United Guaranty Insights United Guaranty Insights

Advanced Predictive Modeling—Applications to Drive Identification of Underwriting Issues
Jim Krueger
By Jim Krueger, Vice President, Risk Management
United Guaranty Corporation
Posted on September 29, 2015

Recently, United Guaranty's Risk Management team was invited to be a presenter at the MBA's Risk Management, QA & Fraud Prevention conference to talk about some of the advanced work we're doing with predictive modeling, which leverages statistics to analyze current and historical data to help manage risks.

I was part of a panel called "New Developments in Mortgage Performance Data and Analytics" and discussed a predictive model United Guaranty is developing to better understand factors that can identify loans that may be at higher risk for material underwriting errors—in the origin phase or later in the loan's life cycle.

United Guaranty leads the mortgage insurance (MI) industry in the quantity and proportion of full-file underwriting we do. With the most experienced underwriting team in the industry, our SecureCertSM program provides the maximum rescission relief available in the MI industry. Lenders trust us to make the MI decision and eliminate the risk of their making an MI underwriting error.

We also have the largest underwriting team in the industry delivering the best customer service—with decisions within 24 hours or less in most cases.

Rather than resting on our laurels, we're working to get even better. In 2012, our parent company, AIG Inc., established a world-class research and development team focused on predictive risk analysis. We joined with AIG's Science Team earlier this year to study mortgage insurance policies going back several years where we discovered material underwriting defects through Quality Assurance reviews.

Through that research, we were able to review the predictive power of more than 60 variables associated with borrowers, collateral and loan instruments. Ultimately, we discovered a number of variables were highly predictive in identifying loans at risk for a material underwriting defect. In fact, the model has proven to be nearly twice as effective in identifying loans with material underwriting defects as random sampling.

We're still in the testing phase of rolling out our predictive model, but the early results are promising. As we add to our data set and update it year over year, we expect our model to improve. We're currently using the model to improve the effectiveness of our Quality Assurance reviews. In the future, we may also be able to use the model to identify loan files with variables that signal higher risk for errors and send those files to the most senior underwriters.

Again, this is still in development, but United Guaranty's focus on using the most advanced risk- reduction strategies is part of our commitment to helping keep uninsurable loans out of the system.

I look forward to reporting on the progress of these efforts in the near future, so watch this space for periodic updates on the development of our predictive modeling application.

Jim Krueger

Jim Krueger is Vice President–Risk Management at United Guaranty. He joined United Guaranty as Vice President–International Marketing and Product Development in 2007 and was named Vice President of International Business Development in 2011 before joining the Risk Management team in November of 2013. Previous to United Guaranty, he worked for another mortgage insurer after beginning his career at GE Capital. He's a graduate of the University of Nevada Las Vegas and earned an MBA from the University of Florida. He holds an Executive Scholar certification in Marketing and Sales from Northwestern University's Kellogg School of Management and has an AMP designation from the Mortgage Bankers Association.

© United Guaranty Corporation 2015. All rights reserved. United Guaranty is a marketing term for United Guaranty Residential Insurance Company and United Guaranty Mortgage Indemnity Company and is a registered mark. Coverage is available through admitted company only.