Risk Scorecards Using Alternative Sources of Data for Credit Risk Applications

This paper will use alternative data sources, such as Telecom postpaid data and Twitter data to develop an approach that can be used to underwrite credit invisible people using machine learning algorithms. A credit risk score gives the probability that a customer will become delinquent in the next x number of days, where the risk definition of the bank determines x, and 90 days is the most common definition used throughout the industry. Banks are missing out on a big chunk of a population for which no credit information is available and are thus credit invisible. Here, we want to test the application of Alternative credit data, information obtained from non-traditional data sources that helps evaluate a consumer’s creditworthiness. The study contributes to the literature by explaining the usefulness of alternative data sources in improving credit scoring models.