Bank Digital Transformation and Enterprise Innovation—Evidence from China

Abstract

With the rapid advancement of digital technology, the banking industry has embarked on a journey of digital transformation. While existing literature primarily examines how these changes impact the banks themselves, our study focuses on a relatively unexplored aspect: the direct influence of bank digital transformation on the performance and behavior of borrowing enterprises. The research objective of this study is to explore the influence of bank digital transformation on the innovation performance of borrowing enterprises and the underlying mechanisms. Leveraging data from Chinese listed companies and commercial banks, we find a positive effect of bank digital transformation on enterprise innovation output as measured by firms’ patent applications. The findings remain robust across alternative model specifications, controls for regional digital economy development levels, and bank financial performance, as well as alternative measures of bank digital transformation. Mechanism tests show that bank digital transformation contributes to corporate innovation by alleviating corporate financial constraints and improving corporate governance. Further research demonstrates that bank digital transformation also helps promote corporate innovation efficiency as measured by the proportion of patent output to total R&D input and corporate innovation output as measured by firms’ invention patent applications and patent grants. Additionally, borrowing firms’ own digital transformations may substitute for bank digital transformation in their effect on innovations.

1. Introduction

Innovation is a key driver of economic development. Since the first industrial revolution in history, every leap in the development of human society has been promoted without exception by breakthroughs in scientific and technological innovation [1]. However, corporate innovation efforts face serious external financing challenges [2] due to a long investment horizon, large capital requirements, and high uncertainty of innovation initiatives [3], as well as their innate information asymmetry [4]. In this regard, real businesses need a new financial service model because traditional finance faces the issue of having insufficient capacity to support the real economy. Moreover, the problem is exacerbated by poor corporate governance, where tunneling behaviors such as insider misuse of funds prevent firms from providing financial support for innovative activities.

In recent years, with the rapid advancement of digital technology, traditional finance has intensified its profound integration with technology and the banking industry has embarked on a journey of digital transformation. Banks are leveraging technologies like big data to modernize and revamp traditional operations, incorporating digital technology into their products and processes. For instance, in China, the banking sector’s investment in information technology reached CNY 207.8 billion in 2020, demonstrating substantial progress in driving digital transformation compared to other financial sectors. Digital transformation has been reshaping the traditional financial services model of China’s banking industry. Research on the economic consequences of bank digital transformation has concentrated on how banks themselves are influenced by it. However, there is a lack of studies directly examining the impact of bank digital transformation on borrowing enterprises’ performance and behavior. The research objective of this study is to investigate the impact of bank digital transformation on the performance and behavior of borrowing enterprises, particularly with regard to their innovation output and efficiency. Through empirical testing using data from Chinese listed companies and commercial banks, we also aim to explore the influencing mechanism by which bank digital transformation contributes to corporate innovation and whether borrowing firms’ own digital transformations can substitute for the effect of bank digital transformation on innovation.

We argue that bank digital transformation can enhance firm innovation performance by alleviating firms’ financing constraints and improving corporate governance. On the one hand, by leveraging digital technology, banks can reduce information asymmetry and improve risk control. This results in a shift from mortgage to credit loans (nonmortgage loans) and lowers the entry threshold for access to bank loans. Furthermore, the cost of bank financing is also reduced through the use of digital technology. As a result, bank digital transformation helps alleviate corporate financing constraints, enabling firms to obtain more funds to support innovation activities and thus improve enterprise innovation performance. On the other hand, bank digital transformation can enhance their roles in corporate governance by curbing the opportunistic conduct of firms’ managers and major shareholders. This reduces the misuse of funds and enhances the effectiveness of capital utilization. It permits firms to retain more funds for innovation activities, ultimately leading to enhanced enterprise innovation performance.

Drawing on the study of Zhao et al. [5], we measure the indicator of bank digital transformation using data on digital transformation-related patents applied for by China’s commercial banking sector between 2010 and 2020. After matching companies with their lending banks year by year based on their loan information, we calculate the annual average digital transformation indicators of the corresponding lending banks. We empirically examine the impact of these digital transformation indicators on the innovation performance of the companies. Through our study, we provide robust evidence to support the positive effect of bank digital transformation on enterprise innovation.

Prior studies have documented the impact of bank digital transformation on banks’ stability [6], nonperforming loan (NPL) ratios [7], operational capabilities [8], production efficiency improvement [9], and financial performance [10]. Existing literature has predominantly concentrated on the economic effects of bank digital transformation on the behavior and performance of banks themselves. However, there has been a notable absence of direct examination on the impact of bank digital transformation on the behavior and performance of the companies that borrow from these banks, leaving a significant gap in our understanding of whether and how bank digital transformation serves the real economy. This research gap holds significance because it limits our ability to fully understand how bank digital transformation contributes to the real economy and to develop effective policies/interventions to address related problems.

Additionally, previous literature has primarily measured the degree of digital transformation based on textual analysis, using the frequency of digital transformation-related keywords in annual reports [10,11,12]. However, this method may overlook the actual practices and effects of banks in the process of digital transformation. Relying only on the number and frequency of keywords to evaluate the level of digital transformation can lead to one-sided and inaccurate assessment results. Moreover, this method may also be affected by the completeness and accuracy of the disclosure of annual reports. Banks may have an incentive to highlight specific keywords related to digital transformation in their reports to create a positive image, or they may not fully disclose information related to digital transformation in annual reports. This can lead to reporting bias, where the actual level of digital transformation may not match the emphasis placed on certain keywords. Furthermore, this method is primarily applicable to listed banks with publicly available annual reports. It may not cover non-listed banks that do not disclose annual reports, limiting the scope of analysis.

Finally, the established literature on the economic consequence of digital transformation solely concentrates on a single digital transformation entity, for example, listed firms or commercial banks [6,7,8,9,10,11,12]. However, existing research has not fully addressed how bank digital transformation interacts with the digital transformation of their enterprise customers, leaving another significant research gap.

Our study aims to fill the above research gaps and is innovative in the following ways: The primary innovation in our study lies in being the first to directly investigate the impact of bank digital transformation on enterprise innovation. We provide empirical evidence to support the positive influence of bank digital transformation on enterprise innovation. We demonstrate that this effect persists even after controlling for the level of regional digital economy development. Furthermore, we introduce and explore the mechanisms through which bank digital transformation benefits the real economy by alleviating financing constraints and enhancing corporate governance. This in-depth analysis sheds light on the effect of bank digital transformation on enterprise innovation in a more comprehensive and detailed manner.

Second, instead of measuring the extent of bank digital transformation by the frequency of digital transformation-related keywords in annual reports of listed banks, we use the data on digital transformation-related patent applications filed by banks. This innovative approach to measuring the extent of bank digital transformation offers several advantages compared to the conventional method of relying on keywords in bank annual reports, as seen in previous literature. To begin with, patent applications directly reflect a bank’s innovations and technological advancements in the digital domain. It is a tangible outcome of their digital transformation efforts. As an objective and quantifiable metric, patent applications furnish concrete evidence of a bank’s investment in and dedication to digital transformation. Moreover, patent data tend to be less susceptible to reporting biases. Lastly, this approach is applicable to both listed and non-listed banks, providing a more comprehensive view of the digital transformation landscape across the banking sector.

Third, we explore the interaction between bank digital transformation and the digital transformation of their enterprise customers. Through this exploration, new avenues of research related to digital transformation are presented, providing innovative ideas and insights for future study.

The remainder of this paper proceeds as follows. Section 2 develops the testable hypotheses. Section 3 explains the data source and the model specifications. Section 4 presents the empirical results of the baseline regression and robustness tests. Section 5 presents the results of the mechanism tests, followed by additional analyses in Section 6. Section 7 concludes with a summary and the discussions. Section 8 discusses the implications, the limitations, and the outlook on future research.

2. Hypothesis Development

We believe that bank digital transformation can play a role in enhancing enterprise innovation performance for the following reasons.

On the one hand, bank digital transformation can enhance firm innovation performance by alleviating firms’ financing constraints. Our study is mainly based on Chinese data to explore the impact of bank digital transformation on borrowing firms. The Chinese financial system is more bank-based [13]. Bank financing still plays a large role in the Chinese financial system, and firms rely heavily on it. Meanwhile, given its long cycle time, high capital consumption, and a high degree of uncertainty and information asymmetry, innovation activity is particularly affected by financing constraints when compared to other investment projects [3,4]. Traditional bank lending practices often prioritize collateral or guarantees, with a focus on the value of fixed assets that can serve as collateral, rather than a firm’s technological or innovative capabilities [14]. This can create challenges for businesses with high growth potential but limited fixed assets to leverage for financing. As a result, these companies may struggle to secure sufficient funding to support their innovative activities. In contrast, bank digital transformation can alleviate firms’ financing constraints and enable them to obtain more funds to support innovation activities, thus improving their innovation performance, which is discussed below:

First, the information asymmetry between banks and borrowing firms is greatly reduced through real-time data collection and processing of multidimensional data. Regarding information collection, cutting-edge technologies, such as Internet of Things and data mining, can help banks obtain multidimensional data in real time, which provides a company’s fundamental information in greater detail [15] and enables banks to accurately portray the risk profile of their customers. In terms of information processing, banks can use big data and artificial intelligence methods to effectively capture nonlinear relationships in massive amounts of data and to harden soft information by enabling traceability management and remote monitoring through Internet of Things and blockchains [16].

Second, digital transformation allows banks to shift their risk control model from collateral- to credit-driven. For example, the big data credit model leverages the information and modeling advantages of big data and artificial intelligence to establish an intelligent risk control framework. This framework enables precise and real-time assessment of the credit risk associated with borrowing firms [17,18]. Subsequently, loans are granted based on this assessment. Within the scope of the big data credit model, banks are better able to identify and manage risks even when making loans to businesses without security or guarantees. Khalifaturofi‘ah et al. [19] demonstrated that banks with a higher investment in digital transformation provide a higher number of cheaper and faster guaranteed loans. As a result, banks have been able to bring more long-tail customers who lack collateral and credit history into their credit services through the use of digital technology [20]. For instance, in 2022, Industrial Bank implemented the “technology flow” evaluation system, leveraging the technical support of “data + model + system”. This groundbreaking system employed digital methods to measure the effectiveness of an enterprise’s science and technology endeavors. It evaluated the growth potential of technology startups and facilitated the conversion of their technological “soft power” into tangible financial resources. Consequently, the system extended credit support to these startups (https://news.sina.com.cn/shangxunfushen/2023-11-02/detail-imztfnvn7913709.shtml (accessed on 8 November 2023)).

Third, through the use of digital technology, banks may fully utilize prelending review techniques and postlending supervisory methods based on big data and artificial intelligence. This can not only improve operational efficiency and reduce banks’ review and monitoring costs but also reduce human intervention and rent-seeking in the loan approval process, ultimately reducing the cost of financing for enterprises [21]. Industrial Bank, for instance, has implemented a digital operation management platform. The platform enables visualized internal control management, replacing manual statistics with intelligent algorithms for complex indicator calculations. Through this platform, managers can obtain a comprehensive understanding of the front-line business operations. The platform dynamically displays key risk indicator data. Furthermore, the internal control management platform clarifies the standardized business management process for each segment. By leveraging the digital operation management platform, the bank reduces the workload of manual operations, supervision, and traceability. It streamlines standardized business management and strengthens the full life-cycle management of business processes. Ultimately, the platform improves the quality and efficiency of operation management and reduces the bank’s operating costs (https://news.sina.com.cn/shangxunfushen/2023-11-02/detail-imztfnvn7913709.shtml (accessed on 8 November 2023)).

In short, through the utilization of digital technology, banks can mitigate information asymmetry and enhance risk management, leading to a transition from collateral-based to credit-based lending and lowering the barrier to entry for accessing bank loans. Additionally, digital technology adoption reduces the cost of bank financing. Consequently, bank digital transformation effectively alleviates firms’ financing constraints, facilitating their access to more funds to bolster innovation activities and to ultimately improve their innovation performance.

On the other hand, by enhancing corporate governance, bank digital transformation can improve enterprises’ innovation performance. To protect the security of loan money, banks have the incentive to discipline the opportunistic behavior of borrowing firms. Commercial banks are relational creditors, according to debt heterogeneity theory [22]. They maintain diverse business relationships with enterprises beyond debt obligations, including fund custodianship and financial consulting [23]. Their aim is to establish enduring partnerships with these enterprises to generate stable income over the long term. As a result, banks are motivated to oversee the borrowing firm and focus on its long-term growth. Due to agency problems, resources that businesses may employ for projects such as technological progress could instead be used by firm managers and large shareholders for their self-interest [24,25]. The adoption of digital technology can potentially enhance banks’ oversight of borrowers, allowing them to effectively prevent misappropriation of funds and ensure that a greater amount is allocated towards innovation investments, ultimately improving businesses’ innovation performance, as discussed further below.

First, by using digital technology, a comprehensive analysis and assessment of the borrowing business could be carried out prior to loan disbursements. For example, through big data analysis and mining, banks can gain insights into various aspects of a borrowing company’s operating conditions, financial performance, and funds flow, which help banks assess the credit risk of a company. Another example is that banks can utilize digital technology to collect multi-dimensional data to multi-dimensionally assess the personal conduct and wealth of significant shareholders, gaining insights into their financial standing and business track record. These data may include real-time transaction data at the point of sale [26], social media data [27], user rating data [28], and satellite imagery data [26], among others. This, in turn, allows for a more accurate estimation of the borrowing company’s risk profile [21]. Based on a full understanding of the risk profile and potential opportunistic behavior of the borrowing enterprise, banks can better achieve the objective of the ex-ante prevention of self-interest behaviors. Loan covenants can be designed to increase the cost to insiders for self-dealing, discouraging capital abuse by management or major shareholders. This encourages firms to reallocate resources from generating private profits to innovation activities, promoting innovation performance [29].

Second, through the use of digital technology, banks are able to conduct comprehensive monitoring and review of borrowers after loan disbursements [8]. This improves the governance of borrowing companies and helps reduce opportunistic behaviors, such as the misuse of funds by firm management or major shareholders. The existence of information asymmetry incurs significant costs for banks to collect information to monitor borrowing firms. The adoption of digital technology can reduce banks’ cost of information collection while improving the efficiency and effectiveness of monitoring [9]. For example, the use of digital technology by banks not only allow them also to convert soft information into hard information that can be transmitted and processed [30] but also to combine traditional financial and nonfinancial data from various data sources to evaluate the performance of their borrowers from multiple dimensions [15]. This can help them identify problems in a timely manner and to strengthen the monitoring of borrowing firms. Banks can also use information technology and data analysis tools to examine the financial data of borrowers to reduce the risk of manipulation and falsification. For example, the Industrial and Commercial Bank of China (ICBC) proposed a new development strategy named E-ICBC 2.0 based on big data technology and internet technology in 2015. By employing emerging technology, the ICBC greatly reduced the credit risks involved with its business [31]. Additionally, by using AI-powered tools and multiple data sources, banks can conduct investigations into firm management, controlling shareholders and related parties more thoroughly and effectively. By doing so, they can better assess their own risk and receive early warnings that will help them lower the likelihood of opportunistic behaviors, such as financial misuse. A notable example is Jiangsu Suning Bank, which has been at the forefront of digital risk control technology and has successfully registered relevant patents. It has developed a core system called “Cloud Opening” and established a comprehensive and rigorous data risk control system. The bank has implemented a digital risk management system that includes the “Lens” intelligent risk control engine, “Unified Credit Limit System”, “Unified Loan Management System”, “Sharp Eye” risk warning system, and “Capture Overdue” intelligent collection system (https://www.sohu.com/a/713815925_407695 (accessed on 8 November 2023)).

In short, bank digital transformation can enhance their roles in corporate governance by curbing the self-interested actions of firms’ managers and major shareholders. Doing so reduces the misuse of funds, enhances the effectiveness of capital utilization, permits enterprises to retain more funds for longer-term investments such as innovation activities, and motivates enterprises to choose wisely regarding innovation projects, all of which ultimately improve enterprise innovation performance.

Based on the above analysis, the following hypothesis is proposed:

Hypothesis 1.

Bank digital transformation helps boost the innovation performance of their borrowing firms.

Hypothesis 2.

Bank digital transformation helps boost the innovation performance of their borrowing firms by alleviating the financial constraints of these firms.

Hypothesis 3.

Bank digital transformation helps boost the innovation performance of their borrowing firms by enhancing the corporate governance of these firms.

3. Research Design

3.1. Sample and Data Source

Using China’s A-share companies from 2011 to 2021 as the initial sample, we obtain 13,052 firm-year observations for estimating the baseline regression. We apply the following sample selection criteria: (1) exclusion of firms in the finance industry, (2) removal of observations with missing data, and (3) exclusion of delisted companies.

The research data in this article come from three sources: listed companies, commercial banks, and the borrowing–lending interactions between these two entities. The data on corporate financial performance indicators, corporate governance indicators, and stock market trading data are all from the China Stock Market Accounting Research (CSMAR) database. The level of bank digital transformation usage is mainly measured using the bank’s digital transformation-related patent data. We follow the following steps to obtain the data on bank digital transformation. ① We collect patent application records from the National Intellectual Property Administration patent retrieval database (https://pss-system.cponline.cnipa.gov.cn/conventionalSearch (accessed on 6 May 2023)) from 2010 to 2020, specifically for those filed by the banks as applicants. These records reveal information such as the patent applicant, application time, and patent summary. The patent summary discloses the primary purpose and technology involved in the patent. ② We use a self-built digital transformation dictionary to further examine each patent application summary. If the summary contains keywords related to digital technology, the patent is identified as a digital transformation-related patent. The digital transformation dictionary is established based on the research by Du et al. [32] (see Appendix A). ③ After identifying digital transformation-related patent applications from 2010 to 2020, we count the number of digital transformation-related patent applications of each bank in each year and organize a panel dataset of “Bank–Year–Digital transformation-related patent applications” from 2010 to 2020. The bank-level financial performance indicators come from the Banking Research database of the CSMAR database.

In addition, to establish the correspondence between banks and their borrowing firms, we use the bank loan data for listed companies from the CSMAR database. First, the loan records are uniformly organized into a dataset of “listed company–year–bank.” Next, these data are matched with the number of digital transformation-related patent applications filed by each bank in that year, resulting in the dataset of “Listed company–Year–Bank–Bank digital transformation-related patent applications”.

3.2. Variables and Models

To test Hypothesis 1, we construct the following OLS (Ordinary Least Squares) regression Equation (1) that includes the industry (IND) and year (YEAR) effects. Controlling for industry effects helps to account for industry-specific factors that could influence the outcome variable, while year effects help account for aggregate fluctuations.

LNPATENT_i,t = a₀ + b₁BANKDIGIT_i,t−1 + b₂CONTROL_i,t−1 + IND + YEAR + ε

(1)

Prior studies have noted that China’s patent statistics are the most suitable indicator for measuring Chinese enterprises’ innovation performance [33,34]. Thus, we use the number of patent applications of listed companies as an indicator of enterprise innovation output. The explained variable, LNPATENT, is defined as the natural logarithm of one plus the number of patent applications of listed companies.

We use the cumulative number of digital transformation-related patent applications filed by the banks corresponding to a company in a given year as a measure of their digital transformation efforts. We calculate the bank digital transformation variable, BANKDIGIT, by dividing the cumulative number of digital transformation-related patent applications of the corresponding bank by the total number of such patent applications filed by all banks in the same year. Specifically, this variable is defined as follows:

$${BANKDIGIT}_{it}={\sum }_{n=1}^N\frac{{Bpatent}_{int}}{{Sum\_bpatent}_t}\times \frac{{Loan}_{int}}{{Sum\_loan}_{it}}$$

(2)

Bpatent_int represents the cumulative number of digital transformation-related patent applications filed by bank n that lends to company i in year t. Sum_bpatent_t represents the sum of the total number of digital transformation-related patent applications of all banks in year t. Loan_int represents the amount of the bank loan borrowed by company i from bank n in year t. Sum_loan_it represents the total amount of the bank loan of company i in year t. The economic interpretation of BANKDIGIT is the proportion of the cumulative number of digital transformation-related patent applications filed by the lending banks of firm i in year t to the total number of digital transformation-related patent applications of all banks in year t. This proportion is weighted by the amount of the bank loan and averaged at the firm level. A higher value of BANKDIGIT indicates that banks from which the sample firm is borrowing have undergone a higher degree of digital transformation.

Following Lin et al. [35], we control for a series of corporate finance and governance variables that may affect enterprise innovation performance: SIZE (firm size, measured as the natural logarithm of total assets of the firm), LEV (debt ratio, measured as the ratio of total debt to total assets), ROA (return on assets, measured as the ratio of net income to total assets), GROWTH (sales growth rate, measured as the growth rate of sales revenue), OCF (operating cash flow, measured as the ratio of cash flow from operating activities to total assets), TANG (tangible assets, measured as the ratio of tangible assets to total assets), AGE (firm age, measured as the natural logarithm of one plus the number of years that the firm has been listed), MANHOLD (management shareholding, measured as the fraction of shares held by management), BOARD (size of the board, measured as the natural logarithm of the number of directors on the board), and BALANCE (equity balance, measured as the shareholding ratio of the second to fifth largest shareholders divided by the shareholding ratio of the largest shareholder). Appendix B provides detailed definitions of the main variables.

4. Empirical Results

4.1. Descriptive Statistics and Baseline Results

Table 1. Summary statistics of the main variables.

VARIABLES	Median	Mean	St. Dev.	Min.	Max.
LNPATENT	0.140	0.283	0.899	0.000	4.234
BANKDIGIT	0.077	0.150	0.196	0.000	1.118
SIZE	22.166	22.295	1.213	18.889	26.610
LEV	0.468	0.474	0.206	0.057	1.081
ROA	0.029	0.022	0.079	−0.354	0.280
GROWTH	0.099	0.169	0.467	−0.682	3.261
OCF	0.040	0.040	0.070	−0.208	0.265
TANG	0.172	0.205	0.158	0.002	0.734
AGE	2.303	2.268	0.703	0.000	3.434
BOARD	2.197	2.117	0.204	1.386	2.890
BALANCE	0.598	0.762	0.627	0.003	4.000
MANHOLD	0.006	0.124	0.184	0.000	0.879

reports the descriptive statistics. All of the variables, except for the dummy variables, are winsorized at the 1st and 99th percentiles. The mean and median of BANKDIGIT variables are 0.150 and 0.077, showing a right-skewed distribution as a whole. The minimum and maximum values of LNPATENT are 0 and 4.234, respectively, with a standard deviation that greatly exceeds its median and mean. This suggests a wide variation in the innovation performance of Chinese-listed companies. Although not reported, the VIF scores are smaller than 10 for all independent variables, indicating no high multicollinearity issue in our study.

Table 2. Baseline results.

	(1)	(2)
VARIABLES	LNPATNET	LNPATNET
BANKDIGIT	0.257 ***	0.155 ***
	(7.921)	(4.229)
SIZE		−0.005
		(−0.933)
LEV		−0.365 ***
		(−11.369)
ROA		−0.544 ***
		(−6.197)
GROWTH		0.028 **
		(2.495)
OCF		−0.257 ***
		(−3.172)
TANG		−0.117 ***
		(−2.710)
AGE		−0.095 ***
		(−12.704)
BOARD		0.006
		(0.441)
BALANCE		0.010
		(1.111)
MANHOLD		0.565 ***
		(13.672)
IND&YEAR	YES	YES
Constant	0.391 ***	0.921 ***
	(5.999)	(7.096)
Observations	13,052	13,052
R-squared	0.046	0.098

Note: **, and *** indicate statistical significance at 5%, and 1%, respectively. The t-statistics are reported in parentheses.

reports the results of the baseline regression Equation (1). Column 1 reflects the regression of the enterprise innovation performance variable, LNPATENT, on the bank digital transformation variable, BANKDIGIT, which shows that enterprise innovation performance is positively related to the level of bank digital transformation. Column 2 reflects the inclusion of the control variables. The coefficient of BANKDIGIT remains significantly positive after including the control variables. Taken together, the results from Table 2 suggest a positive effect of bank digital transformation on enterprise innovation performance, supporting Hypothesis 1.

4.2. Robustness

4.2.1. Propensity Score Matching

High-quality listed companies may have a preference for banks with a high degree of digital transformation, and banks that have undergone digital transformation may also choose high-performing enterprises. Therefore, this research may encounter endogeneity issues stemming from self-selection. As a robustness test, we adopt the propensity score matching (PSM) approach to mitigate potential endogeneity issues. Propensity score matching is a widely employed statistical technique in finance, management, and business research to enhance the comparability of treatment and control groups in observational studies. This method is particularly valuable when conducting non-randomized studies. The core idea behind propensity score matching is to estimate the probability (propensity score) of an individual firm being in the treatment group based on a set of observed covariates, which can include factors like firm size, financial performance, and other relevant characteristics. By estimating these scores, researchers can then match treatment group firms with control group firms that have similar or nearly identical propensity scores, thus creating a balanced comparison group. This matching process helps mitigate selection bias and potential confounding effects, allowing researchers to draw more robust conclusions about the impact of the treatment or intervention on firm outcomes [36]. To apply this approach, firstly, the propensity scores will be estimated, which represent the likelihood of each observation being assigned to the treatment group. Statistical methods, typically logistic regression, are used to calculate these scores based on a set of observed covariates. Next, the observations in the treatment group can be matched with those in the control group that have similar or nearly identical propensity scores. This matching process ensures that the treatment and control groups are more comparable in terms of their observed characteristics. Finally, it is essential to assess the quality of the matching to ensure that the post-matching treatment and control groups exhibit balanced covariate distributions. This verification is necessary to validate the effectiveness of the analysis.

We first run a logit regression to calculate the propensity score based on the matching variables that are the control variables of the baseline regression Equation (1). If the lending banks from which the firm borrows have had digital transformation-related patents for the past five consecutive years, such firms are used as the treatment group, and the rest are in the control group. We then match each treatment observation to a control observation with the nearest estimated propensity score. Panel A of

Table 3. Propensity score matching.

Panel A Covariate Balance between the Matched Pairs
Variables	Treated (1)	Control (2)	Difference (1)–(2)	t-Statistics
SIZE	21.842	21.874	−0.032	−2.27
LEV	0.448	0.450	−0.002	−0.95
ROA	0.034	0.033	0.000	0.98
GROWTH	0.187	0.182	0.004	1.04
OCF	0.049	0.048	0.001	1.09
TANG	0.247	0.249	−0.002	−1.52
AGE	2.119	2.137	−0.018	−1.56
BOARD	2.175	2.174	0.001	0.64
BALANCE	0.676	0.668	0.008	1.48
MANHOLD	0.103	0.101	0.002	1.62
Panel B Results Based on the PSM Sample
			(1)
VARIABLES			LNPATENT
BANKDIGIT			0.129 ***
			(3.448)
SIZE			−0.001
			(−0.273)
LEV			−0.322 ***
			(−9.424)
ROA			−0.513 ***
			(−5.460)
GROWTH			0.026 **
			(2.198)
OCF			−0.247 ***
			(−2.893)
TANG			−0.096 **
			(−2.094)
AGE			−0.082 ***
			(−9.945)
BOARD			0.008
			(0.771)
BALANCE			0.002
			(0.227)
MANHOLD			0.532 ***
			(13.119)
IND&YEAR			YES
Constant			0.518 ***
			(3.434)
Observations			11,226
R-squared			0.090

Note: **, and *** indicate statistical significance at 5%, and 1%, respectively. The t-statistics are reported in parentheses.

demonstrates that the differences in the means for the covariates between the treatment group and the matched control sample are nonsignificant except for firm size, suggesting a successful matching process. Finally, we re-estimate the baseline regression using the treatment samples and the matched control samples. According to Panel B of Table 3, the coefficient on BANKDIGIT remains significantly positive when the baseline regression is estimated based on the matched samples.

4.2.2. Considering the Impact of Confounding Factors

An important omitted variable in this study is the level of regional digital economy development, which acts as both an influential factor on bank digital transformation and a determinant of enterprise innovation output [37]. Without proper control, an omitted variable bias may arise.

Following Li et al. [37], we measure the level of regional digital economy development using the Peking University Digital Inclusive Finance Index. The digital inclusive finance index, DIGINCFIN, which proxies for the level of digital economy development in the firm’s locality, is controlled for in Column 1 of

Table 4. Controlling for the regional digital economy development.

	(1)	(2)	(3)
VARIABLES	LNPATENT	LNPATENT	LNPATENT
BANKDIGIT	0.139 ***	0.137 ***	0.140 ***
	(3.612)	(3.577)	(3.639)
SIZE	−0.002	−0.003	−0.002
	(−0.387)	(−0.438)	(−0.328)
LEV	−0.341 ***	−0.341 ***	−0.341 ***
	(−9.645)	(−9.652)	(−9.648)
ROA	−0.500 ***	−0.500 ***	−0.498 ***
	(−5.245)	(−5.251)	(−5.226)
GROWTH	0.029 **	0.030 **	0.029 **
	(2.409)	(2.410)	(2.391)
OCF	−0.282 ***	−0.283 ***	−0.279 ***
	(−3.210)	(−3.220)	(−3.164)
TANG	−0.064	−0.060	−0.070
	(−1.342)	(−1.262)	(−1.467)
AGE	−0.070 ***	−0.070 ***	−0.071 ***
	(−8.440)	(−8.405)	(−8.515)
BOARD	0.015	0.018	0.011
	(0.343)	(0.372)	(0.300)
BALANCE	0.009	0.009	0.009
	(0.967)	(0.983)	(0.951)
MANHOLD	0.436 ***	0.428 ***	0.517 ***
	(12.640)	(12.555)	(12.738)
NUMFIN
DIGINCFIN	0.000
	(1.302)
COVBRE		0.001 **
		(2.173)
USGDEP			0.000
			(0.135)
IND&YEAR	YES	YES	YES
Constant	0.844 ***	0.836 ***	0.862 ***
	(5.692)	(5.645)	(5.825)
Observations	12,031	12,031	12,031
R-squared	0.087	0.087	0.087

Note: **, and *** indicate statistical significance at 5%, and 1%, respectively. The t-statistics are reported in parentheses.

. Furthermore, Columns 2 and 3 of Table 4 reflect the control for the secondary digital inclusive finance indices, which refer to coverage breadth (COVBRE) and usage depth (USGDEP), respectively. Columns 1–3 of Table 4 show that the coefficients on BANKDIGIT remain significantly positive.

4.2.3. Alternative Measurement of Bank Digital Transformation

In the baseline regression, we measure bank digital transformation by the digital transformation-related patent applications. As part of the robustness testing, we employ an alternative approach to measure this indicator. Specifically, we calculate the frequency of digitization transformation-related keywords within the annual reports of listed banks, where these keywords are derived from the study conducted by Du et al. [32]. As shown in Column 1 of

Table 5. Other robustness tests.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	LNPATENT	LNPATENT	LNPATENT	PATENT	PATENT	LNPATENT
	Alternative Measurement of Dependant Variable	System GMM	Firm Fixed Effect	Poisson Model	Tobit Model	Controlling for Bank Financial Performance
L.LNPATENT		0.537 ***
		(81.103)
BANKDIGIT	0.032 **	0.079 **	0.068 **	0.981 ***	31.689 ***	0.098 *
	(2.014)	(2.400)	(1.994)	(56.221)	(5.133)	(1.823)
SIZE	−0.014 ***	0.182 ***	0.031 ***	0.107 ***	−0.340	−0.017 **
	(−3.382)	(4.577)	(3.049)	(31.805)	(−0.345)	(−1.961)
LEV	−0.299 ***	−0.484 ***	−0.209 ***	−0.821 ***	−47.491 ***	−0.265 ***
	(−12.292)	(−2.964)	(−5.119)	(−42.579)	(−8.337)	(−5.242)
ROA	−0.377 ***	−1.459 ***	−0.000	−1.736 ***	−73.112 ***	−0.506 ***
	(−5.657)	(−4.655)	(−0.003)	(−35.127)	(−4.903)	(−4.006)
GROWTH	0.022 ***	−0.050	−0.003	0.123 ***	5.649 ***	0.033 **
	(2.622)	(−1.368)	(−0.334)	(18.864)	(2.782)	(1.969)
OCF	−0.143 **	−0.157 **	−0.162 **	−0.779 ***	−27.678 *	−0.325 ***
	(−2.395)	(−2.231)	(−2.327)	(−16.166)	(−1.951)	(−2.659)
TANG	−0.037	−0.103 **	0.049	−1.019 ***	−10.705	−0.093
	(−1.141)	(−2.346)	(0.863)	(−36.858)	(−1.388)	(−1.419)
AGE	−0.098 ***	−0.248 ***	−0.306 ***	−0.170 ***	−14.391 ***	−0.084 ***
	(−16.454)	(−10.465)	(−19.831)	(−41.332)	(−11.771)	(−6.764)
BOARD	−0.000	−0.098 ***	−0.056	0.219 ***	−3.512	−0.013
	(−0.164)	(−2.849)	(−1.364)	(12.732)	(−0.685)	(−0.846)
BALANCE	0.008	0.141 *	−0.003	−0.080 ***	0.241	0.008
	(1.209)	(1.879)	(−0.226)	(−17.002)	(0.171)	(0.605)
MANHOLD	0.472 ***	0.728 ***	0.298 ***	1.021 ***	41.818 ***	0.492 ***
	(16.773)	(3.346)	(3.068)	(66.268)	(8.729)	(7.140)
BLEV						0.620 **
						(2.213)
BAGE						0.010
						(0.355)
BSIZE						−0.021 *
						(−1.878)
IND&YEAR	YES	YES	YES	YES	YES	YES
FIRM	NO	NO	YES	NO	NO	NO
Constant	0.606 ***	−2.406 ***	0.674**	−1.167 ***	−30.656	1.032 ***
	(6.187)	(−3.008)	(2.189)	(−11.883)	(−1.312)	(4.915)
AR(1)		0.032
AR(2)		0.745
Observations	6795	13,052	13,052	13,052	13,052	7969
R-squared	0.087		0.063			0.072

Note: *, **, and *** indicate statistical significance at 10%, 5%, and 1%, respectively. The t-statistics are reported in parentheses.

, the coefficient on BANKDIGIT is still significantly positive after changing the method of measuring bank digital transformation.

4.2.4. System GMM Approach

System Generalized Method of Moments (System GMM) is an advanced econometric technique that can be used in panel data settings to address issues related to endogeneity, unobserved heterogeneity, and serial correlation [38]. It is employed in econometrics, finance, and various fields to estimate dynamic models when time series data are persistent and may exhibit serial correlation over time. System GMM is known for its ability to improve the efficiency and consistency of estimates in such settings. Given that enterprise innovation performance may exhibit persistence and may exhibit serial correlation over time, we introduce the one-year lagged variable of innovation performance, L.LNPATENT, into the baseline regression and use a System GMM method for estimation. The results are shown in Column 2 of Table 5. The model passes the Arellano-Bond test, and the coefficient of BANKDIGIT is significantly positive. This indicates that after taking into account the potential serial correlation of enterprise innovation performance, the role of bank digital transformation in promoting enterprise innovation still exists.

4.2.5. Controlling for the Firm Fixed Effect and Applying the Poisson and Tobit Model

In addition to the industry and year effect controlled in the baseline regression, we further control for the firm fixed effect. The results reported in Column 3 of Table 5 do not change substantially from the baseline results. Moreover, we replace the explained variable, LNPATENT, with the number of patent applications, PATENT, and employ the Poisson and Tobit models to re-estimate the baseline regression Equation (1). As shown in Columns 4 and 5 of Table 5, the conclusion still holds.

4.2.6. Controlling for Bank Financial Performance Indicators

We include the weighted average bank financial performance indicators for each firm-year sample as additional control variables. These variables include BLEV (bank leverage, measured as the ratio of total liabilities to total assets), BAGE (bank age, measured as the natural logarithm of one plus the number of years that the bank has been established), and BSIZE (bank size, measured as the natural logarithm of the bank’s total assets). Taking BLEV as an example, we use Equation (3) to calculate this indicator for each firm-year sample as follows:

$${BLEV}_{it}={\sum }_{n=1}^N{Bank\_lev}_{int}\times \frac{{Loan}_{int}}{{Sum\_loan}_{it}}$$

(3)

BLEV_it refers to the weighted average leverage ratio of the lending banks that lend to firm i in year t. Bank_lev_int refers to the leverage ratio of bank n that lends money to company i. Loan_int represents the amount of bank loans that company i borrows from bank n in year t. Sum_loan_it represents the total amount of bank loans that company i has in year t. The other bank financial performance indicators are calculated similarly. Thus, the financial performance indicators for each bank are weighted by the size of the bank loan and averaged to the firm level to generate the bank’s financial performance indicators. These indicators are then added as additional control variables in the baseline regression. Our results remain robust after controlling for the bank’s financial performance indicators, as shown in Column 6 of Table 5.

5. Mechanism Tests

In Section 2, we introduce Hypotheses 2 and 3, asserting that bank digital transformation can enhance firm innovation performance by mitigating firms’ financial constraints and enhancing corporate governance. In this section, we empirically test these mechanisms. Previous studies usually use two methods, mediating effect analysis or grouped regression analysis, to test the influencing mechanisms [39]. In this section, the two main mechanisms through which bank digital transformation affects enterprise innovation will be tested using both the mediating effect analysis and grouped regression analysis.

5.1. Alleviating Financial Constraints

5.1.1. Mediating Effect Analysis

Bank digital transformation can help alleviate corporate financing constraints, allowing businesses to raise more money to support innovative activities and thus enhancing enterprise innovation performance. To test this mechanism, first, we use the mediating effect analysis. This is a statistical technique frequently used in management and business research to explore the underlying mechanisms that explain how an independent variable (X) influences a dependent variable (Y) through one or more mediator variables (M). This method helps researchers understand the process by which the initial effect of X on Y is transmitted through the mediator(s). A stepwise regression approach is often applied in this context to identify and test the significance of these mediation paths. We use the three-step procedure proposed by Baron and Kenny [40] to test the mediating role of financial constraint. The first step is to regress the explained variable LNPATENT (enterprise innovation performance) on the explanatory variable BANKDIGIT (bank digital transformation), as shown in Equation (4). We use Equations (5) and (6) below to test the second and third steps. Following Lamont et al. [41], we measure the level of corporate financial constraint using the KZ index and denote it as KZ. The data on the KZ index are obtained from the CSMAR database. In Equation (5), we regress the mediating variable KZ on the explanatory variable BANKDIGIT. In Equation (6), the explained variable LNPATENT is regressed simultaneously on the explanatory variable BANKDIGIT and the mediating variable KZ.

LNPATENT_i,t = α₀ + α₁BANKDIGIT_i,t−1 + α₂CONTROL_i,t−1 + ε

(4)

KZ_i,t = β₀ + β₁BANKDIGIT_i,t + β₂CONTROL_i,t−1 + ε

(5)

LNPATENT_i,t = θ₀ + θ₁KZ_i,t−1 + θ₂BANKDIGIT_i,t−1 + θ₃CONTROL_i,t−1 + ε

(6)

The results of Equation (4) are reported in Column 3 of Table 2. It shows a positive influence of bank digital transformation on corporate innovation performance. Columns 1 and 2 of

Table 6. Mediating models.

	(1)	(2)
VARIABLES	KZ	LNPATENT
KZ		−0.009 **
		(−2.063)
BANKDIGIT	−0.119 ***	0.156 ***
	(−2.640)	(4.212)
SIZE	−0.342 ***	−0.009 *
	(−37.640)	(−1.660)
LEV	5.861 ***	−0.279 ***
	(107.840)	(−6.382)
ROA	−2.283 ***	−0.541 ***
	(−17.397)	(−5.865)
GROWTH	−0.167 ***	0.029 **
	(−8.715)	(2.489)
OCF	−13.114 ***	−0.369 ***
	(−98.242)	(−3.477)
TANG	1.299 ***	−0.086 *
	(18.354)	(−1.876)
AGE	0.253 ***	−0.068 ***
	(15.893)	(−8.178)
BOARD	−0.117 ***	0.005
	(−2.635)	(0.176)
BALANCE	0.021	0.006
	(1.501)	(0.659)
MANHOLD	−0.570 ***	0.449 ***
	(−9.761)	(13.139)
IND&YEAR	YES	YES
Constant	8.081 ***	0.983 ***
	(34.957)	(6.769)
Observations	12,701	12,701
R-squared	0.790	0.090

Note: *, **, and *** indicate statistical significance at 10%, 5%, and 1%, respectively. The t-statistics are reported in parentheses.

present the regression results of Equations (5) and (6), respectively. As seen in Column 1, the coefficient of BANKDIGIT is significantly negative, showing that the level of bank digital transformation is significantly negatively correlated with the level of corporate financial constraint. This means that bank digital transformation helps reduce the financial constraints of borrowing companies. In Column 2, the coefficients of KZ and BANKDIGIT are observed to be significantly negative and significantly positive, respectively. As such, these results confirm the mediating role of corporate financial constraints.

5.1.2. Grouped Regression Analysis

Next, we conduct a grouped regression analysis to examine how the effect of bank digital transformation on enterprise innovation differs among firms with different levels of financial constraint. Jiang [39] posited that grouped regression is an important method for testing influencing mechanisms. The process first involves a theoretical analysis to suggest that X may affect Y through a specific mechanism. Next, subgroups (M = 1) where the mechanism is more likely to exist and subgroups (M = 0) where it is less likely to exist are identified. This leads to the formation of two groups of subgroups. Following this, grouped regression tests are performed. If a stronger correlation between X and Y is observed in one subgroup (M = 1), i.e., in the regression of Y on X, the absolute value of the estimated regression coefficient of X is significantly larger in one subgroup (M = 1) than the other (M = 0), it validates the mechanism through which X affects Y. Previous mainstream literature has also utilized a grouped regression analysis or similar methods to validate the influencing mechanisms [42,43].

The adoption of digital technology has resulted in a shift from a collateral-based to a credit-based lending model. Given more collateralizable assets, large enterprises have traditionally had easier access to collateralized loans through traditional banking channels, resulting in lower financing constraints and reduced sensitivity to the impact of bank’s adoption of digital technology. Small businesses, on the other hand, have insufficient collateralized assets and must take advantage of the shift in lending patterns brought about by bank’s adoption of digital technology to reduce financing restrictions and boost innovative investments. Thus, if bank digital transformation encourages enterprise innovations by easing financing constraints, then this benefit ought to be more evident for small enterprises that have more severe financing constraints.

To test the above conjecture, we partition the full sample into small and large enterprises according to their total assets. The results in

Table 7. Small versus large enterprises.

	(1)	(2)
VARIABLES	LNPATENT	LNPATENT
	Small Firms	Large Firms
BANKDIGIT	0.171 ***	0.052
	(3.380)	(1.311)
difference of BANKDIGIT	chi2 = 4.84 **
SIZE	0.043 ***	−0.011 *
	(4.196)	(−1.667)
LEV	−0.352 ***	−0.167 ***
	(−7.946)	(−4.221)
ROA	−0.505 ***	−0.498 ***
	(−4.348)	(−4.574)
GROWTH	0.036 **	0.018
	(2.266)	(1.557)
OCF	−0.271 **	−0.111
	(−2.476)	(−1.198)
TANG	−0.002	−0.166 ***
	(−0.030)	(−3.462)
AGE	−0.077 ***	−0.062 ***
	(−7.309)	(−6.604)
BOARD	−0.011	0.016
	(−0.757)	(0.930)
BALANCE	0.028 **	−0.022 **
	(2.272)	(−2.139)
MANHOLD	0.420 ***	0.403 ***
	(10.187)	(7.449)
IND&YEAR	YES	YES
Constant	−0.186	0.462 ***
	(−0.767)	(2.651)
Observations	6986	6045
R-squared	0.104	0.078

Note: *, **, and *** indicate statistical significance at 10%, 5%, and 1%, respectively. The t-statistics are reported in parentheses.

report significantly positive coefficients of the bank digital transformation variable, BANKDIGIT, for small firms in Column 1 but not significant coefficients of BANKDIGIT for large firms in Column 2. The results of the SUR (seemingly unrelated regression) estimation also demonstrate that the coefficient on BANKDIGIT is significantly higher for small enterprises than for large enterprises. These findings demonstrate that the positive effect of bank digital transformation on enterprise innovation performance is stronger for small firms than for large firms.

Firms with lower information disclosure quality have higher information asymmetry with their banks and face higher financing constraints than other firms. Thus, the role of bank digital transformation in providing more financial support for enterprise innovation by reducing information asymmetry and alleviating financing constraints should be more significant for firms with lower information disclosure quality.

We measure the quality of corporate information disclosure in two ways. First, following Dechow et al. [44], we use the absolute value of discretionary accruals, which is a widely used proxy for information quality. Second, as pointed out by studies from China, the information disclosure quality of Chinese listed companies can be measured as the quality rating of corporate information disclosure released by the stock exchanges [45]. This indicator takes values of 1, 2, 3, and 4, which correspond to ratings of A, B, C, and D (i.e., excellent, good, pass, and fail), with higher values indicating lower ratings of information disclosure quality. We use the annual median of these two indicators to split the sample and define company-year samples as having high (low) quality of information disclosure if their values are lower than (equal to or higher than) the annual median.

According to

Table 8. Low versus high quality of information disclosure.

	(1)	(2)	(3)	(4)
VARIABLES	LNPATENT	LNPATENT	LNPATENT	LNPATENT
	Higher Discretionary Accruals	Lower Discretionary Accruals	Lower Rating of Disclosure Quality	Higher Rating of Disclosure Quality
BANKDIGIT	0.176 ***	0.059	0.149 ***	0.019
	(3.774)	(1.405)	(3.910)	(0.273)
difference of BANKDIGIT	chi2 = 3.92 *		chi2 = 3.75 *
SIZE	−0.006	0.002	−0.003	−0.010
	(−0.809)	(0.235)	(−0.609)	(−0.931)
LEV	−0.249 ***	−0.350 ***	−0.295 ***	−0.226 ***
	(−5.956)	(−8.062)	(−8.833)	(−3.181)
ROA	−0.478 ***	−0.659 ***	−0.457 ***	−1.035 ***
	(−4.862)	(−4.078)	(−5.140)	(−5.114)
GROWTH	0.034 ***	0.005	0.017	0.040 **
	(2.619)	(0.301)	(1.534)	(1.983)
OCF	−0.196 **	−0.081	−0.281 ***	0.000
	(−2.138)	(−0.595)	(−3.363)	(0.002)
TANG	−0.093	−0.091	−0.086 *	−0.066
	(−1.586)	(−1.600)	(−1.910)	(−0.696)
AGE	−0.073 ***	−0.078 ***	−0.069 ***	−0.089 ***
	(−7.773)	(−6.930)	(−8.872)	(−5.242)
BOARD	0.015	−0.017	−0.018	0.052
	(0.748)	(−0.813)	(−0.887)	(1.621)
BALANCE	0.006	0.003	0.008	0.004
	(0.477)	(0.298)	(0.842)	(0.223)
MANHOLD	0.433 ***	0.526 ***	0.561 ***	0.402 ***
	(9.427)	(10.686)	(12.935)	(5.272)
IND&YEAR	YES	YES	YES	YES
Constant	0.637 ***	0.506 ***	0.598 ***	0.899 *
	(3.349)	(2.891)	(4.043)	(1.687)
Observations	6919	6042	9307	2486
R-squared	0.095	0.095	0.086	0.138

Note: *, **, and *** indicate statistical significance at 10%, 5% and 1%, respectively. The t-statistics are reported in parentheses.

, the coefficients of BANKDIGIT are significantly positive in Columns 1 and 3 (the group with low information disclosure quality) but not significant in Columns 2 and 4 (the group with high information disclosure quality). The SUR estimation results show that the coefficient on BANKDIGIT is significantly higher in the group with low information disclosure quality than in the group with high information disclosure quality. These results indicate a more significant role of bank digital transformation in improving enterprise innovation performance for firms with lower information disclosure quality.

5.2. Improving Corporate Governance

When corporate governance is weak and the misuse of capital by major shareholders and firm management leads to insufficient financial support for innovation activities, there is more scope for digital technology to improve the corporate governance roles of banks and curb the self-interested behaviors of major shareholders and management so that firms can allocate more capital to support innovation. Thus, if bank digital transformation enhances enterprise innovation by improving corporate governance, we may anticipate that the beneficial impact of bank digital transformation on enterprise innovation will be greater for companies with weak governance.

Following La Porta et al. [46] and Ang et al. [47], we measure the quality of corporate governance by the divergence between the control and cash-flow rights of the controlling shareholder and the ratio of operating expense to sales revenue. Next, we use the annual median of the divergence between control and cash-flow rights, as well as the ratio of operating expense to total assets, to split the sample. A company-year sample is defined as having weak (strong) corporate governance if the divergence between control and cash-flow rights or the ratio of operating expense to total assets is higher than or equal to (lower than) the respective annual median.

As shown in

Table 9. Weak versus strong corporate governance.

	(1)	(2)	(3)	(4)
VARIABLES	LNPATENT	LNPATENT	LNPATENT	LNPATENT
	Higher Divergence	Lower Divergence	Higher Operating Costs	Lower Operating Costs
BANKDIGIT	0.202 ***	0.021	0.205 ***	0.037
	(4.274)	(0.431)	(3.689)	(1.129)
difference of BANKDIGIT	chi2 = 6.66 ***		chi2 = 6.25 **
SIZE	0.002	−0.008	0.015 *	−0.012 **
	(0.260)	(−1.184)	(1.853)	(−2.187)
LEV	−0.242 ***	−0.339 ***	−0.343 ***	−0.121 ***
	(−6.034)	(−7.538)	(−7.394)	(−3.244)
ROA	−0.499 ***	−0.507 ***	−0.601 ***	−0.133
	(−4.652)	(−4.161)	(−5.168)	(−1.177)
GROWTH	0.014	0.032 **	0.028 *	0.028 **
	(1.065)	(2.155)	(1.650)	(2.532)
OCF	−0.169 *	−0.226 **	−0.260 **	−0.177 **
	(−1.735)	(−2.066)	(−2.170)	(−2.124)
TANG	−0.056	−0.163 ***	−0.081	−0.069
	(−1.006)	(−2.699)	(−1.211)	(−1.469)
AGE	−0.091 ***	−0.081 ***	−0.108 ***	−0.046 ***
	(−10.039)	(−6.682)	(−9.458)	(−5.618)
BOARD	−0.005	0.012	−0.010	0.016
	(−0.246)	(0.501)	(−0.299)	(0.721)
BALANCE	0.009	0.003	0.001	0.006
	(0.806)	(0.205)	(0.055)	(0.567)
MANHOLD	0.226 ***	0.607 ***	0.528 ***	0.325 ***
	(3.150)	(13.033)	(10.915)	(7.580)
IND&YEAR	YES	YES	YES	YES
Constant	0.387 **	0.800 ***	0.402 **	0.414 **
	(2.118)	(4.330)	(1.964)	(2.507)
Observations	6,433	6,322	6,527	6,272
R-squared	0.077	0.126	0.106	0.065

Note: *, **, and *** indicate statistical significance at 10%, 5%, and 1%, respectively. The t-statistics are reported in parentheses.

, the coefficient of BANKDIGIT is significantly positive in Columns 1 and 3 (the group with weak corporate governance) but not significant in Columns 2 and 4 (the group with strong corporate governance). Then, the SUR model is used to test if the difference in the coefficient of BANKDIGIT is significant among the two groups. According to the chi2 statistics, the coefficients of BANKDIGIT are significantly larger for the group with weak corporate governance than the group with strong corporate governance. Thus, we have found a more prominent role of bank digital transformation in boosting enterprise innovation for firms with weak corporate governance.

6. Additional Tests

6.1. Different Measurement of Enterprise Innovation

When estimating the baseline regression, we measure enterprise innovation output using the logarithm of one plus the number of total patent applications made by firms, which includes invention patents, utility model patents, and design patents. Invention patents are believed to better reflect the quality of innovation than the other two types of patents, to some extent [48]. Therefore, in this section, we now examine whether bank digital transformation has an enhancing effect on the number of invention patent applications. We also use the number of patent grants, which are certified by the National Patent Office, to measure innovation output. In addition to innovation output, innovation efficiency is also suitable for representing enterprise innovation performance and is thus an effective supplement to innovation output.

We define INVPATENT (GRAPATENT) as the natural logarithm of one plus the number of invention patent applications (patent grants). Following Guan et al. [34], we measure innovation efficiency as the number of total patent applications divided by the average R&D input in the previous four years and denote it as INNEFF. These three variables are used to replace the explained variable, LNPATENT, in the baseline regression Equation (1).

As shown in Columns 1–3 of

Table 10. Additional tests.

	(1)	(2)	(3)	(4)
VARIABLES	INVPATENT	GRAPATENT	INNEFF	LNPATENT
BANKDIGIT	0.108 ***	0.117 ***	0.009 *	0.242 ***
	(3.854)	(3.583)	(1.862)	(4.624)
FDIGIT*BANKDIGIT				−0.069 ***
				(−2.602)
FDIGIT				0.048 ***
				(8.634)
SIZE	0.002	−0.008 *	−0.000 ***	−0.013 **
	(0.511)	(−1.647)	(−4.821)	(−2.405)
LEV	−0.282 ***	−0.287 ***	−0.001 **	−0.362 ***
	(−11.458)	(−10.070)	(−2.464)	(−11.190)
ROA	−0.375 ***	−0.462 ***	−0.003 **	−0.528 ***
	(−5.578)	(−5.936)	(−2.271)	(−5.987)
GROWTH	0.023 ***	0.012	0.001 ***	0.025 **
	(2.668)	(1.205)	(3.961)	(2.250)
OCF	−0.104 *	−0.105	−0.005 ***	−0.259 ***
	(−1.675)	(−1.467)	(−3.105)	(−3.176)
TANG	−0.110 ***	−0.129 ***	−0.001	−0.064
	(−3.347)	(−3.365)	(−1.397)	(−1.461)
AGE	−0.063 ***	−0.079 ***	0.000 **	−0.096 ***
	(−11.041)	(−11.889)	(2.058)	(−12.684)
BOARD	0.006	−0.003	−0.006	−0.000
	(0.478)	(−0.170)	(−0.367)	(−0.029)
BALANCE	0.013 *	0.016 **	−0.000	0.008
	(1.897)	(2.039)	(−1.290)	(0.874)
MANHOLD	0.326 ***	0.351 ***	0.092 ***	0.498 ***
	(10.764)	(11.142)	(7.789)	(13.444)
IND&YEAR	YES	YES	YES	YES
Constant	0.540 ***	0.836 ***	0.014 ***	1.075 ***
	(5.424)	(7.258)	(5.287)	(8.159)
Observations	13,052	13,052	12,887	12,875
R-squared	0.090	0.092	0.072	0.100

Note: *, **, and *** indicate statistical significance at 10%, 5%, and 1%, respectively. The t-statistics are reported in parentheses.

, the variable BANKDIGIT is positively correlated with innovation output, measured by the number of invention patents and patent grants. Additionally, it is positively correlated with innovation efficiency, measured by the proportion of patent output to total R&D input. These findings suggest that bank digital transformation plays a favorable role in enhancing the number of invention patent applications, number of patent grants, and innovation efficiency.

6.2. The Role of Corporate Digital Transformation

Finally, we explore how firms’ own digital transformation interacts with bank digital transformation in promoting enterprise innovation. There are two possibilities. First, firms’ own digital transformation and bank digital transformation may complement each other in improving enterprise innovation. Digital technology strengthens banks’ information processing ability when making loan decisions, which ultimately lowers the entry threshold for access to bank loans and enables banks to provide more financial support for enterprise innovation activities. Such an effect is more significant for firms with a higher level of digital transformation itself and thus accumulate more data available for banks’ information processing. If this is true, we anticipate that the positive relation between bank digital transformation and firms’ innovation output increases with firms’ own level of digital transformation.

Second, in contrast, firms’ digital transformation may substitute bank digital transformation in their effect on enterprise innovation. Prior studies have documented that a firm’s own digital transformation promotes its innovation performance [49,50]. Given this, it could be expected that firms’ digital transformation may lower the marginal effect of bank digital transformation on enterprise innovation. When firms’ level of digital transformation is higher, the room left for bank digital transformation to encourage innovation is relatively limited. Thus, the positive relation between bank digital transformation and firms’ innovation performance is less pronounced for firms with higher digital transformation. Taken together, how firms’ own digital transformation interacts with bank digital transformation in promoting enterprise innovation appears to be an empirical question.

To clarify this empirical question, we measure the level of corporate digital transformation by the frequency of digitization-related keywords in the annual report and denote it as FDIGIT, where digitization-related keywords are obtained from Du et al. [32]. Next, we add the interaction term between FDIGIT and BANKDIGIT to the baseline regression Equation (1). The results in Column 4 of Table 10 show that the coefficient on FDIGIT*BANKDIGIT is significantly negative, which confirms that firms’ own digital transformation may act as a substitute for bank digital transformation in their effect on enterprise innovations.

7. Conclusions

7.1. Summary of the Findings

We empirically test the impact of bank digital transformation on enterprise innovation by collecting data on digital transformation-related patents applied by China’s commercial banking sector from 2010 to 2020 to measure the bank digital transformation indicator. We find that bank digital transformation has a favorable impact on enterprise innovation output, as shown by firms’ patent applications. The mechanism tests’ findings imply that bank digital transformation may promote company innovation by easing financial restrictions and enhancing corporate governance. This is illustrated by not only the mediating role of the KZ index but also the more positive effect of bank digital transformation on enterprise innovation for firms with greater financial constraints and firms with weak corporate governance. Further research shows that bank digital transformation also helps support enterprise innovation efficiency (as measured by the proportion of patent output to total R&D input) and enterprise innovation output (as measured by firms’ invention patent applications and patent grants). Additionally, firms’ digital transformations substitute for bank digital transformation in terms of their impact on enterprise innovations.

7.2. Discussion

Our study directly investigates the impact of bank digital transformation on the innovation performance of borrowing enterprises. From one perspective, our study both parallels and diverges from previous literature in terms of the factors affecting corporate innovation performance. We find common ground with existing research in highlighting that macro-level regional digital economic development and enterprises’ own digital transformations can bolster corporate innovation performance. Yet, when we delve into the underlying mechanisms at play, our study sets itself apart. The influencing mechanism of bank digital transformation on enhancing the innovation performance of borrowing enterprises diverges from that associated with regional digital economic development and firms’ internal digital transformation. In our analysis, we propose that bank digital transformation enhances the innovation performance of borrowing companies by mitigating their financing constraints and bolstering debt governance. Firstly, though earlier literature also hints at the role of regional digital economic development in alleviating financing constraints to enhance innovation [37,51], this effect is somewhat indirect. In contrast, our study places the spotlight on bank digital transformation, which offers a more direct avenue to alleviate financing constraints for borrowing companies since banks serve as the immediate fund providers. Moreover, bank digital transformation not only alleviates financing constraints but also reinforces debt governance on borrowing firms since they serve as relational creditors to these firms [22]. Secondly, prior research highlights that firms’ internal digital transformation predominantly amplifies innovation performance by streamlining internal operational processes, such as production, management control, and product development [11,52]. This stands in contrast to the impact mechanism we explore in the context of bank digital transformation.

On the other hand, previous literature has primarily explored the impact of bank digital transformation on its own operations and performance [6,7,8,9,10], whereas our study takes a distinct perspective by delving into the contribution of bank digital transformation to borrowing companies and the broader real economy. While the focus of the research is on the impact of bank digital transformation on borrowing companies, it indirectly implies potential benefits for banks themselves. Bank digital transformation not only fosters more innovative borrowing companies but also contributes to banks’ own competitiveness and sustainability. By effectively addressing the financing constraints of their borrowing firms and improving corporate governance, banks can build stronger, more reliable relationships with their clients. This, in turn, may lead to reduced default risks and a healthier loan portfolio for the banks. Furthermore, as borrowing companies thrive in their innovative endeavors, they may become more attractive prospects for loans, enabling banks to expand their client base and boost their financial performance. In this regard, our research aligns with previous studies on the economic implications of digital transformation in banking. However, we broaden the scope by examining the larger issue of how digital transformation in banking supports the real economy.

8. Implications, Limitations, and Future Directions

8.1. Implications

This study fills a critical research gap by directly investigating the impact of bank digital transformation on the performance and behavior of borrowing enterprises. While prior literature has predominantly focused on the effects of bank digital transformation on the banks themselves, this research delves into a relatively unexplored dimension, shedding light on the direct influence of bank digital transformation on borrowing enterprises. The findings provide valuable insights into whether and how bank digital transformation serves the real economy.

We utilize an innovative approach to gauge the extent of bank digital transformation. Instead of relying on the frequency of digital transformation-related keywords in annual reports, as often is the case in previous studies, we employ data on digital transformation-related patent applications filed by banks. This methodology offers several advantages, including a more tangible reflection of a bank’s digital innovations, objectivity, and applicability to both listed and non-listed banks, thus presenting a more comprehensive assessment of digital transformation.

Another important theoretical implication of this study is the exploration of how bank digital transformation interacts with the digital transformation of their enterprise customers. The existing literature has primarily concentrated on single digital transformation entities, such as listed firms or commercial banks. This study broadens the scope of inquiry, opening new avenues for understanding the interplay between bank digital transformation and the digital transformation efforts of their enterprise customers.

Overall, these theoretical implications contribute to a deeper understanding of the economic consequences of bank digital transformation and offer innovative perspectives for future research in this domain.

Apart from its theoretical implications, our study also has practical implications for commercial banks and policymakers. Commercial banks should take two key steps. First, they need to enhance their own innovation leadership position by accelerating patent drafting in critical areas and increasing R&D spending on the underlying technology. Additionally, they should encourage the creation of systematic patent technology from the ground up. Second, commercial banks should actively encourage the digital transformation of their traditional financial service divisions, which would help modernize these divisions and ensure that they can meet the evolving needs of businesses. Through this digital transformation, banks can link their digital departments with their business technology, thus empowering them to effectively allocate the credit resources necessary for enterprise innovation. For the government, first, it should actively build a new digital infrastructure to provide the technical foundation for banking and other financial institutions to carry out digital transformation. This can be achieved, for example, by building a data-sharing platform to further release the empowering value of digital technology for financial businesses. Second, it should provide institutional support for banks’ digital transformation and enhance the space for innovation application and innovation risk tolerance.

8.2. Limitations and Future Directions

This study’s primary limitation lies in its specific focus on the Chinese banking industry. Therefore, the generalizability of these findings to other regions or countries with different economic and regulatory environments may be limited. Future research can undertake cross-country comparative analyses to investigate how bank digital transformation affects borrowing firms in different contexts, allowing for the identification of common trends and unique regional variations.

Secondly, this study primarily investigates the influence of bank digital transformation on corporate innovation without delving into the broader economic implications, including aspects like corporate investment efficiency, cash holding, and liquidity management. Future research can investigate the wider economic consequences of bank digital transformation to gain a more comprehensive understanding of its effects on borrowing enterprises and the economy.

Finally, to establish the correspondence between banks and their borrowing firms, we use the bank loan data for listed companies from the China Stock Market Accounting Research (CSMAR) database. The loan data of listed companies in the CSMAR database is sourced from the designated information disclosure website of the China Securities Regulatory Commission, known as CNINFO (http://www.cninfo.com.cn). This data source is both authoritative and reliable. Other studies have also extensively relied on the loan data of listed companies from the CSMAR database [53,54]. Nevertheless, it is important to note that there is a possibility that some listed companies may not have disclosed all of their loan information. Consequently, this study may have a potential limitation in the data. Future research can focus on broader and more extensive data sources, including a wider variety of banks and firms, to provide a more holistic understanding of the effects of digital transformation.