Can Digitalization Foster Sustainable Financial Inclusion? Opportunities for Both Banks and Vulnerable Groups

Abstract

Financial inclusion is a crucial link in achieving the Sustainable Development Goals (SDGs). Strengthening the capability of financial institutions to provide inclusive financial services can help to narrow the inequality gap and increase access to opportunities. Digitalization, with its competitive advantages and rapid growth, may be a powerful tool to foster financial inclusion and sustainable development. This paper examines the effects of bank digitalization on sustainable financial inclusion and explores two underlying incentive mechanisms in banks: profit driven and risk aversion. We construct a basic model and a mechanism model and exploit a nonlinear attempt, heterogeneous estimation as well as supplementary variable and instrument variable methods for a robustness test. The results of the basic model demonstrate that bank digitalization has significant positive effects on financial inclusion and the current financial inclusive effects are sustainable. The mechanism models designed as the mediation effect panel model suggest that digitalization enables banks to expand the business probability frontier of profit-driven behavior and pursuit of credit portfolio diversity in risk aversion behavior, thereby promoting sustainable financial inclusion. As a result of digitalization, vulnerable groups can benefit from sustainable financial inclusion, while financial inclusion feeds back into banks’ sustainable development. This paper conforms to the trend of the development of digitalization and provides theoretical and empirical support for banks to build digitalization and realize sustainable financial inclusion, which contributes to the “triple-win” financial ecology for improving banks’ performance, increasing the rights of vulnerable groups and promoting sustainable development throughout society.

1. Introduction

Financial inclusion, as a key enabler for the Sustainable Development Goals (SDGs), refers to the delivery of accessible and affordable financial services to all social classes and groups in a responsible and sustainable way, which removes the barriers for underserved and vulnerable members of society to meet their financial needs [1,2]. “Digitalization” is the topic du jour in financial inclusion and is being vigorously pursued and encouraged to improve the capability of financial institutions to provide sustainable inclusive financial services. Digital technologies, e.g., big data, cloud computing and artificial intelligence (AI), hold huge potential to reduce information asymmetry, facilitate interactive communication and secure data storage, transmission and sharing [3,4,5]. In China, some internet companies enabled by digital financial technology (Fintech)—later known as Bigtech firms—have established digital platforms to connect excluded and underserved populations and assist small- and medium-sized enterprises as well as low-income individuals to raise funds [5,6]. New Fintech start-ups digitally greet a mass of niche customers who lack access to conventional banks and offer affordable micro-financial services at a lower threshold [3]. The digital approach has played a critical role in achieving financial access and responsible financial inclusion, which unlocks sustainable development opportunities by reaching excluded and underserved populations with a range of formal financial services at a cost affordable to customers and sustainable for providers [1].

There are a large number of relevant studies focusing on the financial contribution of digitalization. For example, digitalization has positive impacts on reducing agency or transaction costs [4,7,8,9,10,11], enhancing financial access and inclusion [12,13] and improving the performance of both the financial sector and inclusive sector [8,14,15,16]. However, few studies refer to the sustainability of financial inclusion, especially in banks. On the one hand, as traditional financial intermediaries are generally subject to prudential regulation, banks have to be cautious about providing loan services, especially to those small micro and unfamiliar customers, avoiding the moral hazard and adverse selection [17]. On the other hand, digital-based innovations have not been widely available to use in traditional financial sectors [18], at least not popularized in banks. It is difficult for banks to weigh the necessity of investing in digital infrastructure and extending liquidity to vulnerable groups. As such, whether banks achieve sustainable financial inclusion (notably sustainable credit inclusion) by leveraging digitalization remains an open problem, although digitalization can be a powerful tool for financial inclusion.

Today, Chinese banks are trying to transform from early e-finance using computers or mobile phones as their complement channels to digital banking based on Fintech that can eliminate the limits of physical contact and soft information and interact with customers intelligently [3,19]. Allowing for the fast-growing development of digital finance and the digital transformation in bank-based financial systems, China is well placed to exemplify the effect that digitalization, especially Fintech, can have on banking [3]. A series of China’s action policies in the past decade, such as innovation-driven development strategy, “Internet Plus” plan of action, financial supply-side structural reform and Digital China, has enhanced digital construction in banking and prepared positive conditions for sustainable financial inclusion. Data quoted in a CNNIC report indicate the number of Chinese mobile Internet users reached 1.047 billion in June 2022 and over 0.904 billion use online payment [20]. The huge netizens and high digital finance penetration project a deep digital trust and acceptance furnishing the endogenous impetus of digital development, which could be significant for individuals and enterprises to access the sustainable-scale dividend of digitalization.

This paper explores whether the digitalization of banks can foster sustainable financial inclusion in China’s banks and provides evidence of how digitalization affects sustainable financial inclusion from banks’ behavior. We analyze the conditions for banks to increase sustainable financial inclusion through digitalization. Banking is an information-intensive business and an early adopter of information technology [21,22]. Fintech leading practitioners boldly explore to exploit long-tail clients who can be understood as uncertain or risky groups and offer financial services, which, in turn, bring them new profit growth points. The increased digital disruption and competition places pressure on banks to reconsider their existing business models and find potential market opportunities [23,24]. Banks can well benefit from digital innovation and spillover to develop inclusive innovation and vitalize data as a crucial resource, which enhances banks’ ability to serve the inclusive market.

We also propose two underlying incentive mechanisms for sustainable financial inclusion that indicate banks’ profit-driven and risk aversion behavior, respectively. The profit-driven mechanism describes digitalization can stimulate banks to expand the business probability frontier and provide credit liquidity to disadvantaged segments of society. The risk aversion mechanism depicts that banks use digitalization to make better portfolio decisions, avoiding the credit risk caused by relatively centralized and single loan objects.

Accordingly, two models are constructed: the basic model of the total effects of bank digitalization on sustainable financial inclusion and the mechanism models reflecting the banks’ profit-driven and risk aversion behavior. The models consist of a sample of 58 bank observations from 2015 to 2020 and apply fixed effects for regression. To improve the robustness, we take a nonlinear attempt, the heterogeneous estimation, as well as supplementary variable and instrument variable methods. The results of the basic model show that the digitalization of a bank has significant positive effects on financial inclusion, and the financial inclusive effects are sustainable and have not reached saturation at the current stage. The mechanism models designed as the mediation effect panel model suggest that bank digitalization enables banks to expand the business probability frontier of profit-driven behavior and pursuit of credit portfolio diversity of risk aversion behavior, thereby promoting sustainable financial inclusion.

As a result of digitalization, banks can provide their sustainable financial inclusive services to vulnerable groups through the profit-driven and risk aversion behavior of banks. Sustainable financial inclusion increases the opportunities of vulnerable groups to access financial services and feeds back into banks’ profit and data sources and risk reduction, which promotes the sustainable development of both vulnerable groups and banks. The “win-win” between vulnerable groups and banks can further lead to sustainable financial inclusion, thereby leading to the “triple-win” model for improving banks’ performance, increasing the rights of vulnerable groups and promoting sustainable development throughout society.

The main contributions of this study are as follows. First, we focus on sustainable financial inclusion and contribute to the effects of bank digitalization on sustainable financial inclusion and the two profit-driven and risk aversion mechanisms. These two incentive mechanisms combine return and risk and reflect how digitalization in banks increases financial inclusion through banks’ behavior, making up the shortage of only focusing on one aspect of return or risk as well as a lack of bank behavior analysis. Second, we propose the “triple-win” financial ecology as a result of digitalization for banks, vulnerable groups and the whole of society. Finally, this study renews the theories and study cases of digital financial inclusion. China’s case proves that the wide application of digitalization can relieve the sequelae of asymmetric information rooted in banks and narrows the distance between banks and niche groups for sustainable development, which can be used as a reference for other countries or regions. However, in some countries and regions, the results may vary due to the digital development conditions and environment [25,26].

The rest of this paper is organized as follows: Section 2 presents our theoretical framework and hypotheses; Section 3 explains the data source and empirical methodologies; Section 4 discusses the estimation results; Section 5 concludes the paper.

2. Theoretical Framework

2.1. Digital Innovation and Spillover

In 2014 and 2015, China’s largest e-commerce company, Alibaba, and tech giant Tencent spread their mobile payment platform Alipay and WeChat from online to a physical store, becoming the two most prominent financial technological firms. They accelerated digital innovation and application and designed a long-term layout in the financial field, offering payment, lending, insurance, wealth management and credit scoring services [3].

Compared with conventional banks, Fintech and Bigtech firms possess superior data processing capability that can screen candidate borrowers via mining data and events, which includes not only historical and ongoing occurrences but also potential and null values, thereby overcoming incomplete information. They leverage cloud computing and machine learning algorithms free of legacy systems to make electronic credit decisions and capture clients based on user data, which document customer behaviors, habits and preferences and remote network that connects both supply- and demand-side participants. Accordingly, more innovative advantages in terms of customer insight, business operation and risk management have been unfolding, leading to potential spillover effects to banks [6,27].

Banks are intensive users of IT and financial technologies that often depend heavily on the use of IT to collect, process and disseminate data, as well as on economic and statistical models to evaluate data [21]. Digital innovation and spillover have emerged to enable banks to achieve more sustainable technological improvement and give full play to their financial functions. For example, the digital system with data sharing and remote identification and authentication can provide an effective and secured interface between the bank and customers [28]; the increase in data analytics by retail banks permits them to estimate credit risks more precisely [29] and respond more quickly to the disadvantaged segments that may not have steady income or traditional credit history [5]. In addition, as digital applications—the introduction of digitalization to payment systems, lending process, investment business, etc.—become integrated and mature, digital innovation and spillover can expedite sustainable inclusive innovation in banking, which benefits not only large banks with strong creativity and absorptive capacity but also those small banks close to niche markets and eager to transform through independent or cooperative digital innovation [21,30,31].

Notably, powered by digital technologies, data are a basic resource needed for social production and management activities, especially in information-intensive sectors. In 2020, the Chinese government announced guidelines to improve the mechanisms for the market-based allocation of production factors, which defined data as the fifth-largest factor in production after land, labor, capital and technology. The combination of data and digital technologies helps to narrow the information differences and improve the efficiency of information management [9]. In practical terms, digital data typically record all kinds of customer information online (e.g., default history, associated personnel) and generate sustainable data trails when electronic transactions happen, thereby increasing transparency, sustainability and robustness in predicting risks and performance [32]. The traceable data further the ability of financial services to formally develop a credit scoring mechanism for informal market participants [5].

Therefore, Hypothesis 1 can be expressed as follows:

Hypothesis 1.

An increased adoption of digitalization would have positive impacts on sustainable financial inclusion.

2.2. Long-Tail Market and Expanding the Business Probability Frontier

For a long time, the Pareto Principle (or the 80–20 rule) dominated the financial market and formed the strategic boundary of traditional financial institutions, which led to the concentration of financial resources and services to high-end customers. Due to the limited technological condition and imperfect financial markets, banks operate prudently and are inclined to lend to those who provide collateral. Low-income crowds and mini-enterprises with no collateral as well as marginal groups with minimal financial identity find it hard to obtain loans, even if they can repay them on time. Thus, credit rationing—a long-term equilibrium phenomenon with excess capital demand in the credit markets—may be a proper choice for banks to avoid a risk before it happens [17]. Fortunately, the use of digital technology and the Internet increases the ability of information gathering and risk identification and makes it possible to focus on the tail groups at a lower search cost [33], thereby creating a long-tail range with high volume [34]. These tail customers offer valuable business data resources that constitute essential strategic factors in digital finance markets, which could generate new profitable revenues (for many Fintech firms, the overall benefits from long-tail customers may even exceed those of the head customers).

As a result of the early adoption of IT-intensive infrastructures, Chinese banks have embarked on flexible and diverse innovations to generate efficiency improvements and expanded the accessibility and convenience of financial services [30,35]. However, the advantages from early technology adoption tend to diminish as the knowledge and technology diffuse with externalities over time. When the technology adoption reaches a certain point, the technology saturation effect may diminish the returns [36]. Figure 1 shows the return of China’s banking sector. The trend of the return on assets (ROAs) and return on equity (ROE) peaked from 2011 to 2013 and then began to decline. During this period, the Internet and mobile technologies had important implications for banks in terms of internal operating and channel upgrading [22], but this did not change the bank’s lending attitude towards new and unfamiliar markets that may bring about sustainable profit growth. Meanwhile, the emergence of Fintech and platform-based competitors with great creative destruction accelerate market game and segmentation, which may be the erosion of banks’ margins and an increase in competitive pressure and contestability in financial markets [3].

To maintain sustainable revenue and long-term development, banks need to expand their business frontier to the long-tail market as new sources of profit growth (see Figure 2). On the one hand, profits from the original big clients of banks may have reached a boundary as the market scale of large customers becomes saturated, while business from the small customer market is booming, with a lot of room for development due to digital technology. On the other hand, digital technology increases information transparency with user data that could even replace collateral as a guarantee of credit, which not only serves to escort banks in the small customer market but enables innovative and strategic initiatives that are aimed at increasing market share [37,38]. In addition, customer data are continuously added to the bank database, which can offset the effects of externalities of technology that may dissolve the technical advantages of banks.

Thus, Hypothesis 2 can be expressed as follows:

Hypothesis 2.

Digitalization↗→ frontier↗→ financial inclusion↗, i.e., digitalization enables banks to expand their business probability frontier, thereby increasing sustainable financial inclusion, constituting the profit-driven mechanism for sustainable financial inclusion.

2.3. Making Better Portfolio Decisions

Practically speaking, entering the credit market is choosing a credit portfolio with uncertainty. A credit portfolio with multiple customers is expected to increase loan diversification, which can reduce unique or unsystematic risks [39]. According to modern portfolio theory (MPT), the familiar financial advice “don’t put all the eggs in one basket”, explained by owning different kinds of financial assets, is less risky than owning just one type, which illustrates this point, but only if the borrowers’ information is easily traceable. When the market is imperfect with high information barriers, extraordinary opacity as well as excessive financial friction, it is difficult for banks to explore the new territory and bear pressure. Thus, the reason why niche markets are “shied away” from by banks is probably not that niche customers really have high default rates and risk, but banks may have no advanced vehicles for picking up information to judge the uncertainty.

After years of efforts, the technological progress in lending services appears to have yielded benefits, not only in terms of small financial provision but the interaction with their niche customers. First, digital technologies improve the “quantity” and “quality” of credit provision by alleviating constraints, increasing opportunities for tail customers and enhancing information processing to screen candidate borrowers. Compared with traditional methods using credit bureau information, new applications depending on digital platform transactions and operations could have better predictive power in loan repayment prospects [40,41]. Second, banks using digital innovations allow for flexible and tailored financial services to meet clients’ specific needs and characteristics [30]. This could attract different types of customers and enrich bank portfolios. Third, increasing the convenience of digital banking services may enhance customers’ stickiness and dependence on a bank. Customers may sustainably increase their use of other business with the bank, e.g., adding deposit business and financial management in the same bank [42], which may further increase information access and credit behavior (new business will help banks hold more information to screen customers).

Given the unavoidable credit risk, finding the balance between credit portfolios and stability is considered an essential link in niche selection, since asset mix and structures may influence the risk and return for banks [43,44]. Portfolio decisions require special banking data sets with information on loan portfolio composition [21]. To some extent, banks no longer perform just as a financial intermediary combining depositors and borrowers to provide financial services, including liquidity supply, payment and transaction, but more like a computing analysis center that mines customer information, estimates return distributions and makes portfolio decisions based on customer data. Digital technologies, as the vehicle for analyzing data and mining information, improve the ability of risk prediction and make risk sharing easier [39]. As digital financial services are forwarded, access to financial services is more convenient and the information of customers and credit portfolios becomes more transparent and traceable. According to MPT, currently, better portfolio decisions involve selecting more diverse customers to avoid correlation risk, which can control bank’s risk-taking behavior [28,45] and contribute positively to stability in the banking sector, as depicted in Figure 3.

Today, more and more commercial banks extend their credit portfolios to niche markets to search for new opportunities in micro-finance operations. With a good reputation and brand recognition [3], as well as the franchise authorized by the central bank and regulators, Chinese banks have had a stable data base with a large amount of customer information, including soft information acquired through the person-to-person relationship at grass-roots branches [30]. However, restricted by obsolete technologies, a great amount of data are not fully utilized and mined to become credit information, so that banks may only select those head applicants with a stable financial state and collateral, which leads to single loan objects and credit concentration [46]. Digital technologies enable banks to obtain credit information by integrating multiple dimensions of data, thereby producing a diversified credit base with less opacity. In addition, personalized services are designed to be more tailor made, as is risk management. Banks can take advantage of advanced digital lending systems to make their credit portfolios, which include not only customers who previously had collateral but also newly entered niche customers with good credit through low collateral barriers to entry. Moreover, the increased financial inclusion also generates new personal information and transaction trails in real time, which makes up their sustainable credit portfolio data sets with sufficient updates, leading to greater bank resilience.

Therefore, Hypothesis 3 can be expressed as follows:

Hypothesis 3.

Digitalization↗→ diversity↗→ financial inclusion↗, i.e., digitalization can improve the pursuit of credit portfolio diversity, then promoting sustainable financial inclusion, constituting the risk aversion mechanism for sustainable financial inclusion.

2.4. “Triple-Win” Sustainability

Financial inclusion is not charitable work but sustainable development following the market. Digital-based financial inclusion is complementary to the traditional financial system, bringing formerly underserved groups into the formal financial system. For vulnerable groups, digitalization promotes financial inclusion, which increases equal rights in accessing financial services for production and livelihood, thereby improving their sustainable development. For banks, digitalization helps to promote banks to expand the business probability frontier to generate new profits and obtain data resources, and allows them to pursue better credit portfolios to reduce credit risk, enabling banks to benefit from financial inclusion to achieve sustainable development. The process of “two-win” furthers sustainable financial inclusion and promotes sustainable development throughout society, leading to the “triple-win” financial ecology (see Figure 4).

3. Data and Methodology

3.1. Data Source

Based on the accessible data from China’s A-share and H-share markets and banks’ annual reports, this paper selects 58 Chinese banks, covering 6 large commercial banks, 11 joint-stock banks, 30 city commercial banks and 11 rural commercial banks. These banks present the typical characteristics of Chinese commercial banks and have a large asset proportion in the banking sector. Considering that the economic effects of technology usually need to reach the critical mass point [36], we choose the years from 2015 to 2020, which are the middle stage of digital development after the first year of Internet finance in China (2013 is generally called the First Year of Internet Finance in China). The bank data are from the Wind database and annual reports, and the bank-specific digitalization data are collected by calling the customer service of banks and calculated manually. The provincial-specific digitalization data using the Digital Financial Inclusion Index are released by the institute of Digital Finance of Peking University. The macroeconomic data come from the national statistical yearbook.

3.2. Basic Model

According to the discussion on digitalization and sustainable financial inclusion, the total effect of digitalization on sustainable financial inclusion as the basic model is constructed in the following form:

$$cinclusio{n}_{i,t}={\alpha }_0+{\alpha }_1\cdot digi{t}_{i,t}+\delta X_{i,t}+{\epsilon }_{i,t}$$

(1)

where i denotes banks and t denotes years; cinclusion is the dependent variable defined as financial inclusion, which is measured by unsecured loans to total gross loans of different banks and supplemented by the logarithmic form of unsecured loans lncred for robust estimation; the independent variable digit is defined as the digital level of banks, and X represents a vector of control variables.

As a measure of ICT technology or infrastructure, digitalization usually has network externalities [47] and the saturation effect [36], which may lead to nonlinear effects. Although there is little possibility of digital externality at the stage of the year from 2015 to 2020, we still add an interaction term of bank-specific digitalization with itself (the quadratic term of digitalization) to the basic model for verification, referring to the approaches in [48,49]:

$$cinclusio{n}_{i,t}={\beta }_0+{\beta }_1\cdot digi{t}_{i,t}+{\beta }_2\cdot digi{t}_{i,t}\cdot digi{t}_{i,t}+\delta X_{i,t}+{\epsilon }_{i,t}$$

(2)

3.3. Profit-Driven and Risk Aversion Mechanism Model

To capture the impact of digital development on sustainable financial inclusion, as discussed in Section 2, we draw on [8,50] and employ the mediation effect panel model and the sequential test. The mediation effect model describes the effect of the independent variable on the dependent variable through the mediator variable [51,52]. The model is constructed as follows.

$$\begin{array}{l}cinclusio{n}_{i,t}=a_0+a_1\cdot digi{t}_{i,t}+a\cdot X_{i,t}+{\epsilon }_{i,t}\\ M_{i,t}=b_0+b_1\cdot digi{t}_{i,t}+u_{i,t}\\ cinclusio{n}_{i,t}=c_0+c_1\cdot digi{t}_{i,t}+c_2\cdot M_{i,t}+c{X}_{i,t}+{\eta }_{i,t}\end{array}$$

(3)

The variables cinclusion and digit are the dependent variable and independent variable of the mediation effect, respectively. The first equation is the total effects of digitalization on sustainable financial inclusion, as described in the basic model above. The second equation is the effect of the independent variable on the mediator variable M_i,t. In our study, the mediator variable M_i,t is induced to reflect the bank’s profit-driven and risk aversion behavior when digitalization affects sustainable financial inclusion. The profit-driven behavior presents an increase in loan scale and proportion, while risk aversion behavior shows the pursuit of credit diversity. The last equation contains the independent variable and the mediator variable, showing the effect of the mediator variable on the dependent variable when controlling the independent variable. According to [50,51,52], if the coefficients of the independent variable and mediator variable, i.e., $a_1$,$b_1$, $c_1$ and $c_2$ are significant, the mediation effect exists. It indicates that the bank’s profit-driven and risk aversion behavior can mediate the effect of the independent variable digit on the dependent variable cinclusion.

3.4. Variables

3.4.1. Dependent Variable

cinclusion is the main dependent variable in sustainable financial inclusion representing a kind of credit structure. Referring to the bank credit structure described in [53,54], we use the unsecured loans to total gross loans in a bank to denote the dependent variable cinclusion. Financial inclusion is not only for individuals and small- and medium-sized enterprises but also for those marginal borrowers lacking financial records or with no collateral. Unsecured loans can reflect the attitude of banks to vulnerable groups, since the procedures of borrowing and repaying depend more on credit than collateral or asset certificates. lncred is the logarithmic form of unsecured loans as a supplementary variable to measure financial inclusion. They are both displayed in the analysis of loans and advances to customers by collateral-type bank annual reports.

3.4.2. Independent Variable

digit represents the bank-specific digital development level that can be measured by the years of mobile banking service. Due to the disunity in digitalization-related indicators published by various banks (for example, electronic banking transaction amount and transaction times are not published by every bank), the service years of mobile banking can be used as a substitute for the digital level of different banks. On the one hand, mobile devices and mobile banks are the main carriers and platforms for digital financial services, which can be regarded as the digital infrastructure development level of banks; on the other hand, mobile banking services and digital financial development were both in the middle stage of digital development during 2015 to 2020 and had similar marginal properties.

3.4.3. Mediator Variable

frontier is the profit-driven mediator variable, representing the bank’s behavior in expanding the business probability market. Similar to the product market, improvements in technology can lead to an increase in production, while advanced digital adoption represents the expansion of credit business in the credit market. This paper considers the scale and proportion of the expanded business frontier and selects the loan-to-deposit ratio as the data for this indicator. It is believed that the increase in digitization will lead to an increase in this variable and, thus, improve sustainable financial inclusion.

diversity indicates the risk aversion mediator variable. Some scholars use the Hirschman–Herfindhl Index (HHI) to measure the degree of credit concentration [44,53,54]. To more intuitively reflect the meaning of the variable diversity, we draw on the research of [55,56] and define diversity as the reciprocal of the HHI index. HHI, here, is the sum of the squares of the four types of loan ratios classified by collateral type, including unsecured loans (s₁), guaranteed loans (s₂), collateralized (s₃) and other secured loans (s₄). To be more specific, the HHI index model shown in the following equation describes the loan concentration:

$$S={\displaystyle \sum _{i=1}^4{s}_i} , HHI={\displaystyle \sum _{i=1}^4{\left(\frac{s_i}{S}\right)}^2}i=1,2,3,4$$

(4)

An increase in HHI suggests a greater portfolio concentration from a bank, with a value range of (1/4, 1). Then, the portfolio diversity is:

$$diversity=1/HHI$$

(5)

with a value range of (1, 4), of which 1 represents the most concentrated and 4 represents the most diversity. Notably, a lower HHI means more diversity, which is expected to be a positive relationship between digitalization and diversity and diversity and sustainable financial inclusion; that is, digitalization increases diversity, thus increasing sustainable financial inclusion.

3.4.4. Control Variable

roa is the average return on total assets, which measures how much the bank uses and manages its resources to generate revenue; cti is the cost-to-income ratio (%), reflecting the bank’s ability to control costs. Referring to previous articles, such as [44,57], size can be expressed in logarithmic form as assets in every bank. inf, as the control variable to adjust the absolute variable lncred, is the inflation growth rate (%) using the CPI consumer price index over the previous year.

cp is a type variable in this model, which is defined as the classification by bank attribute, including four types: 1 = large commercial banks, 2 = joint-stock banks, 3 = city commercial banks, 4 = rural commercial banks.

3.4.5. Instrument Variable

Given the endogeneity in the empirical regression, lndi is induced as the instrument variable of digitalization, which adopts provincial-level data from the Digital Financial Inclusion Index from the institute of Digital Finance of Peking University. The Peking University Digital Financial Inclusion Index of China (PKU-DFIIC) consists of 3 primary indicators and 33 secondary indicators and utilizes Ant Financial’s massive data set on digital financial inclusion [58]. The PKU-DFIIC makes up for the shortcomings of single index construction and insufficient attention to new digital financial services [59]. To avoid heteroscedasticity and obtain a robust data series, lndi takes the logarithmic form of PKU-DFIIC without changing the nature and correlation.

lndi describes the provincial-level digital development where the bank is located. As a general rule, the regional digital development level usually reflects the characteristics of the digitalization level of a local bank. This provincial PKU-DFIIC has linear correlation with the bank’s digitalization level, but it will not be affected by the specific bank’s financial inclusion level. Thus, lndi can be used as the instrument variable to solve the endogeneity problem and estimate the model.

Table 1. Variable definition and descriptive statistics.

Variable	Definition	Obs	Mean	Std. Dev.	Min	Max
cinclusion	Bank unsecured loans to total gross loans (%)	348	16.24	10.87	1.04	51.19
lncred	Logarithm of unsecured loans of the bank	348	6.14	2.33	1.49	11.04
digit	The years of using Mobile Banking Service	348	5.54	2.41	0	12
lndi	Natural logarithm of Peking University Digital Financial Inclusion Index	348	5.70	0.20	5.26	6.07
roa	Average return on total assets (%)	334	0.84	0.24	−0.59	1.81
frontier	Deposit loan ratio (%)	348	71.88	15.12	38.97	123.42
diversity	The reciprocal of HHI index	346	3.07	0.42	1.88	4.30
cti	Cost-to-income ratio (%)	348	30.45	6.42	14.83	66.47
size	Logarithm of total assets of the bank	348	3.93	0.68	2.50	5.52
cp	Type by bank attribute (1 = large commercial banks, 2 = joint-stock banks, 3 = city commercial banks, 4 = rural commercial banks)	348	2.79	0.86	1	4
inf	Inflation growth rate (%)	348	2.08	0.51	1.4	2.9

shows the definitions and descriptive statistics of the selected variables.

4. Empirical Results and Discussion

This section presents the empirical results of the basic and mechanism models. Considering that the bank-level individual effects are probably unobservable and correlated with some explanatory variables, we follow previous works, such as [9,44], to choose fixed effect models for regression.

4.1. The Results of the Basic Model

Table 2. Estimation results for basic model.

Fixed Effects Models
	(1) Cinclusion	(2) Cinclusion	(3) Lncred	(4) Lncred
digit	1.716 *** (0.2780)	1.907 *** (0.5559)	0.143 *** (0.0267)	0.324 *** (0.0466)
cti	−0.0467 (0.0759)	−0.0526 (0.0775)	0.00551 (0.0063)	−0.000292 (0.0062)
roa	2.052 * (1.1749)	2.100 * (1.1830)	−0.0670 (0.0959)	−0.0220 (0.0930)
size	0.788 (5.6482)	−0.138 (6.1185)	3.459 *** (0.4617)	2.597 *** (0.4818)
digit_digit		−0.0131 (0.0331)		−0.0121 *** (0.0026)
inf			−0.0404 (0.0539)	−0.0566 (0.0521)
cons	3.516 (21.6747)	6.732 (23.1667)	−8.258 *** (1.7758)	−5.243 *** (1.8295)
axis of symmetry (=12)		72.786		13.39 ***
N	334	334	334	334

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% * 10%.

describes the changes in bank sustainable financial inclusion caused by digital development in the basic model. Because the digit is measured by years of adopting the mobile banking services that may lead to multiple collinearity with time fixed effects, these models are estimated only using individual fixed effects. Columns (1) and (3) show the basic digital effects of banks, and (2) and (4) use the quadratic term of the explanatory variable digit for regression. As shown in Table 2, the coefficients of digit in all columns are statistically significantly positive, indicating that digitalization has a positive impact on sustainable financial inclusion. The interaction term of digitalization with itself (the quadratic term of digitalization) is added for testing whether the positive effect exists or changes due to the network effect during 2015–2020. It shows the negative effects of the quadratic term digit_digit on the dependent variable cinslusion of column (2) and lncred in column (4). The former coefficient of the quadratic term is not significant, while the latter is significant, rejecting the assumption that the quadratic term coefficient is zero at the level of 1% significance. To further explain the effects, we calculate the digit value at the turning point of this effect by using the coefficient of the quadratic term and the single term coefficient (if a quadratic function is expressed as f(x) = ax² + bx + c (a < 0), the x = −b/2a is the axis of symmetry. When x < −b/2a, f(x) will increase with the increase in x; thus, x has a positive impact on f(x); when x > −b/2a, x has a negative impact on f(x).) The axis of symmetry (digit value) in columns (2) and (4) is 72.786 and 13.39 ***, respectively, which are far greater than the maximum value of digit. That is, the financial inclusion effects will increase with an increase in digitization, and the financial inclusive effects are sustainable and have not reached saturation. In other words, digitization has a positive impact on sustainable financial inclusion at the current stage, which supports Hypothesis 1.

Further, heterogeneous estimation results for the basic model are shown in

Table 3. Heterogeneous estimation results for basic model.

Fixed Effects Models
	(5) Cinclusion	(6) Cinclusion	(7) Lncred	(8) Lncred
i.cp_digit
1	0.954 *** (0.3620)		0.0505 (0.0319)
2	2.232 *** (0.3206)		0.143 *** (0.0289)
3	2.583 *** (0.3231)		0.234 *** (0.0292)
4	1.224 *** (0.3375)		0.191 *** (0.0302)
digit		1.375 *** (0.2063)		0.124 *** (0.0212)
cti	−0.0373 (0.0716)	0.105 (0.0669)	0.00555 (0.0058)	0.00855 (0.0054)
roa	2.961 *** (1.1250)	3.437 ** (1.6815)	0.00118 (0.0905)	0.166 (0.1371)
size	−5.369 (5.5626)	6.860 *** (1.3770)	2.599 *** (0.4481)	2.909 *** (0.1128)
inf			−0.0458 (0.0501)	0.102 (0.0785)
i.cp
2		3.844 ** (1.6750)		0.332 ** (0.1359)
3		−0.716 (2.3743)		0.145 (0.2001)
4		−4.170 (2.7873)		−0.00734 (0.2105)
cons	24.94 (21.2551)	−23.85 *** (7.8984)	−5.109 *** (1.7161)	−6.656 *** (0.6617)
fixed effects	individual	bank type	individual	bank type
N	334	334	334	334

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% ** 5%.

. Classified by the bank attribute, columns (5) and (7) are the varying coefficient estimation using individual fixed effects, while columns (6) and (8) are varying intercept, estimated by bank-type fixed effects instead of individual fixed effects due to the multiple collinearity. Robustly, the coefficients of the independent variable in all the columns, varying coefficient or varying intercept, show the positive effect of sustainable financial inclusion. Specifically, the results of varying coefficient estimation indicate that the effect of bank digitalization on sustainable financial inclusion varies with the type of banks—the effects of joint-stock banks, city commercial banks and rural commercial banks are greater than those of large commercial banks.

4.2. The Results of the Profit-Driven and Risk Aversion Mechanism Model

In the mechanism model, the first function represents the total effects of digitization on sustainable financial inclusion, which was empirically estimated in the basic model of the above sub-section. The mediator variables frontier and diversity, as the dependent variables in the second function regressed by the bank digitalization variable digit, are shown in (9) and (12) in

Table 4. Estimation results for mechanism model.

Mediation Effect Panel Model
	(9) Frontier	(10) Cinclusion	(11) Lncred	(12) Diversity	(13) Cinclusion	(14) Lncred
digit	4.226 *** (0.2145)	1.418 *** (0.3194)	0.103 *** (0.0287)	0.0325 *** (0.0076)	1.880 *** (0.2801)	0.162 *** (0.0262)
cti		−0.0496 (0.0756)	0.00479 (0.0062)		−0.00608 (0.0765)	0.0105 * (0.0062)
roa		2.748 ** (1.2272)	0.0340 (0.0989)		2.397 ** (1.1676)	−0.0244 (0.0932)
size		2.676 (5.7121)	3.726 *** (0.4601)		−3.475 (5.7496)	2.937 *** (0.4593)
inf			−0.0560 (0.0531)			−0.0392 (0.0524)
frontier		0.0595 * (0.0318)	0.00862 *** (0.0026)
diversity					2.630 *** (0.9310)	0.323 *** (0.0743)
cons	48.48 *** (1.2429)	−7.073 (22.3045)	−9.750 *** (1.7990)	2.893 *** (0.0438)	9.742 (21.4894)	−7.494 *** (1.7209)
N	348	334	334	346	332	332

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% ** 5% * 10%.

. The results show that the digitalization of banks has positive effects on frontier and diversity, indicating that digitalization can increase the bank’s possibility frontier and credit portfolio diversity, thereby verifying the existence of the path from independent variable to mediator variable.

The third function of the mechanism model measures the influence of the mediator variable on the dependent variable when the independent variable digit is controlled, thus constituting the complete path of the independent variable, mediator variable and dependent variable. Columns (10) and (11) show the positive effects of frontier on financial inclusion. Combined with the positive result of the second equation, digitalization is judged to have a positive effect on sustainable financial inclusion through the mediator variable frontier; that is, digitalization helps banks to expand the business probability frontier and then increases sustainable financial inclusion, which constitutes the profit-driven mechanism. Similarly, columns (13) and (14) show the positive coefficient of diversity, indicating that diversity can increase financial inclusion. Thus, the whole path of credit portfolio diversity is expressed in that digitalization increases the bank’s pursuit of credit portfolio diversity and then increases sustainable financial inclusion, verifying the bank’s risk aversion behavior.

Thus far, Hypotheses 2 and 3 are confirmed.

4.3. Robustness Test and Endogeneity

In the first two subsections, we examined the effects of digitization on sustainable financial inclusion and the two underlying incentive mechanisms of the bank’s profit-driven and risk aversion behavior. The results are estimated robustly by taking the methods, including the supplementary variables, the quadratic term and heterogeneity regression. First, the dependent variable financial inclusion is supplemented by the variable lncred to regress all models as an additional robustness test, showing a robust positive effect of digitalization on sustainable financial inclusion. Second, the quadratic term of digitalization is added in the model to illustrate whether the positive effect still exists or changes due to the digital externality and the saturation effect. Third, heterogeneity regression sorted by the type of bank confirms the robust results with varying coefficient and varying intercept, respectively.

However, since the independent variable digit is a bank-specific variable, it may vary with the dependent variable financial inclusion that is also defined as a bank-specific variable, which could result in the endogeneity problem. To achieve more reliable results, we select provincial-level data lndi as the instrumental variable of the independent variable digit for a further robustness test. Specifically, the banks we selected are located in 21 different provinces in China where the digital development level of each province can be expressed by the provincial PKU-DFIIC. Since PKU-DFIIC selects data from the Fintech sector, the results using the independent variable and instrument variable also reflect digital innovation and spillover as well as the absorptive capacity of banks.

We use two-way fixed effects (time fixed and individual fixed effects) to estimate the models, and the results of the robustness test are shown in

Table 5. Robust estimation results for basic model.

Fixed Effects Models
	(15) Cinclusion	(16) Cinclusion	(17) Cinclusion	(18) Lncred	(19) Lncred	(20) Lncred
lndi	17.76 *** (2.6401)	62.03 *** (17.9383)	209.0 *** (69.4534)	4.746 *** (1.4681)	1.453 *** (0.2322)	21.54 *** (4.8435)
cti	−0.0617 (0.0751)	−0.0895 (0.0788)	−0.104 (0.0785)	0.00740 (0.0064)	0.00494 (0.0062)	0.00104 (0.0061)
roa	1.533 (1.1583)	0.431 (1.2338)	1.179 (1.2718)	−0.178 * (0.1010)	−0.108 (0.0941)	−0.0330 (0.0931)
size	0.6327 (5.2929)	−10.17 (6.3540)	−13.94 ** (6.5397)	3.057 *** (0.5200)	3.340 *** (0.4330)	2.931 *** (0.4320)
lndi_lndi			−13.11 ** (5.9881)			−1.754 *** (0.4224)
inf					−0.00669 (0.0488)	0.0367 (0.0486)
cons	−86.8 *** (12.2482)	−286.7 *** (92.0512)	−684.1 *** (203.2965)	−32.26 *** (7.5335)	−15.31 *** (1.1554)	−71.19 *** (13.5071)
axis of symmetry(=6.07)			7.971			6.140
bank	yes	yes	yes	yes	yes	yes
year	no	yes	yes	yes	no	no
N	334	334	334	334	334	334

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% ** 5% * 10%.

,

Table 6. Heterogeneous robust estimation results for basic model.

Fixed Effects Models
	(21) Cinclusion	(22) Cinclusion	(23) Cinclusion	(24) Lncred	(25) Lncred	(26) Lncred
i.cp_lndi
1	46.75 *** (17.6806)	9.497 ** (3.7348)		2.763 * (1.4420)	0.371 (0.3128)
2	59.24 *** (16.9565)	22.77 *** (3.1114)		3.651 *** (1.3830)	1.330 *** (0.2639)
3	63.64 *** (17.0344)	25.35 *** (3.0721)		4.627 *** (1.3893)	2.239 *** (0.2571)
4	48.55 *** (16.9962)	11.40 *** (3.2585)		4.127 *** (1.3862)	1.779 *** (0.2731)
lndi			16.88 *** (5.0877)			1.243 *** (0.4072)
cti	−0.100 (0.0739)	−0.0609 (0.0710)	0.0708 (0.0676)	0.00577 (0.0060)	0.00510 (0.0058)	0.00541 (0.0054)
roa	1.354 (1.1652)	2.189 ** (1.1048)	3.689 ** (1.7441)	−0.106 (0.0950)	−0.0605 (0.0886)	0.219 (0.1396)
size	−17.82 *** (6.2604)	−3.891 (5.1791)	7.761 *** (1.3661)	2.067 *** (0.5106)	2.615 *** (0.4190)	3.000 *** (0.1093)
inf					0.0134 (0.0456)
i.cp
2			4.411 *** (1.6908)			0.408 *** (0.1353)
3			0.000494 (2.4401)			0.122 (0.1953)
4			−4.405 (2.8290)			−0.171 (0.2264)
cons	−235.9 *** (87.7361)	−84.31 *** (11.6583)	−115.5 *** (27.8245)	−25.18 *** (7.1557)	−14.33 *** (1.0872)	−13.29 *** (2.2270)
bank type	no	no	yes	no	no	yes
bank	yes	yes	no	yes	yes	no
year	yes	no	yes	yes	no	yes
N	334	334	334	334	334	334

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% ** 5% * 10%.

and

Table 7. Robust estimation results for mechanism model.

Mediation Effect Panel Model
	(27) Frontier	(28) Cinclusion	(29) Lncred	(30) Diversity	(31) Cinclusion	(32) Lncred
lndi	42.85 *** (2.2448)	68.26 *** (17.8967)	1.146 *** (0.2498)	0.335 *** (0.0780)	57.84 *** (17.8312)	1.591 *** (0.2262)
cti		−0.103 (0.0781)	0.00432 (0.0061)		−0.0523 (0.0794)	0.0100 (0.0061)
roa		1.236 (1.2576)	−0.00726 (0.0983)		0.848 (1.2318)	−0.0702 (0.0913)
size		−9.434 (6.2902)	3.538 *** (0.4314)		−13.42 ** (6.3973)	2.889 *** (0.4291)
inf			−0.0331 (0.0489)			0.0461 (0.0474)
frontier		0.0847 *** (0.0320)	0.00773 *** (0.0025)
diversity					2.416 *** (0.9144)	0.321 *** (0.0728)
cons	172.4 *** (12.8011)	−329.4 *** (92.4612)	−14.91 *** (1.1455)	1.162 *** (0.4449)	−260.2 *** (91.6680)	−15.51 *** (1.1155)
bank	yes	yes	yes	yes	yes	yes
year	no	yes	no	no	yes	no
N	348	334	334	346	332	332

Note: Numbers in parenthesis are standard errors. Significance level: *** 1% ** 5%.

. Appendix A compares the scatter plot and fitting line of using the independent variable and its instrument variable. Just like the results estimated using the independent variable digit above, the coefficients of instrument variable lndi are all significantly positive, which indicates that the digitalization of banks is robustly positive to financial inclusion. Similarly, the quadratic terms in columns (17) and (20), measuring the digital network externality and saturation effects, show the negative effects and reject the assumption that the quadratic term coefficient is zero at the level of 5% and 1% significance. The calculated values of the axis of symmetry (lndi value) in the basic robustness model is 7.971 ** and 6.140 ***, respectively, which stands the right of the maximum value of lndi 6.07. Thus, the effect of digitalization on financial inclusion is robustly sustainable. In addition to these similar results, the effects of varying coefficient estimation in Table 6 show that large commercial banks have a smaller slope to other types of banks as well. In addition, we use single individual fixed effects to estimate the model in column (15). By comparing the models with and without time fixed effects, the result indicates that the effects of digitalization on sustainable financial inclusion are much larger when the time effect is controlled.

The results of the mechanism model are also robust when using the provincial-level variable lndi. The instrument variable is significantly positive to the profit-driven mediator variable frontier and risk aversion mediator variable diversity, and both the profit-driven and risk aversion behaviors of banks have significant positive effects on sustainable financial inclusion. These are consistent with the results of the model using the variable digit and, thereby, the theoretical expectation.

4.4. Discussion

The positive effects of digitalization on sustainable financial inclusion were proved in this paper by constructing basic and mechanism models. We exploit a nonlinear attempt, heterogeneous estimation, as well as the supplementary variable and instrument variable methods, and we explore two underlying incentive mechanisms of profit driven and risk aversion to make the results more convincing. The positive results of digital-based financial inclusion are consistent with most of the research [13,15], especially in developing countries [60,61]. The nonlinear attempt indicates that the digital externality and saturation effects [36,47] are inevitable but can be offset by continuous updates in data resources. The heterogeneous estimation results suggest that large commercial banks with the highest digital development have a less intense effect on financial inclusion. The probable cause is that large commercial banks have a large proportion of head customers so that the proportion of financial inclusion in the original customer stock is small, or that large commercial banks with a higher level of average digitalization have diminishing marginal effects. The two profit-driven and risk aversion incentive mechanisms can bring about sustainable financial inclusion, which are not commonly utilized in the literature due to technological limits.

5. Conclusions

In this paper, we contribute to the effects of digital development from banks on sustainable financial inclusion and examine how digitalization affects sustainable financial inclusion through banks’ behavior. The basic model results show a robust positive effect of digitalization on sustainable financial inclusion, and the mechanism model indicates the significant profit-driven and risk aversion behavior of banks.

We quote the case of China to analyze the changes in digital innovation and spillover on banks and the conditions of banks increasing sustainable financial inclusion. Banks can benefit from digitalization to expand the business probability frontier for sustainable profits and make better credit portfolio decisions to share risk. Digital adoption is renewing the Pareto Principle with the long-tail market and making MPT more applicable, which increases the theoretical basis for banks to foster sustainable financial inclusion by leveraging digitalization.

According to the theoretical framework, we construct the model of bank digitalization affecting sustainable financial inclusion. In order to make the results robust, the methods of quadratic term of digitalization, heterogeneous estimation and supplementary dependent variable are adopted. Considering that externalities may lead to nonlinearity, the quadratic term of digitalization is induced to support the result that the digitalization of banks can increase sustainable financial inclusion. Heterogeneous estimation further proves that different types of banks can utilize digitalization to increase sustainable financial inclusion. The supplementary dependent variable is applied as the dependent variable both in the basic model and mechanism model for robustness.

Two underlying incentive mechanisms of digitalization, affecting sustainable financial inclusion in banks, focus on the profit-driven and risk aversion behavior of banks. The profit-driven mechanism is digitalization can stimulate banks to expand the business probability frontier and provide credit liquidity to disadvantaged segments of society, especially marginal people with good credit. The risk aversion behavior mechanism is banks use of digital technologies to make better portfolio decisions, avoiding credit risk caused by relatively centralized and single loan objects. Both incentive mechanisms, as the mediators of the effects of banks’ digitalization on sustainable financial inclusion, were shown to help further increase sustainable financial inclusion.

In addition, endogeneity problems are dealt with using a digitalization instrument variable—the provincial-level data of PKU-DFIIC. The results of the basic model and mechanism model using the instrument variable are consistent with those using the bank-level variable, which further proves the effects of bank digitalization on sustainable financial inclusion.

The findings of this paper conform to the trend of the development of digitalization and provide theoretical and empirical support for banks to build digitalization and realize sustainable financial inclusion. Banks and niches, two entities that have long seemed to be opposed to each other, have moved closer together due to digitalization. The COVID-19 crisis laid the reality that digital connectivity enables countries and consumers to access critical resources and services, which is essential for economic recovery. As the ITU Secretary-General Houlin Zhao proposed at the 2020 G20 Digital Economy Ministerial Meeting, the policy options to support the digitalization of business models during COVID-19 will be an invaluable resource to all nations [62]. Along with the SDGs of the UN, World Bank and ITU, China is committed to digital financial inclusion and the digital economy with ICT applications in financial services.

The case of China indicates a “triple-win” effect for the sustainable development of banks, vulnerable groups and the whole of society, which can be used as a reference for other countries or regions (a simple feedback test of financial inclusion to banks as the sufficient, but not necessary, part is depicted in Appendix B). However, this does not mean that sustainable financial inclusion can be achieved by only increasing digital construction. Some other reasons could offset or hinder this result, for example, a country that has a high digital level but has not increased financial inclusion. This could be related to the type and structure of clients and their acceptance of digital credit services. Thus, further studies can be undertaken to solve more detailed problems.