As regulators in the U.K., Europe, and U.S. have proposed additional disclosure requirements on sustainability and climate-related activities, more and more companies have paid greater attention to understand and measure their climate-related exposures or even incorporate climate risk drivers into their stress testing and underwriting models for capital management and decision-making purposes.
Specifically, capturing credit exposures of mortgage portfolio to physical risk,1 such as cyclones, hurricanes, or floods, and mitigating those risks subsequently, pose significant importance to GSEs (Government-Sponsored Enterprise), banks, and mortgage lenders. In this article, we will discuss why it is important to leverage flood insurance data in physical risk modeling and the key challenges to quantify the impact of flood risks on mortgage loan behaviors (i.e., default, prepay, and losses).
In the U.S., flood risk is covered through a policy from FEMA's National Flood Insurance Program (NFIP), created by Congress in 1968. Flood insurance is mandatory for buildings located in a Special Flood Hazard Area (SFHA2), but optional otherwise. Historically, the highest NFIP payouts belong to Hurricane Katrina in 2005 at $16.3 billion, followed by Hurricane Harvey in 2017, at $8.9 billion3.
How can a flood event affect your mortgage book behavior?
In the area damaged by flood events, one would expect to see an instant spike in delinquency rates. That is because homeowners will have to repair significant damages or even rebuild their properties while paying rent for short-term housing. Figure 1 demonstrates the instant hike in delinquency rates in Texas after Hurricane Harvey in August 2017.
Figure 1 – Freddie Mac delinquency status (2014–2019 vintage) in Texas
30 Day Delinquent
90 Day Delinquent
Source: KPMG LLP
After borrowers receive financial assistance (e.g., forbearance, repayment plans, or loan modifications), a great number of delinquent loans will be resolved before rolling into severe delinquency or default status. The chance of loans becoming severely delinquent or defaulting are mainly influenced by the severity of flood damage (or remaining home equity size) and income status of the borrower.4
Prepayment behavior also surges after the flood events, especially within SFHA zones, because these homeowners can use the NFIP insurance claims to pay off the outstanding loans, and many insured homeowners would like to move out of flood-prone areas by selling the house at a discounted value. This is supported by the 37.2 percent increase5 in house sale activities one year after Harvey hit.
In order to describe the heightened loss and prepayment behavior in the post flood period, the missing link in the traditional credit risk analysis is the use of insurance policies/claims data to answer the following key questions:
- How do we predict property damage caused by flood events?
- How do we predict the insurance coverage change in a loan book over time?
Our analysis below6 will address the first question, with single-family houses as the example.
Since flood depth is the best proxy to estimate property damage, for decades, a conventional way of estimating economic losses caused by flood events is based on depth-damage function, i.e., predicting the damage ratio7 as a function of flood depth. The shape of the depth-damage function is largely determined by the following property characteristics :
Number of floors
It appears that the higher the number of floors (herein, townhouse represents property with three or more floors), the lower the exposure to flood damages. The highest risk belongs to manufactured (mobile) homes or travel trailers on foundations.
Figure 2 – Damage ratio by number of floors (TX and FL) based on NFIP claims data
Damage ratio by Number of Floors
Source: KPMG LLP
Property with no basement (including unknown) tends to show higher risks of damages as compared to the ones with basement structures. As expected, when a basement is more finished and more similar to the building’s upstairs living areas, the property tends to show lower risk of flood damages.
Figure 3 – Damage ratio by basement types (TX and FL) based on NFIP claims data
Damage Ratio by Basement Type
Source KPMG LLP
Flood zone (Flood Insurance Rating Map — FIRM)
FEMA creates and maintains flood zone ratings to assess the flood risks associated with each community. As shown in Table 1 below, flood zones are defined based on risk drivers such as probability of flood hits, area characteristics (e.g., coastal, ponding, etc.), construction standards, etc.
Table 1 – NFIP Flood Zone Rating
|Flood Zones||Flood Risks||Special Flood Hazard Area (SHFA)||Area Characteristics||Subject to inundation by the 1-percent-annual-chance flood event?||Mandatory Flood Insurance Required||Floodplain Management Standards|
|A||High Risk flood zones||Yes||N/A||Yes||Yes||Yes|
|AE, A1-30||High Risk flood zones||Yes||N/A||Yes||Yes||Yes|
|AH||High Risk flood zones||Yes||Ponding||Yes, shallow flooding||Yes||Yes|
|AO||High Risk flood zones||Yes||Sheet flow on sloping terrain||Yes, shallow flooding||Yes||Yes|
|AR||High Risk flood zones||Yes||Areas that result from the decertification of a previously accredited flood protection system||N/A||Yes||Yes|
|A99||High Risk flood zones||Yes||Areas will be ultimately be protected upon completion of an under-construction Federal flood protection system||Yes||Yes||Yes|
|V||High Risk flood zones||Yes||Coastal Areas||Yes, with additional hazards due to storm-induced velocity wave action||Yes||Yes|
|VE,V1-30||High Risk flood zones||Yes||N/A||Yes, with additional hazards due to storm-induced velocity wave action||Yes||Yes|
|Zone X (Shaded), B||Moderate Risk flood zones||No||N/A||No, between 0.2% to 1% annual chance flood event||N/A||N/A|
|Zone X (unshaded), C||Low Risk flood zones||No||N/A||No, less than 0.2% annual chance flood event||N/A||N/A|
|Zone D||Potential Moderate to High Risk, yet to be determined||No||N/A||Unknown||No||No|
Source: KPMG LLP
It is observant that damage ratio is fairly sensitive to the flood zones; however, the highest damage ratios are not necessarily always associated with the high-risk flood zones. In some cases, it is quite the opposite; for example, low- to moderate-risk zones such as Zone C or X have shown high damage ratios instead. This is probably due to the stale flood map information updated by FEMA. FEMA is required to update flood maps every five years, but in some areas, the flood risk condition can evolve much faster than that and the cost of updating flood zone maps is high.8 Also, the methodology they used to come up with the flood maps and associated premium pricing calculation will be updated (i.e., Risk Rating 2.09) for the first time in 50 years in October 2021. The new risk rating methodology is expected to incorporate more flood risk drivers, including flood frequency, multiple flood types—river overflow, storm surge, coastal erosion, and heavy rainfall—and distance to a water source along with property characteristics such as elevation and the cost to rebuild.
Figure 4 – Damage ratio by flood zones (TX and FL) based on NFIP claims data
Damage Ratio by Flood Zone Type
Source: KPMG LLP
When placing the above two figures on top of each other, one can see from the Figure 6 that many areas subject to high damage (blue shades) are actually located outside of the SFHA zone. For example, although Hardin county is located mostly outside SFHA, it was ranked No. 9 of 50 counties hit most devastatingly by Hurricane Harvey.
The empirical analysis above shows the intuitiveness and importance of using insurance data in climate-related credit risk analysis. Outlined in the beginning of this article, as a next step, you may want to explore the depth-damage function to generate property-level annual loss curve under various climate scenarios. FEMA’s Hazus Flood Assessment Structure Tool (FAST), an open source solution, can be a great starting point for you. Although many hydrologists have raised concerns about the hazard data sparsity limitation and criticized the application of the Hazus model to all regions and flood events, it remains the most popular foundation for various climate analytics vendor solutions when it comes to predicting hazard damages.
Another step to be considered is to predict the insurance coverage evolution and incorporate it into climate scenario analysis. By analyzing the historical NFIP policy and claim data, you can model the long-term momentum and instant shocks in the flood insurance coverages by considering the impact of increasing premiums on low-income neighborhoods. By mapping the insurance coverage with your portfolio data, you can quantify the exposure of those underestimated flood risks (i.e., property-level vulnerabilities assessment and geographic concentrations) in your book and thus inform the decision-making process of your senior management of whether to accept, adopt, or mitigate the flood risks.
- As outlined in the Task Force on Climate-Related Financial Disclosure (TCFD) recommendations, there are two main categories of climate risks that can cause financial implications to a company: physical risk and transition risk
- SFHA: aka 100-year flood hazard zone, since these areas have an annual probability of 1 percent being hit by flood events
- Source: Insurance Information Institute, “Facts + Statistics: Flood insurance” (2021).
- Source: Journal of Housing Research, Carolyn Kousky, Mark Palim & Ying Pan (Volume 29, 2020).
- Source: The Wall Street Journal, Nancy Keates (Oct 24, 2018)
- The dataset is sourced from OpenFEMA, FIMA NFIP Redacted Claims — v1.
- Mathematically, this can be expressed as:
Damage ratio = Damage amount caused by flood/Property value
- The Association of State Floodplain Managers (ASFPM) has estimated that the cost for FEMA to maintain and update the map is about $107 million to $480 million per year.
- Source: FEMA, “Risk Rating 2.0: Equity in Action” (July 1, 2021)