Analysis and Design of Credit Scoring Model for Microfinance Service in Telecommunication Industry

Abstract

This project is conducted for Simbrella - the provider of microfinance solutions and services in telecommunication industry. The company heavlly relies on credit scoring techniques within its solutions for dlscrlminatlng loan applicants based on their calculated credibility. There is, however, a doubt that the scoring model that is currently applied by Simbrella is inherently simplistic and suboptimal in terms of the ultimate value creation. Therefore, the project's main goals are reviewing some past academic research on the application of scoring models in microfinance, evaluating the efficiency of the currently applied scoring model, as well as taking certain steps towards optimization of the current model or creating a new scoring model for the company's solutlons. Several academlc papers were reviewed, which concentrate on the applicability of statistical scoring in microfinance and some statistical methodologies aimed at the probabilistic analysis of the microfinance loan applicants’ behavior. Scoring models were found to be suitable enough for microfinance. Also, the two measures of a credit scoring model efflclency - ’true rates’ and ’predictive values’ - were revealed during the literature review. However, the company's available data on the past loans’ performance did not provide the whole picture for the evaluation, as it comprised only the history of the approved, but not the rejected loans. To fill in this gap, the experlment based on random sampling was designed and Implemented in three distinct operational environments of the company. The experiment has provlded the missing data on the performance of the rejected loans. The holistic evaluation of the current scorlng model's efficiency suggested that the model is sufficiently precise in its loan request approving capabillties, but Is suboptimal in terms of rejecting of the loan requests. The new value-centric probabilistic loan appllcant risk assessment framework was suggested as a prospective basis for creating a new scoring model for Simbrella product in future. A new, ready to be applied scoring model was not dellvered In scope of this project. As a result, several recommendations were offered to 5ImbreIIa. Statistical scoring models were fbund well applicable fbr microfinance, hence it was suggested to continue using them in the company's solutions. However, as the currently applied scoring model proved underperformlng In some aspects, it was recommended to invest In the model's Improvement, or to replace it by a new more efficient one. The value-centrlc risk assessment framework devised within the scope of this project was proposed as a possible foundation for creatlng a new scoring model fbr the company's product.