Although the rapid development of global fintech has received global attention, there is no standard definition of fintech (Chen et al., 2019; Cheng & Qu, 2020). Schueffel (2016) reviewed scholarly articles referencing fintech and defined fintech as a new financial industry. The Financial Stability Board (FSB) defined fintech as technologically enabled financial innovation. Subsequently, Leong and Sung (2018) defined fintech as a cross-disciplinary subject combining finance, technology management and innovation management. For Chen et al. (2019), fintech was refered to “the set of recently developed digital technologies that have been applied - or that will likely be applied in the future - to financial services”. Thakor (2020) claimed that fintech is the use of technology to provide new and improved financial services. Wójcik (2021) defines fintech as “a set of innovations and an economic sector that focus on the application of recently developed digital technologies to financial services”. In China, the People's Bank of China (PBOC) officially issued its Financial Technology Development Plan (2019–2021) in September 2019, which defined fintech as the application of emerging technologies.

From the various definitions of fintech, it is evident that the term refers to a broad, complex, and rich phenomenon (Iman, 2020). Some authors focus on the financial services industry that applies technology (Schueffel, 2016; Thakor, 2020; Wójcik, 2021), while others focus on technological innovations (Chen et al., 2019). This paper adopts the fintech definition given by Leong and Sung (2018) since it refers to the combination of finance, technology and innovation, which offers a greater level of detail.

Regarding to the elements of fintech, Ng and Kwok (2017) classify fintech into four categories: efficient payment processes, robo-advisors, peer-to-peer load and deposit platforms, and crowd funding. Moreover, Lee (2019) note that the fintech involves banking technology, pay technology, regression technology, wealth technology, lend technology, and insurance technology. Similar to Lee (2019), Li and Xu (2021) claim that the key technologies of fintech include internet technology, big data, artificial intelligence, distributed technology, and security technology. Focusing on many different contexts of fintech, Lee and Shin (2018) identify five elements of fintech ecosystems: fintech start-ups, technology developers, governments, financial customers, and traditional financial institutions.

To show the level of fintech development, Findexable constructed the Global fintech Index, an overall score indicating the quantity and quality of companies in the fintech ecosystem. The Global fintech Index, which ranks 65 countries, more than 230 cities, and 7000 companies worldwide, is the world's first global ranking of fintech ecosystems. Although the Global fintech Index is widely adopted (Lyons et al., 2021; Wójcik, 2021), it partially covers only the political and economic environment of fintech development (Pauliukevičienė & Stankevičienė, 2021).

In China, the fintech industry has rapidly grown since 2013. Cheng and Qu (2020) applied web crawler technology and word frequency analysis to construct a bank fintech index, which measures the development of fintech in the banking industry. Subsequently, Lee et al. (2021) used principal component analysis to construct a financial technology index composed of four dimensions: (1) credit, deposit and capital-raising services, (2) payments, clearing and settlement services, (3) investment management services, and (4) market support services. The fintech index developed by Lee et al. (2021) only includes the dimension of financial services. Wang et al. (2021) applied factor analysis to comprehensively construct a Chinese fintech index based on the following five attributes: big data, artificial intelligence, distributed technology, technology interconnectedness, and technology security. The fintech index developed by Wang et al. (2021) only focuses on the technology dimension. The fintech index of Deng et al. (2021) was derived from the municipal digital financial index compiled by the Peking University Digital Finance Research Center, which was constructed from the demand perspective, rather than the industry supply perspective (Lee et al., 2021).

From the analysis of the measurement of the fintech index in China, it is evident that the measurement of this index is a complex process. No standard fintech index has been developed in China. Some authors focus on the financial service dimension, while others focus on the technology dimension. According to the definition of fintech adopted in this paper, we evaluate levels of fintech development across multiple dimensions, including the usage depth of digital finance, the level of digitization, and the level of technology innovation, thus offering more detail on finance services, financial innovation and technology development.

Financial technology development plays a positive role in poverty alleviation through numerous channels (Mushtaq & Bruneau, 2019). Financial inclusion has been a very popular channel between fintech and poverty alleviation. Fintech has accelerated increases in financial inclusion because it has the capability to reach financially vulnerable populations (Muneeza et al., 2018). Fintech can offer inexpensive, convenient, and secure financial services to people; thus, it is beneficial to the development of inclusive finance.

In China, the rapid growth of fintech has significantly influenced the development of inclusive finance as one of its key emerging drivers. For instance, mobile banking, third-party payments, Yu'ebao and crowd funding are considered important ways to promote financial inclusion due to their accessibility, affordability, and safety. Since the G20 summit in 2010, financial inclusion has been widely perceived as a dynamic tool in the context of poverty alleviation (Mohammed et al., 2017; Omar & Inaba, 2020). Existing literature at the macro level and household level has generally indicated that the level of inclusive financial development in China has had a significant positive impact on the income poverty (Li, 2019; Yang & Fu, 2019; Peng, 2019; Wang, 2020).

Few studies have focused on whether fintech impacts poverty in China. Wang and He (2020) studied the effects of fintech on farmers’ vulnerability to poverty in China and found that farmers’ use of fintech has positive effects on poverty reduction by improving financial access, disseminating information, promoting social linkages, and promoting rural e-commerce. Zhang et al. (2020) investigated the growth and distribution effects of fintech development, using household survey data from China. The authors argued that fintech development in China has a positive effect in narrowing the urban-rural income gap by expanding access to financial services. It is important to note that academic studies on fintech and poverty in China have focused on the microlevel household perspective (Wang & He, 2020; Zhang et al., 2020). Moreover, Appiah-Otoo and Song (2021) used a panel of data on 31 provinces in China for 2011 to 2017 to examine the direct and indirect effects of fintech and its submeasures of third-party payment and credit on poverty measured by household per capita consumption. The authors confirm that fintech complements economic growth and financial development in reducing poverty in China. Appiah-Otoo and Song (2021) focused on macro provincial perspective to examine the effect of fintech development on poverty alleviation, but little attention has been paid to fintech index construction problems and heterogeneity problems of poverty. Thus, this paper fills the above knowledge gaps by proposing a quantile regression model to explore the effect of fintech development on poverty reduction in China.

To measure fintech development, we adopt a normalization and aggregation methodology (Lyons et al., 2021) to calculate a fintech index for each province in China. This methodology involves two steps. In the first step, we use the min-max method to normalize each dimension for each province using the following equation:

In the second step, the fintech index is calculated by aggregating the normalized scores for the three dimensions using the following geometric mean approach:

To address the potential for unobserved heterogeneity among poverty-stricken regions in China (Ren et al., 2017; Gustafsson, 2009) and the variable selection problem related to high-dimensional dataset, we propose a generalized elastic net Lp-norm support vector quantile regression (GLpSVQR) to examine how fintech affects poverty alleviation in China using a province-level dataset covering 2011 to 2020.

Suppose that Y=(y1,⋯yN)aptaradummyentityaptaradummyx27; is a response variable, X is a known N×K design matrix of covariates, xi=(xi1,⋯xiK)aptaradummyentityaptaradummyx27; is a K-dimensional explanatory variable, and β0 and β=(β1,⋯βK)aptaradummyentityaptaradummyx27; are unknown parameters that need to be estimated. Consider the following linear regression model:

Quantile regression (QR) (Koenker & Bassett, 1978) is an effective method to address regression problems involving heterogeneity, as it estimates β0 and β in a regression function (1) based on a full range of conditional quantile functions (Shim & Lee, 2010). To demonstrate a nonparametric version of QR, Takeuchi et al. (2006) introduced the spirit of support vector regression (SVR) (Drucker et al., 1997) into the QR framework. Subsequently, support vector quantile regression (SVQR) was proposed (Hu et al., 2012; Anand et al., 2020). SVQR maintains the advantages of quantile regression and support vector regression but it lacks the sparseness property related to the variable selection problem for high-dimensional data. Therefore, Ye et al. (2021) proposed an Lp-norm support vector quantile regression (Lp–SVQR) (0 < p < 1).

Lp–SVQR simultaneously realizes estimation and variable selection. However, Lp–SVQR solutions sometimes suffer from the singularity problem (Li et al., 2020). Since minimizing the L2-norm regularization solves the singularity problem, we introduce L2-norm regularization into the Lp–SVQR framework and propose a generalized elastic net Lp-norm support vector quantile regression (GLpSVQR) as follows:

The combination of Lp-norm and L2-norm, that is, λ∥β∥pp+∥β∥22, forms a generalized elastic net that is similar to that of Li et al. (2020), which was used for classification. Therefore, GLpSVQR maintains the advantages of Lp-SVQR and SVQR. GLpSVQR not only has the variable selection ability to solve estimation problems with high-dimensional dataset, but also prevents overfitting. Moreover, GLpSVQR can effectively address heterogeneity problems in the estimation process since quantile estimators are a class of empirical “location” measures for the dependent variable.

Regression problem (2) is not differentiable due to the Lp-norm regularization term. To make it smooth, we introduce the upper-bound variable υ=([υ]1,⋯,[υ]K)T; thus, problem (2) can be written as follows:

Problem (3) is differentiable and can be solved using a convergent successive linear algorithm (SLA). The SLA process starts with a random initial point β^00,β^0,υ^0,ξ^0,ξ^*0, and at the k-th iteration, k=1,2,…,β^0k,β^k,υ^k,ξ^k,ξ^*k is produced by solving the following problem:

To evaluate the regression results of GLpSVQR, let n be the number of samples, y^i be the predicted value of yi, and y¯=∑i=1nyi/n be the average value of y1,…,yn. We use the normalized mean squared error (NMSE) and R2 to evaluate the regression results of GLpSVQR. These values are defined as follows:

In addition, we use the proportion of nonzero estimated coefficients (P1) to evaluate the variable selection ability of GLpSVQR (Ye et al., 2021). The optimal values of these parameters for our experiments are determined using the grid search method.

Fintech can be evaluated across different dimensions, including the usage depth of digital finance, the level of digitization, and the level of technology. Our data on the usage depth of digital finance and the digitization level of financial services come from Guo et al. (2020). To collect data on the finance technology level, we use web crawler technology and word frequency analysis to search fintech-related keywords in the financial operation reports of every province in China. The usage depth of digital finance refers to payments, insurance, monetary funds, investments, credit, and credit investigation. The level of digitization encompasses mobile usage level, affordability, credit level, and facilitation. The finance technological level covers artificial intelligence, distributed technology, and internet technology. The specifications of variables, which reflect fintech activities, are listed in Table 1. We use a normalization and aggregation methodology (Lyons et al., 2021) to calculate a fintech index for each province in China.

We further use the K-means clustering algorithm (Hartigan & Wong, 1979) to aggregate the Chinese provincial fintech indices into 3 clusters (excluding Hong Kong, Macao, and Taiwan): a high-development region, an intermediate-development region, and a low-development region. The provinces included in each region are listed in Table 2. The trend of the fintech index belonging to the same cluster is shown in Figs. 1-3. From Fig. 1, we can see that the fintech high-development region includes Beijing, Zhejiang, and Shanghai. The fintech level of this region is approximately 200. The intermediate-development region includes 16 provinces, such as Jiangsu, Fujian, and Guangdong. The fintech level of this region is approximately 170, as shown in Fig. 2. The low-development region includes 12 provinces, such as Hebei, Guizhou, Yunnan, and Gansu. In Fig. 3, we find that the fintech level of this region is approximately 160. Fig. 4 shows fintech mean level of each cluster. From the figure, we can see that the clustering centers of different groups have different trend. Moreover, the gap of fintech level among different clusters has become larger in recent years. Figs. 1-4 show the heterogeneity in fintech level among Chinese provincial regions.

Fig. 1.

Fintech index of the high-development region.

(0.15MB).

Fig. 2.

Fintech index of the intermediate-development region.

(0.25MB).

Fig. 3.

Fintech index of the low-development region.

(0.24MB).

Fig. 4.

Mean level of fintech index in each cluster.

(0.21MB).

To analyze the impact of fintech on poverty alleviation in China, we take poverty as our dependent variable. Poverty is a diverse and dynamic concept (Wang et al., 2021). This paper adopts the poverty definition given by Ravallion (1998) who defined poverty as inadequate income to gain essential necessities for survival. According to the definition of poverty, we measure poverty by the poverty headcount ration data (H), which refers to the ratio of the people below the poverty line (q) to the total registered population (n) (Liu et al., 2017). Our independent variables include: fintech, measured by the proposed fintech index; income inequality, measured as the proportion of urban per capita disposable income to rural per capita disposable income; economic level, measured as the logarithmic form of per capita GDP; trade, measured as the sum of exports and imports as a share of GDP; education, measured as the education expenditures as a share of GDP; fixed asset investments, measured as fixed asset investments as a share of GDP; foreign direct investment, measured as FDI as a share of GDP; and public budget expenditures, measured as public budget expenditure as a share of GDP. The data of these variables except for the fintech index come from China's National Bureau of Statistics. All annual data range from 2011 to 2020.

We adopt a meta-analysis to test whether poverty and each of our independent variables are heterogeneous. Typically, in meta-analyses, heterogeneity is assessed with the I2 index (Higgins & Thompson, 2002). All the heterogeneity tests of poverty among the provincial regions are implemented in Stata 15.1. I2 values of 25%, 50%, and 75% are interpreted as representing low, moderate, and high levels of heterogeneity. The heterogeneity test results corresponding to each year are similar. Here, we only present the heterogeneity test results corresponding to 2011. The education, fixed asset investment, foreign direct investment, public budget expenditure, and trade I2 values are 99.8%, 97.8%, 100%, 99.6%, and 99.9%, respectively. These values reflect great heterogeneity among the examined provincial regions. The income inequality I2 value is 38.2%, indicating the low level of heterogeneity in relation to this factor among the provincial regions. The I2 values of fintech index and the economy are 0, indicating that there is no heterogeneity among the provincial regions in relation to these factors.

Based on the presence of heterogeneity among the provincial regions, we propose GLpSVQR to examine the potentially differential effects of fintech on poverty alleviation in China. The quantile parameter is chosen from the set {0.25,0.50,0.75}. Fig. 5 provides the regression coefficients of the fintech index, when τ = 0.25, 0.5, and 0.75. Other quantile regression results at different quantile levels are listed in Tables 3-5. The variable selection results are shown in Fig. 6. Figs. 7-8 show the empirical NMSE and R2 results at different quantile levels.

Fig. 5.

Coefficients of the fintech index at different quantile levels.

(0.18MB).

Fig. 6.

The proportion of nonzero estimated coefficients (P1) at different quantile levels.

(0.19MB).

Fig. 7.

R2 at different quantile levels.

(0.21MB).

Fig. 8.

NMSE at different quantile levels.

(0.22MB).

In Fig. 5, the red regression coefficient curve with periodicity, representing the high-income provinces, fluctuates at around −0.1. The green regression coefficient curve with periodicity, representing the moderate-income provinces, fluctuates at around −0.15. The black regression coefficient curve, representing the low-income provinces, fluctuates at around −0.2. The coefficients of fintech at different quantile levels are negative, indicating that fintech reduces poverty in all provinces of China. The estimated coefficients suggest that one-point increase in fintech adoption in a province results in a 10–20% decline in poverty reduction. Moreover, from the trends of these three regression coefficient curves, we find that the absolute values of the regression coefficient have decreased in recent years.

The findings obtained from Fig. 5 have several meaningful interpretations. First, the negative coefficients of fintech suggest that the fintech development has positive effects on poverty reduction. Second, the impact of fintech development on poverty reduction among provincial regions has decreased in the last three years. Third, different fintech quantile levels have different impacts on poverty alleviation, which indicates that the effects of fintech on poverty reduction across regions are heterogeneous. Finally, the poverty-reducing impact is stronger in low-income provinces than in high-income provinces. In other words, fintech in less developed provinces plays a more significant role in alleviating poverty than in high-income provinces since fintech development provides easier access to financial services, and reduces transaction costs, information asymmetry, and household risk (Zhang et al., 2019; Wang & He, 2020; Appiah-Otoo & Song, 2021; Giaretta & Chesini, 2021).

Tables 3-5 show that income inequality is selected in every quantile regression process. The positive coefficients and t-statistic value of income inequality indicate that the income inequality significantly influences poverty. These results imply that income inequality worsens the poverty situation in China. Moreover, the coefficients and t-statistic value in Tables 3-5 also indicate that economic level, trade, education, fixed asset investments, foreign direct investment, and public budget expenditures insignificantly influence poverty in most regression cases. Fig. 6 shows the proportion of nonzero estimated coefficients (P1) to evaluate the variable selection ability of GLpSVQR, which indicates that GLpSVQR selects a few variables in every regression process. Fig. 7 and Fig. 8 show the regression results of R2 and NMSE, respectively. From Figs. 6-8, we find that GLpSVQR selects a few variables in every regression process with large R2 values and small NMSE values. Thus, the very few variables selected by GLpSVQR capture useful statistical information from the samples which indicate that the proposed GLpSVQR is an effective method for solving the representative variable selection problem.

To test the robustness of our results, we use the poverty gap (PG) as an alternative measure of poverty. The poverty gap is defined as follows:

Where H is the poverty headcount ration data, and I is the gap ratio. I is defined as follows:

The results of the proposed support vector quantile regression are reported in Fig. 9 and Tables 6-8. Fig. 9 shows the coefficients of the fintech index when τ=0.25,0.5, and 0.75. The red regression curve with periodicity, which represents the high-income provinces, fluctuates around -0.04. The black regression coefficient curve with periodicity, which represents the low-income provinces, fluctuates around -0.1. These results are consistent with those in Fig. 5, indicating that an increase in the development of fintech is associated with a decrease in poverty at different quantile levels. In summary, these results indicate that in each region, the development of fintech significantly reduces poverty. Tables 6-8 present the results of the proposed quantile regression when τ=0.25,0.5, and 0.75. The variable selection results of GLpSVQR in Tables 6-8 are in line with the results in Tables 3-5. The small values of NMSE and large R2 also indicate that the relevant statistical information is well captured by GLpSVQR with a few selected variables.

Fig. 9.

Coefficients of the fintech index at different quantile levels.

(0.2MB).