How does digital financial inclusion affect the health of older adults? Evidence from China

Abstract

This study explores the impact of digital financial inclusion (DFI) on older adults’ health outcomes and the underlying mechanisms. Using data from the China Health and Retirement Longitudinal Study (CHARLS), we employ a two-way fixed effects model and an instrumental variable approach. The results indicate that advancing DFI significantly improves self-rated health and reduces functional impairment and depression in older adults. DFI primarily improves health outcomes by increasing the ability and willingness to pay for health investments. The positive effect of DFI on health is more pronounced among older adults in Midwestern provinces, with digital access, under 75 years old, and living with their children. These findings highlight the potential of DFI to promote healthy aging, and suggest that policy efforts should prioritize digital financial product innovation, strengthen payment capacity, and integrate health platforms to foster proactive health management. Targeted initiatives to improve digital access and literacy are also essential to ensure that vulnerable groups equitably share in the health benefits of the digital economy.

Highlights

• •This study demonstrates that digital financial inclusion (DFI) significantly enhances the physical and mental health of older adults.
• •DFI primarily improves the health of older adults by increasing their ability and willingness to pay for health investments.
• •The health benefits of DFI are more pronounced among older adults in Midwestern provinces, with digital access, under 75 years old, and living with their children.

1. Introduction

Population aging has emerged as a global phenomenon. China is experiencing rapid demographic aging at a rate surpassing many other Asian countries, as well as high-income countries in Europe and North America (Feng et al., 2020). By the end of 2024, China's population aged 60 and over had reached 310 million, accounting for 22% of the total population (National Bureau of Statistics of China, 2025). These older populations are showing an increase in the incidence and prevalence of chronic non-communicable diseases, functional disabilities (Chatterji et al., 2015), and mental health problems (Reynolds 3rd et al., 2022). These health risks impose a substantial burden on the sustainability of the national healthcare system. Therefore, maintaining and promoting the health of the aging population has become an imperative for China's public health governance.

In recent years, driven by modern information and communications technology, digital financial inclusion (DFI) has experienced rapid development in China (Ahmad et al., 2021). DFI refers to the delivery of formal financial services through digital technologies to populations traditionally excluded from the conventional banking system (Lauer & Lyman, 2015). Both DFI and traditional financial inclusion aim to expand access to financial services for underserved populations. However, DFI enhances breadth of coverage by overcoming geographical barriers, and improves depth of usage by reducing transaction costs and enabling convenient, real-time service delivery. The realization of these advantages depends on digital access and literacy (Ozili, 2018).

A growing body of the literature reveals that financial development has a profound influence on population health (Bhatta, 2013; Chireshe & Ocran, 2020). Recent cross-country studies have examined the effect of financial inclusion on health indicators, including life expectancy, infant mortality rates and the health index. These studies reveal positive health effects (Banerjee et al., 2023; Naveenan et al., 2024). Other forms of financial inclusion, such as access to health insurance, have also been shown to be important for improving self-reported health status and mental well-being (Finkelstein et al., 2012; Sommers et al., 2017).

Despite these population-level benefits, evidence on the health effects of financial inclusion among older adults remains inconclusive. While traditional financial inclusion, such as owning a bank account or maintaining financial engagement, has been found to benefit mental health, its association with physical health outcomes is not consistently supported (Aguila et al., 2016; Gyasi et al., 2019). DFI differs fundamentally from its traditional counterpart in its reliance on digital access and literacy. These are precisely the areas where older adults face significant disadvantages. Specifically, they encounter barriers to device access and internet connectivity due to income constraints and urban-rural disparities (Sen et al., 2022). In addition, they tend to have lower levels of digital literacy (Heinz et al., 2013; Kramer, 2014). Consequently, findings from traditional financial inclusion may have limited applicability to the digital context, which leaves the health effects of DFI on older adults underexplored.

Using panel data from the China Health and Retirement Longitudinal Study (CHARLS), this paper addresses this gap by examining the causal impact of DFI on the health of older adults and its mechanisms. This study makes several contributions to the existing literature. First, it expands the literature on the social determinants of health, focusing on DFI's effect on the health outcomes of older adults, including self-rated health, functional impairment, and mental well-being. Using nationally representative longitudinal data and rigorous empirical strategies, such as a two-way fixed effects model coupled with an instrumental variable approach, it provides robust causal inferences. Second, based on health capital models, it investigates how DFI influences older adults' health, offering practical insights for encouraging health investments and addressing public health challenges posed by rapid population aging. Third, it analyzes the heterogeneity of DFI's health effects across regions and individuals, capturing both the inclusive potential and the digital divide inherent in digital finance. These findings offer empirical evidence to support the development of targeted digital equity policies.

2. Theoretical analysis and research hypothesis

Beyond its role in expanding financial access, a growing body of literature has explored the broader socio-economic implications of DFI, particularly its impact on health outcomes. The literature shows that DFI has long-lasting implications for health (Konou, 2023). It can influence health outcomes through direct and indirect channels (Banerjee et al., 2023).

At the macro level, DFI serves as a driver for economic growth (Thaddeus et al., 2020), poverty reduction, and diminished income inequality (Suhrab et al., 2024; Wong et al., 2023). Extensive research confirms that economic development, income distribution, and poverty are crucial social determinants of health outcomes (Marmot, 2005). Higher economic development facilitates greater investment in public health, while lower poverty and inequality improve living conditions and access to healthcare, both of which are associated with better overall population health (Preston, 1975; Rodgers, 1979; Wagstaff, 2002).

At the household level, the impact of DFI on health is more direct. DFI enables households to mitigate financial risks associated with poor health through improved access to savings and credit (Ayayi et al., 2025; Xu et al., 2025). It also increases the availability of commercial insurance, which can protect households against catastrophic health-related expenses (Hou et al., 2024). This reduces the risk of foregoing necessary healthcare. By providing convenient and secure methods of paying for medical expenses and accessing telemedicine services (Naveenan et al., 2024), DFI addresses some healthcare access barriers, such as geographic distance and limited physical infrastructure. Collectively, these may contribute to improved physical health. Further, literature reveals that expanding access to financial services may improve financial stability, reduce stress, and positively impact mental health (Aguila et al., 2016; Gyasi et al., 2019; Ajefu et al., 2020).

Concurrently, older adults are becoming increasingly proficient in using digital financial services. A report highlights that the amount of mobile payments made by older adults in China increased 50-fold from the first quarter of 2017 to the second quarter of 2021 (Chinese Academy of Social Sciences, 2021). Supported by government initiatives to make financial apps more user-friendly for older people (Ministry of Industry and Information Technology of China, 2020), DFI is expected to exert a positive impact on their health. Therefore, we propose the following hypothesis:

H1

Digital financial inclusion has a positive influence on the health of older adults.

In the health demand model, Grossman (1972) viewed health as a durable capital stock that depreciates over time. Health investment is the primary means of offsetting depreciation and enhancing the stock. This investment decision is constrained by two factors: ability to pay and willingness to pay.

Regarding the ability to pay, older adults face stringent budget constraints, with their health investment levels highly correlated to financial status (Halliday et al., 2019). By overcoming geographical barriers, DFI achieves broader coverage. This expanded reach plays a more extensive role in enhancing household income (Hu et al., 2023). Additionally, DFI lowers transaction costs, allowing older adults to access a wider array of financial products, such as money market and bond funds (Li et al., 2020). Through digital platforms, older adults can manage their funds more efficiently (Yin, Gong, Guo, & Wu, 2019). This improved efficiency is associated with better asset allocation and contributes to higher capital returns. Higher income and asset levels give older adults a strong financial base to cover health-related expenditures. Studies show that being financially capable has a positive influence on mental and physical health (Wagstaff, 2002; Nussbaum, 2009). Therefore, we propose the following hypothesis:

H2

DFI positively affects older adults' health by promoting the ability of paying for health investments.

Regarding willingness to pay, we focus on health investment to mitigate disease risks, which is primarily realized through preventive health expenditures, rather than reactive health spending triggered by illness. According to consumption theory, financial development can help individuals with liquidity constraints smooth their consumption over time, thereby stimulating demand (Campbell & Mankiw, 1991). Building on this, DFI further influences individuals’ consumption habits and health awareness.

DFI can increase the frequency and value of transactions for health-related products and services by leveraging its lower transaction costs and payment convenience (Dong & Zang, 2024). Furthermore, digital financial platforms leverage their extensive user bases to establish integrated financial health ecosystems, providing health education and online consultation services (Iheanachor et al., 2023). These services help improve health literacy and encourage a shift from passive treatment to active prevention among older adults. This cognitive shift may increase their willingness to pay for preventive health investments, such as checkup packages, health supplements, and fitness services. These investments contribute to the accumulation of health stocks. Therefore, we propose the following hypothesis:

H3

DFI positively affects older adults' health by increasing the willingness to pay for health investments.

3. Research design

3.1. Data

The study uses data from the CHARLS, which provides high-quality data representing individuals aged 45 and older in China. CHARLS has conducted five nationwide longitudinal surveys in 2011, 2013, 2015, 2018, and 2020. We analyze the 2013, 2015, and 2018 waves,1 focusing on a sample of individuals aged 60 and over.

We employ the Peking University Digital Financial Inclusion Index to measure the degree of DFI. This index scientifically and accurately depicts the annual development levels of DFI in China. It is also frequently utilized in related research (Shen et al., 2025; Yin & Wang, 2025). We merged individual-level CHARLS data with the city-level index. After removing individuals with missing data, our sample comprised 24,916 observations.

3.2. Variables

3.2.1. Dependent variables

The dependent variable is the health status of older adults. We measure health across three dimensions: self-rated health, functional impairment, and mental health. Self-rated health is a subjective measure of overall health, assessed by asking respondents to rate their health on a scale ranging from “very good” to “very poor.” The options are presented as values ranging from 5 to 1, with 5 indicating “very good” and 1 indicating “very poor.”

Functional impairment, defined as a decline in the ability to perform daily activities, serves as a crucial indicator of health status in older adults (Stuck et al., 1999). We use impairment in activities of daily living (ADL) and impairment in instrumental ADL (IADL) to assess functional status. For each of the six basic ADLs and six IADLs, respondents who answer “no difficulty” receive a score of 0, while those who answer “difficult but manageable,” “difficult with assistance,” or “unable to perform” receive a score of 1. ADL and IADL impairment scores are summed separately, with both variables ranging from 0 to 6. Higher values denote worse physical function.

Mental health is measured using the CES-D10 depression scale. The scale contains 10 questions about the respondent's emotional and physical symptoms over the past week. The total score ranges from 0 to 30. Higher values indicate greater levels of depression.

3.2.2. Independent variable

The independent variable is digital financial inclusion (DFI), which is measured by the Peking University Digital Financial Inclusion Index. The index provides annual data from 2011 to 2021 for 31 provinces and 337 cities, and is built on three dimensions: coverage breadth, usage depth, and level of digitalization. We use the aggregate city-level index to approximate the regional development of DFI. Meanwhile, we lag the DFI index by one year to mitigate endogeneity arising from reverse causality in the regression.

3.2.3. Control variables

Drawing on the health determinants literature, we include control variables at three levels. At the individual level, we control for gender, age, marital status, educational attainment, work status, an independent income source, initial health status,2 exercise habits, and enrollment in public medical insurance. At the household level, we account for household size. Given the considerable economic disparities between Chinese cities that could directly affect health, we also control for GDP growth rates at the city level.

3.2.4. Mediating variables

We measure the ability of older adults to pay for health investments based on their household income and assets. Income represents the “flow” of economic resources, while assets represent the “stock.” Household income includes wages, agricultural income, income from self-employment, and transfer income. Some values in the household income variable are negative. To maximize sample retention, household income is measured on an ordinal scale from 0 to 3,3 with higher values denoting higher income brackets. Household assets are calculated as a couple's net non-housing assets, and its natural logarithmic form is used in the regression.

Willingness to pay for health investments reflects a trade-off between current health consumption and other goods and services (Cropper, 1977). In the absence of stated preference approaches, this willingness can be inferred from actual spending patterns. We focus on preventive health investments, or those made prior to health deterioration, rather than reactive expenditures. Physical examinations are a widely used proxy for preventive health investment (Courbage & Kalouguina, 2020; Margaris & Wallenius, 2024). Similarly, spending on fitness and health supplements has been shown to reduce future medical expenditures and improve quality-adjusted life years cost-effectively (Patel et al., 2011; Smith et al., 2020; GuerreroPatiño, Canchila, & Buendía, 2025). Accordingly, we use physical examinations and fitness expenditures as proxy variables for individuals’ willingness to pay for health investments. Specifically, respondents are assigned a value of 1 for physical examinations if they have undergone routine checkups in the past two years and a value of 1 for fitness expenditures if they have purchased fitness-related products, services, or health supplements in the past year. Table 1 reports variable definitions and descriptive statistics.

Table 1.

Variable definitions and descriptive statistics.

Variable	Definition	Mean	SD
Self-rated health	1 (very poor) to 5 (very good)	2.942	0.964
Impairment of ADL	Impairment in basic activities of daily living	0.506	1.118
Impairment of IADL	Impairment in instrumental activities of daily living	0.733	1.308
Depression	CES-D10 depression index (0-30)	8.463	6.397
DFI	One-year lagged city-level Digital Financial Inclusion Index	158.254	57.000
Gender	Male = 1	0.499	0.500
Age
60-64 years∗	60-64 years = 1	0.354	0.478
65-69 years	65-69 years = 1	0.282	0.450
70-74 years	70-74 years = 1	0.181	0.385
75 years and over	75 years and over = 1	0.182	0.386
Marital status	Spoused = 1	0.806	0.395
Education	High school or above = 1	0.082	0.274
Work status	Working = 1	0.532	0.499
Independent income	Independent income = 1	0.777	0.416
Initial health	Self-reported chronic diseases in the last wave	1.737	1.594
Exercise	Regular exercise = 1	0.572	0.495
Medical insurance	Enrolled in public medical insurance = 1	0.942	0.233
Household size	Number of members in household	2.911	1.581
GDP growth rate	City GDP growth rate (%)	7.955	4.741
Household income	0 (very low) to 3 (very high)	1.485	1.120
Household assets	The natural logarithm of net family assets at the couple level	6.697	3.792
Physical examinations	Have undergone routine physical examinations in the past two years = 1	0.512	0.500
Fitness expenditures	Have fitness expenditures last year = 1	0.067	0.250

Note: ∗ indicates the reference group.

3.3. Empirical methodology

We examine the effect of DFI on older adults’ health using a two-way fixed effects model. This model enables us to account for region and time fixed effects while controlling for unobserved heterogeneity and time-invariant characteristics that could affect health outcomes:

$$Y_{it}={\alpha }_0+{\alpha }_1{DFI}_{c,t-1}+{\alpha }_2{X}_{it}+{\varphi }_p+{\phi }_t+{\mu }_{it}$$ (1)

Where $Y_{it}$ denotes the health outcome of older adult $i$ at time $t$. ${DFI}_{c,t-1}$ is the DFI index for city $c$ in year $t-1$. $X_{it}$ encompasses the set of control variables previously specified. Province and survey-wave fixed effects are represented by ${\varphi }_p$ and ${\phi }_t$, respectively. The error term ${\mu }_{it}$ captures unobserved factors influencing individuals' health.

4. Results

4.1. Baseline regression results

Table 2 displays the baseline regression results, which reveal a significant positive impact of DFI on older adults’ health. Specifically, a one-standard-deviation increase in the DFI is associated with a 0.1995 improvement in self-rated health scores among older adults. This positive effect extends to functional health, corresponding to reductions of 0.2622 points in ADL impairments and 0.3477 points in IADL impairments. Furthermore, mental health also improves significantly, with the same increase in DFI linked to a 1.63 decrease in CESD-10 depression scores. Collectively, these findings strongly support Hypothesis 1.

Table 2.

Two-way fixed effects estimation of DFI on health outcomes.

Variable	(1)	(2)	(3)	(4)
	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
DFI	0.0035∗∗∗	−0.0046∗∗∗	−0.0061∗∗∗	−0.0286∗∗∗
	(0.0008)	(0.0009)	(0.0012)	(0.0061)
Gender	0.0729∗∗∗	−0.0787∗∗∗	−0.2351∗∗∗	−1.6536∗∗∗
	(0.0140)	(0.0173)	(0.0248)	(0.1075)
65-69 years	−0.0112	0.0459∗∗∗	0.0772∗∗∗	−0.0575
	(0.0135)	(0.0157)	(0.0184)	(0.1175)
70-74 years	−0.0380∗∗	0.1282∗∗∗	0.2009∗∗∗	−0.1670
	(0.0187)	(0.0234)	(0.0271)	(0.1310)
75 years and over	0.0107	0.2529∗∗∗	0.4874∗∗∗	−0.6439∗∗∗
	(0.0261)	(0.0296)	(0.0341)	(0.1455)
Marital status	0.0145	−0.0780∗∗∗	−0.0517∗	−0.9242∗∗∗
	(0.0185)	(0.0275)	(0.0305)	(0.1357)
Education	0.1746∗∗∗	−0.2510∗∗∗	−0.3899∗∗∗	−1.8162∗∗∗
	(0.0265)	(0.0233)	(0.0305)	(0.1697)
Work status	0.1881∗∗∗	−0.2485∗∗∗	−0.3439∗∗∗	−0.0787
	(0.0166)	(0.0229)	(0.0278)	(0.1293)
Independent income	0.0634∗∗∗	−0.0683∗∗∗	−0.1307∗∗∗	−0.5255∗∗∗
	(0.0171)	(0.0188)	(0.0196)	(0.1092)
Initial health	−0.1622∗∗∗	0.1155∗∗∗	0.1181∗∗∗	0.7434∗∗∗
	(0.0053)	(0.0069)	(0.0081)	(0.0402)
Exercise	0.0519∗∗∗	−0.2306∗∗∗	−0.2732∗∗∗	−0.3181∗∗∗
	(0.0166)	(0.0181)	(0.0205)	(0.0997)
Medical insurance	0.0709∗∗∗	−0.1662∗∗∗	−0.1358∗∗∗	−0.5847∗∗∗
	(0.0236)	(0.0470)	(0.0417)	(0.1712)
Household size	0.0015	0.0011	0.0368∗∗∗	−0.1070∗∗∗
	(0.0050)	(0.0053)	(0.0068)	(0.0355)
GDP growth rate	−0.0019	0.0050∗	0.0046	0.0053
	(0.0022)	(0.0028)	(0.0030)	(0.0124)
Constant	2.3805∗∗∗	1.4991∗∗∗	1.9685∗∗∗	15.0338∗∗∗
	(0.1336)	(0.1746)	(0.2039)	(1.0332)
Region Fe	Yes	Yes	Yes	Yes
Year Fe	Yes	Yes	Yes	Yes
Observations	24,916	24,916	24,916	24,916
R2	0.1218	0.1007	0.1319	0.1173

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) The reference group for the “Age” variable is individuals aged 60-64 years.

The coefficients of the control variables are consistent with the literature on the health of older adults. It indicates that males are more prone to reporting better health. Physical function decline with age among older adults. Self-rated health is worse among individuals aged 70 to 74, and depression scores are lower among individuals aged 75 and over compared to those aged 60 to 64. Having a spouse is correlated with lower functional impairment and levels of depression. Higher socioeconomic status, as indicated by educational attainment, work status, and independent income sources, is positively related to better health. Initial health, as indicated by chronic conditions in the last wave, is negatively correlated with better health. Older adults who exercise regularly and enrolled in public medical insurance tend to enjoy better overall health. In terms of household characteristics, living with more cohabitants is related to greater impairment in IADL and lower depression levels. Regional economic development shows a positive correlation with older adults’ functional impairment.

4.2. Endogeneity analysis

The two-way fixed effects model may suffer from potential endogeneity concerns arising from omitted variables and reverse causality. To address these concerns, we employ an instrumental variable (IV) approach using two-stage least squares (2SLS) regressions. Specifically, we instrument the DFI with two city-level historical variables, the number of landline telephones per 100 people in 1984 and the number of post offices in 1984.

These instruments are relevant because the diffusion of internet services historically relied on pre-existing telephone infrastructure. Regions with higher fixed-line penetration in the 1980s were consequently better positioned for subsequent digital finance development. Furthermore, post offices were responsible for landline deployment, making their density a plausible predictor of subsequent digital infrastructure development.

Regarding the exclusion restriction, it is unlikely that these historical measures will directly influence current health outcomes. Nevertheless, to guard against potential confounding factors, we include an additional control in the two-stage least squares estimations. Regions with more extensive historical infrastructure may differ systematically in terms of economic development or healthcare provision. In addition to city-level economic conditions in the baseline regression, we control for the supply of medical services by measuring the number of hospital beds per 1000 people to reinforce the credibility of the exclusion restriction.

Table 3 presents the instrumental variable regression results. Column 1 shows the first-stage results, and columns 2 through 5 show the second-stage results for self-rated health, ADL impairment, IADL impairment, and depression. The results demonstrate the IVs are strong. This is evident from the IVs’ statistically significant coefficients at the 1% level (column 1) and the large Cragg-Donald Wald F statistic of 3368.642 in the first-stage regression. Hansen J statistics for overidentification are insignificant across all specifications (columns 2 through 5), confirming that the instruments meet the exclusion restriction. The DFI coefficients in columns 2 through 5 align with the findings of the two-way fixed effects regression, indicating the robustness of our baseline results after addressing potential endogeneity.

Table 3.

Instrumental variable regression results of DFI on health outcomes.

Variable	(1)	(2)	(3)	(4)	(5)
	The first stage: DFI	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Telephone	11.7449∗∗∗
	(2.9275)
Post office	0.0521∗∗∗
	(0.0062)
DFI		0.0054∗∗∗	−0.0053∗∗	−0.0050∗∗∗	−0.0359∗∗∗
		(0.0013)	(0.0020)	(0.0019)	(0.0094)
Hansen J		0.557	0.241	0.087	0.066
Observations	19,393	19,393	19,393	19,393	19,393
R2	0.4404	0.0997	0.0843	0.1104	0.0855

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) All regressions control for individual-level factors, city-level factors (GDP growth rate and number of hospital beds per 1000 people), and region and year fixed effects. (4) The Kleibergen-Paap rk LM statistic is 14.062. The Cragg-Donald Wald F statistic is 3368.642. (5) The details of this table are presented in Appendix Table A1.

4.3. Robustness analysis

We conduct a series of additional tests to validate the robustness of our baseline findings (Table 4). First, we convert the continuous dependent variables into binary measures, and the results are reported in Panel A. Specifically, self-rated health is dichotomized by assigning a value of 1 to scores of 4 or 5 and a value of 0 to all other scores. For ADL and IADL impairment, we assign a value of 1 if the respondent reported difficulty with one or more activities and a value of 0 if they did not. Depression is defined using a cutoff score of 10. Older adults with scores above this cutoff are classified as depressed. Probit models are used in the regression and control for the fixed effects of region and year. Second, the independent variable of aggregate DFI index is substituted with the breadth (Panel B) and depth (Panel C) of DFI. Third, we re-estimate the model using an ordered logit specification to accommodate the ordinal nature of the dependent variable (Panel D). Fourth, we exploit the panel structure of CHARLS and estimate two-way fixed effects models with individual and year fixed effects to account for time-invariant individual health endowments and common national time trends (Panel E). Results in Panels A through D are consistent with baseline estimates. In Panel E, DFI's coefficients for self-rated health and depression become statistically insignificant, while estimates for ADL and IADL impairments remain broadly unchanged. Since DFI is measured at the city-year level, individual fixed effects absorb much of the cross-city variation. This leaves only the within-city temporal variation for identification. The fact that the core results for objective physical limitations persist under this demanding specification reinforces the credibility of our findings.

Table 4.

Robustness test results.

	(1)	(2)	(3)	(4)
	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Panel A: Re-specify the dependent variables
DFI	0.0050∗∗∗	−0.0071∗∗∗	−0.0082∗∗∗	−0.0067∗∗∗
	(0.0015)	(0.0019)	(0.0018)	(0.0017)
Control variables	Yes	Yes	Yes	Yes
Region and year FE	Yes	Yes	Yes	Yes
Log pseudolikelihood	−11348.700	−12045.337	−13941.991	−14455.952
Panel B: Replace the independent variable
Breadth of DFI	0.0025∗∗∗	−0.0034∗∗∗	−0.0047∗∗∗	−0.0221∗∗∗
	(0.0005)	(0.0006)	(0.0007)	(0.0037)
Control variables	Yes	Yes	Yes	Yes
Region and year FE	Yes	Yes	Yes	Yes
Observations	24,916	24,916	24,916	24,916
R2	0.1218	0.1009	0.1322	0.1177
Panel C: Replace the independent variable
Depth of DFI	0.0030∗∗∗	−0.0047∗∗∗	−0.0062∗∗∗	−0.0272∗∗∗
	(0.0008)	(0.0012)	(0.0014)	(0.0074)
Control variables	Yes	Yes	Yes	Yes
Region and year FE	Yes	Yes	Yes	Yes
Observations	24,916	24,916	24,916	24,916
R2	0.1207	0.1002	0.1310	0.1161
Panel D: Ordered logit model
DFI	0.0093∗∗∗	−0.0137∗∗∗	−0.0148∗∗∗	−0.0107∗∗∗
	(0.0022)	(0.0032)	(0.0032)	(0.0025)
Control variables	Yes	Yes	Yes	Yes
Region and year FE	Yes	Yes	Yes	Yes
Observations	24,916	24,916	24,916	24,916
Log pseudolikelihood	−29981.184	−20193.607	−25622.335	−73860.193
Panel E: Two-way (individual and year) fixed effects model
DFI	0.0004	−0.0033∗∗	−0.0038∗∗	0.0022
	(0.0014)	(0.0016)	(0.0018)	(0.0091)
Control variables	Yes	Yes	Yes	Yes
Individual and year FE	Yes	Yes	Yes	Yes
Observations	24,916	24,916	24,916	24,916
R2	0.0151	0.0304	0.0382	0.0127

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) In Panel E, individual and year fixed effects are included in all specifications, absorbing individual time-invariant heterogeneity. Consequently, initial health status is omitted from the regression. Coefficients for gender, education, and the categorical age indicators are not separately identified due to perfect collinearity with the fixed effects. The squared term of age is included to address omitted variable bias, following Zhang et al. (2017).

4.4. Mechanism analysis

Based on the earlier established theoretical framework, DFI may improve the health of older adults by increasing their ability and willingness to pay for health investments. We use household income and household assets as proxies for payment capacity and physical examinations and fitness expenditures as proxies for willingness to pay for health investments. Table 5 shows the regression results for the proposed mediators. Table 6 reports the summary of the total, direct, and indirect effects, as well as the results of Sobel tests for the significance of the indirect effects across all pathways.

Table 5.

Mechanism analysis results.

	(1)	(2)	(3)	(4)	(5)
	Mediating variable	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Panel A: Mediating effect of household income
DFI	0.0095∗∗∗	0.0029∗∗∗	−0.0039∗∗∗	−0.0050∗∗∗	−0.0236∗∗∗
	(0.0026)	(0.0008)	(0.0009)	(0.0010)	(0.0058)
Household income		0.0885∗∗∗	−0.0887∗∗∗	−0.1317∗∗∗	−0.7019∗∗∗
		(0.0076)	(0.0096)	(0.0114)	(0.0540)
Observations	19,505	19,505	19,505	19,505	19,505
R2	0.1880	0.1317	0.1077	0.1410	0.1311
Panel B: Mediating effect of household assets
DFI	0.0215∗∗∗	0.0031∗∗∗	−0.0041∗∗∗	−0.0052∗∗∗	−0.0250∗∗∗
	(0.0066)	(0.0008)	(0.0009)	(0.0011)	(0.0061)
Household assets		0.0289∗∗∗	−0.0311∗∗∗	−0.0475∗∗∗	−0.2446∗∗∗
		(0.0022)	(0.0028)	(0.0034)	(0.0155)
Observations	19,505	19,505	19,505	19,505	19,505
R2	0.1291	0.1344	0.1110	0.1474	0.1371
Panel C: Mediating effect of physical examinations
DFI	0.0022∗∗∗	0.0037∗∗∗	−0.0046∗∗∗	−0.0061∗∗∗	−0.0291∗∗∗
	(0.0006)	(0.0009)	(0.0010)	(0.0012)	(0.0069)
Physical examinations		−0.0047	−0.0611∗∗∗	−0.0856∗∗∗	−0.5589∗∗∗
		(0.0164)	(0.0218)	(0.0244)	(0.1105)
Observations	19,505	19,505	19,505	19,505	19,505
R2	0.0866	0.1232	0.1019	0.1314	0.1208
Panel D: Mediating effect of fitness expenditures
DFI	0.0007∗∗∗	0.0037∗∗∗	−0.0047∗∗∗	−0.0062∗∗∗	−0.0300∗∗∗
	(0.0002)	(0.0009)	(0.0010)	(0.0012)	(0.0070)
Fitness expenditures		0.0210	−0.0423	−0.0830∗∗	−0.4212∗∗
		(0.0271)	(0.0288)	(0.0376)	(0.2009)
Observations	19,505	19,505	19,505	19,505	19,505
R2	0.0364	0.1233	0.1013	0.1306	0.1193

Notes: (1) In column 1, the dependent variables are household income, household assets, physical examinations and fitness expenditures in Panels A, B, C and D, respectively. (2) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (3) Standard errors are clustered at the city level. (4) All regressions control for individual-level factors, city-level factors (such as GDP growth rate), and region and year fixed effects.

Table 6.

Summary of mechanisms.

Mediating variable	Health outcome	Direct effect	Indirect effect	Total effect	IE/TE (%)	Sobel Z-value
Household income	Self-rated health	0.0029∗∗∗	0.0008∗∗∗	0.0037∗∗∗	22.7	3.540∗∗∗
	Impairment of ADL	−0.0039∗∗∗	−0.0008∗∗∗	−0.0047∗∗∗	17.7	−3.448∗∗∗
	Impairment of IADL	−0.0050∗∗∗	−0.0012∗∗∗	−0.0062∗∗∗	20.0	−3.537∗∗∗
	Depression	−0.0236∗∗∗	−0.0067∗∗∗	−0.0303∗∗∗	22.0	−3.571∗∗∗
Household assets	Self-rated health	0.0031∗∗∗	0.0006∗∗∗	0.0037∗∗∗	16.8	3.185∗∗∗
	Impairment of ADL	−0.0041∗∗∗	−0.0006∗∗∗	−0.0047∗∗∗	14.1	−3.150∗∗∗
	Impairment of IADL	−0.0052∗∗∗	−0.0010 ∗∗∗	−0.0062∗∗∗	16.4	−3.195∗∗∗
	Depression	−0.0250 ∗∗∗	−0.0053∗∗∗	−0.0303∗∗∗	17.4	−3.215∗∗∗
Physical examinations	Self-rated health	0.0037∗∗∗	−0.0000	0.0037∗∗∗	−0.3	−0.283
	Impairment of ADL	−0.0046∗∗∗	−0.0001∗∗	−0.0047∗∗∗	2.8	−2.255∗∗
	Impairment of IADL	−0.0061∗∗∗	−0.0001∗∗	−0.0062∗∗∗	3.0	−2.577∗∗
	Depression	−0.0291∗∗∗	−0.0012∗∗∗	−0.0303∗∗∗	4.0	−3.037∗∗∗
Fitness expenditures	Self-rated health	0.0037∗∗∗	0.0000	0.0037∗∗∗	0.4	0.745
	Impairment of ADL	−0.0047∗∗∗	−0.0000	−0.0047∗∗∗	0.6	−1.295
	Impairment of IADL	−0.0062∗∗∗	−0.0000∗	−0.0062∗∗∗	0.9	−1.718∗
	Depression	−0.0300∗∗∗	−0.0003 ∗	−0.0303∗∗∗	0.9	−1.665∗

Notes: IE/TE (%) denotes the proportion of the total effect that is mediated.

We first examine the mediating role of household income.4 As shown in Panel A of Table 5, column 1 indicates that DFI has a statistically significant positive effect on household income. Columns 2 through 5 of the same panel show that, when both DFI and household income are included in the model, the direct effect of DFI on each health outcome is weakened compared to the baseline estimate, while household income remains significantly associated with all four health indicators. These patterns suggest the presence of indirect effects operating through household income. Table 6 formally confirms that the Sobel tests for household income are significant at the 1% level for all health outcomes and the indirect effects (DFI → household income → health outcomes) account for between 17.7% and 22.7% of the total effect. This establishes household income as a partial mediator.

Similarly, household assets also serve as a significant mechanism. In Table 5, column 1 of Panel B shows that DFI has a positive and statistically significant effect on household assets, indicating that DFI facilitates wealth accumulation for older adults. When both DFI and household assets are included in the regressions (columns 2 through 5 of Panel B), the coefficients for both variables remain statistically significant, while the effect of DFI on health is reduced. The mediating role of household assets is further supported by the Sobel tests reported in Table 6. The absolute z-statistics for all health outcomes exceed 2 (p < 0.001), and the indirect effects account for 14.1% to 17.4% of the total effects. These empirical results strongly support Hypothesis 2.5

Turning to mechanisms related to willingness to pay, we first consider physical examinations. As presented in Panel C of Table 5, column 1 reveals that DFI significantly increases the probability of having undergone physical examinations in the past two years. Nevertheless, column 2 shows that the effect of physical examinations on self-rated health is insignificant, implying that mediation through this pathway is not supported. By contrast, columns 3 through 5 of Panel C reveal that physical examinations serve as a partial mediator between DFI and functional impairments (ADL and IADL) as well as depression. The Sobel tests in Table 6 confirm these indirect pathways, with the proportions mediated ranging from 2.8% to 4.0% across these outcomes.

Finally, we examine the role of fitness expenditures. Panel D of Table 5, column 1 indicates that DFI positively and significantly affects fitness expenditures. When both DFI and fitness expenditures are included in the regressions (columns 2 through 5 of Panel D), the coefficients for fitness expenditures are statistically significant only for IADL impairments and depression (columns 4 and 5 of Panel D). The Sobel tests in Table 6 yield z-statistics of −1.718 and −1.665 for these two outcomes, validating the mediating role of fitness expenditures in the pathways through which DFI mitigates declines in instrumental functioning and alleviates depressive symptoms. However, the proportions mediated are small, each below 1% of the total effect. These empirical results provide partial support for Hypothesis 3.

4.5. Heterogeneity analysis

To gain future insights into who benefits more from DFI, we conduct heterogeneity analyses. Table 7, Table 8 present the results by regional and individual characteristics, respectively.

Table 7.

Heterogeneity analysis by regional disparities.

DFI	(1)	(2)	(3)	(4)
	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Panel A
Midwest (N = 16,424)	0.0037∗∗∗	−0.0056∗∗∗	−0.0073∗∗∗	−0.0289∗∗∗
	(0.0010)	(0.0013)	(0.0016)	(0.0082)
East (N = 8492)	0.0028∗	−0.0020∗	−0.0031∗∗	−0.0276∗∗∗
	(0.0015)	(0.0011)	(0.0015)	(0.0068)
Differences	P = 0.128	P = 0.001	P = 0.000	P = 0.422
Panel B
Rural (N = 15,442)	0.0017∗	−0.0029∗∗	−0.0031∗∗	−0.0156∗∗
	(0.0009)	(0.0012)	(0.0014)	(0.0069)
Urban (N = 9474)	0.0021∗∗	−0.0021∗∗	−0.0040∗∗∗	−0.0170∗∗∗
	(0.0009)	(0.0008)	(0.0010)	(0.0062)
Differences	P = 0.279	P = 0.216	P = 0.204	P = 0.366

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) All regressions control for individual-level factors, city-level factors (such as GDP growth rate), and region and year fixed effects.

Table 8.

Heterogeneity analysis by individual disparities.

Variable	(1)	(2)	(3)	(4)
	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Panel A
DFI	0.0038∗∗∗	−0.0036∗∗∗	−0.0047∗∗∗	−0.0286∗∗∗
	(0.0009)	(0.0009)	(0.0012)	(0.0062)
DFI∗Digital access	−0.0004	−0.0009∗∗	−0.0016∗∗∗	0.0002
	(0.0003)	(0.0005)	(0.0006)	(0.0020)
Digital access	0.0752∗	0.0114	0.0110	−0.6057∗
	(0.0438)	(0.0761)	(0.0906)	(0.3204)
Observations	24,791	24,791	24,791	24,791
R2	0.1217	0.1025	0.1357	0.1187
Panel B
DFI	0.0031∗∗∗	−0.0045∗∗∗	−0.0066∗∗∗	−0.0256∗∗∗
	(0.0008)	(0.0010)	(0.0013)	(0.0064)
DFI∗Young-old	0.0005∗	−0.0001	0.0005	−0.0038∗∗
	(0.0003)	(0.0004)	(0.0004)	(0.0017)
Young-old	−0.1047∗∗	−0.1761∗∗∗	−0.4831∗∗∗	1.2512∗∗∗
	(0.0518)	(0.0636)	(0.0772)	(0.2981)
Observations	24,916	24,916	24,916	24,916
R2	0.1217	0.0992	0.1293	0.1174
Panel C
DFI	0.0035∗∗∗	−0.0044∗∗∗	−0.0060∗∗∗	−0.0284∗∗∗
	(0.0008)	(0.0010)	(0.0012)	(0.0064)
DFI∗Live with children	−0.0000	−0.0004∗	−0.0005∗	−0.0008
	(0.0002)	(0.0002)	(0.0003)	(0.0016)
Live with children	0.0192	0.1025∗∗	0.1969∗∗∗	0.4343
	(0.0379)	(0.0431)	(0.0505)	(0.2907)
Observations	24,602	24,602	24,602	24,602
R2	0.1222	0.1008	0.1344	0.1184

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) All regressions control for individual-level factors, city-level factors (such as GDP growth rate), and region and year fixed effects. (4) In Panel B, since the young-old variable is a binary indicator constructed based on the 75-year cutoff, age is no longer separately controlled in the regressions.

First, we divide the sample into Midwestern and Eastern provinces, as well as urban and rural sub-samples based on the residence. The Midwestern provinces are economically less developed than the Eastern region in China. As shown in Table 7, Panel A reveals that the negative effects of DFI on ADL and IADL impairments are significantly more pronounced among older adults in the Midwestern provinces (columns 2 and 3). These inter-group differences are statistically significant, as evidenced by coefficient difference tests. In contrast, Panel B shows that the effects of DFI on health do not differ significantly between urban and rural areas.6

Second, to examine whether the health effects of DFI vary by digital access, we construct a digital access indicator based on the CHARLS questionnaire. Respondents are coded as 1 if they own a mobile phone, computer, or other device used to access the internet, and 0 otherwise. We then include this indicator and its interaction term with DFI in the regression model. As shown in Panel A of Table 8, digital access significantly amplifies the beneficial health effects of DFI, particularly in reducing ADL and IADL impairments, as indicated by the negative and statistically significant interaction coefficients (columns 2 and 3).

Third, to analyze heterogeneity by age and living arrangements, we construct two variables. The first is an indicator for the young-old, defined as those aged 60 to 74 years following the WHO classification. The second is an indicator for whether the respondent lives with their children. The young-old are generally more receptive to digital technologies, and older adults living with children are more likely to receive guidance in digital use. Both characteristics are therefore associated with higher digital literacy.7 As shown in Panels B and C of Table 8, the young-old derive greater self-rated health (interaction coefficient = 0.0005, p < 0.1) and mental health (interaction coefficient = −0.0038, p < 0.05) benefits from DFI. In addition, living with children enhances the protective effects of DFI against functional decline, with significant interaction terms for ADL (coefficient = −0.0004, p < 0.1) and IADL impairments (coefficient = −0.0005, p < 0.1).

5. Discussion

As China's population continues to age, improving the health of older adults has become a major public concern. Against the backdrop of the “Healthy China 2030”8 initiative outlining the importance of strengthening individual responsibility for health, it has become crucial to stimulate residents' ability and willingness to invest in their own health. The development of DFI offers an important entry point for exploring this issue. This study uses three waves of data from the CHARLS, combined with Peking University Digital Financial Inclusion Index. It empirically examines the impact of DFI on the physical and mental health of older adults and its underlying mechanisms.

First, our findings provide ample evidence that DFI significantly improves the health status of older adults. This is manifested in enhanced self-rated health, reduced risk of impaired ADL and IADL, and mitigated depression levels. This conclusion remains generally stable when considering various specifications and endogeneity.

These findings are consistent with the conclusions of Banerjee et al. (2022) and Naveenan et al. (2024), who used macro-level data to examine the impact of financial inclusion on population health in developing countries. Specifically, regarding micro-level health indicators for older adults, Gyasi et al. (2019) used data from Ghanaian adults aged 50 or older, finding traditional financial inclusion positively correlated with self-rated health and negatively correlated with psychological distress. Ajefu et al. (2020), using Nigerian household survey data, found a negative causal impact of traditional financial inclusion on older adults’ depressive symptoms. Our study confirms that DFI similarly has a positive effect on self-rated health and psychological well-being in China. However, this finding should be interpreted with some caution, as the coefficients for self-rated health and depression become statistically insignificant when individual fixed effects are introduced. This pattern suggests that the relationship with subjective health may be more sensitive to individual-level uptake of digital services rather than solely to macro-level DFI exposure.

Regarding physical health, Aguila et al. (2016), using data from the Health and Retirement Study, found that owning a bank account had no significant effect on the physical health of older Hispanic adults. Similarly, Gyasi et al. (2019) reported that after controlling for socioeconomic and health-related factors, financial inclusion did not significantly influence healthcare utilization. In contrast, our study shows that the development of DFI significantly improves the physical health of older adults. Several factors may explain these differences. In the studies by Aguila et al. (2016) and Gyasi et al. (2019), financial inclusion was measured using indicators such as “owning a bank account” or a composite index based on having savings, loans, or a credit card. These measures primarily examined the breadth of financial coverage. In comparison, our research employs a Digital Financial Inclusion Index that not only accounts for coverage breadth but also emphasizes usage depth and digitalization. Usage depth includes sub-indicators such as the number of users and transaction volumes in payments, credit, insurance, and investment. Increased depth of digital finance usage helps disperse financial risks resulting from health shocks and reduces the likelihood of forgoing medical care (Xu et al., 2025). Moreover, digitalization of financial services aligns closely with China's “Internet plus healthcare” policy.9 This has resulted in the development of a distinctive digital healthcare model and enhanced access to high-quality health services (Zhou & Chen, 2021). The widespread integration of health management services into digital financial platforms has further increased public awareness of health management and potentially encourages the utilization of health services. These combined factors may contribute to the alleviation of impairments in ADL and IADL among older adults.

Second, we identify two primary channels through which DFI promotes health: enhancing the ability and willingness to pay for health investments. On the one hand, DFI boosts payment capacity for medical services by increasing household income and asset levels. On the other hand, DFI increases the willingness to pay for preventive health investments, such as check-up packages, health supplements, and fitness services. The mediating effect of payment capacity is more pronounced. DFI has a more direct and potent effect on economic capacity for health investment, such as by facilitating entrepreneurship and optimizing asset allocation, which can alleviate financial constraints. In contrast, the enhancement of health awareness is a longer-term process. The shift for older adults from passive treatment to disease prevention may require more time, resulting in a smaller mediating effect on the willingness to invest in health.

Third, the health-improving effect of DFI is stronger among older adults in Midwestern provinces, with digital access, under 75 years old, and living with their children.

The more pronounced health effects in the Midwestern provinces underscore the inclusive nature of digital finance. DFI markedly improves financial accessibility for older adults in these less developed regions, who have historically been excluded from traditional financial systems (Beck et al., 2007). It enables older adults to afford medical care and preventive expenditures, while promoting healthier lifestyles through convenient digital payments for health-related products and services. Consequently, DFI yields larger marginal health gains for Midwestern older adults than Eastern counterparts, who already benefit from well-developed financial and healthcare infrastructures. This finding is consistent with the literature on the welfare consequences of digital finance (Lei et al., 2023; Shen et al., 2025).

At the individual level, the health benefits of DFI are further shaped by the digital divide. First, regarding digital access, older adults who own smartphones or computers are better positioned to utilize digital financial services for health-related payments, telemedicine, and preventive care, while those without such access remain excluded from these channels (Ding et al., 2023). Second, the health benefits also differ significantly across levels of digital literacy (Sieck et al., 2021). The young-old (aged under 75) are more receptive to digital technologies, and those living with children benefit from intergenerational technical support, both of which amplify the health effects of DFI.

All the findings above are derived from China, where DFI is characterized by uniquely integrated super-apps such as Alipay and WeChat Pay, supported by state-led digital infrastructure. While these institutional features are distinct, they offer valuable insights for other economies undergoing digital transformation.

The underlying mechanisms through which DFI affects health are likely to be applicable more widely. Firstly, alleviating liquidity constraints during health shocks enhances the ability to pay for health investments, and this has been documented in various settings, including Sub-Saharan Africa (Ayayi et al., 2025; Franke et al., 2025). Secondly, reduced transaction costs and improved access to health information via digital platforms contribute to a greater willingness to pay for preventive care. This pattern is supported by global evidence on digital health interventions in low and middle-income countries (Sylla et al., 2025; Van de Vijver et al., 2023).

Thus, while the magnitude of the effects may be China-specific, the main pathways, such as enhanced payment capacity, reduced transaction costs, and improved health literacy, are likely to operate in any context in which digital financial services are widely adopted. However, realizing these benefits requires addressing persistent digital divides in access and literacy. Without doing so, the enhanced ability and willingness to pay enabled by DFI cannot translate into equitable health gains across populations (Sieck et al., 2021).

However, the current study has several limitations. First, DFI is measured at the city-year level. A city with a high level of DFI development does not necessarily have all of its residents actively using digital financial services. The discrepancy between macro-level exposure and micro-level uptake may introduce ecological bias and attenuate the estimated health effects. Future research could address this issue by using more detailed microdata that capture individual-level engagement with digital finance, such as frequency of mobile payments or use of online wealth management services. These data should be paired with individual fixed effects specifications that more fully account for time-invariant unobserved heterogeneity in health endowments. Second, although endogenous issues were addressed through fixed-effects models and instrumental variable approach, eliminating endogeneity entirely remains challenging. There may be a bidirectional causal relationship between digital finance development and resident health, and unobservable omitted variables (such as regional economic dynamism and social capital) may influence both. Future research should explore more exogenous policy shocks or natural experiments to improve causal inference reliability. Third, as a proxy for willingness to pay, fitness expenditures are reported by a relatively small subset of the sample, which could limit statistical power and generalizability. Therefore, findings on this mechanism should be viewed as suggestive rather than definitive. Further research using more comprehensive measures of preventive health investments, such as the purchase of online health checkup packages and wearable device usage, would validate the pathways through which DFI influences the health of older adults.

6. Conclusions

Despite the limitations above, this paper, using a nationally representative data, examines whether and how DFI improves the physical and mental health of older adults in China. It reveals that higher levels of DFI are associated with better self-rated health, lower functional impairment, and reduced depression, with mechanisms operating through increased ability and willingness to pay for health investments. The health benefits of DFI are more pronounced among older adults in Midwestern provinces, with digital access, under 75 years old, and living with their children.

These findings have significant policy implications. First, given DFI's health-promoting effects, continued investment in digital financial infrastructure is warranted. Policymakers should prioritize expanding broadband coverage and promoting mobile payment platforms in less developed regions, such as the Midwestern provinces, to ensure broader accessibility. Simultaneously, strengthening data security and financial regulations is essential to build trust and confidence among older adults in using digital financial services.

Second, the mediating role of payment capacity suggests targeted interventions to strengthen older adults' financial ability to invest in health. Offering low-risk wealth management products tailored to older adults can enhance their long-term financial capacity. Meanwhile, providing health-focused microcredit products, including emergency medical loans and installment plans for preventive services, can directly address immediate liquidity constraints. Regarding willingness to pay, enhancing the convenience of health-related service payments is equally important. Digital finance platforms could simplify payment processes for healthcare expenses and preventive services, reducing transaction friction and making it easier for older adults to act on their health intentions. Additionally, behavioral interventions are needed to cultivate health awareness and proactive prevention. Digital finance platforms could deliver health education and behavioral nudges, such as check-up reminders or incentives for purchasing preventive services. Integrating health information modules into digital finance platforms' interfaces, offering reliable medical advice, appointment booking, and health literacy content would help shift older adults’ mindset from passive treatment to active prevention. Pilot programs combining financial incentives with preventive health behaviors, such as cashback for fitness expenditures, could encourage sustained engagement.

Third, the heterogeneous effects of DFI necessitate differentiated strategies. For older adults in the Midwestern provinces, continued investment in targeted financial education programs is essential to maximize health benefits. For individuals aged 75 and older and those living alone, improving digital access and literacy is urgent. Effective measures include subsidizing internet-enabled devices, developing age-friendly digital interfaces with simplified navigation and voice assistance, and establishing community-based training programs. These efforts would help bridge the digital divide and ensure that the health benefits of digital finance are equitably shared across aging population.

CRediT authorship contribution statement

Di Yuan: Writing – original draft, Formal analysis, Data curation, Conceptualization. Kexin Deng: Writing – review & editing.

Ethical statement

The CHARLS study has gotten approval for interviewing respondents and collecting data from the Biomedical Ethics Review Committee of Peking University (IRB00001052-11015), and informed consent was required to sign by the respondents. As this study was based on the de-identified, publicly available CHARLS data, it does not constitute human subject research. Its institutional review board review was waived because there was no interaction with any individual, and no identifiable private information was used.

Funding

This work has been funded by Chongqing Social Science Planning Fund (Project No. 2025NDQN64).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix.

Table A1.

Instrumental variable regression results of DFI on health outcomes.

Variables	(1)	(2)	(3)	(4)	(5)
	The first stage: DFI	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Telephone	11.7449∗∗∗
	(2.9275)
Post office	0.0521∗∗∗
	(0.0062)
DFI		0.0054∗∗∗	−0.0053∗∗	−0.0050∗∗∗	−0.0359∗∗∗
		(0.0013)	(0.0020)	(0.0019)	(0.0094)
Gender	−0.0351	0.0607∗∗∗	−0.0722∗∗∗	−0.2068∗∗∗	−1.5905∗∗∗
	(0.1687)	(0.0156)	(0.0203)	(0.0285)	(0.1190)
65-69 years	0.0636	−0.0027	0.0486∗∗∗	0.0543∗∗∗	−0.0536
	(0.2508)	(0.0143)	(0.0159)	(0.0183)	(0.1304)
70-74 years	−0.4669	−0.0328	0.1180∗∗∗	0.1724∗∗∗	−0.1819
	(0.3733)	(0.0218)	(0.0244)	(0.0278)	(0.1473)
75 years and over	0.0110	0.0308	0.2530∗∗∗	0.4489∗∗∗	−0.6537∗∗∗
	(0.4545)	(0.0293)	(0.0318)	(0.0368)	(0.1801)
Marital status	0.6939∗∗	0.0261	−0.1004∗∗∗	−0.0444	−1.0998∗∗∗
	(0.2748)	(0.0206)	(0.0331)	(0.0350)	(0.1552)
Education	1.4891∗	0.1682∗∗∗	−0.2370∗∗∗	−0.3710∗∗∗	−1.6689∗∗∗
	(0.7947)	(0.0284)	(0.0255)	(0.0325)	(0.1794)
Work status	−1.4975∗∗	0.1872∗∗∗	−0.2351∗∗∗	−0.3435∗∗∗	−0.0948
	(0.5890)	(0.0194)	(0.0238)	(0.0318)	(0.1444)
Independent income	0.4090	0.0534∗∗∗	−0.0884∗∗∗	−0.1446∗∗∗	−0.5761∗∗∗
	(0.2990)	(0.0197)	(0.0192)	(0.0218)	(0.1214)
Initial health	−0.0235	−0.1630∗∗∗	0.1104∗∗∗	0.1102∗∗∗	0.7454∗∗∗
	(0.1288)	(0.0058)	(0.0074)	(0.0082)	(0.0454)
Exercise	0.1782	0.0685∗∗∗	−0.2321∗∗∗	−0.2676∗∗∗	−0.3482∗∗∗
	(0.2534)	(0.0192)	(0.0203)	(0.0230)	(0.1115)
Medical insurance	0.5174	0.0454∗	−0.1011∗∗	−0.0694	−0.6770∗∗∗
	(0.3859)	(0.0241)	(0.0417)	(0.0433)	(0.2032)
Household size	−0.1871	0.0010	−0.0011	0.0351∗∗∗	−0.0648∗
	(0.1259)	(0.0056)	(0.0055)	(0.0077)	(0.0374)
GDP growth rate	0.1359	−0.0031	0.0027	0.0025	0.0089
	(0.2072)	(0.0023)	(0.0027)	(0.0028)	(0.0139)
Ln (hospital beds)	36.9777∗∗∗	−0.2527∗∗	0.0725	−0.1427	0.0902
	(9.9672)	(0.1150)	(0.1264)	(0.1336)	(0.6837)
Observations	19,393	19,393	19,393	19,393	19,393
R-squared	0.4404	0.0997	0.0843	0.1104	0.0855

Notes: (1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) The reference group for the “Age” variable is individuals aged 60-64 years.

Table A2.

Mediating effect of household income (continuous variable).

	(1)	(2)	(3)	(4)	(5)
	Household income	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
DFI	0.0173∗∗∗	0.0031∗∗∗	−0.0043∗∗∗	−0.0059∗∗∗	−0.0309∗∗∗
	(0.0042)	(0.0009)	(0.0011)	(0.0011)	(0.0077)
Household income		0.0257∗∗∗	−0.0244∗∗∗	−0.0378∗∗∗	−0.2494∗∗∗
		(0.0040)	(0.0057)	(0.0058)	(0.0291)
Observations	13,476	13,476	13,476	13,476	13,476
R2	0.2505	0.1199	0.0993	0.1233	0.1243
IE/TE (%)	-	12.6	8.9	10.0	12.2
Sobel Z-value	-	3.442∗∗∗	−2.947∗∗∗	−3.473∗∗∗	−3.692∗∗∗

(1) Household income is included as a continuous variable in logarithmic form. (2) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (3) Standard errors are clustered at the city level. (4) All regressions control for individual-level factors, city-level factors (such as GDP growth rate), and region and year fixed effects. (5) IE/TE (%) denotes the proportion of the total effect that is mediated.

Table A3.

Mediating effect of household income and household wealth (non-working sample).

	(1)	(2)	(3)	(4)	(5)
	Mediating variables	Self-rated health	Impairment of ADL	Impairment of IADL	Depression
Panel A: Mediating effect of household income
DFI	0.0107∗∗∗	0.0021∗∗	−0.0052∗∗∗	−0.0068∗∗∗	−0.0251∗∗∗
	(0.0028)	(0.0010)	(0.0013)	(0.0015)	(0.0069)
Household income		0.0990∗∗∗	−0.1370∗∗∗	−0.1833∗∗∗	−0.9297∗∗∗
		(0.0115)	(0.0152)	(0.0187)	(0.0778)
Observations	8796	8796	8796	8796	8796
R2	0.2487	0.1306	0.1164	0.1466	0.1433
IE/TE (%)	-	33.8	22.0	22.3	28.4
Sobel Z-value	-	3.463∗∗∗	−3.486∗∗∗	−3.528∗∗∗	−3.605 ∗∗∗
Panel B: Mediating effect of household assets
DFI	0.0265∗∗∗	0.0023∗∗	−0.0055∗∗∗	−0.0071∗∗∗	−0.0280∗∗∗
	(0.0068)	(0.0010)	(0.0013)	(0.0016)	(0.0074)
Household assets		0.0321∗∗∗	−0.0448∗∗∗	−0.0647∗∗∗	−0.2667∗∗∗
		(0.0032)	(0.0049)	(0.0058)	(0.0227)
Observations	8796	8796	8796	8796	8796
R2	0.1826	0.1333	0.1195	0.1540	0.1430
IE/TE (%)	-	27.2	17.8	19.5	20.1
Sobel Z-value	-	3.636∗∗∗	−3.592∗∗∗	−3.688∗∗∗	−3.710∗∗∗

(1) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. (2) Standard errors are clustered at the city level. (3) All regressions control for individual-level factors, city-level factors (such as GDP growth rate), and region and year fixed effects. (4) IE/TE (%) denotes the proportion of the total effect that is mediated.

Data availability

Data will be made available on request.