SOCIAL MEDIA DATA TO DEVELOP DEFAULT PREDICTION MODEL TO SUPPORT LOAN CEILINGS DETERMINATION: CASE STUDY OF THE KEUANGAN MANDIRI & SEJAHTERA SAVING AND LOAN COOPERATIVE

• Main Author: Mauliate, Samuel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/77445
• Institution: Institut Teknologi Bandung

Saving and loan cooperatives in Indonesia as an Indonesian cultural-financial institution have the potential to increase business scale based on internal and external opportunities. However, credit risk is still the primary concern of saving and loan cooperatives, as evidenced by various default cases at these financial institutions. One of the saving and loan cooperatives in Indonesia, Keuangan Mandiri & Sejahtera Saving and Loan Cooperative (not the real name), is categorized as a national primary cooperative in Indonesia that experienced NPL fluctuations in the last five years. To process the credit risk issue, borrowers' social media data assessments have been developed. Therefore, research has aimed to identify social media variables that could be used as default probability predictors and determine the predictability level of credit scoring by using social media data variable assessment in saving and loan cooperative’s credit scoring. The research utilizes the method of logistic regression and credit scorecard to develop a loan default-predicting method with the objective of assisting the saving and loan cooperative in avoiding credit losses. Four independent variables of social media data (number of friends on Facebook, the number of months that a user has been using Facebook since the date of their first posting, the ratio of the total number of Facebook posts to the number of months the account has used Facebook, and the total number of religious accounts followed by the borrower's Facebook account) with twelve control variables consist of six historical payment variables (Tenor, Installment, Collateral, Period, Deposit, and Dividends) and six demographic variables (Gender, Age, Marital Status, District, Employment, and Income) were used to construct goodness of fit model, with criteria of Weight of Evidence (WoE), Information Value (IV), and Logistic Regression model. The results identified four variables that could be considered and used as default probability predictors, which are Employment, Money used as time deposits in saving and loan cooperatives, Number of friends on Facebook, and The ratio of the total number of Facebook posts to the number of months the account has used Facebook. Furthermore, the model increased by 8.60% when the model included a social media data component compared to the model that only used demographic and historical payment variables, while the model that used the four selected variables found a 2.7% increase in rate compared to the initial model. It is recommended to Keuangan Mandiri & Sejahtera Saving and Loan Cooperative focus on the four variables because it is more efficient in using variables and utilizing the use of social media data in the developing loan default predicting method. Furthermore, Keuangan Mandiri & Sejahtera Saving and Loan Cooperative can be more cautious with borrowers who have few social media connections, low monthly frequency of posting on social media, have employed average low salaries, and contribute little to the cooperative's financial condition. The loan applicant with these criteria should be given a small score for their credit scoring to minimize default risk that may arise in the future.