Wildfires, Housing Affordability and Climate Change: Virtuous Cycle or Death Spiral?
● Assuming a business-as-usual carbon emissions trajectory and resulting climate change in the coming decades, catastrophic wildfires are expected to be ~15% more likely on average across the U.S. by 2060, increasing annual average economic damage from wildfires by ~8%. Upper bounds on climate model projections imply increases in wildfire risk could be substantially higher in a worst-case scenario.
● Driven in large part by constricted housing supply, human settlement in wildfire-prone areas has increased significantly over the past several decades. Populations in high wildfire risk locations are disproportionately Latinx, Native American, and Asian; these same populations live in communities where housing is less affordable.
● Lack of access to affordable urban housing also exacerbates automobile dependence and carbon emissions. California has a hidden liability of more than $8B on its balance sheet from on-road emissions alone, and those emissions contribute to the climate change that makes its own wildfire risk worse.
● The Government Sponsored Entities mandated to make home ownership more accessible and affordable are disproportionately exposed to wildfire risk. Ginnie Mae, the only GSE which cannot pass risk on to the market, holds twice the extreme risk on its books compared to mortgage lending institutions.
● Even after accounting for affluence, loan-to-value and other key credit risk metrics, mortgage delinquency is significantly more likely in locations exposed to high wildfire risk, implying higher default likelihood that has direct implications for mortgage-backed securities stakeholders, and, via the likes of Ginnie Mae, for the US taxpayer.
● More effective urban densification centered on affordability can turn a death spiral into a virtuous cycle, at once mitigating wildfire exposure, reducing transportation emissions, and advancing climate and social justice. This is a policy issue at state, county and city levels, but one where the financial community can play an active role.
● In parallel, exposed communities should take a variety of practical, cost-effective measures to manage and mitigate their wildfire risk. In addition to preventing loss of life, injury, and economic loss, debt-issuing communities should disclose these measures to bond investors and credit rating agencies that are rapidly growing more wary of and concerned with wildfire risk.
The basics: How will climate change influence wildfire risk?
risQ collaborated closely with the US Forest Service (USFS) to develop methodology for “climate conditioning” wildfire hazard risk. The climate conditioning methodology uses IPCC Global Climate Model (GCM) output-derived probabilistic “change factors” in weather variables that are linked to the probability of fires. Change factors are used to alter input weather data that (among other data like land cover) drives the USFS FSIM fire simulation software.
risQ’s FSIM modeling reveals that in many areas of the country, climate change is projected to result in warmer weather during both the spring and fall, as well as warmer and drier summers. These seasonal changes will lengthen a region’s fire season, which in turn increases the likelihood that ignitions will occur when weather conditions are conducive to fire spread. The coupling of more ignitions with longer periods of “fire weather” results in both larger and more frequent fires.
These results, which are based on the full physical model, lead to a simplified interpretation that hotter and drier weather are conducive to wildfires, and as such, temperature and relative humidity are key variables in understanding changes in wildfire risk over time. The fossil fuel-intensive climate scenario RCP8.5 projects both hotter temperatures and drier conditions (measured through relative humidity) country-wide — leading to significantly higher wildfire risk. RCP4.5 projections show about half as much warming as RCP8.5, and the signal in relative humidity is mixed — leading to a “mixed bag” of results across the country. It is worth noting that analyses in this sections are based on median futures from thousands of probabilistic simulations based on each of RCP8.5 and RCP4.5, respectively. Upper bound projections of RCP8.5 could be thought of as a worst-case climate scenario, where temperatures would increase double the median and relative humidity would decrease twice as much as the median — amplifying wildfire risk even further. Numbers here should be considered best median case estimates of change in risk, but worst-case warming and drying could double the risks or more. The Appendix provides more detail on risQ’s probabilistic climate projections and their relationship with wildfire probability.
Before examining the financial risks of climate conditioned wildfire, which involves analysis of where human built infrastructure exists, it is important to understand expectations around changes in the patterns of the hazard by itself.
In Figure 1, RCP8.5 shows increased wildfire probability across the coterminous US owing to large increases in temperature and widespread decreases in relative humidity. Leading the country is Texas in increased burn probability which, along with Oklahoma, is expected to be significantly drier under RCP8.5. For these oil and gas heavyweights, the choice to burn fossil fuels now is a choice to burn property later. Several states in the northwestern region, including Washington, Oregon, Idaho and Montana, also show significant increases owing in particular to the well-understood phenomenon of amplification of warming further north. The overall range of state-level increases in burn probabilities under RCP8.5 is +0.7% to +30% and a median of +15%.
The more moderate RCP4.5 scenario, an increasingly improbable storyline where the world transitions away from carbon, is a mixed bag, with some high-risk states like Utah with slight decreases in risk, and most notably California and Nevada showing little change. The overall range of state-level increases in burn probabilities under RCP4.5 is -5% to 15%, with a median of +1%.
Projecting future wildfire losses: California, we have a problem
risQ’s wildfire model consists of simulating hundreds of millions of physically possible events, under both current and future (RCP8.5 and RCP4.5) climate scenarios. To view the relative risk of locations within and between states, industry exposure-weighted annual expected loss costs are created at a 100-meter resolution. Loss cost is an insurance term for the division of annual aggregate property damage by exposure. This metric effectively offers insight into relative insurance premiums per exposures by location. The sum of loss cost in dollars can be thought of as the total market insurance premium that should be collected per year, ignoring any allowances for reinsurance and administration costs. This sum should not be confused with the Annual Average Loss (AAL) obtained from running an event driven catastrophe model, which would typically provide AAL for areas impacted by a subset of events, geared more towards calculating reinsurance costs.
Single family residential (SFR) homes comprise the asset class most vulnerable and relevant to wildfire risk. risQ’s estimate of annual loss cost to SFR under current climate is ~$13B, with California leading the way and representing the majority at ~$8.8B. By 2050 under RCP8.5, risQ projects that annual loss cost will increase to ~$14B (~+8%) over the entire U.S. and to ~$9.4B in California (~+6.8%). Meanwhile, total loss costs in RCP4.5 are expected to net out to be approximately the same by 2050 at ~$13B (it is worth noting that some states would experience increasing losses while others likely the opposite).
By integrating year-over-year upticks in expected loss costs under RCP8.5 from 2021 to 2050, we can estimate a 30-year cumulative marginal “wildfire cost” of RCP8.5 of ~$17B nationally, as compared to RCP4.5.
It is important to note that this analysis is with respect to an assumption of “fixing” the current national stock of SFR. Shortly, we addresses the reality that increasing development in wildfire-prone areas is a major factor in skyrocketing losses in recent years.
The risQ Wildfire Score: One number, everything you need to know
The risQ Wildfire Score is a 0–5 score that translates all probabilities and severities of wildfire and expected property losses to a single score. This becomes a relative wildfire risk metric across the coterminous U.S. for a fixed point in time, wherein every integer increase in the score maps to an approximate doubling of financial risk. Figure 3 shows the score at county level with overlays for large fire perimeters spanning the years 1984–2020. We characterize high wildfire risk as any location that has a risQ Wildfire Score of >= 3.0; ~12% of counties in the U.S. meet or exceed this high-risk threshold. Between 1990 and 2018 wildfires in these high-risk counties have accounted for ~$46 billion in insured losses, which is ~86% of all insured losses from wildfires having caused at least $100 million of loss.
Wildfires have become more frequent, but we’ve built more houses in high-risk places
Academic research from 2018 showed significant growth in the so-called Wildland Urban Interface (WUI) from 1990–2010. Our analysis confirms that this trend has continued over the last decade coinciding with increases in real estate prices and constricted housing supply. Over 2010 to 2019, in the highest 25% wildfire-prone counties, the share of all Single Family Residence (SFR) permits issued went from ~30% to ~37%, after a dip through the global financial crisis (Figure 5). Over 2010 to 2019, a total of 2.2M SFR permits were issued for the highest 25% wildfire-prone counties — led by Florida (526k), California (419k), Texas (247k), Arizona (204k) and Utah (125k).
Wildfires compound socioeconomic and racial inequality
In total, the communities with the highest wildfire risk in the nation, which tend to be disproportionately Latinx, Native American, and Asian, are among the most stretched from an income-to-housing cost perspective.
Relative to their share of the US population as a whole, Latinx, Native American, and Asian populations are disproportionately exposed to wildfire risk. Black and non-Hispanic white populations are less exposed relative to their share of the US population.
Notably:
- While ~18% of the coterminous US population is Latinx/Hispanic, ~37% (about double the rate) of the total population exposed to the nation’s most extreme wildfire (risQ Wildfire Score = 5) is Latinx. Over 2010 to 2019, that share has significantly increased, from ~32% to ~37%.
- Native American and Asian populations also have outsized exposure to wildfire risk. For example, Native American share of population exposed to the most extreme risk (risQ Wildfire Score = 5) is more than ~1.6x their share of the coterminous US total population. Asian populations similarly are overrepresented in the most extreme risk locations by ~1.5x.
- Meanwhile, Hispanic non-white representation in the highest risk locations is lower than the group’s national population share. That share has decreased significantly from 2010 to 2019, in the most extreme wildfire prone locations, from ~63% to ~56%.
Populations in high wildfire risk communities on average spend a higher percentage of income on housing costs. risQ’s Housing Unaffordability Score is a 0–100 composite index of several different census measures of housing costs relative to income. For locations with non-negligible wildfire risk (risQ Wildfire Score > 0), after adjusting for the effect of population density, a one integer point increase in the risQ Wildfire Score is related to a statistically significant ~4–5 point increase in the Housing Unaffordability Score. This is largely driven by California, which leads the nation in both wildfire risk and housing costs. But even for high-risk locations within California, a one integer point increase in the risQ Wildfire Score is related to a statistically significant ~1–2 point increase in the Housing Unaffordability Score.
Wildfire Risk and Carbon Transition Risk Go Hand in Hand
While in Texas and Oklahoma burning conundrums — oil & gas revenue versus wildfire losses — are apparent, California has its own potentially self-inflicting wound to address. Climate gentrification in California owing to constricted housing supply is pushing more single-family development into the Wildland Urban Interface, necessitating even more dependence on cars and consequently carbon emissions. Los Angeles is a visually clear-cut case, showing a stark contrast between the intensity of on-road emissions downtown, with many communities on the periphery sitting in high-risk wildfire zones. In total, California leads the nation in on-road emissions, clocking in at ~2.2% of all the U.S.’s entire carbon footprint (California represents about 10% of all on-road CO2, which is ~22% of all US emissions). Pricing CO2 at the Biden administration’s $51/ton, that would amount to what is currently a “hidden liability” on California’s balance sheet of ~$8–9B annually.
From a Scope 1 perspective, this liability would actually hit the roads on which carbon is combusted — much larger than the state’s 2020–2021 total Local Streets and Roads budget of ~$2.8B. From a Scope 2 perspective, depending on federal and state carbon transition policy decisions, this liability could play at the gas pump, squeezing already economically pinched, disproportionately less white communities.
Wildfire Risk in the Municipal Debt Market
As of May 2021, of the ~$3.9T in outstanding municipal debt, ~$240B (7%) is exposed to significant wildfire risk (risQ Wildfire Score >= 3). Of that, only ~$19B (8%) is insured. Using debt impairment analytics created by risQ’s partner Municipal Market Analytics (MMA), ~8.3% of the almost $300B in outstanding debt held among more than 800 impaired issuers is exposed to significant wildfire risk, a slightly higher rate than the market rate of 7%. Such a difference makes sense given almost 2/3 of the costs of wildfires are covered by state or local government agencies. It is not hard to see how a wildfire could be the last straw for an already stretched balance sheet. Right now, supply of municipal debt relative to demand is historically low, and consequently there is little yield spread — wildfire risk is not correlated with bond yield. Experts are worried about a(nother) potentially catastrophic wildfire season for 2021; meanwhile, investors are not getting paid any more for taking on the risk.
Wildfire Risk in the Residential Mortgage-Backed Securities Market
Adverse mortgage credit outcomes have been historically worse in locations exposed to wildfires. Using a dataset of almost 5 million Fannie Mae home loans originated from 2005–2018, after accounting for standard loan level attributes (loan-to-value ratio, credit score, debt-to-income ratio, loan purpose) and broad economic conditions using a “year” variable as a proxy, a regression analysis showed that for every one-point increase in the risQ Wildfire Score, the probability of a 4+ month delinquency increased by ~9% (95% CI: 8.6–9.6%). The same analysis with almost 3.5 million Freddie Mac loans relates to a ~11% (95% CI: 10.8–12.8%) increase in a 4+ month delinquency probability, per point increase in the risQ Wildfire Score.
GSEs hold an outsized amount of the most extreme wildfire risk on their books, especially Ginnie Mae. For all loan value ($) originated over 2012–2020, ~1.8% fell in extreme fire risk locations — defined here by having a risQ Wildfire Score of greater than or equal to 4.5. Only ~0.9% ($51B) of loan value held by lending institutions falls into this category. Meanwhile, Ginnie Mae holds more than double that rate, at ~2.3% ($35B), and Fannie Mae and Freddie Mac clock in at similar extreme risk shares of ~2.1% ($56B) and ~2.0% ($34B), respectively. In other words, loan originators have sold a disproportionate amount of wildfire risk to the GSEs. Fannie and Freddie transfer at least some of that risk back to fixed income to insurance-like Credit Risk Transfer (CRT) bonds, discussed next. But Ginnie Mae loans are at once the most exposed to the most severe wildfire risks — and, having no CRT-like instruments, are ultimately the burden of the taxpayer.
We tested the hypothesis that, given the uptick in catastrophic wildfire damage in recent years, increasingly aware lenders would increasingly securitize loans in the highest risk locations over the same time period. While research shows that securitization rates are growing in hurricane-prone locations, this does not appear to (yet) be the case for wildfire-prone locations. Over 2012 to 2019, there is no discernible trend in GSEs absorbing an increase in wildfire risk. One hypothesis for this is that, unlike hurricanes and FEMA flood insurance zones, lenders have had less information on what locations are at risk to wildfire. Rather than being prescient and consciously preloading wildfire risk across to the GSE’s, blind luck has delivered the same outcome.
For Fannie Mae (FNM) and Freddie Mac (FHL) CRT bonds, there is a clear relationship flowing from wildfire risk that correlates to probability of severe delinquency (as discussed previously) and, consequently, default risk that can directly hit the bottom lines of investors. As of June 2021, ~5.7% ($83B) of outstanding loan value collateral across 117 CRT pools covered by risQ is exposed to high wildfire risk (risQ Wildfire Score >= 3). But not all deals are created equally: a randomly constructed pool of 10,000 loans has a 95% chance of having between ~5.1% to 6.3% of its loans exposed to high wildfire risk. As Figure 11 illustrates, very few pools actually fall within this range — the full range of at-risk collateral is as low as 1.8% and as high as 9.6%. For CRT investors, this translates to a straightforward risk management opportunity.
The ESG Solution: Affordable Urban Density
As with any challenge of mitigating climate change and its impacts, there are myriad stakeholders with decisions to make. Using “controlling your own controllables” as a framework, there are two stakeholders that control their own destinies and those of their constituents.
State, county and city policy makers in the US have total control over where property is developed, the carbon footprint those locations invoke, the financial risks and costs these decisions amplify, and the degree to which their (often disadvantaged and marginalized) populations are put in harm’s way. The infrastructural costs advantages, climate change mitigation, and climate risk benefits of urban densification are self-reinforcing for the likes of California. That said, while money is cheap and fails to sufficiently differentiate between the merits of sprawl in higher risk locations versus densifying redevelopment in low-risk locations, the party will keep going. That shines a light on a second key participant: the investor community.
Fixed income stakeholders are both exposed to the financial risks outlined in this report but also in a unique position to address them. Municipal market investors should look to tilt their portfolios toward projects that encourage housing density and cleaner forms of public transportation — San Diego’s regional transportation planning agency’s recent $160B proposal may be a timely example. Projects like these are promising from a downside protection perspective — reducing physical and per capita carbon liabilities as described in this report. In contrast, far too many Community Facilities Districts (CFDs) in the state fail the same test, enabling sprawl-ridden new development in higher wildfire risk areas where new loans are then originated and passed willingly into the financial system without anyone being appropriately aware.
In fact, CFDs are a textbook example of the mutually reinforcing shortcomings of all the aforementioned stakeholders. So called Mello-Roos districts were enabled by state-level legislation in 1982. Leveraged by counties and cities to enhance tax collections to help fund new development, they often only need voting approval of the property developer currently owning the land. It is a tax tied to land area, not property value, making it inherently more disadvantageous to lower socioeconomic groups from a tax perspective. Meanwhile, many are on the outskirts of cities and in the WUI, creating enhanced wildfire risk in general and for lower socioeconomic group homebuyers and renters. This debt is being willingly absorbed by supply-starved municipal bond investors likely unaware of the wildfire risk due to poor climate risk disclosure by the debt issuers. Meanwhile, the resulting new development generates new mortgages, many of which will be conforming loans eligible for securitization to and through the GSEs and, by inference, to the US taxpayer. This is no different than coastal locations on the eastern seaboard of the US continuing to build in flood-prone areas, therein generating tax revenues and real estate transactions only for the risk to be passed on to the national tax base via later FEMA bailouts.
Smart density is critical from a long term financial viability perspective — as compared to the well understood risk of financing sprawling suburbanization, which has been likened to a Ponzi scheme and shown to be dramatically tax-inefficient by Strong Towns. Increasing housing supply and affordable housing projects in denser areas should also ease the cost of living for less affluent and disproportionately non-white and marginalized communities. These same fixed income investors also have the market power, working with lenders and GSEs, to place an ESG premium on mortgage debt that solves for, rather than encourages, lending myopia that creates inevitable and avoidable financial toxins that will only get worse with climate change.
What can communities and people do to manage their wildfire risk?
Of course, existing communities exposed to wildfire cannot afford to simply wait for systemic change in terms of policy, the financial system, densification of housing supply, and housing affordability. The USFS Fire Modeling Institute conducts training for and makes resources available to communities surrounding practical, cost-effective wildfire management and risk mitigation practices.
Preventing loss of life and injury is the most important outcome within the control of a community. Having community plans for safe areas to gather in the event of a wildfire, or for timely evacuation, is crucial.
Fuel (combustible land cover) treatment via prescribed burning:
- Prescribed burning near homes in the Wildland Urban Interface (WUI) reduces the probability of future fires reaching the WUI as well as the risk of high-intensity fire in the event a fire does reach the WUI.
- Prescribed burning on private land on larger parcels with homes can also be implemented to reduce risk.
- After a century of fire suppression, forest composition and structure has been altered in many places. Fuel treatments that include prescribed burning or wildfire can restore forest structure and remove fuel buildup.
Simple measures of property maintenance can save homes. Homes can be destroyed by either direct flame contact or ember showers. To prevent the former, woodpiles and flammable vegetation should not be near homes, and dry leaves should be cleared from around homes and under decks. To prevent ignition by ember showers, measures like clearing dry leaves from gutters can reduce risk.
Home construction choices matter. Using less flammable building materials and designs can also reduce risk. Wood shingles are highly flammable, composition shingles are less flammable, and metal roofing is least flammable of all.
Municipal bond investors and rating agencies increasingly scrutinize debt-issuing communities for their wildfire risk and disclosure of it. It will also be beneficial for communities to explore and enact the above practices as appropriate — and ultimately to disclose measures taken in Official Statements for bond fundraising.
Appendix
Probabilistic Climate Change and Wildfire Probability
Figure S1 shows relationships between changes in modeled, state-averaged wildfire Burn Probability (BP) with temperature and relative humidity from 2020 to 2060, both for the fossil fuel intensive scenario RCP8.5 and for the “some climate action” RCP4.5 scenario.
While the relationship between Burn Probability and climate in the full risQ-conditioned FSIM hazard is more complex and varies by region and land cover, this simple regression analysis provides a basic physical intuition for what drives the wildfire risk projections. Holding all else constant, pooling data from both scenarios together into a simple regression model at a state level implies that from 2020 to 2060 (i) for every degree Celsius increase in summer temperature, on average BP change increases by about 2 to 5 percentage points (95% CI); (ii) for every percentage point increase in relative humidity, BP change decreases on average by about 2 to 6 percentage points. Figures S2 and S3 show median projections plus 95% confidence intervals for state-level projected changes in temperature and relative humidity under RCP8.5 and RCP4.5.
