Research Sheds Light on Why the National Flood Insurance Program Is $22 billion in Debt – and How It Could Be Reformed

Study finds that hurricane-driven “hyperclustered” flood events in the last two decades account for more than half of NFIP claims to date 

Embargoed for release until December 9, 2025

For media Inquiries regarding the study, please contact Natalie Judd or Emma Scott

Washington, D.C., December 9, 2025 –  The United States faces a growing disaster insurance crisis, with $22 billion in federal debt accumulated from the National Flood Insurance Program (NFIP) and major insurers retreating from high-risk homeowner’s insurance markets like Florida and California. 

In two separate studies, a research team led by Adam Nayak Ph.D. candidate of Columbia University analyzed millions of flood claims and used machine learning and game theory to understand why the NFIP has been chronically in debt since Hurricane Katrina in 2005. He will report his findings on Dec. 8 at the annual meeting of the Society for Risk Analysis in Washington, D.C. 

Federal regulation mandates flood insurance purchase (through the NFIP) for people living within a 100-year floodplain. These higher risk areas – along ocean coasts and bordering rivers – have historically been defined as having a one percent chance of experiencing a damaging flood in any given year (known as a return period of 100 years). 

“FEMA’s flood maps are not regularly updated and do not account for pluvial flooding, the impact from heavy rains or flash floods, unless tied directly to a water body like riverine and coastal flooding,” says Nayak, “We wanted to understand, what is the actual expected frequency of the rainfall that causes these floods under a changing climate?” 

To examine financial losses of U.S. flood claims from a hydroclimatological perspective, Nayak’s team conducted an analysis of millions of federal insurance claims between the years of 1978 and 2020. 

Key findings: 

• Flood return periods are much shorter than 100 years. On average, return periods of precipitation leading to flood damage are under 5 years for insurance claims and under 20 years for disaster aid disbursements and property buyouts. 
• Chronic annual losses arise from recurrent claims, emphasizing the need for proactive managed retreat from high-risk areas, where properties experience damage year after year. These cost the NFIP $63 million per year, on average. 
• Flood return periods are significantly lower in rural communities compared to urban communities, indicating that the frequency of damage-causing events is higher in these communities. This could be related to differences in protective infrastructure standards between these communities.
• There are significant increasing trends in return periods over time for losses. This may indicate that the intensity of loss-inducing events is increasing, possibly due to a changing climate, increasing populations in floodplains, and infrastructure deterioration. 
• Regions most at risk are large, densely populated metropolitan areas with hurricane risk and coastal regions in the South and Northeast. Significant coastal hotspots of expected net loss are located in Texas, Louisiana, Mississippi, Alabama, New York, and New Jersey. 
• Surprisingly, aggregate risk-based premiums largely recover claim costs in Florida and other counties along the East Coast that may have been expected to be insolvent. This is likely driven by high insurance uptake and increased (i.e. more expensive) risk-based premiums. A similar dynamic is seen in California. 

Hurricane-driven hyperclustered storms account for over 90 percent of losses 

In a second study, the researchers used machine learning, interviews with experts in the reinsurance industry, and game theory to understand the drivers of debt and insurance failure under flood losses over time and space – in relation to the weather events that created them. They found that over 90 percent of losses were associated with what they identify as “hyperclustered” storms: large-scale flood events spanning days to weeks and induced by a common hydrometeorological driver (usually a hurricane). 

“It may be a hurricane in Louisiana, but the atmospheric dynamics may move it inland where it’s not a hurricane anymore but becomes a series of severe storms, heavy precipitation, or even tornadoes that create a lot of damage,” says Nayak. For example, the catastrophic river floods in western North Carolina in September 2024 were the result of extreme precipitation associated with Hurricane Helene. 

“These hyperclustered storms dominate claim volumes often in regions of high asset density and overwhelm a system designed for independent and rare losses,” says Nayak. 

The researchers found that the eight largest of these hyperclustered storms account for more than 50 percent of NFIP insurance claims to date. All have occurred in the last 21 years: Hurricanes Ivan & Jeanne (2004), Katrina (2005), Ike (2008), Sandy (2012), Harvey (2017), Ian (2022), Helene (2024), and the Louisiana Floods of 2016. Their costs include hydro-meteorologically connected storms and damages impacting many more states than those in the direct path of the hurricane. 

Urgent Need: Reforming the NFIP 

As of October 2025, the NFIP has expired and its extension – or reform – is up to Congress. The Trump administration has suggested decentralizing FEMA and placing more financial responsibility on the states. Nayak and his colleagues suggest that changing the way flood risk is predicted – to account for a changing hydrological climate with heavier rainfall and more intense storms – can help policymakers understand what needs to change to reduce the NFIP debt going forward. 

“We identified hydroclimatic hyperclustering and recurrent losses from high-risk properties as primary insolvency triggers in the NFIP, and found that NFIP insolvency is driven largely by catastrophic hyperclustering that is not accounted for in the design of the current risk-based premium schemes,” says Nayak. “Our analysis suggests that, with clever reform, the NFIP could be self-sustaining without chronic insolvency, and millions of properties can remain insured for non-hyperclustered flood-related losses.” 

Specifically, Nayak and his colleagues suggest that hyperclustered events be insured by a secondary, regionally funded catastrophe bond to stabilize counties exhibiting extreme expected losses, and that properties with recurrent losses should be given options for managed retreat and removed from the NFIP pool due to the disproportionate strain these properties place on the system. 

“Our findings support targeted NFIP reform and broader risk management, particularly as climate extremes intensify the homeowners’ insurance crisis,” says Nayak. “We argue that long-term resilience requires aligning financial, structural, and non-structural interventions with distinct regional risk patterns – whether driven by hyperclustering, recurrent losses, or both.”  

### 

EDITORS NOTE: 

This research will be presented on December 9 at 8:30 EST at the Society for Risk Analysis (SRA) Annual Conference at the Downtown Westin Hotel in Washington, D.C. SRA Annual Conference welcomes press attendance. Please contact Emma Scott at emma@bigvoicecomm.com to register. 

About Society for Risk Analysis 

The Society for Risk Analysis (SRA) is a multidisciplinary, global organization dedicated to advancing the science and practice of risk analysis. Founded in 1980, SRA brings together researchers, practitioners, and policymakers from diverse fields including engineering, public health, environmental science, economics, and decision theory. The Society fosters collaboration and communication on risk assessment, management, and communication to inform decision-making and protect public well-being. SRA supports a wide range of scholarly activities, publications, and conferences. Learn more at sra.org. 

Media Contact: 
Emma Scott 
Media Relations Specialist 
Emma@bigvoicecomm.com 
(740)632-0965