How and what kind of cities benefit from the development of digital inclusive finance? Evidence from the upgrading of export in Chinese cities

Figures & data

Figure 1. Digital Financial Inclusion Index and city’s export sophistication in 2013 (left) and 2017 (right).

Source: The Authors.

Figure 2. Traditional finance development and city’s export sophistication in 2013 (left) and 2017 (right).

Source: The Authors.

Table 1. Descriptive statistics of main variables.

Variables	Observations	Mean	Variance	Minimum	Maximum
$\mathit{ln}UG$	1340	10.88	0.13	10.44	11.23
$\mathit{ln}\mathit{DIFI}$	1340	5.13	0.23	4.47	5.65
$\mathit{ln}\mathit{FDIShare}$	1340	−0.03	1.34	−8.64	3.04
$\mathit{ln}\mathit{Finance}$	1340	0.79	0.40	−0.40	2.53
$\mathit{Indus}$	1340	47.19	10.67	8.12	81.89
$\mathit{ln}\mathit{TFP}$	1340	0.15	0.76	−2.25	1.08
$\mathit{ln}\mathit{POP}$	1340	5.94	0.65	3.43	8.13
$\mathit{Gover}$	1340	0.19	0.09	0.04	0.83

Source: The Authors.

Table 2. Benchmark regression results.

Variables	Whole sample		Remove the cities with high degree of $\mathit{DIFI}$ spatial agglomeration
	(1)	(2)	(3)	(4)
$\ln \mathit{DIFI}$	0.0795*		0.0805*
	(0.0408)		(0.0424)
$\ln \mathit{Digitization}$		0.0260*		0.0251*
		(0.0149)		(0.0150)
$\ln \mathit{FDIShare}$	0.0045*	0.0047*	0.0044*	0.0047*
	(0.0024)	(0.0025)	(0.0024)	(0.0025)
$\ln \mathit{Finance}$	−0.0351*	−0.0330*	−0.0353*	−0.0334*
	(0.0190)	(0.0187)	(0.0192)	(0.0190)
$\mathit{Indus}$	0.0003	0.0002	0.0003	0.0002
	(0.0002)	(0.0002)	(0.0002)	(0.0002)
$\ln \mathit{TFP}$	0.0026	0.0023	0.0026	0.0023
	(0.0025)	(0.0025)	(0.0025)	(0.0026)
$\ln \mathit{POP}$	0.0448	0.0413	0.0458	0.0433
	(0.0355)	(0.0343)	(0.0366)	(0.0353)
$\mathit{Gover}$	−0.2173***	−0.2175***	−0.2168***	−0.2160***
	(0.0591)	(0.0591)	(0.0593)	(0.0594)
_cons	10.3754***	10.6832***	10.3630***	10.6746***
	(0.3224)	(0.2260)	(0.3289)	(0.2312)
Time fixed effect	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES
N	1340	1340	1315	1315
R^2	0.8935	0.8933	0.8919	0.8917

Note: (1) The values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represents the significance levels of 10%, 5%, and 1%, respectively.

Source: The Authors.

Table 3. Estimation results of instrumental variables.

Variables	Whole sample		Remove the cities with high degree of DIFI spatial agglomeration
	(1)	(2)	(3)	(4)
	FE-IV	IV-GMM	FE-IV	IV-GMM
$\ln \mathit{DIFI}$	0.4803*	0.4410*	0.5213*	0.4953^a
	(0.2750)	(0.2624)	(0.3123)	(0.3034)
First-stage iv1:	0.0040**	0.0040**	0.0031*	0.0031*
$\ln \mathit{Phone}\_\mathit{internet}$	(0.0018)	(0.0018)	(0.0018)	(0.0018)
First-stage iv2:	0.0092**	0.0092**	0.0092**	0.0092**
$\ln \mathit{Phone}\_\mathit{mobil}$	(0.0045)	(0.0045)	(0.0045)	(0.0045)
First-stage F test	5.58***	5.58***	4.44**	4.44**
	(0.0043)	(0.0043)	(0.0129)	(0.0129)
Cragg-Donald Wald F statistic	11.817	11.817	10.004	10.004
Hansen J statistic	0.5011	0.4984	0.6937	0.6929
Control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES
N	1091	1091	1066	1066
R^2	0.8835	0.8864	0.8790	0.8811

Note: (1) Except that the values in brackets of the first stage F test are the corresponding p values, the other values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

^aThe p value of ln DIFI in column (4) is 0.103, which is approximately equal to 0.1, so this variable can be roughly considered as significant at 10% significance level.

Source: The Authors.

Table 4. Regression results of robustness test.

		(2)
	(1) Dependence on export	Excluding examples with residual error greater than 2.5 times of standard deviation	(3) Excluding provincial capital cities	(4) RE
$\ln \mathit{DIFI}$	2.9807***	0.0865**	0.0732*	0.1578***
	(0.7743)	(0.0377)	(0.0434)	(0.0348)
Control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES
N	1340	1298	1216	1340
R2	0.2242	0.9058	0.8938	0.8921

Note: (1) The values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.

Table 5. Estimation results of city’s heterogeneity analysis.

Variables	City size		City’s wage level		City’s human capital level		Port or non-port city
	(1)		(3)	(4)	(5)	(6)	(7)
	Small and medium size cities	(2) Large cities	Low and medium wage cities	High wage cities	Low and medium human capital cities	High human capital cities	Non- port city	(8) Port city
$\ln \mathit{DIFI}$	0.0783*	0.0399	0.1232**	0.2417	0.0286	0.1899**	0.0734	0.1867***
	(0.0464)	(0.0823)	(0.0517)	(0.1680)	(0.0501)	(0.0815)	(0.0559)	(0.0614)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES	YES	YES	YES	YES
N	982	358	898	442	880	460	860	480
R^2	0.8845	0.9234	0.8734	0.8864	0.8870	0.9075	0.8748	0.9330

Note: (1) The values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.

Table 6. Mechanism results of innovation effect.

	FE		IV
	$\mathit{ln}\mathit{PerPatent}$	$\mathit{ln}UG$	$\mathit{ln}\mathit{PerPatent}$	$\mathit{ln}UG$
	(1)	(2)	(3)	(4)
$\ln \mathit{DIFI}$	0.7748**	0.0718*	4.9258**	0.3901
	(0.3698)	(0.0403)	(2.5115)	(0.2642)
$\ln \mathit{PerPatent}$		0.0126**		0.0157**
		(0.0052)		(0.0065)
First-stage iv1:			0.0040**	0.0038**
$\ln \mathit{Phone}\_\mathit{internet}$			(0.0018)	(0.0018)
First-stage iv2:			0.0094**	0.0092**
$\ln \mathit{Phone}\_\mathit{mobil}$			(0.0045)	(0.0045)
First-stage F test			5.60***	5.23***
			(0.0042)	(0.0060)
Cragg-Donald Wald F statistic			12.077	11.239
Hansen J statistic			0.8862	0.5272
Sobel test	3.692(0.0002)	28.16%
Control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES
N	1335	1335	1086	1086
R-sq	0.7348	0.8946	0.6581	0.8915

Note: (1) Except that the values in brackets of the first stage F test and Sobel test are the corresponding P values, the other values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.

Table 7. Bootstrap sampling mechanism results of innovation effect (${\mathit{\beta }}^{``}\times \mathit{\lambda }$).

	(1)	(2)	(3)	(4)
	Panel A 100 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel B 200 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel C 500 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel D 1000 bootstrap sampling (${\beta }^{``}\times \lambda$)
Indirect effect	0.0932***	0.0932***	0.0932***	0.0932***
	(0.0272)	(0.0247)	(0.0248)	(0.0229)
5% quantile	0.0534	0.0336	0.0488	0.0444
95% quantile	0.1400	0.1400	0.1434	0.1428
Direct effect	0.2378***	0.2378***	0.2378***	0.2378***
	(0.0452)	(0.0401)	(0.0389)	(0.0405)
5% quantile	0.1597	0.1753	0.1671	0.1671
95% quantile	0.3610	0.3083	0.3443	0.3182

Note: (1) The values in brackets are the bootstrap standard errors. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.

Table 8. Mechanism test of market effect.

	FE		IV
	(1)	(2)	(3)	(4)
	ln Market	ln UG	ln Market	ln UG
$\mathit{ln}\mathit{DIFI}$	0.2509***	0.0624	1.8228***	0.3296
	(0.0895)	(0.0428)	(0.5166)	(0.2650)
$\mathit{ln}\mathit{Market}$		0.0979***		0.0847***
		(0.0259)		(0.0294)
First-stage iv1:			0.0039**	0.0037**
$\mathit{ln}\mathit{Phone}\_\mathit{internet}$			(0.0018)	(0.0019)
First-stage iv2:			0.0098**	0.0089**
$\mathit{ln}\mathit{Phone}\_\mathit{mobil}$			(0.0046)	(0.0045)
First-stage F test			5.62***	4.87***
			(0.0042)	(0.0085)
Cragg-Donald Wald F statistic			12.410	10.506
Hansen J statistic			0.6067	0.5198
Sobel test	7.607(0.000)	48.73%
Control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
City fixed effect	YES	YES	YES	YES
N	1309	1309	1081	1081
R^2	0.8054	0.9004	0.6135	0.8952

Note: (1) Except that the values in brackets of the first stage F test and Sobel test are the corresponding P values, the other values in brackets are the robustness standard errors of city level clustering. (2) *, **, and *** represents the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.

Table 9. Bootstrap sampling mechanism results of market effect (${\mathit{\beta }}^{``}\times \mathit{\lambda }$).

	(1)	(2)	(3)	(4)
	Panel A 100 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel B 200 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel C 500 bootstrap sampling (${\beta }^{``}\times \lambda$)	Panel D 1000 bootstrap sampling (${\beta }^{``}\times \lambda$)
Indirect effect	0.1698***	0.1698***	0.1698***	0.1698***
	(0.0211)	(0.0243)	(0.0216)	(0.0225)
5% quantile	0.1409	0.1158	0.1295	0.1251
95% quantile	0.2223	0.2098	0.2194	0.2104
Direct effect	0.1786***	0.1786***	0.1786***	0.1786***
	(0.0375)	(0.0408)	(0.0401)	(0.0373)
5% quantile	0.0974	0.0887	0.0958	0.1138
95% quantile	0.2516	0.2597	0.2534	0.2672

Note: (1) The values in brackets are the bootstrap standard errors. (2) *, **, and *** represent the significance levels of 10%, 5%, and 1%, respectively. (3) The control variables are the same as in Table 2.

Source: The Authors.