.0007 | 0.1556* 0.0057 | -0.1788* 0.0132 | 0.0425 0.4520 | 0.2036* 0.0003 | 0.0541 0.4276 | 0.1101* 0.0509 | -0.1208* 0.0321 | 1.0000 | | |
LnTAc | 0.2387* 0.0014 | -0.2091* 0.0005 | -0.1989* 0.0025 | -0.1924 0.0016 | 0.3212* 0.0000 | 0.0528 0.3507 | 0.3003* 0.0000 | 0.2056* 0.0002 | -0.2402* 0.0000 | 0.0806 0.1537 | 1.0000 | |
LnINFc | -0.2475* 0.0000 | 0.1656* 0.0072 | -0.3022* 0.0000 | -0.1667 0.0005 | 0.1209* 0.0348 | -0.1484* 0.0094 | -0.2798* 0.0000 | 0.0943 0.1001 | -0.0404 0.4826 | -0.2928* 0.0000 | -0.2047* 0.0003 | 1.0000 |
Based on Table 9, an overview of the Spearman correlation matrix that collects with the IBs sample showed a heterogeneous mixture of the various signs, sometimes the variables’ signs coincide with those detected by its conventional counterparts and sometimes they diverge. In what follows, we have listed the conclusions obtained on the effect of each variable on each FP measure.
First, the analysis of the IBs’ profitability (LnRti) highlighted the presence of three variables playing an important role in the process of creating profitability; (LnINDCONSi), (TYi), and (LnAGi). However, (LnTCONSi), (LnREUCONSi), (LnTAi), and (LnINFi) revealed an opposite effect, the more the impacts of these variables grow, the more the IBs’ profitability decreases.
Then, the correlation coefficients relative to the IBs’ efficiency (Eti) signed a very important cumulation of the positive effects generated by some variables mainly due to (LnTCONSi), (ENRADIRi), (LnAGi), and (LnTAi). This has proven that the more the effect of its organizations extends, the more IBs improve its short-term returns and the more they guarantee its sustainable returns.
Turning to the impact of the explanatory variables on the IBs’ liquidity (Lti), we pointed out that (ENRADIRi), (LnINDCONSi), and (LnTAi) played a favorable role of protecting their liquidity capacity. These mechanisms have all been factors for improving monetary dependence. On the contrary, (LnTCONSi), (LnAGi), and (LnINFi) have an adverse effect on the effectiveness of IBs’ liquidity.
Finally, we looked at the dependency analysis between solvency (LnSti) and another explanatory variable, the Spearman test revealed a positive correlation between (LnINDCONSi)and IBs’ solvency. In contrast, Spearman matrix testified to the presence of a considerable but suspicious cumulative effect on the part of (LnTCONSi) and (LnINFi). This led us to conclude that the more significant its effects, the more IBs will keep its solvencies.
The correlation coefficients between the independent variables of the 2nd matrix revealed that the intersection between all the variables did not generate any correlation coefficient greater than 0.8. Hence, there are no collinearity problems between the variables.
Table 9
Spearman correlation matrix of the IBs’ sample
| LnRti | Eti | Lti | LnSti | LnTCONSi | ENRADIRi | LnINDCONSi | LnREUCONSi | TYi | LnAGi | LnTAi | LnINFi |
LnRti | 1.0000 | | | | | | | | | | | |
Eti | 0.2356* 0.0001 | 1.0000 | | | | | | | | | | |
Lti | 0.0461 0.4528 | -0.0040 0.9430 | 1.0000 | | | | | | | | | |
LnSti | 0.0134 0.8269 | -0.2889* 0.0000 | 0.4809* 0.0000 | 1.0000 | | | | | | | | |
LnTCONSi | -0.1452* 0.0574 | 0.3316* 0.0002 | -0.3234* 0.0292 | -0.4313* 0.0004 | 1.0000 | | | | | | | |
ENRADIRi | -0.1280 0.3363 | 0.1292* 0.0028 | 0.1691* 0.0026 | 0.0408 0.4711 | -0.0971* 0.0853 | 1.0000 | | | | | | |
LnINDCONSi | 0.2471* 0.0058 | -0.0863 0.2162 | 0.3133* 0.0003 | 0.2791* 0.0008 | 0.1473* 0.0342 | -0.0987 0.1570 | 1.0000 | | | | | |
LnREUCONSi | -0.2682* 0.0635 | -0.0431 0.4462 | 0.0184 0.7446 | 0.0171 0.7627 | 0.1881* 0.0008 | -0.1585* 0.0048 | 0.2203* 0.0014 | 1.0000 | | | | |
TYi | 0.2356* 0.0005 | 0.0665 0.2389 | 0.1769* 0.0016 | -0.0217 0.7006 | -0.1168* 0.0383 | 0.0426 0.4515 | 0.3381* 0.0000 | -0.1607* 0.0042 | 1.0000 | | | |
LnAGi | 0.1828* 0.0027 | 0.3297* 0.0000 | -0.2425* 0.0655 | -0.0620 0.2741 | 0.0759 0.1807 | 0.1679* 0.0029 | -0.1635* 0.0189 | 0.0248 0.6627 | -0.2066* 0.0002 | 1.0000 | | |
LnTAi | -0.3227* 0.0413 | 0.1179* 0.0279 | 0.4264* 0.0000 | 0.0526 0.3811 | 0.2155* 0.0001 | 0.1591* 0.0050 | 0.1056 0.1300 | 0.0716 0.2086 | -0.1476* 0.0093 | 0.3178* 0.0000 | 1.0000 | |
LnINFi | -0.2621* 0.0000 | 0.0098 0.4360 | -0.2851* 0.0000 | -0.2538* 0.0000 | 0.1285* 0.0248 | -0.0459 0.4241 | -0.2458* 0.0005 | 0.1232* 0.0314 | -0.0201 0.7266 | 0.1029* 0.0738 | -0.1956* 0.0007 | 1.0000 |
4.