1 Introduction

Why do some societies successfully adjust and even thrive amid adversity while others fail to do so (Van Der Vegt et al., 2015)? To answer this question, scholars increasingly recognize the importance of entrepreneurial ecosystems, which refer to a set of interdependent actors and factors that are governed in such a way that they enable productive entrepreneurship (Stam and Van de Ven, 2021). According to Mason & Brown (2014), the entrepreneurial ecosystem has links to “economic gardening” as a metaphor for local economic development, in which specific environments promote not only high rates of new business startups but also high growth firms (Alvedalen & Boschma, 2017). In an environment where crises frequently strike, governments and policymakers seek to implement supportive policies in entrepreneurial ecosystems to encourage entrepreneurship as part of broader economic recovery strategies (Wurth et al., 2022).

As global climate change is unfolding at an unprecedented pace, there is increasing recognition that the rapid increase in natural disasters such as hurricanes, earthquakes, and floodings can have a profound impact on entrepreneurial ecosystems, with some exhibiting remarkable resilience and others struggling to recover (Boudreaux et al., 2023; Monllor & Murphy, 2017). However, how natural disasters might impact regional entrepreneurship and why this influence might vary across different entrepreneurial ecosystems still remain unclear. This lack of consensus persists as scholars have developed competing perspectives on the impact of natural disasters on regional entrepreneurial startup activities. One stream of research, based on a “creative destruction” (Schumpeter, 1939) logic, argues that the destruction caused by disasters creates opportunities for new entrepreneurial ventures to emerge (Brück et al., 2010; Linnenluecke & McKnight, 2017; Marino et al., 2008). Contrastingly, another stream of research contends that the severe disruptions caused by such disasters, including damage to infrastructure (Boudreaux et al., 2022), mental health challenges (Stephan et al., 2023), tightening consumption and investment markets (Lee et al., 2023), impede potential entrepreneurs to initiate startups in the aftermath of natural disasters.

The divergence in perspectives highlights the complexity of the relationship between natural disasters and entrepreneurial activities. To improve understanding of the dynamic interplay between natural disasters and regional entrepreneurship, an emerging line of work has begun to examine how the specific characteristics of entrepreneurial ecosystems may either magnify or buffer the negative effects of natural disasters. For example, recent research suggests that associational diversity (Dutta, 2017) or foreign aid (Doern et al., 2019) can promote the resumption of entrepreneurship activities in the region, but the positive effects of foreign aid may only surface in countries with high-quality governments (Boudreaux et al., 2022). Other work has zoomed in on the individual-level elements of the entrepreneurial ecosystem (Wurth et al., 2022), indicating that natural disasters may promote entrepreneurship among specific ecosystem actors including men, individuals with less human capital, and those who fear failure (Wei et al., 2023).

Despite all the efforts, the literature about the relationship between natural disasters and entrepreneurship startups has largely overlooked one crucial ecosystem element: financial capital availability. Financial capital availability has been researched in the crisis management literature (Cordero, 2023; Islam et al., 2018), but this work tends to dismiss differences between sources of financial capital and also neglects the unique characteristics of natural disasters. The literature has demonstrated that adverse market conditions curtail private-sector venture capital (Rizvi et al., 2020), while government or bank funding does not always align with the immediate financing needs of entrepreneurs who are confronted with sudden resource constraints when a crisis hits (Hammett & Mixter, 2017; Rouhanizadeh et al., 2020). In this situation, potential entrepreneurs’ personal savings, the money that individuals set aside from their income for future use or emergencies, may become the primary source of start-up capital for the pursuit of entrepreneurship in the wake of a natural disaster (Liguori et al., 2019).

The purpose of this paper is to research how regional personal savings, as a source of financial capital in entrepreneurial ecosystems, might moderate the influence of natural disasters on regional entrepreneurship. While extant literature tends to assume that personal savings provide individuals with immediate financial means to start a venture and make them more resilient (Doern, 2017; Liguori et al., 2019; Vasilescu, 2014), we draw on prospect theory and propose an alternative mechanism. Specifically, prospect theory predicts that high personal savings foster a protective attitude toward existing wealth (Annamalah et al., 2019; Karlan et al., 2014), heightening loss aversion and reducing entrepreneurial aspirations during crises. Meanwhile, in low-savings regions, the lack of wealth means that potential financial gains from entrepreneurship are perceived as more valuable compared to the potential losses, which may encourage risk taking (Schaner, 2018; Wake et al., 2020). Thus, we theorize that regional ecosystems characterized by higher savings will counterintuitively be more severely affected by natural disasters. In addition, we posit that this greater entrepreneurial activity in low savings regions shapes regional culture by influencing reference points, thereby exposing individuals to new benchmarks and success stories. This can lead to adjusted expectations and altered risk-taking behaviors as individuals strive to meet or exceed the perceived standards of their peers. Thus, we further theorize that the difference in entrepreneurship rates between high and low regional savings areas increases over time.

Addressing calls to enhance rigor in quantitative entrepreneurship research (Maula & Stam, 2020), we test our theory using a quasi-natural experimental design using the 2008 Wenchuan earthquake. This magnitude 8.0 earthquake, which occurred on May 12, 2008 in the province of Sichuan, China, is the 18th deadliest earthquake ever observed. It resulted in severe destruction of infrastructure and more than 5 million people losing their homes. Using a difference-in-difference-in-difference (DDD) model, our baseline finding reveals that the earthquake—on average—increased subsequent regional entrepreneurship rates. However, we also find that the earthquake’s impact substantially varies depending on the area’s personal savings. Specifically, the earthquake increased regional startup rates in low per capita savings areas, but it reduced regional startup rates in high per capita savings areas. Intriguingly, these between-area differences in earthquake impact also increase over time, confirming the surprising presence of an important feedback loop that triggers virtuous cycles of entrepreneurship in financially constrained areas.

This paper makes several contributions. First, we extend the literature on entrepreneurship and natural disasters by theorizing personal savings as a focal boundary condition that regulates how natural disasters impact entrepreneurship in regional ecosystems. This boundary condition is uniquely different from previously identified boundary conditions (e.g., Boudreaux et al., 2022), allowing for a novel understanding of the specific role of regional wealth endowments in ecosystem resilience. As such, we add to recent calls for theory and evidence that may explain why only certain entrepreneurial ecosystems are able to thrive amid adversity (Iacobucci & Perugini, 2021; Roundy et al., 2017). Specifically, by reconceptualizing the meaning of “savings” from a resource stock that promotes entrepreneurship into a driver of loss aversion that suppresses entrepreneurship in the wake of a natural disaster, our study points to the importance of considering the underlying behavioral mechanisms. Second, we contribute to prospect theory by revealing how reference points are created and changed in the entrepreneurship context. Indeed, the application of prospect theory in economics has suffered from major ambiguity about what reference points people may use to evaluate risk (Barberis, 2013). Here, our work highlights the pivotal role of household savings as a focal reference point for the decision to pursue entrepreneurship in the wake of a crisis. Addressing calls to extend the application of prospect theory to a wider range of economic behaviors (Werner & Zank, 2019), this study adds to our understanding of how expectations and loss aversion shape consumption-savings decisions in various contexts. Specifically, while prior research conducted in non-crisis contexts suggests that wealthier individuals often invest in riskier assets because they are less concerned about potential losses (Calvet & Sodini, 2014), we find that this may not be the case in a crisis context. Our study thus refines the notion that people tend to accumulate precautionary savings when uncertainty about future income increases (Lugilde et al., 2019), offering theory and evidence indicating that this tendency is comparatively weaker for low-wealth individuals who lack opportunities to save when a crisis hits and are thus more likely to be pushed into entrepreneurship.

This paper is structured as follows. In Section 2, we present the theory development and hypotheses. In Section 3, we test our theory using a quasi-natural experimental design. In Section 4, we provide our empirical results. In Section 5, we discuss our study’s theoretical contributions as well as its methodological, economic, and practical implications. We end the paper with research limitations and conclusions.

2 Theory and hypothesis

2.1 Natural disasters and regional entrepreneurial startup activities

A crisis is “a major occurrence with a potentially negative outcome affecting an organization, company, or industry, as well as publics, services or good name. It interrupts normal business transactions and can sometimes threaten the existence of the organization” (Fearn-Banks, 1996, p.1). Crises can vary widely in nature, leading to different classifications, such as man-made and natural crises (Rosenthal & Kouzmin, 1993); sudden and smoldering crises (James & Wooten 2005); and conventional, unexpected, intractable, and fundamental crises (Gundel, 2005). Within the field of entrepreneurship, Miklian and Hoelscher (2022) classified those crises into economic/financial crises, epidemics, natural disasters, societal insecurity and armed conflict, and political violence.

Among all kinds of crises, natural disasters such as hurricanes, floodings, and earthquakes are of great concern since they can be extremely devastating and are expected to occur much more frequently due to global climate change (Tierney, 1997; Kroll-Smith & Couch, 1991). Natural disasters can occur at any time, with little to no warning, and can cause massive destruction in a matter of minutes, resulting in loss of life, injury, and displacement of people. They can also cause significant economic losses, disruption to essential services, and secondary effects such as epidemics (Campbell 2009).

Despite all the concerns given to natural disasters, there are seemingly conflicting views about how they affect subsequent regional entrepreneurship in terms of new business creation. One stream of literature is based on the “creative destruction” perspective (Schumpeter, 1939) and argues that natural disasters destroy old economic relations and create new opportunities for entrepreneurship, even though this process inevitably leads to temporary disruptions and job losses. Natural disasters generate market inefficiencies and ensuing entrepreneurial opportunities within various industries (Monllor & Murphy, 2017), such as the reconstruction of infrastructure, increased awareness of insurance products, new early warning technologies, the reconstruction of the supply chain, improved communication systems, the demand for healthcare services, education programs in disaster management, and so on. Meanwhile, natural disasters can lead to financial strain for incumbents (Lee et al., 2023). Reduced consumer spending, supply chain disruptions, and other economic challenges may result in a decline in revenue, making it difficult for incumbents to sustain their existing workforce (Brück et al., 2010; Linnenluecke & McKnight, 2017). When highly skilled dismissed workers interact and collaborate, they share insights and expertise that can turn entrepreneurial opportunities into action to provide new products and services. As a result, natural disasters powerfully push subsequent regional entrepreneurial startup activities (Brück et al., 2010).

In contrast to the “creative destruction” perspective (Schumpeter, 1939), another stream of literature argues that the temporary disruptions caused by disasters act as barriers to the initiation of new entrepreneurial activities. Natural disasters disrupt daily life and entrepreneurial preparation. Potential entrepreneurs may struggle to adapt to the changes, causing high levels of stress and anxiety. The stress can lead to exhaustion, burnout, and mental health challenges (Stephan et al., 2023), thus inhibiting entrepreneurs from initiating a set of activities to turn their ideas into businesses (Gartner, 1985; Koellinger & Roy Thurik, 2012). Meanwhile, natural disasters often lead to increased market volatility (Lee et al., 2023). The uncertainty surrounding economic conditions, consumer behavior, market demand, and industry trends can make it challenging for investors to assess the potential returns and risks of investments accurately (Shefrin, 2002; Slovic, 1972). In this situation, investors are more cautious and selective in their investment decisions (Lippi & Rossi, 2020; Rizvi et al., 2020). They may choose to allocate capital to safer and more established firms, but not startups. As a result, even when skilled professionals identify entrepreneurial opportunities, they are unable to start entrepreneurial activities due to financial constraints. Thus, natural disasters powerfully hinder subsequent regional entrepreneurial startup activities (Boudreaux et al., 2022).

The preceding discussion thus highlights that natural disasters can theoretically be expected to either promote or reduce entrepreneurial startup activities. Before explicating boundary conditions, we first explore which perspective is on average more predictive of natural disasters’ impact on regional entrepreneurial startup rates by setting up competing hypotheses. By considering competing hypotheses, we avoid confirmation bias (Lehner et al., 2008) which is the tendency to only accept evidence that confirms our existing beliefs (Nickerson, 1998), thus pushing us to consider evidence that might contradict scholarly beliefs about how natural disasters affect regional entrepreneurship. Accordingly, we propose the following:

  • Hypothesis 1a: Natural disasters are positively related to subsequent regional entrepreneurial startup rates.

  • Hypothesis 1b: Natural disasters are negatively related to subsequent regional entrepreneurial startup rates.

2.2 Understanding heterogeneity in entrepreneurial ecosystem resilience

As the pathway between natural disasters and entrepreneurial startups is ambiguous, researching how the specific characteristics and resource endowments of entrepreneurial ecosystems might sway the balance in either direction is important. Entrepreneurial ecosystems constitute “a set of interdependent actors and factors coordinated in such a way that they enable productive entrepreneurship within a particular territory” (Stam & Spigel, 2016, p. 1). Accumulating research insights suggest, however, that not all ecosystems are alike and that they may differ from one another in distinct ways. In particular, Stam (2015) identified 10 key factors that characterize the DNA of an entrepreneurial ecosystem: formal institutions, informal institutions, social networks, physical resources, financial resources, leadership, human capital, knowledge, means of consumption, and producer services. While informative for characterizing the uniqueness of different ecosystems, this framework and related works (see Stam & Van de Ven, 2021) have only recently been leveraged for understanding the origins of ecosystem resilience, i.e., the ability of ecosystems to recover from and adapt to external shocks (Roundy et al., 2017).

Among all possible ecosystem characteristics, this paper emphasizes the importance of financial capital availability for ecosystem resilience (Cordero, 2023; Islam et al., 2018). Entrepreneurial startups indeed require a certain amount of capital for initial investment, including setting up infrastructure, purchasing equipment, and securing necessary licenses (Brown & Rocha, 2020). Without adequate funding, entrepreneurs may struggle to cover these essential startup costs (Berger & Udell, 1998). However, investors become more cautious and selective in their investment decisions after natural disasters (Lippi & Rossi, 2020; Rizvi et al., 2020), and they may choose to allocate capital to safer and more established firms, but not startups. Thus, one of the important ways that natural disasters inhibit entrepreneurial activity is by constraining the financial capital that potential entrepreneurs can access. While other factors might somewhat mitigate this relationship, addressing financial issues is still a crucial determinant in ensuring that entrepreneurial startups can be initiated in crisis environments (Brown & Mason, 2017).

Financial capital in the entrepreneurial ecosystem is defined as “the presence of financial means to invest in activities that do not yet deliver financial means” (Stam & Spigel, 2016). This definition implies that entrepreneurial ecosystems vary in terms of financial resources, such as venture capital (Bonini & Capizzi, 2019), bank loans (Cohen, 2006), government funding (Mason & Brown, 2014), and personal savings (Neumeyer et al., 2019). The intuitive understanding is that, when potential entrepreneurs cannot obtain financial capital from the private-sector, government funding or bank loans can contribute to entrepreneurial recovery after natural disasters (Cordero, 2023; Islam et al., 2018). But studies have shown that there are gaps between the financing needs of potential entrepreneurs after disasters and the availability of government or bank capital (Hammett & Mixter, 2017; Rouhanizadeh et al., 2020). Government agencies often have complex processes that can slow down the allocation of funds (Worldbank, 2016). Government decisions may be influenced by political considerations, which can impact the allocation of venture capital. Political agendas and priorities may not always align with the immediate needs of entrepreneurs and startups after natural disasters (Hammett & Mixter, 2017).

2.3 The moderating role of ecosystem personal savings

Because venture capital becomes scarce and potential entrepreneurs have limited other channels to access capital in times of adversity, scholars have argued that their personal savings are crucial for their entrepreneurship (Doern, 2017; Liguori et al., 2019; Vasilescu, 2014). This line of work thus views savings as a resource and suggests that an increase in personal savings can provide potential entrepreneurs with a greater pool of initial capital, thereby enhancing their ability to initiate and sustain a new venture (Verheul & Thurik, 2001). When individuals accumulate savings, they amass financial resources that can be channeled into entrepreneurial endeavors, covering startup costs, operational expenses, and potential risks associated with business ventures (Cardon et al., 2005). This increased access to capital can empower aspiring entrepreneurs to pursue their business ideas, invest in innovation, and navigate the initial challenges that come with establishing and growing a new enterprise (Gentry & Hubbard 2000; Rotefoss & Kolvereid, 2005).

In contrast, prospect theory suggests that savings may affect risk taking (Kahneman & Tversky, 1979). Prospect theory is a model of decision making under uncertainty, such as natural disasters, which assumes a reference point relative to which outcomes are seen as gains or losses (Figenbaum & Thomas, 1986; Werner & Zank, 2019). In applying prospect theory to the context of ecosystems, we posit that regional savings serves as a key driver of such reference point. That means, natural disasters have the effect of decreasing the probability of gains and increasing the probability of losses, but this effect may be perceived differently depending on an individual’s savings (Levy, 1992).

Prospect theory also introduces the concept of diminishing sensitivity, where the subjective value of gains decreases as their original personal savings increase (Peterson et al., 2021; Tversky & Kahneman, 1992). For example, the subjective value gained from increasing from 1000 to 2000 EUR is greater than the satisfaction gained from increasing from 10,000 to 11,000 EUR. Since high savers already possess substantial wealth, the marginal gains from starting a business are relatively small for them (Kahneman et al., 1991; Wake et al., 2020). They place greater emphasis on risk management and opt for conservative financial strategies to protect their savings (Annamalah et al., 2019), rather than on doing something that could risk losing this wealth (Barberis, 2013; Karlan et al., 2014). But entrepreneurship will be seen as a risky endeavor, particularly in the aftermath of a crisis in which the perceived odds of venture survival sharply decrease. This means entrepreneurship is seen as unattractive due to the risk of losing current savings, rather than potential benefits. Since individuals in regions with high savings could adopt a “wait-and-see” approach when a crisis hits (Stephan et al., 2023, p. 682) as their wealth or the wealth endowments of people in their social networks allows them to survive without doing much until the crisis passes.

Conversely, in low-savings regions, the smaller stock of savings means that additional financial gains from entrepreneurial activities are perceived as more valuable compared to the potential losses, which may encourage risk taking (Schaner, 2018). In particular, low savings dictate that individuals cannot take a wait-and-see attitude as they have to earn money to stay alive and cannot rely on the financial support of others in the local community. Natural disasters often cause extensive damage to infrastructure and result in economic downturns, disrupting the normal functioning of businesses and reducing job opportunities (Brück et al., 2010; Linnenluecke & McKnight, 2017). In this situation, low-savings potential entrepreneurs have no other viable option for licit income than engaging in entrepreneurial startup activities (Dencker et al., 2021; O’Donnell et al., 2024). Thus, entrepreneurship is seen as an attractive pathway as the lack of substantial savings pushes people to be more focused on its potential upside gains compared to its potential downside losses.

While one could argue that low personal savings may indicate insufficient initial capital for entrepreneurship and imply lower personal capabilities (Goyal & Kumar, 2021), this is not necessarily the case. Indeed, the level of initial capital appears more closely tied to the size of entrepreneurial startups (Blank & Dorf, 2020; Klačmer Čalopa et al., 2014). It follows that having substantial financial resources allows for larger ventures, while undercapitalized entrepreneurs may still create smaller-scale ventures (Berger & Udell, 1998; Cassar, 2004). Thus, there are diverse entrepreneurial modes suitable depending on the available capital, which means it would be premature to assume that low personal savings necessarily equate to limited entrepreneurial activity in the wake of a crisis. Accordingly, we propose the following:

  • Hypothesis 2: Natural disasters are positively related to subsequent regional entrepreneurial startup rates in low regional per capita savings areas, but negatively related to subsequent regional entrepreneurial startup rates in high regional per capita savings areas.

2.4 Divergence in entrepreneurial ecosystem resilience over time

Entrepreneurial ecosystems are dynamic systems (Wurth et al., 2022), which means that natural disasters could substantially alter their evolutionary trajectories. While external shocks may set in motion a cascade of new firm formation in some ecosystems, thereby generating a virtuous cycle of entrepreneurship over time, other ecosystems may witness a vicious cycle in which the devastation caused by a disaster triggers a downward spiral of entrepreneurial activity. Here, we argue that a region’s savings may help to explain these divergent pathways.

The literature has shown that entrepreneurial ventures play a pivotal role in shaping regional entrepreneurial cultures by serving as powerful sources of inspiration and role models (Fritsch & Wyrwich, 2018; Stuetzer et al., 2016). Through their efforts and achievements, these startups showcase that innovative ideas can evolve into prosperous ventures, fostering a mindset shift within the local community (Chinitz, 1961). As cultural icons and recognized leaders, the founders of successful startups become relatable figures, demonstrating resilience, adaptability, and problem-solving skills that resonate with aspiring entrepreneurs facing their own challenges (Bikhchandani et al., 1992). The visibility and recognition garnered by these startups elevate the entire regional entrepreneurial ecosystem, encouraging diverse individuals to consider entrepreneurship as a viable and accessible career path (Fornahl, 2003; Wyrwich, 2015; Wyrwich et al., 2016). They share their knowledge and experience with the public, and further contribute to this culture of inspiration by providing valuable insights and guidance to those navigating their entrepreneurial journeys (Figueiredo et al., 2002; Michelacci & Silva, 2007; Stam, 2007). Ultimately, the impact of new firm formation goes beyond individual achievement, contributing to the vibrancy and dynamism of regional entrepreneurial cultures.

Prospect theory (Kahneman & Tversky, 1979) captures that societal norms and cultural expectations can shape what individuals consider to be a “normal” or “acceptable” reference point (Figenbaum & Thomas, 1986; Holmes et al., 2011). Entrepreneurial cultures significantly influence the reference points in prospect theory by exposing individuals to new benchmarks and success stories. When people observe their peers achieving high returns on investments, receiving promotions, or attaining other forms of success, they may recalibrate their own expectations and aspirations (Ruggeri et al., 2020; Sun et al., 2021). This sense of relative deprivation motivates individuals to take greater risks or seek out new opportunities to match or exceed their peers’ successes. For example, an investor initially satisfied with a 5% annual return might shift their reference point to 10% after seeing their peers achieve higher returns, prompting them to take on more risk. In the context of entrepreneurship, observing successful peers can elevate reference points, increasing the appetite for risk as entrepreneurs strive to achieve similar success (Wake et al., 2020). Consequently, in such cultural environments, individuals may be more inclined to change the reference point and choose options with higher potential returns, even if it involves taking on some risk and uncertainty (Bromiley, 2010; Jegers, 1991).

Studies have indeed shown that entrepreneurial cultures tend to be rather persistent and contribute to entrepreneurial activities over longer periods of time (Freytag & Thurik, 2007; North, 1994; Williamson, 2000). This leads us to theorize that when natural disasters promote subsequent entrepreneurship in regional ecosystems with low savings, this will strengthen the entrepreneurial culture of these ecosystems. In turn, entrepreneurial cultures contribute to the sustained and long-term development of entrepreneurial activities, thereby establishing a positive loop in ecosystems with low savings. In contrast, in ecosystems characterized by high levels of personal savings, such a positive feedback loop fails to occur. In this setting, natural disasters fail to ignite an initial burst of new firm formation, which makes individuals even more hesitant to launch their own businesses due to the lack of credible signals about market opportunities and the greater difficulties of imitating or learning from the founding attempts of others. Accordingly, we propose the following:

  • Hypothesis 3: The difference in subsequent regional entrepreneurial startup rates between high regional per capita savings areas and low regional per capita savings areas increases over time.

3 Methods

3.1 Study context

On May 12, 2008 at 2.28 pm, a 8.0-magnitude destructive earthquake struck Wenchuan County and its surrounding areas. The 2008 Sichuan earthquake resulted in the loss of more than 69,000 lives, the injury of more than 370,000 people, the disappearance of more than 17,900 individuals, the displacement of more than 19,930,000 persons, and the overall impact of the disaster on a total of 46,250,000 individuals. This earthquake is one of the deadliest in the twenty-first century (Jia et al., 2010).

3.2 Sample and data

Based on China government announcement, the 2008 Sichuan earthquake totally affected 237 counties that lie across Sichuan, Shaanxi, Gansu, Chongqing, Yunnan, and Ningxia Provinces. Based on the assessment of the affectedness by China government, the degree of affectedness in these counties is categorized into three types: very severe, severe, and slight (see Table 1). Due to the passage of time, some of these counties have been merged or changed; thus, we have to remove them from the dataset. Ultimately, we included the remaining 186 counties (10 very severely affected counties, 34 severely affected counties, and 142 slightly affected counties) as our sample.

Table 1 List of counties categorized by extent of 2008 Wenchuan earthquake damage
Full size table

3.3 Research design

3.3.1 Quasi-natural experiment

We consider the 2008 earthquake as a quasi-natural experiment. The conventional approach to isolate the earthquake’s effect is to select an area unaffected by the earthquake as the control group (Leatherdale, 2019). But this would raise a question of the differences in government support alongside the isolation of the earthquake. The treatment group has both earthquake and government support, while the control group has neither earthquake nor government support. Thus, we cannot be sure whether the result is caused by the earthquake or by government support.

Precise separation of governmental support is a challenge because governmental support always accompanies earthquakes and is implemented more or less differently in each region. To solve the above problem, we take advantage of the classification system used by the China central government to rate the extent of earthquake-induced damages in each of the areas. Specifically, we consider areas rated as “very severe” and “severe” damages as the treatment group and areas rated as “slight” damages as the control group. The benefit of this approach is that all three areas received government support, as outlined in the government’s notification (No. 21 [2008]). Accordingly, our approach allows us to hold government support constant and control other unobservable to isolate the earthquake effect on regional entrepreneurship.

3.3.2 Difference-in-differences-in-differences model

We use the difference-in-differences-in-differences (DDD) model in the quasi-natural experimental research design. The DDD model is derived from the difference-in-differences (DID) model with an additional intervention condition (in our case, high/low regional per capita savings). The key principle of the DID/DDD model is to compare the changes in outcomes between two or more groups over time, where one group has been exposed to the treatment (or intervention) and the other group(s) has not (Goodman-Bacon, 2021).

Quasi-natural experiments are not truly randomized (Antonakis et al., 2010; Hill et al., 2021). In order to ensure that the two sets of data are comparable, the key assumption for consistency of the DDD and DID estimator is the zero correlation assumption which is also called as “parallel trends” assumption (Roberts & Whited, 2013). That means any trends in outcomes for the treatment and control groups prior to treatment are the same. This approach helps control for potential confounding factors that might affect the outcomes by examining how the treatment effect varies before and after the treatment is applied. In other words, DID models provide unbiased effect estimates if the trend over time would have been the same between the treatment and control groups in the absence of the treatment (Stuart et al., 2014). Our analysis confirmed the parallel trends assumption, thus indicating that entrepreneurship rates in the treatment and control groups were comparable prior to the earthquake (see Appendix A1 for details).

3.3.3 Power analysis

Although we include all counties that were at least slightly affected by the earthquake in our sample, we still consider whether this sample size is large enough (Green, 1991). A small sample size directly decreases the effect size and statistical power of a study (Sawilowsky, 2009). Statistical power refers to the probability of correctly rejecting a null hypothesis (Neyman & Pearson, 1928). Effect size quantifies the magnitude or strength of a relationship in a statistical analysis. Following Cohen (2013), we select medium effects size (0.15) and using PASS software estimate the required sample size (Gatsonis & Sampson, 1989). A sample size of at least 144 is required based on a power of 0.9. That is to say, our sample of 186 counties fully meets this requirement.

3.4 Measurement of variables

3.4.1 Regional entrepreneurial startup rates (labeled as startup rate)

Regional entrepreneurial startup rates = (number of entrepreneurial start-ups/regional permanent population) × 10,000. The number of new firms that were established from May 13, 2008 to May 12, 2009, May 13, 2009 to May 12, 2010, and May 13, 2010 to May 12, 2011 in each county represents the number of entrepreneurial start-ups of each county in the first, second, and third years after the earthquake. The number of new firms that were established from May 13, 2007 to May 12, 2008, May 12, 2006 to May 12, 2007, and May 13, 2005 to May 12, 2006 in each county represents the number of entrepreneurial start-ups of each county in the first, second, and third years before the earthquake. We obtained the entrepreneurial start-up information from website search engines and databases compiled by the “China Administration for Industry and Commerce” which is the main governmental administrative department in charge of Chinese businesses. As for the regional permanent population, permanent population = GDP/per capita GDP. We obtained this information from the “Statistical Yearbook” of Sichuan, Shaanxi, Gansu, Chongqing, Yunnan, and Ningxia Provinces. We stopped tracking data until 2012, since another earthquake struck the Lushan area in the year 2013. Another earthquake tends to bring back memories of people who were affected by the earthquake before, thus potentially influencing their behaviors.

3.4.2 Natural disasters (labeled as treat)

As mentioned, we employ a quasi-natural experimental design and generate comparisons within the “seriously/slightly affected” areas of the earthquake. Based on China government announcements and data classifications, we use 10 very severely affected counties and 34 severely affected counties as the seriously affected areas of the earthquake (if county p is a very severely or severely affected county, \({treat}_{p}=1\)). We use 142 slightly affected counties as the control group (\({treat}_{p}=0\)). Besides, DID/DDD models also require a time variable to classify time periods into years before and after the earthquake. After the earthquake (in the years 2008, 2009, and 2010), \({post}_{t}=1\). In the years 2007, 2006, and 2005, \({post}_{t}=0\).

3.4.3 Regional per capita savings (labeled as saving)

Regional per capita savings = total deposits balances of financial institutions/permanent population. The regional per capita savings of each county in the years 2008, 2009, and 2010 represent the first, second, and third years after the earthquake. The regional per capita savings of each county in the years 2007, 2006, and 2005 represent the first, second, and third years before the earthquake. We collected savings data from the “Statistical Yearbook” of Sichuan, Shaanxi, Gansu, Chongqing, Yunnan, and Ningxia Provinces. The DDD model requires converting the factor into dichotomous variables. Therefore, we take the mean value of regional per capita savings. If the regional per capita savings in the county p is higher than the mean value in the year t, \({saving}_{pt}=1\). Otherwise, \({saving}_{pt}=0\).

3.5 Control variables (labeled as cv)

One advantage of the DDD and DID model is that we do not need control variables (Wing et al., 2018). When the parallel trends assumption holds, which in our study is the case (see Appendix A1 for details), it implies that there are no other confounding factors or unobserved variables affecting the treatment and control groups differently before the intervention (Rambachan & Roth, 2019). Therefore, any observed differences in the post-treatment outcomes can be confidently attributed to the treatment effect (Roth et al., 2023).

While DDD/DID models do not require control variables, some researchers still choose to include them. Since we use ratios in our analysis model, as suggested by Wulff and colleagues (2023), we control the permanent population, regional GDP per capita, and government expenditure to increase the robustness of using ratios in statistical models. We obtained these data from the “Statistical Yearbook” of Sichuan, Shaanxi, Gansu, Chongqing, Yunnan, and Ningxia Provinces.

3.6 Data analysis method

To further address potential confounding factors and obtain unbiased treatment effect estimates, we include the unit- and time-fixed effects in the DDD model (De Haan, 2021). Unit-fixed effects (\({u}_{p}\)) capture the differences between different counties that are constant over time (Gobillon & Magnac, 2016). Time-fixed effects (\({\lambda }_{t}\)) capture time-specific factors across all groups (Imai & Kim, 2021). These factors might include macroeconomic conditions, policy changes, or other time-specific shocks that affect all entities similarly over time. Unit- and time-fixed effects help control for unobserved heterogeneity across groups and time periods, and ensure that any observed differences in outcomes are more likely to be attributed to the treatment effect (Goodman-Bacon, 2021; Wooldridge, 2021). Accordingly, we build DDD model as below:

$${startup\;rate}_{pt}=\alpha_0+\alpha_1{saving}_{pt}+\alpha_2{treat}_p\times{post}_t+\alpha_3{treat}_p\times{saving}_{pt}+\alpha_4{saving}_{pt}\times{post}_t+\alpha_5{treat}_p\times{post}_t\times{saving}_{pt}+\alpha_i\sum{cv}_{pti}+\lambda_t+u_p+\varepsilon_{it}$$

4 Empirical results

4.1 Descriptive statistics

Table 2 lists the descriptive statistics for our sample. We tracked 6 years of data for each county, totaling 1116 records. The average startup rate is 3.4 per 10,000 inhabitants every year. Based on 80 years of lifetime, the likelihood of a person starting a business is 3.4 × 80/10,000 = 2.72%, which is similar to Li et al. (2012). And 40.1% of counties are categorized as low regional per capita savings areas.

Table 2 Descriptives
Full size table

4.2 Regression results

Hypothesis 1a predicted that natural disasters promote subsequent regional entrepreneurial startup rates, and hypothesis 1b predicted that natural disasters hinder subsequent regional entrepreneurial startup rates. As shown in Table 3, the two-way interaction term of treat and post (DID) significantly positively affects the entrepreneurial startup rate in the 3 years after the earthquake at the 5% level. The coefficient number is 0.483, which means the startup rate at seriously affected counties is higher than slightly affected counties by 0.483 units. When all other variables are at mean values, the regional entrepreneurial startup rate is 3.276 per 10,000 inhabitants every year in slightly affected counties and 3.759 per 10,000 inhabitants every year in seriously affected counties, which means the startup rate in seriously affected counties increases by 14.7%. Accordingly, our results support H1a and reject H1b.

Table 3 DDD results
Full size table

Hypothesis 2 predicted that natural disasters promote subsequent entrepreneurial startup rates in the low regional per capita savings areas, but hinder subsequent entrepreneurial startup rates in the high regional per capita savings areas. The results show that the three-way interaction term of treat, post, and saving (DDD) significantly negatively affects the entrepreneurial startup rate in the 3 years after the earthquake at the 5% level. When the three-way interaction item increases by one unit, the startup rates decrease by 1.087 units, meaning that the positive effect of the earthquake is canceled out and turns negative in high per capita saving areas. When all other variables are at mean values, the regional entrepreneurial startup rate is 4.572 per 10,000 inhabitants every year in seriously affected low regional per capita savings areas and 3.485 per 10,000 inhabitants every year in seriously affected high regional per capita savings areas, which means the rate in the high regional per capita savings areas after the earthquake decreases by 23.8%. Because startup capital may be the primary consideration in starting a business, the effect size of that factor can surpass 20%. Accordingly, our results support H2.

To test our Hypothesis 3, we need a second analysis of the interaction effect (moderating effect of savings) across different years. Hypothesis 3 predicted that the between-county differences increase over time. The results in Table 4 show that the interaction effects before the earthquake are not significant. After the earthquake, the interaction effects become significant at the 5% level for all the years. The coefficients in the years 2008, 2009, and 2010 are − 1.384, − 1.478, and − 3.363. The result shows that the moderating effect of regional personal savings becomes stronger over time, and the difference in subsequent regional entrepreneurial startup rates between the high regional per capita savings areas and the low regional per capita savings areas increases over time. Accordingly, our results support H3.

Table 4 DDD results in every year
Full size table

The results in Table 4 show that the effects of the earthquake in every year are not significant, but it is significant in Table 3. This is because we compare 3 years before the earthquake (the years 2005, 2006, 2007) with 3 years after the earthquake (the years 2008, 2009, 2010) in the DDD model (Table 3), but we compare the 5 years (the years 2006–2010) with the year 2005 in Table 4. The comparative baseline group changes; therefore, the results seem to be different. This insight underscores the importance of comprehensively comparing the seismic effects over various years by integrating longitudinal data, and relying solely on cross-sectional data may lead to heterogeneous conclusions. For Hypothesis 3, the trend of the difference will not change regardless of any group used as the comparative baseline group (see the Chow test in Appendix A4).

4.3 Robustness analyses

We performed several additional analyses to probe the robustness of the findings. These included (1) a “parallel trends” test to verify whether the treatment and control areas had similar entrepreneurship rates prior to the earthquake (see Appendix A1); (2) a “placebo” test to check the assumption that the treatment and control groups would have similar entrepreneurship rates in same earthquake intensity (see Appendix A2); (3) an examination of the per capita savings trend after the earthquake to verify whether it changed differently for areas with low versus high savings (see Appendix A3); and (4) a “Chow” test to check whether the coefficients for the theorized interaction effects significantly increase in strength over time (see Appendix A4). All supplementary analyses confirmed the robustness of our main findings.

5 Discussion and implications

This study reveals that the impact of natural disasters on entrepreneurial ecosystems can vary substantially over time depending on the ecosystem’s personal savings endowments. Unlike work that has emphasized the buffering role of financial resources during adversity (Karlan et al., 2014; Islam et al., 2018), we theorized and found that a region’s personal savings can importantly limit the resilience of entrepreneurial ecosystems by suppressing firm formation in the wake of a natural disaster. Our study thus alludes to the importance of considering the behavioral implications of wealth endowments, as reflected by its connection with loss aversion, in future scholarship on the entrepreneurship-related consequences of natural disasters. Below we discuss these theoretical implications in more detail as well as the practical implications and limitations of our research.

5.1 Theoretical implications

Firstly, we contribute to the debate about whether and when natural disasters increase or decrease regional entrepreneurship (e.g., Boudreaux et al., 2023; Salvato et al., 2020). Our research establishes that earthquakes can in fact have a longer-term positive impact on regional entrepreneurship, which challenges the popular belief that natural disasters hinder entrepreneurial startup activities immediately after they occur (Boudreaux et al., 2019). Indeed, our analysis provides evidence in support of the “creative destruction” perspective and further delves into the complexities of the relationship. Rather than a unilateral suppression, we contend that natural disasters may serve as pivotal moments for entrepreneurial responses. The heightened uncertainty and shifting landscapes during natural disasters create opportunities for innovative adaptations and the identification of unmet needs, as also for the long-term entrepreneurial culture (Brück et al., 2010; Linnenluecke & McKnight, 2017; Marino et al., 2008). This challenges the notion that entrepreneurial endeavors universally wane in crisis situations, and contributes to a more comprehensive discourse on the multifaceted dynamics between crises and entrepreneurship, revealing that natural disasters can set in motion either an upward or downward spiral of entrepreneurship and that some of this heterogeneity can be explained by the ecosystem’s financial resource endowments.

Secondly, we reconceptualize literature on the meaning of “savings” from a resource to an indicator of loss aversion, and provide a counter-intuitive insight by delineating regional per capita savings as a key boundary condition for the positive and negative impacts of natural disasters on entrepreneurship. Literature on ecosystems has been criticized for still lacking a strong theoretical foundation (Wurth et al., 2022). Perhaps as a result, theoretical understanding of how financial capital endowments might influence a region’s resilience remains limited, with extant literature mainly viewing regional savings as a resource that promotes entrepreneurship (Doern, 2017; Liguori et al., 2019; Vasilescu, 2014). By considering the behavioral mechanisms at play using prospect theory, this study offers the novel insight that regional personal savings actually suppress entrepreneurship in the wake of a natural disaster by triggering loss aversion, adding much needed theoretical depth to the entrepreneurial ecosystem literature. This insight directly responds to calls to better understand mechanisms that might alleviate or exacerbate the adverse consequences of crises on businesses and economic recovery (Van der Vegt et al., 2015).

Besides, we further provide a dynamic perspective on entrepreneurial ecosystems to clarify how regional personal savings influence the link between natural disasters and entrepreneurship over time. The entrepreneurial ecosystem literature has been criticized for applying a static framework that describes relations in entrepreneurial ecosystems without considering the “complexity of the dynamics” (Brown & Mason, 2017, p.26). The dynamic entrepreneurial ecosystem framework needs to make explicit which elements and relations matter and how they influence each other over time (Alvedalen & Boschma, 2017). To understand the evolution of entrepreneurial ecosystems, Shwetzer et al. (2019) provides a dynamic perspective reflecting value creation, the entrepreneurial activity of entrepreneurial ecosystem elements, and relational interactions within institutional environments. Cantner et al. (2021) takes an industry lifecycle perspective and elaborates a dynamic entrepreneurial ecosystem that leads to their birth, growth, maturity, decline, and re-emergence. Wurth et al. (2022) positions dynamic ecosystems in a broader context, within and beyond the domain of entrepreneurship research, and proposes a transdisciplinary research program for ecosystem research and practice. Despite these important contributions, current literature still lacks a theoretical foundation that addresses what might enable or constrain entrepreneurial ecosystems in recovering from and adapting to external shocks (Iacobucci & Perugini, 2021; Roundy et al., 2017). We add to this knowledge void by providing a dynamic account of how regional personal savings influence the relationship between natural disasters and startups over time, thereby addressing recent calls for research that considers the complexity and dynamics of entrepreneurial ecosystems (Fotopoulos, 2023; Wurth et al., 2022).

Thirdly, we contribute to prospect theory by revealing how reference points are created and changed in the entrepreneurship process. Prospect theory involves a fundamental assumption centered around the reference point, relative to which outcomes are perceived as gains or losses. But what exactly determines the reference point and how the reference point changes have been left unspecified, which is regarded as a major shortcoming of prospect theory (Fudenberg, 2006) because it renders the theory “difficult to falsify” (Werner & Zank, 2019, p.731). Empirical evidence suggests that people may use different reference points in different contexts, and identifying the reference point in different contexts has become a key challenge that needs to be addressed to advance prospect theory (Bromiley, 2010). To solve this shortcoming, the literature has studied how reference points are set in the stock market (Meng & Weng, 2018), business strategy (Figenbaum & Thomas, 1986), accounting (Jegers, 1991), and other contexts. Our research adds to this line of work by revealing that, at least in periods of crisis, a region’s household savings serves as a key reference point for people considering to pursue entrepreneurship in that region. We encourage future scholarship in this domain to build on these insights and consider other possible reference points that are salient in the aftermath of a crisis, thereby making prospect theory more empirically testable and applicable in entrepreneurship research.

This theoretical extension also highlights the broader potential of prospect theory for generating a better understanding of various consumption-savings decisions (Barberis, 2013). Research on the savings—risk-taking link, for example—has advocated that wealthier individuals are generally more prone to invest in riskier assets because they can better absorb the potential losses (Calvet & Sodini, 2014). Intriguingly, our study indicates that this pattern may in fact be the opposite when a crisis hits and people face a sudden major increase in uncertainty about their future income. We find that this specific situation particularly triggers loss aversion among wealthy individuals who now focus on protecting their savings by adopting a wait-and-see approach toward weathering the crisis. The finding may initially appear consistent with research that has observed a motive for accumulating precautionary savings to deal with heightened uncertainty (Kőszegi & Rabin, 2009; Lugilde et al., 2019). However, we illuminate that precautionary saving does not seem to be a feasible option for individuals entering a crisis context with limited means. For these individuals, actual survival is the highest priority, which makes the potential upside associated with entrepreneurship much more salient. As such, this deviation from prior research is driven by adjusted reference points and diminishing sensitivity in a crisis context. Individuals with low initial savings exhibit increased sensitivity to further gains due to their precarious financial status (Ruggeri et al., 2020). This can lower the deterrent effect of potential entrepreneurial losses, making them more willing to take risks. Saving involves risk aversion, and entrepreneurship often requires risk taking (Peterson et al., 2021; Wake et al., 2020). By distinguishing between consumption-saving decisions and entrepreneurship decision-making, our research offers a comprehensive framework for understanding individual responses to uncertainty.

5.2 Research design implications

Entrepreneurial ecosystems are macro-dynamic systems. Current research usually uses cross-sectional data and adds time lags to study dynamic effects (e.g., Boudreaux et al., 2019; 2022; 2023; Brück et al., 2010), which comes with several shortcomings. Firstly, relying solely on cross-sectional data may lead to heterogeneous conclusions. To generate additional insights, it is imperative to conduct a comprehensive comparison by synthesizing longitudinal data. Secondly, macro-level studies are often influenced by multiple unobserved factors, including social, economic, cultural, and political dimensions. Traditional regression analysis cannot account for these potentially omitted variables, potentially resulting in omitted variable bias (Antonakis et al., 2010). Thirdly, introducing time lags in regression models may lead to overfitting (Tschernig & Yang 2000), multicollinearity (Chatelain & Ralf, 2014), and spillover effects (Zhang et al., 2020). This can increase standard errors of regression coefficients, reducing the ability to assess the significance of lagged effects.

Our research studies this issue by using a DDD model in a quasi-natural experimental design with longitudinal data. This approach helps to control for potential omitted factors that might affect the outcomes (Roberts & Whited, 2013). By comparing changes in outcomes over time between a treatment group and a control group in longitudinal data, DDD models provide a more robust analysis of how the relationship evolves over time (Stuart et al., 2014). Our experimental design aims and directly responds to recent calls to enhance rigor in quantitative entrepreneurship research (Maula & Stam, 2020), paving the way forward for future work to further refine this design.

5.3 Economic significance and practical implications

Our findings have significant economic implications, particularly in the realm of policy interventions and economic recovery. Different from the traditional view that natural disasters inhibit entrepreneurship, we find that the startup rate in seriously affected counties increases by 14.7% compared to slightly affected counties. By recognizing that natural disasters may serve as catalysts for developing the resilience of entrepreneurial ecosystems, policymakers can design support mechanisms that empower people to navigate crises and seize emerging opportunities. An in-depth dynamic comprehension of the nuances of regional savings allows for the development of targeted policies that address specific needs during crisis periods, whether through financial support, regulatory adjustments, or the facilitation of collaborative networks. Specifically, by understanding that high savings can increase risk aversion during crises, policymakers can design strategies that shift the reference point from savings to an entrepreneurial culture. This can be achieved through education, community support, and government incentives that encourage risk taking and innovation. Such approaches can lead to more robust economic growth and recovery post-crisis. Additionally, maintaining a balance between financial stability and economic dynamism is crucial. Previous research highlights that high savings provide a safety net (Cordero, 2023; Islam et al., 2018), but they can also dampen economic dynamism if they lead to excessive risk aversion. By fostering a supportive entrepreneurial ecosystem, policymakers can promote productive investments and inspire more entrepreneurial activity, thus enhancing overall economic resilience. In essence, our findings caution policymakers to simply encourage households to “save for a rainy day” but instead underscore the need to craft interventions that can reduce people’s loss aversion in the face of adversity.

5.4 Limitations and further research

Despite the theoretical and practical contributions, the study is not without limitations. First, we only examined one natural disaster so naturally it is not immediately clear to what extent our results are generalizable to other spatial and temporal contexts. Future research should therefore consider other types of disasters in other regions where personal savings might also play a different role. Second, future research should directly test the proposed mechanism of loss aversion associated with savings endowments, possibly using individual-level data. This line of work could also examine what factors might suppress these behavioral tendencies, turning wealth into an asset rather than a liability during times of crisis. Third, we only considered the role of regional personal savings as a focal boundary condition, but clearly, ecosystems vary in terms of many different elements. Future studies could build on our work to examine how ecosystems with varying characteristics and configurations of elements might respond differently to natural disasters and what other explanations might exist for the divergent pathways of development that different ecosystems experience in the wake of a disaster. Fourth, we only considered new firm formation as a key indicator of ecosystem resilience, but natural disasters could also impact other entrepreneurship-related outcomes such as the survival or innovativeness of new ventures. Future work may fruitfully build on our study by examining these other outcomes and what are the unique conditions that enable or constrain these distinct outcomes in the aftermath of a disaster.

6 Conclusion

As global climate change is progressing at an unprecedented pace, natural disasters are expected to occur more frequently and with much greater impacts than one could ever imagine. While this may seem like a very gloomy prospect, this study fortunately offers some hope in that it offers important insights into what allows some regions to quickly recover through entrepreneurship while others languish. Perhaps the positive news is that personal savings appear to limit ecosystem resilience, meaning that financial capital constraints may not always be the most pressing concern for policymakers who seek to implement policies to encourage entrepreneurship as part of broader economic recovery strategies after crises. We hope future entrepreneurship scholarship will benefit from these insights, redirecting efforts from a focus on the enabling role of resource endowments toward consideration of the behavioral implications of such endowments.