1 Introduction

Small businesses represent the backbone of the US economy, comprising 99% of all businesses nationwide and employing more than half the workforce (Yoshida and Deyle 2005). Small business owners invest significant time and resources to guarantee the success of their ventures. However, many are extremely vulnerable to natural disasters and approximately 25% never reopen following a major disaster (SBA 2014a). By default, their size makes them vulnerable to extreme events, but small businesses appear to be hit twice, both during and after the event (Runyan 2006). Lack of preparation ex-ante and limited ex-post resources, including loans and disaster assistance, often make it challenging for small businesses to survive in the aftermath (Webb et al. 2000; Alesch et al. 2001; Aldrich and Auster 1986; Schrank et al. 2013). An important factor often overlooked in business recovery studies is that return behaviors of individuals and businesses reinforce each other—businesses reopen if people return and people will likely return faster if businesses are operational. In turn, these private adjustments are influenced by public disaster response and recovery efforts, as well as public investment in disaster protection (Kousky et al. 2006; Boustan et al. 2012).

In this paper, we are interested in understanding the impact of subsidized small business administration (SBA) disaster loans (SBA 2014b), one ex-post recovery mechanism available for small businesses in the USA, as well as the effects of floods on businesses and the extent to which businesses use these infrequent events as a learning process to update their beliefs about risk. Low interest loans allow firms to repay and replace damaged or destroyed real estate, property, machinery or equipment, inventory, and other business assets. Households also qualify and receive SBA disaster loans if disasters destroy their property that is un- or under-insured. There are two channels through which businesses can potentially benefit from the SBA loans: (i) directly—low-interest loans allow businesses to replace damaged capital and inventory and thus rebound quickly; and (ii) indirectly—because SBA loans also help households, there could be an additional stimulus through private household rebuilding. The contradicting hypothesis is that the SBA loans may create additional problems, such as increasing debt and further exacerbating business vulnerabilities (Dahlhamer and Tierney 1996). Examining the efficacy of this ex-post recovery loan program is important in understanding its likely implications in disaster recovery policy for the local economy. Moreover, studying the effects of current and historic weather shocks on small businesses not only helps us explore business vulnerability but also examine if these incidents help them form expectations over the probability of infrequent weather shocks. Furthermore, if this is true, how well do they learn through prior hazard exposure and subsequently adapt.

We use Blundell and Bond’s generalized method of moments (GMM) estimation strategy, which allows accounting for endogeneity in our interest variables, as well as correcting for model bias because of the nature of dynamic dependency. Allocation of SBA disaster loans could be endogenous because economically more exposed areas likely experience more losses and thus, receive more SBA disaster loans. There are several other factors explaining business survival in the aftermath of floods that could also be endogenous. These include population, wealth, financial market conditions, and sustained damages. Research suggests households’ return behavior in the aftermath can be partially driven by the expectation of whether and how fast businesses return (Kousky et al. 2006). Furthermore, because businesses represent engines of economy, their survival affects both the wealth and local financial market conditions. Ignoring these potential endogeneity concerns may bias estimation results.

Our results consistently indicate SBA disaster loans increase small business establishment in a county, while weather shocks have the opposite effect. At the margin, a US$1 increase in SBA loan per establishment results in four additional small business establishments in a county. These effects are primarily seen in metro and urban counties and appear significant for businesses employing fewer than 50 people. Businesses of all employment size appear to be extremely vulnerable to contemporaneous and historic weather shocks. We find that although the adverse effects of weather decays over time, the significance does not disappear for the most part. Furthermore, our results suggest learning processes through which businesses update their belief about uncertain flooding events could be delayed by as much as 5 years, indicating lagged and slow adaptation. Businesses in coastal counties generally exhibit relative resilience to weather shocks and rebound quicker. In the absence of disaster loans, as many as 134 small business establishments will be lost on average; therefore, out of 2492 small business establishments per county in the sample, close to 5% could be lost each year.

Existing research on economic impacts of disasters has disproportionately focused on individuals, households, and the public sector and subsequent adaptation policies.Footnote 1 Remarkably little attention has been paid to business disaster impacts and recovery aftermath (Zhang et al. 2009). The few studies that have examined the effects of SBA guaranteed bank loans (7(a) guaranteed loan and 504 loan programs) on aggregate economic indicators, such as employment and income, suggest these programs are effective.Footnote 2 However, unlike these two primary SBA loan programs, disaster loans are available only after large-scale shocks and are intended to alleviate immediate disaster burden and help communities’ quick recovery. Furthermore, their effects likely differ from the effects of general loans that can be obtained during relative calm. One can argue that disaster loans resemble federal disaster programs in that they are “transitory,” as opposed to other types of loans or traditional federal programs of a “permanent” nature (i.e., “stimulus package”). In theory, these types of temporary programs may be more or less effective than the money spent during normal times.Footnote 3

Our paper contributes to the existing literature in multiple ways. To the best of our knowledge, this is the first study to undertake a systematic approach with a national-level sample to empirically examine the effects of weather shocks and SBA disaster loans on small business survival. We explore the effects of various factors (i.e., population, wealth, cost of operation, and financial market conditions) in the dynamic GMM framework, which allows not only accounting for endogeneity bias but also correcting for bias because of the nature of dynamic dependency in the model. Second, to control for weather shocks, we use precipitation data and construct a county’s annual rainfall anomaly. Incorporating multiple lags of rainfall anomalies make our paper the first to explore the learning processes through past experience businesses use to update their beliefs about flooding events and adapt consequently. Incorporating prior exposure as a way to recover private learning processes to infrequent disaster events has been a common approach in understanding individuals’ hazard insurance purchase behavior (Gallagher 2014; Kousky 2016) as well as whether and how well regions and communities adapt to infrequent events and mitigate disaster risks (Sadowski and Sutter 2008; Hsiang and Jina 2014; Miao and Popp 2014; Davlasheridze et al. 2017). Our study further contributes to disaster economics research by comprehensively assessing small business vulnerability to disaster impacts and bridging business adjustments to individual responses, thereby offering policy prescriptions related to business survival and growth for areas affected by catastrophic flooding events.

The costs of natural disasters have significantly increased in recent years, both globally and nationally (Kunreuther and Rose 2004; Perry 2000). Floods represent the costliest natural hazards affecting almost all regions in the USA (Brody et al. 2007; Walsh et al. 2014). More recent climate models predict these impacts will likely worsen under future climate change (Allan and Soden 2008; Emanuel 2013; Villarini and Vecchi 2013; US Climate Change Science Program 2008). Understanding the implications of business disaster loans and general vulnerability of businesses to frequent disasters will further improve our understanding of the main drivers of community recovery in the aftermath of a catastrophe and the effectiveness of one of the most prevalent US government programs geared toward small businesses.

In the following sections, we provide a conceptual framework of the mechanisms that drive business return behavior in the aftermath of a disaster, describe the data used in the model, present the empirical estimation approach, and explain results and show their robustness and policy implications.

2 Conceptual framework

Conceptually, the decision structure of small businesses in disaster aftermath is determined by the extent of sustained damages; the ability to absorb losses through insurance, private savings, loans, and government assistance; and whether and how fast the customer base will return to impacted areas, particularly if these small businesses produce goods and services consumed locally.Footnote 4 Even if the local customer base returns, disasters also affect individuals’ consumption behavior (Anttila-Hughes and Hsiang 2013); therefore, short-term changes in consumption patterns may have implications for certain categories of small businesses. For example, during the rebuilding period, spending will likely increase in construction and related services sectors, while consumption of other services and goods, such as flowers, may decline—even if the local flower shop does not directly sustain physical damage.

These indirect and direct impacts of disasters ultimately determine whether a business remains open, returns after temporal intermission, starts anew, relocates, or shuts down permanently. These choices are quite different and likely influenced by different factors. Ideally, we would follow individual businesses, determine their post-disaster status, and study the choices they make. However, because we use aggregate-level data, we observe the ex-post outcome, representing the total number of incumbent businesses in a given location. These numbers reflect the consequence of choices made and capture the total of (a) new entrants and relocating businesses that leave another location and move into the reference location and (b) survivors of previous disasters for whom the expected net benefits of continuing operation outweigh disaster losses. Here, we focus on factors affecting entrants and survivors.

The magnitude of damages and loss absorption capacity are the main contributing factors for business survival. The former also may encourage new entrants who seize rebuilding and reconstruction opportunities in the short term. Loss absorption capacity of survivors, in turn, depends on post-disaster financial, infrastructural, and managerial resources and business skills; access to capital for recovery, including available low-interest rate loans and a favorable banking environment; available federal assistance programs to help repair damaged capital; and business characteristics such as size, location, age, ownership, prior exposure to disasters, and readiness to cope with disasters (Dunne et al. 1989; Evans 1987a, b; Hall 1987; Lee 2008).

Large businesses typically possess better infrastructural and financial resources and management skills that increase their survival chances. However, small businesses are vulnerable to cash flow interruption, lack access to capital, and have serious infrastructure problems that may hinder recovery (Runyan 2006). SBA’s disaster loan program is among the public programs for small business recovery in the aftermath of disasters. These federally backed loans have lower interest rates and longer terms than loans typically offered by lenders.

As a way to help communities rebuild, the SBA makes low-interest disaster loans available to eligible businesses and households (i.e., homeowners, renters, personal property owners) whose losses are under- or un-insured. The SBA determines the eligible amount by taking into account advances from existing insurance coverage on household homes or property, which are deducted from the estimated total damage. SBA business loans are typically physical and economic injury loans of up to US$2 million. SBA loans to households are up to US$200,000 and have a 30-year term and can be used to repair or replace damaged primary residence. Renters and homeowners can also qualify for loans up to US$40,000 to replace or repair damaged personal property (e.g., furniture, vehicle, appliances, clothing, etc.). Physical injury loans can be used to repair or replace the items such as real estate, personal property, machinery and equipment, inventory, and business assets that are damaged or destroyed.Footnote 5

Specific to loans given to businesses, the SBA defines whether a business has credit opportunities elsewhere and determines the interest rate based on this (SBA 2014a). Similarly, economic injury disaster loans are given to eligible businesses to provide working capital to meet their financial obligations and pay necessary operating capital expenses caused by disasters. Both businesses and households are required to purchase and maintain flood hazard insurance for possible future catastrophic events to be eligible for subsidized disaster loans. SBA also requires businesses and households to secure loans that exceed US$25,000 with collateral (real estate and other fixed assets such as machinery and equipment are commonly used as collateral for businesses; the first or second mortgage on the damaged real estate is an example of collateral used by households).

Whether subsidized disaster loans promote or deter business survival remains an open question. Dahlhamer and Tierney’s (1996) survey of small businesses after the 1994 earthquake in California suggested that SBA disaster loans make businesses worse off, potentially because of increased indebtedness. However, the impact of catastrophic hazards such as an earthquake may be different from the impacts of chronic and recurring floods. Research on individuals’ adjustment behavior indicates that frequent hazard exposure incentivizes ex-ante mitigation behavior (Sadowski and Sutter 2005, 2008). Communities may also view disaster as “a window of opportunity” and rebuild more resiliently, thus being better prepared for subsequent incidents (Landry and Li 2011).

Both survivors and entrants are encouraged in areas where banking markets are competitive, but the effect of the banking environment on businesses may be ambiguous. One main problem borrowers face in the small business loan market is misinformation that may lead to credit rationing. Loans can be allocated through mechanisms other than market clearing interest rates and lenders may fail to allocate loans efficiently through the market system (Stiglitz and Weiss 1981). If banks foresee small businesses at a higher risk of default, they may be reluctant to give loans. Guaranteed lending programs for small businesses (like the SBA loan) can help businesses establish closer relationships with lenders that would otherwise be costly, helping them evade credit rationing and further improving long-term loan opportunities (Craig et al. 2008). Another source of ambiguity is the uncertainty related to pricing of bank services and availability of services for small businesses (Bartik 1989). Although less competition in the banking market reduces loan availability through high interest rates, banks may give business loans based on personal connections—the service goal becomes more important than profit maximization.

A firm’s entry decisions may also be influenced by differences in costs and availability of inputs such as natural resources, labor, capital, energy, and infrastructure (Bartik 1991). Research regarding the economics of geography also suggests firms are attracted to the same location because geographical proximity generates agglomeration effects (e.g., the availability of specialized skilled labor, capacity and technological spillovers) (Head et al. 1995; Krugman 1991; Schmenner 1980, 1982; Schmenner et al. 1987). In particular, vertical movers (firms re-locating from cities to suburbs) are motivated by lower labor, capital, energy and infrastructure costs. On the other hand, horizontal movers (business re-locating between cities) are mainly influenced by availability of skilled labor, capacity and technology spillovers, indicating agglomeration economics (Burns 1977). Local fiscal policies and tax incentives also may influence new business preferences for specific locations. Although taxes are commonly believed to be disfavored by businesses (Bartik 1989; McGranahan 1998; Newman and Sullivan 1988), provision of public goods and improved public services funded through taxes tend to countervail negative tax effects (Gabe and Bell 2004).

Finally, productivity differences across firms also play a significant role for business entry, survival, and exit decisions. In turn, these differences are explained by firm size, age, ownership, location, firm-specific investments, and learning-by-doing (Baily et al. 1992; Bahk and Gort 1993; Olley and Pakes 1996; Bartelsman and Dhrymes 1998; Bartelsman and Doms 2000). Prior studies show that (1) productivity may increase as a firm gets larger; (2) in general, more capital intensive firms (e.g., large firms) tend to be more productive than their less capital intensive counterparts; (3) different investment types (e.g., replacement, expansionary, and retooling investments), irrespective of a firm size, also lead to productivity differences. For example, replacement investment allows firms replacing depreciated capital with a new, presumably more productive one, while retooling investment type allows them to adopt better technology and make major technological advances.

On the other hand, expansionary types of investments are typically undertaken for the attainment of more capital of the same technology firms use. However, although these types of investments expand firm size, they typically have minimal to no effect on overall businesses productivity (Cooper et al. 1999; Power 1998; Geylani and Stefanou 2013). Productivity differences across firms and industriesFootnote 6 may also widen because of learning and selection mechanisms—productive businesses tend to grow faster and survive compared to their less productive counterparts that will eventually shrink and fail (Jovanovic 1982; Hopenhayn 1992; Ericson and Pakes 1995; Melitz 2003; Foster et al. 2008).

Government subsidy programs could also play a role in business entry and survival decisions, in turn affecting the overall productivity growth in the industry. For example, R&D and various investment subsidies to incumbents and new entrants could help new and returning businesses with high entry costs and boost incumbent firm’s productivity. Low productive firms perhaps are the ones that are under- or un-insured and are likely to qualify for disaster loans in the event of large-scale floods.Footnote 7 Therefore, it is possible these loans help mediocre firms that would otherwise cease to continue operation in the aftermath. Because disruption caused by environmental disasters could stimulate growth and encourage innovation (Skidmore and Toya 2002), subsidized loans could be used to update outdated and less productive capital with more effective ones. Disaster economists often describe this as a “creative destruction” effect after Schumpeter’s creative destruction model (see Cavallo and Noy 2011). Leiter et al. (2009) directly examine the “creative destructiveness” hypothesis in a study dealing with flooding effects on capital, labor, and productivity within European firms. They find evidence that, in the short run, total assets and employment of firms within flood-affected regions exhibit positive growth.

Lack of firm-level productivity data does not permit us to explore potential implications of the SBA disaster loans on business survival through this lens. However, to the extent that large businesses are relatively more productive than the smaller ones, the models estimated in this study for different-sized businesses give us some initial understanding about the differential effects, if any, of SBA loans on firms with varying degrees of productivity.

Based on our conceptual framework, we empirically analyzed the aforementioned factors on business entry and survival decisions after flooding events using county business pattern data. Our main interest variable was SBA disaster loans. Because the SBA loan program’s success depends on the accessibility of loans and the regional banking market, we measured banking competitiveness by calculating the Herfindahl-Hirschman Index (HHI) for a county in our model. Exogenous and objective precipitation anomalies are used to capture the effects of physical hazard incidents and potential learning processes of hazard exposure on small businesses. In addition, we incorporate the extent of disaster impacts, proxied by flood-related economic losses (i.e., property and crop losses) into our empirical model. Business-entry decisions heavily depend on economic and regional factors such as income, customer base, availability of resource endowments, and regional differences in resource costs. We considered manufacturing wage rate as a labor cost to control for cost differences in regions, and used per capita income and population size to investigate whether wealthier and more populous counties attract more businesses.

3 Data

Data for this research were available from various sources. County business pattern data, obtained from the US Census Bureau, report the number of business establishments by employment size, including categories of 1–4, 5–9, 10–19, 20–49, 50–99, and 100 or more employees.Footnote 8 SBA disaster loan information was available from the SBA by the Freedom of Information Act. The data record total loan amounts disbursed to businesses and households in response to flood-related disasters by zip code and county from the years 1998 through 2010. Therefore, we limited our sample frame to the years 1998–2010. The distribution of SBA loans is highly skewed to the right and is distributed unevenly across counties (see Fig. 1). The average loan size in the sample was approximately US$48,081.71, with a median of zero.Footnote 9 The maximum loan size per county is approximately US$126 million. We divided disaster loans by the total number of business establishments, arriving at a final loan variable representing average loan size per establishment in a county. The sample average SBA disaster loan per establishment is US$32.16; with a maximum of US$52,454.76. This variable was log transformed using ln(loan + 1), which allowed us to keep observations with zero loans. This is important because, although some counties may experience disasters, businesses within them may not suffer much damage and thus, not be eligible for SBA disaster loans.

Fig. 1
figure 1

Distribution of the SBA disaster loans per establishment. Notes: This figure shows distribution of average SBA disaster loans per establishment over the sample time frame

To control for market structure in the banking sector at the local level, for each county, we created an HHI. Similar to Craig et al. (2008), we constructed HHI using Federal Deposit Insurance Corporation (FDIC) Summary of Deposit Data (SUMD). For each specified geographical area (state, county, city, and zip code), FDIC SUMD provides information from all FDIC-insured institutions. To calculate HHI for each county, we calculated the market share of each institution in the deposit market, then summed the squared market share of institutions within each county. The HHI ranges from 0 to 10,000 and takes into consideration the relative size and distribution of institutions in a market.Footnote 10 High HHI levels indicate highly concentrated financial markets and may suggest limited access to credit for small businesses. Average HHI in the sample is 2772.83; minimum is 418.23; and maximum 10,000.

To capture relative costs of starting up a business, we included a variable to measure the wage rate per manufacturing sector employee in the county. The demand side of the market was captured by the population and income size, represented by the log of per capita income. Both variables were available from the Bureau of Economic Analysis Regional Economic (BEA REIS) county personal income and population series. Sample average per capita income is 10.25 and the wage rate is 9.15 in log terms.

Instead of using presidentially declared flooding disasters as an indicator for a flooding incident, which presumably is endogenous to economic and political systems, we utilize more objective meteorological data to characterize the physical intensity of flooding conditions at the county level. Following the approach in Felbermayr and Gröschl (2014), we use the precipitation data from the National Climate Data Center (NCDC) Global Historical Climatology Network (GHCN) to create a “rainfall anomaly” variable.Footnote 11 We measure the precipitation anomaly by computing the number of standard deviations (Std_RAIN t ) of a county (i)’s annual precipitation in year t (RAINit ) from its long-run average level over the period between 1950 and 2000 (Mean_RAIN i ), as indicated in Eq. 1. Positive values of anomaly indicate excessive rainfall.

$$ {\mathrm{Rain}\_\mathrm{ano}}_{\mathrm{it}}\kern1em =\frac{{\mathrm{RAIN}}_{\mathrm{it}}\hbox{--} \mathrm{Mean}\_{\mathrm{RAIN}}_i}{{\mathrm{Std}\_\mathrm{RAIN}}_t} $$
(1)

In addition, we measured the extent of disaster impact using total economic losses per capita available from the Spatial Hazards Events and Losses Database of the United States (SHELDUS). The mean sample loss is approximately 0.50 in log terms; minimum is zero and maximum corresponds to 10.26.Footnote 12 Our final sample contains 31,982 observations. Summary statistics of all model variables are reported in Table 1.

Table 1 Summary statistics

4 Empirical approach

To examine the effects of SBA disaster loans on small business establishments, we estimated the dynamic panel data model specified in Eq. (2):

$$ {y}_{\mathrm{it}}={\alpha}_0+\gamma {y}_{\mathrm{it}-1}+\beta {X}_{\mathrm{it}}^{\prime }+\delta\ \mathit{\ln}\left({\mathrm{SBA}}_{\mathrm{it}}\right)+\sum_{t-5}^t{\mu}_{\mathrm{it}}{\mathrm{Rain}\_\mathrm{ano}}_{\mathrm{it}}+{\lambda}_t+{\lambda}_i+{\varepsilon}_{\mathrm{it}} $$
(2)

y it measures the number of business establishments of various employment size in each county i at time tSBA it represents SBA average disaster loans per establishment. δ represents the main coefficient of interest and measures the effect of SBA loans on the number of business establishments. Rain_anoit measures access rainfall events and include contemporaneous as well as lags of rainfall anomalies over the past 5 years. The sequence of multiple lags allows us to explore the extent to which private businesses update their beliefs about risk and potentially learn through prior exposure. \( {X}_{\mathrm{it}}^{\prime } \) corresponds to a vector of county-level variables that vary over time, including per capita total flood related economic losses, manufacturing wage rate, per capita income, size of population, and HHI indices measured in year t. λ t capture time-varying factors common across county such as national-level small business support policies, tax incentives, and the like; λ i corresponds to county-level-specific unobserved fixed effects that are time invariant, such as resource endowment, environment, floodplain areas, available infrastructure, and others. ε it is an error term. Errors are clustered at the county level allowing them to be correlated over time for each clustering unit.

Allocation of the SBA loans may be partially determined by the total number of business establishments in a given county. Thus unobservable factors that determine business establishments in the model (ε it) can be related to the processes that generates SBA loans (SBAit) in our analysis. Other control variables such as population, income, and financial market conditions may also be endogenous because these factors are reinforced by business survival and growth. In addition, total economic damages are endogenous to economic systems and ignoring this potential endogeneity would introduce a bias in the model estimation. Dynamic dependency in the model, captured by the lagged dependent variable entering on the right hand side, is another source of bias in fixed effects panel models with short time T relative to N, referred to as Nickell Bias (Nickell 1981).

To address both the endogeneity and the bias introduced through the dynamic dependency in the model, we employed the Blundell and Bond system GMM estimation approach (Blundell and Bond 1998, 2000; Arellano and Bond 1991). We treated income, population, HHI index, and total losses as endogenous regressors and corrected the endogeneity by employing GMM-type instruments using all available lags starting from the lag five (Roodman 2009a).Footnote 13 The choice of a starting lag (lag five) was dictated by the presence of the second and, to a limited extent, third order serial correlation in errors, as indicated by statistically significant coefficients associated with Arellano and Bond’s test statistics for AR(3) reported in Tables 2, 3, 4, 5, 6, 7, 8, 9, and 10. Bond and Meghir (1994) discussed the general case of the selection of valid instruments in GMM models in which errors are MA(q), or moving average processes of order q with q ≥ 1, and first q autocorrelations are nonzero. For the transformed equation in period t, they suggested untransformed regressors dated t − s, such that s ≥ 2 + q, as valid instruments.Footnote 14

Table 2 Panel fixed effects model (full sample)
Table 3 Blundell and Bond System GMM results, aggregate categories (full sample)
Table 4 Blundell and Bond System GMM results, small establishment categories (full sample)
Table 5 Log-log model, aggregate categories (full sample)
Table 6 Log-log model, small establishment categories (full sample)
Table 7 Blundell and Bond system GMM results, metro sample
Table 8 Blundell and Bond System GMM results, urban sample
Table 9 Blundell and Bond System GMM results, rural sample
Table 10 Blundell and Bond System GMM results, coastal sample

5 Results

In this section, we first present results from the full sample. We also separated counties based on their metro, urban, and rural status designations, to explore important information about local market dynamics using the USDA Rural-Urban Continuum Codes (RUCC). The nine code categories, in ascending order from one to nine, indicate relative ruralness of a county. Codes 1–3 identify metro, codes 4–7 correspond to urban, while the remaining codes represent rural counties.Footnote 15 A county’s rural-urban designation may change over time, given that some counties grow faster than others while some are in decline. USDA updates these RUCC codes every decade. To be consistent with our sample time frame, we used the most recent 2013 RUCC codes corresponding to 2003–2013. Forty percent of the sample counties fall under RUCC codes 1–3, approximately 45.85% are defined as urban, and the remaining 14.15% comprise rural counties.

Last, we present results for the sample of coastal counties exposed to frequent floods, hurricanes, and coastal disasters and naturally receive a larger proportion of disaster loans in the sample. The National Oceanic and Atmospheric Administration (NOAA) defines counties as “coastal” if at least 15% of total land area is located within the nation’s coastal watershed, and a portion of or an entire county accounts for at least 15% of a coastal cataloging unit (NOAA 2011). Approximately 22.63% of sample counties are coastal.

As a robustness check, we also compare results of the system GMM model (Arellano and Bover 1995; Blundell and Bond 1998) with (i) two-step feasible GMM, (ii) Arellano and Bond differenced GMM (Arellano and Bond 1991), and (iii) Anderson and Hsiao (Anderson and Hsiao 1982) models.

5.1 Full sample

We start this section by first presenting the results from the panel fixed effects models reported in Table 2 and compare them to the system GMM model presented in Table 3. Column headings indicate model specifications that differ by establishment size. Column (1) corresponds to the total number of establishments employing 1–49 people, column (2) shows results from the model in which the dependent variable is represented by businesses employing 50–99 people, column (3) presents results for establishments with 100 or more employees, and the last column (4) reports results for all establishments. All dependent variables are scaled by 100. Therefore, marginal effects are multiples of 100 of regression coefficients reported.

As indicated by the results of the fixed effects model, ignored endogeneity of multiple endogenous variables biases coefficients of not only endogenous variables in the model but the estimates on other control variables as well. In addition, the lagged dependent variable on the right-hand-side of the model introduces the so-called Nickel Bias when the data has small T relative to N. Judson and Owen (1999) estimate this bias to go as high as 20% even if T is greater than 20, and show that for data in which time dimension is 10 or less, the GMM performs best.Footnote 16

To address these biases, in Table 3, we present Blondell and Bond System GMM models. Our results suggest the number of total businesses significantly increase in response to an increased spending in SBA disaster loans per establishment in a county. Importantly, this effect appears to be driven by businesses employing fewer than 50 people. If all else is held constant, for every US$1 increase in disaster loan per establishment, an average four businesses of 1–49 employees are added in a county.

Higher labor costs, as captured by manufacturing wage rates, reduce the total number of business establishments of all sizes. Among the different-sized establishments, the largest effect is estimated for the businesses employing 1–49 employees. The more populous the county, the higher the number of total establishments are, of which only the smaller establishment category (1–49) exhibit the statistically significant effect on average. Our results indicate wealth is a primary driver for the growth of all types of business establishments. Prior literature shows the level of economic development in regions plays an essential role in explaining the relationship among government programs, entrepreneurial activity, and various economic performance measures across regions (Craig et al. 2008; Fritsch and Mueller 2008; Van Stel et al. 2005).

In all model specifications, high HHI that proxies lack of competitiveness in financial markets represents an impediment to business survival. This effect is highly significant for all establishment sizes. The negative correlation between business survival and highly concentrated markets can be the consequence of a market dynamic in which fixed costs related to entry barriers lead to a highly concentrated deposits market; in turn, negatively affecting business survival through reduced loan availability.

Small business establishments appear to be particularly vulnerable to weather shocks indicated by negative coefficients associated with contemporaneous as well as the lags of rainfall anomaly variables. Contemporaneous weather shocks exert the highest negative effects on smaller business establishments, which start to decay over time and the significantly negative effect persists for up to 4 years, suggesting limited adaptation and the lack of business-level learning processes. On the other hand, the lag five shows a statistically significant and positive effect on business establishments of all sizes, implying a relatively slower pace of business adjustments and adaptation in the medium term. All else held constant, counties also lose smaller establishments if impacts are severe. At the margin, for every US$10 increase in flood-related losses per capita, as many as 17 small establishments are lost.

In Table 4, we present results for 1–49 establishment sizes. Categorizing establishments by smaller sizes of 1–9, 10–19, and 20–49 allowed us to further separate the differences in the SBA effects dictated by differences in size categories. Small business disaster loans are only statistically significant for the smallest size category 1–9. Importantly, these size establishments are also adversely affected by weather shocks up to lag four and suffer with the magnitude of disaster impacts. Noticeably, all smallest-sized establishments positively increase in response to population and wealth growth in a county, remain highly sensitive to financial market frictions indicated by negatively significant coefficients associated with the HHI index, and decline in response to increased input costs. Size categories 10–19 and 20–49 generally exhibit more resilience to disaster impacts measured by economics losses, and are unaffected by contemporaneous and most recent rainfall anomalies; impacts of weather shocks on these sized establishments are seen in lags three to four. Finding no adverse impacts of immediate shocks on these sized employers could be suggestive of a temporary growth and reconstruction stimulus disasters provide for some businesses, including those operating in construction and related services sectors.

In addition, Tables 5 and 6 report results from the log-log models in which dependent variables are also log transformed. The assumption of log-log models is that elasticities of the SBA disaster loans with respect to business growth are constant. This assumption may be restrictive because the effects of disaster loans on business survival potentially depend on the size of the loan. Importantly, results estimated from the log-log model are consistent with our main models presented in Tables 2 and 3.

5.2 Metro, urban, and rural

Tables 7, 8, and 9 report GMM models corresponding to metro, urban, and rural samples, respectively.Footnote 17 GMM model results indicate SBA disaster loans remain highly significant for business establishments of size 1–49 in metro and urban counties, and insignificant in the rural sample. The effects of SBA disaster loans are estimated statistically significant on size 50–99 establishments in urban sample and marginally significant at 10% significance level in the sample of metro counties. Comparing the effects of the SBA disaster loans across metro and urban sample also show the magnitude of positive effect is larger in metro counties. At the margin, the effect of disaster loans per establishment on business establishments of size 1–49 in metro counties is ten times larger than that in urban areas. Based on the results, businesses in metro counties are also more vulnerable to weather shocks (measured with precipitation anomalies) and suffer more in response to flood damages relative to businesses in urban counties. Increased vulnerability of businesses in metro areas likely indicate that small business in these areas tend to seek and receive disaster loan assistance from SBA more frequently. This in part can explain why businesses in metro areas experience relatively larger impacts of SBA disaster loan on growth than the businesses in urban counties.

The differences in the effects of the SBA disaster loans across metro and urban samples can also be attributed to the regional and economical factors; for example, firms can be attracted to metro counties because of various advantages including transportation cost advantages, availability of skilled labor, capacity, and technology spillovers. Fritsch and Mueller (2008) also find relatively prominent effects of new business formation on regional development in highly agglomerated (e.g., as in metro areas) regions relative to moderately congested urban and rural regions in Germany. This emphasizes the importance of regional and economical differences such as the level of competition in the regions; type of industry mixtures (e.g., knowledge-intensive and/or high-tech industries); the features of new and incumbent businesses; and their ability to absorb the positive effect of entrepreneurial activities.

Demand-side factors (i.e., population and wealth) remain important drivers of business establishment survival in all three samples examined separately. Negative and stastistically significant coefficients associated weather shock variables, up to and including the lag four, suggest small businesses in metro and urban areas remain highly vulnerable to historic hazard incidents and exhibit lack of adaptation through learning. In the sample of metro counties, 1–49 size significantly declines in response to current and prior flooding events up to lag three. HHI is negatively significant in all models in the metro sample, as are the coefficients associated with manufacturing wage rates. Higher manufacturing wage rates negatively impact size category 1–49 in urban and rural samples, while financial market frictions adversely affects the same business categories in the urban sample only. Not surprisingly, small business categories in rural areas are adversely affected by only contemporaneous weather shocks; the past 2-year flooding events show no significant impact on business survival and the adverse effects start to show up from the events happening in the past 3–4 years. Literature shows that floods could induce positive growth on the agriculture sector in the short term because of its effect on land productivity in a subsequent year (Cunado and Ferreira 2011; Loayza et al. 2012).

5.3 Coastal counties

In Table 10, we report results for the sample of coastal counties.Footnote 18 Flood and coastal hazards are most prevalent in these counties, as are disbursements of available SBA disaster loans after flood disasters. Consistent with the results of the full sample, SBA disaster loans appear to significantly impact the growth of business establishments in the 1–49 size category and show statistically no discernable effects on larger sizes. At the margin, every US$1 increase in disaster loans per establishment keeps three additional businesses employing 1–49 people in a coastal county. Consistent with other sample analyses, population and income represent primary drivers for the growth of smaller business categories. Coastal counties are the only counties in which businesses exhibit relative resilience to past disaster shocks. The statistically negative effect is only seen in response to severity of impacts, measured by economic losses and relative to the last year’s weather shock. All previous lags of disaster shocks show no discernable effects on business survival. Relative resilience of coastal counties to hurricane disasters has been suggested in the literature as the potential effect of disasters “motivating mitigation behavior” (Sadowski and Sutter 2008).

5.4 Robustness check

Our results are generally robust to a number of different specifications. As a robustness check, we estimated models in which, instead of adding ones to SBA loans before log transformation, we add arbitrarily small (0.0001) and larger numbers (100). Adopting the approach of Eaton and Kortum (2001), we also tested sensitivity of estimated magnitudes to an alternative strategy that deals with zero observations. Specifically, we replace zero values with the minimum values of SBA loans for each panel unit. This approach is more appropriate for a data-generating process when zero values are a result of censoring. Although potential censoring could be the case for small business loans (i.e., not all businesses experiencing damages are approved for disaster loans), zeros observed in our sample could also be natural realization (i.e., resilient businesses naturally do not suffer with losses and thus receive no loans; furthermore, floods may not happen every year). Our results reported in Appendix Tables A.10 and A.11 are invariant, both in terms of signs and significance, to different log transformations. Adding one generally increases magnitudes of estimated effects relative to models in which we use arbitrarily small numbers or minimum loans for each panel unit, which if anything, reaffirms our conclusions.

We also explore how employing the system GMM improves model efficiency relative to variants of GMM models, including the two-step feasible GMM model, Arellano and Bond differenced GMM and Anderson and Hsiao GMM models. For the ease of exposition, these results are only presented for total establishments, reported in Appendix Table A.12 and for smaller establishment (1–49) categories (Appendix Table A.13). GMM models generally show consistency with the system GMM models, with slight differences in estimated magnitudes. Anderson and Hsiao and two-step feasible GMM models yield lower bounds of estimated effects, while differenced Arellano and Bond GMM yields the magnitude very similar to the Blundell and Bond system GMM models.

Theoretically superior estimates from different GMM models are commonly discussed in reference to the point estimates on the lagged dependent variable entering on the right-hand-side of the model (Bond 2002). In particular, the superior estimates should be bounded between the estimates of the ordinary least square (OLS) and fixed effect (FE) models, because both OLS and fixed effect estimators provide biased estimates and they are biased in opposite directions (e.g., Bond (2002) shows that OLS results in an upward bias while FE gives a downward bias on the coefficient estimate on the lagged dependent variable). For the convenience in columns (5) and (6) of Appendix Tables A.12 and A.13 results from the OLS and FE models are also reported, respectively. OLS and FE ranges of the coefficient associated with lagged dependent variable are (0.8506–0.9899) and (0.8427–0.9895) in small establishments (1–49) and total establishment models, respectively. While AH is a consistent estimate, it performs poorly as the point estimates on lagged dependent variable is outside of suggested range; all other GMM estimates (i.e., two-step feasible GMM, differenced Arellano and Bond GMM and System GMM) are within desirable ranges.Footnote 19

6 Discussion and policy implications

A number of important results emerge from this study. We found strong empirical evidence that disaster loans significantly contribute to the growth and survival of business establishments in a county and are particularly important to the livelihood of the smaller establishment category. Consistent with the literature, our results indicate that small business establishments are extremely vulnerable to current and past weather shocks and decline with the severity of the impacts. To the extent that businesses grow in response to floods experienced 5 years before also suggest business adaptation is slow and generally, small businesses suffer in the short to medium term (1–4 years). Furthermore, business survival is impeded by higher labor costs and financial market frictions. Regarding the demand-side factors, both wealth and population size are important determinants for business survival.

Comparing metro, urban, and rural samples, our results show that SBA disaster loans are particularly significant for establishments in metro and urban areas and exert no discernable effects on those in rural areas. There are number of explanations for this pattern. Businesses of all sizes likely cluster in urban and metro areas because of agglomeration effects and for cost reasons (Burns 1977; Head et al. 1995; Krugman 1991). However, other factors in rural areas more dominant than the SBA disaster loans may affect exit/entry decisions; farmers may rely on crop disaster insurance or other economic development incentives (i.e., development credit programs, tax exemptions on land and capital, infrastructure investment) geared toward restructuring and supporting business growth in rural counties. Importantly, our results indicate that SBA loans are highly important for business survival in coastal counties, where businesses also exhibit relatively more resilience to weather shocks.

Overall, our results indicate that SBA disaster loans are very effective in promoting small business survival. We estimate that, in the absence of disaster loans, counties would lose approximately 134 small businesses per year. This is a staggering number given that new small businesses are paramount as the source of new job creation (Birch 1987; Neumark et al. 2011; Davis et al. 1996; Hatiwanger et al. 2013). In the USA, even though business start-ups account for only a small percent of employment (3%), they are responsible for 20% of gross job creation (Hatiwanger et al. 2013). Further supporting evidence demonstrates that smaller local businesses have a more positive association with local economic performance (e.g., income growth and poverty reduction) than their larger counterparts (Rupasingha 2016).

Like other government industrial programs, the SBA loan programs may have an effect on growth, welfare, employment, and business reallocation in the industries. For example, subsidized disaster loan programs may encourage low productive firms to rebuild their infrastructure, to undertake investments to continue operation and even increase their productivity, and further help businesses return after the disaster. On the other hand, they may hinder economic growth by affecting reallocation structure of industries, i.e., slowing down the reallocation process in the industry by deterring exit of less efficient firms and entrance of more efficient firms into the industry (Acemoglu et al. 2013, Foster et al. 2001, 2006). Without modeling the reallocation process structurally and measuring firm productivity, we cannot clearly discuss welfare implications of the SBA loan programs, which would be an interesting future research topic.

The SBA disaster loans are among the few—if not the only—ex-post recovery resources to which small businesses have access. Except for limited instances, small businesses are typically uninsured against hazards,Footnote 20 hold fewer cash reserves, and are comparatively ill-prepared for hazards, in that they lack hazard contingency, business continuation, and mitigation plans (Alesch et al. 1993). Furthermore, unlike households and individuals who receive generous disaster aid or are entitled to various safety program payments, including income maintenance and unemployment insurance (Deryugina 2013), such types of financial aid are almost nonexistent to support businesses.

In addition to being directly hit, small businesses face impediments to survival after disasters because disasters interrupt supply linkages, disrupt critical infrastructure, transportation and utility services, and adversely impact the potential consumer base (Dahlhamer and Tierney 1996; Webb et al. 2002). However, the literature also indicates that even though the survival rate of small businesses is quite low, if they do not fail, their growth rate is faster than that of large businesses (Dunne et al. 1988). Based on concrete evidence that firm entry, relocation, and survival have a positive effect on regional productivity growth (Brixy 2014; Foster et al. 2006; Brouwer et al. 2005; Robbins et al. 2000),Footnote 21 our main finding, as highlighted throughout the paper, is that post-flood SBA disaster loans are vital for the survival of smaller-sized businesses and help to mitigate the adverse impacts of recurrent disasters.

7 Conclusion

Disasters cause massive disruption of local economies and in the long term may induce various adjustments and behavioral responses, including relocation, permanent outmigration, business shutdowns, and sectoral reallocation of labor (Smith et al. 2006; Hornbeck 2012; Hornbeck and Naidu 2014). Research on business recovery appears fragmented and focused on isolated historic incidents (Dahlhamer and Tierney 1996; Webb et al. 2002), making it challenging to systematically identify factors that affect business survival in the aftermath of disasters. Here, we have presented the first comprehensive study using county-level data about business establishments and loan disbursements to estimate the effects of weather shocks and SBA-subsidized disaster loans available to businesses after floods in the USA.

We employed a dynamic panel GMM model to correct for dynamic dependency and potential endogeneity arising between the disaster loans, other controls, and the number of business establishments. Controlling for county-specific and year-fixed effects and numerous other factors contributing to business start-ups and survival, our results indicate that small business establishments are vulnerable to weather shocks, while SBA disaster loans significantly contribute to their survival. Businesses located in metro and urban areas appear to be most responsive to disaster loans, but they also suffer the most from frequent floods compared to rural counties. We find some evidence that certain size categories of small businesses (10–19 and 20–49) are positively affected by excessive floods in a short term likely because of a rebuilding stimulus disasters generate in the aftermath. Some businesses in rural counties also increase in response to a recent flooding incident. Nonetheless, small business generally exhibit vulnerability to weather shocks for up to 4 years and indicate delayed learning processes and adaptation. Businesses in coastal counties appear to be unaffected by flood incidents, which suggests that frequent exposure motivates adaptation, thereby making coastal businesses more resilient.

Flooding is the major cause of economic and human loss in the USA and impacts are likely to intensify under future climate change. Loss of a single small business may not create huge disturbances in the local economy, but cumulative losses can have a substantial effect on overall employment nationwide. Without SBA disaster loans, significant numbers of small businesses will be lost. Therefore, understanding the mechanisms of the loan disbursement process in different areas is vital for business disaster recovery. Whether subsidized disaster loans imply long-term sustainability of businesses remains an open question we leave for future research. Lack of individual firm-level data also limits our ability to fully explore firm-specific factors that affect the mechanisms behind estimated effects. Examination of sector-specific disaster effects on establishments and SBA loans is another interesting dimension for future research.