Scribd
Upload
42 views

Report Containing Steps of How I Reached The Final Model

The document describes steps taken to build a regression model to analyze the effect of independent variables on NPA ratio (dependent variable). Control variables include repo rate, inflation, and GDP. After initial regression, variables were dropped due to multicollinearity. The model was then tested and found to have heteroskedasticity and autocorrelation. To address this, random and fixed effect GLS regressions were used. The final model was a cross-sectional time-series FGLS regression including GDP and domestic credit as controlled variables.

Uploaded by

SAMARTH TIWARI
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
42 views

Report Containing Steps of How I Reached The Final Model

The document describes steps taken to build a regression model to analyze the effect of independent variables on NPA ratio (dependent variable). Control variables include repo rate, inflation, and GDP. After initial regression, variables were dropped due to multicollinearity. The model was then tested and found to have heteroskedasticity and autocorrelation. To address this, random and fixed effect GLS regressions were used. The final model was a cross-sectional time-series FGLS regression including GDP and domestic credit as controlled variables.

Uploaded by

SAMARTH TIWARI
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 7

Dependent variable – NPA Ratio

Independent variable - Credit-Deposit (%), Net Profit / Total funds (%), Operating Expenses /
Total Funds (%), Log_Advance to Real Estate Sector (Rs. Cr), Log_Advance to Capital Market
Sector (Rs. Cr), % of Net Non-Performing Assets to Net Advance, Return on Assets (%), and
RONW (%)
Control Variable - Repo Rate, Inflation, Gross Domestic Product (GDP)
To find out the effect of independent variables on the dependent variable I run the following
regression model.
Step 1 – Regress all independent variable against NPA
Pooled OLS Regression Model

Step 2 – check for correlation using correlation table and VIF


As it is clearly visible from the correlation and VIF table, that Log(Domestic Credit) and
Log(GDP) have high correlation with all other variables. Therefore, I decided to drop
Log(Domestic Credit) and Log(GDP).

Step 3 – Regress again without Log (Domestic Credit) and Log(GDP)


Pooled OLS Regression

Again, I checked for multicollinearity using VIF to ensure that there is no multicollinearity
among the variables in the new model. As VIF for each variable is less than 5. Hence,
multicollinearity does not exist anymore.
Step 4– Check for Heteroskedasticity using Breusch Pagan and white test.
Null Hypothesis – Homoscedastic in nature
Alternative hypothesis – Heteroscedastic in nature
As p-value < 0.05 in both the cases, we will reject the null hypothesis. Hence,
Heteroscedasticity exist.

As the data is heteroskedastic the pooled OLS regression will give biased and inconsistent
estimator. Therefore, I used Random effect GLS (RGLS) regression and Fixed effect GLS (FGLS)
regression for constructing new model.
Step 5 – Used RGLS and FGLS model to estimate better coefficients. Also, used Hausman test
to determine the more appropriate model between RGLS and FGLS.
Hausman Test
Null hypothesis – RGLS is appropriate
Alternative Hypothesis – RGLS is not appropriate
Fixed effect GLS (FGLS) regression

Random effect GLS (RGLS) regression


Result of Hausman Test

As p-value = 0.1006 > 0.05. Therefore, we will accept the null hypothesis i.e. RGLS is an
appropriate model.
Step 6 – As, many expected variables were not coming to be significant. Therefore, I checked
the model for autocorrelation using Wooldridge test and cross-sectional dependence using
Breusch-Pagan LM test of independence

Result of Breusch-Pagan LM test of independence

As, p-value < 0.05, we will reject null hypothesis i.e. residuals across entities are not
correlated.
Result of Wooldridge Test

As, p-value < 0.05, we will reject null hypothesis i.e. No first-order autocorrelation exists.
Hence, there exist the problem of autocorrelation and cross-sectional dependence.

Step 7 – Use Cross-sectional time-series FGLS regression which solves the issue of
autocorrelation and cross-sectional dependence
Step 8 – Adding controlled variable to the model in order to remove omitted variable bias.
Included Log (GDP) and LOG (Domestic Credit) in the model as controlled variables as their
presence makes the GDP growth rate a statistically significant variable.

Final Estimated model for determining factors effecting NPA of a country


Coefficien
NPA t Standard Error Z-Statistic P- value
Unemployment 0.262 0.072909 3.6 0.000
Interest Rate 0.129 0.039 3.310 0.001
Exchange Rate -0.001 0.013 -0.110 0.909
Log (Total Net Income) -0.463 0.165 -2.810 0.005
Log (Total Deposits) 3.034 0.469 1.450 0.311
Dummy (Dev = 0, Undev =1) 2.307 0.432 5.340 0.000
GDP Growth Rate 0.161 0.070 1.310 0.071
Log (GDP) 4.191 1.178 3.560 0.000
Log (Domestic Credit) -4.389 0.890 -4.930 0.000
_cons -14.505 6.302 -2.300 0.021

You might also like