Residential Flood Risk in the United States

Residential Flood Risk in the United States:

Quantifying Flood Losses, Mortgage Risk and Sea Level Rise

May 2020

Quantifying Flood Losses, Mortgage Risk and Sea Level Rise

Caveat and Disclaimer

The opinions expressed and conclusions reached by the authors are their own and do

not represent any official position or opinion of the Society of

Actuaries or its members. The Society of Actuaries makes no representation or warranty to the accuracy of the information

AUTHORS

David D. Evans, FCAS, MAAA

Consulting Actuary

Milliman, Inc.

Cody Webb, FCAS, MAAA, CPCU

Principal and Consulting Actuary

Milliman, Inc.

Elias Braunstein

GIS Analyst

Milliman, Inc.

Jonathan Glowacki, FSA, CERA, MAAA

Principal and Consulting Actuary

Milliman, Inc.

Andrew Netter

Senior Financial Consultant

Milliman, Inc.

Brandon Katz, M.Sc.

Executive Vice President, Member

KatRisk, L.L.C.

Dag Lohmann Ph.D.

Chief Executive Officer

KatRisk, L.L.C.

SPONSOR

Society of Actuaries

Research Executive Committee

CONTENTS

Introduction.............................................................................................................................................................. 4

Executive Summary .................................................................................................................................................. 5

1. The Flood Insurance Gap ...................................................................................................................................... 7

1.1 How Mortgage Requirements Affect Insurance Purchase ................................................................................ 7

2. Methodology: Estimating the Exposure of U.S. Residences to Flood ..................................................................... 7

2.1 Data and Assumptions ........................................................................................................................................ 7

2.2 Catastrophe Simulation Models – A Primer ....................................................................................................... 8

2.3 The KatRisk Flood Model ..................................................................................................................................... 9

2.4 Applying and Selecting Sea Level Projections .................................................................................................. 11

2.5 Creation of a Market Basket ............................................................................................................................. 12

2.6 Estimating NFIP Take-Up Rate .......................................................................................................................... 13

2.7 The Residential Private Flood Insurance Market ............................................................................................. 15

3. Results: Estimates of Insured and Uninsured Flood Exposure to U.S. Homes ...................................................... 18

3.1 Insured and Uninsured Loss Results ................................................................................................................. 18

3.2 Impacts of Sea Level Rise .................................................................................................................................. 20

3.3 Summary of Results by MSA ............................................................................................................................. 26

4. Projections: Mortgage Default Risks After Catastrophic Flooding ........................................................................ 28

4.1 Potential Bearers of Flood Risk: Beyond Homeowners and Insurers ............................................................. 29

4.2 Methodology: Catastrophe Analytics for Home Loans .................................................................................... 32

4.3 Results: Catastrophe Analytics for Home Loans .............................................................................................. 44

4.4 Historical Results: 2017 Mortgage Performance Following Hurricane Irma .................................................. 48

5. Appendix ............................................................................................................................................................ 58

5.1 Maps: Increase in expected Flood Losses under Sea level rise scenarios ...................................................... 58

5.2 Exhibits ............................................................................................................................................................... 64

5.3 Overview of MIlliman M-Pire: Mortgage Analytic Methodology .................................................................... 77

About The Society of Actuaries ............................................................................................................................... 81

Limitations .............................................................................................................................................................. 82

Residential Flood Risk in the United States

Quantifying Flood Losses, Mortgage Risk, and Sea Level Rise

Introduction

This Society of Actuaries (SOA) report authored by Milliman, Inc. (Milliman) is designed to evaluate the impact and

evolution of societal risk management due to projected future changes in frequency, severity, and variety of

weather-related catastrophes.

One catastrophic peril that could be particularly impacted by these changes is flooding. As sea levels rise, so will the

risk of hurricane-related inundation to coastal properties, driven by storm surge. Additionally, changes to

precipitation patterns could lead to increased risk of inland riverine flooding and urban flash floods.

Under current climactic conditions, flooding stands out among other natural catastrophes in terms of the ongoing

risk that it poses to the financial health of the average United States household. Despite recent efforts to reform

the National Flood Insurance Program (NFIP), most U.S. homeowners do not carry insurance to protect their

properties against the risk of flooding. For most homeowners, the purchase of this coverage is mandatory only if

they live in certain specified high-risk areas. However, significant risk exists in areas where the purchase of flood

insurance is rare. Additionally, even in areas where flood coverage is required, data from the NFIP and private flood

insurers do not indicate high degrees of coverage.

Beyond direct damages to property and communities, the flood insurance protection gap could have many

downstream financial impacts. Because homeowners insurance is integral to protecting the collateral that underpins

the U.S. mortgage system, coverage gaps could create adverse financial exposure to bearers of mortgage risk

including mortgagees, insurers, reinsurers, federal underwriting agencies, and bondholders.

If the frequency or intensity of flooding were to increase, exposed American households could be at more risk in

the future than they are today. Further, some areas historically considered to have low flood risk could become

more exposed, extending the problem’s potential economic impact on the U.S. residential housing stock.

This report examines current countrywide residential exposure to flooding, considers how it could be impacted by

sea level rise, evaluates how this could affect the financial health of residential householders, explores a new

technique to determine whether it could impair their ability to meet their mortgage obligations, and analyzes the

effects of defaults to other parties or institutions.

Executive Summary

This report provides estimates of the insured and uninsured flood exposure of single-family residences in the

contiguous United States from both storm surge and inland flooding. Additionally, storm surge losses have been

produced across sea level rise scenarios that represent how moderate to extreme interpretations of contemporary

scientific projections could affect today’s housing stock.

Results are presented below and in the pages that follow, mostly at the Metropolitan Statistical Area (MSA) level.

While management of flood risk occurs from the federal jurisdiction down to the individual homeowner, MSA-level

results provide an opportunity to identify which local regions may face the most imminent challenges with respect

to developing risk-management strategies to address flooding risk. Note the following key findings with respect to

our analysis of countrywide single-family flood risk:

• Building losses to single-family residences due to flood are expected to cost more than $7 billion annually, and

more than 87% of those losses are estimated to be uninsured by the NFIP. If private flood insurance data were

included with NFIP data in the estimation of the uninsured loss percentage, it is likely that this estimate would

only marginally decrease due to the small size of the residential private flood market relative to the NFIP.

• Building losses to single-family residences due to flood are estimated to average about $78 per single-family

residence per year

. Other costs to homeowners such as damage to contents and other structures, as well as

additional living expenses, mean that

expected total losses for homeowners due to flood are comparable to other

major perils typically insured by a homeowners policy, such as fire and wind/hail.

• Uninsured losses are prevalent across the entire United States. We estimate that 69% of MSAs in the United

States have 90% or more of their expected flood losses uninsured, with only 6% of MSAs having more than 30%

of their expected flood losses insured.

• Countrywide, we estimate that approximately one-third of homes in the SFHA have an NFIP policy, with the

majority of states having less than a 25% take-up rate. Outside the SFHA, every state except Louisiana, Florida,

and Texas has take-up rates of approximately three percent or less, with the majority being less than one

percent.

• The increase in storm surge losses due to sea level rise is significant in all storm surge exposed areas, and highly

sensitive to the amount of sea level rise.

We estimate that sea level rise will increase total storm surge losses 21%

by 2050 in our medium sea level rise scenario, and 66% in our high sea level rise scenario. Local impacts can be

much higher than these averages.

• The severity of extreme flooding events is significantly higher with sea level rise. Half of MSAs exposed to storm

surge currently are expected to have losses from extreme “one-in-500-year” flood events increase by 10% or

more in our medium sea level rise scenario. In our high sea level rise scenario, the increase is 17%.

• MSAs where flooding is already significant relative to income stand to see some of the largest increases in

expected flood losses.

The 10 MSAs with the highest ratio of expected flood losses to annual household income

today are estimated to have flood losses increase by 0.27% of income in our medium sea level rise scenario and

0.75% in the high sea level rise scenario.

• The high percentage of losses uninsured is not expected to improve in the future unless current flood insurance

purchase patterns change. The vast majority of additional losses due to sea level rise will continue to be mostly

uninsured. Sea level rise will cause the amount of uninsured losses to increase on an order similar to total losses.

With uninsured losses already relatively high, homeowners will be faced with either increasing uninsured losses

or paying for flood insurance policies that they do not have today.

An area of growing concern has been the impact of flooding and sea level rise on the ability of homeowners who

financed their residences to pay their mortgages. This report builds on the flood catastrophe modeling results by using

them in conjunction with a financial model predicting homeowner default to estimate the degree of mortgage

impairment that could potentially result from catastrophic flood events. This was estimated using a sample of loans

backing recent credit risk transfer (CRT) securities from the Federal Home Loan Mortgage Corporation (Freddie Mac).

We note the following key observations with respect to our modeled estimates of flood impacts to mortgage risk:

• While current insurance purchasing patterns and the mandatory purchase requirement do mitigate the expected

impact to credit losses,

model estimates indicate that that at least 42% of the expected increase in credit losses

due to extreme flood events could remain after considering the benefits of insurance claim payments.

• Local impacts of some extreme flood events on mortgage defaults could be substantial, with modeled credit

losses an order of magnitude higher post-event.

Even considering current insurance purchasing patterns and

requirements, estimated credit losses for impacted regions increased between approximately four and 23 times

for each of our modeled events.

• Despite the large local impacts, diversification can mitigate most of the credit losses arising from a single extreme

flood event on large portfolios of loans similar to those backing Freddie Mac CRT securities. We estimate that

increased credit losses due to a given extreme flood event ranges between one and five basis points per event

across each scenario for an entire collateral pool, after accounting for flood insurance claim payments. This cost

will vary based on the severity of the event and concentration of loans in the pool for a given event.

• The incremental credit loss impact estimates, while small on the aggregate pool, could deliver a more significant

impact to investors in subordinate tranches of CRT mortgage securities. For a subordinate tranche of high loan-

to-value loans, we estimated flood events could increase principal writedowns by 15% to 95% relative to a

baseline scenario without a flood event. Thus, the relationship between catastrophic and economic

concentrations of risk may be important to consider when evaluating and pricing these transactions.

• Sea level rise can potentially impact credit losses similar to the overall increase in expected losses discussed

above.

Increased credit losses due to extreme flood events increased by 24% and 72% using our medium and

high sea level rise scenarios, respectively.

• Each of the points above is in contrast to historical experience of loan performance. In many cases, such as

Hurricane Irma, historical experience has been more favorable than these estimates would indicate. We believe

there are several intuitive reasons for this, notably

that federal and state financial assistance programs likely

provided a significant buffer against credit losses for prior flood events. Thus, the true credit exposure to these

events is likely lower than models would indicate, so long as assistance continues to be paid at a rate similar to

the one it historically has been. However, if disaster assistance programs were to be reduced or eliminated, the

financial threat posed by this issue could be larger than historical data indicates.

1. The Flood Insurance Gap

1.1 HOW MORTGAGE REQUIREMENTS AFFECT INSURANCE PURCHASE

For homeowners, there is no legal obligation to obtain property insurance. However, insurance purchase

requirements apply to most homeowners as a condition of their mortgages, for which insurance is an essential tool

for lending institutions to mitigate risk to the collateral that secures their loans. They do this by requiring borrowers

to purchase hazard insurance meeting certain guidelines. These guidelines are typically set by federal agencies

because the ultimate guarantors of approximately 60%

of outstanding U.S. mortgage debt are the government-

sponsored enterprises (GSEs). These guidelines shape the coverages and exclusions provided by most insurance

policies and are the primary mechanism society has in place to ensure that homeowners maintain proper insurance.

As a result, the uptake of various coverages is heavily dependent upon such requirements, and homeowners may

incorrectly assume that the insurance they are required to purchase is enough to protect their property from loss.

For flood insurance, purchase is only mandated on federally-backed mortgages for properties located in Special Flood

Hazard Areas (SFHAs), defined by flood maps that use boundaries of “one-in-100-year” floodplains. The requirement

is a “yes or no” mandate, and could provide poor alignment between coverage and risk for many reasons, including:

• The location of a property outside the “100-year floodplain” does not indicate that there is no flood risk or that

the risk is insignificant. The “100-year flood,” or, more accurately, the flood with a 1% annual chance of occurring,

does not fully capture the range of risk. Estimates indicate that an inch of water in a home from a “less-than-100-

year flood” may still be catastrophic, reaching close to $27,000 for a typical home

. The 100-year flood can

happen in any year and has a 26% chance of occurring over the life of a 30-year mortgage, even assuming a stable

climate

• The 100-year floodplain changes over time due to land use and natural factors, but maps only change when

government agencies update them.

• The Flood Insurance Rate Maps (FIRMs) defining the 100-year floodplain mostly do not consider precipitation-

driven flash flooding, tsunami, or the interaction between riverine and coastal flooding.

Accordingly, although there are existing mortgage requirements intended to ensure that flood insurance is in place in

high-risk areas, uninsured exposure to flood risk may exist for many homeowners and mortgages.

2. Methodology: Estimating the Exposure of U.S. Residences to Flood

2.1 DATA AND ASSUMPTIONS

To assess overall flood risk and the flood insurance gap, we paired a realistic cross-section of the U.S. housing stock

with data from the NFIP and a catastrophe simulation model, a tool that insurers use to quantify their financial

exposure to natural catastrophes. Using this data, we are able to produce estimates of total, insured, and uninsured

flood risk not only at the national level but also at a local level so that this issue can be examined for each region and

municipality in the country.

Urban Institute, Housing Finance Policy Center (March 2020). Housing Finance at a Glance: A Monthly Chartbook. Available at:

https://www.urban.org/sites/default/files/publication/101926/housing-finance-at-a-glance-a-monthly-chartbook-march-2020.pdf

Federal Emergency Management Agency. Estimated Flood Loss Potential. Available at: fema.gov/media-library-data/1499290622913-

0bcd74f47bf20aa94998a5a920837710/Flood_Loss_Estimations_2017.pdf

The 26% probability is equal to one minus the probability of no 100-year flood in each year (1 – 0.99

See Federal Emergency Management Agency, Flood Map Revision Processes at https://www.fema.gov/flood-map-revision-processes for details of the

flood map revision process.

We focused on expected flooding losses for buildings of single-family residences in the contiguous United States as

modeled with the KatRisk SpatialKat Flood Model (KatRisk Model). Unless stated otherwise, this definition applies to

all references to expected loss contained in this report.

The KatRisk Model comes from a field of analytical tools known as stochastic catastrophe models, which are used for

pricing and managing the risk of natural catastrophe risk in the insurance industry. With most catastrophes, and

particularly flood, historical data is limited in its ability to predict future events. When provided with input data in the

form of a portfolio of exposure characteristics, these catastrophe models provide estimates of losses from future

events using accepted scientific and engineering principles, but with smoothed results derived from the execution of

thousands of simulation runs.

Insurers typically use these models in conjunction with their own policy data for the purpose of assessing portfolio

risk or negotiating reinsurance treaties. For our analysis, the portfolio in question was the stock of U.S. single-family

residences.

To build a dataset that could be exposed to the KatRisk model to provide estimated countrywide flood losses, we

developed a “market basket” of single-family residences in the contiguous United States. Market baskets are

hypothetical datasets used to represent cross-sections of markets. They rely on the actual locations of properties with

risk characteristics relevant to flood risk that are estimated with an intent to be as realistic as possible.

We also used NFIP’s OpenFEMA data to estimate which properties were likely to have purchased flood insurance from

the NFIP, both inside and outside the SFHAs where purchase is mandated. The impact of private flood insurance is

relatively small compared to the NFIP today and has been excluded from our analysis. For further discussion of this

point, see Section 2.7.

The final market basket contained approximately 1% of all single-family residences in the contiguous United States—

close to 1 million locations. Though this is a credible amount of data when using a catastrophe model for the purposes

of this paper, the statistics presented are sensitive to some inherent uncertainty arising from the sampling of locations,

assignment of property characteristics, and estimates relating to whether NFIP insurance is in place.

KatRisk model output for the market basket serves as the core for the analysis that follows in this report. Detailed

information about the analysis is presented throughout the remainder of the section.

2.2 CATASTROPHE SIMULATION MODELS – A PRIMER

Catastrophe simulation models combine engineering, sciences, and principles of insurance to project future costs

associated with natural disasters. In the insurance industry, they have become a fixture upon which the vast majority

of catastrophe pricing is based. At their core, these models seek to simulate every realistic potential type of disaster

for which a property may be at risk, computing measures of both frequency and severity in the form of return period

and average annual loss statistics. The output of these models is then used by actuaries as a synthetic loss history and,

along with any actual loss history, can be used to inform an underwriter or homeowner of their potential financial risk

over different time horizons. The output from catastrophe models as they pertain to flood risk are especially important

in the United States because publicly available insurance loss history for flood is either summarized, sparse, or volatile,

making it difficult to know the potential risk of a property.

The primary input to a catastrophe model includes information about each property at risk. The most important of

these inputs for computing flood risk include: location, occupancy (e.g., residential or commercial) construction type

(e.g., wood, masonry, steel), presence of a basement, elevation of the first floor above ground, number of stories,

and the value of different types of property (building, contents, appurtenant structure, interruption of use value).

Other important factors may include whether a structure has any enhanced flood defensive measures and any existing

insurance limits and deductibles. How a structure is affected by a flood is a function of the materials and composition

of the building; detailed building information is required for a catastrophe model to provide accurate results.

After importing the building characteristics into an inland flood and storm surge model, every structure is then

exposed to thousands of years of simulated hurricane and rain-storm events. Losses are computed for each of these

events both individually and by other aggregates (ZIP code, state, etc.). For each of these aggregation levels, an event

loss table (ELT) is computed that tracks the total loss by event. With an ELT, which may be thought of as a simulated

loss history, average annual loss (AAL, also known as the pure premium) can be computed. This is determined by

summing all the losses and dividing by the number of simulated years. This is referred to as the pure premium because

it is the minimum amount an insurance company would need to charge over the entire simulation period to, on

average, break even. It is the average loss over every year for a location or aggregation level. Finally, exceedance

probability (EP) may be calculated, which relates a probability or return period year with a given loss. As an example:

if the one-in-100-year flood loss for a property is $100,000, then every year there is a 1% chance of having a $100,000

or greater loss event at that location. It should be noted that there are two different perspectives of EP curves:

occurrence and aggregate exceedance probability (OEP and AEP). OEP simply uses the largest event loss in a given

year to compute the EP curve, whereas AEP uses the sum of all events in a given year. The below figure summarizes

the three output loss statistics presented above.

Figure 1: Exceedance Probability Curve

2.3 THE KATRISK FLOOD MODEL

The KatRisk flood and storm surge model is based on the simulation of 50,000 years of precipitation and hurricane

events, which includes more than 2 million events in the United States and Canada. For the U.S. flood model, primary

input data includes precipitation data from the Center for Climate Prediction (CPC) since 1979, forcing data (such as

temperature, humidity, and radiation) from the North American Land Data Assimilation (NLDAS) since 1979, stream

gauge station data from the United States Geological Survey (USGS), and 10m horizontal resolution elevation data

from the National Elevation Dataset (NED).

After obtaining all required data, simulation of inland flood events is performed by first modeling sea surface

temperature (SST) globally using a corrected approach.

Afterwards, we coupled the principal components of the SST

analysis (often referred to as teleconnections, such as El Niño/La Niña, Atlantic Multidecadal oscillation, etc.) to

Navarra et al.: A stochastic model for SST for climate simulation experiments, Climate Dynamics (1998) 14: 473-487

precipitation by using a VARMAX state-space model.

Next, all major statistical patterns of rainfall and land and sea-

surface temperature are computed using the VARMAX model that was built with observed data. Finally, simulation of

events of 10 thousand seven-year periods are computed using a combination of historic data and the statistical

patterns mentioned above, sometimes referred to as analog models.

These data will then cover both events that

look similar to history and those events that may be more extreme than history.

Once KatRisk has computed 10 thousand seven-year periods of simulated rainfall, the flow of water downhill toward

river outlets must be computed using hydrologic and hydraulic models to capture both the fluvial (river over-topping)

and pluvial (surface water) sources of flooding. To compute pluvial flooding, KatRisk first uses a hydrologic model to

track all sources and sinks of water over time. This includes but is not limited to snow melt/retention,

evapotranspiration, and loss to groundwater. Once the appropriate rainfall/runoff parameters are computed, the

model can then determine what percent of the water that falls in a catchment will flow overland and not be lost to

the aforementioned sinks. Water is then allowed to fall onto the ground and is routed using a two-dimensional finite

volume wave diffusion equation. Computing the fluvial component of flood is accomplished by using the St. Venant

equations, which track how water moves from upstream to downstream and out into the ocean. Once the total flow

is known along the river, the two-dimensional shallow water equations are used to compute flood inundation. This

process is described visually in Figure 2.

In summary, to compute the inland (pluvial and fluvial) components of flood over every simulated year, the general

steps as shown below include (1) modeling rainfall and snowmelt, (2) modeling the fraction of rainfall that results in

run-off, (3) hydraulic modeling of the water over land surface to streams and rivers, and (4) modeling riverine flow

from upstream catchments to downstream catchments. The final step in computing the pluvial/fluvial event set is to

discard the first two years of every simulation to ensure that the model has reached a steady state before data is

retained.

Figure 2: Hydrologic Flow

The above covers flooding from large (synoptic) scale inland flood events. The KatRisk model also simulates flooding

as a result of tropical cyclone-induced extreme precipitation and storm surge. KatRisk has developed a hurricane track

set using the HURDAT dataset from 1950 to 2008, which has been filtered to use only pertinent hurricanes. Cyclone

genesis is then determined using the rates from the historic record during different seasons and climate cycles, the

most important feature being climate patterns related to high and low sea surface temperatures that respectively

increase and decrease hurricane generation rates and strength. The movement and intensity of these cyclones is then

governed by a combination of physics and statistical likelihood functions over a sea surface with spatial and temporal

Kedem, B, and Fokianos, K: Regression Models for Time Series Analysis, Wiley, 2002

van den Dool, Huug: Empirical Methods in Short-Term Climate Prediction, Oxford, 2007.

variation. The landfall rate and intensity of these storms is then analyzed to ensure that they generally match historical

rates during different seasons and climate states. Finally, tropical cyclone rainfall is computed using physical equations

relating statics including but not limited to central pressure and the wind field.

The final track set consists of 50,000 years of tropical cyclone tracks with annual rates and tracks of tropical cyclones

dependent on main development region SST and El Nino-Southern Oscillation (ENSO). The main development region

is the area between 10

N and 20

N, between the coast of Africa and Central America, where most African waves

originate. Each of the 50,000 modeled years is associated with a climate state, making it possible to subset years and

events and resulting model losses corresponding to a given climate state, or range of climate states.

Because these events are generated using the same 50,000 years of simulated sea-surface temperatures as the inland

flood model, they may be directly combined, creating a model that tracks all primary sources of flood in the U.S.,

inland and hurricane-induced. Figure 3 provides a sampling of the KatRisk tropical cyclone track set.

Figure 3: Cyclone Track Set

2.4 APPLYING AND SELECTING SEA LEVEL PROJECTIONS

Beyond modeling insured and uninsured risk under current conditions, this report provides estimates of how this risk

could change as a result of sea level rise. Our aim was not to project sea levels themselves, but rather to extend our

analysis to a future scenario that is plausible and within the range of estimates provided by the contemporary scientific

community. It should be noted that we only sought to estimate the potential for increases in storm surge risk due to

sea level rise.

Potential changes in other sources of flooding arising from precipitation or river flooding extremes under a non-stationary climate would be significantly

harder to estimate. While observations show that “annual precipitation since the beginning of the last century has increased across most of the northern

and eastern United States and decreased across much of the southern and western United States,” (NCA4) the interplay of precipitation with soil moisture,

snow pack, and evapotranspiration is complicated to capture and the uncertainty in estimating the impacts on river floods is an unsolved problem.

Following the IPCC special report (“https://www.ipcc.ch/site/assets/uploads/2019/08/4.-SPM_Approved_Microsite_FINAL.pdf

) “climate change can

According to the Fourth National Climate Assessment:

“Global average sea level has risen by about 7–8 inches (about 16–21 cm) since 1900, with almost

half this rise occurring since 1993 as oceans have warmed and land-based ice has melted. Relative

to the year 2000, sea level is very likely to rise 1 to 4 feet (0.3 to 1.3 m) by the end of the century.

Emerging science regarding Antarctic ice sheet stability suggests that, for higher scenarios, a rise

exceeding 8 feet (2.4 m) by 2100 is physically possible, although the probability of such an extreme

outcome cannot currently be assessed.”

These implications led the National Oceanic and Atmospheric Administration (NOAA) to widen its potential outcomes

in 2100 from previous studies to be between 0.3m and 2.5m of global sea level rise, and formed the basis

of the analysis of potential sea level rise scenarios in this report.

A range of possible outcomes in 2100 for global

sea level rise scenarios of 0.3m, 0.5m, 1.0m, 1.5m, 2.0m, and 2.5m broken down by time horizon between 2020 and

2100, with regionally low, medium, and high estimates form the envelope of possible outcomes.

In addition to current sea levels, selections for this analysis were made for “medium” and “high” sea level rise

scenarios. The medium regional sea level rise by 2050, with 0.5m globally by 2100, was selected as the medium

scenario. The high regional sea level rise by 2050, with 1.5m globally by 2100, was selected as the high scenario.

Once the above selections were determined, the KatRisk storm surge model was modified to reflect the risk under

future sea levels. Because the KatRisk Model stores storm surge events as gridded flood depths above datum, it must

first subtract the local elevation to determine the local flood height above ground. Sea level rise is then simulated by

subtracting local sea level rise from the local elevation. KatRisk then simulates the sea level rise scenarios detailed

above by creating two new exposure data sets, each with their local elevations subtracted by the appropriate value

to simulate each sea level rise scenario. These conditioned exposure datasets are then run through the KatRisk storm

surge and inland flood model. Output is analyzed to obtain AALs and EP curves for all locations and MSA aggregate

areas using the methods and models stated above.

2.5 CREATION OF A MARKET BASKET

To assemble the market basket, Milliman obtained parcel data from county assessor records compiled by a third-party

data vendor. Each parcel has the latitude, longitude, and other attributes of an actual single-family property.

Beyond the characteristics directly obtained from these parcel records, property characteristics were imputed at each

location so that the risk would contain all the necessary input characteristics for catastrophe modeling. To simulate

the value of these characteristics, we used a number of public and private data sources to determine the expected

distribution of each risk characteristic, where risks with certain characteristics are most likely to be located, and the

expected relationships and correlations between characteristic classifications. Data sources include:

 USGS digital elevation maps and FEMA digital flood maps (GIS)

exacerbate land degradation processes (high confidence) including through increases in rainfall intensity, flooding, drought frequency and severity, heat

stress, dry spells, wind, sea level rise and wave action, permafrost thaw with outcomes being modulated by land management.” However, regional impact,

changes in atmospheric patterns as well land management make impact studies very dependent on the underlying assumptions. Many studies that form

the basis of the IPCC and NCA4 assessments show that most likely local precipitation extremes will increase. We therefore choose to alter local

precipitation intensities and asses those impacts by re-scaling local precipitation extremes with the Clausius-Clapeyron equation that relates the impact on

atmospheric water holding capacity with temperature. This gives approximately a 7% increase per 1K temperature increase. Recent studies have found on

a global scale similar trends (e.g. Westra et al., Global increasing trends in annual maximum daily precipitation, J.Clim, 2012).

U.S. Global Change Research Program (November 2018). Fourth National Climate Assessment. Available at: https://www.globalchange.gov/nca4

National Oceanic and Atmospheric Administration (January 2017). Global and Regional Sea Level Rise Scenarios for the United States. Available at:

https://tidesandcurrents.noaa.gov/publications/techrpt83_Global_and_Regional_SLR_Scenarios_for_the_US_final.pdf

 Industry homeowners quote data obtained from a software company

 Public parcel records, compiled by a data vendor

 Residential Energy Consumption Survey (RECS), published 2009

 Multi-hazard Loss Estimation Methodology (HAZUS), current Flood Model Technical Manual developed by

the Federal Emergency Management Agency

 National Association of Insurance Commissioners (NAIC) report: “Dwelling Fire, Homeowners Owner-

Occupied, and Homeowners Tenant and Condominium/Cooperative Unit Owner’s Insurance Report: Data for

2015,” published 2017

Table 1: Market Basket Attributes and Data Sources

Characteristic

Data Source

Year Built

Parcel Records

Construction

Industry Quote Data

Number of Stories

Industry Quote Data

Foundation Type and First Floor Height

Parcel Records, GIS,

RECS, and HAZUS

Building Value and Limit

Parcel Records and NAIC

Site Deductible

Industry Quote Data

2.6 ESTIMATING NFIP TAKE-UP RATE

Once the market basket was developed, we assigned to each risk an assumption as to whether an NFIP policy was

present using the following process. First, we used geocoding and FEMA flood maps to determine whether each risk

in the market basket was inside or outside an SFHA. We then used the NFIP OpenFEMA data, which provides NFIP

policies in force by state, both inside and outside the SFHA, as well as census data, which provides the number of

single-family residences in the same areas, to determine the probability than any given home would have NFIP

coverage for each combination of SFHA and state. The resulting estimated NFIP take-up rates are shown in Figure 4.

Countrywide, we estimate that approximately one-third of homes in the SFHA have an NFIP policy, with the majority

of states having less than a 25% take-up rate in the SFHA. Outside the SFHA, every state except Louisiana, Florida, and

Texas has take-up rates of approximately three percent or less, with the majority being less than one percent.

Figure 4: NFIP Take-Up Rate Estimates

For detailed calculations and data sources, see Appendix, Exhibit 7. South Carolina not shown.

10%

20%

30%

40%

50%

60%

NFIP Take-Up Rate Estimates

Non SFHA SFHA Total

We produced the estimates of insured and uninsured losses in the sections that follow by applying the KatRisk model

and NFIP take-up rate assumptions to the market basket. At the MSA level, our estimates of insured and uninsured

losses are thus dependent on the state(s) in which each MSA is located, and the proportion of single-family residences

in and out of the SFHA for that MSA. Thus, the take-up rates used throughout this study reflect the unique risk profile

of each MSA based on the location of residences and insurance purchasing patterns in their state(s). However, it

should be noted that to the extent that an MSA has unusually high or low take-up rates in the SFHA relative to the

remainder of the state, our results for that MSA may be biased.

Our methodology for take-up rate estimation was selected because the intersection of flood maps, parcel data, and

OpenFEMA data provides for highly variable results below the state level. Some of these results appear to be driven

by data limitations. We found that aggregating to the state level provided more interpretable and stable results. Even

when aggregating to the state and SFHA level, estimates of take-up rates in the SFHA for one state, South Carolina,

were significantly higher than any other state and did not appear credible. We capped South Carolina take-up rates

in the SFHA at the maximum for all other states.

2.7 THE RESIDENTIAL PRIVATE FLOOD INSURANCE MARKET

Recent catastrophes have demonstrated low rates of insurance coverage in affected areas,

and uninsured losses are

often not fully addressed by post-disaster assistance.

Statistics for residential and even single-family homes insured

are published by the NFIP, yet public information on the growing private flood market is scarce.

Best estimates indicate that that the majority of reported premiums for private flood are for commercial lines, and

that for residential insurance, the ratio of private to NFIP flood writings is small. To benchmark our own estimates of

the size of the flood insurance market, and to validate our assumption that private residential flood insurance is

relatively immaterial in terms of its impact on the insurance gap today, we examined a number of estimates by

independent sources and tabulated them in in Exhibit 1 on the following page. Each estimate shown quantifies some

aspect of the residential or private flood insurance market. Because the flood insurance market is growing steadily

and changing over time, it may be important to consider the time at which any estimate was made in addition to the

estimate itself.

Milliman (July 2018). Available at: https://milliman-cdn.azureedge.net/-/media/milliman/importedfiles/uploadedfiles/insight/2018/ny-nj-market

feasibility.ashx

Congressional Budget Office (April 2019). Expected Costs of Damage From Hurricane Winds and Storm-related Flooding. Available at:

https://www.cbo.gov/system/files/2019-04/55019-ExpectedCostsFromWindStorm.pdf

Exhibit 1: Estimates of Private Flood Insurance Written

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Private Flood compared to NFIP

Based on Estimates of Residential Percent of Flood Market

Statutory Accounting Data (Note 2) Wharton III Survey NAIC Survey

NFIP Private Private Flood WSIA Carrier Management Private % of % of Homeowners % of Residences

Earned Flood as % (Note 3) (Note 4) Total Policies with with

Premium Earned Premium

of Private +NFIP

Residential Residential Residential Primary

Flood Insurance

Year (Note 1) (Note 2) (3)/[(2) + (3)] Residential % of Total Residential % of Total (Note 5) (Note 6) (Note 7)

2013 $3,512,987 14%

2014 3,542,525 14%

2015 3,436,750 14%

2016 3,332,142 $239,456 6.7% $120,224 32.0% 12%

2017 3,308,151 599,562 15.3% $220,000 34.9% 3.5% to 4.5%

2018 3,327,327 659,467 16.5% 161,432 34.0% 213,000 31.3% 15%

2019 17%

Notes:

1. https://www.fema.gov/total-earned-premium-calendar-year. Dollar amounts in thousands.

2. (3) from SNL.com database of statutory annual statement for P&C industry. Dollar amounts in thousands.

3. Wholesale and Surplus Lines Insurance Association, Surplus Lines Flood Insurance Market Data and Statistics. January 24, 2017. and February 28, 2019

https://www.wsia.org/docs/PDF/Legislative/Surplus%20Lines%20Market%20Data%20and%20Statistics%201-24-17%20w%20attachement.pdf

https://www.wsia.org/docs/PDF/Legislative/SurplusLinesMarketDataandStatistics2-28-19.pdf

4. Carrier Management - Private Flood Insurance Report - 2019. Dollar amounts in thousands.

https://www.insurancejournal.com/research/app/uploads/2019/06/FINAL-Private-Flood-Insurance-Report-2019.pdf

5. Kousky, et al. "The Emerging Private Residential Flood Insurance Market in the United States" Wharton Risk Managament and Decision Process Center.

http://www.floods.org/ace-files/documentlibrary/committees/Insurance/Emerging_Flood_Insurance_Market_Report-Wharton-07-13-18.pdf

6. Insurance Information Institute (III) - Insurance Factbook 2019, based on Pulse surveys conducted by III.

https://www.iii.org/sites/default/files/docs/pdf/insurance_factbook_2019.pdf

7. https://naic.org/Releases/2019_docs/naic_survey_flood_insurance.htm

A 2017 report from Wharton provided a range of 3.5% to 4.5% (column 9) as the percentage of total residential flood

insurance policies attributable to the private market. NFIP take-up rates countrywide were approximately 3.5% in

2018 for single-family homes.

Accounting for the high end of Wharton’s range suggests that no more than 3.7%

(1.045 * 3.5%) of single-family homes in the United States have flood insurance of any kind, far below the recent

Insurance Information Institute (III) and NAIC survey results of 15% and 17%, shown in columns 10 and 11 above.

While the surveys shown are for slightly different sets of the residential market, single-family homes represent most

of the housing stock in the United States. We note that these survey estimates appear to reflect consumer sentiments,

and it is possible that the results indicate an incorrect belief among some uninsured consumers that flood coverage

is in place. The NAIC has commented that it also expects this to be the case.

We compared the 3.7% single-family flood insurance take-up rate derived above to estimates of the surplus lines and

admitted residential flood markets shown in columns 5 and 7. Summing these premiums, we estimate that the private

residential flood market premium totaled about $374 million in 2018, close to 10% of the total residential flood market

premium when including the NFIP. This is about twice as high as the Wharton estimate, but not necessarily in conflict,

as the Wharton estimate is based on policy counts and not premium.

Whether looking at policy counts or premium, public data supports the notion that the size of the private residential

flood market relative to the NFIP is small as assumed in this report. We find that all data and independent estimates

provide corroborative evidence that despite encouraging recent growth, the private residential flood insurance

market remains relatively small, with almost all homeowners flood policies currently written by the NFIP. Thus, while

our characterization of uninsured flood risk in this report could be overstated because it relies on the assumption that

NFIP is the only provider of residential flood insurance, the small size of this market today means that any bias resulting

from this estimate should also be small.

NFIP single-family policies in 2018 (which include mobile homes) were 3.54 million in 2018 for the United States

(https://bsa.nfipstat.fema.gov/reports/w2rpcnta.htm), compared to 99.97 million single-family and mobile homes in the 2017 five-year American

Community Survey, provided by the United States Census Bureau. 3.54 / 99.97 = 3.5%

As quoted in the NAIC press release (https://naic.org/Releases/2019_docs/naic_survey_flood_insurance.htm) by Eric Cioppa, NAIC president and

superintendent of the Maine Bureau of Insurance, "This disparity perhaps reflects the common, though incorrect, assumption that homeowners insurance

covers flooding."

3. Results: Estimates of Insured and Uninsured Flood Exposure to U.S. Homes

The extreme catastrophic nature of floods can make it problematic to project future losses based on past flood events,

particularly in the face of changing hazard presented by sea level rise. We used a catastrophe simulation model to

produce estimated flood losses for each location in the market basket, with our resulting dataset providing two key

features that cannot be obtained using only historical flood data: 1) An estimate of total losses, including losses for

homes not insured and 2) a projection of future expected losses using future sea level rise scenarios.

3.1 INSURED AND UNINSURED LOSS RESULTS

The output of this process is detailed in Exhibit 8 of the Appendix, which shows modeled output of average annual

losses, the resulting portions that are expected to be insured versus uninsured, and an ultimate estimate of the

percent of losses uninsured for all 380 MSAs in the study area.

We estimate that losses to single-family residences

due to flood are expected to cost more than $7 billion annually, with more than 87% those losses being uninsured by

the NFIP. The combination of insured and uninsured flood losses are significant relative to other perils typically

covered for single-family residences, such as fire, non-flood water, and wind.

Countrywide, the average homeowner

has $78 of expected annual flood losses to their building alone. Costs incurred for losses to contents, additional living

expenses (ALE), and losses to other structures aside from the primary residence would be in addition to the $78

average.

Individual and community-level flood risk is skewed, with portions of the populations having flood risk that is orders

of magnitude higher than others. Even when aggregated at the MSA level, our findings indicate that many areas have

high average expected losses and are mostly uninsured. As seen in Figure 5

, an estimated 69% of MSAs in the United

States have 90% or more of losses uninsured, with approximately 5% of MSAs having more than 30% of their expected

flood losses insured. The MSAs with the highest portion of insured loss are mostly in Florida and Louisiana, with a

handful of others in Atlantic states and Texas. MSAs range from mostly uninsured when total expected flood losses

are high, and nearly entirely uninsured in areas that are generally not exposed to storm surge losses from tropical

cyclones

Multiplying weighted average claim frequency and severity for homeowners from 2013 to 2017 results in estimated average losses of $191, $214, and

$210 for the perils of fire and lightning, wind and hail, and water damage and freezing, respectively. In addition to building losses, these estimates include

losses to contents, additional living expenses, and losses to other structure. Source: Insurance Information Institute. Facts + Statistics: Homeowners and

renters insurance, Average Homeowners Losses, 2013-2017 (1), referencing ISO®, a Verisk Analytics® business. Available at:

https://www.iii.org/fact-

statistic/facts-statistics-homeowners-and-renters-insurance

Using OpenFEMA data, we estimate that contents coverage only accounts for 16% of the total building and contents losses paid by the NFIP since 2010,

and also only accounts for 15% of the insured limits among policies that had losses. This shows that contents losses are roughly the same, relative to their

NFIP insured limits, as building losses. Extrapolating to building value suggests that a homeowner with contents equal to 50% of their building value, for

example, would shave expected building and contents flood losses roughly 50% higher than just building flood losses. ALE and other structures-specific flood

loss data are not available from this data.

Figure 5: Distribution of MSAs by Uninsured Loss Percent

With regard to the magnitude of both insured and uninsured losses, 27 of the 48 states studied have at least one

entire MSA with average expected losses of more than $100 per single-family residence. This statistic does not mean

that the other 21 states have low flood risk, as they tend to be lower-population states with a limited number of MSAs,

such as Rhode Island, Vermont, and Wyoming, which only have four MSAs designated between them. Further

supporting the fact that flood is a countrywide peril, six of the top 20 MSAs by total expected annual flood losses are

in the West or Midwest.

69%

19%

0% 10% 20% 30% 40% 50% 60% 70% 80%

90% to 100%

80% to 90%

70% to 80%

60% to 70%

Less than 60%

Percent of MSAs

Uninsured Loss Percent

Distribution of MSAs by Uninsured Loss Percent

3.2 IMPACTS OF SEA LEVEL RISE

Focusing on exposure to storm surge, we sought to understand how uninsured losses would be impacted by

changes in sea level rise. Table 2 shows that, under each sea level rise scenario, the percent of losses uninsured is

relatively unchanged. The high percentage of losses uninsured is not expected to improve in the future unless

current flood insurance purchase patterns change. Sea level rise will cause the amount of uninsured losses to

increase on an order similar to total losses. With uninsured losses already relatively high, homeowners will be faced

with either increasing uninsured losses or paying for flood insurance policies that they do not have today.

Table 2: Summary of Total Annual Storm Surge Losses by Sea Level Rise Scenario

Annual Storm Surge Losses

Current Sea Levels

Medium

Sea

Level

Rise

High

Sea

Level

Rise

Insured (millions)

$601

$728

$989

Uninsured (millions)

$1,776

$2,155

$2,949

Total (millions)

$2,376

$2,883

$3,938

Uninsured Percent of Total

75%

Change in Total Relative to

Current Sea Levels

N/A

21%

66%

In addition to calculations supporting Table 2 above, MSA level results for storm surge losses by sea level rise

scenario can be found in Exhibits 9 to 12 of the Appendix. Under our medium sea level rise scenario, total storm

surge losses increase an average of 21% and up to a maximum of just over 50% by MSA. In a high sea level rise

scenario these numbers increase to 66% on average, and a maximum of over 200%.

These statistics indicate a high

sensitivity in future storm surge losses due to sea level rise, as the increase in total storm surge losses between the

medium and high scenario averages over 300%.

Figure 6 further illustrates how sea level rise impacts will vary regionally. At current sea levels, we estimate that the

New Orleans-Metairie MSA has the most expected losses due to storm surge. In a medium sea level rise scenario,

Miami-Fort Lauderdale-Pompano Beach would have the highest expected losses, and for the high sea level rise

scenario it would be the New York-Newark-Jersey City MSA.

Figure 6: Storm Surge Losses by MSA and Sea Level Rise Scenario

To view sea level rise impacts visually, we developed Figure 7, which maps the increase in losses due to the impacts

of sea level rise on total flood losses for several Middle Atlantic states. The increase in losses are spatially smoothed

to show how the increase in total flood losses changes with geography, even in areas that may not be represented in

the market basket. Maps for other regions of the country and a description of the spatial smoothing process are

provided in the Appendix.

Lafayette, LA

Sebastian-Vero Beach, FL

Myrtle Beach-Conway-North Myrtle Beach, SC-NC

Beaumont-Port Arthur, TX

Boston-Cambridge-Newton, MA-NH

North Port-Sarasota-Bradenton, FL

Houma-Thibodaux, LA

Punta Gorda, FL

Virginia Beach-Norfolk-Newport News, VA-NC

Jacksonville, FL

Ocean City, NJ

Charleston-North Charleston, SC

Houston-The Woodlands-Sugar Land, TX

Naples-Marco Island, FL

Tampa-St. Petersburg-Clearwater, FL

Hilton Head Island-Bluffton, SC

Cape Coral-Fort Myers, FL

New Orleans-Metairie, LA

Miami-Fort Lauderdale-Pompano Beach, FL

New York-Newark-Jersey City, NY-NJ-PA

Total Storm Surge Average Annual Loss by Sea Level Rise Scenario

MSA

Total Average Annual Storm Surge Losses

Highest 20 MSAs under High Sea Level Rise Scenario

Current Medium High

Figure 7: Increase in Total Flood Losses: High vs. Medium Sea level Scenarios, Middle Atlantic States

While the Mid-Atlantic states tend to have lower storm surge losses than lower Atlantic and Gulf states, the increase

in total flood losses due to sea level rise is estimated to be the highest from North Carolina to New York (above and

Maps 3 to 5 in the Appendix).

By appending census estimates of median household income to the market baskets, we estimated the magnitude of

the costs of flooding relative to current incomes. Table 3 shows these results for selected MSAs exposed to storm

surge. The resulting estimates show the average ratio of location level losses to income, by MSA, under each sea level

rise scenario.

AALs for both storm surge and inland flood are important metrics and are increasingly used by insurers to set prices

for flood insurance premiums. For insurers that do so, changes in AAL would be expected to be correlated with

changes in insurance premiums. Homeowners either would bear the cost of increased prices in insurance premiums

or increased flood losses depending on whether or not they purchase insurance.

The ratio of AAL to median income is highly skewed, with a few MSAs having relatively high ratios. Table 3 illustrates

this, showing the 20 highest ratios at current sea levels, and the associated expected increases as sea levels rise. Flood

is a peril that can present extreme risk in certain areas, which means that every MSA has homeowners that will have

high costs of flooding relative to income. However, these results point to which MSAs may be particularly subject to

either high insurance premiums, possible flood losses, or both. At current sea levels, only Hilton Head Island-Bluffton,

SC is estimated to have AALs average over 2% of median incomes, but In the medium sea level rise scenario, Houma-

Thibodaux, LA and Punta Gorda, FL are also estimated to have AALs exceed 2% of median incomes.

Table 3: Ratio of Average Annual Loss to Census Block Group Median Household Income – Averaged by MSA

MSA

Current Sea Levels

Medium Sea Level

Rise

High Sea Level Rise

Hilton Head Island-

Bluffton, SC

2.13%

2.52%

3.29%

Houma-Thibodaux, LA

1.87%

2.50%

3.32%

Punta Gorda, FL

1.89%

2.21%

2.78%

Naples-Marco Island, FL

1.37%

1.65%

2.16%

Cape Coral-Fort Myers,

1.24%

1.48%

1.88%

New Orleans-Metairie,

1.25%

1.39%

1.57%

Ocean City, NJ

0.88%

1.26%

2.21%

Beaumont-Port Arthur,

0.75%

0.86%

1.00%

Jacksonville, NC

0.73%

0.83%

1.04%

Lake Charles, LA

0.66%

0.80%

0.98%

Homosassa Springs, FL

0.63%

0.71%

0.86%

Sebastian-Vero Beach, FL

0.46%

0.54%

0.74%

Lafayette, LA

0.38%

0.48%

0.58%

Myrtle Beach-Conway-

North Myrtle Beach, SC-

0.37%

0.42%

0.54%

Wilmington, NC

0.36%

0.40%

0.50%

Charleston-North

Charleston, SC

0.31%

0.40%

0.59%

North Port-Sarasota-

Bradenton, FL

0.34%

0.38%

0.50%

For detailed calculations, data sources, and a complete table for storm surge exposed MSAs, see Appendix, Exhibit 13

The risk of extreme events has always had a major impact on the private flood insurance market. This risk, and the

inability to calculate it, has been part of the reason why private flood insurers have not offered significant amounts

of flood coverage in the past. All else equal, affordability and availability of insurance premiums will likely be

adversely impacted as the risk of extreme events occur in an area, as insurers and reinsurers will raise rates or

refuse to accept risks to preserve profitability and solvency.

Exhibit 14 of the Appendix shows increases in the risk of extreme (one-in-500 year)

events under the medium and

high scenarios. Similar to AAL changes, changes in the costs of extreme events are highly correlated between the

medium and high scenarios but are also highly sensitive to sea level assumptions.

The average increase in 500-year

losses with high sea level rise tends to be about three times higher than with medium sea level rise.

Figure 8 below shows the percent change in these extreme flood events for the MSAs with the highest increase in

expected flood losses from extreme events. While Florida and Louisiana have some of the highest risk of storm surge

today, the greatest increases in losses for extreme events tend to be along the Atlantic Coast north of Florida.

The

eight MSAs with the largest percent increases in either sea level rise scenario are in all between Georgia and New

Jersey. New Jersey MSAs are particularly vulnerable, with two showing the highest percent increases in the medium

sea level rise scenario among all MSAs.

One-in-500-year scenarios are based on the OEP curve of event losses. The one-in-500-year loss for a given region is the event loss that has a chance of

being exceeded with an annual probability of 0.2% (1/500).

Figure 8: Change in 500-year Return Period Flood Losses for Sea Level Rise Scenarios Compared to Current Sea Levels

0% 20% 40% 60% 80% 100% 120% 140% 160% 180% 200%

Gulfport-Biloxi, MS

Houma-Thibodaux, LA

Miami-Fort Lauderdale-Pompano Beach, FL

Naples-Marco Island, FL

Lafayette, LA

Jacksonville, FL

Corpus Christi, TX

Lake Charles, LA

Myrtle Beach-Conway-North Myrtle Beach, SC-NC

Hinesville, GA

Norwich-New London, CT

Barnstable Town, MA

Charleston-North Charleston, SC

New Bern, NC

Virginia Beach-Norfolk-Newport News, VA-NC

Brunswick, GA

Savannah, GA

Ocean City, NJ

Salisbury, MD-DE

Atlantic City-Hammonton, NJ

Percent Change in 500 Year Return Period Flood Losses

Highest 20 MSAs under High Sea Level Rise Scenario

Medium High

3.3 SUMMARY OF RESULTS BY MSA

Compiling the model output and metrics developed, we take a broader look at the different risks faced by MSAs with respect to sea level rise in Exhibit 2. The

selected scenario in this case is medium sea level rise.

Exhibit 2: Summary of Flood Losses by MSA – Medium Sea level Rise vs. Current Sea Levels

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Exposure Metrics under Current Sea Level Exposure Increase Metrics under Medium Sea Level Rise

Total Percent of AAL to Percent Percent Increase in

Total Annual Percent of Losses AAL to Income Percent Increase in Percent Increase in 500 Year

Metropolitan Annual Losses Losses Uninsured Income Ratio Increase in Total Loss 500 Year Return Period Loss

Statistical Area Title Losses Rank Uninsured Rank Ratio Rank Total Loss Rank Return Period Loss Rank

(Note 1) (Note 2) (Note 3) (Note 4) (Note 5) (Note 6) (Note 7) (Note 8) (Note 9) (Note 10) (Note 11)

Atlantic City-Hammonton, NJ $4,761,940 54 62.8% 58 0.1% 51 44.1% 1 42.9% 1

Ocean City, NJ 45,806,409 18 72.7% 46 0.9% 7 42.8% 2 30.1% 2

Salisbury, MD-DE 15,973,654 37 60.0% 62 0.1% 43 40.3% 3 29.3% 3

Houma-Thibodaux, LA 55,876,027 15 72.3% 47 1.9% 3 33.6% 4 18.3% 13

Brunswick, GA 7,657,390 47 87.1% 22 0.3% 21 30.8% 5 28.4% 5

Corpus Christi, TX 7,255,456 48 90.2% 10 0.1% 45 29.2% 6 19.2% 11

Charleston-North Charleston, SC 62,080,754 14 67.5% 52 0.3% 17 28.0% 7 22.7% 6

Virginia Beach-Norfolk-Newport News, VA-NC 55,407,985 16 78.2% 37 0.1% 41 28.0% 8 20.9% 8

Savannah, GA 11,081,849 43 86.1% 25 0.1% 39 27.3% 9 28.8% 4

New Bern, NC 4,444,006 55 78.1% 38 0.2% 28 25.7% 10 21.2% 7

Lafayette, LA 26,692,981 25 61.6% 61 0.4% 13 23.6% 11 19.4% 10

Jacksonville, FL 68,047,575 12 80.8% 32 0.2% 26 22.7% 12 13.4% 18

Vineland-Bridgeton, NJ 788,067 61 51.7% 63 0.0% 58 21.8% 13 11.9% 21

Lake Charles, LA 20,273,215 32 67.8% 51 0.7% 10 21.8% 14 20.8% 9

Sebastian-Vero Beach, FL 26,478,775 26 93.0% 5 0.5% 12 20.9% 15 10.7% 30

Naples-Marco Island, FL 96,781,625 8 64.7% 55 1.4% 4 20.2% 16 14.7% 16

Cape Coral-Fort Myers, FL 191,457,026 5 66.5% 54 1.2% 6 18.9% 17 10.9% 27

Hilton Head Island-Bluffton, SC 143,457,248 6 88.7% 16 2.1% 1 17.8% 18 10.8% 29

Hinesville, GA 721,087 63 98.6% 2 0.1% 54 17.7% 19 18.5% 12

Miami-Fort Lauderdale-Pompano Beach, FL 296,245,156 2 77.4% 41 0.3% 20 17.2% 20 13.6% 17

Punta Gorda, FL 68,115,329 11 63.7% 57 1.9% 2 17.1% 21 8.6% 37

Barnstable Town, MA 14,117,678 39 94.0% 3 0.2% 30 16.7% 22 15.9% 14

Beaumont-Port Arthur, TX 42,592,102 19 90.9% 8 0.7% 8 16.2% 23 11.0% 25

Deltona-Daytona Beach-Ormond Beach, FL 29,902,256 22 77.4% 40 0.2% 24 15.8% 24 11.1% 24

North Port-Sarasota-Bradenton, FL 63,139,612 13 68.1% 50 0.3% 16 14.8% 25 10.9% 26

Gulfport-Biloxi, MS 16,960,076 35 83.8% 30 0.3% 19 14.5% 26 12.1% 20

Panama City, FL 3,592,699 56 79.8% 34 0.1% 42 14.5% 27 10.8% 28

New York-Newark-Jersey City, NY-NJ-PA 389,971,283 1 85.4% 27 0.1% 40 14.4% 28 7.6% 38

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 35,561,772 21 72.8% 45 0.4% 14 14.4% 29 13.3% 19

Jacksonville, NC 24,365,009 29 69.7% 48 0.7% 9 14.2% 30 5.1% 46

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Exposure Metrics under Current Sea Level Exposure Increase Metrics under Medium Sea Level Rise

Total Percent of AAL to Percent Percent Increase in

Total Annual Percent of Losses AAL to Income Percent Increase in Percent Increase in 500 Year

Metropolitan Annual Losses Losses Uninsured Income Ratio Increase in Total Loss 500 Year Return Period Loss

Statistical Area Title Losses Rank Uninsured Rank Ratio Rank Total Loss Rank Return Period Loss Rank

(Note 1) (Note 2) (Note 3) (Note 4) (Note 5) (Note 6) (Note 7) (Note 8) (Note 9) (Note 10) (Note 11)

Crestview-Fort Walton Beach-Destin, FL 8,616,116 46 79.2% 35 0.1% 38 13.7% 31 11.5% 22

Tampa-St. Petersburg-Clearwater, FL 136,837,328 7 66.9% 53 0.3% 22 13.4% 32 11.1% 23

Wilmington, NC 23,194,192 30 84.0% 29 0.4% 15 13.3% 33 9.8% 32

Palm Bay-Melbourne-Titusville, FL 27,107,739 24 89.9% 11 0.2% 25 13.0% 34 9.1% 34

Daphne-Fairhope-Foley, AL 13,349,807 41 88.4% 18 0.3% 18 12.6% 35 10.1% 31

Norwich-New London, CT 6,890,794 49 82.6% 31 0.1% 47 12.2% 36 14.7% 15

Baton Rouge, LA 19,909,601 33 64.6% 56 0.1% 35 12.0% 37 8.8% 36

Homosassa Springs, FL 13,655,529 40 68.3% 49 0.6% 11 11.8% 38 5.6% 41

New Haven-Milford, CT 25,364,184 27 87.1% 21 0.1% 37 11.6% 39 9.7% 33

New Orleans-Metairie, LA 258,015,593 3 74.5% 43 1.3% 5 11.4% 40 5.2% 44

Brownsville-Harlingen, TX 5,767,671 51 89.4% 14 0.2% 33 11.1% 41 4.1% 50

Port St. Lucie, FL 22,742,400 31 78.3% 36 0.2% 23 11.0% 42 5.3% 43

Dover, DE 2,120,168 59 88.8% 15 0.1% 52 10.9% 43 0.9% 55

Mobile, AL 14,267,187 38 88.4% 19 0.2% 27 10.6% 44 7.6% 39

Portland-South Portland, ME 24,652,928 28 93.7% 4 0.2% 29 10.6% 45 9.0% 35

California-Lexington Park, MD 2,072,929 60 73.2% 44 0.1% 55 10.4% 46 5.3% 42

Boston-Cambridge-Newton, MA-NH 87,225,356 9 89.8% 12 0.1% 49 9.5% 47 7.3% 40

Bridgeport-Stamford-Norwalk, CT 39,333,937 20 86.1% 26 0.1% 44 9.1% 48 5.2% 45

Houston-The Woodlands-Sugar Land, TX 212,376,082 4 90.8% 9 0.2% 31 8.7% 49 4.9% 47

Hartford-East Hartford-Middletown, CT 16,774,956 36 88.3% 20 0.1% 56 6.9% 50 4.4% 48

Hammond, LA 2,655,297 57 61.7% 60 0.2% 32 6.3% 51 2.4% 52

Pensacola-Ferry Pass-Brent, FL 10,185,699 45 76.7% 42 0.1% 46 6.0% 52 4.1% 49

Tallahassee, FL 5,487,041 53 62.7% 59 0.1% 48 5.6% 53 3.6% 51

Baltimore-Columbia-Towson, MD 18,761,905 34 89.7% 13 0.0% 63 3.5% 54 1.6% 53

Providence-Warwick, RI-MA 11,708,224 42 86.9% 23 0.0% 59 3.5% 55 1.5% 54

Gainesville, FL 6,245,276 50 80.8% 33 0.2% 34 2.5% 56 0.4% 56

Washington-Arlington-Alexandria, DC-VA-MD-WV 55,383,798 17 88.6% 17 0.0% 61 2.5% 57 0.1% 57

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD 74,242,296 10 91.6% 7 0.1% 57 0.6% 58 0.0% 58

Orlando-Kissimmee-Sanford, FL 28,564,743 23 86.6% 24 0.1% 53 0.1% 59 0.0% 60

Richmond, VA 10,827,098 44 85.2% 28 0.0% 60 0.1% 60 0.0% 59

Kingston, NY 5,531,774 52 77.9% 39 0.1% 36 0.0%

61 0.

0% 61

Bangor, ME 743,674 62 100.0% 1 0.0% 62 0.0% 62 0.0% 63

Greenville, NC 2,244,732 58 92.8% 6 0.1% 50 0.0% 63 0.0% 62

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Column (2) = Exhibit 8, Column (8)

3. Rank based on total annual losses for MSAs with storm surge exposure.

4. Column (4) = Exhibit 8, Column (9). Insured and Uninsured Losses are based on estimates of NFIP take-up rates and coverages.

5. Rank based on percent of losses uninsured for MSAs with storm surge exposure.

6. Column (6) = Exhibit 13, Column (3)

7. Rank based AAL to income ratio for MSAs with storm surge exposure.

8. Column (8) = Ratio of inland flood and storm surge losses, medium to current scenario

9. Rank based percent increase in total loss for MSAs with storm surge exposure.

10. Column (10) = Exhibit 14, Column (6)

11. Rank based on percent increase in 500 year return period loss for MSAs with storm surge exposure.

In general, an area will be most affected by sea level rise if both the current exposure and the increase in that exposure

are large. Areas with high current exposure but manageable increases in exposure may be able to adapt to sea level

rise, assuming current strategies of flood risk management are in place and can scale. MSAs with low current exposure

and high increases in exposure may find that implementing new risk-management strategies presents a high value-

add proposition at a feasible cost. However, the remaining MSAs may find unique issues due to the incremental

change in flood risk.

This study illustrates a significant public policy issue: the protection gap of uninsured flood risk is massive. Our results

show that all MSAs have room to greatly increase flood insurance take-up rates and mitigate the disruptive impacts

of uninsured flood losses. MSAs that already have moderate to high AAL to income ratios may find this strategy

difficult, particularly in the face of high increasing AAL and extreme event losses due to sea level rise as the

accompanying affordability and availability of insurance is adversely impacted. Absent an ability or willingness to

finance increasingly expensive flood costs, some areas may be subject to strategies or impacts involving relocation of

residents. These include managed retreat via policies encouraging or requiring relocation away from more flood-

prone areas, and climate gentrification, which occurs as demand for property increases in traditionally lower-income

areas that are less flood-prone.

As examples of the above, our results show that the Atlantic City-Hammonton, Ocean City, and Salisbury MSAs all top

our lists of exposure increase under medium sea level rise. Atlantic City-Hammonton and Salisbury each have relatively

low annual expected losses today both in an absolute context and relative to household incomes. Coupled with a

relatively high rate of insured losses, the area may be able to absorb sea level rise impacts more readily than most.

Ocean City has high losses relative to the number of residences and incomes. Increasing the insurance take-up rates

could be difficult due to affordability issues, and uninsured losses will be more difficult to absorb by homeowners than

the other MSAs discussed in this example.

The MSA level results provide an opportunity to assess how the varying degree of flood risk and exposure to sea level

rise can pose unique challenges to local region’s risk management strategies. They also provide the ability to

understand the magnitude of flood risk at a level that can be readily absorbed. However, decisions on managing flood

risk will be made from the federal to the local level, and from the insurance industry to the consumer.

4. Projections: Mortgage Default Risks After Catastrophic Flooding

When flood insurance is in place, a portion of incurred losses are absorbed by the NFIP, private flood insurers, and

reinsurers. If flood insurance is not in place, state or federal government programs may provide financial disaster

assistance, but these programs are not always available and often provide only partial indemnification. To the extent

that insurance is not in place or assistance is not available, losses are absorbed by the homeowner, either through

out-of-pocket repair costs or through diminishment in home equity if they must finance necessary repairs.

The homeowner is not the only stakeholder in this case. If a homeowner has a mortgage on their residence, diminished

value of the collateral property securing it could increase the risk of delinquency or default. As recent flood

catastrophes

have occurred in non-SFHA areas where flood insurance purchase is exceedingly rare, new attention

has been brought to the relationship between the flood insurance gap and potential mortgage credit risk.

The lack

of flood coverage among U.S. homeowners despite the risk, given the importance of hazard insurance for managing

mortgage risk, indicates that there could be a significant degree of exposure to catastrophic flooding that is borne

Hunn, D. (March 30, 2018). Harvey’s Floods. Houston Chronicle. Available at: https://www.houstonchronicle.com/news/article/In-Harvey-s-deluge-most-

damaged-homes-were-12794820.php

Koning Beals, R. (November 2, 2019). Banks increasingly unload flooded-out mortgages at taxpayer expense. MarketWatch. Available at:

https://www.marketwatch.com/story/climate-change-could-impact-your-mortgage-even-if-you-live-nowhere-near-a-coast-2019-09-30

within the U.S. mortgage system. Though it is apparent that this exposure may exist, estimates of its magnitude and

potential severity are elusive due to sporadic data on historical catastrophic events, lack of detailed public information,

lack of information about historical impacts to household finances after these events, and historically inconsistent

government and lender assistance. Since there is no guarantee that ad hoc government and lender assistance

programs will be in place in the future, past events may provide poor predictors of future ones, potentially resulting

in incomplete estimates of the total exposure of credit losses to extreme flood events.

In this section, simulated property losses from extreme flood events are used to stress test and quantify the potential

downstream financial impacts of a natural catastrophe on a mortgage credit risk portfolio, and how these costs could

be passed on to lenders, GSEs, mortgage guaranty insurers, and investors. While countrywide diversification mitigates

the size of default costs from extreme flood events on a percentage basis, it should be noted that homeowners

themselves are unable to diversify nationwide similar to the diversification benefits of a larger mortgage pool.

4.1 POTENTIAL BEARERS OF FLOOD RISK: BEYOND HOMEOWNERS AND INSURERS

In order to understand how the U.S. single-family mortgage industry

bears flood risk, an understanding of its market

participants is necessary. This section provides an overview of the U.S. mortgage market participants and discusses

two representative credit risk transfer (CRT) deals from Freddie Mac.

When a homebuyer borrows money from a lender to purchase a home, the mortgage loan is collateralized by the

home being purchased. This means the owner of the mortgage has a lien on the property and can seek recourse by

seizing the property if principal and interest are not repaid consistent with the terms of the mortgage. In general, a

mortgage has three types of risk: credit risk, interest rate risk, and prepayment risk. The risk that a borrower stops

paying their mortgage is known as credit risk, the risk that mortgage principal and interest will not be paid back due

to borrower (mortgagor) default. Interest rate risk refers to the risk that mortgage asset prices (that is, the present

value of future cash flows from the mortgage loan) fall as interest rates rise. Prepayment risk refers to the risk that

the borrower will prepay their mortgage and the investor will receive less interest income than expected (the principal

will be returned earlier). In the U.S. mortgage market, each of these risks can be separated and distributed to different

investors with different investment strategies and risk appetite.

For credit risk, the mortgage contract allows the lender (mortgagee) to foreclose on and sell the underlying property

to recover the amount owed on the loan. Therefore, the value of the underlying property directly impacts the party

owning the mortgage asset. Historical data demonstrates that changes in home prices, particularly when they decline,

are one of the most significant drivers of mortgage credit risk. During this discussion and analysis, flood risk is tied to

mortgage credit risk using the following logic: Flood losses that are uninsured could reduce the value of a property

and therefore impact the holder of the mortgage. If a property is affected by a flood, the property could experience

a decrease in value, absent subsequent repair. If the borrower stops paying their mortgage, the value of the collateral

securing the loan would be worth less than it would be absent the event, and the mortgagee could potentially incur

a loss upon sale of the property.

In the U.S. mortgage market, the originator of a given mortgage often sells the mortgage to a third party or obtains

mortgage guaranty insurance to protect themselves against credit risk. In order to understand the impact of flood on

Multifamily, 5+ units, is not evaluated in this analysis.

Choice of Freddie Mac CRT deals based on a review of reference pools and structures. Performing this analysis on Fannie Mae CAS/CIRT transactions

would likely lead to similar conclusions.

the mortgage market and the ultimate bearers of flood risk, the following section provides a discussion of the holders

of mortgage credit risk in the US mortgage market.

Credit Risk Investors in the U.S. Mortgage Market

Mortgages can be originated through a variety of channels. For example, a borrower can go to the physical branch of

their bank to obtain a mortgage (retail), a specialized mortgage company that then sells the mortgage to larger entity

(correspondent), or a specialized mortgage company that originates the mortgage based on the best terms available

from multiple lenders (broker).

In all cases, once the loan is originated, the mortgage can be retained by the lender,

sold as a whole loan to an investor, or securitized along with other similar assets and sold via a mortgage-backed

security (MBS) to investors.

Figure 9 provides a simplified outline of the holders of mortgage credit risk in the U.S. mortgage market.

Figure 9: Illustration of U.S. Mortgage Market

The simplest path for a mortgage to take is to be retained by the lending institution (Path 1 in Figure 9). Under this

path, the lending institution collects principal and interest payments from the borrower for the life of the loan. All

risks (credit, interest rate, and prepayment) associated with the mortgage remain on the balance sheet of the lending

Private mortgage guaranty insurers are major holders of U.S. mortgage credit risk. However, their master policies generally stipulate that they will deny a

claim where physical damage was the principal cause of the default giving rise to the claim. This includes physical damage caused by floods. As a result,

they are not discussed in this analysis.

From the borrower’s perspective, retail and correspondent channels are likely similar experiences (e.g., working directly with the institution originating,

underwriting, and funding the mortgage).

For the purposes of this paper, Government National Mortgage Association securities and private label mortgage-backed securities are excluded.

Whole Loan

Investor/ Portfolio

Lender

Credit Risk

Borrower

Mortgage Bank or

Other Originator

GSEs

Credit Risk

Agency MBS

Investors

Mortgage

CRT

Investors and

Reinsurers

Credit Risk

Path

institution.

As of the first quarter of 2019, balance sheet lending accounted for approximately 30% of the total

outstanding mortgage debt.

The common path for a newly originated mortgage in the United States is to be sold to the Federal National Mortgage

Association (Fannie Mae) or Freddie Mac (GSEs) and subsequently securitized and sold as agency mortgage-backed

securities (agency MBS) to investors (Path 2 in Figure 9). Agency MBS issued by GSEs do not carry credit risk; instead

the GSEs bear the credit risk and collect a portion of the loan’s interest (known as the guaranty fee) in exchange for a

guarantee of timely interest and principal repayment on the agency MBS they issue. As a result, the GSEs historically

have borne a large amount of the credit risk on U.S. mortgages (while transferring portions of the interest rate risk).

Separate and distinct from GSE agency MBS are securities issued from the Government National Mortgage Association

(Ginnie Mae), which are collateralized by mortgages directly insured by government agencies. As of the first quarter

of 2019, agency MBS, including Ginnie Mae, accounted for approximately 60% of the total outstanding mortgage

debt.

As part of their conservatorship and in an effort to transfer mortgage credit risk to the private market, the GSEs

developed debt securities and obtain reinsurance (collectively known as CRT transactions) with market participants.

The first of these transactions were offered in 2013. Therefore, as a result of these CRT transactions, the GSEs now

share credit risk with many private market participants.

GSE CRT

GSE CRT transfers a portion of the credit risk within pools of similar mortgages owned by the GSEs to investors and

reinsurers. The credit risk transfer is achieved by reducing the outstanding principal balance of the CRT securities by

the amount of credit losses sustained by the underlying mortgage pool (or in the case of reinsurance, collecting a

claim to cover losses in exchange for payment of an insurance premium). GSEs generally transfer risk via structured

securities (or reinsurance contacts) where parties invest (reinsure) various tranches of risk. These tranches receive

writedowns based on losses exceeding certain thresholds. This concentrates risk in the junior/subordinated tranches

of the structures. As a result, mortgage principal losses do not impact all CRT investors/reinsurers proportionally;

particular investors and reinsurers in the junior/subordinated tranches of the structures have disproportionately

higher exposure to credit risks relative to holders of the senior tranches. Figure 10 illustrates a typical GSE CRT

structure.

The lending institution can subsequently purchase mortgage guaranty insurance or other forms of credit enhancement.

Urban Institute, Housing Finance Policy Center (August 2019). Housing Finance at a Glance: A Monthly Chartbook. Available at:

https://www.urban.org/sites/default/files/publication/100866/august_chartbook_2019_0.pdf

The GSEs have further protections, mitigating credit risk, not discussed in this paper. Originators represent and warrant that the loans, sold to the GSEs,

meet all eligibility and underwriting guidelines. If the loans are found to be ineligible or defective, the agencies have the ability to require the lender to

repurchase the defective loans.

The remaining 10% of debt includes second liens and private label mortgage-backed securities.

Figure 10: GSE CRT Structure Example

Since 2013, GSE CRT has transferred the credit risk on more than $3 trillion of mortgages to investors and reinsurers.

Today, these structures represent a large and significant portion of the mortgage finance system, and the private

market assumes a significant portion of mortgage credit risk. As discussed above, investors in mortgage credit risk

could be exposed to flood risk through the collateral underlying the mortgages and supporting the CRT securities or

insurance policies. Specifically, flood losses that are uninsured could ultimately impact the holder of mortgage credit

risk. If so, investors and reinsurers in the junior or subordinated tranches of the structures could have

disproportionately high exposure to losses arising from flood. The next section provides an analysis on how flood

losses may impact the holders of mortgage credit risk, including investors participating in the junior/subordinated

tranches of the GSE CRT structures.

4.2 METHODOLOGY: CATASTROPHE ANALYTICS FOR HOME LOANS

To develop an initial assessment of how catastrophic losses could impact loan performance, we used output from the

KatRisk flood model as input to Milliman’s existing M-PIRe loan performance modeling methodology

and estimated

ranges of collateral loss as a result of mortgage borrower default due to flooding using the following steps:

Federal Housing Finance Agency (October 2019). The 2019 Strategic Plan for the Conservatorships of Fannie Mae and Freddie Mac. Available at:

https://www.fhfa.gov/AboutUs/Reports/ReportDocuments/2019-Strategic-Plan.pdf

Milliman’s M-PIRe platform, Mortgage Platform for Investments and Reinsurance, and the technical methodology for modeled mortgage performance is

described in the appendix of this report. M-PIRe includes modules to estimate the future performance of mortgage collateral at the loan level using

advanced statistical models as well as cash-flow structure models to pass the mortgage collateral forecasts through the capital structure of a given CRT

security to estimate investor losses.

100%

425 bps

350 bps

110 bps

60 bps

Losses

10 bps

GSE

Retained

Tranche B2

Tranche B1

Tranche M2

GSE Retained

Tranche B3 (GSE Retained)

Tranche M1

1. Select CRT transactions/underlying mortgage pools for analysis

2. Run the loan level underlying mortgage collateral through the KatRisk model to estimate the potential

property value decline for each loan as a result of natural disasters under various scenarios

3. Adjust the assumed future house price appreciation scenario for each loan to reflect the property value

decline based on modeled flood losses

4. Estimate loan level performance vectors for underlying mortgage pools using adjusted loan-level price

appreciation vectors

5. Summarize the output to estimate the potential impact of flood events on mortgage performance (e.g.,

default frequency increase, loss severity increase, etc.)

6. Run the projected underlying mortgage pools’ performance though CRT deal cash-flow waterfall analyses to

evaluate the impact of the revised cash flows on investors

7. Summarize the output from the projection of CRT investor and reinsurer losses by tranche

The remainder of this section describes each step in the above analysis in detail for a single event. The event is a 500-

year return period flood under current sea levels and takes into account consideration of the benefits of flood

insurance. Section 4.3 provides a summary of the analysis for all modeled events, including those considering our

selected sea level rise scenarios.

1. Select CRT transactions / underlying mortgage pools for analysis

Two recent issue GSE CRT securities were selected for the population of loans used in this analysis. The two deals

selected were STACR/ACIS 2019-DNA2 (low loan to value, or LTV) and STACR/ACIS 2019-HQA2 (high LTV). These

transactions are representative of GSE loan acquisitions, and are issued by Freddie Mac.

The low-LTV deal contains

loans with an initial loan-to-value ratio greater than 60% and less than or equal to 80%. The high-LTV deal contains

loans with an initial loan to value ratio greater than 80% and less than or equal to 97%. The GSEs issue low-LTV and

high-LTV deals separately, so it was important to run one of each to capture both sections of the mortgage market.

The two deals combine to a geographically diverse pool of 168,959 loans with a more than $40 billion unpaid balance

(UPB). Table 4 provides some summary statistics of the collateral underlying these transactions.

Home price appreciation is estimated in M-PIRe at the loan level using MSA-level home price index forecasts. Home price appreciation is a key driver of

estimated mortgage default risk.

Performing this analysis on Fannie Mae CAS/CIRT transactions would likely lead to similar conclusions. Data availability and a review of recent reference

pools informed our choice for this analysis. Freddie Mac STACR/ACIS transactions reference a single pool of loans, while CAS and CIRT transactions

reference disparate pools of loans. Using STACR/ACIS reference pools allowed the analysis to be relevant to two CRT executions.

Table 4: Sample Deal Characteristics

STACR/ACIS 2019-DNA2

(low-LTV)

STACR/ACIS 2019-HQA2

(high-LTV)

Deal Start Date March 2019 April 2019

Original Number of Loans

86,992

81,967

Total Original Unpaid Balance ($B)

$20.5

$19.5

Weighted Average FICO

750

746

Weighted Average Debt to Income

36.4%

37.4%

Weighted Average Loan to Value

76.5%

92.6%

Weighted Average Interest Rate

4.9%

4.8%

Each transaction is collateralized by mortgages with strong credit profiles having weighted average (weighted by loans’

unpaid principal balance) original credit scores of around 750 and weighted average debt-to-income (DTI) ratios

slightly higher than 36%. As expected, the loan-to-value ratio varies between each transaction with the low-LTV deal

having an initial weighted average LTV of 76.5% and the high-LTV deal having an initial weighted average LTV of 92.6%.

The transactions have mortgages with weighted average interest rates of 4.8% to 4.9%.

2. Run the loan level underlying mortgage collateral through the KatRisk model to estimate the potential

property value decline as a result of natural disasters under various scenarios

Locations from the market basket were sampled and assigned to loans based on the state, three-digit ZIP code, and

MSA for each loan and market basket location. Property value was estimated based on the value in the loan

information provided and translated to a building value using data estimates of land share of property value ratios

estimated by the Federal Housing Finance Agency as of November 2019.

Event-level losses were then generated for

each location based on the market basket characteristics and estimated building and property values based on the

loan information.

Event-level flood losses were produced with and without consideration for flood insurance benefits, mostly from

mandatory purchase requirements. Inside the SFHA, the mandatory purchase requirement was assumed to be

adhered to, and homes were assumed to be insured up to the maximum NFIP building limits of $250,000. Outside of

the SFHA, take-up rates in Exhibit 7, driven by voluntary purchase of flood insurance, were assumed.

Figure 11 shows the rank order of estimated property value declines after the example event for loans that had a

flood loss (impacted loans).

Federal Housing Finance Agency (January 2, 2019). Working Paper 19-01: The Price of Residential Land for Counties, ZIP costs, and Census Tracts in the

United States. Available at: https://www.fhfa.gov/PolicyProgramsResearch/Research/Pages/wp1901.aspx

Figure 11: Distribution of Post-Event Home Price Decline

Of the 168,959 loans across the DNA and HQA pools, 1,318 were impacted (0.8% of the total) in the simulated event.

Of those that were impacted, the decrease in property value as a result of the event ranges from a max of 59.5% to

near 0%. The decrease in property value is determined by deducting the estimated uninsured damage incurred on

the building as a result of the event divided by the original value of the property. The weighted average home value

decline is 6.1% within the impacted areas.

Despite the potentially material impact on individual properties, the weighted average impact of a given event on the

value of all the properties underlying the pool of mortgages is estimated to be a modest decline of 0.05%. The impact

is relatively small at the pool level because the pool is geographically diverse and fewer than 1.0% of the loans are

impacted by this simulated event.

3. Adjust the assumed future house price appreciation scenario for each loan to reflect the property value decline

based on modeled flood losses

To estimate the impact that flood events have on mortgage performance, the loan-level simulated decrease in

property value is assumed to occur at time 0 and is treated as a “shock” to the home price. For example, if the original

property value is $100,000 at time 0 and the assumed home price appreciation is 3% per year for the next five years,

the model estimates the property value to be $100,000 at time 0, $103,000 at time 1, $106,090 at time 2, etc. If the

impact to the value of property as a result of the flood is estimated at 20% at time 0, then the value of the property

is estimated to be $80,000 at time 0, $82,400 at time 1, $84,872 at time 2, etc.

Historical data and mortgage performance models indicate that mortgage default events are heavily correlated with

changes in home prices. Specifically, declines in home values result in increased mortgage defaults and severity rates.

The below methodology will rely on these relationships to estimate how mortgages perform following an event.

-70%

-60%

-50%

-40%

-30%

-20%

-10%

0 100 200 300 400 500 600 700 800 900 1000 1100

Home Price Decline as Result of Event

Impact Rank

Sample Event: 500-year Flood w/ NFIP

Home Value Declines by Property

ACIS 2019-DNA2 and HQA2 Impacted Loans

Due to the wide range of shocks stemming from the flood event, the methodology must capture changes in property

values at the loan level. Applying a weighted average 6.1% shock to the entire impacted region is not equivalent to

capturing the full distribution of shocks across the individual loans. The relationship between home price declines and

mortgage defaults is not linear due to optionality of mortgages, borrower resilience to small changes in house prices,

and borrower incentives. If 100 loans all have a 0.5% negative house price shock, the results will not be equivalent to

99 loans having a 0% shock and one loan having a 50% shock.

Figure 12 shows the range of the unadjusted house price index path versus the max decline house price index path,

along with the weighted average (WA) path.

Figure 12: Modeled House Price Forecasts by Scenario

The three paths shown in Figure 12 display the variance in the distribution of property value declines as a result of

the example event.

4. Estimate loan level performance vectors for underlying mortgage pools using adjusted loan-level price

appreciation vectors

Loan-level performance vectors were estimated as a function of each loan’s underwriting characteristics (e.g., original

FICO score, original loan-to-value ratio, debt-to-income ratio, and others) and underlying property value, as measured

by MSA-level home price indices.

In order to estimate the impact of the house price shock due to a flood event, each loan was run through two separate

iterations of the mortgage performance model in M-PIRe. The first run was the unadjusted baseline run for all loans

in the pool. This provides a view of the model’s baseline projected performance under a baseline economic scenario.

100

120

140

160

180

House Price Index

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

House Price Index by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Max Event Impacted - WA Event Impacted - Base

There were no shocks to house prices applied in this iteration, and each loan received a forecasted house price

appreciation path equal to Moody’s baseline forecast for the property’s MSA (or state if a property does not fall in an

MSA).

The second iteration of the model output was performed with the flood-event shocks applied to the loan-level house

price appreciation. After the shocks to house prices were applied, each loan reverted back to receiving a forecasted

house price appreciation path equal to Moody’s baseline forecast for the property’s MSA (or state if a property does

not fall in an MSA). This results in an apples-to-apples comparison to the baseline unadjusted run for the goal of

distilling out the impact of the flood events.

In this analysis, all loans begin as performing at inception of the CRT transaction. In the model, the estimated drop in

the value of the home will be an immediate negative impact to borrower equity. This will in turn result in increased

delinquencies, foreclosure rates, and severity rates for impacted loans.

One aspect of the mortgage performance model that was altered for the performance of this analysis was the

mortgage guaranty insurance proceeds module. In general, for the GSEs to purchase a loan with an LTV greater than

80%, a mortgage guaranty insurance policy must be purchased on that loan (either by the borrower or the lender).

The borrower or lender pays the mortgage guaranty insurer (MI) a premium and in exchange the MI will reimburse

the mortgage holder (investor) for a portion of the lost principal in the event of borrower default.

In MI’s master policy documents, a document that outlines the terms and conditions under which the MI will cover a

loss resulting from a borrower default, there is generally a clause stating that the MI will not be liable for any losses

resulting from physical property damage, excluding normal wear and tear. Many MIs explicitly state flood as one of

the causes of damage that they will not be liable for. As a result, it is likely that under any of the events contemplated

in this analysis, the MIs covering the loans in the impacted region would deny the claims on these properties. This

would result in even higher losses as a result of the event.

To capture this potential outcome, the mortgage guaranty insurance proceeds module (which nets out the MI

proceeds received on high-LTV loans from the loss severity) was overridden—for just those loans projected to incur

flood damage due to the event—to not reflect any benefit from MI coverage.

5. Summarize the output to estimate the potential impact of flood events on mortgage performance

The analysis above results in projections of future collateral performance and investor cash flows over the life of the

collateral. As the collateral for these two deals consists of almost 170,000 individual mortgages, the results need to

be summarized down to standard performance metrics used to evaluate mortgage performance. Table 5 provides a

description of the metrics used in this analysis.

Table 5: Mortgage Model Metrics

Metric

Description

60+ Delinquency Rate The percent of active loans that are two or more payments delinquent. Loans that are

delinquent are leading indicators of future losses, and higher delinquency rates

indicate increased credit risk.

Conditional Default Rate

(CDR)

The percent of active loans estimated to default in a given period. CDRs are typically

annualized, so if the model expects one loan out of 100 loans to default in a given

month, the CDR would be 12% (12% = 1 / 100 * 12).

Loss Rate

Total loss amount from default events divided by the original pool balance.

The model results estimate that immediately following an event, the 60+ day delinquency rate begins to rise for loans

impacted by the event. The M-PIRe mortgage performance model takes the equity position, and change in equity

position, of the borrower at each forecast quarter as a predictor variable. For loans impacted by the event, the equity

position experiences a potentially significant decrease, putting upward pressure on delinquency rates. Figure 13

shows that the event causes the estimated 60+ day delinquency rate of the impacted region to be more than 20%

higher than it would be absent the event.

Figure 13: Modeled Delinquency Rates by Scenario

In the analysis, the event takes place in 2Q 2019, and it takes approximately 18 months for the event to lead to

mortgage defaults (e.g., a foreclosure, short sale, or other credit event). Generally, the loan servicer will work with

the borrower to avoid default, extending the timeline of an ultimate default. If default is inevitable, it still takes

approximately 12+ months from delinquency to claim for the foreclosure process to complete. This timeline varies

based on state level requirements for foreclosure proceedings.

The flood event causes the conditional default rate (CDR) to spike in 2021, as many of the impacted borrowers 18

months earlier are not estimated to recover from the event; the CRT comparison is shown on Figure 14. In late 2021,

the CDR is more than three times greater than it would have been absent the event.

100

150

200

250

300

60+ Day Delinquencies (bps of Original UPB)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

60+ Day Delinquency Rate by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Base Impacted - Event

Figure 14: Modeled Conditional Default Rates by Scenario

Compounding the event's impact on homeowners' inability or unwillingness to pay their mortgage, the loss severity

(shortfall as a percent of the loan's unpaid principal balance) of those mortgages that default is also sharply

increased due to the estimated decline in the value of the underlying property.

Figure 15 provides a visual of the model forecast loss severity.

100

120

140

160

Conditional Default Rate (bps)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

Conditional Default Rate by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Base Impacted - Event

Figure 15: Modeled Severity by Scenario

The losses stemming from a loan default are more than two times higher under the event scenario verses absent the

event. This impact is rather straightforward. If house prices go down, there is less value to recover when a homeowner

is foreclosed on and their home is liquidated. This is further exacerbated due to the lack of mortgage guaranty

insurance recoveries on loans with greater than 80% LTV (see above for rationale for removal of MI proceeds).

Coupling the increased instances of borrower defaults with the increased loss severity given default, losses develop

much more quickly and to a higher ultimate level under the event scenario. Losses within the impacted region are

estimated to be more than three times greater under the event scenario than absent the event, as shown in Figure

16.

10%

15%

20%

25%

30%

35%

40%

45%

Loss Severity (%)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

Loss Severity Rate by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Base Impacted - Event

Figure 16: Modeled Cumulative Loss Rate by Scenario

As the event raises delinquency and adverse outcomes for the borrowers, it also causes borrowers to prepay their

loans more slowly, as shown in Figure 17.

100

120

140

Cumulative Loss Rate (bps of Original UPB)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

Cumulative Loss Rate by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Base Impacted - Event

Figure 17: Post-Event Economic Forecasts

Throughout the life of the DNA2 and HQA2 reference pools, approximately 5% fewer borrowers are estimated to

prepay their mortgage in the impacted area (ultimate prepay rate of 69% versus 74%). Causes of this behavior can

include the inability to refinance their loan, decreased mobility, or negative equity in their property. M-PIRe’s

prepayment model includes variables that capture borrower equity and changes in house prices; as a result of the

flood event, equity and house prices decline and translate into a lower forecast prepayment rate, all else equal.

However, in reality, borrowers may react to the flood event by vacating damaged properties and/or attempting to

move away from the impacted area, thus mitigating or reversing the expected drop in prepayments. Such reactions

to flood events are not contemplated in this analysis.

6. Run the projected underlying mortgage pools’ performance though CRT deal cash-flow waterfalls to evaluate

the impact of the revised cash flows on investors

The above analysis discusses the impact of the flood event on the underlying collateral. The credit risk of these

transactions was offered to the private market through CRT bonds and reinsurance treaties. Table 6 provides a

description of the credit risk for each execution. The table lists the tranches and associated credit enhancement for

each. The credit enhancement represents the amount of credit losses required before a given tranche incurs losses.

For example, for ACIS 2019-DNA2, the B-1H tranche has 0.60% credit enhancement. This means that if the cumulative

loss rate on the transaction (the sum of credit losses divided by the original unpaid balance of the mortgage pool)

exceeds 0.60%, then the B-1H tranche will start to incur losses. The B-1H tranche will incur losses until losses exceed

1.10%, at which time the B-1H tranche would be fully exhausted and the M-2H tranche would start to incur losses.

10%

20%

30%

40%

50%

60%

70%

80%

Prepayments (% of Original UPB)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

Cumulative Prepayment Rate by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

Impacted - Base Impacted - Event

Table 6: Credit Enhancement by Tranche

Credit Enhancement by Tranche

Tranche

ACIS 2019-DNA2

ACIS 2019-HQA2

STACR 2019-DNA2

STACR 2019-HQA2

Retained

4.25%

4.50%

4.25%

4.50%

M-1(H)

3.50%

M-2(H)

1.10%

1.50%

1.10%

1.50%

B-1(H)

0.60%

B-2(H) 0.10% 0.10% 0.10% 0.10%

B-3(H)

0.00%

This waterfall is programmed into M-PIRe for the entire population of possible CRT transactions. The above modeled

cash flows were run through the cash-flow library to estimate investor cash flows.

STACR tranches can be further broken down into A and B classes, as shown in Table 7. For example, B-2B has 0.10%

initial credit enhancement, and B-2A has 0.35% initial credit enhancement. Class B-2B is the class of notes closest to

the first loss position (directly behind the retained B-3 tranche). When analyzing how losses due to specific flood

events can translate into disproportionately higher losses for investors in riskier classes of notes, we focus on the B-

2B tranche.

Table 7: Credit Enhancement by Tranche – STACR A and B Tranches

Credit Enhancement by Tranche

Tranche

STACR 2019-DNA2

STACR 2019-HQA2

Retained

4.25%

4.50%

M-1

3.50%

M-2A

2.30%

2.50%

M-2B

1.10%

1.50%

B-1A

0.85%

1.05%

B-1B

0.60%

B-2A

0.35%

B-2B

0.10%

B-3 0.00% 0.00%

7. Summarize projection of CRT losses by tranche to estimate the impact from the flood event

From the above analysis, the impact on the pool level losses as a result of the event is estimated to be approximately

one basis point for the sample event. The baseline scenario estimates loss rates of approximately 17.4 basis points

without the event. Therefore, a single event could have a material impact on the credit losses for the more

subordinated tranches. Investors in the B-2B tranches for both the high-LTV and low-LTV deals are relying on losses

staying below the 10bps B-3 retention level for full principal repayment. As a result, any incremental loss development

from a natural disaster will disproportionally impact these investors. The sample event, despite only impacting 0.67%

of loans, causes investors to lose more than 3% more of their original principal investment in the B-2 tranche, as

shown in Figure 18.

Figure 18: Modeled Investor Losses by Scenario

Excluding the event in the base case scenario, investors in the B-2 tranche are estimated to lose 30% of principal. With

the event, investors in the B-2 tranche are estimated to lose 33% of principal.

4.3 RESULTS: CATASTROPHE ANALYTICS FOR HOME LOANS

The above analysis walks through the approach utilized to evaluate the impact flood events can have on mortgage

collateral, and the potential impact to bond and reinsurance investors in CRT securities. In total, 16 specific

combinations were analyzed to understand the sensitivity of mortgage impacts to flood event severity, sea levels, and

flood insurance/mandatory purchase requirements. This section compares results across the alternative scenarios to

summarize the range of impact of flood on mortgage collateral performance and investors’ credit risk

Six events were selected as the 100- and 500-year return period events based on flood losses for the countrywide

market basket, for each of the three sea level scenarios. To isolate the impact of sea level rise on an event, we also

analyzed the same 500-year return period event for current sea levels with medium and high sea level rise, adding

two additional return period and sea level combinations. Each of these combinations was run with and without

considering the benefits of insurance, resulting in our final tally of 16 combinations. When considering the benefits of

10%

15%

20%

25%

30%

35%

Principal Losses (% of Original Investment)

Calendar Month

ACIS 2019-DNA2 and HQA2 M-PIRe Model Results

B-2B Investor Principal Losses by Scenario

Baseline Economic Forecast after Event

Source: M-PIRe

All - Base All - Event

insurance, we assumed that all homes in the SFHA purchased flood insurance and that those outside the SFHA

purchased flood insurance at a rate equal to the statewide average rate outside the SFHA.

Results for impacted regions, defined as loans that incurred at least some flood loss, are shown in Exhibit 3. The house

price shocks shown, declines in property value as a result of uninsured building losses, would be substantial under this

scenario, which is expected given the extreme nature of this event. The resulting default frequency increase and

conditional prepayment rate (CPR) decrease would also be substantial. Even when considering insurance claim

payments, we estimate in each scenario that homes with flood damage are at least twice as likely to default on their

home post-event.

Exhibit 3: Collateral Performance for Impacted Regions

The above results are calculated on only the impacted region. However, the collateral pool underlying CRT

transactions is geographically diverse. Exhibit 4 shows the estimated performance of the entire collateral pool under

each scenario. When calculating the impact on the entire collateral pool, the change to average home prices for the

entire collateral pool is small, contributing to less than a 1% decrease to home prices for the entire pool across all

scenarios.

ACIS 2019-DNA2 and HQA2 Model Results

Collateral Performance

Baseline Economic Forecast after Event

Source: M-PIRe

Return Period of Flood

Losses in Ye ars

Sea Level

Scenario

Considers

Benefits of

Insurance

House

Price

Shock

WA CPR

Base

WA CPR

Shock

Ultimate

Default

Freq.

Base

Ultimate

Default

Freq.

Shock

WA Loss

Severity

Base

WA Loss

Severity

Shock

Ultimate

Loss

(bps)

Base

Ultimate

Loss

(bps)

Shock

100 Current No -22.03% 15.04% 11.56% 1.39% 8.83% 16.00% 55.81%

22.23

492.88

100 Current Yes -9.36% 15.04% 13.02% 1.39% 4.62% 16.00% 47.59%

22.23

220.08

500 Current No -14.22% 15.53% 12.20% 1.78% 5.84% 15.62% 42.90%

27.79

250.42

500 Current Yes -6.15% 15.53% 13.57% 1.78% 3.67% 15.62% 35.90%

27.79

131.56

500 (at Current Sea Levels) Medium No -16.04% 15.50% 11.99% 1.77% 6.53% 15.52% 44.65%

27.47 291.65

500 (at Current Sea Levels) Medium Yes -7.07% 15.50% 13.44% 1.77% 3.99% 15.52% 37.69%

27.47 150.51

500 (at Current Sea Levels) High No -19.08% 15.50% 11.72% 1.75% 7.76% 15.47% 47.41%

27.02

367.97

500 (at Current Sea Levels) High Yes -8.69% 15.50% 13.25% 1.75% 4.59% 15.47% 40.78%

27.02 187.28

100 Medium No -17.05% 15.18% 12.14% 1.55% 7.15% 15.19% 53.99%

23.52

386.02

100 Medium Yes -7.62% 15.18% 13.34% 1.55% 4.12% 15.19% 46.55%

23.52

191.82

500 Medium No -13.22% 15.51% 12.37% 1.70% 5.58% 15.15% 42.58%

25.74

237.76

500 Medium Yes -5.83% 15.51% 13.57% 1.70% 3.46% 15.15% 34.37%

25.74

118.79

100 High No -30.97% 14.28% 13.15% 2.03% 20.24% 10.63% 76.90%

21.61 1,556.31

100 High Yes -23.72% 14.28% 13.12% 2.03% 15.78% 10.63% 71.21%

21.61

1,123.60

500 High No -41.65% 16.11% 11.95% 2.00% 22.26% 15.14% 68.90%

30.28

1,533.49

500 High Yes -22.76% 16.11% 13.01% 2.00% 12.62% 15.14% 57.53%

30.28

725.79

Results for Impacted Regions

Exhibit 4: Baseline Economic Performance After Event

Exhibit 5 calculates the percentage increase in credit losses across the scenarios, for both impacted regions and the

entire collateral pool.

Exhibit 5: Increase in Credit Losses After Event; Impacted Regions and Combined Pool

Credit losses in impacted regions are an order of magnitude higher post-event. Considering insurance claim payments,

the impacted regions for a 100-year event under our high sea level rise scenarios are estimated to be more than 50

times larger post-event.

ACIS 2019-DNA2 and HQA2 Model Results

Collateral Performance

Baseline Economic Forecast after Event

Source: M-PIRe

Return Period of Flood

Losses in Ye ars

Sea Level

Scenario

Considers

Benefits of

Insurance

House

Price

Shock

WA CPR

Base

WA CPR

Shock

Ultimate

Default

Freq.

Base

Ultimate

Default

Freq.

Shock

WA Loss

Severity

Base

WA Loss

Severity

Shock

Ultimate

Loss

(bps)

Base

Ultimate

Loss

(bps)

Shock

100 Current No -0.09% 15.05% 15.03% 1.36% 1.39% 12.79% 13.91%

17.39

19.33

100 Current Yes -0.04% 15.05% 15.04% 1.36% 1.37% 12.79% 13.26%

17.39

18.21

500 Current No -0.11% 15.05% 15.02% 1.36% 1.39% 12.79% 13.75%

17.39

19.13

500 Current Yes -0.05% 15.05% 15.03% 1.36% 1.37% 12.79% 13.25%

17.39

18.20

500 (at Current Sea Levels) Medium No -0.13% 15.05% 15.02% 1.36% 1.40% 12.79% 13.98%

17.39

19.56

500 (at Current Sea Levels) Medium Yes -0.06% 15.05% 15.03% 1.36% 1.38% 12.79% 13.36%

17.39

18.40

500 (at Current Sea Levels) High No -0.16% 15.05% 15.01% 1.36% 1.41% 12.79% 14.41%

17.39

20.35

500 (at Current Sea Levels) High Yes -0.07% 15.05% 15.03% 1.36% 1.38% 12.79% 13.57%

17.39 18.78

100 Medium No -0.10% 15.05% 15.03% 1.36% 1.39% 12.79% 14.09%

17.39

19.65

100 Medium Yes -0.05% 15.05% 15.04% 1.36% 1.38% 12.79% 13.41%

17.39

18.44

500 Medium No -0.14% 15.05% 15.01% 1.36% 1.40% 12.79% 14.02%

17.39

19.64

500 Medium Yes -0.06% 15.05% 15.03% 1.36% 1.38% 12.79% 13.34%

17.39

18.38

100 High No -0.13% 15.05% 15.05% 1.36% 1.44% 12.79% 16.73%

17.39

24.06

100 High Yes -0.10% 15.05% 15.04% 1.36% 1.42% 12.79% 15.63%

17.39

22.18

500 High No -0.20% 15.05% 15.03% 1.36% 1.46% 12.79% 17.00%

17.39

24.82

500 High Yes -0.11% 15.05% 15.03% 1.36% 1.41% 12.79% 14.75%

17.39

20.83

Results for Combined ACIS 2019-2 Pools

ACIS 2019-DNA2 and HQA2 Model Results

Collateral Performance

Baseline Economic Forecast after Event

Source: M-PIRe

A B A B

Return Period of Flood

Losse s in Ye a rs

Sea Level

Scenario

Considers

Benefits of

Insurance

House

Price

Shock

Ultimate

Loss (bps)

Base

Ultimate

Loss (bps)

Shock

% Increase

= (B-A) / A

House

Price

Shock

Ultimate

Loss (bps)

Base

Ultimate

Loss (bps)

Shock

% Increase

= (B-A) / A

100 Current No -22.03%

22.23 492.88 2116.7% -0.09%

17.39

19.33 11.1%

100 Current Yes -9.36%

22.23

220.08 889.8% -0.04% 17.39

18.21 4.7%

500 Current No -14.22%

27.79

250.42 801.0% -0.11%

17.39 19.13 10.0%

500 Current Yes -6.15%

27.79

131.56 373.4% -0.05%

17.39 18.20 4.7%

500 (at Current Sea Levels) Medium No -16.04%

27.47

291.65 961.6% -0.13%

17.39 19.56 12.4%

500 (at Current Sea Levels) Medium Yes -7.07%

27.47 150.51 447.8% -0.06%

17.39 18.40 5.8%

500 (at Current Sea Levels) High No -19.08%

27.02 367.97 1261.6% -0.16%

17.39 20.35 17.0%

500 (at Current Sea Levels) High Yes -8.69%

27.02 187.28 593.0% -0.07%

17.39

18.78 8.0%

100 Medium No -17.05%

23.52 386.02 1541.3% -0.10% 17.39

19.65 13.0%

100 Medium Yes -7.62%

23.52 191.82 715.6% -0.05%

17.39 18.44 6.0%

500 Medium No -13.22%

25.74 237.76 823.6% -0.14%

17.39 19.64 12.9%

500 Medium Yes -5.83%

25.74

118.79 361.4% -0.06%

17.39 18.38 5.7%

100 High No -30.97%

21.61

1,556.31 7103.2% -0.13% 17.39

24.06 38.3%

100 High Yes -23.72%

21.61 1,123.60 5100.4% -0.10% 17.39 22.18 27.5%

500 High No -41.65%

30.28

1,533.49 4965.0% -0.20%

17.39 24.82 42.7%

500 High Yes -22.76%

30.28 725.79 2297.2% -0.11%

17.39 20.83 19.8%

Results for Impacted Regions

Results for Combined ACIS 2019-2 Pools

Across the entire collateral pool, the impacts of mandatory purchase (and voluntary purchase for loans outside the

SFHA) can be seen with the reduction in the credit loss increase due to flood losses for an event with and without

consideration of the benefits of flood insurance. While current flood insurance purchasing patterns and the

mandatory purchase requirement do mitigate the expected impact to credit losses,

model estimates indicate that at

least 42% of the expected increase in credit losses due to extreme flood events could remain after considering the

benefits of insurance claim payments.

Following the same 500-year return period event at current sea levels across the medium and high scenarios allows

us to isolate the impact of sea level rise on credit losses for a single major catastrophe. We can calculate the extent

at which sea level rise impacted credit losses due to flood losses by comparing the credit loss impact of combined

results in Exhibit 6

. After considering insurance claim payments, the increase in credit losses for the entire collateral

pool due to a 500-year return period flood event is 4.7% at current sea levels and is 5.8% in a medium sea level rise

scenario. In a high sea level rise scenario, the increase in credit losses is 8.0% for an otherwise similar event.

Thus,

credit losses for this event increased by 24% (5.8% impact versus 4.7% impact) and 72% (8.0% impact versus 4.7%

impact) using our medium and high sea level rise scenarios, respectively.

As discussed above, local impacts of some extreme flood events on mortgage defaults could be substantial, with

modeled credit losses an order of magnitude higher post-event.

Even considering current insurance purchasing

patterns and requirements, estimated credit losses for impacted regions increased between approximately four and

23 times for each of our modeled events. The incremental credit loss impact estimates, while small on the aggregate

pool, could deliver a more significant impact to investors in the subordinate tranches.

Exhibit 6: Increase in Credit Losses After Event; B-2B Bond Performance

Exhibit 6 shows the calculated impact in credit losses for loans of the subordinated B-2B tranche. For high-LTV loans,

even with consideration of flood insurance, extreme flood events could increase principal writedowns by 15% to 95%

relative to a baseline scenario without a flood event. Thus, the relationship between catastrophic and economic

concentrations of risk may be important to consider when evaluating and pricing these transactions.

ACIS 2019-DNA2 and HQA2 Model Results

B-2B Bond Performance

Baseline Economic Forecast after Event

Source: M-PIRe

A B A B

Return Period of Flood

Losse s in Ye a rs

Sea Level

Scenario

Considers

Benefits of

Insurance

House

Price

Shock

Ultimate B-

2B Principal

Loss

Base

Ultimate B-

2B Principal

Loss

Shock

% Increase

= (B-A) / A

House

Price

Shock

Ultimate B-

2B Principal

Loss

Base

Ultimate B-

2B Principal

Loss

Shock

% Increase

= (B-A) / A

100 Current No -22.03%

0.31

0.38 23.7% -0.09%

0.28

0.36 29.2%

100 Current Yes -9.36%

0.31

0.33 8.5% -0.04%

0.28

0.32 13.9%

500 Current No -14.22%

0.31

0.37 20.1% -0.11%

0.28

0.36 27.5%

500 Current Yes -6.15%

0.31

0.33 7.2% -0.05%

0.28

0.32 15.2%

500 (at Current Sea Levels) Medium No -16.04%

0.31

0.39 25.4% -0.13%

0.28

0.38 33.8%

500 (at Current Sea Levels) Medium Yes -7.07%

0.31

0.34 9.4% -0.06%

0.28

0.33 18.5%

500 (at Current Sea Levels) High No -19.08%

0.31

0.42 35.4% -0.16%

0.28

0.41 45.3%

500 (at Current Sea Levels) High Yes -8.69%

0.31

0.35 13.8% -0.07%

0.28

0.35 24.6%

100 Medium No -17.05%

0.31

0.40 28.9% -0.10%

0.28

0.37 32.5%

100 Medium Yes -7.62%

0.31

0.35 12.3% -0.05%

0.28

0.33 16.4%

500 Medium No -13.22%

0.31

0.39 27.8% -0.14%

0.28

0.37 33.3%

500 Medium Yes -5.83%

0.31

0.34 9.6% -0.06%

0.28

0.33 17.6%

100 High No -30.97%

0.31

0.47 54.3% -0.13%

0.28

0.65 132.1%

100 High Yes -23.72%

0.31

0.43 38.6% -0.10%

0.28

0.55 95.3%

500 High No -41.65%

0.31

0.62 100.9% -0.20%

0.28

0.56 100.5%

500 High Yes -22.76%

0.31

0.43 41.1% -0.11%

0.28

0.43 52.9%

Results for ACIS 2019-DNA2

Results for ACIS 2019-HQA2

The 100-year and 500-year return period high sea level rise scenarios, unsurprisingly, all have the highest increased

credits losses due to flooding. However, when the high sea level rise scenario is run with the 500-year event based on

current sea levels, the increased losses isolated due to sea level rise are much smaller. This highlights the variability

of mortgage risk due to flood losses. It is likely that the large increase in losses in the 100-year and 500-year high sea

level rise scenario are partially a result of the event itself impacting a different geographic area and/or specific loans.

4.4 HISTORICAL RESULTS: 2017 MORTGAGE PERFORMANCE FOLLOWING HURRICANE IRMA

The methodology discussed above relies on existing mortgage models that contain parameters to assess potential

cost to mortgage credit holders resulting from home price shocks. This relies on the assumption that home price

shocks as a result of a natural catastrophe would result in a similar spike in mortgage delinquencies, foreclosures, and

claims as shocks from causes other than natural catastrophes. This assumption may be flawed for a number of

reasons. For example, state or federal disaster assistance could mute the impact of underinsurance, or economic

circumstances at the time of any catastrophe could be different than those surrounding generic price shocks.

To provide an evaluation of historical mortgage performance following a catastrophe, this section looks at a portfolio

of mortgages in Florida following Hurricane Irma, which made landfall as a Category 4 hurricane in the Florida Keys

and struck southwestern Florida at Category 3 intensity on September 10, 2017.

The combined effect of storm surge,

rainfall, and the tide produced inundation levels of three to five feet above ground level for several largely populated

cities, with significant flooding in Miami. NOAA estimates the damage caused by Hurricane Irma to be approximately

$50 billion,

of which insured losses totaled approximately $25 billion.

Sample CRT Deals

The performance on the mortgage collateral underlying two mortgage credit risk transfer deals was analyzed to

evaluate the impact Hurricane Irma had on mortgage performance. The two deals used for this analysis are ACIS 2017-

3 / STACR 2017-DNA2 and ACIS 2017-5 / STACR 2017-HQA2. Similar to the above analysis, the ACIS 2017-3 deal

represents low-LTV loans (i.e., loans originated with a downpayment of 20% or greater) and the ACIS 2017-5 deal

represents high-LTV loans (i.e., loans originated with a downpayment of less than 20%). Table 8 provides some

summary statistics of these two transactions.

Cangialosi, J.P., Latto, A.S. & Berg, R. (June 30, 2018). Hurricane Irma. National Hurricane Center Tropical Cyclone Report. Available at:

https://www.nhc.noaa.gov/data/tcr/AL112017_Irma.pdf

NOAA National Centers for Environmental Information (NCEI) U.S. Billion-Dollar Weather and Climate Disasters (2020).

https://www.ncdc.noaa.gov/billions/, DOI: 10.25921/stkw-7w73

Insurance Information Institute. Facts + Statistics: Hurricanes. Available at: https://www.iii.org/fact-statistic/facts-statistics-hurricanes

Table 8: Sample 2017 Deals: Characteristics

ACIS 2017-3 / STACR 2017-DNA2

(low-LTV)

ACIS 2017-5 / STACR 2017-HQA2

(high-LTV)

Deal Start Date

May 2017

July 2017

Original Number of Loans

236,139

129,587

Total Original Unpaid Balance ($B)

$60.7

$31.6

Original Unpaid Balance in All States

Excluding Florida ($B)

$57.7

$29.9

Original Unpaid Balance in Florida ($B)

$3.0

$1.7

Percent of Loans in Florida

4.95%

5.32%

Average FICO

751

747

Average Debt to Income

34.7%

35.6%

Average Loan to Value

75.9%

91.6%

Average Interest Rate

3.8%

For both deals, loans in Florida comprise between 5% and 6% of the total origination amount. Each transaction is

collateralized by mortgages with strong credit profiles having average original credit scores of about 750 and average

debt-to-income ratios less than 36%. As expected, the loan-to-value ratio varies between each transaction, with the

low-LTV deal having an initial average LTV of 75.9% and the high-LTV deal having an initial average LTV of 91.6%. Both

transactions have mortgages with low average interest rates of 3.8%. Based on our review and modeling of these

transactions, under a baseline economic scenario and without consideration of any flood events, we would expect

relatively low levels of collateral losses in the range of 10 to 20 basis points.

The first payment date for each deal was May 2017 and July 2017 for the low- and high-LTV transactions, respectively.

This means the underlying mortgages for each transaction were originated approximately six months prior to the deal

start date. Hurricane Irma landed in Florida on September 10, 2017, so the mortgages underlying these two deals

were approximately one year seasoned when Hurricane Irma landed.

Delinquency Rates

One early indicator of future losses on mortgage collateral is the status of the loan at a given date. Loans that are 30

days delinquent are more likely to result in a foreclosure or loss to the investor relative to loans that have not missed

any payments. Similarly, loans that are 60 days delinquent are more likely to result in a foreclosure or loss to the

investor relative to loans that are 30 days delinquent, and so forth. Credit analysts often look at the delinquency

development for a portfolio of mortgages as an indication of future losses.

To evaluate the impact of Hurricane Irma on potential future losses, we considered the development of delinquencies

before and after the event. Figure 19

shows the unpaid principal balance on loans that are delinquent divided by the

unpaid principal balance of all mortgages in the low-LTV transaction by delinquency status (30 days delinquent, 60

days delinquent, and 90+ days delinquent) and calendar month.

Figure 19: Delinquency Rates: Low LTV Loans Excluding Florida

This figure excludes loans in Florida from the data in order to demonstrate the difference in performance between

loans in Florida (i.e., loans potentially impacted by Hurricane Irma) and loans not in Florida (i.e., loans not impact by

Hurricane Irma).

The figure shows that for the life of this deal from inception through December 2019, the

cumulative percent of loans that were delinquent started at 0.2% in the first month and has gradually increased to

around 1.0% as of December 2019. The majority of delinquencies are mortgages that are 30 days delinquent, with

relatively few loans transitioning into 60 and 90+ day delinquencies.

Figure 20 provides the same information, but this

figure only includes loans in Florida.

We note that there were additional natural catastrophes during this time, namely Hurricane Harvey and Maria, but either the concentration of loans was

not large or the impact on delinquencies was smaller relative to Hurricane Irma. Including or excluding loans impacted by these disasters does not

materially change the results of the analysis.

UPB of Delinquent Loans / Active UPB

Calendar Month

ACIS 2017-3 / STACR 2017-DNA2

All Loans Excluding Florida

Delinquency Rate (Percent of Active UPB) by Status

Source: Freddie Mac, M-PIRe

30 Day Delinquency 60 Day Delinquency 90+ Day Delinquency

Figure 20: Delinquency Rates: Low-LTV Loans Florida Only

In stark contrast to the non-Florida loans in Figure 19, Figure 20 demonstrates that loans in Florida experienced a

spike in delinquencies starting in November 2017, peaking in January 2018 at slightly less than 6%, and gradually

declining through October 2018. Significantly, many of the loans that went delinquent quickly transitioned from 30

days delinquent to 60 days delinquent and 90 or more days delinquent. Following October 2018, the percent of loans

that are 30- and 60-days delinquent are consistent with the rest of the country, but there still remains an elevated

level of defaults that are 90 days delinquent or more in Florida for this transaction.

These results are consistent with the model estimates presented above where a natural disaster results in a shock to

the market where the disaster occurs. The shock leads to elevated defaults, which should then result in an increased

level of credit losses.

Figure 21 and Figure 22 provide identical information as above but for the high-LTV transaction.

UPB of Delinquent Loans / Active UPB

Calendar Month

ACIS 2017-3 / STACR 2017-DNA2

Florida Loans Only

Delinquency Rate (Percent of Active UPB) by Status

Source: Freddie Mac, M-PIRe

30 Day Delinquency 60 Day Delinquency 90+ Day Delinquency

Figure 21: Delinquency Rates: High-LTV Loans Excluding Florida

10%

UPB of Delinquent Loans / Active UPB

Calendar Month

ACIS 2017-5 / STACR 2017-HQA2

All Loans Excluding Florida

Delinquency Rate (Percent of Active UPB) by Status

Source: Freddie Mac, M-PIRe

30 Day Delinquency 60 Day Delinquency 90+ Day Delinquency

Figure 22: Delinquency Rates: High-LTV Loans Florida Only

For this transaction, the trends and conclusion are the same as the low-LTV transaction, but the level of defaults is

greater for the high-LTV transaction, with defaults peaking at more than 9% in January 2018. This result is expected,

as the original loan-to-value ratio of a mortgage is a significant predictor of future default rates. Borrowers with larger

downpayments have more equity in their property and incentive (and possibly financial resources) to maintain the

property, and empirical data confirms that they have lower default rates relative to borrowers with small

downpayments.

Typically, loans that are 90 days delinquent transition into foreclosure and result in a loss at a relatively high “roll rate”

ranging from 20% to 70%. The roll rate depends on, among other factors, the equity position and creditworthiness of

the borrower. Once a loan goes through the foreclosure process, the loss on the mortgage is not the entire unpaid

principal balance. The loss is equal to the unpaid principal balance plus delinquent interest and expenses minus the

recovery amount collected from selling the property and minus mortgage guaranty insurance payments, if applicable.

In recent years, this value, known as the severity, has averaged approximately 15% of the unpaid principal balance.

Table 9 estimates the losses that would have been expected using these ranged on the delinquent collateral for each

transaction for loans located in Florida. This table represents a rough approximation of losses for demonstrative

purposes.

10%

UPB of Delinquent Loans / Active UPB

Calendar Month

ACIS 2017-5 / STACR-HQA2

Florida Loans only

Delinquency Rate (Percent of Active UPB) by Status

Source: Freddie Mac, M-PIRe

30 Day Delinquency 60 Day Delinquency 90+ Day Delinquency

Table 9: Sample 2017 Deals: Expected Performance

Calculation

ACIS 2017-3 / STACR 2017-

DNA2 (low-LTV)

ACIS 2017-5 / STACR 2017-

HQA2 (high-LTV)

Peak 90 Day Delinquency

Rate

A 3.2% 5.6%

Roll Rate from 90 to

Foreclosure

20%

Foreclosure

C = A * B

0.64%

1.12%

Severity

15%

Expected Losses

E = B * D

0.10%

0.17%

Table 9 uses fairly optimistic estimates for the expected loss, as it only considers 90-day delinquencies (it does not

include a provision for 30- or 60-day delinquencies), the roll rate assumption at 20% is on the low-end of the range,

and the severity rate at 15% is optimistic, as these loans became delinquent as a result of a natural disaster. Loans

impacted by a natural disaster are more likely to have some physical damage that would impact the value of the

property, and this is not reflected in the above estimates. Nevertheless, 21 months after the peak 90-day delinquency

rate, based on the above analysis, we would have expected at least 10 basis points of loss on the low-LTV transaction

and 17 basis points of loss on the high-LTV transaction.

Figure 23 shows the actual cumulative losses through December 2019 for the low-LTV transaction.

Figure 23: Post-Event Low-LTV Net Loss Rates

The losses are segmented between loans in Florida (yellow bars) and loans outside of Florida (blue bars). Through

December 2019, actual cumulative losses on this transaction were less than 0.3 basis points for both populations of

loans, significantly less than we would have expected based on the assumption that shocks arising from a catastrophe

could be treated the same as generic ones. Part of the reason for this is because following Hurricane Irma, like other

catastrophes, homeowners may receive financial assistance or grace on loan repayment. For example, Freddie Mac

offers disaster relief options to borrowers whose loan is owned by it and whose property is located in a presidentially-

declared major disaster area. The relief offers suspending foreclosure for a period of 12 months by providing

forbearance for up to 12 months following the event, among other benefits, including waiving penalties or late fees

and not reporting missed payments to the credit bureaus. Hurricane Irma landed in September 2017, and the disaster

relief ended in late 2018. This is likely why

Figure 23 does not show any losses for Florida until early 2019, despite the

large rate of delinquencies of 90+ days observed throughout 2018. Nevertheless, the comparison between the

expectation presented in

Table 9 and the observation in Figure 23 does not support the assumption that home price

shocks following a natural disaster would have a similar impact on credit risk as other shocks. That is, if home price

shocks following a disaster could be expected to behave in this fashion, then this transaction should have incurred

significantly more losses relative to actual observations by the end of 2019, which includes 12 months after the

expiration of the disaster relief assistance.

0.00

0.05

0.10

0.15

0.20

0.25

0.30

Loss Rate (Net Loss / Original UPB) (bps)

Calendar Month

ACIS 2017-3 / STACR 2017-DNA2

Resolved Loans

Net Loss Rate

Source: Freddie Mac, M-PIRe

All Loans Excluding Florida Florida

Figure 24: Post-Event High-LTV Net Loss Rates

Figure 24

provides similar information for the ACIS 2017-5 / STACR 2017-HQA2 transaction. For this transaction,

cumulative losses in Florida are actually negative, indicating the amount recovered from foreclosures plus mortgage

guaranty insurance has exceeded the defaulting principal balance plus expenses and delinquent interest.

These results are not consistent with the model assumptions in the prior section, which expected an increase in actual

losses following a natural disaster. This type of performance is not unique to Hurricane Irma, and similar results have

been observed following other events. Specifically, the data indicates that while natural disasters may result in a spike

in delinquencies, the majority of impacted loans either cure or prepay following the event. Few mortgages result in a

realized loss to the investor. It is difficult to pinpoint the reason for this observation, but potential contributors are:

1. Some borrowers did have flood insurance during these events.

2. The disaster assistance offered by mortgage investors provided effective relief to borrowers.

3. Some uninsured borrowers received state or federal assistance to repair their homes.

4. Some borrowers may have been uninsured for flood losses but insured for wind losses and may have received

payment from their insurance company for some of their flood losses even though such protection was not

provided under their policies.

5. Some historical natural disasters occurred during periods of relative strength in the housing market. That is, for

the sparse record of observable extreme events, most did not occur while home prices were declining and the

borrowers were “underwater” on their mortgages. Thus, relying on the historical record may understate the true

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

Loss Rate (Net Loss / Original UPB) (bps)

Calendar Month

ACIS 2017-5 / STACR-HQA2

Resolved Loans

Net Loss Rate

Source: Freddie Mac, M-PIRe

All Loans Excluding Florida Florida

potential risk associated with this issue if a disaster were to occur under stressed economic conditions, or if future

disaster assistance programs did not provide the same relief as past ones.

6. Homeowners and communities may have a stronger desire to rebuild following a natural disasters than during

other circumstances that could affect their mortgages.

The disconnect between model estimates resulting from the methodology presented in this paper and historical

performance indicates that more study is still needed on this issue. Nevertheless, the estimates and methodology can

provide a useful benchmark in determining a priori expectations for mortgage investors to which historical

performance can be compared. Based on our examination of Hurricane Irma, it appears that the threat of natural

catastrophes to mortgage credit risk may be smaller than this a priori expectation would indicate, possibly due to the

fact that assistance or forgiveness programs have acted as mitigating factors to prevent a realization of the potential

downside associated with uninsured homeowners defaulting on their mortgages.

Note that these empirically observed outcomes are only particular to one disaster. Different disasters and different

impacted regions may have materially different outcomes that what was observed after Hurricane Irma. For instance,

the relatively benign to-date outcomes of the impacted loans may be partially driven by Florida’s longer-than-average

foreclosure timelines. Reviewing similar data post-Hurricane-Harvey shows moderately higher losses in the region (as

compared to similar data post-Irma). Texas has a quicker foreclosure process, and delinquencies may translate into

losses more quickly than in Florida, all else equal. Nevertheless, our review finds that actual losses from recent natural

catastrophes are relatively small as compared to the model estimates and methodology presented in the above

sections. This is consistent with the finding from Irma: The threat of natural catastrophes to mortgages credit risk may

be smaller than this a priori expectation would indicate.

However, if the frequency of natural disasters increases with climate change, then it would also become increasingly

costly for taxpayers and GSEs to continue to provide these types of assistance to affected borrowers, and if the same

degree of assistance were not to continue, then the performance of mortgages following these events could be worse.

Furthermore, borrowers in areas impacted by climate change may be less resilient in the future. They may decide not

to rebuild because of the threat that their properties could continue to be impacted by similar events, so the transition

from delinquency to foreclosure and loss could rise in the future.

Because of limitations in the historical data available to analyze mortgage performance during disasters across

different credit cycles, it could be possible that natural catastrophes may pose more mortgage credit downside

relative to recent experience. But the historical record simply lacks any “perfect storm” observations that contain the

combination of event severity and economic circumstances that could give rise to significant impairment in mortgage

investments, despite the possibility that such an event could occur. Thus, any effort to quantify the potential effects

of this issue should recognize the limitations of the historical record and employ a methodology that recognizes the

full range of possible events that could occur.

5. Appendix

5.1 MAPS: INCREASE IN EXPECTED FLOOD LOSSES UNDER SEA LEVEL RISE SCENARIOS

Each map in this report shows the increase in total flood losses under sea level rise scenarios. Results are spatially

smoothed using inverse distance weighting (IDW) interpolation performed on the modeled market basket locations.

The ratios of total flood losses for the market basket locations were calculated between different sea level rise

scenarios and were run through ESRI’s IDW tool. IDW converts an input dataset of points into a continuous one-

kilometer grid layer by calculating a weighted average of the 10 closest data points at each cell location. The tool is

able to smooth the values of cells at locations that lack sampled points by assuming that the variable being mapped

decreases in influence with distance and by weighting the surrounding values based on their inverse distances from

that location.

Map 1: Ratios of Expected Flood Losses; Sea Level Rise Scenarios to Current Sea Level - Gulf of Mexico

Map 2: Expected Flood Losses; Sea Level Rise Scenarios to Current Sea Level - Florida to Georgia

Map 3: Expected Flood Losses; Sea Level Rise Scenarios to Current Sea Level – Carolinas

Map 4: Expected Flood Losses; Sea Level Rise Scenarios to Current Sea Level – Mid-Atlantic

Map 5: Expected Flood Losses; Sea Level Rise Scenarios to Current Sea Level – Northeast

5.2 EXHIBITS

Exhibit 7: Development of NFIP Take-Up Rate Assumptions

(1) (2) (3) (4) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Estimated Single Family Home

NFIP Take-up NFIP Policy Count Number of Single Family Homes Single Family Homes not Insured by NFIP

State Total SFHA Non SFHA Total SFHA Non SFHA Total SFHA Non SFHA Total SFHA Non SFHA

Alabama 1.7% 23.5% 0.9% 32,517 16,735 15,782 1,863,170 71,137 1,792,033 1,830,653 54,402 1,776,251

Arizona 1.1% 28.0% 0.5% 25,304 14,520 10,784 2,332,748 51,781 2,280,967 2,307,444 37,261 2,270,183

Arkansas 1.1% 14.0% 0.4% 12,344 7,545 4,799 1,133,948 53,777 1,080,171 1,121,604 46,232 1,075,372

California 1.9% 31.6% 1.1% 185,387 85,848 99,539 9,628,644 272,066 9,356,578 9,443,257 186,218 9,257,039

Colorado 0.8% 19.3% 0.6% 13,753 4,377 9,376 1,720,260 22,702 1,697,558 1,706,507 18,325 1,688,182

Connecticut 2.4% 29.7% 1.1% 23,599 13,243 10,356 985,578 44,629 940,949 961,979 31,386 930,593

Delaware 5.0% 41.0% 2.2% 17,257 10,106 7,151 348,602 24,626 323,976 331,345 14,520 316,825

Florida 15.5% 51.3% 9.0% 999,832 507,579 492,253 6,437,823 989,468 5,448,355 5,437,991 481,889 4,956,102

Georgia 2.2% 24.2% 1.4% 74,149 30,042 44,107 3,335,899 124,220 3,211,679 3,261,750 94,178 3,167,572

Idaho 0.8% 12.4% 0.4% 4,667 2,263 2,404 594,681 18,320 576,361 590,014 16,057 573,957

Illinois 0.8% 22.7% 0.4% 29,013 16,626 12,387 3,580,795 73,396 3,507,399 3,551,782 56,770 3,495,012

Indiana 0.8% 14.7% 0.4% 18,492 9,908 8,584 2,322,891 67,300 2,255,591 2,304,399 57,392

2,247,007

Iowa 0.8% 13.7% 0.4% 8,705 4,342 4,363 1,115,953 31,717 1,084,236 1,107,248 27,375 1,079,873

Kansas 0.7% 10.3% 0.4% 7,467 3,094 4,373 1,030,486 30,160 1,000,326 1,023,019 27,066 995,953

Kentucky 1.0% 15.0% 0.4% 16,087 10,372 5,715 1,604,688 69,224 1,535,464 1,588,601 58,852 1,529,749

Louisiana 25.8% 51.5% 19.8% 424,592 160,350 264,242 1,647,501 311,322 1,336,179 1,222,909 150,972 1,071,937

Maine 1.0% 11.7% 0.6% 6,076 2,586 3,490 596,095 22,110 573,985 590,019 19,524 570,495

Maryland 2.0% 45.3% 1.2% 35,239 14,392 20,847 1,806,078 31,761 1,774,317 1,770,839 17,369 1,753,470

Massachusetts 2.2% 24.6% 1.2% 36,375 17,075 19,300 1,671,803 69,436 1,602,367 1,635,428 52,361 1,583,067

Michigan 0.4% 12.3% 0.2% 16,850 10,606 6,244 3,756,861 86,119 3,670,742 3,740,011 75,513 3,664,498

Minnesota 0.4% 8.5% 0.2% 7,254 2,692 4,562 1,875,707 31,847 1,843,860 1,868,453 29,155 1,839,298

Mississippi 4.7% 21.1% 3.1% 52,221 20,150 32,071 1,117,509 95,311 1,022,198 1,065,288 75,161 990,127

Missouri 0.7% 16.7% 0.3% 14,586 8,233 6,353 2,208,296 49,306 2,158,990 2,193,710 41,073 2,152,637

Montana 1.1% 9.9% 0.8% 4,519 1,418 3,101 419,516 14,365 405,151 414,997 12,947 402,050

Nebraska 1.0% 16.8% 0.4% 6,763 4,468 2,295 659,090 26,518 632,572 652,327

22,050

630,277

Nevada 1.1% 32.1% 0.5% 9,745 5,587 4,158 850,493 17,423 833,070 840,748 11,836 828,912

New Hampshire

1.0% 15.3% 0.5% 4,616 2,505 2,111 468,372 16,356 452,016 463,756 13,851 449,905

New Jersey 5.4% 48.2% 1.4% 123,495 93,404 30,091 2,303,561 193,976 2,109,585 2,180,066 100,572 2,079,494

New Mexico 1.4% 13.8% 0.5% 10,832 6,889 3,943 787,714 49,745 737,969 776,882 42,856 734,026

New York 3.0% 37.9% 1.8% 123,657 54,793 68,864 4,074,892 144,683 3,930,209 3,951,235 89,890 3,861,345

North Carolina 2.9% 40.5% 1.2% 107,484 62,893 44,591 3,727,599 155,106 3,572,493 3,620,115 92,213 3,527,902

North Dakota 3.2% 10.8% 2.7% 8,344 1,870 6,474 256,899 17,381 239,518 248,555 15,511 233,044

Ohio 0.6% 16.3% 0.3% 24,911 13,602 11,309 3,980,674 83,460 3,897,214 3,955,763 69,858 3,885,905

Oklahoma 0.8% 11.2% 0.4% 10,833 4,931 5,902 1,441,948 43,988 1,397,960 1,431,115 39,057 1,392,058

Oregon 1.6% 23.4% 0.8% 20,780 10,516 10,264 1,320,983 44,951 1,276,032 1,300,203 34,435 1,265,768

Pennsylvania 0.9% 16.1% 0.5% 40,193 18,656 21,537 4,506,700 116,128 4,390,572 4,466,507 97,472 4,369,035

Rhode Island 3.4% 43.2% 1.8% 9,370 4,491 4,879 276,748 10,384 266,364 267,378 5,893 261,485

South Carolina 7.8% (Note 7) 3.4% 142,467 83,533 58,934 1,834,704 102,795 1,731,909 1,692,237 (Note 7) 0

South Dakota 0.9% 10.9% 0.5% 2,752 1,203 1,549 308,314 11,040 297,274 305,562 9,837 295,725

Tennessee 0.

9% 19.3% 0.5% 21,994 9,476 12,518 2,352,974 49,165 2,303,809 2,330,980 39,689 2,291,291

Texas 8.4% 33.7% 7.1% 669,789 133,754 536,035 7,978,382 397,440 7,580,942 7,308,593 263,686 7,044,907

Utah 0.3% 7.3% 0.3% 2,708 590 2,118 820,270 8,067 812,203 817,562 7,477 810,085

Vermont 0.9% 16.8% 0.5% 2,335 1,177 1,158 256,291 7,012 249,279 253,956 5,835 248,121

Virginia 3.1% 41.6% 1.8% 84,188 35,793 48,395 2,703,730 85,942 2,617,788 2,619,542 50,149 2,569,393

Washington 1.3% 32.2% 0.6% 28,092 15,321 12,771 2,233,400 47,631 2,185,769 2,205,308 32,310 2,172,998

West Virginia 1.6% 13.5% 0.6% 12,513 8,301 4,212 781,324 61,348 719,976 768,811 53,047 715,764

Wisconsin 0.5% 10.4% 0.2% 9,780 4,985 4,795 1,985,427 47,783 1,937,644 1,975,647 42,798 1,932,849

Wyoming 0.6% 10.5% 0.4% 1,485 564 921 230,105 5,371 224,734 228,620 4,807 223,813

Total (Note 6) 3.6% 34.2% 2.1% 3,545,411 1,553,404 1,992,007 99,276,126 4,399,795 94,876,331 95,730,715 2,894,761 92,907,315

Notes:

4. Single family homes inside and outside the SFHA were calculated by allocating the total home counts with distributions based on Market Baskets.

6. Total row does not include Alaska, District of Columbia, Hawaii, or Territories of the United States.

7. Take-up rate estimates inside and outside SFHA for South Carolina are currently unavailable.

NFIP single family dwelling policies, in force as of September 30, 2018, were determined using the June 2019 release of OpenFEMA NFIP policy data. These include policies for

mobile homes. Policies were counted in total, inside the SFHA, and outside of the SFHA.

Take-up in Columns (2) through (4) is the ratio of NFIP single family dwelling policies to single family homes. Policies and homes were counted in total, inside the Special Flood

Hazard Area (SFHA), and outside the SFHA.

Total single family homes were determined using the 2017 American Community Survey 5 year estimate. Single family attached, single family detached, and mobile home counts

were included.

Each Market Basket represents a sample of actual locations of real risks in the marketplace. Market Baskets contain approximately 10% of the single family home locations in

each state.

Exhibit 8: Modeled Insured and Uninsured Losses by MSA

(1) (2) (3) (4

) (5) (6) (7) (8) (9)

Metropolitan Single Family Average Annual Losses Annual Losses

Percent of

Statistical Area Title Residences Insured Uninsured Total Insured Uninsured Total Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 2) (Note 2) = (2) * (3) = (2) * (4) = (5) + (6) = (6) / (7)

Hilton Head Island-Bluffton, SC 72,147 $225 $1,763 $1,988 $16,251,239 $127,206,009 $143,457,248 88.7%

Houma-Thibodaux, LA 60,417 256 669 925 15,450,541 40,425,486 55,876,027 72.3%

Naples-Marco Island, FL 95,685 357 654 1,011 34,175,531 62,606,094 96,781,625 64.7%

Punta Gorda, FL 70,708 350 614 963 24,735,052 43,380,277 68,115,329 63.7%

Napa, CA 41,795 9 538 547 377,749 22,485,975 22,863,724 98.3%

Cape Coral-Fort Myers, FL 238,495 269 534 803 64,053,011 127,404,015 191,457,026 66.5%

New Orleans-Metairie, LA 371,993 177 516 694 65,886,138 192,129,455 258,015,593 74.5%

Sebastian-Vero Beach, FL 52,438 35 470 505 1,841,690 24,637,086 26,478,775 93.0%

Ocean City, NJ 75,589 166 440 606 12,511,248 33,295,161 45,806,409 72.7%

Wenatchee, WA 37,375 6 329 335 209,661 12,303,994 12,513,655 98.3%

Beaumont-Port Arthur, TX 120,529 32 321 353 3,884,590 38,707,512 42,592,102 90.9%

Jacksonville, NC 53,490 138 318 456 7,378,136 16,986,873 24,365,009 69.7%

Santa Cruz-Watsonville, CA 76,110 7 295 303 569,633 22,477,038 23,046,671 97.5%

Santa Rosa-Petaluma, CA 159,048 14 242 256 2,262,874 38,457,232 40,720,106 94.4%

St. George, UT 52,060 0 225 225 0 11,735,854 11,735,854 100.0%

Lake Charles, LA 62,822 104 219 323 6,522,451 13,750,763 20,273,215 67.8%

Wilmington, NC 93,347 40 209 248 3,711,650 19,482,542 23,194,192 84.0%

Bend, OR 65,818 1 199 200 46,605 13,108,708 13,155,313 99.6%

Charleston-North Charleston, SC 212,033 95 198 293 20,196,030 41,884,724 62,080,754 67.5%

San Jose-Sunnyvale-Santa Clara, CA 433,002 13 196 209 5,505,301 85,069,419 90,574,720 93.9%

Santa Maria-Santa Barbara, CA 101,415 5 191 197 552,315 19,402,243 19,954,557 97.2%

Oxnard-Thousand Oaks-Ventura, CA 214,030 16 180 196 3,434,263 38,537,513 41,971,776 91.8%

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 52 177 229 66,996,364 229,248,792 296,245,156 77.4%

Grants Pass, OR 27,766 18 176 194 496,217 4,898,611 5,394,828 90.8%

Hot Springs, AR 34,799 0 176 176 5,908 6,130,445 6,136,353 99.9%

Brunswick, GA 37,880 26 176 202 986,323 6,671,067 7,657,390 87.1%

Daphne-Fairhope-Foley, AL 70,318 22 168 190 1,544,017 11,805,790 13,349,807 88.4%

Homosassa Springs, FL 55,727 78 167 245 4,329,909 9,325,620 13,655,529 68.3%

Los Angeles-Long Beach-Anaheim, CA 2,637,020 2 166 169 5,450,959 438,945,936 444,396,895 98.8%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 156,898 62 165 227 9,680,192 25,881,580 35,561,772 72.8%

Charleston, WV 91,961 11 164 175 976,720 15,075,368 16,052,087 93.9%

North Port-Sarasota-Bradenton, FL 263,971 76 163 239 20,142,829 42,996,783 63,139,612 68.1%

Lake Havasu City-Kingman, AZ 69,892 20 163 182 1,375,208 11,361,314 12,736,521

89.2%

uba City, CA 46,641 3 161 164 136,787 7,490,238 7,627,026 98.2%

Flagstaff, AZ 43,823 9 146 155 373,357 6,416,249 6,789,606 94.5%

Bridgeport-Stamford-Norwalk, CT 238,970 23 142 165 5,463,805 33,870,132 39,333,937 86.1%

San Francisco-Oakland-Berkeley, CA 1,053,639 5 130 136 5,544,929 137,406,918 142,951,847 96.1%

Riverside-San Bernardino-Ontario, CA 1,142,048 4 129 133 4,244,809 147,637,490 151,882,300 97.2%

Eugene-Springfield, OR 108,311 9 129 137 931,339 13,933,106 14,864,445 93.7%

Jacksonville, FL 429,019 31 128 159 13,090,295 54,957,280 68,047,575 80.8%

Palm Bay-Melbourne-Titusville, FL 190,403 14 128 142 2,725,675 24,382,064 27,107,739 89.9%

Austin-Round Rock-Georgetown, TX 496,025 13 124 137 6,538,712 61,379,423 67,918,136 90.4%

Port St. Lucie, FL 147,959 33 120 154 4,935,659 17,806,741 22,742,400 78.3%

Portland-South Portland, ME 193,001 8 120 128 1,559,702 23,093,226 24,652,928 93.7%

Bowling Green, KY 49,659 3 117 120 144,621 5,825,760 5,970,381 97.6%

Houston-The Woodlands-Sugar Land, TX 1,649,856 12 117 129 19,638,279 192,737,803 212,376,082 90.8%

Lafayette, LA 141,177 73 116 189 10,257,000 16,435,980 26,692,981 61.6%

Madera, CA 40,942 6 111 116 231,268 4,537,215 4,768,484 95.2%

Gulfport-Biloxi, MS 130,147 21 109 130 2,750,538 14,209,538 16,960,076 83.8%

Longview, WA 30,847 0 109 109 421 3,367,033 3,367,454 100.0%

Tampa-St. Petersburg-Clearwater, FL 849,853 53 108 161 45,283,390 91,553,938 136,837,328 66.9%

Pittsfield, MA 46,831 11 107 118 517,556 5,011,584 5,529,140 90.6%

Reno, NV 125,517 8 105 113 951,172 13,185,536 14,136,708 93.3%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 31 105 135 6,747,605 23,154,651 29,902,256 77.4%

Sioux City, IA-NE-SD 45,531 1 105 106 49,585 4,771,404 4,820,989 99.0%

Beckley, WV 44,391 7 104 111 296,101 4,638,646 4,934,747 94.0%

New Haven-Milford, CT 213,870 15 103 119 3,264,959 22,099,225 25,364,184 87.1%

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 17 102 120 57,037,811 332,933,472 389,971,283 85.4%

San Luis Obispo-Paso Robles, CA 87,318 4 98 101 308,638 8,517,183 8,825,820 96.5%

Barnstable Town, MA 138,885 6 96 102 846,437 13,271,242 14,117,678 94.0%

Prescott Valley-Prescott, AZ 77,916 2 95 97 165,379 7,394,507 7,559,886 97.8%

Albany-Lebanon, OR 35,729 2 94 96 87,088 3,342,659 3,429,747 97.5%

Williamsport, PA 39,686 3 93 96 129,168 3,681,444 3,810,613 96.6%

San Antonio-New Braunfels, TX 608,824 11 92 103 6,823,906 56,180,166 63,004,073 89.2%

Salinas, CA 97,116 12 92 104 1,156,612 8,941,956 10,098,568 88.5%

Cumberland, MD-WV 35,918 4 90 94 127,897 3,238,922 3,366,819 96.2%

Mobile, AL 139,996 12 90 102 1,653,555 12,613,633 14,267,187 88.4%

Tucson, AZ 306,712 3 88 91 1,072,406 26,935,244 28,007,650 96.2%

Savannah, GA 109,497 14 87 101 1,538,723 9,543,127 11,081,849 86.1%

Billings, MT 56,865

2 86 87 85,

618 4,863,330 4,948,948 98.3%

Medford, OR 64,532 0 85 86 17,400 5,506,433 5,523,832 99.7%

Chambersburg-Waynesboro, PA 49,867 0 84 84 4,884 4,172,498 4,177,382 99.9%

New Bern, NC 41,778 23 83 106 971,149 3,472,857 4,444,006 78.1%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 23 83 106 12,096,352 43,311,633 55,407,985 78.2%

Mount Vernon-Anacortes, WA 40,761 4 83 86 148,860 3,368,352 3,517,213 95.8%

(1) (2) (3) (4) (5

) (6) (7) (8) (9)

Metropolitan Single Family Average Annual Losses Annual Losses

Percent of

Statistical Area Title Residences Insured Uninsured Total Insured Uninsured Total Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 2) (Note 2) = (2) * (3) = (2) * (4) = (5) + (6) = (6) / (7)

Springfield, MO 151,362 2 29 31 294,621 4,397,124 4,691,745 93.7%

Albany-Schenectady-Troy, NY 243,598 2 29 31 443,180 7,068,891 7,512,071 94.1%

Grand Junction, CO 48,184 0 29 29 8,314 1,396,021 1,404,336 99.4%

Tulsa, OK 311,486 1 29 30 206,192 9,005,964 9,212,156 97.8%

Dallas-Fort Worth-Arlington, TX 1,774,642 3 28 32 5,713,468 50,555,812 56,269,280 89.8%

Eau Claire, WI 52,052 0 28 28 2,093 1,477,808 1,479,901 99.9%

Jackson, TN 61,377 0 28 28 2,354 1,740,277 1,742,631 99.9%

Lansing-East Lansing, MI 170,090 1 28 29 186,394 4,804,856 4,991,250 96.3%

Hickory-Lenoir-Morganton, NC 112,229 3 28 31 289,500 3,151,409 3,440,909 91.6%

Idaho Falls, ID 40,678 0 28 28 0 1,141,355 1,141,355 100.0%

Cleveland, TN 37,264 0 28 28 959 1,041,016 1,041,975 99.9%

Columbus, OH 587,034 1 28 28 532,251 16,186,391 16,718,642 96.8%

Blacksburg-Christiansburg, VA 47,600 18 27 46 871,150 1,305,687 2,176,837 60.0%

Terre Haute, IN 63,600 3 27 31 214,466 1,741,337 1,955,803 89.0%

Greenville-Anderson, SC 259,626 8 27 35 2,098,696 7,050,206 9,148,901 77.1%

Ann Arbor, MI 94,907 0 27 27 2,336 2,566,434 2,568,770 99.9%

Omaha-Council Bluffs, NE-IA 284,447 0 27 27 43,071 7,690,637 7,733,708 99.4%

Worcester, MA-CT 242,324 1 27 28 141,537 6,543,531 6,685,068 97.9%

El Centro, CA 38,508 1 27 28 30,355 1,038,890 1,069,245 97.2%

Laredo, TX 55,325 6 27 33 312,234 1,490,884 1,803,118 82.7%

Saginaw, MI 68,876 1 27 27 35,097 1,855,129 1,890,225 98.1%

Augusta-Richmond County, GA-SC 179,180 2 27 29 379,153 4,824,769 5,203,922 92.7%

Gettysburg, PA 34,673 0 27 27 5,824 928,257 934,082 99.4%

Elkhart-Goshen, IN 58,286 10 27 37 607,532 1,558,758 2,166,290 72.0%

Champaign-Urbana, IL 60,883 0 27 27 2,792 1,619,582 1,622,374 99.8%

Lewiston, ID-WA 19,993 1 26 27 20,101 528,525 548,626 96.3%

Minneapolis-St. Paul-Bloomington, MN-WI 1,021,609 1 26 27 569,651 26,711,165 27,280,816 97.9%

Florence-Muscle Shoals, AL 53,242 4 26 31 235,210 1,391,748 1,626,957 85.5%

Lancaster, PA 157,703 7 26 33 1,028,179 4,120,851 5,149,030 80.0%

Indianapolis-Carmel-Anderson, IN 634,827 1 26 27 858,778 16,464,041 17,322,819 95.0%

Charlotte-Concord-Gastonia, NC-SC 728,959 4 26 30 2,702,463 18,893,738 21,596,201 87.5%

Cincinnati, OH-KY-IN 656,376 0 26 26 283,382 16,828,298 17,111,680 98.3%

Niles, MI 60,741 0 26 26 64 1,549,685 1,549,749 100.0%

Bloomington, IL 47,664 0

25 25 684 1,

211,399 1,212,084 99.9%

Atlanta-Sandy Springs-Alpharetta, GA 1,621,740 2 25 27 2,485,022 40,784,706 43,269,728 94.3%

Providence-Warwick, RI-MA 404,392 4 25 29 1,539,531 10,168,692 11,708,224 86.9%

Joplin, MO 60,391 6 25 31 354,509 1,512,491 1,867,000 81.0%

Columbia, SC 235,164 4 25 29 1,042,109 5,873,176 6,915,285 84.9%

Utica-Rome, NY 89,908 1 25 26 124,586 2,215,031 2,339,617 94.7%

Albany, GA 42,795 2 25 26 79,161 1,052,893 1,132,054 93.0%

York-Hanover, PA 146,752 1 25 26 207,109 3,608,997 3,816,106 94.6%

Durham-Chapel Hill, NC 176,197 2 24 26 311,514 4,309,944 4,621,457 93.3%

Kansas City, MO-KS 684,093 1 24 25 665,309 16,602,317 17,267,626 96.1%

Pueblo, CO 55,343 0 24 24 451 1,342,872 1,343,324 100.0%

South Bend-Mishawaka, IN-MI 111,987 0 24 24 5,421 2,712,906 2,718,327 99.8%

Richmond, VA 390,291 4 24 28 1,597,396 9,229,702 10,827,098 85.2%

Macon-Bibb County, GA 70,206 1 23 24 83,400 1,631,034 1,714,435 95.1%

Dalton, GA 35,740 1 23 24 24,165 829,595 853,761 97.2%

Birmingham-Hoover, AL 348,098 1 23 24 397,704 8,055,312 8,453,016 95.3%

Olympia-Lacey-Tumwater, WA 81,895 2 23 25 174,067 1,886,079 2,060,146 91.6%

Colorado Springs, CO 210,551 0 23 23 21,183 4,839,676 4,860,859 99.6%

Lakeland-Winter Haven, FL 183,132 4 23 27 656,094 4,200,766 4,856,860 86.5%

Logan, UT-ID 32,839 0 23 23 0 750,229 750,229 100.0%

Green Bay, WI 101,075 0 23 23 7,394 2,296,619 2,304,013 99.7%

Spartanburg, SC 89,595 3 23 25 249,447 2,033,145 2,282,592 89.1%

Pittsburgh, PA 848,872 2 22 24 1,429,786 18,987,968 20,417,754 93.0%

Denver-Aurora-Lakewood, CO 769,541 0 22 23 234,889 17,116,733 17,351,622 98.6%

Walla Walla, WA 17,451 1 22 23 16,771 379,492 396,263 95.8%

Canton-Massillon, OH 140,756 0 22 22 28,422 3,058,414 3,086,836 99.1%

Provo-Orem, UT 126,241 0 22 22 470 2,737,920 2,738,390 100.0%

Detroit-Warren-Dearborn, MI 1,460,406 0 22 22 545,185 31,666,096 32,211,281 98.3%

Oshkosh-Neenah, WI 52,178 1 22 23 66,363 1,127,390 1,193,753 94.4%

Ames, IA 31,807 0 21 22 2,007 683,400 685,408 99.7%

Lima, OH 34,734 0 21 21 356 740,920 741,277 100.0%

Racine, WI 59,802 0 21 21 0 1,252,923 1,252,923 100.0%

Fort Wayne, IN 131,241 0 21 21 17,398 2,744,139 2,761,538 99.4%

Pocatello, ID 26,373 3 21 24 84,814 548,023 632,837 86.6%

Florence, SC 57,428 2 21 23 120,186 1,192,909 1,313,094 90.8%

Memphis, TN-MS-AR 408,730 0 21 21 141,057 8,478,334 8,619,391 98.4%

Columbus, IN 25,959 1 21 22 36,831 532,505

569,336 93.

Reading, PA 132,322 0 20 20 2,893 2,698,490 2,701,383 99.9%

Rockford, IL 110,588 3 20 24 377,319 2,247,550 2,624,869 85.6%

Davenport-Moline-Rock Island, IA-IL 129,399 1 20 21 113,171 2,628,222 2,741,393 95.9%

Jackson, MI 54,228 4 20 24 192,639 1,099,469 1,292,108 85.1%

Battle Creek, MI 45,127 2 20 22 69,009 907,318 976,327 92.9%

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Metropolitan Single Family Average Annual Losses Annual Losses

Percent of

Statistical Area Title Residences Insured Uninsured Total Insured Uninsured Total Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 2) (Note 2) = (2) * (3) = (2) * (4) = (5) + (6) = (6) / (7)

Springfield, OH 48,584 0 10 10 308 463,692 464,000 99.9%

Watertown-Fort Drum, NY 37,853 0 9 9 5,458 351,501 356,959 98.5%

Warner Robins, GA 54,033 2 9 11 84,896 493,155 578,051 85.3%

Danville, IL 28,655 0 9 9 2,192 256,246 258,437 99.2%

Bremerton-Silverdale-Port Orchard, WA 81,736 0 8 8 2,800 635,800 638,600 99.6%

Total All MSAs 74,570,040 $10 $70 $80 $766,115,468 $5,188,121,769 $5,954,237,237 87.1%

Total Outside MSAs 16,212,126 $7 $61 $68 $110,374,631 $989,912,415 $1,100,287,046 90.0%

Total 90,782,166 $10 $68 $78 $876,490,099 $6,178,034,184 $7,054,524,283 87.6%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Columns (3) to (5) contain loss data that from KatRisk catastrophe model runs on a subset of Milliman market basket locations. Modeling runs were set to a standard sea level rise scenario.

3. Insured and Uninsured Losses are based on estimates of NFIP take-up rates and coverages.

Exhibit 9: Change in Storm Surge Losses for Sea Level Rise Scenarios Compared to Current

(1) (2) (3) (

4) (5) (6) (7) (8) (9)

Single Family Percent Increase in Single Family Residences

Residences Exposed Exposed to Storm Surge Percent Increase in Storm Surge Uninsured Loss Percent Increase in Storm Surge Loss

Metropolitan Single Family to Storm Surge, Medium Sea Level Rise High Sea Level Rise Medium Sea Level Rise High Sea Level Rise Medium Sea Level Rise High Sea Level Rise

Statistical Area Name Residences Current Scenario vs. Current Sea Level vs. Current Sea Level vs. Current Sea Level vs. Current Sea Level vs. Current Sea Level vs. Current Sea Level

(Note 1) (Note 1) (Note 2) (Note 3) (Note 4) (Note 5) (Note 6) (Note 7) (Note 8)

Atlantic City-Hammonton, NJ 84,496 30,099 1.0% 3.0% 46.7% 189.4% 50.6% 203.2%

Corpus Christi, TX 121,799 35,100 7.7% 12.3% 48.3% 109.0% 49.1% 111.2%

Orlando-Kissimmee-Sanford, FL 649,296 400 25.0% 50.0% 46.9% 116.4% 48.5% 156.2%

Salisbury, MD-DE 161,632 61,074 3.4% 7.2% 47.0% 177.7% 46.8% 174.8%

Ocean City, NJ 75,589 64,290.64 2.6% 5.6% 43.1% 152.4% 42.9% 150.7%

California-Lexington Park, MD 35,784 4,598 0.0% 8.7% 21.9% 94.2% 36.8% 155.3%

Baltimore-Columbia-Towson, MD 854,053 33,298 4.5% 15.3% 34.7% 151.2% 34.6% 149.0%

Houma-Thibodaux, LA 60,417 60,417 0.0% 0.0% 33.8% 77.9% 33.6% 77.2%

Washington-Arlington-Alexandria, DC-VA-MD-WV

1,422,224 6,899 5.8% 11.6% 34.2% 127.2% 33.5% 125.5%

Brunswick, GA 37,880 28,785 0.3% 1.7% 32.0% 100.5% 32.0% 100.2%

Vineland-Bridgeton, NJ 40,871 6,096 3.3% 4.9% 14.8% 53.8% 31.8% 107.4%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 323,017 3.3% 6.0% 30.4% 103.3% 31.0% 105.7%

New Bern, NC 41,778 21,189 2.4% 9.4% 28.1% 97.2% 29.3% 99.9%

Charleston-North Charleston, SC 212,033 102,416 1.8% 3.5% 30.1% 96.9% 28.8% 92.1%

Savannah, GA 109,497 78,898 2.3% 5.7% 28.1% 89.9% 28.2% 90.3%

Dover, DE 52,288 2,399 4.2% 12.5% 27.6% 87.4% 28.0% 88.8%

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 340,686 2.4% 7.0% 26.9% 101.2% 27.3% 102.5%

Jacksonville, FL 429,019 101,404 3.6% 10.0% 26.6% 87.8% 27.1% 89.0%

Lafayette, LA 141,177 120,065 1.4% 2.6% 25.6% 52.7% 25.1% 51.4%

Hinesville, GA 19,678 4,595 0.0% 2.2% 23.4% 69.1% 23.5% 69.4%

Sebastian-Vero Beach, FL 52,438 8,906 4.5% 11.2% 23.8% 80.4% 23.5% 79.0%

Providence-Warwick, RI-MA 404,392 18,704 3.7% 13.4% 22.3% 95.9% 23.4% 100.3%

Norwich-New London, CT 85,235 8,503 3.5% 11.8% 23.4% 95.5% 23.3% 93.4%

Richmond, VA 390,291 1,800 11.1% 22.2% 26.9% 82.9% 23.3% 74.1%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 76,920 0.5% 1.7% 21.9% 70.9% 22.1% 71.9%

Barnstable Town, MA 138,885 25,497 3.5% 9.0% 21.6% 81.0% 22.1% 82.7%

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 477,787 3.0% 7.5% 22.4% 66.5% 22.1% 65.4%

Lake Charles, LA 62,822 59,721 0.0% 0.0% 22.2% 49.6% 22.1% 49.2%

Panama City, FL 55,465 16,290 7.4% 20.9% 21.2% 84.6% 21.2% 84.1%

Hartford-East Hartford-Middletown, CT 332,941 6,601 1.5% 3.0% 21.4% 80.4% 20.8% 77.8%

Port St. Lucie, FL 147,959 20,094 9.0% 28.9% 21.8% 76.2% 20.6% 70.2%

Bridgeport-Stamford-Norwalk, CT 238,970 12,598 4.0% 5.6% 21.4% 84.7% 20.6% 81.2%

Baton Rouge, LA 241,791 90,134 1.4% 3.6% 19.5% 50.3% 20.4% 52.9%

Naples-Marco Island, FL 95,685 94,485 0.0% 0.1% 21.1% 60.1% 20.2% 56.9%

Boston-Cambridge-Newton, MA-NH 1,042,909 36,597 4.6% 13.7% 20.0% 79.2% 19.5% 76.2%

Crestview-Fort Walton Beach-Destin, FL 92,891 23,598 5.9% 12.7% 19.1% 72.2% 19.3% 72.5%

Brownsville-Harlingen, TX 98,581 38,593 6.7% 10.6% 18.5% 42.7% 19.3% 44.5%

Cape Coral-Fort Myers, FL 238,495 225,495 0.2% 0.8% 19.4% 54.0% 18.9% 52.4%

New Haven-Milford, CT 213,870 18,397 1.6% 6.0% 18.7% 68.6%

18.6% 68.

Palm Bay-Melbourne-Titusville, FL 190,403 48,301 1.7% 3.7% 18.2% 57.1% 18.6% 58.4%

Wilmington, NC 93,347 19,489 4.1% 13.3% 18.1% 62.9% 18.2% 62.8%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC

156,898 40,099 2.7% 10.0% 18.6% 63.4% 18.2% 61.8%

Hilton Head Island-Bluffton, SC 72,147 55,836 1.8% 5.6% 17.8% 52.3% 17.9% 52.6%

Houston-The Woodlands-Sugar Land, TX 1,649,856 321,730 1.1% 2.3% 17.5% 42.2% 17.6% 42.8%

North Port-Sarasota-Bradenton, FL 263,971 158,183 2.0% 5.5% 17.0% 58.2% 17.3% 59.1%

Punta Gorda, FL 70,708 70,308 0.0% 0.4% 17.0% 48.2% 17.2% 48.5%

Pensacola-Ferry Pass-Brent, FL 151,893 15,199 3.9% 11.2% 15.1% 55.1% 16.9% 62.3%

Tampa-St. Petersburg-Clearwater, FL 849,853 285,451 1.3% 3.5% 16.0% 53.6% 16.5% 55.1%

Beaumont-Port Arthur, TX 120,529 105,425 0.0% 0.2% 16.5% 36.6% 16.4% 36.4%

Daphne-Fairhope-Foley, AL 70,318 14,404 0.7% 1.4% 15.4% 55.3% 16.1% 58.5%

Jacksonville, NC 53,490 6,799 5.9% 10.3% 16.6% 51.8% 15.5% 47.4%

Portland-South Portland, ME 193,001 5,804 3.6% 14.5% 15.5% 65.1% 15.2% 63.7%

Gainesville, FL 75,798 1,700 0.0% 0.0% 15.6% 43.8% 15.1% 42.9%

Gulfport-Biloxi, MS 130,147 92,462 0.2% 0.3% 14.9% 44.6% 14.9% 44.6%

Mobile, AL 139,996 29,399 3.1% 9.5% 14.8% 44.0% 14.6% 43.4%

Homosassa Springs, FL 55,727 11,906 0.8% 0.8% 14.3% 43.5% 14.2% 43.2%

Tallahassee, FL 100,908 8,292 0.0% 2.4% 12.1% 33.3% 12.2% 33.6%

New Orleans-Metairie, LA 371,993 359,590 0.0% 0.2% 12.1% 27.8% 11.4% 26.1%

Hammond, LA 34,396 15,198 5.3% 8.6% 12.8% 31.8% 11.1% 29.1%

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD

1,830,424 67,

497 7.0% 15.6% 6.0% 33.2% 6.5% 32.7%

Kingston, NY 61,429 100 0.0% 100.0% 27.1% 749.8% 4.7% 86.1%

Greenville, NC 44,156 400 0.0% 50.0% 0.2% 1.7% 0.3% 1.9%

Total All MSAs 74,520,514 4,420,220 2.0% 4.9% 21.0% 65.0% 20.9% 64.5%

Total Outside MSAs 16,212,126 265,317 2.2% 5.1% 28.9% 86.8% 28.5% 86.0%

Total 90,732,640 4,685,537 2.0% 5.0% 21.4% 66.1% 21.3% 65.7%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Column (3) = Exhibit 10 Column (4)

3. Column (4) = (Exhibit 11 Column (4) – Exhibit 10 Column (4)) / Exhibit 10 Column (4)

4. Column (5) = (Exhibit 12 Column (4) – Exhibit 10 Column (4)) / Exhibit 10 Column (4)

5. Column (6) = (Exhibit 11 Column (8) – Exhibit 10 Column (8)) / Exhibit 10 Column (8). Insured and Uninsured Losses are based on estimates of NFIP take-up rates and coverages.

6. Column (7) = (Exhibit 12 Column (8) – Exhibit 10 Column (8)) / Exhibit 10 Column (8). Insured and Uninsured Losses are based on estimates of NFIP take-up rates and coverages.

7. Column (8) = (Exhibit 11 Column (9) – Exhibit 10 Column (9)) / Exhibit 10 Column (9)

8. Column (9) = (Exhibit 12 Column (9) – Exhibit 10 Column (9)) / Exhibit 10 Column (9)

Exhibit 10: Storm Surge Losses by MSA - Current Sea Levels

(1) (2) (3

) (4) (5) (6) (7) (8) (9) (10)

Percent of Single Family Single Family Average Annual Average Annual Total Annual Total Annual Total Annual Percent of

Metropolitan Single Family Residences Exposed Residences Exposed Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge

Statistical Area Name Residences to Storm Surge to Storm Surge Insured Losses Uninsured Losses Insured Losses Uninsured Losses Losses Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 3) (Note 4) (Note 4) = (4) * (5) = (4) * (6) = (7) + (8) = (8) / (9)

New Orleans-Metairie, LA 371,993 96.7% 359,590 $183 $533 $65,676,785 $191,826,500 $257,503,284 74%

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 37.0% 477,787 126 369 59,975,377 176,068,959 236,044,336 75%

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 10.4% 340,686 127 479 43,196,373 163,172,697 206,369,069 79%

Cape Coral-Fort Myers, FL 238,495 94.5% 225,495 284 563 63,962,263 126,892,431 190,854,694 66%

Hilton Head Island-Bluffton, SC 72,147 77.4% 55,836 291 2,272 16,249,038 126,876,180 143,125,219 89%

Tampa-St. Petersburg-Clearwater, FL 849,853 33.6% 285,451 146 248 41,645,688 70,824,114 112,469,802 63%

Houston-The Woodlands-Sugar Land, TX 1,649,856 19.5% 321,730 29 300 9,351,423 96,477,511 105,828,935 91%

Naples-Marco Island, FL 95,685 98.7% 94,485 362 662 34,168,067 62,573,515 96,741,581 65%

Punta Gorda, FL 70,708 99.4% 70,308 352 616 24,735,052 43,287,792 68,022,843 64%

Charleston-North Charleston, SC 212,033 48.3% 102,416 195 395 19,920,847 40,405,905 60,326,752 67%

Jacksonville, FL 429,019 23.6% 101,404 118 450 11,932,615 45,583,548 57,516,163 79%

Houma-Thibodaux, LA 60,417 100.0% 60,417 257 667 15,553,261 40,322,766 55,876,027 72%

North Port-Sarasota-Bradenton, FL 263,971 59.9% 158,183 120 225 18,979,822 35,590,809 54,570,631 65%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 61.7% 323,017 35 121 11,422,922 39,012,630 50,435,552 77%

Ocean City, NJ 75,589 85.1% 64,291 195 517 12,510,098 33,221,645 45,731,744 73%

Boston-Cambridge-Newton, MA-NH 1,042,909 3.5% 36,597 199 971 7,271,034 35,545,426 42,816,460 83%

Beaumont-Port Arthur, TX 120,529 87.5% 105,425 36 364 3,747,084 38,328,495 42,075,579 91%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 156,898 25.6% 40,099 230 474 9,207,030 18,995,587 28,202,616 67%

Lafayette, LA 141,177 85.0% 120,065 83 131 9,950,321 15,732,642 25,682,963 61%

Sebastian-Vero Beach, FL 52,438 17.0% 8,906 171 2,500 1,527,440 22,268,952 23,796,391 94%

Jacksonville, NC 53,490 12.7% 6,799 1,034 2,279 7,029,454 15,491,426 22,520,880 69%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 34.8% 76,920 70 208 5,372,068 15,963,444 21,335,512 75%

Lake Charles, LA 62,822 95.1% 59,721 108 227 6,451,543 13,574,303 20,025,846 68%

Portland-South Portland, ME 193,001 3.0% 5,804 209 3,132 1,214,294 18,176,396 19,390,690 94%

Palm Bay-Melbourne-Titusville, FL 190,403 25.4% 48,301 38 361 1,852,033 17,430,437 19,282,470 90%

Bridgeport-Stamford-Norwalk, CT 238,970 5.3% 12,598 263 1,119 3,318,801 14,097,575 17,416,377 81%

Wilmington, NC 93,347 20.9% 19,489 170 708 3,307,824 13,796,429 17,104,253 81%

Gulfport-Biloxi, MS 130,147 71.0% 92,462 30 149 2,729,900 13,803,842 16,533,742 83%

New Haven-Milford, CT 213,870 8.6% 18,397 127 730 2,338,065 13,437,334 15,775,399 85%

Salisbury, MD-DE 161,632

37.8% 61,

074 96 130 5,880,899 7,930,213 13,811,112 57%

Port St. Lucie, FL 147,959 13.6% 20,094 208 421 4,185,044 8,453,749 12,638,794 67%

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD 1,830,424 3.7% 67,497 41 141 2,763,814 9,535,125 12,298,939 78%

Baton Rouge, LA 241,791 37.3% 90,134 51 82 4,575,227 7,375,051 11,950,278 62%

Homosassa Springs, FL 55,727 21.4% 11,906 337 618 4,010,747 7,362,398 11,373,145 65%

Barnstable Town, MA 138,885 18.4% 25,497 32 390 816,343 9,956,553 10,772,895 92%

Savannah, GA 109,497 72.1% 78,898 19 117 1,526,499 9,223,203 10,749,702 86%

Mobile, AL 139,996 21.0% 29,399 54 304 1,585,854 8,937,413 10,523,266 85%

Daphne-Fairhope-Foley, AL 70,318 20.5% 14,404 80 646 1,149,824 9,301,685 10,451,509 89%

Brunswick, GA 37,880 76.0% 28,785 33 223 964,260 6,416,524 7,380,784 87%

Crestview-Fort Walton Beach-Destin, FL 92,891 25.4% 23,598 68 196 1,612,384 4,632,408 6,244,792 74%

Hartford-East Hartford-Middletown, CT 332,941 2.0% 6,601 138 708 910,798 4,674,313 5,585,111 84%

Corpus Christi, TX 121,799 28.8% 35,100 13 114 460,554 4,009,256 4,469,810 90%

Washington-Arlington-Alexandria, DC-VA-MD-WV 1,422,224 0.5% 6,899 20 594 140,188 4,097,234 4,237,422 97%

Atlantic City-Hammonton, NJ 84,496 35.6% 30,099 56 82 1,688,498 2,477,984 4,166,482 59%

New Bern, NC 41,778 50.7% 21,189 45 139 954,171 2,953,738 3,907,908 76%

Pensacola-Ferry Pass-Brent, FL 151,893 10.0% 15,199 77 167 1,173,628 2,543,037 3,716,665 68%

Brownsville-Harlingen, TX 98,581 39.1% 38,593 9 85 356,609 3,282,949 3,639,559 90%

Norwich-New London, CT 85,235 10.0% 8,503 89 336 755,231 2,860,894 3,616,125 79%

Panama City, FL 55,465 29.4% 16,290 33 122 543,940 1,993,770 2,537,710 79%

Tallahassee, FL 100,908 8.2% 8,292 156 148 1,296,648 1,231,196 2,527,844 49%

Baltimore-Columbia-Towson, MD 854,053 3.9% 33,298 14 52 468,141 1,725,498 2,193,639 79%

Hammond, LA 34,396 44.2% 15,198 46 74 696,574 1,130,015 1,826,589 62%

Providence-Warwick, RI-MA 404,392 4.6% 18,704 20 74 374,995 1,377,582 1,752,578 79%

Gainesville, FL 75,798 2.2% 1,700 244 359 414,699 611,008 1,025,707 60%

Dover, DE 52,288 4.6% 2,399 60 284 144,956 681,484 826,439 82%

California-Lexington Park, MD 35,784 12.8% 4,598 83 45 380,407 204,718 585,125 35%

Hinesville, GA 19,678 23.4% 4,595 2 117 7,194 537,063 544,257 99%

Vineland-Bridgeton, NJ 40,871 14.9% 6,096 62 26 380,331 161,218 541,549 30%

Bangor, ME 50,205 0.4% 200 0 459 0 91,787 91,787 100%

Orlando-Kissimmee-Sanford, FL 649,296 0.1% 400 187 9 74,856 3,720 78,576 5%

Richmond, VA 390,291 0.5% 1,800 6 24 11,138 43,070 54,208 79%

Greenville, NC 44,156 0.9% 400 8 72 3,081 28,891 31,972 90%

Kingston, NY 61,429 0.2% 100 207 10

20,685 1,

034 21,719 5%

Total All MSAs 74,570,040 5.9% 4,420,220 $126 $381 $554,924,738 $1,684,625,068 $2,239,549,806 75%

Total Outside MSAs 16,212,126 1.6% 265,317 173 343 45,920,279 90,934,090 136,854,369 66%

Total 90,782,166 5.2% 4,685,537 128 379 600,845,017 1,775,559,158 2,376,404,175 75%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Any given location is deemed to have storm surge exposure if its combined inland flood and storm surge ground up loss is greater than its inland flood loss alone.

3. Column (4) = Column (2) * Column (3).

4. Loss data is ground up storm surge losses sourced KatRisk catastrophe model runs on a subset of Milliman market basket locations. Modeling runs were set to a standard sea level rise scenario

5. Insured and Uninsured losses are based on estimates of NFIP take-up rates and coverages.

Exhibit 11: Storm Surge Losses by MSA - Medium Sea Level Rise Scenario

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Percent of Single Family Single Family Average Annual Average Annual Total Annual Total Annual Total Annual Percent of

Metropolitan Single Family Residences Exposed Residences Exposed Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge

Statistical Area Name Residences to Storm Surge to Storm Surge Insured Losses Uninsured Losses Insured Losses Uninsured Losses Losses Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 3) (Note 4) (Note 4) = (4) * (5) = (4) * (6) = (7) + (8) = (8) / (9)

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 38.1% 491,987 $148 $438 $72,702,425 $215,446,576 $288,149,001 75%

New Orleans-Metairie, LA 371,993 96.7% 359,690 200 598 71,828,224 214,964,793 286,793,018 75%

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 10.7% 348,786 160 594 55,640,594 207,057,408 262,698,002 79%

Cape Coral-Fort Myers, FL 238,495 94.8% 225,995 334 670 75,493,291 151,507,272 227,000,563 67%

Hilton Head Island-Bluffton, SC 72,147 78.8% 56,837 339 2,629 19,264,725 149,441,373 168,706,098 89%

Tampa-St. Petersburg-Clearwater, FL 849,853 34.0% 289,050 169 284 48,821,605 82,163,743 130,985,348 63%

Houston-The Woodlands-Sugar Land, TX 1,649,856 19.7% 325,231 34 348 11,109,222 113,340,195 124,449,417 91%

Naples-Marco Island, FL 95,685 98.7% 94,485 428 802 40,457,185 75,786,850 116,244,035 65%

Punta Gorda, FL 70,708 99.4% 70,308 413 720 29,036,338 50,654,107 79,690,445 64%

Charleston-North Charleston, SC 212,033 49.2% 104,216 242 504 25,170,034 52,554,982 77,725,017 68%

Houma-Thibodaux, LA 60,417 100.0% 60,417 342 893 20,680,598 53,954,652 74,635,250 72%

Jacksonville, FL 429,019 24.5% 105,005 146 550 15,354,340 57,722,069 73,076,409 79%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 63.7% 333,517 46 153 15,177,416 50,876,659 66,054,075 77%

Ocean City, NJ 75,589 87.3% 65,990 270 720 17,804,339 47,525,042 65,329,381 73%

North Port-Sarasota-Bradenton, FL 263,971 61.1% 161,282 139 258 22,391,853 41,627,904 64,019,757 65%

Boston-Cambridge-Newton, MA-NH 1,042,909 3.7% 38,297 222 1,114 8,484,175 42,666,416 51,150,591 83%

Beaumont-Port Arthur, TX 120,529 87.5% 105,425 41 424 4,328,433 44,657,069 48,985,502 91%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 156,898 26.3% 41,199 263 547 10,823,462 22,521,384 33,344,846 68%

Lafayette, LA 141,177 86.3% 121,766 101 162 12,350,384 19,767,775 32,118,159 62%

Sebastian-Vero Beach, FL 52,438 17.7% 9,307 195 2,962 1,814,158 27,563,855 29,378,012 94%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 35.0% 77,320 85 252 6,594,446 19,463,547 26,057,992 75%

Jacksonville, NC 53,490 13.5% 7,199 1,104 2,508 7,947,629 18,055,536 26,003,165 69%

Lake Charles, LA 62,822 95.1% 59,721 132 278 7,854,002 16,587,823 24,441,825 68%

Palm Bay-Melbourne-Titusville, FL 190,403 25.8% 49,101 46 420 2,253,864 20,610,376 22,864,239 90%

Portland-South Portland, ME 193,001 3.1% 6,015 222 3,490 1,335,354 20,993,964 22,329,318 94%

Bridgeport-Stamford-Norwalk, CT 238,970 5.5% 13,098 297 1,307 3,896,463 17,115,216 21,011,679 81%

Salisbury, MD-DE 161,632 39.1% 63,173 136 185 8,618,012 11,656,090 20,274,103 57%

Wilmington, NC 93,347 21.7% 20,288 194 803 3,931,373 16,293,081 20,224,454 81%

Gulfport-Biloxi, MS 130,147 71.2% 92,662 34 171 3,141,877 15,862,710 19,004,587 83%

New Haven-Milford, CT 213,870

8.7% 18,

697 148 853 2,761,143 15,949,430 18,710,572 85%

Port St. Lucie, FL 147,959 14.8% 21,894 226 470 4,949,942 10,298,150 15,248,093 68%

Baton Rouge, LA 241,791 37.8% 91,434 61 96 5,574,135 8,814,787 14,388,921 61%

Savannah, GA 109,497 73.7% 80,698 24 146 1,973,098 11,810,547 13,783,645 86%

Barnstable Town, MA 138,885 19.0% 26,397 40 459 1,046,991 12,104,007 13,150,998 92%

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD 1,830,424 3.9% 72,197 41 140 2,984,847 10,109,507 13,094,354 77%

Homosassa Springs, FL 55,727 21.5% 12,006 381 701 4,575,794 8,413,449 12,989,243 65%

Daphne-Fairhope-Foley, AL 70,318 20.6% 14,504 97 740 1,404,984 10,731,875 12,136,859 88%

Mobile, AL 139,996 21.6% 30,299 59 338 1,802,121 10,256,092 12,058,212 85%

Brunswick, GA 37,880 76.3% 28,885 44 293 1,269,146 8,471,170 9,740,316 87%

Crestview-Fort Walton Beach-Destin, FL 92,891 26.9% 24,998 77 221 1,932,043 5,517,300 7,449,343 74%

Hartford-East Hartford-Middletown, CT 332,941 2.0% 6,701 160 847 1,072,402 5,674,936 6,747,338 84%

Corpus Christi, TX 121,799 31.0% 37,800 19 157 715,724 5,946,838 6,662,562 89%

Atlantic City-Hammonton, NJ 84,496 36.0% 30,399 87 120 2,638,428 3,635,082 6,273,510 58%

Washington-Arlington-Alexandria, DC-VA-MD-WV 1,422,224 0.5% 7,299 22 753 159,747 5,498,451 5,658,198 97%

New Bern, NC 41,778 51.9% 21,689 59 174 1,270,608 3,784,012 5,054,620 75%

Norwich-New London, CT 85,235 10.3% 8,804 105 401 926,024 3,531,602 4,457,626 79%

Pensacola-Ferry Pass-Brent, FL 151,893 10.4% 15,799 90 185 1,418,732 2,926,782 4,345,514 67%

Brownsville-Harlingen, TX 98,581 41.8% 41,192 11 94 448,986 3,891,378 4,340,364 90%

Panama City, FL 55,465 31.5% 17,489 38 138 659,057 2,416,631 3,075,688 79%

Baltimore-Columbia-Towson, MD 854,053 4.1% 34,798 18 67 628,790 2,324,643 2,953,433 79%

Tallahassee, FL 100,908 8.2% 8,292 175 166 1,454,837 1,380,672 2,835,509 49%

Providence-Warwick, RI-MA 404,392 4.8% 19,404 25 87 477,164 1,685,350 2,162,514 78%

Hammond, LA 34,396 46.5% 15,998 47 80 754,155 1,274,969 2,029,124 63%

Gainesville, FL 75,798 2.2% 1,700 279 415 474,665 706,190 1,180,856 60%

Dover, DE 52,288 4.8% 2,499 75 348 188,400 869,641 1,058,041 82%

California-Lexington Park, MD 35,784 12.8% 4,598 120 54 551,172 249,565 800,737 31%

Vineland-Bridgeton, NJ 40,871 15.4% 6,296 84 29 528,593 185,158 713,751 26%

Hinesville, GA 19,678 23.4% 4,595 2 144 9,169 662,868 672,037 99%

Orlando-Kissimmee-Sanford, FL 649,296 0.1% 500 222 11 111,198 5,465 116,663 5%

Bangor, ME 50,205 0.4% 200 0 459 0 91,820 91,820 100%

Richmond, VA 390,291 0.5% 2,000 6 27 12,167 54,646 66,812 82%

Greenville, NC 44,156 0.9% 400 8 72 3,116 28,949 32,065 90%

Kingston, NY 61,429 0.2% 100 214 13

21,429 1,

315 22,744 6%

Total All MSAs 74,520,514 6.0% 4,507,928 $148 $452 $669,104,628 $2,037,741,745 $2,706,846,373 75%

Total Outside MSAs 16,212,126 1.7% 271,117 217 432 58,700,794 117,183,666 175,884,460 67%

Total 90,732,640 5.3% 4,779,045 152 451 727,805,422 2,154,925,411 2,882,730,833 75%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Any given location is deemed to have storm surge exposure if its combined inland flood and storm surge ground up loss is greater than its inland flood loss alone.

3. Column (4) = Column (2) * Column (3).

4. Loss data is ground up storm surge losses sourced KatRisk catastrophe model runs on a subset of Milliman market basket locations. Modeling runs were set to a medium sea level rise scenario.

5. Insured and Uninsured losses are based on estimates of NFIP take-up rates and coverages.

Exhibit 12: Storm Surge Losses by MSA - High Sea Level Rise Scenario

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Percent of Single Family Single Family Average Annual Average Annual Total Annual Total Annual Total Annual Percent of

Metropolitan Single Family Residences Exposed Residences Exposed Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge Storm Surge

Statistical Area Name Residences to Storm Surge to Storm Surge Insured Losses Uninsured Losses Insured Losses Uninsured Losses Losses Losses Uninsured

(Note 1) (Note 1) (Note 2) (Note 3) (Note 4) (Note 4) = (4) * (5) = (4) * (6) = (7) + (8) = (8) / (9)

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 11.2% 364,585 $246 $901 $89,611,876 $328,321,022 $417,932,899 79%

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 39.8% 513,786 189 571 97,226,532 293,214,517 390,441,049 75%

New Orleans-Metairie, LA 371,993 96.8% 360,190 221 681 79,649,017 245,125,855 324,774,871 75%

Cape Coral-Fort Myers, FL 238,495 95.3% 227,395 420 859 95,507,309 195,442,677 290,949,985 67%

Hilton Head Island-Bluffton, SC 72,147 81.7% 58,938 427 3,279 25,181,584 193,249,254 218,430,838 88%

Tampa-St. Petersburg-Clearwater, FL 849,853 34.8% 295,449 222 368 65,648,641 108,753,552 174,402,192 62%

Naples-Marco Island, FL 95,685 98.9% 94,585 546 1,059 51,602,207 100,171,782 151,773,988 66%

Houston-The Woodlands-Sugar Land, TX 1,649,856 20.0% 329,231 42 417 13,922,089 137,221,066 151,143,155 91%

Charleston-North Charleston, SC 212,033 50.0% 106,017 343 751 36,345,891 79,571,570 115,917,461 69%

Ocean City, NJ 75,589 89.8% 67,890 453 1,235 30,787,988 83,853,367 114,641,355 73%

Jacksonville, FL 429,019 26.0% 111,505 207 768 23,096,487 85,584,729 108,681,216 79%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 65.5% 342,518 71 232 24,426,584 79,325,266 103,751,850 76%

Punta Gorda, FL 70,708 99.9% 70,608 522 909 36,883,999 64,153,239 101,037,238 63%

Houma-Thibodaux, LA 60,417 100.0% 60,417 451 1,187 27,250,881 71,739,610 98,990,491 72%

North Port-Sarasota-Bradenton, FL 263,971 63.2% 166,882 183 337 30,522,010 56,296,798 86,818,808 65%

Boston-Cambridge-Newton, MA-NH 1,042,909 4.0% 41,596 282 1,531 11,737,707 63,692,170 75,429,877 84%

Beaumont-Port Arthur, TX 120,529 87.6% 105,625 48 496 5,040,445 52,342,089 57,382,534 91%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 156,898 28.1% 44,099 331 704 14,589,288 31,031,278 45,620,566 68%

Sebastian-Vero Beach, FL 52,438 18.9% 9,907 244 4,055 2,413,047 40,173,651 42,586,698 94%

Lafayette, LA 141,177 87.2% 123,167 121 195 14,845,505 24,029,604 38,875,109 62%

Salisbury, MD-DE 161,632 40.5% 65,472 243 336 15,936,859 22,019,817 37,956,676 58%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 35.4% 78,220 120 349 9,387,808 27,288,947 36,676,755 74%

Jacksonville, NC 53,490 14.0% 7,499 1,291 3,136 9,683,576 23,512,590 33,196,166 71%

Portland-South Portland, ME 193,001 3.4% 6,648 259 4,514 1,724,381 30,009,777 31,734,158 95%

Bridgeport-Stamford-Norwalk, CT 238,970 5.6% 13,298 415 1,958 5,512,782 26,038,707 31,551,489 83%

Palm Bay-Melbourne-Titusville, FL 190,403 26.3% 50,101 63 547 3,164,806 27,381,662 30,546,468 90%

Lake Charles, LA 62,822 95.1% 59,721 160 340 9,572,733 20,312,648 29,885,382 68%

Wilmington, NC 93,347 23.7% 22,087 243 1,017 5,375,201 22,472,474 27,847,675 81%

New Haven-Milford, CT 213,870 9.1% 19,497 201 1,162 3,920,461 22,655,347 26,575,808 85%

Gulfport-Biloxi, MS 130,147

71.3% 92,

762 43 215 3,959,114 19,954,365 23,913,479 83%

Port St. Lucie, FL 147,959 17.5% 25,893 256 575 6,624,452 14,891,632 21,516,085 69%

Savannah, GA 109,497 76.2% 83,398 35 210 2,942,412 17,510,302 20,452,715 86%

Barnstable Town, MA 138,885 20.0% 27,797 60 648 1,659,821 18,026,162 19,685,983 92%

Baton Rouge, LA 241,791 38.6% 93,335 77 119 7,185,651 11,084,038 18,269,689 61%

Daphne-Fairhope-Foley, AL 70,318 20.8% 14,604 145 989 2,112,344 14,448,919 16,561,262 87%

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD 1,830,424 4.3% 77,997 47 163 3,627,771 12,697,447 16,325,218 78%

Homosassa Springs, FL 55,727 21.5% 12,006 477 880 5,725,726 10,565,862 16,291,588 65%

Mobile, AL 139,996 23.0% 32,199 69 400 2,219,603 12,867,893 15,087,496 85%

Brunswick, GA 37,880 77.3% 29,285 65 439 1,915,828 12,863,185 14,779,014 87%

Atlantic City-Hammonton, NJ 84,496 36.7% 30,999 176 231 5,460,689 7,170,061 12,630,750 57%

Crestview-Fort Walton Beach-Destin, FL 92,891 28.6% 26,597 105 300 2,793,703 7,976,690 10,770,393 74%

Hartford-East Hartford-Middletown, CT 332,941 2.0% 6,801 220 1,240 1,498,861 8,433,652 9,932,512 85%

Washington-Arlington-Alexandria, DC-VA-MD-WV 1,422,224 0.5% 7,699 32 1,209 245,810 9,310,125 9,555,934 97%

Corpus Christi, TX 121,799 32.3% 39,400 27 213 1,061,635 8,378,563 9,440,198 89%

New Bern, NC 41,778 55.5% 23,188 86 251 1,987,483 5,825,496 7,812,980 75%

Norwich-New London, CT 85,235 11.2% 9,504 147 589 1,398,681 5,593,205 6,991,886 80%

Pensacola-Ferry Pass-Brent, FL 151,893 11.1% 16,899 124 233 2,088,794 3,944,828 6,033,622 65%

Baltimore-Columbia-Towson, MD 854,053 4.5% 38,398 29 113 1,126,757 4,334,489 5,461,246 79%

Brownsville-Harlingen, TX 98,581 43.3% 42,692 13 110 574,315 4,686,010 5,260,325 89%

Panama City, FL 55,465 35.5% 19,688 50 187 991,476 3,681,251 4,672,727 79%

Providence-Warwick, RI-MA 404,392 5.2% 21,205 38 127 811,115 2,699,086 3,510,201 77%

Tallahassee, FL 100,908 8.4% 8,492 204 193 1,736,592 1,640,948 3,377,540 49%

Hammond, LA 34,396 48.0% 16,498 53 90 868,259 1,489,468 2,357,727 63%

Dover, DE 52,288 5.2% 2,699 105 473 283,603 1,277,009 1,560,612 82%

California-Lexington Park, MD 35,784 14.0% 4,998 219 80 1,096,129 397,474 1,493,603 27%

Gainesville, FL 75,798 2.2% 1,700 345 517 586,833 878,405 1,465,237 60%

Vineland-Bridgeton, NJ 40,871 15.6% 6,395 137 39 874,974 247,988 1,122,962 22%

Hinesville, GA 19,678 23.9% 4,695 3 193 13,665 908,118 921,782 99%

Orlando-Kissimmee-Sanford, FL 649,296 0.1% 600 322 13 193,265 8,051 201,316 4%

Richmond, VA 390,291 0.6% 2,200 7 36 15,641 78,761 94,402 83%

Bangor, ME 50,205 0.4% 200 0 460 0 91,949 91,949 100%

Kingston, NY 61,429 0.3% 200 158 44 31,639 8,791 40,430 22%

Greenville, NC 44,156 1.4% 599 5 49

3,199 29,

395 32,594 90%

Total All MSAs 74,520,514 6.2% 4,638,556 $195 $599 $904,283,533 $2,778,984,685 $3,683,268,218 75%

Total Outside MSAs 16,212,126 1.7% 278,918 304 609 84,708,898 169,861,279 254,570,177 67%

Total 90,732,640 5.4% 4,917,473 201 600 988,992,430 2,948,845,964 3,937,838,395 75%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Any given location is deemed to have storm surge exposure if its combined inland flood and storm surge ground up loss is greater than its inland flood loss alone.

3. Column (4) = Column (2) * Column (3).

4. Loss data is ground up storm surge losses sourced KatRisk catastrophe model runs on a subset of Milliman market basket locations. Modeling runs were set to a high sea level rise scenario.

5. Insured and Uninsured losses are based on estimates of NFIP take-up rates and coverages.

Exhibit 13: Ratio of Average Annual Loss to Census Block Group Median Household Income – Averaged by MSA

Exhibit 14: Change in 500 Year Return Period Flood Losses for Sea Level Rise Scenarios Compared to Current Sea Levels

(1) (2) (3) (4) (5) (6) (7)

500 Year Return Period Event 500 Year Return Period Event

500 Year Return Period Event Total Losses Percent Total Loss Increase

Metropolitan Single Family Current Medium High Medium Sea Level Rise High Sea Level Rise

Statistical Area Name Residences Sea Level Sea Level Rise Sea Level Rise vs. Current Sea Level vs. Current Sea Level

(Note 1) (Note 1) (Note 2) (Note 2) (Note 2) = ((4) - (3)) / (3) = ((5) - (3)) / (3)

Atlantic City-Hammonton, NJ 84,496 $362,430,776 $517,973,620 $1,022,388,888 42.9% 182.1%

Ocean City, NJ 75,589 2,640,784,247 3,434,388,722 5,161,489,799 30.1% 95.5%

Salisbury, MD-DE 161,632 971,553,505 1,256,196,402 1,905,236,917 29.3% 96.1%

Savannah, GA 109,497 1,042,258,575 1,342,122,747 1,921,665,551 28.8% 84.4%

Brunswick, GA 37,880 646,274,261 829,589,873 1,191,184,136 28.4% 84.3%

Charleston-North Charleston, SC 212,033 4,185,234,641 5,133,704,055 7,019,067,066 22.7% 67.7%

New Bern, NC 41,778 331,955,342 402,342,741 558,266,630 21.2% 68.2%

Virginia Beach-Norfolk-Newport News, VA-NC 523,327 2,646,227,542 3,199,814,645 4,468,560,601 20.9% 68.9%

Lake Charles, LA 62,822 2,131,971,128 2,575,623,077 3,048,489,169 20.8% 43.0%

Lafayette, LA 141,177 2,039,553,711 2,435,222,403 2,836,245,454 19.4% 39.1%

Corpus Christi, TX 121,799 645,844,971 769,941,306 906,521,476 19.2% 40.4%

Hinesville, GA 19,678 86,598,049 102,660,873 132,150,416 18.5% 52.6%

Houma-Thibodaux, LA 60,417 3,427,551,097 4,054,513,220 4,694,094,855 18.3% 37.0%

Barnstable Town, MA 138,885 645,013,230 747,408,192 1,046,199,745 15.9% 62.2%

Norwich-New London, CT 85,235 413,603,226 474,407,657 658,782,871 14.7% 59.3%

Naples-Marco Island, FL 95,685 8,418,978,839 9,652,428,169 11,700,316,167 14.7% 39.0%

Miami-Fort Lauderdale-Pompano Beach, FL 1,292,465 20,340,408,122 23,113,537,953 28,106,226,804 13.6% 38.2%

Jacksonville, FL 429,019 3,804,710,357 4,315,520,562 5,303,511,334 13.4% 39.4%

Myrtle Beach-Conway-North Myrtle Beach, SC-NC 156,898 2,268,212,275 2,569,841,942 3,245,084,009 13.3% 43.1%

Gulfport-Biloxi, MS 130,147 1,645,325,078 1,844,341,416 2,196,462,593 12.1% 33.5%

Vineland-Bridgeton, NJ 40,871 56,962,003 63,763,884 77,758,669 11.9% 36.5%

Crestview-Fort Walton Beach-Destin, FL 92,891 622,399,371 693,838,989 881,080,867 11.5% 41.6%

Tampa-St. Petersburg-Clearwater, FL 849,853 10,552,205,581 11,726,579,381 14,170,266,275 11.1% 34.3%

Deltona-Daytona Beach-Ormond Beach, FL 220,757 2,128,314,643 2,363,626,803 2,846,803,715 11.1% 33.8%

Beaumont-Port Arthur, TX 120,529 4,136,931,309 4,591,625,309 5,107,839,059 11.0% 23.5%

North Port-Sarasota-Bradenton, FL 263,971 4,892,636,151 5,425,892,573 6,616,914,002 10.9% 35.2%

Cape Coral-Fort Myers, FL 238,495 17,578,399,236 19,488,597,567 22,350,598,971 10.9% 27.1%

Panama City, FL 55,465 350,669,368 388,420,836 493,870,890 10.8% 40.8%

Hilton Head Island-Bluffton, SC 72,147 6,351,094,038 7,034,450,781 8,177,827,700 10.8% 28.8%

Sebastian-Vero Beach, FL 52,438 1,466,731,259 1,624,349,559 1,899,142,929 10.7% 29.5%

Daphne-Fairhope-Foley, AL 70,318 724,944,865 798,099,880 968,551,673 10.1% 33.6%

Wilmington, NC 93,347 1,461,575,060 1,604,176,624 1,942,715,791 9.8% 32.9%

New Haven-Milford, CT 213,870 1,463,463,262 1,606,085,966 1,921,764,452 9.7% 31.3%

Palm Bay-Melbourne-Titusville, FL 190,403 2,073,766,310 2,263,058,733 2,667,315,832 9.1% 28.6%

Portland-South Portland, ME 193,001 1,381,335,025 1,505,869,025 1,851,182,778 9.0% 34.0%

Baton Rouge, LA 241,791 1,608,658,881 1,749,674,462 1,969,420,766 8.8% 22.4%

Punta Gorda, FL 70,708 6,560,020,807 7,126,167,679 8,008,198,581 8.6% 22.1%

New York-Newark-Jersey City, NY-NJ-PA 3,262,366 12,814,631,896 13,790,870,482 16,466,194,589 7.6% 28.5%

Mobile, AL 139,996 706,914,457 760,477,719 852,008,287 7.6% 20.5%

Boston-Cambridge-Newton, MA-NH 1,042,909 3,175,352,710 3,406,121,712 4,134,800,792 7.3% 30.2%

Homosassa Springs, FL 55,727 1,021,531,705 1,078,550,868 1,177,613,475 5.6%

15.3%

alifornia-Lexington Park, MD 35,784 152,341,838 160,360,516 183,087,226 5.3% 20.2%

Port St. Lucie, FL 147,959 1,507,227,983 1,586,452,485 1,751,038,904 5.3% 16.2%

New Orleans-Metairie, LA 371,993 13,737,164,558 14,454,744,384 15,314,118,953 5.2% 11.5%

Bridgeport-Stamford-Norwalk, CT 238,970 1,952,259,033 2,054,111,847 2,346,150,126 5.2% 20.2%

Jacksonville, NC 53,490 1,026,480,191 1,079,084,217 1,162,936,642 5.1% 13.3%

Houston-The Woodlands-Sugar Land, TX 1,649,856 12,303,369,318 12,901,889,116 13,843,318,052 4.9% 12.5%

Hartford-East Hartford-Middletown, CT 332,941 934,100,657 975,360,784 1,085,135,448 4.4% 16.2%

Pensacola-Ferry Pass-Brent, FL 151,893 743,214,785 773,403,621 855,264,199 4.1% 15.1%

Brownsville-Harlingen, TX 98,581 489,992,331 509,879,804 534,085,364 4.1% 9.0%

Tallahassee, FL 100,908 437,015,592 452,710,654 481,798,780 3.6% 10.2%

Hammond, LA 34,396 243,081,634 248,949,976 259,753,682 2.4% 6.9%

Baltimore-Columbia-Towson, MD 854,053 1,156,427,858 1,175,000,239 1,227,743,826 1.6% 6.2%

Providence-Warwick, RI-MA 404,392 786,868,304 799,017,652 852,565,894 1.5% 8.3%

Dover, DE 52,288 118,657,037 119,729,080 121,670,348 0.9% 2.5%

Gainesville, FL 75,798 357,238,027 358,516,564 359,952,367 0.4% 0.8%

Washington-Arlington-Alexandria, DC-VA-MD-WV 1,422,224 3,391,657,515 3,394,115,057 3,398,066,680 0.1% 0.2%

Philadelphia-Camden-Wilmington, PA-NJ-DE-MD 1,830,424 4,340,819,317 4,342,270,803 4,351,138,771 0.0% 0.2%

Richmond, VA 390,291 781,569,090 781,640,458 782,122,857 0.0% 0.1%

Orlando-Kissimmee-Sanford, FL 649,296 2,808,656,721 2,808,865,318 2,809,292,195 0.0% 0.0%

Kingston, NY 61,429 305,322,894 305,336,995 305,377,933 0.0% 0.0%

Greenville, NC 44,156 218,971,172 218,980,565 218,994,557 0.0% 0.0%

Bangor, ME 50,205 51,483,429 51,483,429 51,483,429 0.0% 0.0%

Notes:

1. MSAs and residential populations are sourced from the 2017 five year American Community Survey, provided by the United States Census Bureau.

2. Total Losses in Columns (3), (4) and (5) are ground up inland flood and storm surge losses sourced from Katrisk model runs simulating an event with a

500 year return period under standard, medium and high sea level rise scenarios, respectively.

5.3 OVERVIEW OF MILLIMAN M-PIRE: MORTGAGE ANALYTIC METHODOLOGY

Milliman’s Mortgage Platform for Investments and Reinsurance (M-PIRe) is an integrated platform hosted on

Microsoft’s Azure cloud and was built by Milliman specifically to evaluate mortgage credit risk transfer securities and

reinsurance exposures. M-PIRe is used by GSE CRT market participants to perform key functions when evaluating

specific deals and holistically managing their portfolio.

M-PIRe’s robust modeling framework and customization flexibility made it an appropriate tool to use for performing

the analyses in this paper. Specific M-PIRe features leveraged for this analysis include the Loan level Mortgage

Performance Model, Loan Level Economic Scenario Engine, and CRT Deal Cash Flow Waterfall Library.

Loan level Mortgage Performance Model

M-PIRe’s Loan Level Mortgage Performance Model combines loan-level data, loan-level econometric models, and

economic scenario forecasts to produce deterministic cash-flow estimates for each loan (and ultimately each CRT

exposure). Specifically, M-PIRe estimates quarterly conditional prepayment, default, and loss severity (also referred

to as “loss given default”) rates for each active loan included in the reference pools underlying the transactions for all

periods beyond the evaluation date. These estimates are generated using various economic scenarios and our

econometric mortgage performance model. The econometric mortgage performance model was developed using

publicly available Fannie Mae and Freddie Mac data and translates the historical mortgage performance data and

economic data into future estimates. After generating the mortgage collateral performance estimates, M-PIRe passes

the aggregated estimates to the cash-flow model to generate CRT cash flows under the various deterministic

economic scenarios.

The model estimates the performance of loans through a binomial logistic regression framework where the risk of

prepayment and default are competing risks. The figure below provides a visual of the main components of the model.

The first stage of the framework, the

Performing Loan Model, estimates the

probability of a loan transitioning from

performing status to either

prepayment or default—where default

is defined as 180 days or more

delinquent. The second stage of the

framework, the Non-Performing Loan

Model, estimates the probability of a loan transitioning from default to disposition (i.e., title transferred to investor

and sold through the foreclosure process or real estate owned inventory). The transition probabilities generate

dynamic estimates that vary according to a loan’s underwriting characteristics, age, and economic influences. For

loans that are estimated to result in a property disposition, the third stage of the framework is a Loss Severity Model

to estimate the ultimate losses that will flow through the CRT structure as a result of the loan default.

Performing Loan Model

The Performing Loan Model estimates the quarterly conditional probability of a loan transitioning from performing

status to either prepayment or default (defined as 180 days or more delinquent). This model was parameterized using

a combination of the Fannie Mae single-family loan performance data and the Freddie Mac single-family loan level

dataset. Both datasets contain mortgage loans that are fixed rate and fully amortizing. The dataset contains

Figure 25: Model Flow

performance results for more than 60 million mortgages acquired by the GSEs. Once expanded to the quarterly panel

dataset, the final estimation dataset contained more than 375 million quarterly observations of loan performance.

Table 10: Performing Loan Model Variable Overview - Dynamic Economic Factors

Below is a high-level description of the predictor

variables in the model and their impact on

prepayment and default probabilities. Note that the

variables are partitioned into two tables to illustrate

the static loan characteristic that drive mortgage

performance versus the dynamic economic factors

that drive mortgage performance.

Section 4 describes the analysis methodology in this

paper where the dynamic economic factors are

altered to correspond to specific flood events. This is

the main channel (HPA variables in Performing, Non-

Performing, and Gross Severity models) for

translating the flood events into M-PIRe and producing estimates of mortgage performance (and ultimately CRT deal

performance) under the events.

Table 11: Performing Loan Model Variable Overview – Static Loan Characteristics

Variable Value Prepayment Default

Cumulative

Home

Price

Appreciation

High HPA

Low HPA

Increase

Decrease

Increase

One-Year

Change in

HPA

High HPA

Low HPA

Increase

Decrease

Increase

Spread relative

to 10-Year

Interest Rate

High Spread

Low Spread

Increase

Decrease

N/A

Unemployment

Rate

High Rate

Low Rate

N/A

Increase

Decrease

Variable

Value

Prepayment

Default

Loan-to-Value Ratio

High Ratio

Low Ratio

Decrease

Increase

Decrease

Credit Score High Score

Low Score

Increase

Decrease

Increase

Debt to Income

Ratio

High Ratio

Low Ratio

N/A

Increase

Decrease

Occupancy Status

Owner

Investor

Second

Reference

Decrease

Reference

Increase

Decrease

Property Type

SFR

Condo

Co-op

Manufactured

PUD

Unknown

Reference

Decrease

Reference

Decrease

Increase

Decrease

Loan Purpose

Purchase

Cash out

Rate/term

Unknown

Reference

Decrease

Increase

Reference

Increase

Term

360

LT360

GT360

Reference

Increase

Decrease

Reference

Decrease

Increase

Original Loan

Amount

High Amount

Low Amount

Reference

Decrease

Reference

Increase

Number of

Borrowers

1 Borrower

2+ Borrowers

Unknown

N/A

Increase

Reference

Increase

Number of Units

N/A

Reference

Non-Performing Loan Model

The Non-Performing Loan Model estimates

the lifetime probability of a loan

transitioning from the current delinquency

status to 180-day delinquency status (e.g.,

30-day to 180-day), and subsequently foreclosure/disposition.

The model estimates the probabilities of each status transition to 180 days delinquency using linear regression. The

linear regression has a single explanatory variable: borrower equity at the time of delinquency. Equity is measured as

initial equity (one minus combined loan-to-value ratio) adjusted for cumulative home price appreciation. As stated

above, HPA is the mechanism for translating the flood events into M-PIRe and producing performance (and ultimately

CRT deal performance) under the event

This roll-rate/transition methodology is a key component used in Section 4 to provide estimates of loans 60+ days

delinquent as a result of the flood event.

Gross Severity Model

The gross severity, or loss given default, model estimates the ground-up loss for a default as a percent of original loan

balance. Note that this model estimates the indicated severity of loss prior to the estimation of recoveries on any MI

policies that may be in place. M-PIRe has a separate component to estimate and net out MI recoveries. The gross

severity model is comprised of three components:

1. Unpaid principal balance at termination

2. Delinquent interest plus disposition expenses

3. Net sale proceeds

Figure 26 provides a

visual of the model

components. In the

illustrative example, the

severity rate estimates

of 25% of the original

loan balance is equal to

the UPB at delinquency

(95%) plus delinquent

interest and disposition

expense (15%) minus net

sale proceeds (85%).

HPA is a key predictor

variable when estimating

the net sale proceeds.

Increase

Decrease

First Time Home

Buyer

Yes

Unknown

Decrease

Reference

Decrease

Reference

Decrease

Spread at

Origination

High Spread

Low Spread

N/A

Increase

Decrease

Figure 26: Loan Level Performance

Loan Level Economic Scenario Engine

M-PIRe includes access to

Moody’s Analytics scenarios

as well as custom user-

defined stress scenarios. M-

PIRe has the ability to map

economic scenarios at the

loan level to specific loans

so as to reflect differences

in economics at the granular

geographic level. This

feature is key to the analysis

described in Section 4 when

mapping property value

shocks as a result of flood

events to specific homes

included in the modeled

CRT transactions.

CRT Deal Cash Flow Waterfall Library

M-PIRe includes a complete library of the cash-flow waterfalls for all credit risk transfer deals from Freddie Mac and

Fannie Mae completed to date. The waterfalls include details on the capital structure, including deal triggers, the

ability to turn on or off optional call features, and historical collateral performance and bond/insurance cash flows.

These waterfalls are crucial to

turning the projected losses

of the underlying collateral

into cash flows to/from CRT

investors/reinsurers. Figure

28 provides a visual of the

structure pay-down on a

hypothetical CRT deal

structure.

Together, these components

of M-PIRe allowed us to run

new flood event scenarios

through the M-PIRe platform

at the loan level to derive

estimates of loan and deal

impacts as a result of the

event.

Figure 27: Loan Level Model Diagram

Figure 28: Forecast Loss by Quarter and Tranche

About The Society of Actuaries

The Society of Actuaries (SOA), formed in 1949, is one of the largest actuarial professional organizations in the world

and is dedicated to serving more than 30,000 actuarial members and the public in the United States, Canada, and

worldwide. In line with the SOA Vision Statement, actuaries act as business leaders who develop and use mathematical

models to measure and manage risk in support of financial security for individuals, organizations, and the public.

The SOA supports actuaries and advances knowledge through research and education. As part of its work, the SOA

seeks to inform public policy development and public understanding through research. The SOA aspires to be a trusted

source of objective, data-driven research and analysis with an actuarial perspective for its members, industry,

policymakers, and the public. This distinct perspective comes from the SOA as an association of actuaries who have a

rigorous formal education and direct experience as practitioners as they perform applied research. The SOA also

welcomes the opportunity to partner with other organizations in our work where appropriate.

The SOA has a history of working with public policymakers and regulators in developing historical experience studies

and projection techniques as well as individual reports on healthcare, retirement, and other topics. The SOA’s research

is intended to aid the work of policymakers and regulators and follow certain core principles:

Objectivity: The SOA’s research informs and provides analysis that can be relied upon by other individuals or

organizations involved in public policy discussions. The SOA does not take advocacy positions or lobby specific policy

proposals.

Quality: The SOA aspires to the highest ethical and quality standards in all of its research and analysis. Our research

process is overseen by experienced actuaries and nonactuaries from a range of industry sectors and organizations. A

rigorous peer-review process ensures the quality and integrity of our work.

Relevance: The SOA provides timely research on public policy issues. Our research advances actuarial knowledge while

providing critical insights on key policy issues, and thereby provides value to stakeholders and decision makers.

Quantification: The SOA leverages the diverse skill sets of actuaries to provide research and findings that are driven

by the best available data and methods. Actuaries use detailed modeling to analyze financial risk and provide distinct

insight and quantification. Further, actuarial standards require transparency and the disclosure of the assumptions

and analytic approach underlying the work.

Society of Actuaries

475 N. Martingale Road, Suite 600

Schaumburg, Illinois 60173

www.SOA.org

Limitations

Data Reliances

In performing this analysis, we relied upon information obtained from other sources. We have not audited or verified

this data and information. If the underlying data or information is inaccurate or incomplete, the results of our analysis

may likewise be inaccurate or incomplete. In that event, the results of our analysis may not be suitable for the intended

purpose.

We performed a limited review of the data used directly in our analysis for reasonableness and consistency. We did

not find material defects in the data. If there are material defects in the data, it is possible that they would be

uncovered by a detailed, systematic review and comparison of the data to search for data values that are questionable

or relationships that are materially inconsistent. Such a detailed review was beyond the scope of our assignment.

Model Reliances

Our analysis is based on the KatRisk catastrophe model. We have reviewed the model output for reasonableness and

consistency. However, no catastrophe model is entirely accurate. To the extent that the model is biased, the resulting

analysis may be biased.