The impact of digital finance on household consumption: The impact of digital finance on household consumption: Evidence from China Evidence from China

Using panel data from the China Household Finance Survey (CHFS) in 2013, 2015, and 2017 and the digital inclusive finance index developed by Peking University, this study examined impacts of the digital inclusive finance on household consumption and explored its mechanisms. Results suggest that the digital inclusive finance could promote households consumption. A heterogeneity analysis showed that households with fewer assets, lower income, less financial literacy and in third-and fourth-tier cities experienced larger facilitating effects of digital finance on consumption compared to their counterparts. For consumption categories, digital finance was positively correlated with food, clothing, house maintenance, medical care, and education and entertainment expenditures. In terms of consumption structure, digital finance mainly promoted the recurring household expenditures rather than the non-recurring expenditures. Further analyses based on the mediating model found that online shopping, digital payment, obtainment of online credit, purchase of financing products on the internet and business insurance, were the main mediating variables through which digital finance affected household consumption.


1.Introduction
Since China's economy entered the new normal, consumption has been gradually becoming an important driving force for economic development. In recent years, the government has been working on the expansion of residents' consumption demand. However, Chinese households' consumption behavior indicates that the consumption demand is still low. The household consumption rate has declined from 47. 5% in 2000 to 35. 6% in 2010, which is far below the world average 1 . Inadequate consumption has become an important restraining factor for economic transformation and sustainable development in China. How to promote the growth of household consumption has become a major subject of concern among both policy makers and the academia. Existing research on inadequate consumption has suggested that liquidity constraints (Kuijs, 2005), imperfect security systems (Meng, 2003), and income inequality (Schmidt-Hebbel and Serven, 2000) are important enforcing factors for inadequate consumption.
Therefore, financial development can relieve consumers from liquidity constraints through reasonable and efficient resource allocation and realization of inter-temporal smoothing of consumption, thus increasing consumption demand (Levchenko,2005).
In recent years, with the deep integration between Internet technology and finance, the new digital finance model supported by information technology is gradually becoming an indispensable part of China's financial system as it can help decrease the degree of information asymmetry, reduce transaction costs, improve availability of financial services and optimize resource allocation in the financial market. According to the report released by a research group from the Institute of Digital Finance of Peking University in 2016, the digital inclusive finance index increased from 40 in 2011 to 220 in 2015. Digital finance has developed rapidly in just a few years. Meanwhile, the household consumption rate has picked up slightly in recent years, reaching 39.3% in 2016. Therefore, is the rapid development of digital finance able to significantly influence household consumption? Which consumer groups are most affected? How about the influence path? This paper examines these questions. Answers to these questions will not only help to understand the impacts of digital finance on China's economic development at 1 According to the World Development Indicators database of World Bank, the world average consumption rate is 57. 9% in 2010. household level, but will also provide useful information on the growth of China's household consumption and a basis for improving relevant policies.
Digital finance including online loans, mobile payment, Internet finance, Internet insurance and other kinds of innovative products may impact household consumption from various aspects.
First of all, online credit makes it possible to match the financial demand side with the supply side where the parties may be geographically disparate (Pierrakis and Collins, 2014).
Consumption credit services represented by Alipay, cash loans and many kinds of P2P platforms and other new types of financial models have expanded the channels for obtaining funds, changed the traditional mode of credit services, lowered the bar for financial services and improved the borrowing convenience, thus relieving households from the constraints of credit to a certain extent. Then, relieving liquidity constraints promotes household consumption. Secondly, the rapidly developed Internet financing market represented by Yu'E Bao 1 has expanded the channels for people to invest using small funds, increased the rate of return on investment and promoted the growth of household wealth, thus increasing household consumption. Meanwhile, rapidly developed digital payment platforms have greatly reduced the transaction and time costs of financial services, improved the efficiency of payment and transfer for household consumption. Additionally, the development of digital finance has not only promoted the upgrade of the service mode of traditional insurance companies, but also led to the emergence of Internet insurance companies such as Zhong'an Insurance, thus breaking the geographical barriers of the former offline outlet mode and improving insurance accessibility. Meanwhile, the application of big data technology has reduced operating costs, which may encourage residents to purchase insurance, improve residents' social security, and reduce uncertainty losses, thus increasing consumption.
This study uses panel data from the China Household Finance Survey in 2013, 2015 and 2017, and the digital inclusive finance index developed by Peking University to examine the impacts of digital finance on household consumption and further explore its mechanisms. This paper also chooses appropriate instrumental variable to solve the endogenous problem of digital finance.
1 Yu'E Bao is an internet financing product owned by Ant Financial Services Group.
Results suggest that digital finance can significantly promote household consumption, especially for households with fewer assets, lower income, and less financial literacy and in third-and fourth-tier cities, compared to their corresponding counterparts. The results of a further analysis on the influencing mechanism imply that digital finance has promoted household consumption mainly through online shopping, digital payment, obtaining loans via the Internet, purchasing financing products on the Internet, and buying commercial insurance.
The main contributions of this paper include the following. First, this paper examined Chinese families' consumption from the perspective of digital finance development, relying on the data from a nationwide large-scale household survey and the digital inclusive finance index. It has not only deepened the discussion about the problem of inadequate consumption of Chinese families, but also enriched literature relating to digital finance. Secondly, this paper reported nuanced results regarding heterogeneous impacts of digital finance on household consumption in terms of consumption structure, family characteristics, and geographical features. Thirdly, this paper made an important addition to existing literature by examining the path of influence of digital finance on household consumption using the mediating model to examine possible mechanisms of how digital finance affects household consumption through consumption channels, smoothing effects, and wealth growth effect from the aspects of online shopping, online payment, obtainment of Internet loans, and purchase of financing products on the Internet.
The remainder of this paper is arranged as follows. Section 2 reviews related literature. Section 3 introduces the data, variables, and model. Section 4 presents the results. Section 5 concludes.

Literature Review
China's household consumption demand has been low for a long time. Scholars have explained it from various perspectives. The first explanation assumes that the households are facing liquidity constraints due to the underdevelopment of the financial market and forced consumption reduction based on the theory of liquidity constraints (Kuijs, 2005;Aziz and Cui, 2007); the second explanation attributes this problem to demographic structure factors based on the life cycle hypothesis (Modigliani and Cao, 2004;Curtis et al., 2015); the third explanation assumes that the imperfect medical care, endowment, education and housing systems have intensified the uncertainty of residents for their future, thus strengthening the precautionary saving motivation and reducing resident consumption based on the theory of precautionary saving (Meng, 2003;Chamon and Prasad, 2010); the fourth explanation assumes that income inequality is an important reason for inadequate consumption from the perspective of income distribution (Schmidt-Hebbel and Serven, 2000;Jin et al.,2011); the fifth explanation involves cultural traditions and consumption habits (Modigliani and Cao, 2004); while the sixth explanation is the hypothesis of competitive saving (Wei and Zhang, 2011). Additionally, debts are also an important factor that affects consumption (Dynan et al., 2012;Scholnick, 2013). Many scholars had explored factors that impacted consumption growth. In terms of financial development, according to the theoretical analysis, it was believed the expansion of consumption credit services could relieve residents from liquidity constraints, thus facilitating consumption (Cochrane, 1991) and the development of the financial market could promote consumption growth (Bayoumi, 1993;Levchenko, 2005). Empirical research found that residents living in an area where the financial market was poorly developed were facing more severe liquidity constraints, whereas residents living in an area where the financial market was well developed were able to ease liquidity constraints and smooth consumption through consumption credit services (Jappelli and Pagano, 1989). Research by Ludvigson (1999) indicated that household consumption was positively correlated with consumption credit services when the qualification for consumption credit services was loosened. And the research by Karlan and Zinman (2010), and Dupas and Robinson (2013) indicated that the income and consumption of a person with low income would be increased if she sets up an account in a financial institution and used it frequently. In terms of assets and wealth, different types of assets have different impacts on household consumption and the mechanisms are also different (Carroll et al., 2001). According to the life cycle hypothesis (Ando and Modigliani, 1963), the higher the household asset level, the higher the consumption level. In terms of insurance, commercial insurance can reduce residents' uncertain expenditure in the future to some extent; purchase of insurance can help residents maintain a healthy consumption level and increase the average consumption propensity of the whole society (Arrow, 1963). Engen and Gruber (2001) proved that insurance could lower the saving ratio based on the variation of insurance coverage resulting from policy changes. Zhao (2019) also found that health insurance could increase household daily consumption. Besides, Kang (2019) found that social networks could promote household consumption.
Existing research on digital finance focuses on its impacts on the economy, the traditional financial market, enterprise financing, and household economics and finance. In terms of the economy, research suggests that digital inclusive finance can help improve residents' income, lower poverty rates, reduce the degrees of income inequality, and narrow down the gap between urban and rural areas (Sarma and Pais, 2011;Anand and Chhikara, 2013). For the traditional financial market, the development of digital finance will transform traditional financial departments, improve the quality and diversity of banking services, and increase the efficiency of financial services (Berger, 2003;Cortina and Schmukler, 2018). In terms of financing, research indicates that big data-based risk evaluation can help save transaction cost and decrease the degree of information asymmetry, thus helping small-and micro-businesses secure financing (Moenninghoff and Wieandt, 2013). In terms of household economics and finance, a research by Beck et al. (2018) found that mobile payments could help improve entrepreneurship execution and decrease the degree of information asymmetry, thus improving entrepreneurial performance. Grossman and Tarazi (2014) found that digital finance was helpful for peasant households in It can be seen from the above review that, as an important component of the financial market, digital finance has infiltrated all aspects of daily life. Its innovative development in online credit, Internet financing, Internet insurance, mobile payment, and credit investigation can help improve the penetrability of financial services, improve the availability of financial services to residents, relieve residents from liquidity constraints, promote income growth, facilitate residents' living consumption, thus likely promoting household consumption. However, at present, little research has examined the impacts of digital finance on consumption. Therefore, this study examined the potential impact of digital finance on household consumption based on several aspects of digital finance, and explored its influence path.

Data
The household data used in this paper was obtained from the China Household Finance Meanwhile, the consumption data, assets, and income were winsorized by eliminating samples among the top 1‰ and the bottom 1‰. Considering the abnormal fluctuations of household consumption and income, samples with household consumption variation rate (defined by the household consumption variation rate for the current year compared with that in the prior year), and household income variation rate (defined by the household income variation rate for the current year compared with that in the prior year) lower than 0 and higher than 10 were excluded.
In addition, samples with missing values for relevant variables were excluded.

Variables
Per capita household expenditure and household consumption rate were used as dependent variables in this paper to measure the level of household consumption. Per capita household expenditure was defined as the value obtained by dividing the aggregate household expenditure by the number of family members. Household consumption rate was defined as the ratio obtained by dividing the aggregate household expenditure by the disposable household income.
CHFS has kept a detailed record about household consumption, including expenditures for food, clothing, daily necessities and housekeeping services, house maintenance, transportation and communication, medical care, entertainment, and education, etc. Considering the possible non-normality of per capita household expenditure, these variables were transferred to logarithms in the regression. This paper mainly used the data at municipal level for regression analyses and used county level data for the robustness check. In addition, in the regression analyses, the digital finance development index that lagged two periods were adopted and all indexes were divided by 100.
Since existing literatures listed multiple factors that impacted household consumption (Carroll, 1994;Attanasio and Weber, 1995;Zhao, 2019), the following control variables were used: household demographic characteristics namely, age and the square of age considering the possible non-linear influence, sex, marital status, education years, health condition and risk attitude of the householder; household characteristics such as family size, children's dependency ratio, and the elderly's dependency ratio. Household resource variables including household assets and income, considering the possible non-linear influence, the assets and income were transformed to logarithms. Economic development variables include per capita GDP and financial development level measured by the ratio of outstanding loans in RMB of financial institutions to GDP of the province where the family was located. Additionally, dummy variables of provinces were included to control provincial fixed effects. The detailed variable descriptions are shown in Table 1.
The descriptive statistics of the main variables are shown in Table 2

Model
In the basic regression, the dependent variables were continuous variables. Therefore, the OLS model was used as follows: In the above equation, COMSUMPit represents the dependent variable: per capita consumption expenditure of household i in year t. Additionally, the household consumption rate is used as an alternative dependent variable in the robustness examination. INDEXit-2 represents the digital finance development index of year t-2 in the area where household i is located and is used to measure the level of digital finance development in this city. β1 is the corresponding regression coefficient, representing the marginal effect of digital finance development on per capita household consumption expenditure. Xit represents a series of control variables, including householder characteristics, household wealth, regional economic development, etc. δt represents time fixed effect. εit is a random disturbing term.
Previous studies show that the development of the financial market (Levchenko, 2005), improvement of security level (Engen and Gruber, 2001) and convenient payment (McCallum and Goodfriend, 1988) can facilitate household consumption. In addition, online shopping arising with the rapid development of e-commerce has decreased the degree of information asymmetry and expanded supply in the consumer market, thus likely influencing household consumption purchase decisions. Since digital finance has infiltrated many aspects of daily life, this study explored the mechanism by which digital inclusive finance affects household consumption from the aspects of online shopping, online payment, online credit, internet financing, and commercial insurance, with the meditating model for examination. What needs to be noted is that, since variables relating to online payment and online credit in the CHFS data were inquired only in 2017, and the inquiry on online shopping in 2013 and 2017 was "did your family shop online last year," while that in 2015 was "did your family shop online last month," making the data incomparable. Only the cross-section data in 2017 were used for the analyses in this part. The mediating model was set as follows (Baron and Kenny, 1986):  is insignificant, but 2  is still significant, it means that the mediating variable has played the role of a full mediator.

Baseline results
Based on the examination results from the Hausman Test, the values of p is 0.000, which significantly reject the null hypothesis, thus this part adopts the fixed effect model to examine the impacts of the development of digital finance on household consumption expenditures. Table 3 reports the baseline regression results. The first column shows the regression results of the total index of digital finance. As shown in the table, the regression coefficient of digital finance on per capita household expenditure is significantly positive at a magnitude of 0.108, which indicates that digital finance has significantly promoted household consumption.
However, there is possible endogenous problem caused by a reverse causality issue. To overcome this problem, we use the number of mobile phones per person in the province as the instrumental digital finance variable. On the one hand, mobile phones have facilitated the use of financial services by residents, and can therefore be correlated with the level of digital finance development in a place. On the other hand, the average number of mobile phones in provinces hardly affects the consumption expenditure of households. Besides, we also did some tests to verify the validity of the instrumental variable. As shown in the second column of The coefficient of digital finance is still significantly positive.
For the control variables, the coefficients of total household assets and total household income are significantly positive, which indicates that the higher the household assets and income level, the higher the household consumption level and is consistent with the hypothesis of permanent income and inter-temporal consumption smoothing under the life cycle hypothesis. The regional financial development also has positive impacts on the level of household consumption, suggesting that regional financial development may promote household consumption.
Since digital finance is a multi-dimensional concept, this paper not only examined the impacts of the total index of digital finance on household expenditure, but also used second-level and third-level indices in the regression analyses. The second-level indices are coverage breadth, and use depth, and the regression results are shown in the second and third columns of Table 3; the third-level indexes namely, insurance, investment, credit investigation and the regression results are shown in the fourth, fifth, and sixth columns in Table

Heterogeneity results
This section reports the heterogeneity of the impacts of digital inclusive finance on household consumption among families in terms of household assets, household income, householder's financial literacy level, and urban development level, and the regression results are shown in Table 4. For a more reasonable division of the household sample, the balanced panel data was used in the regression for asset, income, and financial literacy heterogeneity, means these  Table 4. The coefficient of interaction of digital finance index and families in third-and fourth-tier cities is 0.046, which is also significantly positive, indicating that digital finance has larger impacts on the household consumption of families in third-and fourth-tier cities than in households living in first-and second-tier cities. The possible interpretation for the above results is that, in comparison to families with fewer assets, low income and lower financial literacy levels, and in third-and fourth-tier cities, families with more assets, high income and higher financial literacy levels and in first-and second-tier cities are facing less liquidity constraints, thus being less influenced by the marginal effect of digital finance on consumption.

[INSERT TABLE 4 ABOUT HERE]
We further discuss the heterogeneous impacts of digital finance on eight categories of household consumption and the results are shown in

Mechanism results
This section explored possible mechanisms by which digital finance affects household consumption. With the development of e-commerce, online shopping channels can help households improve the convenience of shopping and obtain abundant and cost-efficient commodities, thus likely promoting consumption. Table 7  Liquidity constraints are an important factor that restricts household consumption (Aziz and Cui, 2007). The development of online credit under digital finance can relieve consumers from the constraints of micro loans, thus likely promoting household consumption. Table 9 reports the results for which online credit is selected as the mediating variable through which the development of digital finance affects household consumption. The results in the second column show that the coefficient of the impacts of the index of digital finance on access to online credit is significantly positive, which indicates that digital finance has improved the access to online credit in households. The findings in the third column show that the access to online credit has significant positive impacts on household expenditure. Meanwhile, the coefficient of the index of digital finance is significantly positive and slightly lower than the result in the first column, implying that access to online credit is a mediating variable through which digital finance affects household consumption. The result of the Sobel mediating effect test is also significant at a magnitude of 1.91%, which supports the conclusion.

[INSERT TABLE 9 ABOUT HERE]
The value-added effect of digital finance on household consumption is delivered mainly through a wealth and income effect and realized through Internet financing (Zhang and Tu,2017). Digital finance has expanded the investment channels for residents. Internet financing products with both good profitability and liquidity represented by "Yu'e Bao" can promote household Internet investment and increase the rate of return on investment made by residents, thus likely promoting household consumption. Table 10   Since China's social security system is presently imperfect, there are higher uncertainty risks in households. Commercial insurance can help reduce the household expenditures on uncertainty risks like diseases and accidents, thus likely promoting household consumption. With the support of big data and information technology, the launching of more and more products online by traditional insurance companies and the gradual rise of Internet insurance may also promote more convenient insurance purchase by households. Table 11 reports the results of the mediating effect of commercial insurance purchase. The results in the second column show that the coefficient of the index of digital finance on the probability of households purchasing commercial insurance is significantly positive, which indicates that digital finance has promoted the purchase of insurance by households. The results in the third column show that the coefficient of the effect of the purchase of commercial insurance on per capita household expenditure is significantly positive. And, the coefficient of the index of digital inclusive finance is smaller than the first column, which indicates that insurance purchase plays a mediating role in the relationship between digital finance and household consumption. However, the results of the Sobel mediating effect test is significant at only 0.83% mediating effect, which is much weaker than in the other mediating variables.
[ INSERT TABLE 11 ABOUT HERE]

Robustness checks
This section performed a robustness check using the index of digital inclusive finance at the county level as the alternate index for digital finance development and the household consumption rate as the measure of household consumption level. The regression results are shown in Table 12 and Table 13. Since the digital finance indices at the county level are only for 2014, only the household samples for 2017 were selected in the regression. In Table 12, the first column shows the impacts of the total index of digital finance on per capita household expenditure; the second column shows the impacts of the index of digital finance coverage breadth on per capita household expenditure; the third column shows the impacts of the index of digital finance use depth on per capita household expenditure; the fourth column shows the impacts of the index of insurance on per capita household expenditure; the fifth column shows the impacts of the index of investment on per capita household expenditure; and the sixth column shows the impacts of the index of credit investigation on per capita household expenditure. It can be seen that the regression coefficients of all indices of digital finance are significantly positive, which indicates that the higher the digital finance development level, the higher the household consumption expenditure; and that digital finance has significantly promoted household consumption, consistent with the results in previous sections.
[ INSERT TABLE 12 ABOUT HERE] Table 13 reports the impacts of the index of digital inclusive finance on household consumption rate. The first, second, third, fourth, fifth and sixth columns report the regression results of the total index of digital inclusive finance and the indexes of coverage breadth, use depth, insurance, investment and credit investigation on household consumption rate, respectively. It can be seen that the coefficients of all indexes of digital finance are significantly positive, which indicates that the digital finance has significantly promoted the household consumption rate, consistent with the results before mentioned.
[INSERT TABLE 13 ABOUT HERE]

Conclusion
In recent years, there has been rapid development of digital finance based on big data, cloud computing and other digital technologies. By combining the data of CHFS and the digital finance index, this study performed empirical analyses on the impacts of digital finance on household consumption and explored its influencing mechanisms.
The results show that digital finance can significantly promote household consumption, especially for recurring items and households with fewer assets, lower income, less financial literacy, and those that live in third-and fourth-tier cities. The mediating model suggest that online shopping, digital payment, access to online credit, purchase of financing products on the internet and commercial insurance are all mediating variables in the relationship between digital finance and household consumption, which indicate the impacts of digital finance on household consumption mainly through relieving liquidity constrain, facilitating payment and transaction, expanding investment channels and increasing income, and enhancing security. Relevant government agencies shall actively promote the development of digital finance and focus on the role of digital finance in improving the consumption level of low-and middle-income families and underdeveloped areas.   Notes: *, ** and *** respectively indicate significance at the level of 10%, 5% and 1%; words in brackets mean the standard deviation of heteroskedasticity-robust of the cluster at the municipal level; what is reported in the table is the estimated marginal effect. Hereinafter the same. Because we use the index data with two years lag in the regression, and the start year of investment index data is 2014 and credit investigation index data is 2015, which means these two index data could only be matched with the 2017 CHFS data. Thus, in the regression of investment index and credit investigation index on household consumption, we only use the 2017 CHFS data in which the observations are 23070.  (1) and (2) are both 42658 which is less than the baseline regression of 66789. As for column (3), the observations are 42088 which are less than those in column (1) and (2) because there are missing values for financial literacy.

Note:
Considering the stability of data, we use the samples of household consumption rate between 0 and 1. Thus, the observations in column (1), (2), (3) and (4) are 37705 which are less than the baseline regression of 66789. The same reason as the column (5) and (6) of table 3, column (5) and (6) only use the 2017 CHFS data in which the observations are 14017.