Broadband Access and Household Risky Assets Investment: A Double/Debiased Machine Learning-Based Difference-in-Difference Approach

Abstract

This study analyzes the impact of broadband access on household investments in high-risk assets by employing the Double/Debiased Machine Learning-based Difference-in-Difference (DMLDiD) method. The results demonstrate a significant increase in these investments, propelled by enhanced broadband access. This increase is particularly evident among younger, more educated households in urban areas and Eastern regions, highlighting broadband’s influence on financial behaviors across diverse demographics. The study’s novel implementation of the DMLDiD approach in analyzing household behavior enhances the current literature. It provides a comprehensive view of the influence of digital infrastructure on economic decisions across various demographic segments. This research enhances comprehension of the multifaceted effects of technology on household financial strategies, highlighting the necessity of targeted digital infrastructure development.

JEL Classification: D14, G11, O33.

Keywords

broadband access risky assets investment double/debiased machine learning difference-in-difference analysis household financial behavior

Introduction

The rapid global expansion of broadband internet represents a pivotal shift in the digital era, significantly impacting multiple facets of life and the economy (Koutroumpis, 2019). The rapid and extensive expansion of broadband infrastructure in China has significantly influenced this worldwide phenomenon (Ito, 2019). This digitalization initiative in China is transforming household financial behaviors as broadband internet access becomes increasingly essential in day-to-day life. The Chinese context exhibits a substantial capacity for broadband to influence household financial decisions, especially in risky assets investment (Liang & Guo, 2015). This transformative influence of broadband internet in China highlights its influence as both a communication tool and a pivotal factor in households’ financial decision-making processes, thereby signifying an essential domain of study in household finance (Shim & Shin, 2016).

A burgeoning corpus of research explores the internet’s influence on household finance; however, a substantial knowledge gap persists concerning the specific impact of broadband internet on household investments in financial assets, especially within the Chinese context. Despite the recognition of broadband’s influence on areas such as income, consumption, and employment (Alam & Imran, 2015; Houngbonon & Liang, 2021; Jin et al., 2023; Khanal & Mishra, 2016; Mora-Rivera & García-Mora, 2021; Wan et al., 2021), its direct correlation with household financial assets investment decisions remains underexplored. This gap is particularly conspicuous in China, where rapid digitalization and a unique economic landscape create a specific context for examining the impact of broadband access on households’ financial asset allocation, particularly regarding riskier investments. This study seeks to address this gap by providing new perspectives on the complex relationship between broadband availability and household financial decision-making related to risky asset investment in China.

This study examines critical research questions regarding the impact of broadband internet on households’ investments in risky financial assets within China’s rapidly evolving digital landscape. It aims to comprehensively analyze how enhanced broadband access influences Chinese households’ financial behaviors, particularly regarding their propensity to invest in riskier assets. It seeks to ascertain the magnitude of this impact, its variations across various demographic groups, and the influence of broadband access on such investment decisions. This investigation is crucial for understanding the dynamic interplay between technological advancement and household financial strategies, significantly impacting household finance.

This research examines China as a unique case study due to its remarkable journey of digitalization and its specific implications for household financial decision-making (Ahmad et al., 2021; Xu et al., 2022). The country’s rapid adoption and expansion of broadband technology create a unique framework for exploring how enhanced digital connectivity impacts households’ investment decisions, particularly concerning high-risk financial assets. China’s broadband expansion, characterized by its scale and speed (Jiang & Murmann, 2022), provides a valuable context for investigating these dynamics and enhancing comprehension of the relationship between technological advancement and household finance in a rapidly digitalizing economy.

This study meticulously examines the impact of broadband internet on household investments in risky financial assets in China using the innovative Double/Debiased Machine Learning-based Difference-in-Difference (DMLDiD) method. Particularly, the DMLDiD approach is appropriate for this analysis because it integrates the robustness of traditional difference-in-difference techniques with the advanced functionalities of machine learning (Chang, 2020; Chernozhukov et al., 2018; Sant’Anna & Zhao, 2020). This method enables a more accurate treatment effect estimation by more effectively addressing potential biases and confounding factors than standard methods. The complex relationship between broadband access and household financial decisions, which may influence numerous unobserved factors, renders its application in this context crucial. Therefore, the DMLDiD method offers an advanced analytical framework for elucidating these intricate dynamics and deriving more precise and reliable insights.

A comprehensive empirical strategy is utilized to guarantee this study’s robustness. This strategy includes several measures to mitigate potential biases and validate the approach within China’s unique digital landscape. First, we perform a rigorous pre-treatment trend analysis to validate the parallel trends assumption, a crucial condition for ensuring the credibility of the difference-in-differences methodology. This involves analyzing the patterns of household investments in high-risk assets before the extensive adoption of broadband internet. Additionally, the study utilizes various robustness checks, including sensitivity analyses and alternative model specifications, to test the results’ consistency. These checks mitigate concerns regarding unobserved confounding variables and potential biases associated with the staggered implementation of broadband access in various regions. Furthermore, the application of the DMLDiD method is rigorously validated within the context of China’s distinct economic and technological environment, ensuring the findings’ applicability and relevance to the Chinese context. This meticulous approach reinforces the study’s credibility and emphasizes its contribution to household finance.

This study highlights the significant impact of broadband access on the investment behaviors of Chinese households in high-risk financial assets. The impacts are mainly due to increased household income, enhanced social networks, and reduced credit constraints. Enhanced connectivity unlocks economic opportunities and broadens market access, thereby increasing household income (Bauer, 2018; Khanal & Mishra, 2016; Mora-Rivera & García-Mora, 2021). Additionally, it reinforces social networks, which are essential for investment decisions (Evans et al., 2021; Geraci et al., 2022; Liang & Guo, 2015), and mitigates credit constraints (H. Chen & Yoon, 2022; Y. Chen et al., 2018). The effects of broadband access differ markedly across various demographic and regional segments, exhibiting a more pronounced influence in urban areas and among younger, more educated households. This underscores the diverse impact of digital infrastructure, emphasizing its impact on investment behaviors among different societal strata.

The subsequent sections of this paper are structured as follows: Section 2 presents the literature review and research hypotheses, outlining the theoretical foundations and key mechanisms by which broadband access may influence household investment in high-risk assets. Section 3 outlines the institutional background, detailing the Broadband China policy and its relevance to this study. Section 4 describes the research methodology, elucidating the empirical strategy and the application of the DMLDiD approach. Section 5 delineates the research results, analyzing the empirical findings and their ramifications. Finally, Section 6 summarizes the key insights and discusses their broader significance for financial decision-making and digital infrastructure policy.

Literature Review and Research Hypotheses

Effect of Broadband Internet on Household Investment Behavior

The expansion of broadband internet has significantly transformed household financial behaviors, particularly regarding asset allocation and investment decisions. Theories of economic decision-making emphasize the significance of information availability, access to financial markets, and the minimization of transaction costs in influencing investment behavior (Sood et al., 2025). Broadband internet, as a type of digital infrastructure, can substantially enhance households’ access to information and financial services, thereby directly influencing their investment decisions (Thomas & Finn, 2018; Wan et al., 2021). Broadband access has liberated households from reliance on traditional, physical sources of financial information. They can now leverage online platforms, real-time market data, and financial services that enable them to diversify their portfolios and explore higher-risk financial assets such as stocks, mutual funds, and bonds (Hvide et al., 2024; Zareei, 2019).

According to portfolio theory, access to a wider array of information enables households to make more informed asset allocation decisions. Access to real-time financial news, stock market data, and online financial advice online mitigates the uncertainty that households encounter when contemplating risky investments, thereby increasing the probability of their engagement with such assets (Ye et al., 2022). Furthermore, the theory of financial inclusion posits that access to digital financial services can diminish barriers to participation in financial markets, especially for households in underbanked or underserved areas. Broadband internet facilitates household access to these services, potentially enhancing participation in financial markets, especially among those previously excluded or hesitant due to limited information or perceived high costs (Beck et al., 2018; Gomber et al., 2017).

Current literature indicates that digital infrastructure, such as broadband, can significantly enhance economic decision-making. Studies across multiple sectors highlight that broadband access fosters more informed and diversified decision-making, and these principles can also be applied to household finance (Shin et al., 2020). Broadband enables households to learn more about investment opportunities, monitor market trends, and engage with financial advisors or online communities that share investment strategies (Hvide et al., 2024). This access empowers households to expand their investment horizons, especially in riskier assets that they may have otherwise shunned due to information asymmetries or lack of access to suitable platforms.

Therefore, as broadband access expands, the proportion of households investing in financial assets is expected to increase, particularly in higher-risk investments. This shift is due to increased confidence in making informed financial decisions, facilitated by easy access to extensive financial information, advice, and services.

Hypothesis H1: Access to broadband internet raises the probability that households engage in high-risk asset investment.

Household Income and Financial Decision-Making

Household income is a key determinant of financial decision-making, influencing the ability and willingness to invest in high-risk assets. Higher-income levels enhance financial stability and expand households’ capacity to assume investment risks (Shi & Lim, 2024). Broadband internet can enhance income growth by expanding job opportunities, enhancing productivity, and facilitating access to digital financial services. As income increases, households acquire greater financial flexibility, thereby increasing their propensity to invest in high-risk assets such as stocks and bonds (Gallardo et al., 2021; Stockinger, 2019). Therefore, broadband-driven income growth functions as a crucial mechanism linking internet access to household investment behavior.

Hypothesis H2: Broadband internet access increases household income, resulting in an increased propensity to invest in high-risk assets.

Social Networks and Investment Behavior

Social network theory posits that the robustness of a household’s social ties can significantly influence its investment decisions. Broadband internet facilitates the expansion of social networks by enabling individuals to connect more effortlessly via online platforms (Anderson, 2008; Zhou et al., 2023). These enhanced social networks enable households to obtain financial advice, exchange investment strategies, and discover investment opportunities from peers, family members, and online communities (Lei & Ramos Salazar, 2022; Ostrovsky-Berman & Litwin, 2019). With the enhancement of broadband access, households are anticipated to leverage these networks to make more informed financial decisions, including the decision to invest in higher-risk financial assets. Therefore, the expansion of social capital via broadband is anticipated to positively influence households’ participation in financial markets.

Hypothesis H3: Broadband internet access enhances social networks, thereby increasing the likelihood of households investing in high-risk assets.

Credit Constraints and Household Investment

Broadband internet access may also alleviate credit constraints encountered by households. Broadband enables households to access a broader range of financial services, including online lending platforms, credit assessments, and financial advice (Acolin et al., 2019; Jia, 2025; Lin et al., 2015). This enhanced access to financial services can assist households in surmounting traditional barriers to credit, enabling them to invest more in higher-risk assets. Additionally, households with broadband internet are more inclined to be cognizant of diverse financial products and services that can alleviate credit constraints and expand investment opportunities (Peng et al., 2021; Weng et al., 2022). The ability to secure loans, credit, or financial products with greater ease may encourage households to invest more resources in high-risk financial assets.

Hypothesis H4: Broadband internet access alleviates credit constraints, thereby increasing household investment in high-risk assets.

Contribution to Existing Literature

This study enhances the literature on household finance and impacts digital infrastructure in two significant ways. First, it contributes to the existing literature by focusing specifically on how broadband internet affects household investment in high-risk financial instruments. Although much of the existing literature has concentrated on the broader economic impacts of broadband, such as its effects on income or general household welfare (Bahia et al., 2020; Dettling, 2017; Ford, 2018; Hambly & Rajabiun, 2021; Jung & López-Bazo, 2020; Koutroumpis, 2019; Ma et al., 2020; Masaki et al., 2020; Valentín-Sívico et al., 2023; Wan et al., 2021), this study presents a novel perspective by exploring how broadband access directly influences household financial behavior, particularly regarding risky assets allocation. This addresses a critical gap in the literature concerning the microeconomic implications of broadband in rapidly digitalizing economies such as China.

Second, this study presents an innovative methodological approach by utilizing the DMLDiD method. This approach enhances conventional Difference-in-Differences (DID) and regression techniques, which frequently encounter issues related to omitted variables and selection bias (Abadie, 2005; Chang, 2020). The DMLDiD method leverages machine learning techniques to improve the accuracy of causal inference, thereby enhancing the reliability and precision of the findings. This methodological innovation strengthens the analysis by providing more reliable evidence on how broadband access influences household investment decisions.

Institutional Background

The Broadband China policy, initiated by the Chinese government on August 17, 2013, is a strategic program aimed at significantly enhancing the country’s broadband infrastructure. This policy, which constitutes an integral component of the national strategy, formulates a comprehensive 8-year development blueprint focusing on the fundamental significance of broadband in fostering economic and social advancement. It establishes specific targets for a phased implementation. By 2015, the goals included establishing extensive urban fiber-optic connectivity, expanding broadband access in rural areas, achieving a 50% household fixed broadband penetration rate, and attaining a 32.5% penetration rate for 3G/LTE users. Further, it targeted 95% broadband coverage in administrative villages. For 2020, the goals were expanded to encompass full broadband coverage in urban and rural areas, targeting a 70% fixed broadband penetration rate, an 85% 3G/LTE user rate, and broadband coverage in administrative villages that exceeds 98%.

The initiation of pilot projects between 2014 and 2016 was a pivotal element of the Broadband China policy. The launch of these pilot projects in various cities and regions signified a critical phase of policy implementation. They were instrumental in assessing the feasibility and impact of the policy on a smaller scale before its broader application. These pilots provided significant insights into the challenges and opportunities of broadband development, impacting the subsequent phases of policy implementation. In 2014, for instance, the government designated 39 cities and city groups for the initial series of pilot projects. These encompassed major metropolitan areas such as Beijing, Shanghai, Tianjin, and other regional hubs. In 2015, the scope expanded, incorporating additional cities and regions into the pilot program, thereby demonstrating a commitment to enhancing broadband accessibility across diverse geographic and demographic areas. By 2016, the pilot projects had been expanded to encompass a more diverse range of urban and rural settings. These pilots were instrumental in evaluating various facets of broadband deployment, including infrastructure buildouts, service delivery, and usage patterns.

These pilot projects assessed both the technical and logistical implementation of the Broadband China policy and gauged the socio-economic impacts of enhanced broadband access. By implementing this strategic approach, China has successfully ensured that the benefits of broadband expansion are widespread and equitably distributed. This aligns with the country’s vision of digital inclusivity and its contribution to economic growth and societal advancement. The insights and lessons derived from these pilot phases have been vital in refining the policy’s trajectory and ensuring its efficacy in attaining its ambitious goals.

Research Methodology

Data Source

This study employs data from the China Household Finance Survey (CHFS) to empirically analyze the impact of broadband internet on household investment in high-risk assets. The CHFS, administered by the China Household Finance Survey and Research Center at the Southwestern University of Finance and Economics, offers an extensive study dataset for this study. The survey, which was initiated in 2011, is executed biennially and offers extensive financial data on Chinese households.

This study utilizes panel data from the survey waves of 2013, 2015, and 2017. The selection of these specific years is strategic: although the 2011 survey offers valuable insights, notable methodological discrepancies in questionnaire design and sample selection between the 2011 survey and subsequent waves may undermine the consistency necessary for thorough longitudinal analysis. Additionally, to eliminate confounding effects associated with the COVID-19 pandemic—which significantly influenced household financial behaviors—survey data gathered post-2019 were deliberately excluded.

To enhance robustness and consistency in our DMLDiD analysis, we included only households observed in a minimum of two survey waves, thereby excluding those surveyed only once. The final dataset comprises 77,337 observations, with 36.4% of households observed in precisely two waves and 63.6% in all three waves. Specifically, the dataset comprises 20,765 households in 2013, 30,416 in 2015, and 26,436 in 2017. This approach, while not perfectly balanced, ensures adequate longitudinal continuity, mitigating potential biases associated with incomplete sample tracking and establishing a robust foundation for rigorous empirical estimation.

This study carefully links CHFS data with regional information to determine household broadband access based on coverage under the Broadband China policy. The matching is conducted at the city level to locate households situated in areas affected by the policy rollout. This produces a robust dataset for the DMLDiD analysis. The subsequent sections of this paper will provide a detailed exposition of the methodology, including the matching techniques and the definition and construction of variables. This comprehensive approach enables a nuanced exploration of how broadband access, as outlined in the Broadband China policy, impacts household decisions regarding risky assets investment.

DMLDiD Strategy

The DMLDiD estimator builds on the framework established by Abadie (2005), which defines the Average Treatment Effect on the Treated (ATT) under two key assumptions related to repeated outcomes. The first, referred to as the conditional parallel trends assumption, states that conditional on a set of covariates $X_{i}$ , the expected outcomes of treated and control groups would have followed similar paths had the treatment not occurred. This relationship is formally represented as:

E (Y_{1} (0) - Y_{0} (0) | X, D = 1) = E (Y_{1} (0) - Y_{0} (0) | X, D = 0)

(1)

The second assumption concerns the common support condition, which requires that the conditional probability of treatment for treated units lies within the range of propensity scores observed among control units. This overlap condition ensures sufficient comparability between the two groups. Under this assumption, the ATT can be identified as follows:

θ_{0} = E (\frac{Y_{1} - Y_{0}}{P (D = 1)} \cdot \frac{D - P (D = 1 | X)}{1 - P (D = 1 | X)})

(2)

The ATT can be estimated using:

\hat{θ} = \frac{1}{N} \sum_{i = 1}^{N} \frac{Y_{i 1} - Y_{i 0}}{\hat{p}} \cdot \frac{D_{i} - \hat{g} (X_{i})}{1 - \hat{g} (X_{i})}

(3)

where $\hat{p}$ is the estimator of $p_{0} = P (D_{i} = 1)$ , and $\hat{g} (X_{i})$ denotes the estimator of the propensity score $g_{0} (X_{i}) = P (D_{i} = 1 | X_{i})$ .

Abadie’s semi-parametric DID estimator achieves $\sqrt{N}$ -consistency and asymptotic normality when traditional nonparametric techniques are used to estimate the propensity score. However, when $\hat{g} (X_{i})$ is obtained via a machine learning (ML) algorithm, the estimator $\hat{θ}$ may lose this $\sqrt{N}$ -consistency due to the score function in Equation 2 having a non-zero directional derivative with respect to the propensity score. Moreover, ML estimators often converge at a rate slower than $N^{- \frac{1}{2}}$ , primarily due to regularization bias (Chang, 2020). To overcome these issues, a DMLDiD estimator has been developed based on the cross-fitting procedure and Neyman-orthogonal score proposed by Chernozhukov et al. (2018):

φ = \frac{Y_{1} - Y_{0}}{P (D = 1)} \cdot \frac{D - P (D = 1 | X)}{1 - P (D = 1 | X)} - θ_{0} - c

(4)

where $c$ refers to an adjustment term. In this approach, using Neyman-orthogonal scores ensures consistent estimation of the parameter of interest, whereas utilizing cross-fitting eliminates the bias due to overfitting (for more details, refer to Chernozhukov et al. (2018)).

Specifically, the procedures for using DMLDiD to estimate the ATT can be stated as follows (Chang, 2020; Jia, 2025; Sant’Anna & Zhao, 2020; Zhang et al., 2022):

Step 1: Randomly divide the full sample $S$ into $K$ random folds to facilitate cross-fitting.

Step 2: Using the auxiliary subset $S_{k}^{c}$ , estimate the nuisance components $p_{0}$ , $g_{0}$ and $l_{0}$ , where $l_{0} = E (Y_{1} - Y_{0} | X, D = 0)$ . In particular, the LightGBM algorithm is employed to estimate $l_{0}$ , as it is well-suited for capturing complex nonlinearities and handling variable heterogeneity in this context.

Step 3: Apply the remaining observations in fold $S_{k}$ to compute the intermediate, debiased estimator of the ATT:

{\tilde{θ}}_{k} = \frac{1}{n} \sum_{i \in S_{k}} \frac{D_{i} - {\hat{g}}_{k} (X_{i})}{{\hat{p}}_{k} (1 - {\hat{g}}_{k} (X_{i})} \cdot (Y_{i 1} - Y_{i 0} - {\hat{l}}_{k} (X_{i}))

(5)

Step 4: Combine the estimators from all $K$ folds to obtain the final DMLDiD estimator, defined as ${\tilde{θ}}_{DML} = \frac{1}{K} \sum_{k = 1}^{K} {\tilde{θ}}_{k}$

This estimator is shown to be $\sqrt{N}$ -consistent and asymptotically normal, provided that the employed machine learning methods exhibit convergence rates faster than $N^{- \frac{1}{4}}$ . As a result, it enables consistent estimation of the asymptotic variance of ${\tilde{θ}}_{DML}$ , supporting valid statistical inference.

This study employs the DMLDiD method to estimate the causal effect of broadband access on household investment in risky assets while addressing potential endogeneity and nonlinear relationships. The implementation adheres to a structured four-step process to ensure robustness.

First, the sample is randomly partitioned into five folds (with robustness checks using 10-folds) to facilitate cross-fitting, preventing overfitting and enhancing estimation accuracy. Second, propensity scores and conditional outcome functions are estimated using LightGBM, enabling the model to capture complex relationships between broadband access, household characteristics, and investment behavior. These nuisance parameters are then utilized to construct the debiased ATT estimator, ensuring that model specification errors do not bias the estimated treatment effect. Finally, estimates from all folds are aggregated to compute the final DMLDiD estimate, leveraging asymptotic normality and $\sqrt{N}$ -consistency for valid statistical inference.

This approach ensures that broadband’s effect on risky assets investment is estimated without strong parametric assumptions, reducing concerns regarding omitted variable bias and facilitating a more credible causal interpretation.

Summary Statistics

Table 1 outlines the key variables used to examine the effects of broadband internet, treated as an exogenous intervention resulting from the implementation of the Broadband China policy. Specifically, broadband access is defined as a binary variable equal to one if the household resides in a region covered by the policy, and 0 otherwise. This regional-level definition follows a widely adopted empirical strategy in the policy evaluation literature, which leverages government-led, geographically targeted infrastructure policies as quasi-natural experiments within Difference-in-Differences frameworks. While this measure does not directly capture individual household broadband adoption or usage, it reflects exogenous variation in broadband availability that is plausibly independent of household-specific financial behavior. Using regional policy coverage as a proxy allows the study to mitigate endogeneity concerns associated with voluntary broadband subscription decisions, which may be correlated with unobservable household characteristics such as financial literacy or investment propensity.

Table 1.

Variable Definitions for Broadband Impact Analysis.

Variables	Definitions	Mean	SD
Explanatory variable
Broadband	Equals 1 if the household is located in a region covered by the Broadband China policy; equals 0 otherwise.	0.312	0.463
Dependent variables
RiskInvest	Equals 1 if the household participates in high-risk financial markets (e.g., stocks, funds, corporate bonds, or gold); 0 if not.	0.091	0.288
RiskVal	Natural logarithm of the household’s total holdings in risky financial assets.	0.003	0.022
RiskShare	Proportion of risky assets relative to the household’s total asset portfolio.	0.003	0.026
Household characteristics
Gender	Equals 1 if the household head is male; equals 0 if female.	0.789	0.407
Age	Age of the household head, measured in years.	54.349	13.795
AgeSq	Squared term of the household head’s age, used to capture nonlinear effects.	3144.153	1528.456
EduYears	Number of completed years of formal education by the household head, based on degree equivalents.	8.894	4.094
Married	Equals 1 if the household head is married; 0 if not.	0.866	0.340
FinLit	Equals 1 if the respondent shows high interest in financial and economic topics; equals 0 otherwise.	0.070	0.255
RiskAverse	Equals 1 if the household prefers low or no-risk investment options; equals 0 otherwise.	0.940	0.237
Rural	Equals 1 if the household is situated in a rural area; 0 if urban.	0.560	0.496
LaborNum	Number of household members actively participating in the labor force.	1.708	1.249
ChildNum	Number of children under the age of 16 within the household.	0.715	0.896
HHSize	Total number of individuals living in the household.	2.966	1.750
TotAssets	Logarithmic transformation of the household’s total assets.	12.482	1.732
Income	Standardized value of the household’s total annual income.	0.000	1.000
SocNet	Log-transformed total of social gift income and gift-related household expenditures.	6.596	3.339
CreditRej	Equals 1 if the household experienced a loan rejection by a financial institution; 0 if not.	0.044	0.206

Importantly, regional broadband access can influence household investment behavior through multiple channels. Increased broadband infrastructure in the area typically improves service quality, reduces access costs, and facilitates market entry by internet service providers—factors that collectively lower the threshold for household-level adoption. Even for non-subscribing households, regional broadband expansion can promote broader digital and financial ecosystem development, such as enhanced access to financial information, peer learning, and digital financial services. Thus, regional broadband policy coverage, although not a direct measure of utilization, serves as a meaningful and policy-relevant proxy for household exposure to improved digital infrastructure and its associated behavioral incentives.

This study defines risky assets as investments in stocks, funds, corporate bonds, and gold, in accordance with prior literature that utilize the CHFS dataset (Gu & Zhu, 2024; Liao et al., 2017; Yang et al., 2024; Zhu & Xiao, 2022). This definition adheres to widely accepted empirical approaches for assessing household participation in high-risk financial markets. Notably, because the CHFS questionnaire excludes specific survey items associated with short-term savings as investment vehicles, this study does not classify short-term savings as investments. By embracing this definition, the study conforms to established research on household financial decision-making, thereby ensuring comparability with prior findings regarding risky asset holdings in China.

Table 1 shows that approximately 9.1% of households in the sample engage in risky assets investment, a proportion that aligns closely with existing studies on household investment behavior in China. This study defines Risky Assets Investment as a binary variable that indicates whether a household participates in risky investments, thereby capturing various aspects of risky asset engagement. Risky Assets Value uses the logarithm of total risky asset holdings to measure the magnitude of such investments, assigning zero value to non-investing households. Similarly, the Risky Assets Ratio, which represents the share of risky assets within a household’s total financial portfolio, is set to zero for non-investors. This methodological approach ensures comprehensive inclusion of all households, effectively distinguishing between market participants and non-participants while accurately capturing both the extensive and intensive margins of household risky assets investment.

Additionally, several control variables, including household demographics, financial literacy, risk attitudes, and socio-economic status, are incorporated to mitigate potential confounding factors that could otherwise bias the estimation. Household demographics, such as age, education level, and family size, capture fundamental characteristics that may influence financial decision-making. Financial literacy is essential for households to effectively analyze investment-related information, whereas risk attitudes influence their willingness to engage in risky financial activities. Socio-economic status, including household income and employment conditions, contributes to variations in financial resources and market accessibility. By adopting this comprehensive approach, the analysis effectively isolates the impact of broadband internet access on household investment behavior. This ensures that the estimated effects accurately reflect genuine changes in financial decision-making following the implementation of the Broadband China policy.

Research Results

Baseline Results

Table 2 presents the empirical results derived from the DMLDiD analysis, specifically assessing the impact of broadband access on households’ investments in high-risk assets. The estimations were progressively refined by incorporating varying degrees of complexity in the control variables, ranging from basic household characteristics to quadratic and cubic forms, to comprehensively capture potential nonlinear relationships.

Table 2.

Impact of Broadband Access on Household Risky Assets Investment.

Variables	(1)	(2)	(3)	(4)
Variables	RiskInvest
Broadband	0.0223*** (0.0027)	0.0501*** (0.0026)	0.0494*** (0.0026)	0.0494*** (0.0026)
Control variables		YES	YES	YES
Quadratic control variables			YES	YES
Cubic control variables				YES
Observations	77,337	77,337	77,337	77,337

Note. This table presents the DMLDiD estimation results on the impact of broadband access on household risky asset investment. All specifications control for household fixed effects, year fixed effects, and county-specific linear time trends. The LightGBM algorithm is employed for estimating nuisance parameters. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

Column (1) provides baseline estimates with household and year fixed effects, excluding supplementary household-level controls. The estimated broadband coefficient is 0.0223, statistically significant at the 1% level, indicating that broadband policy implementation significantly increases households’ likelihood of engaging in risky assets investment by approximately 2.23 percentage points relative to households in non-covered regions. This preliminary result underscores the influence of broadband access on household investment decisions.

In Column (2), the incorporation of household control variables—such as demographic characteristics, household income, education, and employment status—enhances the explanatory power and refines the estimation. The broadband coefficient remains highly significant (p < .01) and rises to 0.0501, indicating that when relevant household-level heterogeneity is considered, broadband’s influence on risky assets investment is further elucidated. Incorporating quadratic terms (Column 3) and cubic terms (Column 4) of these controls addresses potential nonlinearities and interactions among control variables, thereby ensuring robustness. Consequently, in the most comprehensive specification (Column 4), the broadband coefficient remains robust and significantly positive at 0.0494, indicating that households in broadband-covered regions are approximately 4.94 percentage points more inclined to invest in risky financial assets compared to households without broadband access. This stable coefficient across increasingly complex models underscores the robustness and reliability of our empirical findings.

Overall, the consistently significant positive effect of broadband access across various model specifications robustly supports our main hypothesis (H1), confirming that the expansion of broadband infrastructure significantly increases households’ participation in risky asset markets. These results align closely with theoretical predictions, indicating that broadband facilitates enhanced access to financial information, diminishes entry barriers to financial markets, and ultimately empowers households to engage confidently in higher-risk investments. Thus, broadband expansion may substantially alter household financial behavior, promoting financial market participation and diversification at the household level, particularly in rapidly digitizing contexts such as China.

Parallel Trends Test

The visual evidence presented in Figure 1 robustly substantiates the parallel trends assumption and reinforces the causal interpretation that broadband internet access significantly affects households’ investment decisions, leading to increased participation in risky financial assets.

Figure 1.

Parallel trends test for risky assets investment before and after broadband implementation.

During the period preceding the implementation of broadband—denoted as pre3 to pre1—the estimated coefficients remain near zero and exhibit no statistical significance. This lack of significant deviation from zero before the treatment indicates that any observed alterations in the outcome variable are random fluctuations rather than trends. The confidence intervals for these estimates include the zero point, reinforcing the conclusion that no pre-treatment differences in the investment behavior were evident between the treated and control groups.

The introduction of broadband, marked by post0, demonstrates a notable increase in the estimated coefficients, reflecting the treatment’s immediate effect. This shift is crucial because it signifies the divergence in trends that DMLDiD aims to capture. The years following the introduction of broadband, labeled post1 and post2, exhibit positive and statistically significant coefficients, with confidence intervals that do not intersect with zero. This positive trend indicates that the treatment group’s likelihood of investing in risky assets has risen compared to the control group, implying a causal relationship between broadband access and investment behavior.

The consistency of the post-treatment effect, coupled with the validation of the parallel trend’s assumption, underscores the robustness of the DMLDiD design in this context. The graphical evidence strongly supports the assertion that enhanced broadband connectivity has significantly influenced households’ financial decisions regarding riskier asset investments.

Robustness Checks

Table 3 presents a series of robustness checks using alternative dependent variables and methodological approaches to validate the reliability and consistency of the main empirical findings. Columns (1) and (2) show alternative measures of households’ participation in risky assets markets, presenting further dimensions to evaluate broadband’s impact on investment decisions beyond the mere binary decision to invest. Column (3) conducts a placebo test to assess the validity of the causal inference, while Column (4) examines sensitivity to a key methodological parameter in the DMLDiD analysis.

Table 3.

Robustness Checks for the Impact of Broadband Access on Household Risky Assets Investment.

	(1)	(2)	(3)	(4)	(5)
Variables	RiskVal	RiskShare	RiskInvest	RiskInvest	FinVal
Broadband	0.4045*** (0.0252)	0.0016*** (0.0002)	0.0010 (0.0019)	0.0496*** (0.0026)	0.5187*** (0.0312)
Observations	77,337	77,337	77,337	77,337	77,337

Note. This table reports robustness checks of the DMLDiD estimation on broadband access and household risky asset investment. Columns (1) and (2) examine alternative dependent variables related to risky asset value and allocation. Column (3) presents a placebo test, and Column (4) adjusts the fold parameter used in model training. Column (5) uses total financial assets value to capture the broader effect of broadband on general investment participation. All specifications include control variables (and their quadratic and cubic terms), household fixed effects, year fixed effects, and county-specific linear time trends. The LightGBM algorithm is used to estimate nuisance components. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

In Column (1), the dependent variable represents the aggregate value of households’ investments in high-risk assets. The broadband coefficient is significantly positive (0.4045) at the 1% level, indicating that broadband access increases both the likelihood of investing and the overall monetary value households invest in risky assets. Economically, this implies that broadband connectivity enhances both market participation and financial engagement intensity. Column (2) further examines this relationship by analyzing the ratio of risky assets in the household’s overall financial portfolio. The broadband coefficient remains significantly positive (0.0016, p < .01), confirming that broadband access prompts households to invest a greater proportion of their total assets to riskier financial products. Collectively, these results reinforce the initial findings, offering substantial evidence that broadband significantly enhances both the extensive margin (participation) and the intensive margin (investment magnitude and portfolio allocation) of risky assets investment.

To further ascertain causal validity, Column (3) conducts a placebo test in which broadband treatment assignment is randomly distributed among households, deliberately disrupting the true causal relationship. In this placebo scenario, the broadband coefficient is minimal (0.0010) and statistically insignificant, precisely as anticipated if the actual broadband effect were genuine. This placebo test’s absence of statistical significance indicates that the main findings are not the product of random fluctuations or chance assignments, thus reinforcing confidence in the initial causal interpretation.

Column (4) addresses the sensitivity of methodological choices within the DMLDiD estimation procedure. Specifically, the fold parameter for cross-validation, initially set at 5 in baseline analyses, is increased to 10-folds to evaluate the extent to which the results are contingent upon this analytical decision. Despite this methodological adjustment, the broadband coefficient remains positive (0.0496) and statistically significant at the 1% level, closely aligning with the initial baseline estimate. This consistency confirms that the primary findings are robust and not sensitive to variations in the cross-validation strategies employed during the DMLDiD modeling.

Column (5) broadens the scope of analysis by examining the effect of broadband access on total household financial assets value. This specification captures the broader financial engagement of households, reflecting whether broadband exposure encourages households to participate in investment activities more generally, beyond the domain of risky assets. The broadband coefficient is significantly positive (0.5187, p < .01), indicating that broadband access is associated with an increase in total asset holdings. This result suggests that broadband may play a role in facilitating initial entry into financial markets, potentially converting non-investing households into active investors. Such an expansion in participation provides context for the observed increases in risky asset engagement, as greater overall investment activity can shift both the average and the distributional profile of household portfolios.

Overall, these comprehensive robustness checks reinforce the credibility and stability of the central conclusion: broadband internet access significantly and positively influences household financial behavior. In addition to strengthening both the value and proportion of risky assets held, broadband access appears to encourage broader investment participation, as reflected in increased total asset holdings. Together, the findings suggest a two-step dynamic: broadband facilitates household entry into financial markets and subsequently shapes their investment composition, particularly in favor of riskier financial products.

To address the large number of zero observations in the outcome variables—particularly among households that do not invest in risky assets—we employ Tobit models as a robustness check. Unlike linear models, Tobit estimation accounts for the potential censoring of the dependent variable at zero, allowing for more appropriate handling of cases where non-investment may reflect a threshold or corner solution rather than continuous variation. While our main estimation strategy treats zero values as valid economic outcomes and relies on advanced machine learning techniques to improve causal identification, the Tobit model serves as a useful alternative specification to verify the consistency of the estimated effects.

Table 4 presents the results of Tobit regressions examining the impact of broadband access on risky asset holdings. Across all specifications, the coefficient on the broadband variable remains positive and highly significant at the 1% level. In Columns (1) and (2), broadband access is associated with substantial increases in the expected value of risky asset holdings, with estimates of 0.6560 and 0.4395, respectively. Columns (3) and (4) reveal similar effects on the proportion of risky assets in household portfolios. These consistent results reinforce the robustness of the main findings, indicating that the observed effects are not driven by model specification or the distributional characteristics of the dependent variables. Overall, the Tobit-based results support the conclusion that broadband access plays a significant role in shaping household investment in riskier financial instruments.

Table 4.

Robustness Checks Using Tobit Model.

Variables	(1)	(2)	(3)	(4)
Variables	RiskVal		RiskShare
Broadband	0.6560*** (0.0202)	0.4395*** (00199)	0.0029*** (0.0002)	0.0018*** (0.0002)
Control variables		YES		YES
Observations	77,337	77,337	77,337	77,337

Note. This table reports robustness checks using Tobit models to account for the censored nature of the dependent variables, which include a large number of zero observations. Columns (1) and (2) use the level of risky asset value as the dependent variable under alternative specifications, while Columns (3) and (4) use the risky asset ratio. All specifications control for household fixed effects, year fixed effects, and county-specific linear time trends. Tobit models are employed throughout. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

To enhance the verification of the robustness and consistency of our baseline results, Table 5 presents an analysis of broadband’s impact on household risky assets investment through various ML algorithms. Employing diverse ML approaches ensures that the estimated broadband effects are not confined to a specific algorithm but rather reflect genuine underlying relationships within the data. Specifically, we employ four established algorithms—Lasso Regression (Column 1), Random Forest (Column 2), Decision Tree (Column 3), and Logistic Regression (Column 4)—each addressing data characteristics, non-linearities, and interactions in distinct manners.

Table 5.

Robustness Checks Using Alternative Machine Learning Algorithms.

	(1)	(2)	(3)	(4)
Variables	RiskInvest
Broadband	0.0146*** (0.0023)	0.0621*** (0.0039)	0.0787*** (0.0173)	0.0943*** (0.0026)
Algorithm type	Lasso	Random Forest	Decision Tree	Logistic Regression
Observations	77,337	77,337	77,337	77,337

Note. This table reports robustness checks of the DMLDiD estimation using alternative machine learning algorithms. Columns (1) to (4) apply Lasso Regression, Random Forest, Decision Tree, and Logistic Regression, respectively. All specifications include control variables along with their quadratic and cubic terms, household fixed effects, year fixed effects, and county-specific linear time trends. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

In all four alternative models displayed in Table 4, broadband consistently exhibits a positive and statistically significant effect on households’ investments in risky assets. The coefficients vary moderately owing to the distinct assumptions and data complexity management of each algorithm. Specifically, Lasso Regression, which is known for imposing regularization to address multicollinearity and feature selection, yields a coefficient of 0.0146 (p < .01). This coefficient indicates broadband’s significant positive effect, albeit relatively modest in magnitude. Random Forest and Decision Tree algorithms, which are recognized for their efficacy in capturing complex interactions and nonlinear relationships, generate higher broadband coefficients of 0.0621 and 0.0787, respectively (both significant at the 1% level). This increase in magnitude underscores the potential for broadband’s effect on investment behavior to be amplified by interactions among household characteristics or nonlinear patterns that simpler models may not fully capture.

Logistic Regression, which models the probability of binary outcomes, yields the highest broadband coefficient of .0943, which is also statistically significant at the 1% level. This result highlights that when explicitly modeling household investment as a binary decision, broadband access significantly enhances the propensity to engage in risky asset markets, potentially indicating broadband’s role in reducing informational and transaction barriers to financial market participation.

Comparing these findings with the baseline DMLDiD estimates derived from LightGBM (Table 2 coefficients ranging from 0.0223 to 0.0501) reveals a consistent significance and direction across various ML algorithms, thereby affirming the robustness and validity of our core conclusions. Despite minor fluctuations in coefficient magnitude resulting from the unique methodological characteristics of each ML algorithm, the primary conclusion remains unequivocal: broadband internet access significantly enhances the likelihood of households investing in risky assets.

Possible Mechanisms

To further explore how broadband internet access influences households’ decisions to invest in risky assets, Table 6 systematically examines three potential channels: household income, social networks, and credit constraints. This mechanistic analysis validates the robustness of the observed effects and offers critical insights into the pathways through which broadband access translates into altered investment behaviors.

Table 6.

Mechanisms Underlying the Impact of Broadband on Household Risky Assets Investment.

Variables	(1)	(2)	(3)
Variables	Income	SocNet	CreditRej
Broadband	0.1887*** (0.0099)	0.0828*** (0.0302)	−0.0067*** (0.0022)
Observations	77,337	73,900	48,154

Note. This table examines the mechanisms through which broadband affects household risky asset investment using the DMLDiD approach. Columns (1) to (3) analyze the effects on household income, social networks, and credit constraints, respectively. All specifications include control variables (along with their quadratic and cubic terms), household fixed effects, year fixed effects, and county-specific linear time trends. The LightGBM algorithm is used to estimate nuisance parameters. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

Column (1) specifically examines the influence of household income. The significantly positive coefficient (.1887, p < .01) indicates that broadband access considerably increase household income levels. Economically, this implies that broadband connectivity may facilitate households in accessing superior employment opportunities, enhancing productivity, or benefiting from online economic activities. An increase in household income directly enhances financial capacity and diminishes the perceived risks associated with engaging in higher-risk financial markets (Bricker et al., 2021). Thus, enhanced broadband infrastructure can facilitate risky assets investment partly by increasing household financial resources and risk-bearing capacity.

Column (2) examines the impact of broadband on households’ social networks, yielding a significant and positive coefficient (.0828, p < .01). This result indicates that broadband enhances both the extent and quality social interactions among households, potentially facilitating increased information exchange, financial advice sharing, and peer influence on investment strategies (Lei & Ramos Salazar, 2022). Enhanced social networks serve as vital conduits through which broadband access reduces information asymmetry, encourages peer learning, and increases the likelihood of households participating in risky asset markets.

In Column (3), the analysis investigates whether broadband mitigates households’ credit constraints. The estimated coefficient for broadband is significantly negative (−0.0067, p < .01), indicating a decrease in household credit constraints following broadband expansion. This adverse correlation indicates that broadband enables households to access online financial platforms, digital lending services, and other financial products more efficiently, thereby alleviating previously existing financial constraints. Mitigating credit constraints can substantially empower financially restricted households, enabling them to engage more actively in higher-risk investments due to enhanced liquidity and credit availability (Gomes et al., 2021; Peng et al., 2021).

Overall, the findings Table 6 offer robust empirical evidence for various mechanisms that contribute to broadband’s positive influence on household risky asset investments. Broadband access significantly increases household income, strengthens social networks, and alleviates credit constraints, thereby enhancing households’ financial capacity, informational advantages, and market accessibility. These multifaceted mechanisms collectively illuminate how broadband internet influences financial decision-making, offering enhanced understanding of the relationship between technological infrastructure and household financial behaviors. Furthermore, the results support Hypotheses 2, 3, and 4, indicating that broadband access increases household income (H2), facilitates the expansion of social networks that enhance investment decision-making (H3), and reduces credit constraints, thereby fostering greater participation in risky asset investments (H4).

Heterogeneity Analysis

Table 7 presents subgroup analyses to examine the varying impacts of broadband access among diverse groups, categorized by household location (urban vs. rural) and geographical region (Eastern vs. Western and Central China). Exploring these disparities is crucial for understanding how diverse socio-economic and regional contexts influence the effectiveness and implications of broadband expansion.

Table 7.

Regional Heterogeneity in the Impact of Broadband on Household Risky Assets Investment.

Variables	(1)	(2)	(3)	(4)
	Urban	Rural	Eastern Region	Western and Central Regions
	RiskInvest
Broadband	0.0773*** (0.0044)	0.0194*** (0.0019)	0.0514*** (0.0045)	0.0279*** (0.0030)
Observations	33,974	43,363	35,027	42,310

Note. This table examines the regional heterogeneity in the impact of broadband on household risky asset investment using the DMLDiD approach. Columns (1) and (2) compare urban and rural households, while Columns (3) and (4) analyze differences between the Eastern region and the Western and Central regions. All specifications include control variables (along with their quadratic and cubic terms), household fixed effects, year fixed effects, and county-specific linear time trends. The LightGBM algorithm is used to estimate nuisance parameters. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

Columns (1) and (2) explicitly examine the differentiation between urban and rural households. In urban households (Column 1), the broadband coefficient is significantly high (.0773, p < .01), indicating a robust and positive broadband effect on risky assets investment in urban areas. In contrast, the broadband coefficient for rural households (Column 2) is positive and statistically significant, yet substantially smaller (.0194, p < .01). This apparent urban-rural disparity reflects variations in financial infrastructure, digital literacy, and financial market accessibility. Urban households generally possess superior financial literacy, a broader array of accessible investment opportunities, and more advanced financial institutions. Consequently, broadband in urban settings amplifies pre-existing advantages in accessing financial information and investment platforms, resulting in more pronounced investment responses.

The analysis in Columns (3) and (4) further investigates regional disparities between economically developed Eastern regions and comparatively underdeveloped Western and Central regions. The broadband coefficient for households in the Eastern region (.0514, p < .01, Column 3) significantly surpasses that of households in the Western and Central regions (.0279, p < .01, Column 4). This regional variation aligns with the expectation that more economically advanced regions exhibit enhanced financial market infrastructure, increased digital penetration, elevated financial sophistication, and stronger integration into digital economic activities. These characteristics likely empower households in the Eastern region to more effectively leverage broadband connectivity to engage in high-risk financial markets compared to their counterparts in less developed areas.

Overall, the heterogeneity analyses in Table 7 highlight that broadband’s impact on risky assets investment varies significantly based on socio-economic context and geographic location. The pronounced effects observed in urban and economically advanced Eastern regions underscore how pre-existing economic conditions, financial market access, and digital infrastructure significantly mediate broadband’s influence on household investment behavior. These findings emphasize the necessity of considering urban-rural divides and regional economic disparities when evaluating the broader economic consequences of broadband infrastructure expansion.

Table 8 examines the differential impact of broadband access on risky assets investment decisions across demographic groups, particularly distinguishing households by educational attainment and age to further explore individual-level heterogeneity. Investigating these individual differences is essential for comprehending how demographic characteristics influence households’ responses to broadband availability and for guiding targeted policy interventions.

Table 8.

Individual Heterogeneity in the Impact of Broadband on Household Risky Assets Investment.

Variables	(1)	(2)	(3)	(4)
	Higher education	Lower education	Younger	Elder
	RiskInvest
Broadband	0.1018*** (0.0056)	0.0186*** (0.0022)	0.0801*** (0.0045)	0.0430*** (0.0028)
Observations	24,485	52,852	19,193	58,144

Note. This table examines individual heterogeneity in the impact of broadband on household risky asset investment using the DMLDiD approach. Columns (1) and (2) compare households with higher and lower education levels, while Columns (3) and (4) distinguish between younger and elder households. All specifications include control variables (along with their quadratic and cubic terms), household fixed effects, year fixed effects, and county-specific linear time trends. The LightGBM algorithm is used to estimate nuisance parameters. Standard errors are in parentheses.

***

p < .01, **p < .05, *p < .1.

Columns (1) and (2) juxtapose the broadband effect on households with higher education (≥12 years of schooling) against those with lower education (<12 years). The broadband coefficient is significantly higher for households with higher education (0.1018, p < .01, Column 1) than for those with lower education (.0186, p < .01, Column 2). This significant disparity likely arises from higher-educated households’ superior financial literacy, digital competencies, and greater capacity to leverage broadband-enabled access to financial information and digital investment platforms. Such households can more effectively translate enhanced connectivity into active participation in riskier, more intricate financial markets. Conversely, while still benefiting significantly, lower-educated households seem less capable of fully leveraging broadband-driven opportunities due to comparatively limited financial and digital literacy.

Columns (3) and (4) assess the broadband impact across age groups, comparing younger households (age ≤ 45) with older households (age > 45). Younger households exhibit a relatively higher broadband coefficient (.0801, p < .01, Column 3) than that of older households (.0430, p < .01, Column 4). This age-based variation is primarily attributed to technological adoption, digital familiarity, and risk tolerance differences. Generally, younger individuals exhibit greater technological proficiency, increased comfort with digital platforms, and a higher propensity for financial risk, rendering them more responsive to broadband-induced investment opportunities. While still significantly influenced, older households may exhibit more cautious financial behavior and encounter greater obstacles in embracing digital technologies, resulting in a comparatively modest broadband impact.

The findings presented in Table 8 underscore the significant influence of individual demographic characteristics on households’ financial responses to broadband internet access. The pronounced differences based on education and age highlight the necessity of considering demographic heterogeneity in the design and deployment of broadband and financial market policies. Policymakers seeking to maximize the financial benefits of digital infrastructure must recognize these variations to enact complementary strategies, such as financial education programs and age-specific digital training initiatives, to ensure equitable and widespread access to the benefits of technological advancement.

Conclusion

In this study, we conducted an empirical analysis to determine the impact of increased broadband access on household investments in high-risk assets. We innovatively applied a DMLDiD approach, integrating advanced econometric techniques with ML algorithms to enhance causal inference and address confounding variables. The findings demonstrate that broadband access significantly elevates households’ engagement in risky assets investment. The study’s rigorous validation, including various robustness checks and subgroup analyses, reinforces the positive correlation between household financial market participation and broadband penetration. This pivotal relationship demonstrates the paradigm-shifting impact of digital infrastructure on investment behaviors and economic outcomes.

The research emphasizes substantial policy implications, particularly regarding government investment in broadband infrastructure. It recommends that policymakers contemplate broadband expansion to enhance households’ access to financial markets and diversify their investment portfolios. These findings indicate that prioritizing investments in digital infrastructure can aid policymakers in fostering a more inclusive financial system and enhancing economic democratization. Consequently, this can enable individuals to exploit a wider array of investment opportunities and enhance their financial well-being.

This study advances the literature by demonstrating that enhanced digital connectivity significantly and positively influences households’ financial decisions by linking broadband internet access to risky assets investment. The methodological perspective it presents to economic research is a novel application of the DMLDiD approach, which addresses potential confounding variables that were previously overlooked, thereby providing a more nuanced comprehension of the causal relationship.

The study’s limitations primarily stem from the nature and scope of the available data. First, the identification strategy relies on regional-level broadband policy coverage as a proxy for access, rather than direct household-level measures of broadband subscription or usage. While such policy-based exposure captures exogenous variation useful for causal inference, it may not fully reflect actual utilization of digital services at the household level. Consequently, the results should be interpreted as the effects of improved broadband availability rather than confirmed usage, which could lead to attenuation or heterogeneity in the estimated impacts. Second, the analysis is limited to survey waves from 2013 to 2017, which constrains the ability to capture long-term behavioral shifts or the delayed effects of digital infrastructure investment. Additionally, the focus on specific regions within China may limit generalizability to other institutional or economic contexts.

Future research should seek to incorporate more granular data on actual broadband adoption and internet usage behaviors within households. This would help distinguish the effects of mere access from those of active digital engagement. Moreover, extending the time horizon beyond the initial years of broadband policy implementation would allow for the assessment of sustained or evolving impacts on investment behavior. Further cross-regional or cross-country comparisons could also reveal how the influence of digital infrastructure varies depending on financial literacy, regulatory environments, or levels of digital inclusion across populations.

Footnotes

ORCID iD

Haowen Jia

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The China Household Finance Survey (CHFS) data utilized in this study is sourced from the Survey and Research Center for China Household Finance of the Southwestern University of Finance and Economics. To register and apply for data, researchers can log in to .

References

Abadie

(2005). Semiparametric difference-in-differences estimators. Review of Economic Studies, 72(1), 1–19. https://doi.org/10.1111/0034-6527.00321

Acolin

Goodman

Wachter

S. M.

(2019). Accessing homeownership with credit constraints. Housing Policy Debate, 29(1), 108–125. https://doi.org/10.1080/10511482.2018.1452042

Ahmad

Majeed

Khan

M. A.

Sohaib

Shehzad

(2021). Digital financial inclusion and economic growth: Provincial data analysis of China. China Economic Journal, 14(3), 291–310. https://doi.org/10.1080/17538963.2021.1882064

Alam

Imran

(2015). The digital divide and social inclusion among refugee migrants: A case in regional Australia. Information Technology and People, 28(2), 344–365. https://doi.org/10.1108/itp-04-2014-0083

Anderson

(2008). The social impact of broadband household Internet access. Information Communication & Society, 11(1), 5–24. https://doi.org/10.1080/13691180701858810

Bahia

Castells

Cruz

Masaki

Pedros

Pfutze

Rodriguez-Castelan

Winkler

Pfutze

(2020). The Welfare Effects of Mobile Broadband Internet: Evidence from Nigeria. World Bank.

Bauer

J. M.

(2018). The Internet and income inequality: Socio-economic challenges in a hyperconnected society. Telecommunications Policy, 42(4), 333–343. https://doi.org/10.1016/j.telpol.2017.05.009

Beck

Pamuk

Ramrattan

Uras

B. R.

(2018). Payment instruments, finance and development. Journal of Development Economics, 133, 162–186. https://doi.org/10.1016/j.jdeveco.2018.01.005

Bricker

Krimmel

Ramcharan

(2021). Signaling status: The impact of relative income on household consumption and financial decisions. Management Science, 67(4), 1993–2009. https://doi.org/10.1287/mnsc.2019.3577

10.

Chang

N.-C.

(2020). Double/debiased machine learning for difference-in-differences models. Econometrics Journal, 23(2), 177–191. https://doi.org/10.1093/ectj/utaa001

11.

Chen

Yoon

S. S.

(2022). Does technology innovation in finance alleviate financing constraints and reduce debt-financing costs? Evidence from China. Asia Pacific Business Review, 28(4), 467–492. https://doi.org/10.1080/13602381.2021.1874665

12.

Chen

Gong

Chu

C.-C.

Cao

(2018). Access to the Internet and access to finance: Theory and evidence. Sustainability, 10(7), 2534. https://doi.org/10.3390/su10072534

13.

Chernozhukov

Chetverikov

Demirer

Duflo

Hansen

Newey

Robins

(2018). Double/debiased machine learning for treatment and structural parameters. Econometrics Journal, 21(1), C1–C68. https://doi.org/10.1111/ectj.12097

14.

Dettling

L. J.

(2017). Broadband in the labor market: The impact of residential high-speed internet on married women’s labor force participation. ILR Review, 70(2), 451–482. https://doi.org/10.1177/0019793916644721

15.

Evans

W. N.

Perry

Factor

(2021). The impact of broadband access on social and economic integration: Results from the refugee mobile pilot. Journal of Refugee Studies, 34(1), 915–935. https://doi.org/10.1093/jrs/fez067

16.

Ford

G. S.

(2018). Is faster better? Quantifying the relationship between broadband speed and economic growth. Telecommunications Policy, 42(9), 766–777. https://doi.org/10.1016/j.telpol.2018.05.006

17.

Gallardo

Whitacre

Kumar

Upendram

(2021). Broadband metrics and job productivity: A look at county-level data. Annals of Regional Science, 66(1), 161–184. https://doi.org/10.1007/s00168-020-01015-0

18.

Geraci

Nardotto

Reggiani

Sabatini

(2022). Broadband Internet and social capital. Journal of Public Economics, 206, 104578. https://doi.org/10.1016/j.jpubeco.2021.104578

19.

Gomber

Koch

J.-A.

Siering

(2017). Digital Finance and FinTech: Current research and future research directions. Journal of Business and Economics, 87(5), 537–580. https://doi.org/10.1007/s11573-017-0852-x

20.

Gomes

Haliassos

Ramadorai

(2021). Household finance. Journal of Economic Literature, 59(3), 919–1000. https://doi.org/10.1257/jel.20201461

21.

Zhu

(2024). Homeownership and household risky asset holdings: Moderating role of social security and the mediating role of subjective well-being. Finance Research Letters, 66, 105664. https://doi.org/10.1016/j.frl.2024.105664

22.

Hambly

Rajabiun

(2021). Rural broadband: Gaps, maps and challenges. Telematics and Informatics, 60, 101565. https://doi.org/10.1016/j.tele.2021.101565

23.

Houngbonon

G. V.

Liang

(2021). Broadband Internet and income inequality. Review of Network Economics, 20(2), 55–99. https://doi.org/10.1515/rne-2020-0042

24.

Hvide

H. K.

Meling

T. G.

Mogstad

Vestad

O. L.

(2024). Broadband Internet and the stock market investments of individual investors. Journal of Finance, 79(3), 2163–2194. https://doi.org/10.1111/jofi.13335

25.

Ito

(2019). Digital China: A fourth industrial revolution with Chinese characteristics? Asia-Pacific Review, 26(2), 50–75. https://doi.org/10.1080/13439006.2019.1691836

26.

Jia

(2025). How broadband internet shapes household tourism decisions: A double/debiased machine learning-based difference-in-difference approach. Journal of Hospitality and Tourism Technology, 16(2), 326–345. https://doi.org/10.1108/jhtt-02-2024-0102

27.

Jiang

Murmann

J. P.

(2022). The rise of China’s digital economy: An overview. Management and Organization Review, 18(4), 790–802. https://doi.org/10.1017/mor.2022.32

28.

Jin

Zhang

(2023). Impact of fast internet access on employment: Evidence from a broadband expansion in China. China Economic Review, 81, 102038. https://doi.org/10.1016/j.chieco.2023.102038

29.

Jung

López-Bazo

(2020). On the regional impact of broadband on productivity: The case of Brazil. Telecommunications Policy, 44(1), 101826. https://doi.org/10.1016/j.telpol.2019.05.002

30.

Khanal

A. R.

Mishra

A. K.

(2016). Financial performance of small farm business households: The role of internet. China Agricultural Economic Review, 8(4), 553–571. https://doi.org/10.1108/caer-12-2014-0147

31.

Koutroumpis

(2019). The economic impact of broadband: Evidence from OECD countries. Technological Forecasting and Social Change, 148, 119719. https://doi.org/10.1016/j.techfore.2019.119719

32.

Lei

Ramos Salazar

(2022). Use of social networks in stock investment. International Journal of Bank Marketing, 40(1), 110–127. https://doi.org/10.1108/ijbm-04-2021-0158

33.

Liang

Guo

(2015). Social interaction, Internet access and stock market participation—An empirical study in China. Journal of Comparative Economics, 43(4), 883–901. https://doi.org/10.1016/j.jce.2015.02.003

34.

Liao

Xiao

J. J.

Zhang

Zhou

(2017). Financial literacy and risky asset holdings: Evidence from China. Accounting and Finance, 57(5), 1383–1415. https://doi.org/10.1111/acfi.12329

35.

Lin

Whinston

A. B.

Fan

(2015). Harnessing Internet finance with innovative cyber credit management. Financial Innovation, 1(1), 5. https://doi.org/10.1186/s40854-015-0004-7

36.

Masaki

Granguillhome Ochoa

Rodriguez-Castelan

(2020). Broadband Internet and household welfare in Senegal. World Bank.

37.

Nie

Zhang

Renwick

(2020). Impact of Internet use on economic well-being of rural households: Evidence from China. Review of Development Economics, 24(2), 503–523. https://doi.org/10.1111/rode.12645

38.

Mora-Rivera

García-Mora

(2021). Internet access and poverty reduction: Evidence from rural and urban Mexico. Telecommunications Policy, 45(2), 102076. https://doi.org/10.1016/j.telpol.2020.102076

39.

Ostrovsky-Berman

Litwin

(2019). Social network and financial risk tolerance among investors nearing and during retirement. Journal of Family and Economic Issues, 40(2), 237–249. https://doi.org/10.1007/s10834-018-9592-5

40.

Peng

Y. L.

Ren

H. J.

(2021). Do credit constraints affect households’ economic vulnerability? Empirical evidence from rural China. Journal of Integrative Agriculture, 20(9), 2552–2568. https://doi.org/10.1016/s2095-3119(20)63557-2

41.

Sant’Anna

P. H. C.

Zhao

(2020). Doubly robust difference-in-differences estimators. Journal of Econometrics, 219(1), 101–122. https://doi.org/10.1016/j.jeconom.2020.06.003

42.

Shi

Lim

T. S.

(2024). Household risky financial asset allocation in China: The influence of financial literacy dimensions. Journal of Infrastructure, Policy and Development, 8(8), 6786. https://doi.org/10.24294/jipd.v8i8.6786

43.

Shim

Shin

D.-H.

(2016). Analyzing China’s Fintech industry from the perspective of actor–network theory. Telecommunications Policy, 40(2-3), 168–181. https://doi.org/10.1016/j.telpol.2015.11.005

44.

Shin

S. H.

Kim

K. T.

Seay

(2020). Sources of information and portfolio allocation. Journal of Economic Psychology, 76, 102212. https://doi.org/10.1016/j.joep.2019.102212

45.

Sood

Pathak

Gupta

(2025). How do the determinants of investment decisions get prioritized? Peeking into the minds of investors. Kybernetes, 54(4), 2175–2203. https://doi.org/10.1108/k-04-2023-0662

46.

Stockinger

(2019). Broadband internet availability and establishments’ employment growth in Germany: Evidence from instrumental variables estimations. Journal for Labour Market Research, 53(1), 7. https://doi.org/10.1186/s12651-019-0257-0

47.

Thomas

Finn

(2018). Households’ valuation of new broadband networks: The impact of the perceived benefits of and access to e-services. Internet Research, 28(3), 675–695. https://doi.org/10.1108/intr-10-2016-0317

48.

Valentín-Sívico

Canfield

Low

S. A.

Gollnick

(2023). Evaluating the impact of broadband access and internet use in a small underserved rural community. Telecommunications Policy, 47(4), 102499. https://doi.org/10.1016/j.telpol.2023.102499

49.

Wan

Nie

Zhang

(2021). Does broadband infrastructure really affect consumption of rural households? – A quasi-natural experiment evidence from China. China Agricultural Economic Review, 13(4), 832–850. https://doi.org/10.1108/caer-12-2020-0303

50.

Weng

Wang

Heerink

van Den Berg

(2022). Credit constraints and rural households’ entrepreneurial performance in China. Emerging Markets Finance and Trade, 58(2), 570–583. https://doi.org/10.1080/1540496x.2020.1788538

51.

Tao

Zhang

Xie

Tai

(2022). Review of digital economy research in China: A framework analysis based on bibliometrics. Computational Intelligence and Neuroscience, 2022, 1–11. https://doi.org/10.1155/2022/2427034

52.

Yang

Wang

Liu

(2024). What affects the financial asset allocation of the elderly? From the perspective of financial literacy and risk attitude. Finance Research Letters, 63, 105327. https://doi.org/10.1016/j.frl.2024.105327

53.

Xiong

(2022). The impact of digital finance on household participation in risky financial markets: Evidence-based study from China. PLoS One, 17(4), e0265606. https://doi.org/10.1371/journal.pone.0265606

54.

Zareei

(2019). Network origins of portfolio risk. Journal of Banking & Finance, 109, 105663. https://doi.org/10.1016/j.jbankfin.2019.105663

55.

Zhang

Ren

(2022). Quantifying the social impacts of the London night tube with a double/debiased machine learning based difference-in-differences approach. Transportation Research Part A Policy and Practice, 163, 288–303. https://doi.org/10.1016/j.tra.2022.07.015

56.

Zhou

Vatsa

(2023). Impact of internet use on value and diversity of risky financial asset portfolios. Quarterly Review of Economics and Finance, 89, 188–193. https://doi.org/10.1016/j.qref.2023.03.012

57.

Zhu

Xiao

J. J.

(2022). Consumer financial education and risky financial asset holding in China. International Journal of Consumer Studies, 46(1), 56–74. https://doi.org/10.1111/ijcs.12643