Model for assessing SME credit risk in India

Abstract

Aim:The research aims to create an accurate and efficient model for assessing the Credit Risk of SMEs in India to predict the probability of default and additionally, identify and reduce the credit risk by predicting an accurate measure of the willingness of the borrowers. newlineMethodology:The research methodology is divided into two parts, Firstly, collecting data from the credit officers with a minimum of 3 years of experience of SME credit risk with Indian Banks through a comprehensive questionnaire, which includes validation of the identified and classified variables. Through the questionnaire, 138 responses were received and these responses were subjected to principal component analysis to group the variables suitable for further model creation. The principal component analysis grouped the variables into a total of 26 variables and 6 components viz, 3 demographic, 4 behavioural, 10 financial, 6 business, 2 collateral and 1 organizational indicator. newlineResults and Discussions: The identified component variables were kept as the key indicators for model creation and were used to collect the data from the Banks. A total of 1186 accounts data were collected from Banks which included 690 standard (not distressed) accounts, 229 overdue ( bad but not yet NPA) accounts and 267 NPA or distressed SME sample accounts were utilized. For the analysis NPA and overdue accounts were considered as NPA. The training data set included 483 standard accounts and 347 NPA accounts i.e., in total 830 accounts in addition to a total of 356 accounts was separated as testing data set. The data was put through Multiple Discriminant Analysis (MDA) to create the model equation. Alongside this, the data was put through Logistic regression and Multi-Layer Perceptron (MLP) to test the validity of the MDA model. newlineConclusion:The results reveal the credit scores for the default probability and the model shows an accuracy of 78.8% with training data set and 77.7% with testing data set, which I believe is higher than existing models in the Industry.