Proposta de um modelo de Credit Scoring para uma carteira de crédito consignado visando ações de Cross-Sell.

Abstract

This work has the objective to analyze the efficiency of the credit scoring model in cross-selling action to provide greater profitability aligned with the risk of new product. This study differs from others by using a database of clients who Payroll-linked loan from the conventional modeling of a Credit Scoring offer another product, the credit card that requires a better profile for meeting payments. The study resulted in 3 of profitability and performance scenarios. In Scenario 1 without use of shoring showed profitability of R$ 0.5 million and delinquencies of 16.1%. In the others scenarios with the use of the yields scores exceeded R$ 2.3 million and delinquencies below 9%. Scenarios 2 and 3 with just score Bureau companies. Scenario 4 includes Credit scoring model developed in this work, we showed the best discrimination between good and bad customers and the highest rate of approval, 75% against 64% of the best Bureau. For this, we used data provided by a financial institution. Using SPSS and statistical techniques, the risk analysis Relative, construction of dummies and Spearman correlation analysis, generated the model Logistic Regression Binary, validated with the Kolmogorov-Smirnov test, the ROC curve and others. The model developed credit scoring showed good results as to their power of customer classification. The effectiveness of Logistic Regression as credit performance prediction tool enables the application of the use of credit scoring model by the financial institution provider of data to improve profitability and default of the customer portfolio by credit card coming from the customer base of payroll loan.