UTILIZATION PROCESS PROPOSAL FOR PT X TO MAKE CREDIT DECISION USING ITS E-COMMERCE DATA

• Main Author: MUHAMMAD PRATAMA NIM 14414016 , PATIH
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/29914
• Institution: Institut Teknologi Bandung

Indonesia e-commerce competition level is escalating through the roof propulsed by the combination of the enormous size of population and the constant rising of middle class segment. One of the big players in the industry, PT X, are looking to increase the buying power of their users by disbursing loan to its users including those deemed not creditworthy by financial institution. PT X have been storing vast amount of data from their users’ activities on its platforms: marketplace, digital, and fintech. The next step for PT X is to be able to use those data as the competitive advantage in achieving their goal: to increase the buying power of all their users. <br /> <br /> This research proposes steps for PT X to utilize their e-commerce data to decide whether a certain user shall be granted a loan based on their calculated default probability. Without the availability of historical data, the research uses Monte Carlo simulation approach to addresses three factors of uncertainty faced by PT X: the prediction quality of their internal variabels to loan performance, the level of missingness in a variable, and the number of sample needed to build the prediction model. In the research two phase simulation will be conducted. First phase is to decide whether the logit model is suitable and the condition where it could be used. Second phase is to decide whether missingness level in a variable affects its prediction quality and also the number of sample needed to build the prediction model. <br /> <br /> From the first phase simulation, the research concludes that logit model is a very stable classifier in every scenario tested excluding the first scenario where there exists very little correlation between the independent variables and the dependent variables. The second phase simulation results on that the level of missingness should also be considered in selecting a variable to be used in the prediction model while PT X should be able to build a good prediction model just as good with much less number of samples. By using the result of the research, PT X shall be able to use their data to predict their users’ future loan performance.