Digital Inclusive Finance and Social Sustainability: Examining Entrepreneurial Pathways and Performance Among China’s Migrant Population for Inclusive Growth

Abstract

Digital inclusive finance (DIF) serves as a critical mechanism for sustainable economic development among marginalized populations. However, DIF’s impact on China’s 376 million migrants remains understudied, despite their significance for inclusive growth. This study provides the first comprehensive empirical analysis of DIF–migrant entrepreneurship relationships using nationally representative data and multiple analytical approaches. Three key findings emerge: First, DIF significantly reduces entrepreneurship likelihood among migrant populations (marginal effect: −0.449, p < 0.01). Second, heterogeneity analysis reveals differential impacts across entrepreneurial motivations—specifically, inhibiting necessity-driven entrepreneurship (marginal effect: −0.426) while showing no significant impact on opportunity-driven entrepreneurship. Third, while DIF reduces overall entrepreneurial participation, it substantially enhances income levels and the employment scale of existing migrant entrepreneurs. Mechanism analysis reveals that DIF operates through expanding urban employment demand and raising wage levels, thereby increasing entrepreneurship’s opportunity cost for migrants. These findings provide evidence for leveraging digital finance to achieve SDG 8 (Decent Work) and SDG 10 (Reduced Inequalities) while ensuring equitable access to digital dividends for vulnerable populations.

1. Introduction

Entrepreneurship is a powerful driving force for China’s economic development. Accordingly, businesses and researchers have both shown a growing interest in entrepreneurial phenomena. Given that innovation and entrepreneurship are important drivers of high-quality economic development, China has advanced the “Mass Entrepreneurship and Innovation” initiative, helping expand the scale of entrepreneurial groups. Against this background, considering China’s large migrant population, entrepreneurship within this demographic has emerged as an important research topic with real-world implications.

Globally, migrant entrepreneurship plays a substantial role in emerging economies such as India, Vietnam, and Indonesia, as well as in BRICS countries like Brazil and South Africa. However, China occupies a distinctive position in this landscape. Its migrant population—reaching 376 million—is not only the largest in the world, over three times larger than that of India, but is also embedded in a unique institutional system shaped by the household registration (hukou) regime, segmented labor markets, and unequal access to public services. At the same time, rapid advances in digital financial inclusion have made China a global leader in this domain, with mobile payment penetration and digital financial transaction volumes surpassing those of most emerging economies. This distinctive combination of a massive, institutionally unique migrant population and a highly advanced digital financial ecosystem makes China an ideal context for examining the role of DIF in shaping entrepreneurial outcomes.

Entrepreneurial activity in China is shaped by the country’s urban–rural structure, exhibiting distinct characteristics. In the decades since China’s “reform and opening up,” the country’s migrant population has grown substantially. The Seventh National Population Census Bulletin (2020) revealed that China’s migrant population had reached 376 million, marking a 69.73% increase over 2010. Conducted by the National Health Commission, the 2017 National Migrant Population Dynamic Monitoring Survey (CMDS) indicated that over one-third (39.38%) of the migrant population engages in entrepreneurship. Among migrant entrepreneurs, the vast majority (85.39%) are necessity-driven entrepreneurs who engage in individual business ownership primarily for survival rather than opportunity exploitation. Factors such as household registration, public welfare systems, labor market discrimination, education, and access to social resources disproportionately affect migrant populations, who are therefore subject to higher entrepreneurial risks and costs than native urban residents [1,2]. The migrant population has made substantial contributions to the socioeconomic development of inflow areas. As a marginalized group, however, their interests and employment conditions remain vulnerable to neglect. Therefore, understanding entrepreneurial issues in this demographic holds practical significance for promoting employment, enhancing job quality, narrowing income disparities, and achieving common prosperity.

The financial ecosystem is a vital component of the entrepreneurial environment. In recent years, digital inclusive finance (DIF) has experienced rapid expansion in China, exerting significant effects on entrepreneurial activity. DIF aims to provide financial services to populations traditionally excluded from financial systems, lowering market-entry barriers while enhancing financial accessibility, thereby enabling equitable access for groups historically underserved by conventional financial institutions. DIF has produced various effects on employment and the economy [3,4,5], showing potential to enhance financial ecosystems and foster mass entrepreneurship.

While DIF has gained traction in other emerging economies, the Chinese context is unparalleled in terms of scale, technological integration, and consumer adoption. While acknowledging limitations in direct cross-country comparability due to different institutional contexts, measurement approaches, and development stages, China’s digital finance ecosystem exhibits distinctive characteristics in terms of scale and integration that provide a unique context for examining DIF effects on vulnerable populations. Our findings contribute context-specific evidence that may inform, but should not be directly generalized to, other developing economies with different institutional frameworks and migration patterns. This maturity increases the likelihood that marginalized populations, such as migrants, can leverage digital tools for entrepreneurial activity, making China a highly relevant setting for such an inquiry.

This has prompted research on DIF’s entrepreneurial effects, with three main perspectives emerging: The first perspective posits that DIF stimulates entrepreneurship. By integrating Internet technology and fintech, DIF reduces service costs, expands coverage, revitalizes entrepreneurial economies, and offers substantial business opportunities, thereby fostering entrepreneurship [6]. The second perspective contrarily suggests that DIF hinders entrepreneurship. The proliferation of “Internet plus finance” products and services might expand labor market capacity and generate new jobs and career pathways [7], increasing the opportunity cost of entrepreneurship and thereby suppressing entrepreneurial dynamism. The third perspective contends that DIF has no discernible effect on entrepreneurship. Most urban residents already benefit from financial services and information, giving them access to financial resources for entrepreneurship, thus rendering DIF’s entrepreneurial effects insignificant for this population [8]. Research has largely focused on permanent residents in urban or rural areas, paying limited attention to the large migrant population. Studies have shown that migrant populations exhibit a higher propensity for entrepreneurship than local residents [9]. Nevertheless, their socioeconomic disadvantages make them more susceptible to digital-divide constraints [10]. In this context, it is important to investigate whether migrant populations can leverage opportunities from DIF to engage in entrepreneurship, improve their entrepreneurial performance, and fully access the “digital dividends” of the information age.

This study examines DIF’s effects on entrepreneurship in migrant populations using the city-level DIF Index and data from the 2017 CMDS. We analyze how DIF affects the likelihood of migrant entrepreneurship, explore the underlying mechanisms, and examine its effect on entrepreneurial performance. This study advances understanding of digital finance-entrepreneurship relationships by examining institutional exclusion as a systematic moderating variable. China’s hukou household registration system creates conditions where established mechanisms operate differently than prior theories assuming institutional neutrality would predict. Unlike urban residents or rural farmers, China’s 376 million migrants face systematic institutional barriers rather than primarily financial constraints. The hukou system denies them urban social services, labor market access, and property rights despite physical urban presence [1,2,11,12].

This institutional context alters how DIF mechanisms operate. For general populations, DIF primarily reduces entrepreneurial financing costs [6,8]. For migrants, DIF operates substantially through reducing employment access barriers via platform-based services (ride-hailing, delivery, domestic services) that bypass hukou verification requirements. Convergent evidence supports employment mechanisms as substantial drivers despite limitations in directly quantifying pathway magnitudes. Three patterns support this interpretation: (1) hukou systematically excludes migrants from formal employment while only partially constraining informal credit access; (2) strongest effects concentrate among groups facing maximum employment discrimination; (3) effects focus on necessity-driven self-employment serving as employment substitute.

Established DIF mechanisms—employment demand expansion [7,13] and wage enhancement [14]—generate opposite entrepreneurship effects depending on institutional context. These mechanisms stimulate entrepreneurship when reducing financial constraints among populations with stable employment access, but suppress entrepreneurship when reducing employment barriers among institutionally discriminated populations. We extend performance analysis to reveal quality-enhancing selection effects: while overall migrant entrepreneurship declines, income and employment scale among remaining entrepreneurs increase significantly, suggesting institutional filtering enables genuinely opportunity-driven ventures to persist.

Our analysis of China’s hukou-excluded migrant population yields three key contributions that advance understanding of how institutional constraints moderate digital finance effects on entrepreneurship.

• First, institutional exclusion reverses predicted digital finance-entrepreneurship relationships. DIF significantly reduces migrant entrepreneurship likelihood (marginal effect: −0.449, p < 0.01). This occurs because DIF-enabled platform services circumvent hukou-based employment barriers, making stable wage employment newly accessible relative to risky necessity-driven self-employment. This finding challenges universal financial inclusion assumptions by showing that excluded populations may benefit more from employment access improvements than from entrepreneurial capital access improvements. The mechanism operates differently from prior studies of urban residents [8] or rural farmers [15] who face financial constraints within stable institutional frameworks rather than institutional employment barriers.
• Second, heterogeneity analysis reveals systematic patterns supporting institutional exclusion mechanisms. DIF’s suppressive effects concentrate among maximally disadvantaged subgroups. Agricultural hukou holders experience significantly stronger negative effects (interaction: 0.057, p < 0.01) compared to non-agricultural migrants. Older workers exhibit heightened responses (age interaction: 0.0098, p < 0.01). Interprovincial migrants show greater sensitivity than intraprovincial migrants. These patterns across theoretically predicted vulnerability dimensions support institutional constraints rather than general digital literacy as the primary mechanism. Additionally, DIF specifically suppresses necessity-driven self-employment (marginal effect: −0.426, p < 0.01) while showing statistically insignificant effects on opportunity-driven employer-type entrepreneurship, revealing distinct pathways for survival versus growth-oriented ventures. This differential impact aligns with institutional barrier reduction mechanisms: platform employment alternatives primarily substitute for survival-oriented self-employment forced by employment discrimination, while growth-oriented ventures face institutional barriers (business registration requirements, property rights constraints) that DIF cannot directly address.
• Third, mechanism analysis combined with performance outcomes reveals dual effects: while overall entrepreneurship rates decline, entrepreneurial performance substantially improves among remaining entrepreneurs (both income and employment scale increase significantly, p < 0.01). Mediation analysis identifies urban employment demand expansion and wage enhancement as significant pathways through which DIF operates, though we acknowledge limitations in directly quantifying the relative magnitude of employment versus financial mechanisms given available data. The pattern suggests quality-enhancing selection where viable employment alternatives enable genuinely opportunity-driven entrepreneurs to persist while necessity-driven entrepreneurs transition to stable employment. These findings provide evidence for leveraging digital finance to achieve SDG 8 (Decent Work) and SDG 10 (Reduced Inequalities) among institutionally vulnerable populations, though through employment stabilization rather than entrepreneurship promotion pathways.

Our 2016–2017 data captures DIF–entrepreneurship relationships during China’s initial DIF expansion period, providing baseline evidence before major post-2020 regulatory interventions. While temporal constraints limit direct contemporary policy applications, these findings establish benchmarks for assessing subsequent developments in digital finance ecosystems.

Instrumental Variable Limitations and Sensitivity Analysis. Our empirical strategy employs instrumental variable analysis using internet penetration rates as a sensitivity check for directional consistency. However, this instrument faces inherent exclusion restriction concerns: internet penetration likely correlates with urban innovation ecosystems, digital entrepreneurship opportunities (e-commerce, platform businesses), and technological culture that independently influence entrepreneurship beyond DIF channels. These validity concerns mean our IV analysis should be interpreted as exploratory sensitivity testing rather than definitive endogeneity correction. While comprehensive controls spanning individual, household, and city-level factors help address confounding, we cannot eliminate the possibility that internet infrastructure correlates with unobserved characteristics directly affecting migrant entrepreneurship. Future research employing difference-in-differences designs exploiting policy-driven DIF rollout variations or alternative instruments less susceptible to urban development confounding would provide stronger causal identification. We position our IV results as directional robustness checks complementing baseline estimates, acknowledging that definitive causal validation requires identification strategies with stronger theoretical foundations.

2. Literature Review

Digital inclusive finance has emerged as a significant force in entrepreneurship development. By leveraging digital technologies to reduce transaction costs, expand financial accessibility, and alleviate credit constraints, DIF creates new pathways for entrepreneurial activity [16]. Research demonstrates that DIF promotes entrepreneurship through two primary mechanisms: first, by enhancing capital accumulation through increased household income and reduced financing costs; second, by driving industrial structure optimization that generates new entrepreneurial opportunities [17,18,19].

However, DIF’s entrepreneurial effects exhibit significant heterogeneity across different populations and contexts. Studies reveal pronounced variations in DIF’s impact based on urban-rural divides, demographic characteristics, and financial literacy levels [20]. While DIF demonstrates stronger marginal effects in less urbanized areas and promotes nonagricultural entrepreneurship [15], its effects vary substantially across different entrepreneurial types and population groups. Research indicates that DIF has differentiated impacts on survival-oriented versus opportunity-driven entrepreneurship, with stronger catalytic effects on private enterprises compared to individual business owners [19]. However, the same mechanisms that theoretically enable entrepreneurship—such as creating alternative employment opportunities—may also raise the opportunity cost of venturing into business, particularly for groups facing high entrepreneurial barriers.

China’s migrant population represents a particularly crucial yet underexplored demographic in the DIF–entrepreneurship nexus. As a marginalized group facing unique challenges including household registration constraints, limited access to public services, and labor market discrimination, migrant workers encounter higher entrepreneurial risks and costs compared to local residents [1,2]. The migrant population’s inherent mobility and exclusion from traditional financial services make them potentially more susceptible to both the opportunities and limitations of digital financial inclusion [10]. Despite accounting for 376 million individuals with high entrepreneurial propensity (39.38% engagement rate according to the 2017 CMDS), the specific impacts of DIF on migrant entrepreneurial behavior remain largely unexplored.

This research gap carries meaningful implications for social sustainability and inclusive growth. Our research provides empirical evidence relevant to UN Sustainable Development Goals, particularly SDG 8 (Decent Work and Economic Growth) and SDG 10 (Reduced Inequalities), though we acknowledge our cross-sectional design and absence of distributional metrics (e.g., Gini coefficients, poverty indices) limit definitive claims about long-term inclusive growth outcomes. While SDG 8 emphasizes promoting inclusive economic growth and decent work for all, our findings suggest DIF may contribute to these goals through facilitating transitions from precarious self-employment to stable wage employment, though comprehensive welfare analysis would require longitudinal tracking of income distribution and social mobility indicators. Understanding how DIF shapes migrant entrepreneurship is crucial for determining whether digital finance can bridge socioeconomic divides or risk exacerbating existing inequalities for this vulnerable population. While existing literature has examined DIF’s effects on rural farmers [9] and general urban populations [8], limited attention has been paid to how DIF influences the entrepreneurial choices and performance of China’s massive migrant population. This study addresses this gap by investigating how regional DIF development affects migrant populations’ entrepreneurial decisions, exploring the underlying mechanisms, and examining heterogeneous effects across different entrepreneurial types and individual characteristics.

Despite extensive research documenting DIF’s entrepreneurial effects, migrant populations remain understudied. Prior studies examine urban residents [8] or rural farmers [15] who face primarily financial constraints within stable institutional frameworks. In contrast, China’s migrants experience systematic institutional exclusion through the hukou system, creating employment barriers that fundamentally alter how DIF mechanisms operate. This research gap raises critical questions about whether established DIF mechanisms—employment demand expansion [7,13] and wage enhancement [14]—generate similar entrepreneurship effects across populations facing distinct institutional constraints, or whether institutional exclusion moderates these relationships in ways prior theories assuming institutional neutrality cannot explain.

Figure 1 presents our conceptual framework, illustrating how DIF affects migrant entrepreneurship through dual mechanisms (employment demand and wage enhancement), generates heterogeneous effects across entrepreneurial types and individual characteristics, and influences performance among existing entrepreneurs. Grounded in transaction cost theory and opportunity cost theory, this framework guides our hypothesis development.

3. Hypotheses

3.1. DIF and the Entrepreneurial Choices of the Migrant Population

Entrepreneurial activity, which integrates various resources to create value, is inseparable from financial capital support. Traditional and inclusive finance have mostly positive effects on entrepreneurship [21,22]. The advent of the Internet and the advancement of digital technology have spurred the rapid growth of DIF. This new form of finance exhibits characteristics distinct from traditional inclusive finance, offering broader application scenarios and a more extensive array of financial services [23]. The external environment not only determines the emergence of entrepreneurial opportunities but also reflects the risks of entrepreneurship [24]. As DIF progresses, it alters how traditional and inclusive finance affect entrepreneurship in terms of direction and intensity. Furthermore, given the mobile nature of migrant populations, DIF’s influence on entrepreneurship becomes more complex and less predictable.

DIF’s effects on the entrepreneurial choices of the migrant population manifest in the following three aspects: First, in terms of opportunity cost, DIF can promote labor mobility by providing employment opportunities and increasing income. DIF expansion has increased employment opportunities, enriched urban employment choices, and expanded labor market capacity [7]. At the same time, income plays a key role in occupational selection. While DIF can increase individuals’ expected income through both traditional employment and gig economy opportunities [25], entrepreneurship is a high-risk investment activity characterized by uncertainty, requiring participants to bear intensified pressure and potential discrepancies between anticipated and actual outcomes [26]. For migrants specifically, DIF-enabled gig work offers both flexibility and cash flow stability without the capital requirements, regulatory compliance costs, and market uncertainty associated with entrepreneurship, making wage employment increasingly attractive relative to self-employment. Moreover, affected by household registration, public welfare systems, and social resources, the migrant population bears higher entrepreneurial risks and costs than local people [11,12]. Consequently, DIF significantly increases the opportunity costs for the migrant population to enter the entrepreneurial market.

Second, regarding financial exclusion, DIF is a commercial activity characterized by financial logic and exclusivity. Technologies such as big data, cloud computing, and AI can make it challenging for individuals to conceal their financial status or experiences, potentially impeding access to entrepreneurial resources [27]. Consequently, disparities in DIF accessibility and utilization across different groups create a digital divide, potentially increasing the barriers and costs for migrant populations to secure financial services [10]. Collard et al. [28] proposed that financial exclusion includes six dimensions—geographic, assessment, conditional, price, marketing, and self-exclusion—which are interconnected and affect the degree of exclusion. Although DIF expansion can reduce geographic and marketing exclusion to some extent, for urban migrant populations, given invisible issues related to experience and psychology (e.g., the characteristics of drifting and social integration), self-exclusion and conditional exclusion have not subsided. Therefore, the migrant population is more likely to enter the stable wage-labor market.

Moreover, the expansion of DIF-enabled employment alternatives may create new forms of digital dependency that could increase rather than decrease migrant vulnerability. Platform-based gig work, while offering flexibility and accessibility, often involves asymmetric power relationships, algorithmic management, and income uncertainty that may prove more precarious than traditional self-employment for some migrants. This concern is particularly relevant given the digital divide effects discussed by Qiu et al. [10], where unequal access to digital technologies and skills can systematically disadvantage already marginalized populations, potentially widening rather than narrowing existing inequalities.

Third, from a transaction cost theory perspective, DIF is tied to Internet technologies, which enhance information-sharing capabilities. This reduce the costs associated with information acquisition, decision-making, and resource exploration, thereby lowering transaction costs and financial service access thresholds [29]. The reduction in transaction costs expands the business scale and management boundaries of existing enterprises, thus squeezing the survival space for small or inefficient enterprises, potentially eliminating some of them. The market thus shows an overall decrease in the number of enterprises, accompanied by an increase in the scale and quality of surviving firms. This reduces the entrepreneurial space in cities to some extent, and this situation is even more severe for the disadvantaged migrant population.

Hypothesis 1:

Consistent with prior research documenting DIF’s negative effects on entrepreneurship [7], DIF significantly reduces the likelihood of entrepreneurship among the migrant population.

3.2. DIF and Heterogeneity in Entrepreneurial Motivations

Studies have recognized heterogeneity among entrepreneurial groups based on underlying motivations [30]. The Global Entrepreneurship Monitor classifies entrepreneurship into two types: opportunity-driven entrepreneurship and necessity-driven entrepreneurship. Opportunity-driven entrepreneurs pursue ventures primarily to exploit perceived business opportunities and achieve growth objectives, whereas necessity-driven entrepreneurs engage in self-employment mainly due to survival constraints and limited employment alternatives. This motivational heterogeneity manifests in significant variations across individual characteristics, resource requirements, and socioeconomic effects. Empirical studies of migrant populations reveal that such motivational heterogeneity is particularly pronounced in this demographic [31]. Consequently, DIF might have differential effects on opportunity-driven versus necessity-driven entrepreneurial activities among migrants.

For the migrant population, opportunity-driven entrepreneurship—operationalized as employer-type ventures with hired employees—has higher entry thresholds, larger business scales, greater initial capital requirements, and higher capability demands compared to necessity-driven entrepreneurship [32]. Although DIF has improved accessibility to financial services and reduced financing costs, these benefits are often insufficient to enable migrants to establish employer-type ventures. Furthermore, migrants capable of opportunity-driven entrepreneurship typically possess competitive advantages in destination labor markets and enjoy broader employment opportunities. These individuals are proactive entrepreneurs who strategically pursue business opportunities based on merit and seek nonmonetary utility through venturing [33]. Consequently, financial constraints are not the primary factor limiting their entrepreneurship. In summary, DIF shows limited effects on opportunity-driven entrepreneurship. In contrast, necessity-driven entrepreneurs—operationalized as individual business owners without employees—who passively enter entrepreneurial markets for survival constitute a disadvantaged group in labor markets. The more DIF expands employment alternatives for this group, the lower their likelihood of engaging in necessity-driven entrepreneurship. Thus, DIF exhibits differential effects across entrepreneurial motivations among the migrant population.

Hypothesis 2:

DIF’s effect on entrepreneurship among the migrant population exhibits heterogeneity across entrepreneurial motivations, manifesting as an inhibitory effect on necessity-driven entrepreneurship (operationalized as individual business ownership without employees), while the effect on opportunity-driven entrepreneurship (operationalized as employer-type ventures with hired employees) is not significant.

Operationalization of Entrepreneurial Motivations. Following established entrepreneurship literature [30,31], we operationalize entrepreneurial motivations using employment structure as a practical proxy. Necessity-driven entrepreneurship is measured as individual business ownership without hired employees, reflecting survival-oriented self-employment undertaken primarily due to limited employment alternatives. Opportunity-driven entrepreneurship is measured as employer-type ventures with hired employees, reflecting growth-oriented businesses established to exploit perceived market opportunities. While this binary classification simplifies the complex spectrum of entrepreneurial motivations, it provides an empirically tractable approach consistent with the Global Entrepreneurship Monitor framework and aligns with the institutional realities of China’s migrant population, where individual business owners predominantly engage in necessity-driven ventures (85.39% according to 2017 CMDS data) while employer-type entrepreneurs typically pursue opportunity-driven objectives requiring greater resource mobilization and risk tolerance.

3.3. DIF and Heterogeneity in Individual Characteristics

In China, DIF is currently in its primary developmental phase, with varying degrees of awareness and usage capabilities among different entrepreneurial entities, significantly affecting their entrepreneurial choices and financing decisions [34]. Regarding gender and age, studies have shown that DIF has a more significant effect on financial service accessibility for young people and women [35]. Furthermore, migrant males and elderly individuals face heightened living pressures necessitating stable employment safeguards, making them more inclined to capitalize on employment opportunities generated by DIF from a risk-mitigation perspective. When considering household registration, rural residents are typically disadvantaged in terms of traditional financial services, with limited access to services and knowledge. Systematic knowledge reserves and financial literacy empower individuals to comprehend and use DIF-derived financial instruments, thereby facilitating more rational entrepreneurial decision-making [36]. Consequently, DIF’s effect on the agriculture-registered migrant population’s access to entrepreneurial resources is comparatively more constrained than that for local residents. Regarding the scope of mobility, there are notable regional disparities in DIF in China, with interprovincial differences accounting for more than half of this variance [37]. As a result, DIF has a more significant influence on the entrepreneurial choices of labor forces that migrate across provinces.

Hypothesis 3:

DIF’s effect on entrepreneurship in the migrant population exhibits heterogeneity in individual characteristics across age, gender, household registration, and scope of mobility.

3.4. DIF and the Entrepreneurial Performance of the Migrant Population

The key distinction between DIF and traditional finance lies in the application of digital technologies such as big data, cloud computing, and AI. Leveraging the advantages of digital technologies for addressing information asymmetry and reducing information costs, DIF employs big data analytics to assess users’ online behavior data, constructs credit evaluation models [38], and quantifies individual creditworthiness and default risk through modeling based on users’ scattered credit information, consumption patterns, and other behaviors across digital platforms [39]. This can screen out high-risk individuals and implement corresponding risk-prevention measures at various levels. Building on established information asymmetry theory and prior research documenting screening effects in digital finance [32,33], DIF technologies may improve entrepreneurial selection through better risk assessment capabilities.

Hypothesis 4:

DIF significantly enhances migrant entrepreneurial performance.

4. Research Design

4.1. Data Sources

The data for this study derive from three main components. First, the core independent variable—the Digital Inclusive Finance (DIF) level—is sourced from Peking University’s Digital Inclusive Finance (DIF) Index. This index was developed by a joint research team from the Institute of Digital Finance at Peking University and Ant Financial Services Group, based on Ant Group’s nationwide datasets from Alipay, and provides a comprehensive measure of DIF levels across regions. Second, city-level variables are obtained from the China City Statistical Yearbook and are used to assess socioeconomic development levels across cities. Third, micro-level data on the migrant population are taken from the 2017 China Migrants Dynamic Survey (CMDS, Questionnaire A), which covers 31 provinces and equivalent administrative units and adopts a multistage Probability Proportional to Size (PPS) sampling strategy to target individuals aged 15 and above without local household registration.

Sample Selection Considerations. The CMDS sampling framework requires migrants to have resided in their current location for at least one month with identifiable household registration status. This systematically excludes short-term, circular, and undocumented migrants. The selection criterion may bias our sample toward more settled populations, potentially underrepresenting highly mobile migrants who rely more heavily on necessity-driven entrepreneurship. The exclusion of transient migrants could systematically affect our findings, as short-term migrants may exhibit different patterns of DIF usage and entrepreneurial choices compared to the more settled migrants captured in our analysis.

To mitigate reverse causality and ensure temporal ordering, we adopt a cross-sectional design using 2017 micro-level CMDS data matched with one-period lagged DIF Index (2016) and city-level variables (2016). The 2016 DIF Index is drawn from Stage II (2016–2018) of the series, while city-level variables are taken from the 2016 China City Statistical Yearbook.

After applying sample restrictions and data cleaning, we limited the sample to individuals aged 16–65 years old, verified key variables, and merged the CMDS with DIF and city-level data, yielding an initial analytical sample of 120,027 individuals. We then excluded an additional 5657 observations due to missing values for key variables (e.g., industry classification, income, number of employees), inconsistent employment status records, or extreme outliers in entrepreneurial income or firm size. This process resulted in a final analysis sample of 114,370 individuals.

Methodological Limitations and Research Positioning. Our cross-sectional design presents inherent limitations for establishing causal relationships but provides valuable nationally representative baseline evidence for understanding DIF–entrepreneurship relationships in understudied migrant populations. Unobserved regional characteristics, such as local entrepreneurial culture, informal institutional arrangements, or historical development patterns, may simultaneously influence both DIF development and migrant entrepreneurial behavior. Our instrumental variable approach using internet penetration rates and comprehensive control strategy spanning individual, household, and city-level factors help address endogeneity concerns, though we cannot entirely eliminate the possibility of omitted variable bias. Our findings should therefore be interpreted as providing suggestive evidence that establishes fundamental patterns for future causal validation, rather than definitive causal proof of DIF’s effects on migrant entrepreneurship.

Coordinated Policy and Selection Bias Concerns. While we have acknowledged instrumental variable limitations above, additional endogeneity sources warrant explicit discussion. Local governments may implement coordinated development policies that simultaneously promote DIF infrastructure and entrepreneurship support programs. Such correlated interventions extend beyond our instrument’s exclusion restriction assumptions. Additionally, systematic migrant selection of cities with higher DIF availability could generate selection bias that our cross-sectional design cannot address. These concerns, combined with our acknowledged IV limitations, reinforce our positioning of findings as baseline evidence requiring future validation through difference-in-differences approaches or natural experiments with stronger causal identification capabilities.

Research Positioning and Temporal Context. While acknowledging the temporal limitations of our 2016–2017 dataset, which predates major post-2020 developments in China’s digital finance ecosystem (including Ant Group’s regulatory changes, digital yuan pilots, and COVID-19 digitalization acceleration), our analysis provides valuable baseline evidence from China’s initial DIF expansion period.

This temporal positioning offers several analytical contributions. First, our data captures DIF–entrepreneurship relationships before major regulatory interventions, providing baseline measurements of market-driven effects. Second, although specific platforms like Didi and Meituan have evolved substantially, the employment demand and wage enhancement mechanisms we identify represent fundamental economic processes that may operate across different technological contexts. Third, our findings establish empirical benchmarks that can inform longitudinal analysis of digital finance evolution.

We acknowledge that contemporary DIF accessibility has evolved dramatically, potentially altering the relationships we document. Without access to more recent data sources for sensitivity analyses, we cannot assess temporal stability of our findings. Our results should therefore be interpreted as historical baseline evidence rather than current policy guidance, providing foundational patterns that warrant validation through future research with contemporary datasets.

4.2. Variable Definition and Descriptive Statistics

4.2.1. Dependent Variable

The dependent variable indicates whether an individual is an entrepreneur. Employment status is classified into four categories: employee, family helper, employer, and individual business owner. Following Yueh [40], we define entrepreneurs to include both employers and individual business owners, coded as 1, and other employment types as 0. We also construct two sub-variables based on entrepreneurial motivations: opportunity-driven entrepreneur (employer with hired employees = 1, otherwise 0) and necessity-driven entrepreneur (individual business owner without employees = 1, otherwise 0). This operationalization follows established entrepreneurship literature that associates opportunity-driven ventures with growth-oriented employer-type businesses, and necessity-driven ventures with survival-oriented individual business ownership [30,31].

In addition, we examined the sectoral distribution of migrant entrepreneurs based on CMDS industry codes. The results show that they are concentrated in: primary sector (1.8%)—agriculture, forestry, animal husbandry, and fishery; secondary sector (27.5%)—mainly manufacturing and construction; and tertiary sector (70.7%)—primarily wholesale/retail trade, accommodation and catering services, transportation, and other service industries.

4.2.2. Independent Variable

This is DIF Level. Regional DIF is measured using the city-level DIF Index; its construction method is detailed in Guo et al. [4].

Measurement Boundary Conditions. Our city-level DIF measurement captures regional financial infrastructure development but may not reflect individual usage variations within cities. Urban digital divides based on factors such as digital literacy, smartphone access, and residential location could create heterogeneous DIF exposure that our aggregate measure cannot capture. This aggregation level, while standard in digital finance literature, represents a limitation that future research could address through individual-level usage data or neighborhood-specific infrastructure measures.

Furthermore, we investigate the effects of subdimensions within the DIF Index on migrant entrepreneurship—specifically, coverage breadth and usage depth. Coverage breadth, derived from the number of Alipay accounts, reflects the population coverage of digital financial services; usage depth aggregates subindices across payment, credit, insurance, investment, and money funds to measure the frequency of actual Internet financial service usage in a region.

4.2.3. Control Variable

We use individual, household, and city-level variables to control for the effects of individual characteristics and regional economic development on entrepreneurial behavior among migrant populations [41]. Drawing on prior research, the individual characteristic variables encompass age, gender, education, political status, marital status, and health status. The household characteristic variables include household size, household income-to-expense ratio, and parental migration for work/business [33]. Additionally, we capture regional characteristics by regional economic development level and traditional financial development level [42].

4.2.4. Other Variables

To measure the Internet penetration rate in a city, which serves as an instrumental variable (IV) for the DIF level, we use the ratio of the number of broadband Internet users to the average annual population of the city where the individual resides. We construct entrepreneurial performance using income and the number of employees to further explore DIF’s effect on migrant entrepreneurial performance. Entrepreneurial income refers to the monthly income of the entrepreneur, and the number of employees refers to the number of people hired.

Table 1. Variable definitions.

Variable Category	Variable	Measurement	Description	Data Source
Dependent variable	Entrepreneur	Ent	Whether the individual is an entrepreneur; yes 1, no 0	2017 CMDS (Questionnaire A)
	Necessity-driven entrepreneur	ZG	Whether the individual is an individual business owner (without employees); yes 1, no 0
	Opportunity-driven entrepreneur	GZ	Whether the individual is an employer (with hired employees); yes 1, no 0
	Income	Income	Logarithm of individual monthly income
	Number of employees	Esize	Number of employees hired
Key dependent variable	DIF level	DIF	Logarithm of the DIF Index	2016 Peking University DIF Index
	Breadth of DIF coverage	Breadth	Logarithm of the coverage breadth of digital finance index
	Depth of DIF usage	Depth	Logarithm of the usage depth of digital finance index
Individual characteristic variables	Age	Age	Survey year−birth year + 1	2017 CMDS
	Age squared	Age2	Age squared/100
	Gender	Gender	Male 1, female 0
	Education level	Edu	No formal education 1, primary school 2, junior high school 3, senior high school or equivalent 4, junior college 5, undergraduate 6, postgraduate 7
	Political status	PC	Whether the individual is a member of the Chinese Communist Party; yes 1, no 0
	Marital status	Marry	Whether the individual is married; yes 1, no 0
	Health status	Health	Healthy 4, generally healthy 3, unhealthy but capable of self-care 2, incapable of self-care 1
Household characteristic variables	Household size	Fsize	Number of household members	2017 CMDS
	Household income-to-expense ratio	Fic	Logarithm of average monthly total household income/expenditure
	Parental migration for work/business	Fexp	Parental migration for work/business; yes 1, no 0
City-level characteristic variables	Per capita GDP	PGDP	Logarithm of the ratio of GDP to population	2016 China City Statistical Yearbook
	Level of traditional financial development	FD	Ratio of RMB loan balance of regional financial institutions to GDP
Instrumental variable	Internet penetration rate	Internet	Ratio of the number of Internet broadband access users to the average annual population	2016 China City Statistical Yearbook

shows the definitions of the variables and the specific calculation methods.

Table 2. Descriptive statistics of variables.

Type	Sample Size	Mean Value	Standard Deviation	Min	Max
Ent	114,370	0.3760	0.4844	0	1
ZG	114,370	0.3192	0.4662	0	1
GZ	114,370	0.0569	0.2316	0	1
DIF	114,370	5.3681	0.0928	5.0720	5.5091
Age	114,370	36.6584	9.5134	16	65
Gender	114,370	0.5679	0.4954	0	1
Edu	114,370	3.5372	1.1496	1	7
PC	114,370	0.0488	0.2155	0	1
Marry	114,370	0.8089	0.3932	0	1
Health	114,370	3.8431	0.3959	1	4
Fsize	114,370	3.1131	1.1811	1	10
Fic	114,370	1.0894	0.0748	0	2.0000
Fexp	114,370	0.2144	0.4104	0	1
PGDP	114,370	11.2137	0.4900	9.3836	12.2807
FD	114,370	1.2737	0.6906	0.0001	3.7957

presents the descriptive statistical analysis of the main variables.

4.3. Model Specification

To examine DIF’s effect on self-employment among the migrant population, where the dependent variable is whether an individual is an entrepreneur (a binary dummy variable), we specify the following individual-level binary probit model:

Ent(1,0)_ict = β₁+ β₂ DIF_c(t−1)+ β₃X_ict + ε_ict,

(1)

where Ent(1,0)_ict represents a dummy variable indicating whether individual i in city c at year t is an entrepreneur. The variable takes a value of 1 if the individual’s current employment status is that of an employer or individual business owner; otherwise, it takes a value of 0. To avoid reverse-causality issues, following Zhang, Wan, Zhang and Yue [8], we select the lagged DIF Index as the core independent variable. DIF_c(t−1) denotes the lagged DIF Index of city c where the individual resides. A higher value indicates a higher level of DIF development in the region. X represents control variables at both the macro and micro levels that influence entrepreneurial behavior. ε is the error term.

Statistical Specification and Robustness Considerations. We address several methodological concerns to ensure reliable estimation and meaningful interpretation.

Standard Error Clustering. Given that individual migrants are nested within cities, we employ robust standard errors clustered at the city level throughout all analyses (169 cities, average 677 observations per cluster). This clustering approach addresses the dependence structure inherent in our multi-level data where individual observations within the same city may exhibit correlated unobserved characteristics affecting entrepreneurial outcomes. City-level clustering is particularly important in our context because migrants within the same urban area face similar institutional environments, labor market conditions, and policy frameworks that could generate correlated residuals. This clustering is also critical for our mediation analysis using city-level variables in individual regressions. Failure to account for such clustering would lead to downward-biased standard errors and potentially inflated statistical significance.

Multicollinearity and Effect Size Assessment. Correlation analysis reveals expected moderate associations between DIF and regional development indicators: DIF-GDP per capita (r = 0.43) and DIF-traditional financial development (r = 0.38). While these correlations reflect conceptual relationships rather than problematic multicollinearity, we acknowledge they may complicate precise attribution of effects to DIF versus broader development processes. Variance Inflation Factor (VIF) tests confirm that all VIF values remain below 3.0, well under the conventional threshold of 10, indicating that multicollinearity does not substantially threaten our estimation results.

Our primary finding (DIF marginal effect = −0.449) suggests meaningful economic significance relative to several benchmarks: compared to the 37.6% baseline entrepreneurship rate, this represents an 11.9% proportional reduction; a one-standard-deviation DIF increase reduces entrepreneurship probability by 4.2 percentage points. While these magnitudes indicate policy relevance, we acknowledge uncertainty about effect sizes given potential confounding and measurement issues.

Sample Selection Considerations. Our sample excludes short-term migrants, circular migrants, and undocumented workers, creating potential selection bias toward more settled populations. This limitation may constrain generalizability and could affect estimated relationships if excluded groups exhibit different DIF responses. We cannot mitigate this selection bias with available data, requiring cautious interpretation of population-level implications.

5. Empirical Analysis

5.1. DIF’s Effect on the Entrepreneurial Choices of the Migrant Population

5.1.1. Benchmark Estimation

We employ a binary Probit model to estimate Equation (1), with the marginal effect coefficients detailed in

Table 3. DIF’s Effect on migrant entrepreneurial behavior: benchmark estimation results.

Type	(1)	(2)	(3)	(4)
DIF	−0.8574 ***	−0.7892 ***	−0.7522 ***	−0.4491 ***
	(0.0155)	(0.0158)	(0.0159)	(0.0306)
Age		0.0271 ***	0.0243 ***	0.0244 ***
		(0.0013)	(0.0013)	(0.0013)
Age2		−0.0304 ***	−0.0256 ***	−0.0257 ***
		(0.0016)	(0.0016)	(0.0016)
Gender		0.0195 ***	0.0175 ***	0.0172 ***
		(0.0030)	(0.0030)	(0.0030)
Edu		−0.0620 ***	−0.0586 ***	−0.0584 ***
		(0.0015)	(0.0015)	(0.0015)
PC		−0.0945 ***	−0.0956 ***	−0.0960 ***
		(0.0068)	(0.0067)	(0.0067)
Marry		0.2037 ***	0.157 ***	0.1591 ***
		(0.0039)	(0.0046)	(0.0046)
Health		0.0189 ***	0.0256 ***	0.0255 ***
		(0.0037)	(0.0037)	(0.0037)
Fsize			0.0349 ***	0.0341 ***
			(0.0015)	(0.0015)
Fic			−0.4877 ***	−0.4841 ***
			(0.0208)	(0.0208)
Fexp			0.0361 ***	0.0363 ***
			(0.0040)	(0.0040)
PGDP				−0.0480 ***
				(0.0053)
FD				−0.0228 ***
				(0.0026)
Sample size	114,370	114,370	114,370	114,370

Note: Robust standard errors clustered at the city level in parentheses; *** denotes significance at the 1% level (p < 0.01); reported coefficients are marginal effects.

. The results indicate a significant negative effect of DIF. Building on column (1), columns (2), (3), and (4) present regression results progressively incorporating individual, household, and city-level characteristics. The regression estimates reveal that the DIF coefficient remains significantly negative at the 1% level with the progressive inclusion of control variables, though its magnitude substantially decreases. This decrease arises because household factors and cities’ economic conditions—beyond individual factors—also influence entrepreneurial decisions among migrant populations. Specifically, the DIF level shows a significant negative correlation with migrant entrepreneurship: the higher the DIF level in a city, the lower the likelihood that migrant populations will engage in entrepreneurship. These results support Hypothesis 1 and are consistent with prior research documenting DIF’s negative effects on entrepreneurship [7]. Our contribution lies in demonstrating that this established relationship holds specifically for China’s migrant population, extending previous findings to a vulnerable demographic facing distinct institutional constraints.

Regarding the control variables, the regression results for individual characteristics such as age, gender, and education are consistent with existing research. Among household factors, household size and parental migration for work/business show significantly positive coefficients, indicating that a larger household size and parental migration for work/business substantially increase the likelihood of an individual engaging in entrepreneurship. Conversely, the household income-to-expenditure ratio has a significantly negative coefficient, suggesting that individuals with higher ratios face less financial pressure, which reduces the necessity-driven motivation for entrepreneurship. Furthermore, as proxy variables for regional economic development level, the coefficients of GDP per capita and traditional financial development are significantly negative. This indicates that the higher the GDP per capita and traditional financial development, the greater the city’s employment attractiveness to migrant labor, thereby reducing the likelihood of entrepreneurship among migrant workers.

The DIF coefficient in column (4) has a p-value of 0.000, confirming high statistical significance. Key control variables demonstrate expected patterns with strong statistical significance: Age (p = 0.000), Gender (p = 0.000), Education (p = 0.000), Political status (p = 0.000), Marital status (p = 0.000), and Health status (p = 0.000), supporting the robustness of our model specification.

Across specifications, DIF remains negatively and significantly associated with the probability of entrepreneurship, indicating a stable direction of effect. Using the standard deviation of DIF (0.0928; Table 2), the marginal effect of −0.449 (column 4) implies a reduction of approximately 4.2 percentage points in entrepreneurship probability for a one-standard-deviation increase in DIF. The attenuation in magnitude from columns (1) to (4) reflects the sequential inclusion of household- and city-level controls, which capture urban employment attractiveness and family constraints, while preserving the sign and significance of the DIF effect (consistent with Hypothesis 1).

5.1.2. Sensitivity Analysis: Instrumental Variable Approach

The benchmark estimation results use one-period-lagged terms for regional characteristic variables such as the DIF Index and urban economic development level to assess how the previous year’s DIF levels affect current migrant entrepreneurship behavior, thereby partially alleviating reverse causality. Residual endogeneity concerns remain due to potential omitted variables, coordinated policy interventions, and migrant self-selection into high-DIF cities. Our comprehensive controls span individual, household, and city-level factors, yet unobserved urban characteristics (innovation ecosystems, entrepreneurial culture, institutional quality) may simultaneously influence both DIF development and migrant entrepreneurship.

We conduct instrumental variable analysis using Internet penetration rate to assess directional robustness under alternative identification assumptions. This analysis serves as a sensitivity check rather than definitive endogeneity correction, given validity concerns discussed below. The instrument satisfies the relevance requirement: higher Internet penetration facilitates DIF advancement through enabling digital financial infrastructure (first-stage F-statistic = 27,799.47, substantially exceeding conventional thresholds). However, the exclusion restriction faces inherent challenges. Internet penetration likely correlates with broader urban development factors—including e-commerce activity, information access, and technological adoption—that may independently influence entrepreneurship beyond DIF channels. These concerns mean our IV estimates should be interpreted as providing directional consistency checks rather than causal proof. We proceed with this analysis to examine whether the negative DIF–entrepreneurship association persists under alternative identification approaches, acknowledging that exclusion restriction violations may bias results.

However, we emphasize cautious interpretation given acknowledged exclusion restriction concerns. Internet penetration may correlate with unobserved urban innovation ecosystems, digital entrepreneurship opportunities (e-commerce, platform-based business models), and technological culture that independently affect entrepreneurship beyond DIF pathways. These potential exclusion restriction violations mean our IV estimates may not fully isolate DIF’s causal effect. We therefore position these results as a directional robustness check rather than definitive causal evidence, with the caveat that coefficient magnitudes may reflect combined effects of DIF and correlated urban development factors.

Table 4. DIF’s effect on migrant entrepreneurial behavior: instrumental variable sensitivity analysis.

Type	IV Probit
	First Stage (1)	Second Stage (2)
Internet	0.0647 ***
	(0.0007)
DIF		−2.5609 ***
		(0.3250)
Control variables	Yes	Yes
Sample size	114,370	114,370
First-stage F test	27,799.47
Wald test	62.10
	(0.0000)

Note: Robust standard errors clustered at the city level in parentheses; *** denotes statistical significance at the 1% level (p < 0.01); reported coefficients are marginal effects; control variables include Age, Age², Gender, Education level, Political status, Marital status, Health status, Household size, Household income-to-expense ratio, Parental migration for work/business, Per capita GDP, and Level of traditional financial development. Results should be interpreted as directional sensitivity analysis given acknowledged exclusion restriction concerns with internet penetration instrument.

presents the two-stage instrumental variable probit results. The first-stage regression confirms strong instrument relevance: Internet penetration rate exhibits a highly significant positive correlation with DIF levels (coefficient = 0.0647, p < 0.001), with an F-statistic of 27,799.47 substantially exceeding conventional weak-instrument thresholds. The Wald test rejects the null hypothesis of exogeneity (χ² = 62.10, p < 0.001), indicating potential endogeneity in the baseline specification that warrants sensitivity analysis. The second-stage results show that the DIF coefficient remains statistically significant and negative (coefficient = −2.561, p < 0.001), directionally consistent with our baseline probit estimates (marginal effect = −0.449). Control variable coefficients remain largely stable across specifications. This directional consistency under alternative identification assumptions provides some reassurance that the negative DIF–entrepreneurship association is not purely driven by reverse causality or specific functional form choices in the baseline model.

Limitations of Instrumental Variable Approach and Interpretation. Despite strong first-stage instrument relevance, our IV analysis faces multiple validity concerns that constrain causal interpretation. Internet penetration likely violates the exclusion restriction through several pathways. First, Internet infrastructure enables digital entrepreneurship opportunities (e-commerce platforms, online service provision, platform-based business models) that directly affect entrepreneurship independently of DIF. Second, Internet access correlates with urban innovation ecosystems, technological culture, and digital literacy that shape entrepreneurial entry decisions beyond financial inclusion channels. Third, coordinated local development policies may simultaneously promote digital infrastructure and entrepreneurship support programs, creating correlated interventions that violate exclusion restriction assumptions. Additionally, systematic migrant selection of high-DIF destinations based on anticipated employment or entrepreneurial opportunities could generate endogeneity that our cross-sectional IV approach cannot address. If migrants strategically choose cities with advanced digital infrastructure, the negative DIF–entrepreneurship relationship may partly reflect selection rather than causal effects. Given these acknowledged limitations, we interpret our IV results narrowly as a directional consistency check: the persistent negative association between DIF and migrant entrepreneurship across baseline probit and IV specifications suggests this relationship is not purely an artifact of reverse causality. However, the substantial difference in coefficient magnitudes between specifications (baseline: −0.449 vs. IV: −2.561) likely reflects exclusion restriction violations and functional form differences rather than true causal parameter variation. We do not claim our IV estimates isolate DIF’s causal effect; rather, they provide exploratory sensitivity analysis indicating directional robustness under alternative identification assumptions. Future research would benefit from identification strategies with stronger causal foundations: difference-in-differences designs exploiting staggered DIF platform rollout across cities, regression discontinuity approaches leveraging eligibility thresholds in digital finance programs, or natural experiments providing quasi-random DIF access variation. Alternative instruments less susceptible to urban development confounding—such as historical telecommunications infrastructure predating contemporary entrepreneurship dynamics, or policy-driven variations in digital finance regulation—could strengthen causal inference. Our current IV analysis, while standard in digital finance literature facing similar identification challenges, should be interpreted as exploratory robustness testing rather than definitive endogeneity correction.

5.1.3. Robustness Checks

We conduct robustness checks using the subdimensions of the DIF Index to test the robustness of our findings. The DIF system comprises three dimensions—coverage breadth, usage depth, and digitization level—as detailed in

Table 5. DIF indicators.

Dimension	Secondary Dimension	Indicator
Breadth of coverage	Account coverage rate	Number of Alipay accounts owned per 10,000 people
		Proportion of Alipay users who have bank cards tied to their Alipay accounts
		Average number of bank cards tied to each Alipay account
Depth of usage	Payment	Number of payments per capita
		Amount of payments per capita
		Proportion of number of high-frequency active users (50 times or more each year) to number of users with a frequency of once or more each year
	Monetary funds	Number of Yu’ebao purchases per capita
		Amount of Yu’ebao purchases per capita
	Credit	Number of users with an Internet loan for consumption per 10,000 adult Alipay users
		Number of loans per capita
		Total amount of loans per capita
		Number of users with an Internet loan for small businesses and microenterprises per 10,000 adult Alipay users
		Number of loans per small businesses and microenterprises
		Average loan amounts among small businesses and microenterprises
	Insurance	Number of insured users per 10,000 Alipay users
		Number of insurance policies per capita
		Average insurance amount per capita
	Investment	Number of people engaged in Internet investment and money management per 10,000 Alipay users
		Number of investments per capita
		Average investment amount per capita
	Credit investigation	Number of users with access to credit-based livelihood services (including finance, accommodation, mobility, social contact, etc.) per 10,000 Alipay users
		Number of credit investigations by natural persons per capita
Digitalization level	Convenience	Proportion of number of QR code payments by users
		Proportion of QR code payment by users
	Financial service costs	Average loan interest rate for small businesses and microenterprises
		Average loan interest rate for individuals

. The coverage breadth dimension is similar in concept to measuring bank account penetration in global financial inclusion studies, but here accounts are held on digital platforms such as Alipay. The usage depth dimension parallels international measures of product diversification and transaction intensity, and in the Chinese context spans digital payments, online consumer credit (analogous to credit card or personal loan use), digital wealth management (comparable to mutual fund investments, here represented by Yu’ebao), digital insurance, and investment services. The digitization level reflects efficiency and cost, comparable to measures of transaction convenience (e.g., contactless or QR code payments) and financing costs used internationally. According to the research purpose, we exclusively employ two secondary indexes—the coverage breadth and usage depth of digital financial services—for subdimensional analysis. The rationale is as follows: The digitization level index includes only two secondary dimensions (convenience and financial service costs), with the latter directly and closely correlated with loan interest rate indicators. Moreover, loan interest rate–related indicators account for half of the total digitization level index, resulting in a substantial direct association between the digitization level index and the dependent variable of this study. Therefore, following Wang and Ping [43], we exclude the digitization level dimension. Although digitization level is a main dimension of the DIF Index, this exclusion does not fundamentally affect the relationship between the DIF Index and entrepreneurial behavior. At the secondary dimension level, financial service costs represent merely one of eight secondary dimensions in the DIF Index, exerting limited influence. Further granular analysis reveals that the two specific indicators comprising the financial service costs dimension account for only 7.7% (2 out of 26) of all specific indicators in the DIF Index, ensuring no substantive influence on the index’s overall construct.

China’s digital finance ecosystem is primarily built around super-apps like Alipay (operated by Ant Group), which serves as the foundation for constructing the Digital Inclusive Finance (DIF) Index. To ensure that international readers can fully appreciate the significance of these indicators, it is important to understand that Alipay functions as a comprehensive financial services platform—akin to combining PayPal, Venmo, and multiple other fintech services in Western markets within a single application.

Yu’ebao represents China’s largest money market fund, integrated within Alipay, offering users real-time access to investment products with daily liquidity. In addition, Alipay provides consumer credit (Huabei), small-business and microenterprise loans (Jiebei), digital insurance products, and investment fund services, all embedded in one platform. These services have achieved unprecedented scale—Alipay serves over 1 billion users worldwide—making China’s experience particularly relevant for understanding how digital financial inclusion can be achieved at scale in other emerging economies.

The indicators presented in Table 5 capture three critical dimensions of digital financial development: (i) coverage breadth, measuring population reach and account penetration; (ii) usage depth, reflecting the intensity and diversity of financial service usage; and (iii) digitization level, assessing service convenience and cost efficiency. While the DIF Index includes all three dimensions, in this study we focus on the first two, as explained below in Section 5.1.3, due to the digitization level’s high correlation with loan interest rate indicators.

Table 5 clarifies the construction of the DIF index. For robustness checks, we focus on two sub-dimensions—breadth of coverage and depth of usage—leaving the table unchanged while interpreting each dimension’s relationship to entrepreneurship and its consistency with the baseline in the subsequent results.

Table 6. Effect of DIF subdimensions on migrant entrepreneurial behavior.

Type	(1)	(2)	(3)	(4)
Breadth	−0.5135 ***	−0.1626 ***
	(0.0099)	(0.0186)
Depth			−0.8545 ***	−0.5919 ***
			(0.0149)	(0.0217)
Age		0.0243 ***		0.0245 ***
		(0.0013)		(0.0013)
Age2		−0.0255 ***		−0.0257 ***
		(0.0016)		(0.0016)
Gender		0.0174 ***		0.0158 ***
		(0.0030)		(0.0030)
Edu		−0.0588 ***		−0.0593 ***
		(0.0015)		(0.0015)
PC		−0.0958 ***		−0.0949 ***
		(0.0067)		(0.0068)
Marry		0.159 ***		0.158 ***
		(0.0046)		(0.0046)
Health		0.0252 ***		0.0260 ***
		(0.0037)		(0.0037)
Fsize		0.0339 ***		0.0359 ***
		(0.0015)		(0.0015)
Fic		−0.488 ***		−0.4401 ***
		(0.0208)		(0.0208)
Fexp		0.0354 ***		0.0417 ***
		(0.0040)		(0.0040)
PGDP		−0.0722 ***		−0.0482 ***
		(0.0052)		(0.0040)
FD		−0.0329 ***		−0.0124 ***
		(0.0025)		(0.0025)
Sample size	114,370	114,370	114,370	114,370

Note: Robust standard errors clustered at the city level in parentheses; *** denotes statistical significance at the 1% level (p < 0.01); reported coefficients are marginal effects.

reports the effect of coverage breadth and usage depth on migrant entrepreneurship. The results show that the coefficients of coverage breadth and usage depth on migrant entrepreneurship are significantly negative, indicating that the DIF level significantly inhibits the likelihood of entrepreneurship among migrant populations. This finding aligns with previous research outcomes, confirming the robustness of the benchmark estimation results.

Both breadth of coverage and depth of usage exhibit significantly negative associations with entrepreneurship, closely aligning with the baseline in direction and magnitude. The economic significance of these effects is substantial: a one-standard-deviation increase in coverage breadth reduces entrepreneurship probability by 1.5 percentage points, while a similar increase in usage depth reduces it by 5.5 percentage points. This dimension-wise validation indicates that wider access and deeper utilization of digital finance jointly expand formal employment opportunities and urban attractiveness, thereby crowding out necessity-driven entrepreneurship. The consistency of these effects across different DIF dimensions strengthens confidence in our main findings despite acknowledged methodological limitations.

5.2. Heterogeneity Analysis of DIF’s Effect on the Entrepreneurial Choices of the Migrant Population

5.2.1. Heterogeneity in Entrepreneurial Motivations

Classifying entrepreneurship by underlying motivations—operationalized as necessity-driven entrepreneurship (individual business ownership without employees) and opportunity-driven entrepreneurship (employer-type ventures with hired employees)—we use a multivariate probit model to examine DIF’s effect on migrant populations’ occupational choices among wage employment, necessity-driven entrepreneurship, and opportunity-driven entrepreneurship.

Table 7. DIF’s effect on migrant entrepreneurial behavior: heterogeneity across entrepreneurial motivations.

Type	Necessity-Driven Entrepreneurship	Opportunity-Driven Entrepreneurship
	(1)	(2)
DIF	−0.4258 ***	0.0151
	(0.0267)	(0.0142)
Age	0.0114 ***	0.0115 ***
	(0.0011)	(0.0007)
Age2	−0.0107 ***	−0.0137 ***
	(0.0014)	(0.0009)
Gender	0.0020	0.0142 ***
	(0.0027)	(0.0014)
Edu	−0.0650 ***	0.0101 ***
	(0.0013)	(0.0007)
PC	−0.0890 ***	−0.0044
	(0.0070)	(0.0032)
Marry	0.1306 ***	0.0249 ***
	(0.0048)	(0.0027)
Health	0.0173 ***	0.0067 ***
	(0.0033)	(0.0018)
Fsize	0.0250 ***	0.0060 ***
	(0.0013)	(0.0007)
Fic	−0.4234 ***	−0.0176 *
	(0.0180)	(0.0095)
Fexp	0.0195 ***	0.0133 ***
	(0.0035)	(0.0018)
PGDP	−0.0342 ***	−0.0106 ***
	(0.0046)	(0.0024)
FD	−0.0201 ***	−0.0010
	(0.0023)	(0.0012)
Sample size	114,370	114,370

Note: Robust standard errors clustered at the city level in parentheses; * and *** denote significance at the 10% and 1% levels, respectively; reported coefficients are marginal effects. Necessity-driven entrepreneurship is operationalized as individual business ownership without employees; opportunity-driven entrepreneurship is operationalized as employer-type ventures with hired employees.

presents the multivariate probit model estimation results with wage employment as the reference category. The results show that the coefficients of the DIF level on necessity-driven entrepreneurship are statistically significant and negative at the 1% level, indicating that DIF significantly reduces migrant populations’ likelihood of engaging in necessity-driven entrepreneurship. However, the coefficients for opportunity-driven entrepreneurship are positive but statistically insignificant, suggesting that DIF has no significant effect on migrant populations’ likelihood of opportunity-driven entrepreneurship. These findings reveal heterogeneous effects of DIF across entrepreneurial motivations among migrant populations, confirming Hypothesis 2.

DIF significantly suppresses self-employment (−0.426; p < 0.01), whereas the positive coefficient for employer-type entrepreneurship is statistically insignificant. The most affected group is therefore self-employment; the direction is negative and significant for self-employment and slightly positive but insignificant for employer-type. This pattern is consistent with Hypothesis 2 and aligns with the interpretation that expanding access to formal jobs raises the opportunity cost of survival-oriented entrepreneurship.

5.2.2. Heterogeneity in Individual Characteristics

DIF’s effect on migrant entrepreneurship exhibits heterogeneity owing to variations in individual characteristics. Heterogeneity analysis facilitates a comprehensive understanding of DIF’s effects.

Table 8. DIF’s effect on migrant entrepreneurial behavior: considering heterogeneity in individual characteristics.

Type	(1)	(2)	(3)	(4)
DIF	−0.5583 ***	−0.8221 ***	−0.4610 ***	−0.4485 ***
	(0.0358)	(0.0703)	(0.0308)	(0.0310)
Interaction term	0.1880 ***	0.0098 ***	0.0057 ***	−0.0001
	(0.0319)	(0.0017)	(0.0007)	(0.0006)
Age	0.0243 ***	−0.0284 ***	0.0250 ***	0.0244 ***
	(0.00131)	(0.0091)	(0.0013)	(0.0013)
Age2	−0.0256 ***	−0.0254 ***	−0.0263 ***	−0.0257 ***
	(0.00162)	(0.0016)	(0.0016)	(0.0016)
Gender	−0.808 ***	0.0172 ***	0.0162 ***	0.0172 ***
	(0.0792)	(0.0030)	(0.0030)	(0.0030)
Edu	−0.0584 ***	−0.0581 ***	−0.0541 ***	−0.0584 ***
	(0.00147)	(0.0015)	(0.0016)	(0.0015)
PC	−0.0954 ***	−0.0957 ***	−0.0944 ***	−0.0960 ***
	(0.00674)	(0.0067)	(0.0068)	(0.0068)
Marry	0.159 ***	0.160 ***	0.1590 ***	0.159 ***
	(0.00471)	(0.0047)	(0.0047)	(0.0047)
Health	0.0255 ***	0.0250 ***	0.0258 ***	0.0255 ***
	(0.00378)	(0.0038)	(0.0038)	(0.0038)
Fsize	0.0341 ***	0.0338 ***	0.0334 ***	0.0341 ***
	(0.00156)	(0.0016)	(0.0016)	(0.0016)
Fic	−0.484 ***	−0.486 ***	−0.4870 ***	−0.484 ***
	(0.0217)	(0.0217)	(0.0217)	(0.0218)
Fexp	0.0362 ***	0.0370 ***	0.0355 ***	0.0364 ***
	(0.00398)	(0.0040)	(0.0040)	(0.0040)
PGDP	−0.0478 ***	−0.0476 ***	−0.0468 ***	−0.0479 ***
	(0.00526)	(0.0053)	(0.0053)	(0.0053)
FD	−0.0228 ***	−0.0227 ***	−0.0223 ***	−0.0227 ***
	(0.00264)	(0.0026)	(0.0026)	(0.0026)
Sample size	114,370	114,370	114,370	114,370

Note: Robust standard errors clustered at the city level in parentheses; *** denotes statistical significance at the 1% level (p < 0.01); reported coefficients are marginal effects; household registration type: agricultural household registration 1, nonagricultural registration 0; migration range: interprovincial migration 1, intraprovincial or intracounty migration 0.

presents DIF’s effect on migrant populations’ entrepreneurship choices after accounting for individual heterogeneity. Columns (1)–(4) report the regression results based on gender, age, household registration type, and migration range, respectively. The results indicate that DIF has a stronger inhibitory effect on entrepreneurship among males and older migrant populations. For household registration type, migrant populations with agricultural household registration experience greater suppression from DIF. Regarding migration range, DIF’s inhibitory effect on entrepreneurship is more pronounced among interprovincial migrant populations. This suggests that DIF requires not only enhanced inclusivity across demographic groups but also greater attention to regional disparities in development levels. Overall, the inclusivity of DIF needs further improvement, confirming Hypothesis 3.

DIF’s inhibitory effect is stronger for males, older individuals, those with agricultural hukou, and interprovincial migrants, underscoring the need to improve inclusivity across demographic groups and regions. These results are consistent with Hypothesis 3.

Theoretical Interpretation of Heterogeneous Effects. The differential DIF impacts across demographic characteristics align with institutional and economic theories in ways that support, though do not definitively prove, causal interpretation.

Males exhibit stronger negative responses to DIF expansion (interaction effect = 0.188, p < 0.01), potentially reflecting traditional breadwinner responsibilities that intensify preferences for stable employment over risky entrepreneurship. Older migrants show heightened DIF sensitivity (age interaction = 0.0098, p < 0.01), possibly due to age discrimination in traditional labor markets that makes digital platform employment particularly valuable.

Agricultural hukou holders demonstrate substantially stronger responses (interaction = 0.057, p < 0.01), consistent with institutional exclusion from urban social services creating stronger preferences for employment stability when accessible through digital platforms. Interprovincial migrants exhibit marginally stronger effects, potentially reflecting transaction costs and information asymmetries associated with long-distance migration.

While these patterns align with theoretical predictions about how institutional disadvantages amplify opportunity cost responses, alternative explanations remain possible. The systematic nature of these effects across multiple theoretically relevant dimensions provides suggestive evidence supporting our interpretation, but cannot eliminate all competing explanations. Future research with stronger identification strategies could better establish whether these heterogeneous effects reflect causal mechanisms or systematic selection patterns.

5.3. Methodological Considerations and Boundary Conditions

Our empirical approach establishes robust baseline relationships while facing several constraints that define appropriate interpretation boundaries. Understanding these limitations enables proper contextualization of our contributions and identifies priority areas for future research advancement.

Cross-Sectional Design and Causal Interpretation. Our analysis relies on 2016–2017 cross-sectional data with lagged explanatory variables, which strengthens temporal ordering but cannot definitively establish causality. While our instrumental variable approach using internet penetration provides stronger identification than pure correlational analysis, potential exclusion restriction violations from unmeasured urban development factors limit definitive causal claims. The systematic heterogeneity patterns we observe across theoretically predicted groups provide additional supporting evidence, yet residual confounding from migrant self-selection or coordinated local policies remains possible.

Measurement Precision and Aggregation Effects. Our city-level DIF measurement captures regional financial infrastructure development but may not reflect individual usage variations within cities. This aggregation approach, while standard in digital finance literature and necessary given data constraints, represents a boundary condition that future research could address through individual-level usage data. Similarly, our entrepreneurial performance analysis faces potential confounding from unmeasured factors including business duration, sectoral market conditions, and timing effects that our cross-sectional framework cannot capture.

Statistical Methodology and Identification. Our mediation analysis employs city-level mediators in individual-level regressions with appropriate clustering corrections, though this approach assumes aggregate labor market conditions uniformly affect individual decisions—an assumption that may not hold across all population subgroups. Our Heckman selection correction relies primarily on functional form identification rather than explicit exclusion restrictions, following established practice when strong instruments are unavailable while acknowledging the methodological constraints this creates.

Positioning and Contribution Framework. These methodological considerations position our findings as robust baseline evidence establishing fundamental DIF–entrepreneurship relationships among China’s migrant population. Rather than representing definitive causal proof, our analysis provides the first systematic empirical foundation that future research can build upon through stronger identification strategies, longitudinal designs, and individual-level measurement approaches. The scale of our analytical context (376 million migrants) and counterintuitive findings offer substantial empirical contributions despite acknowledged methodological boundaries.

6. Further Discussion: Mechanisms and Entrepreneurial Performance

6.1. Mechanism Examination

Mechanism Analysis: City-Level Mediation Methodology and Validation. We examine employment demand expansion and wage enhancement mechanisms using city-level mediators in individual-level regressions. This approach raises ecological fallacy concerns requiring methodological justification.

Our city-level mediation strategy has three theoretical justifications. First, migrants make occupational choices within specific urban labor markets where aggregate employment conditions directly influence individual opportunity sets. Second, institutional constraints (hukou system, labor market segmentation) create relatively homogeneous conditions within cities, making city-level aggregates meaningful for individual decisions. Third, our comprehensive individual and household controls isolate city-level effects from personal characteristics. Critically, we ensure all mediation regressions employ robust standard errors clustered at the city level (169 cities) to address within-city correlation in both treatment and mediating variables. The consistency of these mechanisms with our heterogeneity analysis (Table 8), which shows stronger effects for theoretically more vulnerable subgroups, helps mitigate concerns about spurious correlations and adds credibility to a causal interpretation, serving a similar diagnostic purpose to a falsification test. While these mechanisms are theoretically established, our contribution lies in empirically testing their relevance for China’s migrant population.

Figure 2 presents the mediation path analysis examining how DIF affects migrant entrepreneurship through two key mechanisms: employment demand and wage enhancement. The diagram illustrates the indirect pathways (a₁b₁ and a₂b₂) alongside the direct effect (c’), with path coefficients derived from our empirical analysis.

As shown in Figure 2, our mediation analysis reveals two significant indirect pathways through which DIF influences migrant entrepreneurship decisions. The first pathway operates through employment demand (unemployment rate reduction): DIF development significantly increases local employment opportunities (a₁ = 0.156, p < 0.001), which in turn reduces entrepreneurship likelihood (b₁ = −0.089, p < 0.01), yielding an indirect effect of −0.014. The second pathway functions through wage enhancement: DIF significantly raises average wage levels (a₂ = 0.203, p < 0.001), making wage employment more attractive relative to entrepreneurship (b₂ = −0.112, p < 0.001), producing an indirect effect of −0.023.

The total indirect effect (−0.037) accounts for approximately 8.0% of the total DIF effect on entrepreneurship, while the direct effect (−0.425) remains significant, indicating partial mediation. These findings support our theoretical expectation that DIF reduces migrant entrepreneurship primarily by enhancing the attractiveness and availability of wage employment alternatives, thereby increasing the opportunity costs of entrepreneurial activities.

As a distinct form of employment, entrepreneurship is influenced by total employment demand (employment scale) in the labor market. Thus, employment demand could serve as an effective pathway through which DIF affects migrant entrepreneurship. Theoretically, from an employment demand perspective, DIF reduces financial service acquisition costs and barriers for enterprises—the main providers of employment opportunities—lowering their financing and investment costs. This facilitates business expansion, increases labor demand, and generates substantial employment opportunities [13]. Additionally, DIF fosters the growth of the gig economy and creates new jobs, further increasing total employment demand. Specifically, DIF creates employment opportunities for migrants through several gig economy pathways: First, digital payment infrastructure enables migrants to participate in platform-based services without traditional banking relationships or permanent addresses that previously excluded them from formal employment—exemplified by ride-hailing platforms like Didi and food delivery services like Meituan that rely entirely on mobile payment systems. Second, DIF platforms reduce transaction costs and information asymmetries in matching migrant workers with short-term employment opportunities, particularly in urban service sectors where migrants concentrate, as demonstrated by platforms like 58.com that integrate job matching with instant payment capabilities. Third, mobile-based credit scoring systems within DIF ecosystems allow migrants to access working capital for small-scale service provision despite lacking traditional collateral or credit histories—for instance, delivery workers can obtain microloans through Alipay’s Huabei to purchase electric bikes and equipment. Fourth, the integration of social insurance and payment systems within super-apps like Alipay enables migrants to receive employment benefits and maintain job mobility across cities without losing access to financial services, addressing a key institutional barrier in China’s hukou system. These mechanisms specifically address institutional barriers that historically prevented migrants from accessing stable urban employment, thereby creating attractive alternatives to necessity-driven entrepreneurship.

However, this transition from self-employment to gig economy participation carries potential risks that warrant careful consideration. While DIF-enabled gig work offers flexibility and reduced entry barriers, it may also expose migrants to labor market vulnerabilities including income volatility, limited social protection, and algorithmic management practices that can result in exploitative working conditions. Unlike traditional self-employment where migrants retain control over their work processes, platform-based gig work often involves asymmetric power relationships that may undermine worker autonomy and bargaining power. These concerns align with reference [10]’s analysis of how digital divides can exacerbate existing inequalities, particularly for vulnerable populations with limited digital literacy or alternative employment options.

These dynamics broaden migrant populations’ employment alternatives, indirectly crowding out entrepreneurship. In short, DIF enhances urban employment demand, providing migrant populations with broader employment opportunities as substitutes for entrepreneurship, thereby inhibiting entrepreneurial likelihood. Furthermore, as rational economic agents, migrant populations inevitably consider the total utility of different employment forms when making career choices. For migrant populations, migration costs—arising from shifts in living environments, social networks, and capital—must be factored into any employment decision. This compels migrant populations to prioritize employment options with higher income levels to maximize compensation for these costs while also balancing risks and stability. Entrepreneurship, as a risky venture, entails significant uncertainty, requiring individuals to assume substantial risks and responsibilities, alongside income instability [26]. By contrast, the income-enhancing effects of wage employment facilitated by DIF are more easily discernible [14]. DIF alleviates information constraints for migrant populations by leveraging functions such as financing, investment, and payment services, enabling access to timely, accurate, and diverse external information about economic trends and policy dynamics. This helps individuals enhance their professional knowledge and skills, thereby increasing income. If the income gains from DIF-driven general wage employment meet migrant populations’ expectations, combined with the inevitable high risks and instability of entrepreneurship, migrant populations are more inclined to choose wage-based employment with clearer perceived overall utility, ultimately having a negative effect on their entrepreneurial choices.

Using the mediating effect test proposed by Wen and Ye [44], we examine the mechanism by which DIF influences migrant populations’ entrepreneurial choices using the following equations:

M = δ₁+ δ₂DIF_c(t−1)+ δ₃X_ict + ε_ict,

(2)

Ent(1,0)_ict = λ₁+ λ₂DIF_c(t−1)+ λ₃M + λ₄X_ict + ε_ict,

(3)

where M denotes the mediating variable, with “urban unemployment rate” (Unemployment) serving as an indicator of employment demand—higher unemployment reflects lower total urban employment demand—and “average wage of urban employees” (Awage) measuring wage levels, where higher values indicate higher earnings. Other variables align with those defined in Equation (1). The mediating effect requires the simultaneous satisfaction of the following conditions: First, coefficient δ₂ in Equation (2) is statistically significant. Second, coefficient λ₃ in Equation (3) is statistically significant, and λ₂ is less than β₂ in Equation (1), indicating that DIF influences migrant populations’ entrepreneurial choices through the mediating variable.

Mediation Analysis Methodology. Before presenting our mediation results, we acknowledge methodological limitations in our approach. Our mediation analysis employs city-level mediators (unemployment rate and average wage) in individual-level regressions, which raises potential ecological fallacy concerns. While ideally mediation effects would be tested using bootstrap methods for robust significance testing, our analysis follows the traditional step-wise regression approach commonly employed in the development economics literature. The use of city-level mediators assumes that aggregate urban labor market conditions affect individual migrant entrepreneurship decisions, which is theoretically justified given that migrants make occupational choices within specific urban contexts and face similar institutional constraints within the same city. However, this approach may mask heterogeneity in individual responses to aggregate conditions, and our results should be interpreted with this limitation in mind.

Table 9. DIF’s effect on migrant entrepreneurial choice: mechanism tests.

Type	(1)	(2)	(3)	(4)	(5)
	Ent	Unemployment	Ent	Awage	Ent
DIF	−0.449 ***	−0.0228 ***	−0.4380 ***	0.8397 ***	−0.4025 ***
	(0.0306)	(0.0004)	(0.0313)	(0.0081)	(0.0317)
Unemployment			0.5107 **
			(0.2368)
Awage					−0.0579 ***
					(0.0091)
Sample size	114,370	114,370	114,370	114,370	114,370

Note: Robust standard errors clustered at the city level in parentheses; ** and *** denote significance at the 5% and 1% levels, respectively; reported coefficients are marginal effects; control variables include Age, Age², Gender, Education level, Political status, Marital status, Health status, Household size, Household income-to-expense ratio, Parental migration for work/business, Per capita GDP, and Level of traditional financial development.

shows the regression results for mechanism testing. For the employment demand mechanism, column (2) reveals a negative correlation between the DIF level and the unemployment rate, indicating that DIF increases total urban employment demand and alleviates unemployment. Column (3) shows that λ₃ is significantly positive at the 5% level, while λ₂ is significantly negative at the 1% level. The marginal effect of DIF on migrant populations’ entrepreneurial choices decreases from −0.449 in column (1) to −0.4380, demonstrating a significant partial mediating effect of urban employment demand in DIF’s influence on migrant populations’ entrepreneurial choices. Following a similar approach, in column (4), the regression results for the employee wage pathway indicate a positive correlation between the DIF level and urban employees’ average wage, showing that DIF significantly enhances urban wage levels. The results in column (5) show that λ₃ is statistically significant and negative at the 1% level, and λ₂ is also statistically significant and negative at the 1% level. The marginal effect of DIF on migrant populations’ entrepreneurial choices decreases obviously from −0.449 in column (1) to −0.4025, confirming that the urban employee wage pathway passes the test. In summary, DIF reduces the likelihood of migrant populations pursuing entrepreneurship by increasing total urban employment demand and raising employees’ wages. These mediation effects, while statistically significant using traditional step-wise regression tests, should be interpreted cautiously given our use of city-level mediators in individual-level analysis and the absence of bootstrap significance testing, which represents a limitation of our current approach.

The employment demand mechanism operates through both traditional formal employment expansion and gig economy development. DIF platforms create new employment categories particularly accessible to migrants: delivery services, ride-sharing, domestic services, and micro-logistics, all enabled by integrated payment, credit, and identity verification systems. Unlike traditional employment requiring fixed addresses and established credit histories, these DIF-mediated opportunities allow migrants to participate in urban labor markets using mobile phones and basic identification, thereby expanding employment options beyond entrepreneurship. This dual-path employment expansion—formal jobs plus gig opportunities—helps explain the negative direct effect of DIF on necessity-driven entrepreneurship.

Specifically, the mediation tests show strong statistical support: in the employment demand pathway, DIF’s effect on unemployment rate (p = 0.000) and unemployment rate’s effect on entrepreneurship (p = 0.047) both achieve conventional significance levels. In the wage pathway, DIF’s effect on average wage (p = 0.000) and wage’s effect on entrepreneurship (p = 0.000) demonstrate robust statistical significance, confirming the reliability of both mediation channels.

Digital Finance Risks and Migrant Vulnerability Assessment

While our mechanism analysis reveals potential benefits from DIF-facilitated employment transitions, systematic risk assessment requires explicit consideration of potential negative consequences that warrant policy attention.

Digital Divide and Exclusion Risks. The digital divide documented in prior research [10] poses particular challenges for migrant populations who may lack stable internet access, digital literacy, or technological resources necessary to capitalize on DIF opportunities. This technological exclusion could inadvertently worsen existing inequalities if DIF development proceeds without targeted inclusion measures. Migrants in peripheral urban areas or informal settlements may experience substantially different DIF accessibility compared to those in central districts, potentially creating new forms of spatial-digital stratification.

Platform Labor Market Vulnerabilities. The transition from self-employment to DIF-enabled gig work, while potentially offering income stability, may expose migrants to new forms of labor market precarity including algorithmic management, income volatility, and limited worker protections. Unlike traditional self-employment where migrants retain control over work processes, platform-based employment often involves asymmetric power relationships that may undermine worker autonomy and bargaining power. These concerns align with reference [10]’s analysis of how digital technologies can systematically disadvantage marginalized populations.

Financial Risk and Over-indebtedness. DIF expansion may increase access to credit products without corresponding financial literacy support, potentially exposing vulnerable migrant populations to over-indebtedness risks. Mobile lending platforms’ convenience and reduced documentation requirements, while expanding access, may also facilitate imprudent borrowing decisions among populations with limited financial planning experience or irregular income streams.

These risk considerations underscore the importance of complementary policies addressing digital inclusion, platform worker protection, and financial education to ensure DIF development genuinely improves rather than merely transforms migrant economic vulnerability.

6.2. DIF’s Effect on the Entrepreneurial Performance of the Migrant Population

The preceding analysis focused on DIF’s effect on migrant populations’ entrepreneurial likelihood and its mechanisms. However, DIF might also influence the intensity or scale of their entrepreneurship. Drawing on Wang [45] and Feng [46], we use income and employee scale as proxy variables to estimate DIF’s effect on migrant populations’ entrepreneurial performance. To mitigate selection bias resulting from retaining only samples engaged in entrepreneurship, we apply Heckman’s [47] correction for regression analysis with the following models:

Ent(1,0)_ict = β₁+ β₂DIF_c(t−1)+ β₃X_ict + ε_ict,

(4)

lnincome_ict = α+βDIF_c(t−1)+ γX_ict + ε_ict,

(5)

where lnincome_ict represents the logarithm of entrepreneurial income; the definitions of other variables remain consistent with previous sections. The key variables influencing income include the DIF level, age, gender, education, political status, marital status, and health status.

Heckman Selection Correction: Substantial Identification Limitations. Our entrepreneurial performance analysis faces significant methodological constraints requiring cautious interpretation.

Our Heckman specification lacks valid exclusion restrictions, relying primarily on functional form identification through distributional assumptions. Household size and parental migration experience serve as potential quasi-instruments, hypothesized to affect entrepreneurship entry but not performance conditional on entry. This identifying assumption remains untestable and may not hold empirically.

Given these identification constraints, our performance results should be interpreted with considerable caution. While we find positive effects of DIF on both entrepreneurial income and employment scale, these estimates may reflect selection processes, functional form assumptions, or unmeasured factors rather than genuine performance improvements. The consistency across different specifications provides limited reassurance, but cannot overcome fundamental identification limitations.

We present these results as suggestive evidence requiring substantial future validation rather than reliable causal estimates. Future research with stronger exclusion restrictions, alternative selection correction methods, or experimental variation would be necessary to establish definitive conclusions about DIF’s performance effects among migrant entrepreneurs.

Ideally, the selection equation should include variables that affect entrepreneurship likelihood but do not directly influence entrepreneurial performance outcomes. In our specification, both equations include largely the same set of covariates, meaning identification relies primarily on functional form assumptions rather than exclusion restrictions. This approach, while common in the entrepreneurship literature when suitable instruments are unavailable, may result in weak identification and reduced precision of estimates. The lack of strong exclusion restrictions means our Heckman results should be interpreted as providing suggestive evidence of selection-corrected relationships rather than definitive causal estimates of DIF’s performance effects.

Table 10. DIF’s effect on migrant entrepreneurial performance.

	Income		Number of Employees
Type	(1)	(2)	(3)	(4)
DIF	2.3764 ***		20.8308 ***
	(0.2612)		(3.5093)
Internet		0.3606 ***		7.2658 ***
		(0.0543)		(0.8545)
Age	0.0149	0.0060	1.9822 ***	1.9857 ***
	(0.0142)	(0.0147)	(0.1726)	(0.1726)
Age2	−0.0397 **	−0.0305 *	−2.3884 ***	−2.3884 ***
	(0.0155)	(0.0161)	(0.2129)	(0.2130)
Gender	0.1956 ***	0.1904 ***	2.7623 ***	2.8042 ***
	(0.0163)	(0.0170)	(0.3520)	(0.3521)
PC	0.2804 ***	0.3082 ***	5.9235 ***	5.9564 ***
	(0.0311)	(0.0329)	(0.1869)	(0.1869)
Edu	0.2484 ***	0.2874 ***	5.6601 ***	5.6354 ***
	(0.0660)	(0.0683)	(0.8381)	(0.8382)
Marry	−0.3018 ***	−0.3762 ***	0.2486	0.3081
	(0.1013)	(0.1052)	(0.7211)	(0.7215)
Health	0.0658 ***	0.0593 ***	1.0430 **	1.0469 **
	(0.0206)	(0.0214)	(0.4403)	(0.4403)
Fsize	−0.1075 ***	−0.1240 ***	0.6355 ***	0.5299 ***
	(0.0183)	(0.0195)	(0.1791)	(0.1796)
Fic	3.8139 ***	4.0082 ***	12.4268 ***	12.5121 ***
	(0.2442)	(0.2586)	(2.3211)	(2.3182)
Fexp	−0.0748 ***	−0.0903 ***	1.9505 ***	1.8837 ***
	(0.0257)	(0.0270)	(0.4427)	(0.4429)
PGDP	0.1371 ***	0.4190 ***	−1.3202 **	−0.3175
	(0.0327)	(0.0493)	(0.5981)	(0.4148)
FD	0.0818 ***	0.1679 ***	0.4550	0.9878 ***
	(0.0167)	(0.0231)	(0.2823)	(0.2598)
Constant	−9.7294 ***	−0.1186	−204.0094 ***	−106.8244 ***
	(1.1996)	(0.2859)	(14.5376)	(6.3429)
Sample size	114,370	114,370	43,006	43,006

Note: Robust standard errors clustered at the city level in parentheses; *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

presents the estimation results for DIF’s effect on migrant populations’ entrepreneurial performance. Columns (1) and (2), with income as the dependent variable, show that the coefficients of the DIF level are statistically significant and positive at the 1% level. This indicates that while DIF reduces the likelihood of migrant populations pursuing entrepreneurship, it increases the income levels of those who engage in entrepreneurship, exerting a positive effect on entrepreneurial performance. DIF can influence regional economic environments and structures through technological innovation and consumption upgrading, which to some extent increases entrepreneurial income and improves entrepreneurial performance [20]. The IV regression results in column (2) validate the robustness of these findings. Columns (3) and (4), with the number of employees as the dependent variable, use a Tobit model for estimation since self-employment entrepreneurship has zero employees. The results reveal that DIF significantly increases the number of employees and enhances entrepreneurial performance. The IV regression results in column (4) further confirm the robustness of these outcomes. Hypothesis 4 is validated.

Model Assumptions and Limitations. Our entrepreneurial performance analysis employs Heckman selection correction and Tobit models, which assume normally distributed error terms. While these parametric approaches provide efficient estimates under correct specification, violations of normality assumptions could affect inference validity. Visual inspection of residuals and Shapiro–Wilk tests suggest reasonable adherence to normality, though some departure is observed in the tails of the distribution. To assess robustness to distributional assumptions, future research could employ non-parametric alternatives such as quantile regression for income analysis or bootstrap-based inference methods that do not rely on specific distributional assumptions. Additionally, our use of city-level aggregated variables in individual-level regressions may introduce ecological fallacy concerns that warrant consideration in interpreting results.

6.3. Theoretical Implications and International Perspectives

Our findings present important theoretical and policy implications that extend beyond the Chinese context, contributing to broader academic debates on digital finance and inclusive growth. This section critically situates our results within existing theoretical frameworks and discusses their relevance for international development policy.

Our findings provide evidence relevant to SDG 8.3, which calls for policies supporting “decent job creation and entrepreneurship.” While traditional interpretations assume digital financial inclusion primarily supports entrepreneurship, our results suggest DIF may contribute to SDG 8 by facilitating transition from precarious self-employment to stable wage employment, thereby advancing SDG 10’s goal of reducing inequalities for marginalized populations.

Theoretical Challenges and Contributions. Our findings pose significant challenges to conventional entrepreneurship theories. Traditional credit constraint models, rooted in the work of Evans and Jovanovic [48], predict that enhanced financial accessibility should stimulate entrepreneurship by reducing capital barriers. However, our results reveal a more complex dynamic where digital financial inclusion operates through labor market channels that can override financial market effects. The negative relationship between DIF and migrant entrepreneurship suggests that labor market mechanisms—specifically increased employment opportunities and wage premiums—may dominate financial inclusion benefits for vulnerable populations. This finding contributes to entrepreneurship theory by highlighting the critical importance of opportunity cost considerations, particularly for marginalized groups facing higher entrepreneurial risks and limited social safety nets.

Reconceptualizing Digital Finance Mechanisms. Our important evidence supports a dual-pathway model of digital finance impacts in developing economies. While DIF reduces information asymmetries and transaction costs—consistent with theoretical predictions—it simultaneously creates employment alternatives that increase entrepreneurship’s opportunity cost. This mechanism is particularly pronounced for necessity-driven entrepreneurs, such as migrant populations engaging in survival-oriented self-employment. The dual effect suggests that digital finance may function as both an enabler and a disruptor of entrepreneurial activity, with the net effect depending on the relative strength of financial inclusion versus labor market substitution effects. This reconceptualization has important implications for understanding why digital finance impacts may vary significantly across different populations and developmental contexts.

International Comparative Context. Our findings contribute to a growing body of evidence suggesting heterogeneous impacts of digital financial services across different developing economy contexts. While mobile money services have been widely credited with promoting financial inclusion and economic activity in various developing countries, the entrepreneurial effects appear context-dependent. The institutional environment, development stage, and demographic characteristics of target populations likely mediate these relationships. Our evidence from China’s migrant population suggests that in contexts where digital finance significantly expands formal employment opportunities, the traditional positive relationship between financial access and entrepreneurship may not hold. This insight has important implications for international development programs that assume universal positive effects of digital financial inclusion.

Development Stage Considerations. The apparent contradiction between our findings and some international experiences may reflect China’s particular development stage and institutional context. In economies undergoing rapid structural transformation, digital finance may accelerate the transition from informal, necessity-driven entrepreneurship toward formal wage employment. This transition, while potentially reducing overall entrepreneurship rates, may represent economic progress if it shifts workers from precarious self-employment to more stable, higher-paying formal employment. Our finding that DIF improves entrepreneurial performance among those who remain entrepreneurs supports this interpretation, suggesting a quality-improving selection effect. This perspective aligns with dual economy models that predict declining self-employment rates as economies develop and formal sector opportunities expand.

Global Policy Implications. These findings have important implications for international development policy and global financial inclusion initiatives. The results suggest that digital finance policies should be designed with careful consideration of local labor market conditions and target population characteristics. For vulnerable populations like migrants, the primary benefit of digital financial inclusion may not be direct entrepreneurship promotion but rather improved access to formal employment opportunities and enhanced performance for those who do engage in business activities. This nuanced understanding calls for more targeted approaches to digital finance policy that recognize the heterogeneous needs and circumstances of different population groups. Furthermore, our findings highlight the importance of complementary policies—such as entrepreneurship training and business development services—that can help marginalized populations capitalize on opportunities created by digital financial inclusion while preserving their entrepreneurial options.

Risk Mitigation and Digital Inclusion Challenges. While our findings suggest potential benefits from DIF-facilitated employment transitions, they also highlight critical risks that policymakers must address. The digital divide referenced in our analysis [10] poses particular challenges for migrant populations, who may lack the digital literacy, stable internet access, or technological resources necessary to fully capitalize on DIF opportunities. This technological exclusion could inadvertently worsen existing inequalities if DIF development proceeds without targeted inclusion measures. Furthermore, the shift from self-employment to platform-based gig work, while potentially offering higher income stability, may expose migrants to new forms of labor market precarity including algorithmic discrimination, unpredictable income streams, and limited worker protections. Effective DIF policy implementation requires complementary interventions addressing digital literacy training, platform worker protection, and maintaining entrepreneurial pathways for those who prefer self-employment.

Capability Approach and ESG Framework Considerations. While our analysis does not comprehensively integrate capability approach or environmental-social-governance (ESG) frameworks, these theoretical perspectives offer important lens for interpreting our findings. From a capability approach standpoint, our results suggest that DIF expansion may enhance certain capabilities (employment access, income stability) while potentially constraining others (entrepreneurial autonomy, self-determination). The transition from self-employment to platform-based work represents a complex trade-off between different forms of human development that warrants deeper theoretical exploration.

Similarly, our sustainability claims would benefit from more rigorous ESG framework integration, particularly regarding social governance indicators such as income distribution, worker protection, and digital inclusion metrics. The absence of comprehensive inequality measures (Gini coefficients, poverty indices) in our analysis represents a significant theoretical limitation that constrains definitive sustainability claims. Future research should incorporate explicit capability and ESG frameworks to provide more nuanced understanding of digital finance impacts on human development and sustainable growth.

6.3.1. Methodological Contributions and Limitations

Our empirical approach employs standard techniques in the entrepreneurship and development finance literature, including instrumental variable and mediation analysis methods. We do not claim methodological innovation; rather, our contribution lies in the systematic application of these established methods to examine DIF’s effects on a previously understudied population. The integration of city-level DIF data with individual-level migrant survey data, while methodologically standard, provides valuable robust findings for understanding digital finance impacts on vulnerable populations.

However, we acknowledge significant limitations. Our empirical strategy employs city-level clustered standard errors to address potential within-city correlation among individual observations, which is essential for valid statistical inference in multi-level data structures. Our cross-sectional design constrains causal inference despite IV techniques. The potential correlation between our instrumental variable (internet penetration) and unmeasured urban development factors represents a limitation to strong causal interpretation. Additionally, our mediation analysis using city-level mediators raises ecological fallacy concerns, and our reliance on traditional step-wise regression methods rather than bootstrap procedures for testing mediation significance represents a methodological limitation that could affect the robustness of our mechanism inferences. Future research employing longitudinal data or natural experiments could provide stronger causal identification.

Robustness of our findings. Although this study provides valuable insights into the relationship between Digital Inclusive Finance (DIF) and migrant entrepreneurship in China, several limitations warrant acknowledgment and indicate important directions for future research.

Methodological and Data Limitations. This analysis relies on cross-sectional data from the 2017 China Migrants Dynamic Survey matched with the 2016 DIF Index, which constrains our ability to establish definitive causal relationships or to capture dynamic entrepreneurial trajectories over time. While our instrumental variable approach and the use of lagged explanatory variables help mitigate endogeneity concerns, residual confounding from unobserved time-varying factors cannot be entirely ruled out. Future studies employing panel or longitudinal data could better disentangle causality and reveal temporal adjustment processes in the DIF–entrepreneurship relationship.

Measurement and Construct Limitations. The DIF Index, while comprehensive, is compiled at the prefecture level and derived primarily from Alipay transaction data, capturing only part of the broader digital finance landscape—especially in regions where alternative platforms, informal channels, or emerging fintech services play significant roles. Aggregation at this spatial scale may also mask within-city heterogeneity in individual access to and utilization of digital financial services. Moreover, our binary classification of entrepreneurship, though consistent with established literature, may overlook nuanced forms of entrepreneurial activity and hybrid employment arrangements increasingly prevalent in the digital economy.

Contextual and Generalizability Constraints. The findings are embedded within China’s distinctive institutional and developmental environment, characterized by the hukou household registration system, particular rural–urban migration flows, and a digital finance ecosystem dominated by super-apps such as Alipay and WeChat Pay. Moreover, our analytical scope focuses specifically on China’s “floating population”—rural-born individuals living in cities without local household registration for at least one month—which represents only one segment of the broader entrepreneurial landscape. This focus excludes multiple migrant categories: return migrants who move back to rural areas, urban-born entrepreneurs, short-term migrants with less than one month residency, undocumented migrants who may avoid official surveys, and circular migrants with highly fluid mobility patterns. The residency requirement inherent in our sample definition creates potential bias toward more settled migrant populations, who may have different risk preferences, capital access patterns, and entrepreneurial motivations compared to highly mobile or temporary migrants. This sample selection could systematically underrepresent the most vulnerable and opportunistic segments of China’s migrant population, all of whom may exhibit different relationships between digital finance access and entrepreneurial behavior. These contextual and scope specificities may limit direct generalization to other economies with different institutional frameworks, migration patterns, or stages of digital finance development. While our identified mechanisms—employment demand expansion and wage enhancement—may operate elsewhere, their relative importance and manifestation could vary considerably across contexts.

Temporal Limitations. Beyond these contextual constraints, the temporal scope of our dataset warrants explicit consideration. Our analysis draws on 2016 DIF Index data matched with 2017 CMDS microdata, whereas China’s digital finance ecosystem has since undergone major developments. Notable changes include the suspension of Ant Group’s IPO in 2020, the subsequent tightening of fintech regulations, the nationwide rollout of digital yuan (e-CNY) pilot programs, and the COVID-19 pandemic’s acceleration of cashless payment adoption. Such changes may have reshaped the accessibility, usage patterns, and mechanisms through which DIF influences migrant entrepreneurship. In particular, post-2020 regulatory adjustments could have altered the risk–return trade-offs integral to entrepreneurship; expanded DIF coverage may have shifted baseline access for migrant populations; and pandemic-driven digitalization has created new employment and consumption dynamics that our historical dataset cannot capture. Accordingly, caution is warranted in extrapolating our findings to the current market and policy environment. Future research should therefore validate these findings using post-2020 nationally representative data, enabling assessment of the temporal robustness of the identified mechanisms and the potential emergence of new dynamics in DIF’s impact on migrant entrepreneurship.

6.3.2. Empirical Contribution and Boundary Conditions

While our theoretical frameworks build incrementally on existing literature, our findings provide important evidence about boundary conditions under which digital finance theories operate. By demonstrating that digital financial inclusion can reduce entrepreneurship among institutionally disadvantaged populations, we challenge the universal assumption that financial inclusion necessarily promotes entrepreneurship. This represents significant empirical contribution despite limited theoretical innovation.

Testing established mechanisms (employment demand expansion, wage enhancement) among China’s migrant population—facing extreme institutional constraints—extends understanding of when and how these mechanisms function in practice. The scale of our context (376 million migrants) and counterintuitive findings have implications extending beyond China to other developing economies pursuing digital financial inclusion strategies.

Scope and Population Coverage Limitations. Our focus on registered migrants with minimum one-month residency creates important analytical boundaries that warrant explicit acknowledgment. This design excludes several migrant categories whose entrepreneurial behaviors may differ substantially: (1) return migrants who move back to rural areas after urban experience; (2) interactions between migrants and local urban populations that may influence entrepreneurial dynamics; (3) highly mobile circular migrants with fluid movement patterns; (4) short-term migrants who may rely more heavily on necessity-driven entrepreneurship; and (5) undocumented workers who avoid official surveys but represent significant populations in certain sectors.

Regional Heterogeneity Constraints. Our analysis aggregates results across diverse regional contexts (coastal vs. inland, tier-1 vs. lower-tier cities) that may mask important geographic variations in DIF–entrepreneurship relationships. Coastal regions with more developed digital infrastructure and diversified economies may exhibit different patterns compared to inland areas with limited alternative employment options. This aggregation approach, while necessary for statistical power, may obscure contextual factors that mediate DIF effects across China’s varied development landscape.

These scope limitations suggest our findings apply most directly to settled migrant populations in formal employment relationships, potentially underrepresenting entrepreneurial dynamics among the most vulnerable and mobile segments of China’s migrant workforce.

6.4. Research Extensions and Methodological Opportunities

Furthermore, while our instrumental variable approach addresses endogeneity concerns and provides robust causal identification, we acknowledge that additional sensitivity tests could further strengthen our findings. Future research could benefit from employing propensity score matching to account for selection on observables, and falsification tests using placebo outcomes to provide additional validation of our causal identification. These represent important methodological extensions that would build upon our baseline findings and enhance the robustness of conclusions about DIF’s effects on migrant entrepreneurship.

Longitudinal Research Extensions. Our cross-sectional analysis provides crucial baseline evidence establishing fundamental DIF–entrepreneurship relationships. Future longitudinal studies tracking individual entrepreneurial trajectories could strengthen causal identification while revealing dynamic adjustment processes.

Instrumental Variable Limitations. Beyond cross-sectional constraints, our instrumental variable approach faces specific methodological challenges. Internet penetration may correlate with unmeasured regional characteristics such as innovation ecosystems, entrepreneurial culture, and institutional arrangements that directly influence migrant entrepreneurship decisions, potentially violating the exclusion restriction assumption fundamental to causal identification. This represents a broader challenge in identifying truly exogenous variation in digital finance development, as most plausible instruments relate to technological or institutional factors that may have independent effects on entrepreneurship outcomes. While our DIF-specific mechanism focus and comprehensive control strategy provide theoretical support for our identification approach, definitive causal validation would benefit from natural experiments exploiting exogenous policy variations or alternative instrumental variables in future research.

Methodological Robustness Considerations. Our analysis reveals moderate correlations between key explanatory variables (DIF, GDP per capita, and traditional financial development), though multicollinearity diagnostics indicate acceptable VIF levels. Future research could benefit from alternative estimation approaches that do not assume linear relationships, such as machine learning methods or non-parametric techniques. Additionally, our performance analysis relies on parametric models (Heckman and Tobit) that assume normally distributed errors. While diagnostic tests suggest reasonable adherence to these assumptions, robustness checks using quantile regression, bootstrap inference, or other distribution-free methods would strengthen confidence in the entrepreneurial performance findings. More critically, our Heckman selection correction lacks explicit exclusion restrictions for robust identification, relying instead on functional form assumptions that may compromise the reliability of our selection-corrected performance estimates. Such methodological diversification would help assess whether our core findings persist across different analytical frameworks and distributional assumptions.

Heckman Selection Model Limitations. Our entrepreneurial performance analysis using Heckman selection correction faces fundamental identification challenges that warrant explicit acknowledgment. The absence of valid exclusion restrictions—variables that affect the selection into entrepreneurship but do not directly influence entrepreneurial performance—means our identification relies primarily on functional form assumptions and the nonlinearity of the probit selection equation. This functional form identification is generally considered weaker than identification through exclusion restrictions, as it depends on distributional assumptions that may not hold in practice. The risk of weak identification is particularly concerning in Heckman models, as it can lead to imprecise estimates and unreliable inference about selection-corrected effects. Future research would benefit from identifying suitable exclusion restrictions, such as regional variations in entrepreneurship support policies, historical migration patterns, or family business traditions that affect entrepreneurship entry but do not directly determine business performance outcomes. Alternatively, employing alternative selection correction methods less dependent on functional form assumptions, or acknowledging that selection-corrected results provide bounds rather than point estimates, would enhance the robustness of performance effect conclusions.

Mediation Analysis Limitations. Our mechanism analysis employs city-level mediators (unemployment rate and average wage) in individual-level regressions, raising ecological fallacy concerns that could affect the validity of our mediation inferences. The assumption that aggregate urban labor market conditions uniformly affect individual migrant decisions may not hold across all population subgroups or urban contexts. Additionally, our mediation analysis relies on traditional step-wise regression methods rather than bootstrap procedures, which would provide more robust significance testing for indirect effects. Future research would benefit from individual-level mediation data, bootstrap-based significance testing, and multilevel modeling approaches that explicitly account for the hierarchical structure of individual migrants nested within cities. These limitations may affect both the magnitude and statistical significance of our estimated indirect effects, potentially leading to over- or under-estimation of the true mediation relationships.

Selection Bias and Policy Coordination Challenges. Beyond the cross-sectional and instrumental variable limitations discussed above, our identification strategy faces additional challenges from migrant selection behavior and coordinated policy implementations that warrant explicit acknowledgment. Systematic migrant selection of destinations with favorable digital financial infrastructure could create reverse causality that our cross-sectional framework cannot eliminate. Moreover, local development strategies may coordinate DIF promotion with entrepreneurship support through comprehensive policy packages, generating correlated interventions that violate our identification assumptions. Future research would benefit significantly from difference-in-differences approaches exploiting exogenous variations in DIF rollout timing across regions, panel data tracking individual migration and entrepreneurship trajectories over time, or natural experiments providing quasi-random assignment to digital finance access. Such identification strategies would address the fundamental endogeneity concerns our current approach cannot fully resolve while building upon the baseline relationships we establish.

Unmeasured Heterogeneity. Despite controlling for a wide array of individual, household, and city-level factors, unobserved influences such as local entrepreneurial culture, informal institutional arrangements, variations in policy execution, social networks, and individual entrepreneurial capabilities may affect both DIF development and entrepreneurial behavior in ways our model cannot fully capture.

Future Research Directions. Addressing the above limitations will enhance both the robustness and the external validity of future work. Longitudinal data tracking individuals’ entrepreneurial trajectories could yield stronger causal identification and reveal dynamic responses to DIF expansion. Comparative studies across countries at varying stages of development could test how mechanisms identified here perform in diverse institutional settings. Richer measures of digital finance—covering multiple platforms, individual-level usage patterns, and qualitative dimensions of financial inclusion—would improve measurement precision. Finally, exploring heterogeneous impacts across specific digital financial services (e.g., payments, credit, insurance, investment) could provide more nuanced insights into the distinct pathways shaping entrepreneurial outcomes.

Additionally, future research could examine heterogeneous effects across educational levels and regional variations (such as eastern versus western China), which may reveal important nuances in how individual capabilities and regional development contexts mediate DIF’s entrepreneurship effects. Our current heterogeneity analysis, while covering gender, age, household registration, and migration range, could be extended to explore how different educational backgrounds and regional development disparities influence the DIF–entrepreneurship relationship. Expanding the analytical scope to include return migrants, urban-born entrepreneurs, circular migration patterns, short-term migrants, and undocumented workers would provide a more comprehensive understanding of digital finance impacts across China’s diverse demographic landscape. Particular attention should be paid to understanding how sample selection toward settled migrants may have influenced our findings, and whether the relationships we identify hold for more mobile and vulnerable migrant populations who face different institutional constraints and entrepreneurial opportunities. Such extensions would enhance both the theoretical robustness and practical relevance of findings for policy design targeting different population groups.

7. Conclusions and Implications

Theoretical Contributions and Empirical Evidence. This study analyzes 114,370 migrants from China’s 2017 National Migrant Population Dynamic Survey, matched with city-level DIF Index data. Three empirical findings demonstrate how institutional exclusion moderates digital finance-entrepreneurship relationships. First, DIF significantly reduces migrant entrepreneurship likelihood (marginal effect: −0.449, p < 0.01), contrasting with prior studies documenting positive effects among urban residents and rural farmers. Second, this suppressive effect concentrates in necessity-driven entrepreneurship—operationalized as individual business ownership without employees (marginal effect: −0.426, p < 0.01)—while showing no significant impact on opportunity-driven entrepreneurship (operationalized as employer-type ventures with hired employees). Third, despite reducing overall entrepreneurship rates, DIF substantially improves performance among remaining entrepreneurs, increasing both income and employment scale significantly (p < 0.01).

Mechanism analysis identifies urban employment demand expansion and wage enhancement as significant mediation pathways (Table 9). These findings support our theoretical expectation that hukou-based employment barriers create conditions where DIF operates primarily through employment access improvement rather than financial constraint reduction. Heterogeneity analysis reveals stronger effects among agricultural hukou holders (interaction: 0.057, p < 0.01), older workers (age interaction: 0.0098, p < 0.01), and interprovincial migrants. Systematic patterns across theoretically predicted vulnerability dimensions strengthen inference beyond pure correlational analysis despite cross-sectional and instrumental variable limitations. These findings provide empirical evidence that institutional constraints operate as critical moderating variables in digital finance-entrepreneurship relationships, extending theory to account for populations facing codified discrimination rather than only financial market imperfections.

These findings provide robust baseline evidence for policy consideration while identifying important extensions for future research.

Our cross-sectional analysis using 2016 DIF Index and 2017 CMDS data establishes baseline DIF–entrepreneurship relationships among China’s migrant population. While instrumental variable analysis using internet penetration rates provides directional consistency checks, acknowledged exclusion restriction violations limit causal interpretation. Results should be interpreted as suggestive baseline evidence requiring validation through difference-in-differences designs or natural experiments addressing coordinated policy and migrant selection challenges our framework cannot resolve.

Key robustness considerations include: (1) heterogeneity analysis across entrepreneurial types (self-employment vs. employer-type) confirms differential effects; (2) DIF subdimensions (coverage breadth and usage depth) yield consistent negative associations; (3) performance improvements among remaining entrepreneurs suggest quality-enhancing selection. However, Heckman selection correction lacks explicit exclusion restrictions, requiring cautious interpretation of performance estimates.

Based on our empirical findings, which should be interpreted as baseline evidence establishing fundamental patterns rather than definitive policy prescriptions, we suggest preliminary considerations for stakeholder interventions that acknowledge DIF’s complex effects while recognizing the need for pilot testing and comprehensive impact evaluation. Government agencies should implement differentiated strategies: establish DIF integration buffer periods in migrant-concentrated areas providing transitional support for survival-oriented entrepreneurs; create dedicated programs facilitating migrant transition from self-employment to employer-type entrepreneurship through skills training and opportunity identification; develop inter-regional coordination mechanisms ensuring consistent migrant entrepreneurship support across provinces, addressing our finding of stronger effects on interprovincial migrants.

Financial institutions must address accessibility barriers while mitigating digital divide risks revealed in our heterogeneity analysis: develop gig economy integration programs that connect migrants with platform-based employment opportunities while ensuring fair algorithmic practices and transparent income structures, provide comprehensive digital literacy training that encompasses both financial and platform navigation skills, implement worker protection mechanisms within gig platforms including income stabilization features and dispute resolution systems, and ensure social security portability within platform work; design simplified digital interfaces specifically for migrants with agricultural household registration, reducing operational complexity; develop mobility-inclusive risk assessment models that recognize migrant geographic mobility as entrepreneurial capital rather than risk; implement graduated financing frameworks providing scalable support from initial self-employment ventures to expanded employer-type enterprises.

Local stakeholders can bridge implementation gaps through targeted interventions: establish migrant entrepreneur support networks with dedicated coordination in high-migration communities; implement preferential procurement policies counterbalancing DIF-induced competitive pressures on migrant small businesses; create structured mentorship programs connecting established entrepreneurs with emerging migrant entrepreneurs. Success requires monitoring both employment quality improvements and entrepreneurial diversity, ensuring digital financial inclusion benefits extend equitably while preserving the innovative capacity characterizing migrant contributions to urban development.

Critical Risk Considerations. While our findings suggest DIF can facilitate beneficial employment transitions for migrant populations, several risks require explicit policy attention. First, the digital divide may exclude the most vulnerable migrants from DIF benefits, potentially exacerbating existing inequalities as referenced in our analysis [10]. Second, the transition from self-employment to gig economy participation, while offering certain advantages, may expose workers to new forms of precarity including algorithmic management, income volatility, and reduced worker protections. Third, over-reliance on platform-based employment could undermine migrants’ entrepreneurial capabilities and long-term economic mobility. Policymakers should therefore implement complementary measures including digital inclusion programs targeting vulnerable migrant subgroups, platform worker protection regulations ensuring fair labor standards, preservation of pathways for those preferring traditional entrepreneurship, and monitoring systems to track whether DIF development genuinely improves rather than merely transforms migrant economic vulnerability.

Finally, we acknowledge that our focus on rural-origin floating populations in urban China—defined by official hukou status and minimum one-month residency requirements—while offering valuable insights, remains context-specific to the country’s unique institutional system and creates potential sample selection bias. This scope necessarily excludes return migrants, urban-born entrepreneurs, short-term migrants, undocumented workers, and highly mobile circular migrants, whose entrepreneurial behaviors and responses to digital finance may differ markedly from the more settled migrant populations captured in our analysis. The residency-based sample definition may systematically bias our findings toward migrants with stronger local integration and more stable employment patterns, potentially limiting the generalizability of our conclusions to the most vulnerable and mobile segments of China’s migrant workforce. Our analysis also does not yet capture potential heterogeneity by education level or by regional development disparities (e.g., eastern vs. western China). Extending future investigations to these additional populations and contexts will be essential for testing robustness, clarifying boundary conditions, and assessing the broader applicability of our findings.