Predicting Early Mortgage Payments with AI

This post was originally part of the DataRobot Community. Visit now to browse discussions and ask questions about the DataRobot AI Platform, data science, and more.

The Problem

When a borrower takes out a 30-year mortgage, usually they won’t finish paying back the loan in exactly thirty years. It could be later, it could be earlier, or the borrower could refinance. For regulatory purposes—and to manage their liabilities—banks need to be able to accurately forecast the effective duration of any given mortgage.

Between general economic data and individual mortgage records, there’s plenty of data available to predict early loan prepayment. The challenge lies in figuring out which features, in which combination, with which modeling technique, will yield the most accurate model. Furthermore, federal regulations require that models be fully transparent so that regulators can verify that they are non-discriminatory and robust.

How Can DataRobot Help?

DataRobot automatically sifts through a universe of machine learning algorithms, derived features, and preprocessing techniques to find the most accurate model for the data in minutes instead of weeks. Mortgage loan traders can combine their practical experience with DataRobot’s brain and brawn to understand which mortgages are likely to be repaid early. Paired with a comprehensive array of explanatory visualizations, DataRobot’s model diversity ensures that these traders will be able to meet regulatory requirements without sacrificing accuracy.

Examples

Cecile is a loan trader for a bank that invests in mortgages, and she sees hundreds of mortgages every day. Cecile’s job is to decide which mortgages to invest in and which mortgages to divest. She collects data from Fannie Mae showing the default performance of prior loans, collects economic data from the Fed, consolidates the data, and prepares it for analysis. However, with so many mortgages to choose from, and so much information accompanying each, it is not humanly possible for her to model and select the optimal portfolio. Because models become outdated faster than the data science team can tweak them to better suit Cecile’s requirements, she turns to DataRobot to build models that predict mortgage prepayments and to score each new mortgage she sees.

Training data
Prediction data
Data Dictionary

About the author

Linda Haviland

Community Manager

Meet Linda Haviland