The Impacts of Flooding and Business Activity and Employment: A Spatial Perspective on Small Business

1. Introduction

Flooding is one of the most common types of natural hazards and can occur across different types of geographies from inland to coastal areas. While scientific knowledge, information technology, and accurate warning systems can limit fatalities, significant economic damage, such as to crops, businesses, and infrastructure, often occurs. This is particularly true when more significant weather events lead to larger floods. One recent example occurred in portions of the Midwestern United States (US) in 2019. This event named “The Great Flood of 2019”, occurred primarily along the Missouri River and affected over 13 million residents across 11 states. This flood disaster resulted in four fatalities and caused about US$ 6.3 billion in damage to property and agriculture.¹ While many industries in the region were disrupted, agricultural production in the Southern Plains states was significantly affected by major flooding and persistent heavy rainfall. Additionally, the flooding event’s extended duration reduced crop planting by millions of acres. There was also considerable infrastructure damage in many of the cities and towns across the region, and high water levels disrupted barge and other transportation traffic, negatively affecting many industries.

Major flooding events have both short-term and long-term impacts which may vary depending on the existing infrastructure, community planning, and the recovery effort. From a spatial framework, impacts caused by a major flood are likely greater than the initially observed direct losses within the flooded area due, in part, to longer-term changes in population migration and the allocation of public and private resources. It may be that industries dependent on larger labor pools, such as manufacturing or the service sector, are more flexible in terms of relocating after a major flooding event, relative to industries more dependent on natural resources or amenities in the affected region. If a local manufacturing or large service-related firm were to relocate after suffering significant flood-related losses, the result would be a permanent reduction in contributions made by such a firm to the local economy (e.g., to employment, labor income, or gross regional domestic product). On the other hand, the likelihood of a permanent loss of agriculture-related activity (e.g., agricultural services) resulting from the flooding may be relatively low in regions dominated by agriculture. However, agricultural land may have to be cleaned and restored, which may result in post-flooding economic losses due to delays in production activity.

This study augments the existing empirical literature on the impacts of natural disasters and regional economies by emphasizing the effects of flooding on local employment and business establishments in the aftermath of an event. However, in contrast to previous empirical analyses, we examine the relative impacts of a range of flooding magnitudes. Additionally, while many previous studies examine the long-run economic effects of natural disasters on economic activity using highly aggregated data, this paper focuses on the short-run and medium-run flood effects on US county-level employment and number of establishments. Hence, we examine flood impacts on the business employment and business survival at a more disaggregated level. To accomplish the study goals, we examine how business activity and employment respond to major and minor floods, both in the immediate impacted period as well as in subsequent years. We use county-level data for the period 1992–2012 to estimate changes in economic activity, including spatial interactions, where business activity is measured by alternative variables such as local employment by sector and the numbers of small and larger establishments. We employ a Spatial Durbin Model (SDM) to account for spatial dependence in the data.

2. Literature Review

According to the Intergovernmental Panel on Climate Change (IPCC), climate change may alter precipitation patterns in many regions worldwide and it is likely that heavy precipitation and its frequency will increase over most areas (Schröter et al.2005). As the recent IPCC report states, such events would adversely affect food production, water supply, and, hence, also potentially have a negative influence on human health. Disruption of business activity and property damage are also likely consequences for the industrial sector due to an anticipated increase in extreme weather.

The literature on the economic impacts of natural disasters spans from cross-country analyses to within-country subnational analyses to sectoral analyses, to household/business levels. In our review, we briefly summarize the most salient researches from the cross-country works. We then turn our attention to several key studies that use subnational and sectoral data from the US.

2.1. Economic impacts of disasters

There are now a number of cross-country studies evaluating the short- and longer-run macroeconomic impacts of disasters. The general findings in the empirical literature on the impact of disasters on economic output can be summed up as follows. As a direct effect, natural disasters can destroy or impede the factors of production — labor (e.g., Anbarci et al.2005; Kahn 2005; Halliday2006) and physical capital (Albala-Bertrand1993). These direct impacts cause business interruptions within the affected establishments and set off additional indirect effects on companies up- and down-stream in the supply chain (Rose2004). The aftershock period can follow several paths that have been simulated in a numerical model as shown by Tol and Leek (1999). If lost capital is not replaced, then the level of production is permanently reduced. If the lost capital stock is replaced, while the output might drop in the immediate aftermath of the event, it is expected to increase at an even higher rate in subsequent periods. Economic researchers (Crespo-Cuaresma et al. 2008; Okuyama2003; Skidmore and Toya2002) suggest that the primary impetus for the potential rise in long-run output comes from updates in technology and/or factor composition. Furthermore, several authors highlight the possibility of positive employment effects (Ewing et al.2003, 2007). The magnitude of a disaster impact is not solely determined by the level of the disaster itself (Richter scale, water level, etc.) but also by company-specific factors such as investment strategies (Tol and Leek1999; Skidmore and Toya2002), factor composition, the level of technology (Crespo-Cuaresma et al. 2008), and disaster relief (Sobel and Leeson2006; Shughart2006).

Okuyama (2003) put forward the idea that older capital stock is more vulnerable to natural disasters and suggested taking into consideration capital productivity when analyzing the short-term effects of disasters on production. A few scholars emphasized short-run effects in their empirical analyses and mostly focused on a single aspect such as employment (Ewing et al.2007) or productivity (Raschky 2007). In an analysis of the effects of sudden disaster shocks, Hallegatte et al. (2007) point out the caveats of applying the standard Solow economic growth framework with its long-run perspective when an area suffers from a severe natural disaster. For example, even if the probability of the occurrence of extreme weather is small, ignoring the existence of the possibility of such events in the Solow growth model will lead to inaccurate predictions. This is so because disaster losses cannot simply be replaced using average cost or other economic values, not to mention the follow-up impacts and various economic and policy challenges involved in the restoration process.

Murlidharan (2003) shows in various simulations that a change in capital assets depends on the time elapsed after a hazardous event has occurred. This evaluation supports the notion that short-run and long-run analyses may come to different conclusions regarding the impacts of disasters on firm performance. By controlling for different levels of the exposure of capital assets to flooding and distinguishing between tangible and intangible capital, Leiter et al. (2009) focus on the short-run flood effects on European firms by using company-level information in flooded regions. Their findings show that the growth of firm assets and employment was affected by flooding in the short term, and that these impacts are also positively correlated with the share of intangible assets.

2.2. Evaluation of disasters using subnational and sector data

Turning to subnational analyses of disasters in the US such as those by Corey and Deitch (2011) and Xiao and Van Zandt (2012), research points to business survival following disasters as being important to long-term community recovery. Xiao and Feser (2014) use time-series techniques to identify the impacts of the 1990 US Midwest Flood on unemployment in affected counties. They find that the flood caused a significant spike in unemployment, but the effect was temporary. Community recovery also depends on the degree of physical damage and disruptions to critical services such as power and utilities (Lam et al.2009; Webb et al.2000, 2002). The impact of disasters within a community varies by industry. Restaurants and retail outlets rely on local customers and are thus very dependent on a quick community recovery (Tierney1997; Chang and Falit-Baiamonte2002; Webb et al.2002). On the other hand, as shown by Belasen and Polachek (2008, 2009), Chang (2010), Corey and Deitch (2011), Brown et al. (2015), and others, construction and engineering sectors may thrive as rebuilding ensues. Lee (2020) used a survey approach to examine business recovery patterns in different sectors in Aransas County (Texas) and Monroe County (Florida) following the Hurricanes Harvey and Irma. Lee documented how industries that support rebuilding activities during recovery experience increased activity, whereas restaurants and personal services tended not to fully recover until the broader community was fully functional. In other words, some industries may benefit in the immediate aftermath of a natural disaster while others are hurt and may take longer to recover.

A number of studies demonstrate the importance of considering the spatial aspects of economic impacts and recovery. For example, LeSage et al. (2011) evaluated business reopening in New Orleans following Hurricane Katrina, documenting the importance of spatial interdependence of reopening decisions. Lee (2021) used a duration model with spatial effects to examine business reopening decisions in Southern Texas following Hurricane Harvey. Lee highlights the role of social networks and spatial relationships in decisions to reopen following the hurricane. While there is a large body of literature that demonstrates the importance of spatial relationships within industry sectors [some examples include agriculture (Debolini et al.2013), hospitality (Lee et al.2018), and manufacturing (Xue et al.2020)], the studies cited above illustrate the role of spatial relationships specifically in the recovery process. In the context of our research, the scope is at the county level rather than the business level. Nevertheless, counties contiguous to flooded areas may experience a different impact relative to the directly affected counties. For example, businesses in neighboring counties that assist in recovery could experience significant increases in the activity following disasters. Similarly, while closed restaurants may suffer in the directly affected areas, would-be customers could very well shift demand to neighboring county restaurants. Thus, it is both useful and important to incorporate spatial relationships into our evaluation using appropriate spatial econometric tools.

2.3. Business activity impacts

Small businesses are more vulnerable to natural disasters compared to larger establishments due, in part, to their inability in the short term to quickly adapt to extreme circumstances (Davlasheridze and Geylani2017). Further, there are concerns that increased severe weather events from growing climate volatility will disproportionally and negatively impact small businesses as a result of the higher degree of natural disaster vulnerability. While federal and state governments emphasize local community preparedness to help address this increased risk, there may also be polarization between community members and local authorities in terms of implementing disaster protection measures (Henderson et al.2020). At the same time, it is necessary that all community stakeholders, including local small businesses, are engaged in community resilience plans for such plans to be effective (Begg et al.2015).

Much of the literature exploring the impact of natural disasters on small businesses relies on post-disaster surveys to better understand factors that contribute to small establishment resiliency (Danes et al.2009; Marshall et al.2015; Runyan2006). Key findings show that poor planning on the part of the small business, local area infrastructure issues, disruptions to the business cash flow, externalities from federal assistance, and barriers to access recovery-related capital negatively impacted establishment resilience in the wake of a natural disaster (Runyan2006; Wiatt et al.2021). Additionally, women-owned, minority-owned, and veteran-owned businesses appear more likely not to survive a natural disaster (Marshall et al.2015). However, establishments that were older, larger in size, with owners who had greater industry experience and owners with prior experience navigating natural disaster, were more likely to survive.

We only found a small number of studies that examined the longer-term impacts of natural disaster events on US small businesses while also considering a range of industries in the affected region (Xiao2011; Xiao and Drucker2014). Xiao (2011) and Xiao and Drucker (2014) both explored the longer-term effects from natural disasters, specifically the 1993 Midwest Flood, on the local economy. The overall findings were that the longer-term impact of flooding on local economies was negligible (Xiao2011), and this was especially true in regions with greater economic diversity (Xiao and Drucker2014). However, the effect on the agriculture sector was negative and for longer term for some communities (Xiao2011).

3. Theoretical Framework, Data, and Empirical Approach

Before introducing the data and our empirical strategy, we first discuss the theoretical framework to help guide our expectations for the evaluation. In principle, various natural disasters will have different impacts on consumption and production in a local economy; our focus in this examination is on production, including the effects on business establishments, employment conditions, and payroll status.

Figure 1 offers a summary of the paths by which a flood hazard can affect a local economy. A flood event may result in physical damage to the building stock, essential facilities, transportation systems, utilities, and agricultural products and vehicles. The initial damage results in induced physical damage and direct social and economic losses, which in turn lead to other indirect economic losses.

Our evaluation seeks to measure these losses in terms of business establishments and employment. Potential measures of business establishments are relatively straightforward and intuitive; there are two possible outcomes, shutting down or continuing to operate. The damages of company assets and production materials, locations, related industries, scale of firms, as well as governmental disaster relief policies, will have an impact on the vulnerability and resiliency of establishments. However, new demand may arise due to reconstruction and recovery efforts, prompting a short-term boom in some sectors.

Mapping changes in employment and payroll status is more complicated. The impact of flood events may vary across industry and time, and the analysis needs to consider the potential changes in supply and demand in the local labor market. Within a closed local economy, as illustrated in Figure 1, we assume a constant population and thus the supply of labor can be viewed as constant in the short run. Employment shocks faced by flood-affected industries, such as agricultural services establishments, some types of manufacturing companies with damaged properties, as well as restaurants and other services, lead to reduced employment demand. This will generate a new labor market equilibrium, leading to a potential reduction in employment in those sectors. However, as mentioned earlier, there may be short-term demand growth for some occupations in the post-disaster reconstruction process, such as road repairs, house repairs, and industries that may be involved in post-disaster reconstruction. We seek to identify the direct, indirect, and induced effects.

Extending the model to an open economy environment, migration and relocation of enterprises enter consideration. In addition, we also consider potential spatial spillover effects to other communities, both positive and negative, which can be incorporated into the conceptual illustration (Figure 1). Within this broader framework, we examine flood impacts on the number of business establishments and employment by sector in flood-affected counties as well as in neighboring counties using spatial econometric tools. We model the number of establishments and employment using a reduced-form regression estimation where the number of establishments and employment are a function of flood events and a variety of control variables including economic, demographic, and local governmental factors.

3.1. Data and model specification

The dependent variables on business activity, which include number of establishments in different industries (agricultural services, manufacturing, broader service, etc.) of different sizes and employment information, come from the County Business Patterns (CBP) provided by the United States Census Bureau. Specifically, for local establishments we have also carried out a detailed classification according to the size of the company and the industry to which it belongs, paying special attention to small establishments such as those with 1–9 employees and 10–49 employees.² While our establishment data do not include farming, agricultural services are included which may provide some insights to impacts on farming and related agriculture production.

The US county-level flooding data come from the National Oceanic and Atmospheric Administration (NOAA) dataset of storm events. We divide flood events into two categories, major and minor flooding, to characterize the floods in a county. The threshold for determining major versus minor floods is US$ 5 million.³ We control for property tax burden as well as socioeconomic characteristics. Property tax burden is measured as the property tax revenue as a percentage of total tax revenue. The total property tax revenue and total tax used to calculate the property tax burden are from the United States Census Bureau’s county finance package. These data include government tax revenue information from 1992 to 2012 (with a five-year gap), and our constructed $r$ ratio allows us to manage potential endogeneity caused by tax variables without using an instrumental variable regression approach.⁴ The county-level personal income data come from the Bureau of Economic Analysis. Data on other demographic variables (the share of population aged 20–64, the share of population aged 65 and over, and the share of the non-White population) all come from the United States Census Bureau. Variable definitions, data sources, and summary statistics are provided in Tables A.1 and A.2, respectively.

As discussed in our review of the literature (i.e., Lee2021; LeSage et al.2011), our investigation includes a spatial analysis. LeSage (2014) recommends using a spatial model with global spillover effects, namely the SDM, when there are potential reactions over time and changes in behavior (i.e., interaction effects that are endogenous) due to an event, and that such changes may traverse the geographic unit in question. In our case, this is the county level of observation. Diagnostics for spatial dependence show strong spatial autocorrelation of residuals based on Moran’s I scores. Also, the Lagrange Multiplier (LM) test shows the presence of both types of spatial dependence, so the spatial error model and the spatial lag model are applicable for our evaluation.⁵ Thus, the SDM was adopted to examine the effects of the flood events on business activity. Given the guidance from LeSage (2014) and our diagnostics, we constructed an SDM specification for our regression analysis:

Here, ${\theta }_k$ is the exogenous spatial lag parameter and $W_{ij}$ is the spatial weight matrix, which together are used to generate the spatial lag variable. Also, $ln{Y}_{it}$ represents local business activity and includes five measures: total business employment, total number of establishments, number of establishments with 1–9 employees, number of establishments with 10–49 employees, and number of establishments with 50–99 employees. $X_{it}$ represents our key explanatory variables (major floods and minor floods) and the remaining control variables (property tax burden, per-capita income, and demographic variables). For the SDM model, the total effect is the sum of the direct and indirect effects. However, the values of direct and indirect effects now depend on three parameters as specified in the SDM model: the exogenous parameter, ${\beta }_k$; the endogenous spatial lag parameter, $\rho$; and the exogenous spatial lag parameter, ${\theta }_k$. The model includes county-level fixed effects (${\mu }_i$) and temporal effects (${\gamma }_t$). Finally, ${ϵ}_{it}$ is the error term.

The estimated coefficients in the SDM model, with a spatial lag of both dependent and independent variables, are a combination of the direct and indirect effects. Direct effects are the effects on the county experiencing the major or minor flood event. Indirect effects are the spillover effects of these flood events on other neighboring counties. See Elhorst (2009) for a description of the interpretation of the direct, indirect, and total effects in the SDM model. With the SDM model, the direct and indirect effects of each of the explanatory variables depend on the coefficient estimate of the spatially lagged value of each explanatory variable. This means that there are no restrictions imposed on the magnitudes of the direct and indirect effects, and thus the ratio between the direct and indirect effects can be different for each of the explanatory variables.

3.2. Results

In Table 1, we report our regression results for all sectors aggregated together. As discussed in the previous subsection, we use SDM in all the regressions in which we also control for both county-level and temporal fixed effects. In Tables 2 and 3, we show regression results for the agricultural services and broader services sectors, respectively. Both sectors had statistically significant findings, which was not true for other sectors considered in our preliminary work.⁶ The results tables include direct, indirect, and total effects, and the dependent variables used in each model are abbreviated in the title, to include employment, total establishments, the number of establishments with 1–9 employees, and the number of establishments with 10–49 employees. Note that we also include estimates of lagged major and minor flood events. Thus, we have an overall estimate of flood impacts plus some indication of the dynamics of those impacts. Generally, the initial effects of flooding can be negative until the flooding dissipates and economic activity resumes. Once the activities resume and funds from insurance, government, etc. flow in, positive economic effects can emerge. However, when the resource inflow slows, economic activity can slow down as well. Given that we have annual data, we cannot precisely map out the short-run dynamics. For example, suppose there are two floods where one occurs in December of one year and the other in January of the next. The impacts of those two floods will be different in terms how they are reflected in the annual data. For this reason, we do not focus too closely on the lagged effects, but we do offer a limited, more specific discussion in the context of agricultural services.

We focus our attention first on the major and minor flood coefficients. The results in Table 1 (all sectors combined) suggest that the net effect from flooding had little immediate or longer-run overall impact on our business activity measures. However, this finding may be misleading in the sense that certain sectors may be affected differently. In fact, we see mostly a positive relationship between minor floods and the business activity variables, where establishments with only 1–9 employees have equally significant positive correlations. The indirect effect shows that neither large floods nor small floods have a significant impact on local employment and establishments. For the total effects, this positive relationship is consistently significant in all regressions except for the establishments with 10–49 employees. The significance varies for direct and indirect effects for major flood events, however, apart from showing that it will have a small and positive impact on the establishments with 1–9 employees, no significant effect was observed. Taken together, the results using data that include all sectors combined show that the total impact of the flood is positive but very small. Further, impacts tend to dissipate quickly over time.

Before turning to the analysis of the different business sectors, consider the coefficient estimates on the control variables. As expected, higher population is associated with greater employment and establishments. Somewhat surprisingly, higher per-capita income is associated with lower employment and establishments. One possible explanation is that places with rapidly growing wages may experience slower growth in jobs and business activity. Controlling for overall population and income, a greater female population, working age population, and elderly population is associated with less employment and fewer establishments. Lastly, greater reliance on property taxation is associated with lower employment and establishments. While we do not discuss these results in detail, generally these coefficient estimates are in line with the expectations.

Turning to the flood variables in Table 2, we find a strong and negative direct impact of minor flood events on local agricultural services business activity. Major floods have a much greater negative impact on these establishments, particularly for smaller operations. While our data do not include actual farming operations, one could speculate that a similar effect may also occur for agriculture producers. Looking at the total effect and indirect spillover effects, we do not find a significant impact of flooding on the number of agricultural service providers. Given the land is immobile, agriculture is more vulnerable to natural disasters than some other industries and this is reflected in the effect on agricultural services. The lagged effects offer some additional insight. Specifically, the one-year lag estimates show an initial positive direct effect, which is then followed by a significant negative effect in the two-year lag coefficients. Note, however, that the overall direct effect is negative for both major and minor floods. One interpretation of these results is that while the overall effect of flooding is negative, there appears to be a temporary boost in the first year following the flooding, perhaps due to an inflow of funds from insurance, government, and other sources.

In Table 3, we see substantial changes in the service sector in the wake of flooding. The results of the direct effects suggest that both major and minor floods increase employment in the service industry in the short term and the number of service sector establishments also increases. For the indirect effects, the outbreak of major floods also appears to promote the service sector activity in the surrounding counties. However, minor floods are associated with a reduction in employment and the number of establishments in the surrounding areas. One possible explanation is that the impact of major floods is so substantial that service sector businesses in counties adjacent to flooded areas expand to meet the increased market demand. In contrast, minor floods do not have such a significant influence in terms of creating new market demand in the service sector. Possible reasons for service sector growth in adjacent counties may be the result of increased service needs after the disaster, or the increased demand for some services during the recovery process.

4. Conclusions

In this article, we examine how business activity responds to flood events. We use nationwide county-level data over the 1992–2012 period to examine the different impacts of floods on overall employment and the number of establishments as well as for different sectors. Specifically, we examine the impact of floods on local business activity within a panel data framework using spatial econometric methods to account for spatial dependence.

We find evidence of a positive relationship between minor floods and overall local business activity. While the most severe floods can be very costly and alter the long-run trajectory of economic activity in an affected region, our findings suggest that for most flood events communities are resilient in the sense that overall economic activity remains strong in the aftermath. However, an examination of subsectors offers a more nuanced evaluation. While the net effect of flooding on local business activity is positive but small, there is an offset between direct and indirect effects and the differential impacts across sectors. Specifically, while flooding appears to generate positive employment effects in the service sector, we find that the agricultural sector is negatively affected and that the impact is most pronounced for the large floods and among small establishments.

Our analysis also suggests that any negative effects are not long lasting. Further, our findings reveal important spatial impacts. For example, we find that the service sector in counties adjacent to an area that experienced major flooding tends to have an uptick in service sector activity following flood events. This finding suggests that neighboring counties provide needed services to the affected areas in the wake of flooding.

This analysis offers clear policy implications in that smaller agriculture service establishments are most vulnerable and thus, targeted short-term support from local, state, and national authorities is worthy of consideration. The evaluation also shows that regional economies are resilient in the sense that impacts tend to be short-lived and that the businesses in nearby counties provide needed services in the wake of disasters. Generally, this research adds to our understanding of how businesses and economic activity are affected by and respond to floods.

5. Acknowledgement

We thank the North Central Regional Center for Rural Development (USDA) for financial support. We also thank two anonymous referees and the editors for very helpful comments.

Notes