Impacts of Climate Hazards on Employment and Earnings by Hazard Type, Industry, Race, and Ethnicity

BACKGROUND AND LITERATURE

A natural disaster can shape a local economy through several pathways. First, a hazard could lead to capital depreciation. Capital depreciation occurs directly as a result of destroyed infrastructure, such as roads, ports, and rail lines. An inflow of aid can counteract these impacts and may even lead to improvements in infrastructure in the longer run, or aid could forestall these negative impacts for a few years. Second, a natural disaster could reduce productivity. This can arise directly by reducing a worker’s ability to engage with the local economy if, for example, their own home is ruined. Again, the amount and nature of aid can shape this response in the longer run. A larger amount of aid in the short run could support construction efforts and thus the longer-term impacts on the local economy may not be observed until aid runs out. Third, a natural disaster could change the nature of local amenities. For example, if a nearby forest is burned down, this could impact tourism in the area, or cause people to move away since a key amenity is no longer a part of the community. Alternatively, local investments that arise as a result of the disaster could also change this trajectory, perhaps even creating new amenities (Bilal and Rossi-Hansberg 2023; Boustan et al. 2020). Ultimately, how local economies respond to natural hazards, and the timing of this response are an empirical question and perhaps a policy choice.

There is a growing body of knowledge examining the economic impacts of natural disasters. A recent review of the literature by W. J. Wouter Botzen and colleagues (2019) synthesizes the significant negative direct and indirect economic consequences of these events, which include outcomes ranging from gross domestic product (GDP), GDP growth rate, trade flows, death counts, employment, per capita income, expenditures, migration, housing and other asset values, and government transfers. Botzen and colleagues (2019) highlight the lack of consensus within this literature. To provide some examples, one key finding in this body of work is a significant negative impact on GDP growth following a disaster. Gabriel Felbermayr and Jasmin Gröschl (2014) find that a hazard in the top 1 percent of the disaster index distribution reduces the GDP growth rate by 7 percent, but that these large impacts only hold for the most significant events. Another well-identified study by Solomon M. Hsiang (2010) examines the effect of cyclone intensity on economic activity in twenty-eight Caribbean-basin countries by sector, finding that agriculture, wholesale, retail, and tourism sectors are all impacted negatively by cyclones, whereas the construction sector grows. More recently, Stephanie Lackner (2018) shows that earthquakes reduce per capita GDP for low- and middle-income countries but may boost it for high-income countries. In an attempt to unify this literature, Laura Bakkensen and Lint Barrage (2025) develop an empirical-structural approach to studying the impacts of climate change on growth, highlighting the need for empirical researchers to understand impacts on the structural determinants of growth rather than growth itself. They provide empirical evidence on impacts of cyclones, finding a wide range of results from substantial negative effects on economic growth in vulnerable small island states to small gains, particularly in countries where risks are predicted to decline.

Examining the set of papers studying local labor market impacts within the United States, which is the focus of this article, we also find heterogeneity in outcomes. Anna Rhodes and Max Besbris (2022) provide a unique view into the responses of households in a middle-class community in Texas following Hurricane Harvey. After following fifty-nine households for two years, they bring to light a few key factors that shaped households’ decision-making. Importantly, they note the strong connections many households feel to their community are one of the primary reasons many provide for staying. They also highlight that in addition to the availability of federal support, private insurance is also a key determinant of the households’ ability to rebuild. In fact, they find that for households with flood insurance, some residents considered the experience an “opportunity” to rebuild and perhaps even added to the household’s savings, whereas those without this additional insurance often had to take on additional debt. There are also a number of quantitative case studies of specific disasters, which again provide mixed results, highlighting the need for work that provides a broader view (see, for example, Vigdor 2008; Hornbeck 2012; Gallagher and Hartley 2017; Kirchberger 2017; Deryugina et al. 2018; Groen et al. 2020; Meltzer et al. 2021). There are a small set of quantitative studies which examine a wider set of natural hazards within the US, which we summarize in table 1.

Focusing first on studies examining the impacts of hurricanes in the US, the broad consensus is that hurricanes lead to longer-term declines in county-level employment but perhaps increases in earnings. Ariel R. Belasen and Solomon W. Polachek (2008) study hurricanes in Florida using county-level employment data and find that earnings of the average worker in a Florida county rise by over 4 percent within the first quarter of being hit by a major Category Four or Category Five hurricane relative to counties not hit, and simultaneously employment falls between 1 and 5 percent depending on hurricane strength. Brigitte Roth Tran and Danel J. Wilson (2025) examine a wider set of hazards across the US but, when focusing only on hurricanes, find similar results for earnings. Eric Strobl (2011) examines hurricanes across the coastal USA and finds that hurricanes reduce a county’s annual economic growth rate by 0.45 percentage points, with 28 percent of it due to richer individuals moving away from affected counties. Elizabeth Fussell and colleagues (2017) find that counties with past population declines hardly experience changes in population as a result of a hurricane, but in growing counties hurricanes suppress future population growth. Ilan Noy and Eric Strobl (2023) find that hurricanes lead to a temporary boost in damage-mitigating patents a few years after the event, but there is a long-term general reduction of innovation after a damaging storm.

Looking next at studies examining the impacts of wildfires on local economies, here we see results are more mixed. Raphaelle G. Coulombe and Akhil Rao (2025) examine local labor market impacts of fires and find that increased fire exposure causes lower employment growth in the short and medium run, with medium run impacts linked to migration. Tran and Wilson (2025) find no statistically significant impacts of fires on earnings. Margaret A. Walls and Matthew Wibbenmeyer (2023) find positive impacts of major wildfires on county-level employment growth for up to one year after a fire, but no significant effects after that.

There are also a few studies that examine the impacts of a wide range of natural hazards on local economies on a related but different set of outcomes, including mobility and inequality, again finding a mixed set of outcomes. Leah Platt Boustan and colleagues (2020) construct a long panel examining a broad set of natural hazards and find that disasters increase out-migration rates at the county level by 1.5 percentage points and lower housing prices and rents by 2.5–5 percent. They find that migration responses to milder disasters are smaller but increasing over time. When focusing on particular hazard types, they find that wildfires and hurricanes encourage out-migration, while floods actually attract in-migrants to an area.² Storms and tornadoes have no effect on migration flows. Regina Pleninger (2022) examines the impacts of a wide range of natural hazards on county-level income distributions and finds that impacts are most damaging for middle incomes, with little impact on overall income distribution. She finds results are mostly driven by heavy storms and hurricanes.

Most closely related to our article, Tran and Wilson (2025) examine the impacts of a wide range of natural hazards on earnings and employment and find that disasters increase total and per capita income over the longer run. They find that this effect is driven initially by an employment boost and in the longer run by higher wages. They find that these results are primarily driven by tornadoes and hurricanes. They also find that over the longer run, house prices increase while population is roughly flat, especially in areas with inelastic housing supply. They find that the longer-run increase in income is largest for the most damaging disasters.

Overall, the results from these studies suggest that local economic impacts of climate hazards vary widely, particularly in relation to our key economic constructs of interest, earnings and employment, with some studies finding positive impacts on earnings, others finding no impacts, some finding positive impacts on employment, and others finding negative impacts on employment. One reason that these findings may vary is that impacts of natural hazards will affect individual sectors of a local economy differently. For example, in construction we expect to see increases in both earnings and employment following a natural disaster, but in other industries, particularly those providing services, we may expect to see declines in both earnings and employment.

Only a few of these studies examined impacts on earnings and employment within particular sectors. Beginning with construction, a sector where theory most strongly predicts an increase in employment and earnings, due to the strong demand generated in this sector post-natural hazard, authors find some evidence of a positive impact on earnings and employment and some evidence of no impact. Belasen and Polachek (2008) find positive impacts on earnings and no impacts on employment in the construction sector. In contrast, Tran and Wilson (2025) find the initial boost in employment they observe is driven by this sector, with larger impacts for natural hazards causing more damage. Similarly, Walls and Wibbenmeyer (2023) find positive impacts on construction employment growth for up to six quarters after the fire.

In contrast, we expect employment and earnings to be negatively impacted in most other industries, at least in the short run, as a result of individual negative consequences of natural hazards, which is reflected in the literature. Belasen and Polachek (2008) find negative or no impacts on earnings and employment growth in manufacturing, trade, transportation, utilities, finance, investment, and real estate but positive impacts on employment and earnings in the service sector. Walls and Wibbenmeyer (2023) find negative or no effects on employment growth in all non-construction industries, including the service sector. Tran and Wilson (2025) do not separately examine earnings and employment impacts for industries other than construction. Our work examines earnings and employment outcomes in each of these sectors by natural hazard type for the most severe natural hazards to help reconcile existing findings and gain further understanding of how natural hazards shape local economies.

Our work also builds on our understanding of the heterogeneity of impacts by race and ethnicity, an area where there has been even less research. One notable exception is an article by Jeffrey A. Groen and colleagues (2020), who use the Longitudinal Employer-Household Dynamics (LEHD) to examine the impacts of both Hurricane Katrina and Hurricane Rita on individual outcomes. Overall, they find that in the short term, earnings of affected individuals suffer, but in the long term, earnings of affected individuals outpaced the earnings of individuals in the control sample. They include evidence that these earnings gains were a result of wage growth in affected areas relative to control areas, driven by limited labor supply and growing labor demand, particularly in sectors tied to rebuilding. When looking separately at outcomes for Black workers, they find that in the short run Black workers experience larger earnings losses than White workers and in the longer run, though both White and Black workers experience gains in wage growth; Black workers experience less wage growth than White workers. Our study is the first to estimate heterogeneous earnings and employment impacts of a wide range of natural hazards across counties in the US for Black and Latino workers who overall have less household wealth and therefore less of a buffer against the economic hardships brought on by climate-related disasters (Ganong et al. 2020).

Overall, our work extends existing knowledge by examining a wider set of natural disasters, studying a wider set of industries, and investigating whether there are heterogeneous effects among economically vulnerable populations, specifically Black and Latino individuals.

DATA AND METHODOLOGY

Our analysis draws primarily on two datasets. One is a publicly available data source, the QWI. The other is a proprietary dataset, the SHELDUS, currently supported by the Center for Emergency Management and Homeland Security at Arizona State University (ASU).

The QWI contains county-level data on employment, wages, hiring, and layoffs by industry type and worker demographics from 1990 for some states, through the present. These data are generated from unique job-level data, the LEHD, that link workers to their employers. For this reason, these data are available by race and ethnicity, in addition to county and industry. We use data on total county-level employment and earnings, as measured at the start of each quarter. Our baseline sample includes 3,109 counties from the 48 continental states plus Washington, DC. Of these counties, 849 are metropolitan counties.

In tables 2 and 3 we present descriptive statistics drawn from the QWI on employment and earnings. Table 2 shows that across all counties and year-months between 2000 and 2022, the average county had 42,321 workers, who earned on average $3,698 per month (in 2022 dollars). We see that the highest industry-level employment at the county level is in the health care and social assistance category, and the highest average earnings occur in the utilities sector. Table 3 presents earnings and employment by race and ethnicity. We see that Asian workers have the highest average earnings per month ($4,509) among the identified racial groups, and White workers comprise the largest share of workers in these data. Average monthly earnings in a county by racial and ethnic groups are also presented in table 3. Here we see that Asian workers have the highest average earnings and White workers are just below. The average earnings are lowest for Black and Hispanic or Latino workers.

SHELDUS is a county-level hazard dataset for the country beginning in 1960 that covers disasters such as thunderstorms, hurricanes, floods, wildfires, and tornadoes, in addition to perils such as flash floods and heavy rainfall. These data include information on the type, date, and county of an event as well as the direct losses caused by the event (property and crop losses as well as fatalities). We then assign each hazard to a quarter to align it with the quarterly QWI data. We have acquired SHELDUS data through 2022.³

We present some descriptive statistics on the SHELDUS data in tables 4, 5, and 6 and figures 1, 2, and 3. In table 4 we show that our data cover sixteen different types of natural hazards, with the most common category being wind events, followed by flooding. Figure 1, panel A shows the distribution of each of these events over the course of our study period. Table 5 presents descriptive statistics on the three types of damages in our data: property damages, crop damages, and fatalities. The average property damage is almost $3 million, but this high average is driven by a few events that caused severe damages, the most significant one being Hurricane Katrina. When looking at crop damages, we see that the averages are already much lower. When looking at fatalities, we see that the county experiencing the largest loss from a natural hazard, lost 638 people, again, attributable to Hurricane Katrina. In table 6 we classify hazards by the amount of property damage caused. We create five different categories and for the remainder of our analysis we focus on those hazards which we classify as extreme hazards. To create this cutoff we searched for the closest round number that included at least the top 2 percent of hazards, which we found to be $10 million in property damages. Figure 1, panel B shows the distribution of these hazards over our study period and we see a significant amount of variation both in the number and type of hazard over the study period. Figure 2 presents a tree diagram of property damages for extreme hazards, showing that flooding causes the most amount of damage, followed by hurricanes or tropical storms and then wildfires, tornadoes, and hail. Figure 3 shows that for these extreme hazards we see variation within states as well as across states, with variation both in terms of the intensity of the damages caused by these hazards as well as the type of hazard.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 1.

Climate Hazard Counts by Type, 2000–2022

Source: Authors’ calculations from SHELDUS data.

Note: Seismic activity includes earthquakes, tsunamis, and volcanic eruptions. The optimal way to view this figure is in color. We refer readers of the print edition of this article to https://www.rsfjournal.org/content/12/4/134 to view the color version.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 2.

Extreme Hazards Property Damage Shares by Hazard Type, 2000–2022

Source: Authors’ calculations from SHELDUS data.

Note: Seismic activity includes earthquakes, tsunamis, and volcanic eruptions. Damages are measured in real 2022 dollars. The optimal way to view this figure is in color. We refer readers of the print edition of this article to https://www.rsfjournal.org/content/12/4/134 to view the color version.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 3.

Minimum and Maximum Property Damages by State of Extreme Hazards, 2000–2022

Source: Authors’ calculations from SHELDUS data.

Note: Seismic activity includes earthquakes, tsunamis, and volcanic eruptions. The optimal way to view this figure is in color. We refer readers of the print edition of this article to https://www.rsfjournal.org/content/12/4/134 to view the color version.

To shed light on how natural hazards shape the local labor market immediately after the event, we start with a set of descriptive analyses. We then perform a more systematic examination by following Tran and Wilson (2025) who model the impacts of natural hazards on aggregate county-level earnings and employment as impulse response functions by using a panel-data variation of Òscar Jordà’s (2005) local projection method:

In this model, Y_c,t represents our two outcome variables—employment and earnings—in county c, at time t. We measure employment at the quarterly level as the number of workers who are employed in a county at the beginning of the quarter. We measure real earnings in 2022 dollars as the quarterly average of average monthly earnings per worker. Both outcome variables are logged. The dependent variable y_c,t+h – y_c,t–1 then measures the cumulative differences in the outcome variables from a baseline quarter (t−1) over a horizon of h quarters, where h spans from twelve quarters (three years) before the hazard event to twenty quarters (five years) after the hazard event. A separate regression is run for each value of h. The coefficient of interest is β^h, which measures the relationship between the occurrence of an extreme hazard event in county c at time t—as represented by the indicator variable D_c,t—and the cumulative change in the outcome at time horizon h. Using the log form of the outcomes variables means we interpret β^h as a percent change in the outcome variable relative to the baseline value.

The second term on the right-hand side of the regression model represents a series of indicator variables that control for whether any other natural hazard with nonzero property damages occurred p quarters before the extreme hazard event. We choose twelve quarters as the value of p. We define the pre-disaster period as the twelve quarters (three years) preceding the event because this window provides a stable baseline against which to measure post-disaster deviations, capturing typical economic dynamics while avoiding confounding from earlier hazard events. A three-year period also aligns with the temporal scale used in prior disaster economics studies (Belasen and Polachek 2008; Strobl 2011) and provides sufficient data points for testing pre-trends in our event-study framework.

The third term is a continuous variable that controls for how many additional climate hazards with nonzero property damages occurred in county c between the extreme hazard of interest at time t = 0 and the estimation period t + h. The variables α_c and α_q represent county and quarter fixed effects, respectively. Including county fixed effect controls for any non-time-varying county characteristics, while including quarter fixed effects controls for the cyclicality of earnings and employment across quarters. We also control for region by quarter fixed effects (α_r(c),t) to account for any regional trends in employment or earnings that may occur simultaneously with the climate hazard horizon. To identify heterogeneity of these relationships by disaster type, industry, and race and ethnicity, we perform subgroup analyses of the same estimation model.

We split our data into two groups, counties with an extreme hazard and those with a less severe hazard (we drop four counties that have no hazard during the full study period). We then adjust our data so that time zero is the date of the first extreme hazard, or other moderate hazard in that county for the remaining counties. We then examine mean employment and earnings three years before and five years after the hazard.

RESULTS

We begin by presenting uncontrolled means for both employment and earnings in figure 4, separating extreme hazards from all other hazards. In panel A we see that on average employment growth seems to flatten after a county experiences an extreme natural hazard, but we see no change in slope in counties experiencing less severe hazards. When looking at earnings, in figure 4, panel B, it appears that there is no change in slope after a county experiences a natural hazard, either of the extreme type or one that causes less damage.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 4.

Means of Outcome Variables over Time by Severity of Climate Hazard Event

Source: Authors’ calculations from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

We next present results from our regression models. We begin with aggregate results and then present results by hazard type, industry, and race and ethnicity. We present aggregate results for employment in figure 5, panel A. We see here that in the first few quarters after an extreme hazard, employment drops before experiencing a small upward trend, relative to previous trends, but that by the three-year mark it appears that employment growth slows, relative to trends previous to the hazard. These results appear to persist up through five years post hazard. Our estimates show that three to five years post natural hazard, employment growth declines by 0.5 percent. These results are economically meaningful and in line with the literature, which could suggest a slow and perhaps permanent reallocation of labor away from the market following a climate hazard. Specifically, our results align with those of Belasen and Polachek (2008) who look at employment outcomes post hurricane, as well as those of Coulombe and Rao (2025) and Walls and Wibbenmeyer (2023), who examine outcomes post fire exposure. The results are also similar to those of Tran and Wilson (2025) when looking at the five-year time horizon for all hazards. However, Tran and Wilson observe possible gains in employment when looking further out as areas have more time to recover from adverse climate events, and when focusing only on extreme hazards. Given this long time period, the causal link between the climate event and the hazard itself becomes less clear, and thus the majority of the economic literature, including our analysis, focuses on the three- to five-year period post hazard.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 5.

Impulse Response Functions for Employment and Earnings

Source: Authors’ calculations from model (1) using data from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

In figure 5, panel B, we see a slight, but sustained, increase in earnings post hazard. When examining each subsequent layer of heterogeneity, however, earnings trends seem largely unchanged by the occurrence of an extreme hazard. For this reason, we focus the remainder of the analysis on results for employment. We include results for earnings in appendix tables and pull out a few key results that appear to be driving the main earnings result in figure 5, panel B.

Figure 6 presents results by hazard type. Here we see that trends look very different for each of these disaster types. Specifically, we see that in the example of Hurricanes and Tropical Storms, there is a decrease in employment in the immediate aftermath of the hazard, but employment actually increases over the course of the three to twelve quarters post hazard before leveling out to pre-hazard levels. These post-hurricane and tropical storm gains are also realized in earnings as sustained positive earnings growth for the full twenty quarters (see online appendix figure A.1).⁴ No other hazard type exhibits positive earnings growth, leading us to believe the overall earnings results in figure 5, panel B, are driven by the results for hurricanes and tropical storms. We hypothesize that this increase in employment and earnings is driven by federal and local investments resulting from the occurrence of named and damaging storms. In contrast, when looking at storms and floods that are not classified as hurricanes and thus possibly do not receive the same type of aid, but in this case also cause over $10 million in damages, we find no significant increase in employment post hazard. Surprisingly, we see no evidence of employment impacts post wildfires in contrast to studies by Coulombe and Rao (2025) and Walls and Wibbenmeyer (2023). We also see no impacts post hail or tornadoes. We do, however, find some evidence that suggests build back better may describe results post seismic activity.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 6.

Impulse Response Functions for Employment by Hazard Type

Source: Authors’ calculations from model (1) using data from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

To gain additional insights into the mechanisms driving these findings, we examine employment impacts by industry, presenting results in figure 7. Not surprisingly, we find the increases in employment immediately following a hazard appear to be driven by changes in construction employment. We also see some evidence of elevated churn in the real estate industry, but these impacts are not statistically significant. We do, however, see longer-term declines in both the accommodations and food services industry as well as the health-care and social assistance industries. We do expect to see these longer-term declines in accommodations and food services as places impacted by natural hazards may become less attractive for tourists in the years following a hazard. Results in the health care and social assistance sector appear more surprising and perhaps could be a result of longer-term losses in local budgets that result from the hazard.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 7.

Impulse Response Functions for Employment by Industry

Source: Authors’ calculations from model (1) using data from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

Finally, figure 8 presents results by race and ethnicity. Here we see that the slight increase in employment that occurs in the three to twelve quarters after a natural hazard is primarily experienced by Hispanic households. This could be driven by the fact that they are overrepresented in the construction industry. Three years after an extreme hazard, however, Hispanic workers begin to experience declines in employment that eventually become as large as the declines experienced by all other groups. In contrast, we see little medium-term employment impacts for Black workers, though patterns do follow national trends with a small decrease in the immediate aftermath of the hazard. These results suggest that shorter-term benefits associated with hazards are relatively concentrated, whereas the longer-term costs appear to be more spread out over these different groups, though overall patterns do look similar for each racial and ethnic group.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 8.

Impulse Response Functions for Employment by Race/Ethnicity

Source: Authors’ calculations from model (1) using data from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

FEMA AID AS A MECHANISM

To gain some insight into the impacts of public assistance on county-level employment patterns post-natural hazard, we have matched the extreme hazards we identify in the SHELDUS data with the FEMA Presidential Disasters Declarations (PDD) dataset, using FEMA OpenFEMA data. This publicly available dataset is used to track when and where federal assistance is authorized in response to a major disaster declaration issued by the US president. We matched extreme hazard events from these two datasets by county, year, and quarter. For example, if an extreme hazard occurred in the same county and year in both datasets and within a one-quarter window before or after, we considered the extreme hazard in the SHELDUS dataset to be a FEMA declared presidential hazard. We present the results of this merge in table 7. We see that overall 64.5 percent of our extreme hazards from the SHELDUS data are identified as FEMA PDDs. We use these matched disasters as a proxy measurement of natural hazards that receive significant aid. Specifically, we are going to argue that an extreme hazard in the SHELDUS dataset likely received FEMA aid if it matches to a hazard in the FEMA PDDs dataset. If there is no match between the two datasets, we are going to argue that the SHELDUS extreme hazard did not receive FEMA aid. Since we do not have the actual aid dollars received, our second-best alternative is this proxy.

Our strongest predictions about how FEMA aid impacts employment or earnings are through increased investment in the construction industry to build back after a natural disaster. To examine this relationship, we then stratify our sample by whether or not the SHELDUS extreme disaster was identified in the FEMA dataset and estimate the impulse response functions by industry separately for FEMA-supported disasters and those that did not receive FEMA aid. We present these results in figure 9.

• Download figure
• Open in new tab
• Download PowerPoint

Figure 9.

Impulse Response Functions for Employment by Industry and FEMA Support

Source: Authors’ calculations from model (1) using data from the Quarterly Workforce Indicators, Longitudinal Employer-Household Dynamics, US Census Bureau, and SHELDUS.

Note: Earnings are quarterly averages of monthly earnings, adjusted for inflation and reported in 2022 dollars.

When looking at results within the construction industry, we do see clear differences in employment responses when disasters are FEMA declarations versus not FEMA declarations. The initial bump observed within construction is only apparent in the FEMA declarations, not in the disasters that are not matched to FEMA. This provides some suggestive evidence that initial employment growth is tied to assistance. Additionally, the longer-term decline in employment is only observed in the counties with no FEMA disasters, again suggesting that aid may be a central part of the story. Looking next at the real estate industry, results in the FEMA and no FEMA categories do not appear statistically different from one another, but there are slight differences in trends that perhaps align with theory here. We do see slight increases in employment in real estate in the three- to five-year window after a hazard for FEMA declarations, but no such impact in the non-FEMA hazards. Again, this provides some suggestive evidence that aid does bolster local economic recovery, particularly in the industries where aid is directed. In both the accommodations and health-care industries, we do not see much of a difference in trend in matched versus unmatched hazards, which is perhaps not so surprising as the aid would not be directed toward these industries.

CONCLUSIONS

On average climate-related natural hazards appear to reduce employment growth over the longer term at the county level and have slight but sustained positive impacts on earnings. When examining results by hazard type, we find increases in employment one to two years following hurricanes and then a return to baseline, while we find steady declines in employment following severe storms and floods. We find no impacts on employment or earnings following hail, tornadoes, or wildfires. We find longer-term increases in employment following extreme seismic activity events with no impacts on earnings. This increase in employment is consistent with a build back better mechanism where areas that experience extreme levels of property damage due to seismic events may see an influx of people and capital during the rebuilding period. Previously mentioned work by Erda (2025) uses detailed micro-level firm data to examine how manufacturing plants respond to floods. Her work provides some evidence that surviving firms do invest in capital quickly and that federal disaster spending plays an important role here.

When examining results by industry, we find a post-hazard bump for the construction industry followed by longer-term declines. This is consistent with construction being a necessary industry for rebuilding after an extreme climate hazard event. We find some evidence of elevated employment one to two years out in the real estate industry that we hypothesize is related to turnover in sales and rentals of properties as some residents decide to rebuild while others decide to sell and relocate after a hazard event. For health and accommodation related industries, we see declines in employment both immediately post-hazard and over the longer term. Our analysis of FEMA-supported hazards compared with those that did not receive FEMA aid underscores these industry results by demonstrating that the bump in employment in the construction industry (and suggestively in the real estate industry) is driven by those counties that received FEMA aid.

Lastly, when looking at results by race and ethnicity, we find increases in employment (as well as earnings) for Hispanic and Latino workers followed by longer-term employment declines. We attribute this to the high concentration of Hispanic and Latino workers in the construction industry (US Bureau of Labor Statistics 2023), which sees increases in employment post-hazard, with eventual declines. For other racial groupings, we see very little change one to two years after an extreme hazard but larger declines in employment in the longer term.

Our research is important for informing policy responses to climate-related disasters because a large sum of federal dollars is allocated to aiding individuals impacted by disasters each year, and there is evidence that federal emergency aid is exacerbating wealth inequality. Junia Howell and James R. Elliott (2019) find that economic damages from natural disasters increase wealth inequality in counties that experience the disaster, and critically, they find that federal emergency aid further exacerbates that inequality. They argue that this inequality may occur because the majority of funding for disaster recovery is allocated toward property owners. They highlight that for less privileged residents and non-property owners, hazards may lead to increased financial liabilities that result from an increasing risk of losing one’s job (Elliott and Pais 2006), an increased chance of mobility (Elliott and Howell 2017), and increased rents that arise from the reduced availability of housing stock (Vigdor 2008).

Our results also highlight key levels of heterogeneity, which may contribute to inequality in the recovery from extreme natural hazards. For example, we see declines in employment in two key sectors, accommodation and food services, as well as health care and social assistance. The most vulnerable workers in these sectors may be the most impacted by these hazards. More work is needed to identify which workers are impacted within these sectors. In future work, we will rely on individual-level employment and earnings to more clearly identify these potential pathways. By focusing on individual workers, we will be able to learn much more about distributional heterogeneity of longer-term impacts of these climate hazards, which may reveal disparities in vulnerability and recovery pathways. Additionally, we will be able to identify whether workers displaced by hazards are more likely to reallocate across sectors or locations, which is not observable in these aggregate county-level statistics. Overall, understanding these individual level patterns could further inform the impact of climate hazards on the relative distribution of earnings and employment, improving our understanding further on how hazards and our policy responses to them may be contributing to patterns of inequality.