Will air pollution affect entrepreneurial activity? Evidence from China

ABSTRACT

This is the first study to examine the empirical evidence on the effect of air pollution on entrepreneurial activity, which fills the research gaps in the relevant literature and provides perspectives on both pollution avoidance behaviour and influence on entrepreneurial activity. Moreover, we explored the innovative and unconventional big data of entrepreneurial activity, which was measured at the county level from 2000 to 2019, and used thermal inversion as an instrumental variable to reduce endogenous bias. Based on the regression analysis, haze pollution inhibits entrepreneurial activity. Meanwhile, the results also show geographic differences. In underdeveloped and non-resource-based regions, the inhibiting effect is more severe. Therefore, air pollution management is an adequate assurance of increasing entrepreneurial activity.

KEYWORDS:

• Air pollution
• entrepreneurial activity
• thermal inversion
• regional development

JEL CLASSIFICATION:

• K32
• M13
• O13
• R11

I. Introduction

Over the years, there have been considerable discussions on air pollution, especially in developing countries such as China. Numerous studies have demonstrated that air pollution, particularly haze pollution, adversely affects human health as well as economic development. Particularly, it causes respiratory ailments, cardiovascular diseases, strokes, and several types of cancers, leading to premature deaths (Sharma, Chandra, and Kota 2020). Moreover, it induces depression (Zhang, Zhang, and Chen 2017), cognitive decline (Lai et al. 2021) and a decrement in labour productivity (He, Liu, and Salvo 2019). These physical and mental health damages inevitably lead to an increase in medical expenditure (Williams and Phaneuf 2019) and pollution prevention expenses incurred towards, for example, anti-smog face masks and air cleaners (Zhang and Mu 2018), and migration costs (Chen, Oliva, and Zhang 2022). Besides, interestingly, several scholars have paid attention to the newly found impacts of air pollution on risk tolerance and pessimistic expectations. For example, Klingen and van Ommeren (2021) find that haze pollution increases stress hormones and causes people to take fewer risks. Further, haze pollution decreases business participation and investment efficiency due to pessimism (Zhang et al. 2019; Zhang, Jiang, and Guo 2016). Based on preceding studies, we believe that haze pollution may also increase the financial and psychological costs of entrepreneurial activity, which deserves our attention.

Is haze pollution an obstacle to entrepreneurial activity? Is the effect geographically different? Our study contributes to several strains of the relevant literature. To our knowledge, this is the first empirical study to examine air pollution avoidance in entrepreneurial activity, which is the core of our understanding of economic growth. Our analysis contributes to the growing literature that seeks to understand entrepreneurial activity from a new perspective, particularly in developing countries. In addition, we contribute to the approach of exploring unconventional data from the Internet, which is crucial for studying the behavioural responses to air pollution in developing countries where there is lack of data (He, Luo, and Zhang 2022). Furthermore, the use of an instrumental variable (IV) and the long-duration panel in two-way fixed effect models reduce endogenous bias caused by omitted variables, measurement errors, and bidirectional causality.

II. Materials and methods

Data and variables

Our research sample includes 2877 counties in mainland China, excluding Taiwan, Hong Kong, and Macao, from 2000 to 2019. The explained variable, entrepreneurial activity, is measured by the number of newly registered companies and self-employers (denoted by Newfirm and SelfEm in the regression) from the Aiqicha enterprise statistics database (https://aiqicha.baidu.com/). The primary explanatory variable is air pollution measured by the annual mean concentration of fine particulate matter (PM25; particles smaller than 2.5 microns) given by the Atmospheric Composition Analysis Group at Washington University in St. Louis. The instrument variable, thermal inversion (Inver) (counting the annual number of days with at least one thermal inversion), comes from the Modern-Era Retrospective Analysis for Research and Applications, version 2, created by the NASA Global Modelling and Assimilation Office. It is a widely used technique to solve the endogenous problem of air pollution effects (Chen, Oliva, and Zhang 2022; Fu, Viard, and Zhang 2021). In addition, the model needs to consider other factors, such as local weather conditions and business environments. The National Meteorological Information Center monitoring spots provides the annual mean temperature (Temp) and precipitation (Precip) data. Additionally, the county statistical yearbook provides the proportion of secondary and tertiary industry value in the GDP (2&3GDP), local government expenditure-revenue ratio (FER), and per capita household savings (Savepc). In Table 1, we present the descriptive statistics.

Table 1. Descriptive statistics.

Variables	Unit	Obs.	Mean	Std. Dev.	Min.	Max.
Newfirm	Count	57,540	2974	5570	0	217,242
SelfEm	Count	57,540	1859	3524	0	198,127
PM25	ug/m^3	56,460	44.47	19.2	1.102	139.2
Inver	Days/year	56,080	130.5	63.22	0	291
Temp	Centigrade	56,420	13.73	5.387	−4.063	26.75
Precip	mm/m^2	56,460	945.9	452.1	0	3389
2&3GDP	Ratio	44,413	0.769	0.142	0.111	1
FER	Ratio	46,809	5.866	61.06	0.106	8157
Savepc	Yuan/person	41,288	14,706	17,324	0.75	356,480

Model specification

We construct the following empirical equation to test the hypothesis:

$Y_{\mathit{it}}={\beta }_0+{\beta }_1\mathit{PM}{2.5}_{\mathit{it}}+{\beta }_2\mathit{Contro}{l}_{\mathit{it}}+{\mu }_t+{\rho }_i+{\epsilon }_{\mathit{i},\mathit{t}}$

here, $Y_{\mathit{it}}$, $\mathit{PM}{2.5}_{\mathit{it}}$, and $\mathit{Contro}{l}_{\mathit{it}}$ represent the number of newly official registered companies and self-employers, the concentration of PM2.5, and the control variables of local weather conditions and business environments of the i-th city in the t-th year, respectively. Moreover, ${\mu }_t$, ${\rho }_i$, and ${\epsilon }_{\mathit{i},\mathit{t}}$ represent the time fixed effect, entity fixed effect, and the error term, respectively. Our panel two-way fixed effects model can effectively control all time-invariant omitted variables; meanwhile, we use an IV to reduce endogenous bias due to the measurement error in the explanatory variable resulting from a lack of accurate information and the possible bidirectional causality between the explanatory variable and the response variable.

III. Results

Regression results and heterogeneity analysis

Table 2 shows the regression results. As we gradually introduce control variables in columns (1) and (4), we find that the coefficient of PM2.5 is always negative. This indicates that air pollution inhibits local entrepreneurial activity, which is consistent with the pollution avoidance behaviour phenomenon (Graff Zivin & Neidell, 2009).

Table 2. Basic regression and heterogeneity.

	(1)	(2)	(3)	(4)	(5)	(6)
Dependent variables	lnNewfirm	lnNewfirm	lnNewfirm	lnSelfEm	lnSelfEm	lnSelfEm
lnPM25	−0.433***	−0.446***	−0.428***	−0.487***	−0.824***	−0.457***
	(0.0514)	(0.0751)	(0.0539)	(0.1310)	(0.1789)	(0.1350)
lnTemp	0.109**	0.085	0.083	0.114	0.064	0.064
	(0.0545)	(0.0534)	(0.0531)	(0.0836)	(0.0823)	(0.0821)
lnPrecip	−0.160***	−0.111***	−0.109***	−0.294***	−0.238***	−0.220***
	(0.0269)	(0.0255)	(0.0256)	(0.0652)	(0.0656)	(0.0651)
ln2&3GDP	0.534***	0.516***	0.511***	0.868***	0.899***	0.885***
	(0.0963)	(0.0886)	(0.0886)	(0.2404)	(0.2367)	(0.2370)
lnFER	0.039**	0.015	0.014	0.222***	0.200***	0.194***
	(0.0185)	(0.0177)	(0.0177)	(0.0515)	(0.0517)	(0.0516)
lnSavepc	0.215***	0.230***	0.231***	0.259***	0.250***	0.256***
	(0.0329)	(0.0300)	(0.0300)	(0.0643)	(0.0666)	(0.0665)
lnPM25*dev		0.156**			0.750***
		(0.0703)			(0.1569)
lnPM25*res			0.261***			0.390**
			(0.0621)			(0.1538)
_cons	6.352***	5.755***	5.686***	5.007***	4.413***	4.191***
	(0.3962)	(0.3672)	(0.3655)	(0.8322)	(0.8430)	(0.8397)
Time&County FE	Yes	Yes	Yes	Yes	Yes	Yes
N	39357	37,965	37,965	39,538	38,039	38,039
adj. R-sq	0.746	0.764	0.764	0.582	0.597	0.597
F	805.943	817.000	809.397	420.312	413.540	412.577

* p < 0.1, ** p < 0.05, *** p < 0.01. The robust standard errors clustered at the county level are in parentheses.

Additionally, the inhibitory effect of air pollution may differ depending on the region in which the counties are located (Chen, Oliva, and Zhang 2022; Li, Chen, and Li 2020). To examine whether there is heterogeneity at the county level, we add the interaction terms PM2.5*dev and PM2.5*res into the regression. Specifically, dev represents the dummy variable for eastern and central developed areas. When the county is in an eastern or central developed region, it is taken as 1, and if it is in a western underdeveloped region, it is taken as 0. Similarly, res represents the dummy variable for resource-based counties. The results are shown in columns (2), (3), (5), and (6), which confirm that geographic differences can affect the relationship between air pollution and entrepreneurial activity. Air pollution has a weaker negative effect on entrepreneurial activity in eastern and central developed areas and resource-based regions. Thus, air pollution increases the inequality in entrepreneurial opportunities and costs in different regions with varying levels of economic development and resource abundance.

Endogeneity issue and robustness checks

Based on the two-stage least square method, we use thermal inversion as an IV to eliminate endogeneity due to the potential bidirectional causality between air pollution and entrepreneurial activity. Since thermal inversion is a random weather variation, it is directly related to haze pollution but is unrelated to the other factors affecting human economic activity (Chen, Oliva, and Zhang 2022; Fu, Viard, and Zhang 2021). Accordingly, in columns (1) and ⁠(2) of Table 3, we find that air pollution still has a significant negative impact on entrepreneurial activity, and the estimated coefficients using the IV approach are larger. It should be noted that in related studies in both developed and developing economies, IV estimates are usually larger than OLS estimates due to the measurement error of air pollution (He, Liu, and Zhou 2020).

Table 3. Regression results based on the two-stage least square method and robustness tests.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	lnNewfirm	lnSelfEm	Newfirm	SelfEm	lnNewfirm	lnSelfEm	lnNewfirm	lnSelfEm
lnPM25	−2.527***	−5.767***			−5.099***	−4.699***	−2.785***	−6.680***
	(0.6114)	(1.6369)			(0.7867)	(0.8873)	(0.5609)	(1.5114)
PM25			−413.307***	−270.339***
			(49.3807)	(34.6760)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time&County FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	39106	39,287	39,287	39,287	32,793	32,384	38,046	38,046
adj. R-sq	0.720	0.538	0.045	0.152	0.620	0.689	0.730	0.534
F	735.404	342.032	93.359	104.959	557.755	617.020	760.696	328.419

* p < 0.1, ** p < 0.05, *** p < 0.01. The robust standard errors clustered at the county level are in parentheses.

Furthermore, to test the robustness of our findings, we convert the dependent and independent variables to their non-logarithmic forms in columns (3) and (4) of Table 3. Besides, we winsorize the sample at the 5th and 95th percentiles in columns (5) and (6), and transfer the dataset into a balanced panel data in columns (7) and (8). The regression results stating that worsening air pollution inhibits entrepreneurial activity are robust.

IV. Discussion

Based on the PM2.5 data and the registration information of start-up companies and self-employers at the county level in China from 2000 to 2019, this study discusses the effect of haze pollution on entrepreneurial activity. We have found that haze pollution inhibits entrepreneurial activity, and the inhibiting effect is more severe in underdeveloped and non-resource-based regions. The study is the first comprehensive examination of the impact of air pollution on entrepreneurial activity. Its strengths include its long duration and the unconventional and innovative use of big data to study entrepreneurial activity. The findings provide new insights on pollution avoidance behaviour and add to the substantially increasing number of studies on factors influencing entrepreneurship. However, a limitation of this study, arising from restrictions with regard to article length and data, is that a more in-depth investigation of the influencing mechanism of entrepreneurial behaviour could not be conducted; therefore, further research should be conducted with micro-data.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Appendix

In Table 4, according to the first stage of 2SLS regression results of columns (1)-(2) in Table 3, the t-test and F-test values, underidentification test, and weak identification test are statistically significant, so the instrumental variable is valid.

Table 4. Summary results for first-stage regressions of (1)-(2) in Table 3.

Dependent variable: thermal inversion	Time&County fixed effects estimation with control variable				Underidentification test		Weak identification test
	Coef.	t	F(1,2113)	P-value	Stock-Yogo Chi-sq (1)	P-value	Stock-Yogo F(1, 2113)	10% maximal IV size
lnPM25 (lnNewfirm)	0.0005565***	12.18	148.34***	0.0000	148.50***	0.0000	148.34***	16.38
lnPM25 (lnSelfEm)	0.0005423***	11.83	140.06***	0.0000	140.22***	0.0000	140.06***	16.38

* p < 0.1, ** p < 0.05, *** p < 0.01. Statistics robust to heteroskedasticity and clustering on county level.