Coastal Flood Risk in the Mortgage Market: Storm Surge Models' Predictions vs. Flood Insurance Maps
CCoastal Flood Risk in the Mortgage Market:Storm Surge Models’ Predictions vs. Flood Insurance Maps ∗ Amine Ouazad † May 2020
Abstract
Prior literature has argued that flood insurance maps may not capture the extent of floodrisk. This paper performs a granular assessment of coastal flood risk in the mortgage marketby using physical simulations of hurricane storm surge heights instead of using FEMA’s floodinsurance maps. Matching neighborhood-level predicted storm surge heights with mortgagefiles suggests that coastal flood risk may be large: originations and securitizations in stormsurge areas have been rising sharply since 2012, while they remain stable when using floodinsurance maps. Every year, more than 50 billion dollars of originations occur in storm surgeareas outside of insurance floodplains. The share of agency mortgages increases in storm surgeareas, yet remains stable in the flood insurance 100-year floodplain. Mortgages in storm surgeareas are more likely to be complex: non-fully amortizing features such as interest-only or ad-justable rates. Households may also be more vulnerable in storm surge areas: median house-hold income is lower, the share of African Americans and Hispanics is substantially higher, theshare of individuals with health coverage is lower. Price-to-rent ratios are declining in stormsurge areas while they are increasing in flood insurance areas. This paper suggests that un-covering future financial flood risk requires scientific models that are independent of the floodinsurance mapping process. ∗ The author thanks the National Oceanic and Atmospheric Administration for access to the SLOSH simulationdata, Drechsler, Itamar, Alexi Savov, and Philipp Schnabl for access to the bank-level data, as well as Matthew E. Kahn,Jesse M. Keenan for comments on early versions of the paper. The usual disclaimers apply. The author acknowledgessupport from the HEC Montreal foundation. † Associate professor of economics, HEC Montreal. Research professorship in Real Estate and Urban Economics,funded by the HEC Foundation. Email: [email protected]. a r X i v : . [ ec on . GN ] J un Introduction
What is the amount of mortgage credit potentially exposed to the risk of catastrophic hurricanestorm surges? Evidence suggests that flood insurance maps published by FEMA may not providean accurate description of risk. New estimates of floodplain boundaries (Wing, Bates, Smith,Sampson, Johnson, Fargione & Morefield 2018) suggest that up to 41 million Americans live withinthe 100-year floodplain, substantially above the number of Americans living within the 100-yearfloodplain of FEMA’s flood insurance maps. As statistics from the Federal Reserve’s Flow of Fundssuggest that Americans owed about 11.2 trillion dollars of mortgage debt in 2019 (Goodman 2020),the exposure of lenders, securitizers, and households may be higher than suggested by flood in-surance maps. Flood risk may cause defaults or prepayments among borrowers, and cause lossesamong lenders and securitizers. Yet, estimating the exposure of the mortgage market is challeng-ing as (i) not all communities participate in FEMA’s National Flood Insurance Program and thusflood risk may not be mapped comprehensively; (ii) in participating communities, the observedfrequency of flooding due to precipitation events, fluvial flooding, or hurricane storm surges maynot match the predicted frequency of FEMA’s National Flood Hazard Layer; (iii) parts of the 100-year floodplain assume protection by a levee; and (iv) publicly available data may not includesimple measures of mortgage structure or performance for loans outside of the conventional loansingle-family market.This paper provides a transparent and replicable assessment of flood risk in the mortgagemarket by matching the numerical simulations of a model of hurricane storm surges with individ-ual mortgage files, household demographics, house prices and rents, and lenders’ balance sheets.Such simulations predict storm surge height at a granular level, an assessment of flood risk thatis independent of communities’ willingness to participate in the flood insurance program. Suchmodel acknowledges the impact of levees, yet does not exclude leveed areas from the simulation.Relying on scientific models rather than on flood insurance maps is key, as flood insurance mapsare the outcome of a political economy process of community participation in the program as wellas of community investment in mitigation efforts. Hence scientific models help us tease out the Pralle (2019) suggests that three quarters of houses damaged during hurricane Harvey were outside of the 100-yearfloodplain; and that half of the buildings in New York City affected by Sandy were outside of the 100-year floodplain.Kousky (2018) provides a discussion of the design of Flood Insurance Rate Maps. Kousky & Kunreuther (2010) arguesthat better flood maps are required in St Louis. A future test of climate adaptation in the mortgage mar-ket is to test the impact of such information disclosures on lenders’ underwriting standards andhousehold demand for mortgage credit.The paper proceeds as follows. Section 2 describes the data sets, their strengths and limitations,and the methodology that matches them to mortgage data. Section 3 then presents the core sixfacts of this paper. Section 4 concludes.
Assessing the vulnerability of real estate assets, mortgage debt, and lenders’ financial statementsto coastal flood risk requires four sources of data. First, data on the impact of hurricane stormsurge heights or sea level rise at a fine-grained geographic level. Second, data on individual mort-gages’ location, loan-to-value or equity, amortization. Third, information on households’ andindividuals’ vulnerability. This includes household income, the type of dwelling (mobile home),minority status, education, and other relevant variables that are correlated with a household’sability to repay the mortgage. Fourth, information on lenders’ vulnerability, by matching mort-gage originations to lenders’ status (bank or non-bank lender), net income, balance sheet. Wedescribe these information sources one by one. Keenan (2019) introduces climate adaptation in asset management (chapter 2) and in funding and financing options(chapter 4). .1 Coastal Flood Risk Neighborhood-Level Hurricane Storm Surge Heights
NOAA’s Sea Lake and Overland Surgefrom Hurricanes (SLOSH) simulates the impact of hurricanes on storm surge heights using thetransport equations and hurricanes’ measures. The initial conditions of such equations are hur-ricanes’ characteristics such as pressure, speed, and track. The transport equations have beenpresent in the literature at least since Ekman’s (1902) seminal work, published in English in Ek-man (1905), which were not initially used for the forecasting of storm surge heights. A practicalapplication to the modeling of storm surge dynamics is presented later in the work of Jelesnianski(1970). In this latter work, the model is tested for three storms of the Atlantic Seaboard on AtlanticCity, NJ: the September 1944 Storm, hurricane Donna, and hurricane Carol. The availability of measured surges using gage records helps in comparing the predictions of the equations with therealizations of storm surge heights. In these three key examples, the model performs well and,when deviating from the observations, tends to underestimate storm surge heights. Jelesniansky& Chen (n.d.) compares measures from 570 tide gage and high water mark observations with themodel’s predictions for a larger set of storms. It reports that the model’s predictions are within ± of observed heights. A recent assessment of model driven forecasts of storm surge heightsis presented in Kalourazi, Siadatmousavi, Yeganeh-Bakhtiary & Jose (2019).Ekman transport equations are at the core of the SLOSH model, which has become a cen-tral tool for NOAA’s National Hurricane Center forecasts of storm surge heights (Glahn, Tay-lor, Kurkowski & Shaffer 2009). Parameters provided by hurricane forecasters lead to finite-difference simulations of the Ekman transport equations using storm position, the radius of max-imum winds, and the pressure difference between the central and the peripheral pressure. Using model-driven predictions of storm surge heights rather than historical observationsof water gage levels is key. There is indeed evidence that hurricane intensity has been increas-ing (Kossin, Knapp, Olander & Velden 2020), consistent with the prediction of numerical modelsthat a warmer world leads to a higher intensity of hurricanes. Hence using historical observationsof storm surge heights may not be a good indication of future storm surge risk. In this paper,we use SLOSH simulations to provide estimates of worst-case scenarios that may materialize as Ekman was a student of the founder of modern meteorology Vilhelm Bjerknes. Historical measures every few hours are provided in the National Hurricane Center’s HURDAT2 data set. ).The MOM simulations are obtained for 35 different basins, from the Penobscot Bay in Maine,down South to the Laguna Madre basin at the border between the United States and Mexico. Suchbasins include the New Orleans basin, as well as basins for Palm Beach and Biscayne Bay, amongothers. Basins cover all parts of the Atlantic coastline and coastal areas of the Gulf of Mexico.When basins overlap, we take the maximum of the MOMs for the two basins. Sea Level Rise Forecasts
Our second source of coastal flood risk is the series of Sea Level Riselayers provided by NOAA. In its “Global Sea Level Rise Scenarios for the United States NationalClimate Assessment,” (Parris, Bromirski, Burkett, Cayan, Culver, Hall, Horton, Knuuti, Moss,Obeysekera et al. 2012) NOAA states that “we have very high confidence ( > There are minor differences between the boundaries of ZCTA5s and the boundaries of postal ZIPs. The current version of this paper does not consider the basins of Puerto Rico, the Virgin Islands, and Hawaii forcomputational reasons.
Flood Insurance Rate Maps: FEMA’s Special Flood Hazard Areas
The third source of informa-tion on coastal flood risk is FEMA’s Flood Insurance Rate Maps. In such FIRM maps, the SpecialFlood Hazard Areas are strict boundaries of the 100-year floodplain, in which the annual proba-bility of a flood is 1%. This paper uses the 2017 National Flood Hazard Layer.Flood Insurance Rate Maps are relevant for the vulnerability of the mortgage market to floodrisk. First, households borrowing using an agency-guaranteed mortgage are required to buy floodinsurance since the 1973 Flood Disaster Protection Act. Second,Yet, such FIRM maps have some significant limitations. First, they provide a strict binaryboundary for the 100-year floodplain, which may give a false sense of security for borrowers inthe immediate vicinity of the external boundary of the 100-year floodplain. In contrast, SLOSHmodels provide smooth predictions of hurricane storm surges. Measures from water gages arecontinuous as well. Second, some communities do not participate in the National Flood InsuranceProgram, which leads to (i) households’ inability to purchase federal flood insurance and (ii) theabsence of flood mapping. We compare flood zones below. Section 103, (3), (B) “GOVERNMENT-SPONSORED ENTERPRISES FOR HOUSING.--The Federal National Mort-gage Association and the Federal Home Loan Mortgage Corporation shall implement procedures reasonably designedto ensure that, for any loan that is-- [...] purchased by such entity, the building or mobile home and any personalproperty securing the loan is covered for the term of the loan by flood insurance in the amount provided in paragraph(1)(A).” The list of participating communities is presented in the “National Flood Insurance Program Community StatusBook.” omparing Flood Zones Figure 1(a) maps the storm surge areas obtained using a Category 4hurricane at high tide in the New Orleans basin. Hurricane Katrina was a Category 5 floodplain, sothis could be considered a moderate scenario. Figure (b) presents the areas of the flood insurancemaps’ 100-year floodplain.This visual representation sheds lights on the potential limits of using the flood insurancemaps to assess flood risk in the mortgage market. First, while a Category 4 storm surge wouldaffect all parts of the New Orleans metropolitan area except its northern part, the flood insur-ance areas cover a substantially smaller part of the metropolitan area. This is the National FloodHazard Layer provided by FEMA in 2017. The discrepancies are substantial: a stretch from NewOrleans to Laplace is unmapped, while the City of Kenner is assumed to be protected by a leveeand is thus outside of the 100-year floodplain. The southern part of the state of Louisiana is mostlyunmapped. Second, storm surge simulation models provide a continuous and granular visual-ization of storm surge risk, with surges ranging from no surge (in the northern counties of theNew Orleans MSA) to more than 22 feet (in the city of New Orleans, on the Eastern side of LakePontchartrain).Figure 2 on the next page presents the 3 feet sea level rise scenario; in New Orleans the 6 feetand 3 feet scenarios are virtually identical at this scale. Such layers also suggest that the downtownparts of the metropolitan area will not be affected by the slow-moving sea level rise, even as theymay be affected by more frequent and higher storm surges. 6 feet of sea level rise is on the upperbound of the likely scenarios, i.e. in the unmitigated global warming scenarios in 2100.Overall, evidence suggests that NOAA’s storm surge models provide a more conservativeforecast of coastal flood risk.
Limits of the Measures
This paper does not present fluvial flooding measures, which a newversion of this paper will present at a granular level for the conterminous United States. Levee reliability is the focus of a complex and extent literature with both engineering, cost-benefit, and risk prefer-ence considerations, see for instance Tobin (1995), Wolff (2008), Rogers, Kemp, Bosworth Jr & Seed (2015) . .2 Mortgage data This paper matches the predictions of storm surge models with instrument-level financial andeconomic data.
Home Mortgage Disclosure Act data
Mortgage-level information on mortgage applications,originations and securitizations is provided by the Federal Financial Institutions and ExaminationCouncil and by the Consumer Financial Protection Bureau for the period of analysis (2012-2018).Data is collected according to the 1975 Home Mortgage Disclosure Act (codified in 12 USC Banksand Banking). In the so-called Loan Application Register (LAR), for each application, the data in-clude the loan amount, the unique lender identifier (Respondent ID), the applicant income, race,gender, ethnicity, the census tract of the house , the regulating agency, the loan type (conven-tional, FHA, VA-guaranteed, Farm Service Agency or Rural Housing Service), the property type(1-to-4 family, manufactured housing, multifamily), the loan purpose (home purchase, home im-provement, refinancing), the owner-occupancy status. The outcome of the application is recorded(origination, denial, withdrawal by the household, incomplete application), as well as the poten-tial securitization of the mortgage either by an agency (Fannie Mae, Ginnie Mae, Freddie Mac,Farmer Mac, thereafter the “GSEs”), or by private institutions. The file also reports securitiza-tions independently of applications. In this paper we do not restrict or filter the file and includeall originations, regardless of their nature. One important feature of HMDA is the stability of itscodebook since 2004, providing a unique way to assess the evolution of mortgage originationsand securitizations by location over time. McDash
The mortgage data come from the McDash data set compiled by Black Knight financialusing data from the servicing industry. This data set covers about 75% of the mortgage market. This data source is unique as it provides us with a granular view of the composition of the mort-gage market since 1989, at the 5-digit ZIP code level. In addition, we obtained exclusive accessto 5-digit ZIP code data, when other researchers have used 3-digit ZIP code data. An alternative The file uses the Census’ 2000 Tract boundaries until 2012, then adopts the 2010 Tract boundaries. We adjust theintersections accordingly. This coverage fluctuates across years, is highest during the housing boom of 2001-2006, and somewhat lower duringthe housing bust. Hence increases in originations in McDash during the latter period are likely underestimated.
American Community Survey
Household, individual, and housing characteristics are extractedfrom the 5-year averages of the American Community Survey, at the Zip code tabulation area(ZCTA5) level. We use survey weights provided by the U.S. Census Bureau. The ZCTA5 level wasused as the finest level of geographic disaggregation available in the McDash mortgage data set.More granular analysis at the census tract level is available.
The Federal Reserve of Chicago’s Commercial Bank data
Data collected in accordance withthe HMDA also provides a mortgage-level crosswalk with the identity of its lender (RSSDID) inthe reporter panel until 2016 inclusive. Lenders submit a transmittal sheet alongside the loanapplication record (LAR). Such transmittal sheet is linked to the Federal Reserve’s RSSDID. TheFederal Reserve of Chicago provides researchers with reports reports of condition and income forall banks regulated by the Federal Reserve System, Federal Deposit Insurance Corporation, andthe Comptroller of the Currency. As such we do not observe the liquidity and capital levels ofnon-bank lenders. We access such formatted using Drechsler, Savov & Schnabl’s (2018) consistentdata set. 10
Six Facts
Table 1 computes the total volume of originations in billions of dollars, in storm surge areas, i.e.more than 5 feet of storm surge during a Category 4 hurricane, panel (a); in special flood hazardareas, i.e. in the flood insurance maps 100-year floodplain, panel (b); in the 3 feet sea level risearea, panel (c); and in the 6 feet sea level rise area, panel (d). The volume of originations in storm surge areas jumps from 210 billion dollars in 2012 to 249billion dollars annually in 2018, i.e. by about 18.5%. In contrast the total volume of originationsdeclines from 2.1 trillion dollars to 2.0 trillion dollars, a 4.8% decline. This decline may be dueto changes in reporting requirements. Yet these changes in reporting requirements also affectthe volume of originations in storm surge areas, whose share in the total volume increases from9.8% to 12.5%, a 2.7 percentage point increase. The volume of agency originations in storm surgeareas is lower in 2018 than in 2012. This reflects the downward trend in agency originations.Overall, as the volume of agency originations in storm surge areas declines proportionately lessthan the volume of agency originations overall, the share of agency originations in storm surgeareas increases between 2012 and 2018, from 8.6% to 11%, a 2.4 percentage point increase.Similar findings obtain when focusing on areas with more than 10 feet of storm surge duringa category 4 hurricane, instead of areas with more than 5 feet; or when focusing on category 5hurricanes. Indeed, the volume of originations in such areas with more than 10 feet of surge in acat. 4 hurricane is 221 billion dollars in 2018 (compared to 249 billion dollars in areas with morethan 5ft of surge). There is also an increase in originations: from 181 billion dollars of originationsin 2012, to 222 billion dollars of originations in 2018. Similar trends obtain when focusing on areaswith more than 20 feet of storm surge (from 99 billion dollars in 2012, to 133 billion dollars in2018). The total origination and agency origination volumes are of the same order of magnitude as those reported by theUrban Institute’s Housing Finance at a Glance series. 2018 is the last year with public disclosure of HMDA. The 2019data was not available at the time of writing this paper. .2 Fact This contrasts with the findings obtained when focusing on flood insurance maps’ 100-year flood-plains. Table 1(b) suggests that the volume of originations in Special Flood Hazard Areas (the100-year floodplain) is remarkably stable between 2012 and 2018, at 197.2 billion dollars versus197 billion dollars, with some fluctuations in-between these two years. The share of mortgageoriginations in SFHAs is also stable, oscillating between 9.2% and 9.9% throughout this period.A similar stability obtains when focusing on agency originations in SFHAs. The volume ofagency originations in SFHAs declines, from 141 billion dollars to 109 billion dollars, with a sharein SFHAs also remarkably stable between 9.2% and 9.9%.
The HMDA mortgage data presented above present a satisfactory aggregate overview of mort-gage originations. They do not provide the structure of the mortgage, in particular whether themortgage is a “simple” mortgage, fully amortizing with a fixed rate, or whether the mortgagehas any of the complex features described in Amromin, Huang, Sialm & Zhong (2018): whetherthe mortgage is interest-only (and thus does not lead to reductions in principal amount), whetherthe mortgage is a fixed or adjustable rate mortgage (ARM), whether the mortgage was approvedusing full document or using low documentations, e.g. for self-employed individuals.Table 2 and Figure 3 present aggregate statistics on complex mortgage originations by stormsurge area. As before, a storm surge area is defined here as at least 5 feet of storm surge aboveground level during a category 4 hurricane. A Zip code is in the storm surge area if the maximumsurge height across computation cells is greater than 5 feet. The author has checked that otherdefinitions of storm surge areas, for instance using 10 feet of surge, or a category 5 hurricane (suchas Katrina) yield similar stylized facts.The table finds a few such robust stylized facts. First, there is a persistently higher share ofinterest-only loans in Storm Surge Areas: 10.2% vs 2.3%. There is a persistently lower share of fixedrate mortgages in Storm Surge Areas: 79.8% vs 89.9%. There are more full documentation loans12nd fewer no Income no Asset loans in Storm Surge Areas vs the rest of the US. This may suggestthat while lenders give more complex mortgages, they ensure that households have better creditcharacteristics. The next section suggests that households in such areas are in fact significantlymore vulnerable.
Table 3 compares household, individual, or housing characteristics in storm surge areas (SLOSH),sea level rise areas, and Special Flood Hazard areas. Data is from the 5-year average of the 2018American Community Survey, at the ZCTA5 level. The “Rest of the United States” is made ofareas that are in none of the other three categories.Households in Storm Surge areas tend to be poorer than households in the rest of the U.S.,with differences in household income ranging between $1,295 (5ft Storm surge vs. rest of theU.S.) and $2,055 (15ft Storm surge vs. rest of the U.S.). Despite such lower median incomes,households face higher monthly dollar owner costs, which translate into higher monthly ownercosts as a percentage of income. Households in storm surge areas tend to be less likely to be livingin owner-occupied housing. Yet we saw in the previous section that the total volume of mortgageoriginations is higher than in SFHA areas. This is due to the higher price of housing, as householdsin storm surge areas tend to live in more expensive metropolitan areas.As the previous analysis suggests, focusing on storm surge areas is key in describing mortgagevulnerability to coastal flood risk. This is also the case here for households’ vulnerability. House-holds in sea level rise areas have significantly higher income, likely related to the higher amenityvalue of coastal areas not exposed to the short-term risk of hurricane storm surges.The lower panel of the descriptive table (Demographics) shows that storm surge areas displaysubstantially higher fractions of African Americans (up to 7.1 ppt higher in storm surge areas vs.the rest of the United States), higher fractions of Hispanics (18.4% vs. 17.9% in the rest of theUnited States), and lower fractions of Whites (70.5% vs. 76.6% in the rest of the United States).Individuals living in storm surge areas tend to be older, between 0.9 and 2.4 years older thanindividuals living in the rest of the United States. While older individuals may have higher sav- With 36,721 Zip code areas overall, and more than 3,000 per area, the margins of errors (MOEs) do not affect thesignificance of the difference between flood zones in this table.
Finally we turn to the evolution of housing markets in storm surge areas and in FEMA’s SFHAs.We proceed by looking at the price-to-rent ratio as an indicator of future rental risk as the priceincorporates current and future values of rents. In the Gordon (1959) growth model applied tohousing, the price of a real estate asset reflects its flow of rents net of taxes and maintenance costs,discounted by the difference of the discount factor and the growth rate of rents. In this framework,an increase (resp. a decline) in the price to rent ratio reflects the market’s perception of rising (resp.declining) rents. Rents reflect the current flow utility of amenities.If storm surge areas are facing more future risk than flood insurance SFHAs, we should observethat price-to-rent ratios are experiencing lower growth than flood insurance SFHAs. If, on theother hand, the current amenity value of the coast in storm surge areas is similar to the amenityvalue of the coast in flood insurance areas, we should observe no difference in the evolution ofrents.This is what we test by using Zillow’s House Value Index (ZHVI), and its Zillow Rental Index(ZRI). Such price index is available for a subset of 7,439 Zip codes, 339 of which have more than Two other price indices are the Case Shiller index and the FHFA HPI index (formerly called the OFHEO index). Adetailed comparison between the ZHVI and the Case-Shiller is provided by Zillow in this document. current amenity value.This supports the hypothesis that storm surge areas may be exposed to risk that is not typicallymeasured when focusing on flood insurance areas.
Table 5 presents the characteristics and key ratios of lenders in storm surge areas and in floodinsurance SFHA areas. The ratios are computed using the first quarter of 2012, the beginning ofthe expansion of mortgage credit in storm surge areas in Table 1. Each ratio is obtained by takingthe mean of bank characteristics weighted by their volume of mortgage originations in the area.These ratios and characteristics are available for bank lenders. The last row of the table presentsthe share of non-bank lenders in the sample. The non-bank origination share is large overall,consistent with the evidence of Goodman (2020).The evidence presented in this table supports the hypothesis that lenders in storm surge areastend to be larger banks (about 3% larger, 719b$ vs. 672$) than in flood insurance areas. This istrue both on average and for the median bank lender. There is a positive relationship betweenthe storm surge height and the size of the bank lender. Such monotonicity between bank lendercharacteristics and surge height is typical of the findings of this paper.15onsistent with this larger size, the table suggests that banks in storm surge areas have marginallylower return on assets (0.238% vs. 0.243%) and marginally lower return on equity (2.26% vs. 2.3%).They tend to rely less on loans in their portfolio (58.2% in loans over assets vs. 58.4%), and theyhave similar leverage (equity over assets of 10.8% in all four zones).These findings are substantially different than the ones from Beltratti & Stulz (2012). In theirwork, they found that banks that performed better between July 2007 and December 2008 tendedto have lower returns and lower leverage. This paper finds that banks with a portfolio of mortgageloans most exposed to hurricane storm surges tend to have lower returns and similar leverage.One common finding with Beltratti & Stulz (2012) is that banks exposed to storm surges tend torely less on their loan portfolio.
This paper presents an assessment of coastal flood risk in the mortgage market that relies onscientific simulations of storm surges rather than on flood insurance maps, which are inherentlydependent on a community’s decision to participate in the program. Relying on independentmeasures of storm surge risk presents a substantially different picture of coastal flood risk: thereis a large and rising volume of mortgages at risk (constant in flood insurance areas), householdsare more likely to be African American or Hispanic, house prices exhibit lower trends, and lenderstend to be larger and are more likely to be commercial banks.These findings suggest the importance of relying on measures of flood risk that are indepen-dent from flood insurance maps. Future research could explore the endogeneity of flood insurancemapping to the demographics of the neighborhoods, conditional on objective flood risk measures.Future research could also explore the optimal risk taking of financial institutions facing ambigu-ous flood risk in coastal areas.
References
Amromin, G., Huang, J., Sialm, C. & Zhong, E. (2018), ‘Complex mortgages’,
Review of Finance (6), 1975–2007. 16eltratti, A. & Stulz, R. M. (2012), ‘The credit crisis around the globe: Why did some banks performbetter?’, Journal of financial economics (1), 1–17.Drechsler, I., Savov, A. & Schnabl, P. (2018), Banking on deposits: Maturity transformation withoutinterest rate risk, Technical report, National Bureau of Economic Research.Ekman, V. W. (1902),
Om jordrotationens inverkan p˚a vindstr¨ommar i hafvet... , Centraltryck.Ekman, V. W. (1905), ‘On the influence of the earth’s rotation on ocean-currents.’.Genovese, E. & Green, C. (2015), ‘Assessment of storm surge damage to coastal settlements insoutheast florida’,
Journal of Risk Research (4), 407–427.Glahn, B., Taylor, A., Kurkowski, N. & Shaffer, W. A. (2009), ‘The role of the slosh model in nationalweather service storm surge forecasting’, National Weather Digest (1), 3–14.Goodman, L. (2020), ‘Housing finance at a glance: A monthly chartbook: March 2020’.Gordon, M. J. (1959), ‘Dividends, earnings, and stock prices’, The review of economics and statistics (2), 99–105.Hong, H., Karolyi, G. A. & Scheinkman, J. A. (2020), ‘Climate finance’, The Review of FinancialStudies (3), 1011–1023.Horton, B. P., Rahmstorf, S., Engelhart, S. E. & Kemp, A. C. (2014), ‘Expert assessment of sea-levelrise by ad 2100 and ad 2300’, Quaternary Science Reviews , 1–6.Jelesnianski, C. P. (1970), ‘Bottom stress time history in linearized equations of motion for stormsurges’, Monthly Weather Review (6), 462–478.Jelesniansky, C. & Chen, J. (n.d.), Shaffer (1992) slosh: sea, lake and overland surges from hurri-canes, Technical report, NOAA Technical Report NWS 48, National Oceanic and AtmosphericAdministration.Kahn, M. E. (2014), Climate change adaptation: lessons from urban economics, Technical report,National Bureau of Economic Research. 17ahn, M. E. (2016), ‘The climate change adaptation literature’, Review of Environmental Economicsand Policy (1), 166–178.Kalourazi, M. Y., Siadatmousavi, S. M., Yeganeh-Bakhtiary, A. & Jose, F. (2019), ‘Simulating tropi-cal storms in the gulf of mexico using analytical models’, Oceanologia .Keenan, J. M. (2019),
Climate adaptation finance and investment in California , Taylor & Francis.Kossin, J. P., Knapp, K. R., Olander, T. L. & Velden, C. S. (2020), ‘Global increase in major trop-ical cyclone exceedance probability over the past four decades’,
Proceedings of the NationalAcademy of Sciences .Kousky, C. (2018), ‘Financing flood losses: A discussion of the national flood insurance program’,
Risk Management and Insurance Review (1), 11–32.Kousky, C. & Kunreuther, H. (2010), ‘Improving flood insurance and flood-risk management: in-sights from st. louis, missouri’, Natural Hazards Review (4), 162–172.Melton, G., Gall, M., Mitchell, J. T. & Cutter, S. L. (2010), ‘Hurricane katrina storm surgedelineation: implications for future storm surge forecasts and warnings’, Natural hazards (2), 519–536.Mercado, A. (1994), ‘On the use of noaa’s storm surge model, slosh, in managing coastal hazards– the experience in puerto rico’, Natural hazards (3), 235–246.Niedoroda, A., Resio, D., Toro, G., Divoky, D., Das, H. & Reed, C. (2010), ‘Analysis of the coastalmississippi storm surge hazard’, Ocean Engineering (1), 82–90.Parris, A. S., Bromirski, P., Burkett, V., Cayan, D. R., Culver, M. E., Hall, J., Horton, R. M., Knuuti,K., Moss, R. H., Obeysekera, J. et al. (2012), ‘Global sea level rise scenarios for the unitedstates national climate assessment’.Pralle, S. (2019), ‘Drawing lines: Fema and the politics of mapping flood zones’, Climatic change (2), 227–237. 18ogers, J. D., Kemp, G. P., Bosworth Jr, H. & Seed, R. B. (2015), ‘Interaction between the us armycorps of engineers and the orleans levee board preceding the drainage canal wall failures andcatastrophic flooding of new orleans in 2005’,
Water Policy (4), 707–723.Shepard, C. C., Agostini, V. N., Gilmer, B., Allen, T., Stone, J., Brooks, W. & Beck, M. W. (2012),‘Assessing future risk: quantifying the effects of sea level rise on storm surge risk for thesouthern shores of long island, new york’, Natural hazards (2), 727–745.Tobin, G. A. (1995), ‘The levee love affair: a stormy relationship?’, JAWRA Journal of the AmericanWater Resources Association (3), 359–367.Wing, O. E., Bates, P. D., Smith, A. M., Sampson, C. C., Johnson, K. A., Fargione, J. & Morefield,P. (2018), ‘Estimates of present and future flood risk in the conterminous united states’, Envi-ronmental Research Letters (3), 034023.Wolff, D. (2008), ‘Reliability of levee systems’, Reliability-based design in geotechnical engineering pp. 448–496.Woodruff, J. D., Irish, J. L. & Camargo, S. J. (2013), ‘Coastal flooding by tropical cyclones andsea-level rise’,
Nature (7478), 44–52. 19igure 1: Two Approaches to Flood Zones: Storm Surge model and Flood Insurance Maps
The upper panel shows a map of the simulated Maximum of MEOWs storm surge heights for a Category 4hurricane at high tide. The storm surge heights in feet are above ground level. MEOW: Maximum EnvelopeOf Water. This is presented for the basin of New Orleans. The grey area is the New Orleans-Metairie, LAMetropolitan Statistical Area. The black boundaries are the Zip code tabulation areas (ZCTA5) used in themortgage analysis. The bottom panel shows the boundaries of the 100-year floodplain in the National FloodHazard Layer, provided by FEMA in 2017. (a) Storm Surge Simulation – Category 4, High Tide – Surge Above Ground Level in Feet(b) Flood Insurance 100-Year Floodplain – Federal Emergency Management Agency’s National Flood Haz-ard Layer
These figures focus on New Orleans. The paper considers the comprehensive set of basins of storm surgesimulations. Source: mapping by the author using NOAA SLOSH simulations, FEMA NFHL, and USCensus Bureau shapefiles.
This panel present the areas flooded in the case of a 3-feet sea level rise for the New Orleans-Metairie, LAMetropolitan Statistical Area. (a) 3 Feet of Sea Level Rise, NOAA’s Simulations, Louisiana
Source: Mapping by the author using US Census Bureau and NOAA shapefiles.
These figures compare the fraction of complex mortgages for storm surge areas (at least 5 feet of surge duringa Category 4 hurricane) vs. other areas. (a) Share of Interest-Only Mortgages . . . . . . year s ha r e_ i o Not in Storm Surge Areain Storm Surge Area (b) Share of Fixed Rate Mortgages . . . . . . year s ha r e_ f i x ed (c) Share Full Documentation . . . . . year s ha r e_ f u ll _do c u m en t (d) Share No Income No Asset . . . . . year s ha r e_n i na Source: Author’s calculation using the McDash data set from Black Knight financial, NOAA’s Storm SurgeSLOSH simulations, US Census Bureau shapefiles.
The upper panel presents the evolution of the price-to-rent ratio relative to the national trend in SpecialFlood Hazard Areas (FEMA’s insurance map 100-year floodplain, black points); and the evolution of thesame ratio in storm surge areas (SLOSH model, white circles). The bars represent 95% confidence interval,clustering by Zip code and by year. Vertical axis: 0.05 is 5%. Red dotted line for national trend. (a) Price-to-Rent Ratio Relative to the National Trend − . . . Year R e l a t i v e P r i c e / R en t T r end SFHAStorm Surge Area (b) Rent Relative to the National Trend − . . . . . . Year R e l a t i v e R en t T r end SFHAStorm Surge Area
Source: Zillow’s ZHVI house price index and ZRI rental index. Monthly data by postal ZIP.
These four panels present the unweighted sum of mortgage originations in the HMDA data, for four differ-ent types of flood zones: storm surge (at least 5 feet in a cat. 4 hurricane), flood insurance areas (SFHAs),and 3-6ft sea level rise. Data includes all originations. Agency originations: from Fannie Mae, FreddieMac, Ginnie Mae, and Farmer Mac. (a) Storm Surge AreasYear OriginationAmount (b$) In Surge Area(b$) In Surge Area(%) AgencyAmount (b$) Agency inSurge Area(b$) Agency inSurge Area(%)2012 2,135 210 9.80 1,528 131 8.602013 1,903 203 10.60 1,609 159 9.902014 1,386 151 10.90 826 77 9.302015 1,848 197 10.70 1,138 105 9.202016 2,181 225 10.30 1,375 125 9.102017 1,930 211 10.90 1,170 115 9.802018 1,993 249 12.50 1,100 121 11.00(b) Special Flood Hazard AreasYear OriginationAmount (b$) In SFHA (b$) In SFHA (%) AgencyAmount (b$) Agency inSFHA (b$) Agency inSFHA (%)2012 2,135 197 9.20 1,528 141 9.242013 1,903 180 9.50 1,609 152 9.482014 1,386 134 9.70 826 79 9.662015 1,848 177 9.60 1,138 108 9.572016 2,181 207 9.50 1,375 130 9.512017 1,930 190 9.90 1,170 115 9.852018 1,993 198 9.90 1,100 109 9.93(c) 3ft Sea Level Rise AreasYear OriginationAmount (b$) In 3 ft SLR(b$) In 3 ft SLR (%) AgencyAmount (b$) Agency in 3ftSLR (b$) Agency in 3ftSLR (%)2012 2,135 34 1.60 1,528 21 1.372013 1,903 31 1.60 1,609 23 1.432014 1,386 23 1.60 826 11 1.332015 1,848 30 1.60 1,138 15 1.322016 2,181 35 1.60 1,375 18 1.312017 1,930 32 1.60 1,170 16 1.362018 1,993 33 1.70 1,100 15 1.37(d) 6ft Sea Level Rise AreasYear OriginationAmount (b$) In 6 ft SLR(b$) In 6 ft SLR (%) AgencyAmount (b$) Agency in 6ftSLR (b$) Agency in 6ftSLR (%)2012 2,135 67 3.10 1,528 42 2.752013 1,903 64 3.30 1,609 46 2.862014 1,386 46 3.30 826 22 2.662015 1,848 62 3.30 1,138 31 2.722016 2,181 71 3.30 1,375 37 2.692017 1,930 64 3.30 1,170 32 2.742018 1,993 70 3.50 1,100 31 2.81
Source: author’s calculations from FFIEC’s HMDA, NOAA’s SLOSH, FEMA’s NFHL, NOAA’s Sea LevelRise layers, and US Census Bureau shapefiles.
Storm surge area: at least 5 feet of storm surge above ground level during a Category 4 hurricane at hightide. Sample: First mortgages, in 5-digit ZIP areas with at least 10 loans, and on mortgages for owner-occupied housing, where the property value at origination is above $ $ (a) Storm Surge AreasYear Total InterestOnly InterestOnly FixedRate FixedRate FullDoc. FullDoc. N.I.N.A. N.I.N.A.M$ M$ % M$ % M$ % M$ %2006 130,058 33,128 25.5% 57,775 44.4% 28,617 22.0% 6,003 4.6%2007 89,887 24,617 27.4% 53,698 59.7% 28,386 31.6% 4,871 5.4%2008 46,444 6,341 13.7% 35,770 77.0% 27,268 58.7% 1,638 3.5%2009 49,223 2,162 4.4% 44,377 90.2% 25,555 51.9% 779 1.6%2010 41,824 3,056 7.3% 34,050 81.4% 27,157 64.9% 373 0.9%2011 37,329 4,348 11.6% 26,386 70.7% 24,512 65.7% 662 1.8%2012 54,484 5,208 9.6% 41,330 75.9% 33,942 62.3% 182 0.3%2013 51,983 5,933 11.4% 38,721 74.5% 32,583 62.7% 113 0.2%2014 34,266 3,713 10.8% 24,947 72.8% 10,639 31.0% 84 0.2%2015 51,535 5,050 9.8% 40,253 78.1% 10,587 20.5% 95 0.2%2016 56,364 5,771 10.2% 44,994 79.8% 10,596 18.8% 87 0.2%(b) Other ZIP CodesYear Total InterestOnly InterestOnly FixedRate FixedRate FullDoc. FullDoc. N.I.N.A. N.I.N.A.M$ M$ % M$ % M$ % M$ %2006 1,685,407 406,480 24.1% 948,216 56.3% 419,192 24.9% 111,374 6.6%2007 1,429,807 324,624 22.7% 1,047,208 73.2% 472,041 33.0% 87,932 6.1%2008 903,184 66,331 7.3% 797,292 88.3% 484,876 53.7% 53,654 5.9%2009 1,148,235 16,648 1.4% 1,099,416 95.7% 582,970 50.8% 43,295 3.8%2010 984,994 17,594 1.8% 903,601 91.7% 559,498 56.8% 35,691 3.6%2011 851,573 18,178 2.1% 745,714 87.6% 461,347 54.2% 40,153 4.7%2012 1,293,600 20,246 1.6% 1,193,174 92.2% 681,359 52.7% 17,253 1.3%2013 1,043,271 21,058 2.0% 950,015 91.1% 548,348 52.6% 18,840 1.8%2014 594,731 13,523 2.3% 516,695 86.9% 166,558 28.0% 8,177 1.4%2015 793,719 17,128 2.2% 710,655 89.5% 165,605 20.9% 9,299 1.2%2016 811,134 18,553 2.3% 729,140 89.9% 143,375 17.7% 6,694 0.8% Source: Author’s calculations using the McDash data provided by Black Knight financial.
This table presents summary statistics for households and individuals in Zip code tabulation areas (ZCTAs),depending on storm surge heights in the ZIP code tabulation area (columns 2–4) and depending on the shareof the ZIP code in Special Flood Hazard Areas, the 100-year floodplain of insurance maps. (1) (2) (3) (4) (5)Rest of theU.S. Storm Surge5ft, Cat 4Hurricane Storm Surge10ft, Cat 4Hurricane Storm Surge15ft, Cat 4Hurricane Special FloodHazard Area
Housing Units
House value ‡ $252,288 $317,241 $315,413 $318,604 $228,930Monthly owner cost ‡ (cid:63) $1,226.1 $1,360.8 $1,347.1 $1,347 $1,160.3As % of income ‡ (cid:63) ‡ $1,055.5 $1,244.6 $1,241.8 $1,232.5 $1,047.1% Owner occupied 64.8% 59.5% 59.5% 59.0% 65.1%% Mobile homes 5.93% 5.53% 5.7% 5.77% 7.19% Demographics
Household income ‡ $64,811 $63,516 $63,205 $62,756 $62,698Age ‡ ‡ : median. (cid:63) : households with a mortgage. Source: ZCTA5 5-year average of the 2018 American Commu-nity Survey. NOAA’s Sea LevelrRise layer, NOAA’s SLOSH MOMs, FEMA’s Special Flood Hazard Areasfrom the 2017 National Flood Hazard Layer (NFHL). This table presents bank lenders’ key ratios in four types of flood zones (defined before). Each ratio isthe average ratio of bank lenders in each zone in the first quarter of 2012, weighted by lenders’ mortgageorigination volume in that zone.
Flood Zone:(1) (2) (3) (4)Storm Surge 5ft Storm Surge 10ft Storm Surge 20ft SFHAAverage Assets (M$) 719,933 717,457 743,404 672,091Median Assets (M$) 330,227 330,227 330,227 206,808Return on Assets (Quarterly) 0.238% 0.238% 0.233% 0.243%Return on Equity (Quarterly) 2.26% 2.26% 2.22% 2.3%Loans over Assets 58.2% 58.2% 57.5% 58.4%Deposits over Assets 69.7% 69.7% 69.1% 71.2%Liquidity over Assets 19.3% 19.2% 19.3% 19.9%Equity over Assets 10.8% 10.8% 10.8% 10.8%Share of Non-Bank Lenders 62.66% 62.35% 60.0% 62.9%