A Model for predicting credit card loan defaulting using cardholder characteristics and account transaction activities
Muchiri, Lewis Munene
MetadataShow full item record
Defaulting of credit card loans is an area of great concern amongst banks and financial institutions. This is because loan portfolio is considered an asset to the institution and one which directly impacts the firm’s profitability, hence effective measures need to be put in place to ensure that default risk is at manageable levels. Credit card accounts are normally classified as “good” or “bad” depending on whether the cardholders are able to repay their debts within the agreed time or not. Good accounts continually settle their debts as per the agreement with the bank and consequently receive a higher credit limit over time allowing them have more credit at their disposal while bad accounts default their payments and end up with blocked credit cards or have more punitive measures taken on them. The latter bring losses to the banks. It is therefore imperative for these financial institutions to improve their credit management processes with the aim of optimizing their provisioning for bad debts, minimizing losses resulting from defaults and maximizing on revenue that would be generated from the “good” accounts. Traditional credit scoring techniques have relied on static models of loan default prediction which produce classes of cardholder accounts according to default risk but which give no insight to the changes in loan states over time. Changes in loan states may reveal much about the terminal state of the credit card loans which is to be expected with respect to the defined duration of time and was hence a good consideration for study. This study applied Logistic Regression Analysis to develop a model that learned patterns from observed account transaction activities and performed predictions on subsequent credit loan states based on the learned information. Successful predictions were obtained with an accuracy level of 85.28%, a recall level of 83.71% and a precision level of 77.87%, indicating the value in considering the states of loans over time and the events leading up to the terminal default or non-default states as opposed to a singular focus on the final default or non-default events.