The Impact of Risk Aversion and Migrant Work Experience on Farmers’ Entrepreneurship: Evidence from China

Abstract

Investing in entrepreneurship may be costly, and therefore risky, and entrepreneurship is also an economic endeavor that is highly dependent on entrepreneurial ability and risk appetite. In this study, data from 669 famers in southwest China were used as the sample, and we used three different methods to measure farmers’ risk aversion level, including DOSPRET (Domain-Specific Risk-Taking), SOEP (Simple Self-Report), and BRET (Bomb Risk Elicitation Task). This paper studies the impact of farmers’ risk aversion on entrepreneurial choices and the moderating impact of the migrant work experience (MWE) on the relationship between risk aversion and entrepreneurship. The results can be summarized as follows: Firstly, entrepreneurial farmers have lower average levels of risk aversion than non-entrepreneurial farmers. Secondly, risk aversion has a significant negative impact on farmers’ choice of entrepreneurship, corporate entrepreneurship and portfolio entrepreneurship. Thirdly, MWE can reduces the negative impact of risk aversion on a decision to enter entrepreneurship or portfolio entrepreneurship. Fourth, MWE in local or nearby areas reduces the negative effects of risk aversion on entrepreneurship and portfolio entrepreneurship, while MWE in coastal and developed cities increases the negative effects.

1. Introduction

Entrepreneurship holds significant importance not only in terms of fostering innovation and facilitating long-term economic growth within a nation [1,2], but also plays a crucial role in tackling employment challenges prevalent in society and optimizing the employment situation [3,4,5]. Moreover, cultivating and promoting of entrepreneurship in rural regions is deemed the primary driving force of economic development in rural areas [6,7]. Due to the ongoing growth of the rural economy, Chinese farmers have more diversified entrepreneurial choices, various new types of agricultural and non-agricultural business entities are developing rapidly, and new trends of rural entrepreneurship are constantly emerging. Firstly, “Agricultural entrepreneurship” and “Non-agricultural entrepreneurship” have developed together [8]. Secondly, “Self-employed entrepreneurship” and “corporate entrepreneurship” are emerging at the same time [9]. Thirdly, an increasing number of migrant farmers are opting to return to their home villages to set up their businesses [10]. The primary challenges faced by Chinese farmers in the realm of entrepreneurship have yet to be successfully overcome, including credit constraints and poor ability to identify entrepreneurial opportunities. These issues are chiefly exhibited in the following ways: On the one hand, farmers appear to lack enthusiasm to embark on entrepreneurial activities. According to the statistical data from the China National Bureau of Statistics, China’s migrant worker population amounted to 292 million individuals in 2021; in contrast, during the same period in 2022, only about 9.9 million returned to their home villages to embark on entrepreneurial activities, constituting a mere 3.38% of the previous year’s total migrant worker population. Lack of innovation and mature skills are important factors hindering entrepreneurship among Chinese farmers. On the other hand, the marginalization and insufficient level of farmers’ entrepreneurship present obstacles to business expansion, resulting in weak growth of operating income. Addressing the challenge of farmers’ entrepreneurship necessitates tackling the issue of the indigenous drive of such entrepreneurship. At the same time, we can also find some favorable signs for rural entrepreneurship in the employment trends of Chinese migrant workers (Figure 1). The number of migrant workers in China has almost always been on an upward trend, with some choosing to stay home to work in agriculture in 2020 due to the impact of COVID-19. The number of migrant workers choosing to work in other provinces has been declining since 2014, as local jobs are becoming more attractive to migrant workers due to the growth of China’s inland urban economy, as well as the economy of rural areas.

The concept of “Risk attitude” is a classic underlying assumption in the field of economics. Risk attitudes affect individual decision-making in a myriad of diverse contexts [11]. Being less susceptible to risk aversion is deemed a highly noteworthy “individual characteristic” that motivates entrepreneurial pursuits [12], and individuals with lower risk aversion may be inclined towards speculative activities [13]. Similarly, farmers’ risk attitudes play a fundamental role in entrepreneurial selection. Moreover, the probability of farmers embracing novel agricultural production techniques and forms of production organization increases with the decrease in risk aversion. Due to the existence of China’s urban–rural dual economic system, farmers who work in cities cannot have access to equivalent public welfare as urban residents. With the optimization of rural economic policies, the number of migrant workers returning to their hometowns has increased year by year due to various reasons. Migrant workers’ working experience enables them to accumulate physical and human capital and expand their access to information [14,15,16]; such resources are potential entrepreneurial motivation. However, the endogenous motivation of farmers’ entrepreneurship remains to be further discussed because of the low level of Chinese farmers’ average education, the prevalence of risk-neutral and risk-averse preference attitudes, the weak entrepreneurial consciousness, and the lack of pioneering spirit.

Given this, it is of great practical significance to investigate in depth the influence of farmers’ risk aversion on entrepreneurial behavior and whether migrant work experience alleviates the inhibitory effect of farmers’ risk aversion on entrepreneurship.

2. Literature Review

2.1. Risk Preferences and Its Measurement

Risk preference theory belongs to a particular branch of risk theory proposed by Friedman and Savage in 1948 [17]. Markowitz [18] conducted an analysis of this theory in 1952, while Pratt [19] and Arrow [20] further discussed measures of risk preference, respectively. Subsequently, Ross refined the theory in 1981 [21]. Risk aversion is derived from risk attitude, where individuals who are averse to risk have difficulty adapting to uncertainty and display a low tolerance for ambiguity. Such individuals are inclined towards accepting and consciously enjoying established realities rather than embracing new theories or innovative methods; thus, risk aversion leads to a preference for less risky choices [22]. The existing literature on measuring risk aversion can be broadly delineated into three distinct categories, contingent upon both the context within which the respondents are evaluated and the design of the measurement questions. The first category is the independent reporting method. This category can be further subdivided into three specific methods: Simple independent reporting, the situational questioning method, and the comprehensive evaluation method. (1) Simple independent reporting is a method in which participants are posed limited inquiry-based questions regarding their risk attitude. For example, Caliendo et al. measured the existence of risk aversion through a predetermined question sourced from the data of the German Socio-Economic Survey (GSOEP) [23]. “How do you see yourself: a person who is fully prepared to take risks or tries to avoid taking risks?”. Some scholars used the simple self-report method [24,25] in early population research and entrepreneurial research, but this method is also considered to be the risk measurement method with the largest error. (2) The situational questioning method is a method in which respondents’ risk attitudes are measured by choosing from presented options with different levels of risk, such as hypothetical investment opportunities and betting opportunities [12,26]. (3) The comprehensive evaluation method entails posing inquiries that employ indicators that indirectly measure an individual’s risk aversion across different aspects, including personal life, financial matters, healthcare, and safety. The most representative comprehensive evaluation method is DOSPERT (Domain Specific Risk-Taking Scale), proposed by Weber in 2002 [27]. The second category is the proxy variable approach. Out of distrust of autonomous reporting methods, some scholars use explicit variables of individuals or organizations in specific products and business environments to approximate risk aversion level. For example, Brown uses a portfolio of entrepreneurs and constructs alternative indices that express risk aversion [28]. Another example is Hvide and Panos, who use stock market participation, individual leverage, and the proportion of wealth invested in the stock market to measure risk aversion [29]. The third category is the controlled experiment method. The current mainstream experimental methods to measure risk aversion are mainly “lottery selection experiments” and “risk game experiments”. The “lottery choice experiment” is represented by the method of Holt and Laury [30]. Some scholars have used experimental gambling choice methods to measure risk aversion, furnishing a set of deterministic matters and four additional options with linearly increasing risk and payout [31,32]. The BRET (Bomb Risk Elicitation Task) experiment, proposed by Crosetto and Filippin in 2013 [33], is also representative of a risk game experiment.

2.2. Risk Aversion, Migrant Work Experiences, and Farmers’ Entrepreneurship

Some researchers have studied the relationship between risk aversion, migrant work experience, and farmers’ entrepreneurship. Kihlstrom and Laffont were the first to construct a general entrepreneurial equilibrium model and predict the effect of risk-taking on employment patterns [34]. Based on this model, Feng and Rauch developed an available simplified form for the entrepreneurial equilibrium model and extended it [35]. Astebro et al. outline risk preference within the standard expected utility model, highlighting that an individual’s risk preference constitutes a pivotal factor in determining the farmers’ entrepreneurship [36]. Risk preference can facilitate the adoption of new technology and farmers’ entrepreneurship [37,38]. Farmers less averse to risk are more likely to embrace innovative alternatives for continuing agricultural projects with potential termination [39]. Several scholars have concluded that risk aversion reduces the likelihood of new technology adoption, based on research data from Ethiopia, Nigeria, and Indonesia [37,39]. The study results in China are similar to those of other countries. Entrepreneurs have a requirement to seek risk-sharing opportunities to undertake entrepreneurial risk effectively [40]. Moreover, risk propensity has a significant positive influence on entrepreneurial intention [41,42]. Due to the more considerable expected investment, it is a greater challenge for adventurous farmers to start businesses in their hometowns, and many are less likely to do so [43]. One study revealed that farmers who prefer risk tend to opt for industrial and commercial entrepreneurship, whereas risk-averse individuals are more inclined towards agricultural entrepreneurship [44]. Furthermore, the positive effect of risk preference on Chinese farmers’ entrepreneurial propensity is more significant among the new generation of migrant workers born after 1978 [45].

Regarding the potential effects of migrant work experience on farmers’ entrepreneurship, pertinent research conducted on a global scale has predominantly centered on migration. In contrast, not much research has been dedicated to studying those farmers who migrate to cities and then return to their home villages to start businesses. Mesnard highlights that under certain conditions, migrant work can allow farmers to accumulate more physical and human capital, facilitating migrants to return home villages for entrepreneurship [46]. McCormick and Wahba contend that there is heterogeneity in migrants’ human capital accumulation and human capital has no significant effect on returning migrants with low levels of education [47]. Wahba and Zenou argue that the phenomenon of out-of-state workers results in a depletion of social capital in their respective home villages, but this adverse impact is effectively compensated for by other forms of capital [42]. In the Chinese institutional environment, it is common for farmers to work outside and return to their home villages [48]. Scholars maintain two perspectives concerning the correlation between these two: on the one hand, according to Murphy, migrant work can foster the farmers’ comeback to enterprise in various ways [49]. In contrast to local left-behind villagers, migrant workers receive substantial remuneration, leading to a greater initial endowment [50]. Moreover, migrant work experience can assist migrant workers in overcoming challenges such as limited entrepreneurial opportunities and false information [51]. The facilitating effect is also particularly seen in off-farm entrepreneurship [14]. On the other hand, undoubtedly, long-term migrant work damages hometown social capital, which is closely related to local society and kinship, and causes the loss of personal credit, thus inflicting some negative effects on returning migrants’ entrepreneurship [10], reducing their entrepreneurial success rate [52]. It is worth noting that returning migrants who cannot establish themselves in urban areas because of economic factors or policy restrictions have the same attitude toward self-employment as farmers who stay in their hometown [48].

3. Theoretical Hypothesis

As mentioned in the literature review, farmers with higher levels of risk preference have higher risk tolerance thresholds. Consequently, when faced with entrepreneurial decisions necessitating the absorption of risk shocks, farmers with high-risk preference are more likely to accept the entrepreneurship risks. Secondly, according to Knight et al. [37], farmers with greater risk tolerance exhibit greater confidence in modern technologies and are more inclined to embrace novel production techniques. Consequently, adopting new technologies fosters modifications in the scope and structure of farming practices that can potentially stimulate farmers’ entrepreneurial activities. Finally, identifying opportunities is a necessary precondition for entrepreneurial choice. The more risk-averse farmers are inclined to avail themselves of potentially lucrative opportunities in uncertain production and operational activities. Based on the above analysis, we propose the following hypothesis (H1):

H1. 

Farmers are more inclined to choose entrepreneurship (non-farm entrepreneurship or farm entrepreneurship) as their level of risk aversion decreases.

Farmers can choose to start a family-based entrepreneurship and do not need to register for a business license. In addition, farmers can choose corporate entrepreneurship and need to register a business license, which allows them to carry out larger scale production and commercial operations, as well as access to more legal protection. Corporate entrepreneurship has the advantage of higher selling price and lower scale production cost, but company entrepreneurship has higher requirements for farmers’ operational capacity and risk management ability. Based on the prevailing concept of risk attitude, it can be inferred that farmers with lower levels of risk aversion are more inclined to accept higher management expenses, including the policy cost of business registration, labor cost, social reputation maintenance cost, etc. Therefore, we propose the following hypothesis (H2):

H2. 

Farmers are more inclined to choose corporate entrepreneurship as their level of risk aversion decreases.

Individual risk aversion also influences their choice of entrepreneurial form. The decision of farmer entrepreneurs to pursue portfolio entrepreneurship is related to their level of innovation and tolerance towards risk. When farmers choose portfolio entrepreneurship, a farming-based livelihood allows them to have leisure time outside the farming season, which will lead them to choose both farming entrepreneurship and off-farm entrepreneurship. Similar to the motivating factors behind farmers’ decision to become entrepreneurs, among the multiple entrepreneurship methods of breeding, planting, and non-agricultural entrepreneurship, farmers with a wider range of entrepreneurial options have better opportunity recognition ability and better risk management capabilities. At the same time, investment risk can be diversified through different entrepreneurial portfolios. The motivations for portfolio entrepreneurship can be categorized into two opposite possibilities: (1) Farmers with lower levels of risk aversion have stronger opportunity recognition and risk management capabilities, and they will tend to choose portfolio entrepreneurship. (2) Farmers with higher levels of risk aversion choose portfolio entrepreneurship in order to diversify investment risks. Based on the above analysis, we propose hypotheses (H3a and H3b):

H3a. 

Farmers are more inclined to choose more complex portfolio entrepreneurship as their level of risk aversion decreases.

H3b. 

Farmers are more inclined to choose more complex portfolio entrepreneurship as their level of risk aversion increases.

Other endogenous and exogenous characteristics are closely related to farmers’ risk aversion in the mechanisms promoting farmers’ entrepreneurship. Rural households’ human and physical capital can also directly affect risk attitudes and entrepreneurial choices. It is assumed that farming households possess a more stable level of risk preference, and farmers with migrant experience compared with local left-behind villagers receive rich labor remuneration. Upon fulfilling their everyday consumption requirements, the residual savings can be utilized for investment purposes, thereby providing returning migrant workers with initial endowment. Moreover, the majority of migrant workers have gained professional skills through long-term practice, rendering them more easily adaptable to entrepreneurial needs. The migrant work experience also improves the ability of migrant workers to obtain market information and reduces the effect of information bias. Therefore, farmers’ long-term migrant work experience may mitigate the inhibitory effect of risk aversion on entrepreneurial choice. Based on the above analysis, we propose the following hypothesis (H4):

H4. 

The long-term migrant work experience can positively moderate the relationship between risk aversion and entrepreneurial choice.

Based on the literature review and analysis above, a graphical framework on risk aversion, migrant work experience, and farmers’ entrepreneurship can be presented (Figure 2).

4. Empirical Design and Descriptive Statistics

4.1. Sample

For this research, we selected the four provinces and cities of Sichuan, Chongqing, Guizhou, and Yunnan in southwest China as the sample area. To conduct the sampling process, a combination of stratified sampling and judgmental sampling was utilized. (1) Based on the economic zoning of each province, the municipalities were divided into regions by geographic location. (2) All counties in each city were ranked according to the mean value of the per capita disposable income of rural residents, and counties above and below the mean value were selected. (3) Probability proportional to size sampling was applied to determine the size of sampled households in each county based on the ratio of the size of the rural population of the sampled counties. (4) In judgmental sampling, the rural households in the village that met the definition of farmer entrepreneurship were identified and randomly surveyed; then, the sample of non-entrepreneurial farmer households were collected in the same village according to income and living situation elements. The judgmental sampling process required the investigator to make observations and judgments as well as obtaining the help of the village headman. During September–November 2019, investigators distributed a total of 700 questionnaires. After screening for contradictory or missing contents, the final number of valid questionnaires was determined to be 669, representing 95.57% of the total sample.

4.2. Risk Aversion Measurement Experiments and Statistics

The Domain-Specific Risk-Taking (DOSPERT) method, initially utilized by Weber et al., is a synthesis measurement of risk attitude based on self-reported autonomy [27]. It adopts a five-point Likert scale with five points, wherein each individual’s likelihood of engaging in risky activities is measured on a scale of one to five. As the principal focus of this study is risk aversion, the inverse metric was employed in the five-point scale. A higher numerical value suggests a heightened degree of risk aversion, while a lower numerical value reflects a greater preference for risk. DOSPERT is a canonical implementation of showing risk preference. Appendix A presents the 17 questions used for the DOSPERT synthesis report, covering five different risk behavior domains: ethics, finance, health and safety, recreation, and social. In order to reduce the “learning effect” of the scale questionnaire, the questions were disorganized into a random order during the research.

A five-point Likert scale and a weighted average method were applied to calculate the DOSPERT index. As shown in

Table 1. Statistics of DOSPERT.

	Risk Aversion Type	Obs	Ratio	Mean	Min	Max	SD
All farmers	High	39	5.83%	3.94	3.73	4.47	0.193
	Moderate	586	87.59%	3.00	2.60	3.64	0.325
	Low	44	6.58%	2.23	2.07	2.33	0.090
	Total	669	100%	3.002	2.07	4.47	0.432
Entrepreneurial farmers	High	31	7.81%	3.95	3.73	4.47	0.209
	Moderate	323	81.36%	2.91	2.53	3.67	0.367
	Low	43	10.83%	2.23	2.07	2.33	0.091
	Total	397	100%	2.92	2.07	4.47	0.499
Non-entrepreneurial farmers	High	8	2.94%	3.92	3.73	4.13	0.122
	Moderate	263	96.69%	3.11	2.40	3.67	0.222
	Low	1	0.37%	2.20	2.20	2.20	-
	Total	272	100%	3.13	2.20	4.13	0.265

, the majority of farmers were in the medium risk aversion group, with entrepreneurial farmers having lower average level of risk aversion compared to non-entrepreneurial farmers and a higher percentage of farmers in the low risk aversion group.

4.3. Variables

All defining criteria for farmers’ entrepreneurship are shown in

Table 2. The standard of famers’ entrepreneurship.

Type	Defining Criteria
Non-Farm Entrepreneurship	Rural households have established non-farm enterprises or long-term non-farm self-employed activities
	Rural households engaged in commodity circulation and various commercial services, including but not limited to: catering, logistics, passenger transport, handicrafts, farmhouse operation, rural tourism, grocery stores, retail of agricultural and sideline products, processing of agricultural and sideline products, agricultural services, hairdressing, etc.
	Annual profit of at least CNY 30,000
Plantation/Forestry Entrepreneurship	Grain, oilseed, cotton, and other planting seeding area of 30 acres or more
	Vegetables, fruit (dried fruit), flowers, seedlings, tea, Chinese herbal medicine planting more than 10 acres
	Family management of forest land area of 5 acres or more and clearly have forest land inflow
Breeding Entrepreneurship	Annual slaughter capacity of 30 pigs and above, 100 piglets and above
	Raising more than 10 head of beef cattle, annual slaughter capacity 3 head and above Raising 50 or more meat sheep, annual slaughter capacity of 50 or more sheep
	Raising more than 200 rabbits
	Raising broiler chickens, meat ducks, meat geese with an annual slaughter capacity of 100 or more
	Raising more than 100 laying ducks and laying hens

. The selection and statistical description of the variables are shown in

Table 3. Definition of variables and descriptive statistical analysis results.

Variable	Symbol	Definition	Mean	S.D.
Explained Variables
Whether to enter entrepreneurship	Entrep	0 = No entrepreneurship, 1 = Choose entrepreneurship	0.593	0.492
Corporate entrepreneurship	Corporate	0 = No entrepreneurship, 1 = Corporate entrepreneurship	0.260	0.439
Portfolio entrepreneurship	Portfolio	1 = Single entrepreneurship, 2 = Any portfolio of two entrepreneurial approaches (any portfolio of two entrepreneurial approaches means the combination of any two of planting entrepreneurship, breeding entrepreneurship, or non-entrepreneurship), 3 = Planting + breeding + non-agricultural entrepreneurship	1.139	1.423
Explanatory variable
DOSPERT Index	DOSPERT	Risk aversion index	3.002	0.432
Moderating variable
Long-term migrant work experience	Migration	0 = No, 1 = Yes	0.697	0.460
Control variables
Family dependency rate	Rise	Number of (minors + seniors)/Total number of household members	0.598	0.390
Household income in the previous year	Income	CNY 10,000	17.36	38.29
Gender	Gender	0 = Female, 1 = Male	0.855	0.352
Party membership	Party	0 = No, 1 = CCP members	0.223	0.416
Health level	Health	0 = Unhealth, 1 = health	0.756	0.430
Failed entrepreneurial experience	Experience	0 = No-entrepreneurship, 1 = Once establish entrepreneurship but failed	0.211	0.408
Number of siblings	Sibling	Number of siblings	3.363	1.663
Average age of couples	Age	Age	47.88	9.719
Average years of education for couples	Education	Year	8.196	2.302
Sample size		669

.

There are three explained variables: ① Whether farmers choose entrepreneurship (Entrep); non-entrepreneurial farmers = 0, entrepreneurial farmers = 1. ② Corporate entrepreneurship (Corporate) is defined as the entrepreneurial farmers who have registered a company. ③ In the portfolio entrepreneurship type (Portfolio), farmers’ entrepreneurial portfolios can be selected as farming, breeding, and non-farm entrepreneurship. Single entrepreneurship was assigned a value of 1, a combination of any two entrepreneurship types was assigned a value of 2, and a combination of all three entrepreneurship was assigned a value of 3.

Core Explanatory Variable: Determined by applying a five-point Likert scale and weighted average method to calculate the DOSPERT index.

Moderating Variable: The moderating variable is the presence of long-term migrant work experience, whereby the time condition for meeting the long-term criterion is determined by migrant work experience longer than 1 year.

Control Variables: In addition to the usual variables related to household economic decision-making, such as gender, age, and education, there are several control variables that need to be given more explanation. ① Family dependency rate: Elderly people and children in rural families in China receive fewer benefits than families in urban areas, and rural families with higher family dependency ratios have more financial needs, so this variable affects the entrepreneurial choices of young and middle-aged people. ② Household income in the previous year: Household income status influences the economic decisions of rural households, and retained funds from the previous period are an important basis for starting and sustaining the entrepreneurship. ③ Health level: This variable is the farmer’s self-assessment, standardized by comparison with peers around them. The level of health of the main household labor is an important factor influencing livelihood choices, and the lower the level of health of a farmer, the less able he or she is to carry out entrepreneurial activities. ④ Number of siblings: The effect of the number of siblings on farmers’ entrepreneurship is twofold; on the one hand, more siblings represent more resources and information available in the social network, which has a positive effect on entrepreneurship. On the other hand, farmers with more siblings have less access to parental help, such as educational resources and family assets, which has a negative effect on entrepreneurship.

4.4. Econometric Models and Estimation

4.4.1. Benchmark Model and Moderating Effect Model

Considering the setup of the explanatory variables, the Probit model was selected as the basic econometric model. Equation (1) calculates the effect of risk aversion on farmers’ entrepreneurial choices, and Equation (2) calculates the moderating effect of migrant work experience on risk aversion and farmers’ entrepreneurship. The equations are as follows:

$${Entrep}_i={\alpha }_0+{\alpha }_1{DOSPERT}_i+{\alpha }_{control}Controls+C_i+{\epsilon }_i$$

(1)

$${Entrep}_i={\alpha }_0+{\alpha }_1{DOSPERT}_i+{\alpha }_2{Migrant}_I+{\alpha }_3{DOSPERT\times Migrant}_i+{\alpha }_{control}Controls+C_i+{\epsilon }_i$$

(2)

where ${Entrep}_i$ is the explanatory variable representing farmers’ entrepreneurial choices, which are whether to choose entrepreneurship, and whether to choose the corporate entrepreneurship or portfolio entrepreneurship; ${DOSPERT}_i$ represents the value of risk aversion obtained through the DOSPRET method; ${Migrant}_i$ represents migrant work experiences. ${DOSPERT\times Migrant}_i$ represents the interaction term of risk aversion and migrant work experience. ${\alpha }_1,{\alpha }_2{,and\alpha }_3$ are the estimated coefficients corresponding to the explanatory variables, where ${\alpha }_1$ < 0, ${\alpha }_2$ > 0, ${and\alpha }_3$ < 0 are necessary conditions for the hypothesis to be provable. $Controls$ represents control variables;${\alpha }_{control}$ is the coefficient of the control variable. $C_i$ represents county fixed effects, and ${\epsilon }_i$ represents the random error term.

4.4.2. Robustness Test Processing Model

According to the literature review, there are a variety of measures of risk preference or risk aversion. We chose the simple assessment method (SOEP) and the gambling experiment method (BRET) to replace the DOSPERT index. We established the following robustness test model:

$${Entrep}_i=b_0+b_1{SOEP}_i+b_{2}Controls+C_i+{\epsilon }_i$$

(3)

$${Entrep}_i=b_0+b_1{SOEP}_i+b_2{Migrant}_i+b_3{SOEP\times Migrant}_i+b_{control}Controls+C_i+{\epsilon }_i$$

(4)

$${Entrep}_i={\beta }_0+{\beta }_1{BRET}_i+{\beta }_{2}Controls+C_i+{\epsilon }_i$$

(5)

$${Entrep}_i={\beta }_0+{\beta }_1{BRET}_i+{\beta }_2{Migrant}_i+{\beta }_3{BRET\times Migrant}_i+{\beta }_{control}Controls+C_i+{\epsilon }_i$$

(6)

where ${SOEP}_i$ and ${BRET}_i$ represent the value of risk aversion obtained through the $\mathrm{SOEP}$ method and the BRET method, and ${SOEP\times Migrant}_i$ and ${BRET\times Migrant}_i$ represent the interaction term of risk aversion and migrant work experience. $b_1$, $b_2$, $b_3$ and ${\beta }_1$, ${\beta }_2$, ${\beta }_3$ are the estimated coefficients corresponding to the explanatory variables. $Controls$ represents control variables; $b_{control}$ and ${\beta }_{control}$ are coefficients of the control variable.

4.4.3. Endogenous Processing Model

Due to possible endogeneity problem in the benchmark model, the ratio of household safe assets (non-risky assets) to total assets is used as an instrumental variable to replace risk aversion. In order to avoid the problem of missing asset data for the same period, IV is calculated using asset data from the period prior to the farmer’s entrepreneurial venture. We established the following endogenous processing model:

$$I{V}_i={\gamma }_0+{\gamma }_1{DOSPRET}_i+{\gamma }_{2}Controls+C_i+{\epsilon }_i$$

(7)

$${Entrep}_i={\delta }_0+{\delta }_1{\widehat{IV}}_i+{\delta }_{2}Controls+C_i+{\theta }_i$$

(8)

In Equation (7), $I{V}_i$ is the instrumental variable (IV), indicating the ratio of household safe assets (non-risky assets) to total assets, $C_i$ represents county fixed effects, and ${\epsilon }_i$ represents the random error term, and the rest of the variables are the same as those in the benchmark model. In this equation, $I{V}_i$ is related to ${\epsilon }_i$, and ${\widehat{IV}}_i$ is not related to ${\epsilon }_i$. Equation (7) is the first-stage result of the two-stage estimation (2SLS) of the IV, which aims to test the correlation between the IV and DOSPRET.

In Equation (8), ${\widehat{IV}}_i$ is the fitted value of $I{V}_i$, ${\theta }_i$ represents the random error term, and the rest of the variables are the same as those in the benchmark model. Equation (8) is the second-stage result of the two-stage estimation (2SLS) of the IV, which aims to demonstrate the reliability of the results of the estimation results.

5. Results

5.1. Benchmark Results

The results shown in

Table 4. Results of the benchmark regression model.

	Entrep	Corporate	Portfolio	Entrep	Corporate	Portfolio
	Probit	Probit	Oprobit	Probit	Probit	Oprobit
	(1)	(2)	(3)	(4)	(5)	(6)
DOSPERT	−1.138 ***	−0.892 ***	−0.878 ***	−1.075 ***	−0.646 ***	−0.694 ***
	(0.147)	(0.155)	(0.123)	(0.178)	(0.170)	(0.131)
Rise				0.222	0.447***	0.147
				(0.173)	(0.147)	(0.112)
Sibling				−0.234 ***	−0.0573	−0.167 ***
				(0.0388)	(0.0385)	(0.0300)
Income				0.155 ***	0.102 ***	0.125 ***
				(0.0376)	(0.0308)	(0.0248)
Gender				−0.513 ***	0.314 *	−0.370 ***
				(0.169)	(0.189)	(0.129)
Party				0.503 ***	0.180	0.270 **
				(0.160)	(0.138)	(0.112)
Health				−0.389 ***	−0.206 ***	−0.319 ***
				(0.0607)	(0.0621)	(0.0473)
Experience				0.345 **	0.840 ***	0.297 ***
				(0.160)	(0.135)	(0.113)
Age				0.0083	0.0119 *	0.0072
				(0.00707)	(0.00705)	(0.00546)
Education				0.201 ***	0.112 ***	0.136 ***
				(0.0356)	(0.0301)	(0.0238)
Constant	4.121 ***	2.346 ***		3.674 ***	−0.295
	(0.508)	(0.521)		(0.844)	(0.780)
Fixed Effect	YES	YES	YES	YES	YES	YES
R2	0.0682	0.0448	0.0322	0.3548	0.2365	0.1940
Sample	669	669	650	669	669	650

Note: Values in parentheses are standard errors, * p < 0.01, ** p < 0.05, *** p < 0.01. All regressions control for county fixed effects. The empirical notes in the later section are consistent with this table, so other regressions are not repeated in the later section.

were calculated by Stata 16SE and passed the VIF test and White test. Regressions 1, 2, 4, and 5 employed the Probit model to assess the variables “whether to choose entrepreneurship” and “corporate entrepreneurship”, while Regressions 3 and 6 utilized the Oprobit model to analyze portfolio entrepreneurship. According to the empirical findings presented in Table 4, there is a negative correlation between individual farmers’ risk aversion and their likelihood of choosing entrepreneurship as a career. This relationship is supported by a significant regression coefficient of −1.075, which passes the 1% significance test. This result is in accordance with the theoretical hypotheses as well as the discoveries of Chen Bo and Wang Yong [43,45]. Based on our analysis, it can be concluded that hypothesis 1 is supported. Furthermore, our findings indicate that risk aversion plays a significant role in inhibiting the choice of entrepreneurship and portfolio entrepreneurship at the 1% level with negative coefficients of −0.646 and −0.694, respectively. As a result, we can confirm that hypotheses 2 and 3a are also validated.

Further calculations of the marginal effects yield the results in

Table 5. Marginal effects of benchmark regression.

	Entrep	Corporate	Portfolio	Entrep	Corporate	Portfolio
	(7)	(8)	(9)	(10)	(11)	(12)
DOSPERT	−0.438 ***	−0.284 ***	−0.342 ***	−0.390 ***	−0.188 ***	−0.267 ***
	(0.0562)	(0.0491)	(0.0479)	(0.0634)	(0.0493)	(0.0503)
Controls	NO	NO	NO	YES	YES	YES
Fixed Effect	YES	YES	YES	YES	YES	YES
Sample	669	669	650	669	669	650

Note: Values in parentheses are standard errors, *** p < 0.01.

, which remain largely consistent with the benchmark regression results. Risk aversion has an inhibitory effect on entrepreneurship, which are all significant at the 1% level.

5.2. Robustness Test

As described in the literature review, there are a variety of measures of risk preference or risk aversion. In this section, we chose the simple assessment method (SOEP) and the gambling experiment method (BRET) to replace the data of the DOSPERT; SOEP is more subjective, and BRET is based on experiments in Behavioral Economics. SOEP means that farmers could choose a level of risk aversion among 11 numbers from 0–10. Among them, “0” indicates an extreme risk aversion, and the higher the number, the lower the level of risk aversion. BRET is a “grid-flipping game”, adapted from Crosetto and Fillippin, which provides each participant with a game bonus corresponding to the score [33]. The BRET game needs to be played in the investigator’s mobile phone, with technical support from our team members. The game interface has a 10 × 10 square matrix with 100 squares, and a randomly assigned bomb grid is hidden among these 100 squares. Once the test is initiated, the participant is to flip the grids consecutively at one grid per second for each unit of time. Successfully revealing a grid without a hidden bomb will result in the accrual of 1 point. However, if you flip to the “hidden bomb grid”, the game is over, the total score is zero, and the prize is also zero. Before the official game, the interviewed sample was given 2–5 trial opportunities and the cash reward for each grid was randomly assigned to each questionnaire according to three grades of CNY 0.5, CNY 1 and CNY 2. In the BRET experiment, as the marginal risk to be taken for each additional benefit unit is increasing, we reverse-converted the value of risk preference obtained from the BRET experiment into the value of risk aversion.

The regression results, as shown in

Table 6. Robustness test of the replacement risk aversion index.

	Entrep	Corporate	Portfolio	Entrep	Corporate	Portfolio
	(13)	(14)	(15)	(16)	(17)	(18)
SOEP	−0.408 ***	−0.319 ***	−0.305 ***
	(0.0434)	(0.0428)	(0.0318)
BRET				−0.0308 ***	−0.0211 ***	−0.0169 ***
				(0.00378)	(0.00340)	(0.00256)
Rise	0.262	0.480 ***	0.173	0.151	0.452 ***	0.133
	(0.173)	(0.144)	(0.113)	(0.182)	(0.154)	(0.112)
Sibling	−0.227 ***	−0.0349	−0.156 ***	−0.248 ***	−0.0567	−0.167 ***
	(0.0408)	(0.0398)	(0.0305)	(0.0404)	(0.0395)	(0.0301)
Income	0.127 ***	0.0634 **	0.0889 ***	0.143 ***	0.0952 ***	0.114 ***
	(0.0407)	(0.0316)	(0.0253)	(0.0390)	(0.0313)	(0.0249)
Gender	−0.586 ***	0.289	−0.425 ***	−0.521 ***	0.328 *	−0.326 **
	(0.177)	(0.194)	(0.129)	(0.175)	(0.187)	(0.128)
Party	0.424**	0.0849	0.193 *	0.541 ***	0.134	0.239 **
	(0.170)	(0.143)	(0.115)	(0.166)	(0.141)	(0.113)
Health	−0.315 ***	−0.154 **	−0.272 ***	−0.342 ***	−0.160 **	−0.289 ***
	(0.0633)	(0.0653)	(0.0484)	(0.0626)	(0.0643)	(0.0478)
Experience	0.339 *	0.861 ***	0.285 **	0.327 **	0.881 ***	0.299 ***
	(0.173)	(0.140)	(0.115)	(0.166)	(0.138)	(0.114)
Age	0.0137 *	0.0179 **	0.0127 **	0.0123 *	0.0165 **	0.00917 *
	(0.00745)	(0.00745)	(0.00559)	(0.00738)	(0.00729)	(0.00552)
Education	0.197 ***	0.107 ***	0.128 ***	0.213 ***	0.123 ***	0.138 ***
	(0.0378)	(0.0312)	(0.0242)	(0.0369)	(0.0312)	(0.0239)
Constant	2.037 ***	−1.150 *		1.552 **	−1.730 ***
	(0.641)	(0.594)		(0.630)	(0.574)
Fixed Effect	YES	YES	YES	YES	YES	YES
R2	0.4262	0.2967	0.2359	0.3974	0.2704	0.2040
Sample	669	669	650	669	669	650

Note: Values in parentheses are standard errors, * p < 0.01, ** p < 0.05, *** p < 0.01.

, all obtained the same conclusion as in the benchmark regression that risk aversion has a significant inhibitory effect on farmers’ entrepreneurship, despite the substitution of the risk aversion measure.

5.3. Endogeneity Test

To test for possible endogeneity problems in the model, the ratio of household safe assets (non-risky assets) to total assets was used as an instrumental variable to replace risk aversion in the endogeneity test section; the model has been explained in Section 4.4.3. Endogenous Processing Model. The results are shown in

Table 7. Endogeneity test results.

	First Stage	Second Stage
	DOSPERT	Entrep	Corporate	Portfolio
	(19)	(20)	(21)	(22)
DOSPERT		−2.236 ***	−1.413 ***	−3.381 ***
		(0.430)	(0.316)	(0.783)
IV	0.351 ***
	(0.066)
Rise	−0.008	−0.0228	0.0964 *	0.0484
	(0.034)	(0.0782)	(0.0574)	(0.127)
Sibling	0.009	−0.0327 *	0.00102	−0.0772 **
	(0.008)	(0.0198)	(0.0145)	(0.0330)
Income	−0.002	0.0186	0.0216 *	0.0586 **
	(0.007)	(0.0172)	(0.0126)	(0.0289)
Gender	−0.164 ***	−0.482 ***	−0.137 *	−0.715 ***
	(0.037)	(0.110)	(0.0810)	(0.188)
Party	−0.010	0.0769	0.0360	0.182
	(0.032)	(0.0738)	(0.0542)	(0.122)
Health	0.023 *	−0.071 **	−0.105 ***	−0.134 **
	(0.013)	(0.0331)	(0.0243)	(0.0534)
Experience	0.006	0.0738	0.256 ***	0.192
	(0.033)	(0.0746)	(0.0548)	(0.123)
Age	0.003 **	0.00712 *	0.00618 **	0.0124 **
	(0.002)	(0.00368)	(0.00270)	(0.00627)
Education	−0.003	0.0287 *	0.0226 *	0.0692 ***
	(0.007)	(0.0159)	(0.0117)	(0.0264)
Constant	3.221 ***	8.170 ***	4.542 ***	12.34 ***
	(0.123)	(1.475)	(1.083)	(2.687)
Fixed Effect	YES	YES	YES	YES
Anderson Corr		27.802	27.802	22.030
Cragg-Donald		28.533	28.533	22.418
F Test		11.92	11.89	17.26
Sample	669	669	669	650

Note: Values in parentheses are standard errors, * p < 0.01, ** p < 0.05, *** p < 0.01.

. IV passes the Underidentification test and the Weak identification test in all three regression results, with F-values greater than the critical values. The regression results for the instrumental variables remain largely consistent with those in the benchmark regression, all of which inhibit farmers’ entrepreneurial behavior at the 1% significance level.

5.4. Moderating Effect Results

This section examines the possibility of migrant work experience (MWE) serving as a moderator in the relationship between risk aversion and farmers’ entrepreneurial choices. To achieve this, the paper seeks to demonstrate this issue by including the interaction term between MWE and risk aversion. The findings from

Table 8. MWE moderating effect.

	Entrep	Corporate	Portfolio	Entrep	Corporate	Portfolio
	(23)	(24)	(25)	(26)	(27)	(28)
DOSPERT	−0.527 **	−0.816 ***	−0.461 **	−0.593 **	−0.641 **	−0.359 *
	(0.237)	(0.251)	(0.183)	(0.291)	(0.280)	(0.191)
MWE	3.803 ***	0.355	2.623 ***	2.855 **	0.0870	2.065 **
	(1.092)	(1.075)	(0.844)	(1.313)	(1.175)	(0.874)
DOSPERT×MWE	−1.290 ***	−0.127	−0.900 ***	−0.999 ***	−0.00618	−0.683 ***
	(0.314)	(0.319)	(0.247)	(0.376)	(0.349)	(0.256)
Rise				0.246	0.443 ***	0.173
				(0.175)	(0.147)	(0.112)
Sibling				−0.207 ***	−0.0604	−0.151 ***
				(0.0396)	(0.0388)	(0.0304)
Income				0.151 ***	0.103 ***	0.123 ***
				(0.0381)	(0.0309)	(0.0248)
Gender				−0.369 **	0.292	−0.301 **
				(0.174)	(0.191)	(0.131)
Party				0.443 ***	0.185	0.237 **
				(0.162)	(0.138)	(0.113)
Health				−0.402 ***	−0.204 ***	−0.316 ***
				(0.0625)	(0.0621)	(0.0475)
Experience				0.315 *	0.848 ***	0.269 **
				(0.163)	(0.135)	(0.114)
Age				0.00426	0.0128 *	0.00521
				(0.00731)	(0.00714)	(0.00553)
Education				0.209 ***	0.114 ***	0.135 ***
				(0.0367)	(0.0304)	(0.0240)
Constant	2.436 ***	2.143 **		2.360 **	−0.426
	(0.831)	(0.852)		(1.170)	(1.127)
Fixed Effect	YES	YES	YES	YES	YES	YES
R2	0.1247	0.0455	0.0535	0.3548	0.2365	0.1940
Sample	669	669	650	669	669	650

Note: Values in parentheses are standard errors, * p < 0.01, ** p < 0.05, *** p < 0.01.

reveal that the results of the regression analysis on risk aversion and entrepreneurial choice remain consistent with those observed in the benchmark regression. MWE has a positive effect on whether to enter entrepreneurship or portfolio entrepreneurship, which is consistent with the conclusions of Démurger and Hui [16], but in the regression of entrepreneurs, no significant results were obtained. Upon examination of the interaction term, as evidenced by regressions 26, 27, and 28, it can be determined that MWE assumes a negative moderating role. However, this moderating effect does not prove significant for corporate entrepreneurship. The regression coefficients of the interaction term on whether to choose entrepreneurship and portfolio entrepreneurship are −0.999 and −0.683, both significant at the 1% level. This suggests that MWE can moderate the relationship between risk aversion and entrepreneurship, as well as risk aversion and portfolio entrepreneurship. Hypothesis 4 is not fully confirmed.

Corporate entrepreneurship in rural areas is a type of entrepreneurship that relies heavily on financial, human, and social capital. Merely assessing an individual’s long-term migrant work experience does not offer a reliable measure of the heterogeneity. Therefore, we used a variable of migrant location instead of long-term migrant experience.

The hometowns of the samples in this paper are all located in southwest China. The southwestern provinces of Sichuan, Yunnan, Guizhou, and Chongqing are geographically close to each other, and there is no apparent contradiction in the lifestyles and dialects of the residents of these four provinces. Therefore, we categorized all migrant workers working in southwest China as a proximity group. The samples working in coastal provinces and in the Beijing–Tianjin–Hebei Economic Circle were combined and categorized in the coastal area group. In addition to these two groups, there are also five samples working in other inland provinces with poorer economic conditions than in the southwest, which cannot be categorized and tested separately, so these five samples were excluded. The regression results are shown in

Table 9. Moderating effects of MWE region.

	Entrep	Corporate	Portfolio	Entrep	Corporate	Portfolio
	(29)	(30)	(31)	(32)	(33)	(34)
DOSPERT	−0.527 **	−0.816 ***	−0.477 ***	−0.583 **	−0.640 **	−0.385 **
	(0.237)	(0.251)	(0.184)	(0.289)	(0.281)	(0.191)
MWE region
Southwest	4.312 ***	2.788 **	3.381 ***	3.722 **	2.609 *	2.736 ***
	(1.246)	(1.241)	(0.992)	(1.473)	(1.354)	(1.026)
Coastal	2.134	−2.948 **	0.529	1.093	−3.041 **	0.429
	(1.433)	(1.310)	(1.055)	(1.671)	(1.421)	(1.096)
Interaction
DOSPERT× Southwest	−1.532 ***	−0.872 **	−1.219 ***	−1.306 ***	−0.745 *	−0.953 ***
	(0.360)	(0.368)	(0.290)	(0.423)	(0.402)	(0.301)
DOSPERT ×Coastal	−0.607	0.906 **	−0.121	−0.353	0.950 **	−0.0924
	(0.418)	(0.391)	(0.314)	(0.484)	(0.425)	(0.327)
Rise				0.237	0.451 ***	0.170
				(0.176)	(0.148)	(0.113)
Sibling				−0.203 ***	−0.0594	−0.145 ***
				(0.0404)	(0.0394)	(0.0307)
Income				0.153 ***	0.0957 ***	0.114 ***
				(0.0390)	(0.0314)	(0.0250)
Gender				−0.335 *	0.326 *	−0.246 *
				(0.176)	(0.194)	(0.132)
Party				0.361 **	0.128	0.223 *
				(0.164)	(0.141)	(0.114)
Health				−0.418 ***	−0.205 ***	−0.331 ***
				(0.0634)	(0.0629)	(0.0480)
Experience				0.316 *	0.883 ***	0.299 ***
				(0.163)	(0.137)	(0.114)
Age				0.00719	0.0133 *	0.00832
				(0.00747)	(0.00725)	(0.00561)
Education				0.181 ***	0.117 ***	0.118 ***
				(0.0382)	(0.0311)	(0.0245)
Control variables	YES	YES	YES	YES	YES	YES
Fixed Effect	YES	YES	YES	YES	YES	YES
Constant	2.436 ***	2.143 **		2.412 **	−0.494
	(0.831)	(0.852)		(1.166)	(1.131)
R2	0.1861	0.0729	0.0913	0.4042	0.2577	0.2180
Sample	664	664	645	664	664	645

Note: Values in parentheses are standard errors, * p < 0.01, ** p < 0.05, *** p < 0.01.

.

There is no consistent moderating effect of migrant work region on the relationship between risk aversion and whether to enter entrepreneurship in both categories of regions. In the case of farmers who work in the southwest, their proximity to their workplace mitigated the inhibitory effect of risk aversion on entrepreneurship. However, the inhibitory effect of risk aversion on corporate entrepreneurship in coastal cities is intensified. Hence, through the division of migrant work regions of farmers, it is evident that working in the nearby areas provides greater benefits for preserving local social capital compared to working in coastal regions. This demonstrates the dependence of corporate entrepreneurship on local social capital and on local entrepreneurial information. Farmers working in coastal areas do not compensate for the loss of local entrepreneurial capital by accumulating more physical capital.

6. Conclusions, Implications and Discussion

6.1. Conclusions

Using data from 669 field surveys collected from southwest China, we examined the relationship between farmers’ risk aversion and entrepreneurial choices, and whether farmers’ experience of working outside the home can play a moderating role. Key findings can be summarized as follows:

(1)

Entrepreneurial farmers were found to have a lower average level of risk aversion compared to non-entrepreneurial farmers.

(2)

Farmers with lower risk aversion display a higher inclination towards entrepreneurship and are more likely to choose corporate and portfolio entrepreneurship.

(3)

MWE can reduce the negative impact of risk aversion on a decision to enter entrepreneurship and portfolio entrepreneurship.

(4)

MWE in local or nearby areas reduces the negative effects of risk aversion on entrepreneurship and portfolio entrepreneurship, while MWE in coastal and developed cities increases the negative effects.

6.2. Implications

Local governments should encourage farmers to participate in entrepreneurial activities and pay attention to the entrepreneurial willingness and ability of returning migrant workers. At the same time, local governments should actively implement a range of incentive policies to attract and support migrant workers and other returning entrepreneurs, and to enable local farmers with entrepreneurial aspirations and abilities to receive entrepreneurial competency training. It is imperative that the government take on the role of guiding innovation and entrepreneurial elements, including technology, information, capital, and management, to allocate and gather in villages effectively. This guidance should also improve entrepreneurs’ risk perception associated with entrepreneurial activities. Additionally, the government should emphasize the assessment of farmers’ risk attitudes and conduct comprehensive assessments of entrepreneurial and prospective entrepreneurial farmers. Farmers should be guided to a proper understanding of the different levels of entrepreneurial risk associated with different modes of production and organization.

6.3. Discussion

There are some limitations in this paper. (1) The sample size is only 669 and all samples are from the southwestern provinces of China, so it is limited in terms of representativeness. (2) The contents about entrepreneurial organizations, strategies, and forms of entrepreneurial need to be considered for further discussion in the future. (3) There are a number of traits that can have an impact on entrepreneurship that need more discussion. Risk attitudes have implications for both the survival strategies and the scale that farmers choose when starting a business. The higher the affinity for risk, the more likely it is that an entrepreneurial strategy of expansion and diversification is more likely to be adopted [53]. Personal traits such as trust, patience, and individualism are also important factors that affect entrepreneurship [54], and we can use experimental methods to study other personal traits as well.