From the initial regression model, we have found that two of the correlation coefficients is significant at 1% significant level. Beside this coefficient of regression is not very high. So as the next process of analysis we are going to test whether the data has shown the heteroskedasticity and multicollinearity test.

## Graph 1: Residual Plot Test

As the process of checking heteroskedasticity we have relied on the Bruesh-Pegan Test. From the Bruesh-Pegan test result we see that there is no heteroskedasticity as the probability of chi squared value is greater than 0.10.

**Table 3: Bruesh-Pegan Test**

## Multicollinearity Test Result

We are now concentrating to test the multicollinearity test of the data. As the process of multicollinearity test the Pearson correlation power test is used. From the correlation matrix we see that there is high correlation between the two in dependent variables namely the Total Liabilities and ROA. So, there is the problem of multicollinearity.

**Table 4: Power Correlation Test**

## Model Transformations

As there is multicollinearity problem but no heteroskedasticity problem so the model required to be transformed. As the process of model transformation and speciation we are now relied on the dropping variable process and run the regression. Followings are some results of data dropping scenario and regression result.

## Model Transformation # 1

Dropping the total liability, we estimate the following linear regression model:

## LCij = α + β1NPij + β2ROEij + β3NWCij + β4TAij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 5

Variable | Pooled Ordinary Least Square | Random Effect Model |

NP | -0.04817** (1.1804) | 0.2991* (0.0341) |

ROE | 0.6184 (0.1588) | -0.5184 (0.1566) |

NWC | 0.2258 (1.2314) | 0.3287 (1.3455) |

TA | -0.0017 (0.0591) | -02451 (0.1989) |

**1% Significance

*5% Significance

From the regression model, we have found that only one of the coefficients is significant at 1% significant level. this coefficient of regression is not very high. Regression coefficient is also very low.

*Model Transformation # 2*

Dropping the return on equity, we estimate the following linear regression model:

## LCij = α + β1NPij + β2NWCij + β3TAij + β4TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 6

Variable | Pooled Ordinary Least Square | Random Effect Model |

NP | -0.3355** (0.0879) | 0.1850** (0.0489) |

NWC | 1.0461 (0.3258) | -1.2555 (0.3145) |

TA | 2.3554* (0.8411) | 2.1245 (0.8867) |

TL | -7.1208 (5.3495) | -7.6841 (4.8756) |

**1% Significance

*5% Significance

From the regression model, we have found that only two of the coefficients is significant. Beside this coefficient of regression is not very high. Regression coefficient is also very low. Beside this model is not significant.

*Model Transformation # 3*

Dropping return on equity and total assets, we estimate the following linear regression model:

## LCij = α + β1NPij + β2NWCij + β3TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance. ‘

Table 7

Variable | Pooled Ordinary Least Square | Random Effect Model |

NP | -0.0374 (0.7126) | 0.0025 (0.0760) |

NWC | -0.5705 (0.4.3191) | -0.0067 (0.2587) |

TL | -4.3694 (4.0074) | -3.5487 (4.1578) |

**1% Significance

*5% Significance

From the regression model, we have found that none of the coefficient is significant. Beside this coefficient of regression is not very high. Regression coefficient is also very low.

*Model Transformation # 4*

Dropping net profit margin and total assets, we estimate the following linear regression model:

## LCij = α + β1ROEij + β2NWCij + β3TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 8

Variable | Pooled Ordinary Least Square | Random Effect Model |

ROE | -0.3476 (1.1322) | 0.4870 (0.6860) |

NWC | 0.0125 (0.0129) | -0.0769 (0.0037) |

TL | -0.0532 (0.0214) | 0.0396 (0.9906) |

**1% Significance

*5% Significance

From the regression model, we have found that none of the coefficient is significant. Beside this coefficient of regression is now high. Regression coefficient is also high now.

*Model Transformation # 5*

Dropping the profitability variable, we estimate the following linear regression model:

## LCij = α + β1ROEij + β2NWCij + β3TAij + β4TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 9

Variable | Pooled Ordinary Least Square | Random Effect Model |

ROE | -0.3401 (0.1.132) | 0.2770** (0.0240) |

NWC | -0.1393 (3.5629) | -0.2223 (0.9875) |

TA | 1.7533* (0.6493) | -1.2856 (0.9856) |

TL | -7.0294 (5.8909) | -7.3521 (0.0206) |

**1% Significance

*5% Significance

From the regression model, we have found that two of the coefficients is significant. Beside this coefficient of regression is not very high. Regression coefficient is also very low.

*Model Transformation # 6*

Dropping the profitability variables, we estimate the following linear regression model:

## LCij = α + β1NWCij + β2TAij + β3TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 10

Variable | Pooled Ordinary Least Square | Random Effect Model |

NWC | -0.2576 (3.2620) | 0.6874 (0.5492) |

TA | 1.7522* (0.6372) | -0.3544** (0.6987) |

TL | -5.7008 (2.7365) | -2.6687 (0.2354) |

**1% Significance

*5% Significance

From the regression model, we have found that coefficient TA is significant. Beside this coefficient of regression is not very high. Regression coefficient is also very low.

*Model Transformation # 7*

Dropping the profitability and return on equity variables, we estimate the following linear regression model:

## LCij = α + β1ROEij + β2TLij+ ϵ

In this equation *i* refers to a specific bank, *j* refers to a year, LCij refers to L/C loan margin and is the observation of bank *i* in a particular year *j*. and ϵ is a normally distributed random variable disturbance term or error term with zero variance.

Table 11

Variable | Pooled Ordinary Least Square | Random Effect Model |

ROE | -0.4215 (0.9969) | 0.8736 (1.1245) |

TL | -0.9935* (0.8752) | -0.42587* (0.9855) |

**1% Significance

*5% Significance

From the regression model we have found that coefficient of total liabilities is significant. Beside this coefficient of regression is not very high. Regression coefficient is also very low.

Form the above data dropping scenario, we see that only when we consider the three variables namely the financial score, behavioral score, and non-behavioral score the model is properly specified as in this case, all the respective coefficient is significant, and the overall coefficient of determination level is high. Under all other scenario most of the coefficient is not significant and overall coefficient of determination is low.