Abstract:
In Ghana, some local banks lack a comprehensive credit risk management
framework that includes the use of credit scoring; hence, the purpose of this
study is to develop a credit scoring tool derived from a binary logistic
regression model to reduce credit risk exposure for banks in general and local
banks in particular. A review of the literature on credit scoring models or
classifiers revealed that the specificity and sensitivity of the developed models
are not explored further to reveal insights into the optimal cut-off point of the
model. This study seeks to fill this gap by further exploring the specificity and
sensitivity of the developed model and offers explanations and insights about
the variations of the optimal cut-off point. The study makes a case for using
the optimal cut-off point as a practical decision point for financial institutions.
Secondary data on borrowers were obtained from a local bank and examined
to identify its retail customers' demographic and behavioural characteristics
based on the minimum Know Your Customer (KYC) required by the central
bank. The binary logistic regression model developed from the data had an
overall classification accuracy of 93% at a cut-off point of 0.5; however, the
sensitivity measure was barely 23%. Typically, resampling techniques are
employed to deal with the imbalance, however, in this study a plot of
sensitivity and specificity against the probability of default is used to derive an
optimal cut-off point. The performance of the logistic regression model at the
derived optimal cut-off point was found to be similar to other binary models’
performance that used resampling techniques.
