The purpose of this project was to build a loan repayment simulator for a lending company. The goal was to estimate the loan repayment curve for various products that the client was backing. Additionally, the client wanted to estimate the repayment curve for different quantiles of outcomes - a good case scenario, a likely case scenario and a bad case scenario.
We ended up designing and building a Monte Carlo simulation model that predicted 1) whether or not each account would make a payment each month and 2) the size of each payment. The model was a machine learning model - one whose parameters were entirely dictated by their existing accounts and how those account behaved in the past. If their accounts improve or decline over time, the model will learn and reflect those changes in its future predictions.
We also wrapped the model in an easy-to-use Shiny web app with descriptive analyses of the input data. Our client’s team simply uploads data, runs the model, and gets the output they need without having to touch any back-end code.