Natural disasters are a given anywhere in the world that cause billions of damage. And with climate change, the impacts are only getting worse.
The most recent NOAA National Centers for Environmental Information disaster report, released in January, confirms rising costs from weather extremes in the United States: From 2016-2022, 122 separate billion-dollar disasters have killed at least 5,000 people and cost more than $1 trillion in damage.
But in a new bit of research from University of Maryland's Robert H. Smith School of Business Professor Clifford Rossi and his students, have developed a new way to quantify risk for homeowners and mortgage lenders alike.
"Climate risk is complicated because it's a global risk," says Professor of the Practice Clifford Rossi, who spent 25 years in risk management at banks and government agencies before joining the finance faculty at Smith. "This is not just about managing risk, it's also about managing uncertainty—a much harder game."
Rossi led a group of 11 students in Smith's Master of Quantitative Finance program in a recent experiential learning project with government-sponsored mortgage enterprise Freddie Mac. The students built a model that leverages machine learning to pinpoint the regions of the country with the highest climate risk and the implications for homeowners and the mortgage industry.
The students created an interactive dashboard of all 13 million single-family mortgage loans originated in 2021 and merged it with FEMA's National Risk Index tool for all 78,000 U.S. Census tracts across 18 different climate hazards, including earthquakes, wildfires, hurricanes, coastal and river flooding, tornados and drought. Then they randomly selected 1 million mortgage loans and used a battery of different machine learning models and performance statistics to analyze the data, looking for such effects as adverse selection against GSEs Freddie Mac and Fannie Mae, impact on low- and moderate-income borrowers, minorities and other key effects.
"Anyone can point and click on any county in the U.S., for any climate hazard type and get figures on borrower characteristics such as race, age, income," says Rossi. "It is quite amazing."
Machine learning was also used in the creation process to differentiate high hazard risk areas from others based on borrower, tract and other characteristics. The Smith School plans to make the tool available online.
So how could this information be used? Fannie Mae and Freddie Mac, insurance companies, and policy makers could all use this information during the underwriting process and for figuring out which areas need more government resources/intervention.
Rossi called the project "the finest student presentation on a highly technical subject I have ever been involved with." He said senior leaders at Freddie Mac were also impressed with the extent and complexity of the students' analysis, and their ability to explain difficult concepts, calling the presentation better than some by employees and consultants.
"This is a growing area," says Rossi. "Many governments around the world are involved in trying to get their arms around this, and these students would need those tools adapted from our standard finance concepts but focusing specifically on solving climate-related financial issues."