Digital transformation has the potential to overshadow a company’s commitment to environmental sustainability. The process of digitalization often requires significant investments in assets, personnel, materials, and funds. Similarly, the implementation of green strategies within a company, aimed at achieving energy conservation, emission reduction, and pollution prevention (Gee and McMeekin, 2011; García-Pozo et al., 2015), demands investments in new technologies, knowledge, and innovative ideas. As a result, there may be a conflict between utilizing digital resources and investing in green initiatives. Given that the benefits of green investments often take a significant amount of time to materialize (Chen and Ma, 2021), combined with the performance and cash flow challenges associated with green development, companies may prioritize limited innovation resources towards digitalization rather than greenization. Unfortunately, this can lead to a shortage of green resources necessary for driving sustainable transformations.

Digital transformation has the potential to weaken a firm’s ability to discover, integrate, and utilize resources. While digital technology can generate and unlock value for businesses, it can also result in a significant increase in external transaction costs and internal organizational costs (Guo et al., 2023), ultimately hindering the improvement of resource search capabilities and damaging the identification of firm resources. Furthermore, digital transformation may negatively impact the ability to adjust or reconfigure resources. Although companies can monitor and analyze real-time data on resource procurement channels, price fluctuations, and supply and demand conditions through the use of data analysis models and computational algorithms, the adoption of new technologies may lead to a new monopoly where entities with higher digital power in the innovation ecosystem may more easily acquire and integrate resources (Li et al., 2016), ultimately resulting in unfair resource allocation. Similarly, when digital transformation is too broad and too fast, existing organizational capabilities may not meet the requirements (Gebauer et al., 2020), leading to underutilization or even waste of resources. In such situations, enterprises may be reluctant to invest in green resources considering the externality and public goods attribute of green activities (Chen et al., 2022). Therefore, we propose the following hypothesis:

H1

Both BODT and DODT are negatively correlated with IOCG overall.

Digital transformation can bring the green resources needed for green transformation. From a resource discovery standpoint, digital transformation facilitates enterprises in establishing and expanding their value networks, and a large amount of information and data are constantly exchanged between enterprises and among enterprises’ internal entities, reducing their reliance on complementary green resources to a large extent (He et al., 2023). Simultaneously, digitalization promotes the upgrading of enterprise value chains and optimization of industrial structures (Li X. et al., 2023; Zhang W. et al., 2023; Wang J. et al., 2023), creating conditions for firms to obtain green resources from a wider range of channels. In terms of resource transfer, the use of digital and intelligent technologies helps firms build digital supply chains with partners, creating digital platforms among internal entities, which not only alleviates information asymmetry but also greatly reduces the cost of knowledge sharing (Zhang Q. et al., 2022). Regarding resource absorption, based on new technologies such as artificial intelligence and big data, firms can extract valuable information from large-scale, real-time interactive data, strengthening risk control in the process of resource utilization. Additionally, digitization helps enterprises to update innovative resources in real-time or restructure the relationship architecture between resources. According to the resource-based theory, the discovery, transfer, and absorption of green resources play a powerful role in promoting enterprises to achieve green innovation goals.

Digital transformation may also enhance green innovation capabilities and reduce the costs of green product development (Yang et al., 2022). On one hand, digital transformation encourages enterprises to reduce the production cost of green products through routine updates, resource bricolage, and synergy effects. Routine update is an essential process involved in the digitalization of enterprises, which entails modifying and revamping existing procedures and patterns. The combination of emerging technologies and production processes prompts companies to update or adopt new management models when offering products to customers. This change in conventions creates favorable conditions for firms to engage in green innovation, such as “source reduction” and “end-of-pipe cleaning” (Liu X. et al., 2023), ultimately accelerating the speed of green transformation. Resource bricolage refers to the strategic identification and integration of resources by utilizing modules and interface rules offered by digital platforms, helping to alleviate institutional and resource constraints. Compared to acquiring resources from external networks or creating unique resources, embedding in digital platforms ensures resource specificity and reduces acquisition costs. Synergy effects occur when firms and other entities cooperate and gain benefits that exceed what they would achieve as independent units. Within an ecosystem, network effects facilitate the sharing of product design inspiration between firms and other entities, driving iterative improvements based on market response and ultimately reducing the cost of acquiring new ideas.

On the other hand, digital transformation has the potential to enhance firms’ capabilities for green innovation through the renewal of their knowledge base, iteration of intellectual capital, and innovation in their business models. Knowledge base renewal involves the reconstruction and updating of skills, knowledge, tools, and other resources within a firm, facilitated by digital technologies, to meet the specific needs and requirements of the firm. Throughout the process of digital transformation, enterprises can acquire new skills for task completion, learn novel methods of value creation, and even generate entirely new knowledge. According to the resource-based theory, the updating of the knowledge base is expected to significantly expand enterprises’ capacity for green innovation, ultimately enabling the realization of green transformation. Intellectual capital iteration refers to the evolution of human capital, organizational capital, and relational capital driven by the application of digital technologies. Serving as a powerful driver of the synergistic transformation of various elements within firms, digital change and innovation are accompanied by continuous investments in intellectual capital (Li J. et al., 2023). The investment in intellectual capital will inevitably enhance the innovation capacity of green processes and products, thereby exerting a positive influence on the greening of enterprises. Business model innovation entails the adjustment of the value creation process based on changes in business logic, supported by digital technology. As digital transformation continues to advance, firms have the opportunity to innovate their business models based on customer feedback, surpassing previous experiences and limited rationality. Aligned with business model innovation, the capacity for green innovation is strengthened, ultimately overcoming barriers to the transformation into environmentally friendly enterprises. Therefore, we propose the following hypothesis:

H2

Both BODT and DODT are positively correlated with OOCG overall.

BODT and DODT may have distinct influences on the greening of enterprises. BODT, which stands for breadth of digital asset deployment (Ye et al., 2022), refers to the number of technologies and tools adopted by enterprises in their digitization efforts (Yang Z. et al., 2023). On the other hand, DODT, or depth of digital asset deployment (Ye et al., 2022), reflects the level of attention and resource investment dedicated to digital asset deployment. While BODT indicates the adoption of specific digital technologies, DODT signifies the extent to which the potential of these technologies is fully utilized (Blichfeldt and Faullant, 2021). When a technology is fully leveraged, it demonstrates a firm’s solid foundation in knowledge and expertise, acting as a critical catalyst for driving innovation and progress within the firm (Katila and Ahuja, 2002). Regarding the relationship between digitalization and greenization, a broader BODT enables firms to explore more combination innovations in green processes and products. Conversely, a deeper DODT suggests that firms are more familiar with new technologies and possess a stronger knowledge base tailored to these technologies (Blichfeldt and Faullant, 2021), increasing the likelihood of implementing green innovation strategies. Furthermore, some studies have compared the effects of BODT and DODT, finding that DODT has a stronger impact. For instance, in the context of the relationship between digital asset deployment and supply chain agility, the depth of digital asset deployment has a significant positive effect on supply chain agility, rather than the breadth of deployment (Ye et al., 2022). Based on the above, we propose the following hypothesis:

H3

Compared with BODT, the relationship between DODT and IOCG/OOCG is stronger.

The presence of slack resources can help mitigate conflicts and tensions between advocates of digital transformation and proponents of green investment. As self-interested individuals with different priorities, the concerns of digital transformation advocates and green investment supporters may not align, leading to competition for limited corporate innovation resources. In situations where resources are scarce, these stakeholders may engage in bargaining and political struggles, potentially diverting management decisions away from strategic goals and towards serving their own interests (Moch et al., 1977). However, when an organization has abundant slack resources, the needs and demands of both digital transformation initiatives and green investment proponents can be more adequately met. With sufficient resources available, the intensity of their competition for resources is reduced. This, in turn, weakens the negative impact of digital transformation on the implementation of green initiatives, as both parties can find a more harmonious balance between their respective goals.

The presence of surplus resources can influence an organization’s inclination towards adopting digital and green solutions, as well as the level of commitment exhibited by managers towards digital and green risks. On one hand, when considering performance objectives, digital transformation primarily serves short-term performance goals, whereas greening initiatives aim to achieve harmonious development between the organization and the environment. In situations where resource limitations are a concern, enterprises may prioritize enhancing short-term performance and hesitate to allocate necessary resources towards greening efforts. Conversely, in resource-abundant scenarios, enterprises are more inclined to invest substantial funds, even if greening initiatives do not yield immediate results. This shift occurs because the selection criteria for enterprise transformation plans transition from “maximization” to “satisfactory” (Simon, 1967). On the other hand, both digitalization and greening initiatives entail significant risks for organizations. When faced with the choice between the two, companies may prioritize digitalization over greening. However, when enterprises possess surplus resources, they need not excessively worry about the high risks associated with innovation and change (Baird and Thomas, 1985; Singh, 1986). This allows them to consider both digital innovation and green development, rather than neglecting one in favor of the other.

Slack resources help enterprises to reduce resource crowding and misappropriation brought by digital transformation while advancing digitalization in an orderly manner under uncertainty. By acting as a “buffer” for both the organization and the environment, organizational slack create a margin of improvement that allows the enterprise to navigate environmental changes more effectively (Levinthal and March 1981). This, in turn, enables the enterprise to gradually or cautiously implement digital transformation initiatives. Consequently, the core operations of the enterprise experience minimal disruption (Thompson, 1967), and the overall shock caused by the transformation is mitigated, thereby reducing the adverse effects of digital transformation on greening efforts. Based on these observations, the hypothesis is proposed:

H4

Both USR and ASR play a negative moderating role in the impact of BODT/DODT on IOCG, that is, the more ASR and USR, the weaker the negative correlation between BODT/DODT and IOCG.

While slack resources can provide certain benefits, it is important to acknowledge that they can also exacerbate information asymmetry and create conditions that encourage executives to engage in moral hazard behaviors to serve their own interests. As a buffer, redundant resources introduce a barrier between external investors and executives, making it more challenging for investors to observe and predict the utilization of green resources (Zhu et al., 2022). Particularly in situations where there is an expectation gap or organizational decline, idle resources can transmit ambiguous or even misleading signals to stakeholders, reducing the regulatory pressure faced by executives and weakening their sense of urgency in implementing corporate strategies (Wang et al., 2016). Furthermore, these surplus resources enable executives to allocate resources towards the development of innovative products in a manner that prioritizes their personal interests, without adequately considering the overall economic interests of the company (Child, 1972). As agents operating under the separation of ownership and control, executives are motivated to pursue personal goals such as power and prestige. In the presence of slack resources, they may become overly confident or excessively optimistic, evading corporate responsibilities related to environmental protection and sustainable development (John et al., 2008). They may disregard the long-term value that can be derived from the enterprise’s green innovation (Tan and Peng, 2003), leading to inefficient resource utilization, wastage, and ultimately, a negative impact on the enterprise’s capacity to produce green patents and green products (Arena et al., 2018).

In addition to exacerbating information asymmetry and moral hazard, slack resources also increase the discretion of executives, resulting in higher corporate governance costs. On one hand, while these resources can provide protection against external environmental impacts (Liu et al., 2014), they also enable top management teams (TMTs) to exercise greater decision-making autonomy. When executives harness innovative resources, they often rely on their instincts rather than engaging in trial-and-error to identify the most effective resource utilization strategies. Consequently, in the context of digital transformation, the presence of slack resources may lead managers to over-diversify their resource utilization practices (Salge and Vera, 2013). This not only diminishes their ability to effectively search for, integrate, and utilize resources but also increases the management costs associated with resource transfer, allocation, and related procedures. As a result, the positive impact of digital transformation on OOCG is weakened. On the other hand, the existence of slack resources requires enterprises to invest significant time and energy in transforming these resources into products or services, which also increases the costs for stakeholders to supervise the utilization of these resources to a certain extent. When the scope of resource integration extends beyond firm boundaries or becomes excessively large, the time and cost associated with resource selection, integration, and utilization escalate. This may even result in diseconomies of scale and scope in technology application (Breschi et al., 2003). Moreover, when enterprises possess excess resources, executives gain more opportunities to prioritize their own “favored projects,” potentially impeding resource owners or providers from effectively monitoring executive behavior due to false resource utilization (Li-Ying et al., 2014). Consequently, agency costs rise, significantly diminishing the role of digital transformation in promoting environmentally sustainable output. Therefore, we propose the following hypothesis:

H5

Both USR and ASR play a negative moderating role in the impact of BODT/DODT on OOCG, that is, the more ASR and USR, the weaker the positive correlation between BODT/DODT and OOCG.

Both ASR and USR have an impact on the relationship between digitalization and greenization, but their effects may differ. ASR are resources that exist for a specific business or process during product production and are difficult to use for other purposes once they are applied to the production process of a product (Sharfman et al., 1988). In contrast, USR are not limited to a specific technical field, business operations, or production management (Tan and Peng, 2003) and can be used for more innovative activities. In the context of digitalization influencing green development, ASR has a higher degree of specificity, greater coordination cost, and is more likely to generate resource stickiness and structural rigidity (Greve, 2003). When resources are excessively invested in business operations, it is more likely to produce a “siphon effect,” which can lead to enterprises falling into a dilemma of resource allocation. Therefore, the moderating role of ASR may be greater than that of USR. On the other hand, ASR has a lower conversion efficiency than USR, as it is less flexible and liquid, and its utilization and transformation also require higher costs. When a firm faces organizational inertia (Henderson and Clark, 1990), inadequate management incentive systems, or internal control failures (Hitt et al., 1991), ASR may exhibit a higher degree of inefficient allocation. Thus, compared to USR, ASR has a stronger moderating effect. Based on these considerations, we propose the following hypothesis:

H6

Compared with USR, ASR has a stronger moderating effect.

Corporate greenization is characterized by a dynamic and evolving process. In this study, we adopt the classification proposed by Qi et al. (2023) and categorize it into two dimensions: IOCG and OOCG. IOCG, as defined by Huang and Lei (2021), includes the investments and expenditures made by enterprises in environmental protection equipment, cleaning technology, and pollution control measures. To address the inherent right-skewness in green investment data (Fan et al., 2023), we apply a natural logarithm transformation after adding 1 to the total green investment of listed companies. Regarding OOCG, we adopt the analytical perspective put forth by He et al. (2023) and Feng et al. (2022), which utilizes the number of green patent applications as a proxy measure. The choice of using the number of green patent applications is motivated by its intuitive reflection of the efficiency of green resource utilization by enterprises. Furthermore, it is considered more stable, reliable, and timely compared to patent grant data (Li and Shen, 2021). Specifically, we measure OOCG by considering both the number of green patents applied for by individual companies and those applied for jointly with other companies.

Previous literature has employed three distinct methodologies to measure digital transformation. The first approach involves using a binary variable to indicate whether an enterprise has undertaken digital transformation (Wang C. et al., 2023). However, this method has been criticized for its limited ability to comprehensively capture the extent of organizational transformation (Fan et al., 2023). The second method involves selecting a restricted set of financial or non-financial indicators (Wen et al., 2021), which may not fully reflect the multi-faceted nature of digital transformation. The third method involves administering questionnaires to subjects, but this approach has limitations in terms of objectivity.

In line with the methodology proposed by Li J. et al. (2023) and Zhang G. et al. (2023), this study employs a textual analysis of annual reports from publicly traded companies to evaluate the content and extent of digital transformation. The procedural sequence is as follows: Firstly, a digital transformation text database is created. To ensure the comprehensiveness, accuracy, and authority of the texts, we utilized a compilation of texts from Chinese and English literature, as previously explored by Wu et al. (2021) and Ren et al. (2022). These articles encompassed various sources of text, including policy documents, research reports, academic papers, and corporate annual reports. Notably, Wu et al. (2021) extracted text from sources such as the “Special Action Plan for Digital Enablement of SMEs”, the “2020 Digital Trends”, and the government work report. Subsequently, pairwise comparisons were conducted on the collected texts, resulting in the identification of 120 non-duplicate keywords associated with digital transformation. To illustrate the interrelationships among these keywords, we categorized them into two groups: “underlying technology” and “practical application.” The “underlying technology” category encompasses texts related to four key technologies, namely, artificial intelligence, big data, cloud computing, and blockchain. On the other hand, the “practical application” category includes texts describing new tools, patterns, or phenomena arising from the integration of these technologies with various business or scenario contexts. In summary, the aforementioned five types of keywords, totaling 120, constitute the digital transformation text database. Table 1 presents the keywords associated with digital transformation.

Secondly, the annual reports of publicly traded companies were manually collected. Annual reports serve as authoritative documents disclosed by listed companies, providing insights into their development direction and strategic decisions (Donovan et al., 2021). Therefore, we utilized these reports as a foundation for assessing the digital transformation of listed companies. Initially, we downloaded the annual reports from various platforms (e.g., eastmoney.com) and then cross-referenced them with the stock codes in the CSMAR database to identify any missing reports. Subsequently, we conducted a thorough search for the corresponding annual reports of the companies with missing reports, supplementing our dataset accordingly.

Thirdly, we employed the Jieba function in Python for word segmentation. Utilizing a language model, we identified text sections within the company’s annual reports that were relevant to digital transformation. Subsequently, we performed segmentation on the identified text, eliminating insignificant modal particles, conjunctions, and duplicate content.

Finally, the segmented text is compared against the digital transformation text database, and the corresponding statistical results are utilized to measure BODT and DODT. In measuring BODT, if the segmented company annual report text does not contain any digital transformation keywords, a value of 0 is assigned. If there is one digital keyword present, a value of 1 is assigned. Similarly, if there are two digital keywords, a value of 2 is assigned, and so on. In other words, BODT is determined by the number of digital transformation texts disclosed within a company’s annual report.

Yang Z. et al. (2023) utilized the summation of digital technology-related phrases divided by the number of digital technology types implemented by enterprises to measure DODT. They posited that this approach identified the average effort of enterprises in each digital technology. However, this method solely reflects the degree of emphasis and resource allocation of a single enterprise towards digitalization, without considering the overall impact of digital transformation and potential spillover effects. To achieve a more accurate and comprehensive measurement, we propose an enhancement to Yang Z. et al. (2023)’s method by employing the following formula: D O D T i , t = n i , t ∑ n i , t (1)

In formula 1, n i , t is the number of digital transformation texts disclosed by listed company i in its annual report for year t; ∑ n i , t is the sum of the digital transformation texts disclosed by all listed companies in their annual reports in year t.

Based on theoretical analysis, we introduce slack resources as moderating variables, specifically USR and ASR. Following the approach of Iyer and Miller (2008), Peng et al. (2010) and Wu and Hu (2020), we employ the quick ratio to quantify USR. The quick ratio is calculated as the disparity between current assets and inventories divided by current liabilities. As for ASR, it is assessed by determining the ratio of the total management expenses and sales expenses to the operating revenue.

To control for various factors that may influence enterprise greenization, we consider their own attributes, financial performance, and governance arrangements, as outlined by (Wang C. et al., 2023). The own attributes we examine are age (Age) and size (Size). Age is included to control age-related effects (Czarnitzki and Hottenrott, 2011), while size is considered to address the scale effect (Xiang et al., 2022a). Regarding financial performance, we incorporate the following variables: leverage ratio (Lev), return on assets (Roa), cashflow (Cashflow), growth rate (Growth), sales revenue to operating expenses ratio (Mpower). These variables have been found to have some influence on corporate environmental performance (Xue et al., 2022; Li J. et al., 2023). In terms of corporate governance, we adopt the recommendations of Choi et al. (2020) and Feng et al. (2022). Specifically, we include integration of two functions (Dual), number of directors (Board), proportion of independent directors (Indep) and proportion of the largest shareholder (Top1). Additionally, we consider whether the enterprise is state-owned or not (Soe). Please refer to Table 2 for a detailed explanation and measurement of each variable.

To examine the impact of digital transformation on the greening of firms, we employ ordinary least squares (OLS) estimation. The regression model controls for various firm and industry characteristics that may affect greenization. Additionally, year and industry dummy variables are included to account for potential time and industry-specific effects. The results of the basic regression analysis are presented in Table 5. In columns 1 and 2, the coefficient of the digital transformation dimension on IOCG is found to be negative and statistically significant at the 1% level. These findings support hypothesis H1, suggesting that digital transformation indeed utilizes green resources, thereby weakening green innovation and creating a “crowding out effect” on green investment. This result aligns with previous studies, such as Truby (2018) who argued that the energy-saving effect of digital transformation is limited. In fact, the application of digital communication technology can even lead to a rebound effect, resulting in increased energy consumption.

In columns 3 and 4, the coefficient of the digital transformation dimension on OOCG is positive and statistically significant at the 1% level, supporting hypothesis H2. These results suggest that while digital transformation may crowd out IOCG, it can also enrich green resources, enhance green technology innovation, and ultimately lead to green innovation outcomes for enterprises. These findings are consistent with previous studies, such as He et al. (2023), Feng et al. (2022), and Zhang G. et al. (2023).

The coefficient of BODT on IOCG is −0.2910, while the coefficient of DODT is −884.1720. Similarly, the coefficient of BODT on OOCG is 0.2193, while the coefficient of DODT is 660.1011. These results indicate that both BODT and DODT have a consistent direction of influence on enterprise greening. However, the absolute value of the DODT coefficient is significantly larger than that of the BODT coefficient, suggesting that DODT is a more powerful driver of enterprise greening. This finding supports hypothesis H3. Similar results have been found in previous studies. For instance, Khayer et al. (2023) analyzed the relationship between cloud computing deployment and enterprise performance and found that the coefficient of cloud computing assimilation depth is greater than the breadth.

Table 6 presents the moderating effect of slack resources on the relationship between digital transformation and IOCG. In columns 1 to 4, the coefficients of BODT*ASR, BODT*USR, DODT*ASR and DODT*USR are all positive and statistically significant at the 1% level. This indicates that the presence of more slack resources weakens the impact of digital transformation on IOCG, thereby confirming hypothesis H4. These findings suggest that slack resources can help mitigate internal stakeholder conflicts, influence the risk attitudes of enterprises, act as a “buffer” between the environment and enterprises, and weaken the resource preemption effect of digital transformation.

Table 7 presents the moderating effect of slack resources on the relationship between digital transformation and OOCG. In columns 1 to 4, the coefficients for BODT * ASR, BODT * USR, DODT * ASR, and DODT * USR are negative and statistically significant at the 1% level. This indicates that as slack resources increase, the impact of digital transformation on OOCG is reduced, thus confirming hypothesis H5. The results suggest that an increase in slack resources can enhance managerial discretion, induce higher governance costs, and ultimately weaken the positive effect of digital transformation on OOCG. This finding contrasts with some existing literature that emphasizes the strengthening effect of slack resources on the relationship between digitization and greenization (e.g., Ying et al., 2023).

From a comparative perspective, in Table 6, the coefficient of BODT*ASR is significantly larger than that of BODT*USR, and the coefficient of DODT*ASR is also significantly larger than that of DODT*USR. Similarly, in Table 7, the coefficient of the interaction term between digital transformation and ASR is much larger than that between digital transformation and USR. These results indicate that ASR has a stronger effect on the relationship between digitization and greenization compared to USR, thus confirming hypothesis H6.

In order to mitigate the bias resulting from the measurement of single variables, we conducted a substitution of independent variables and re-evaluated the analysis. Specifically, we revised the measurement approach for BODT, shifting from a quantitative assessment based on the quantity of digital transformation keywords to a categorical evaluation. For instance, if we extract 10 digital transformation keywords from the relevant text of a company’s annual report, with each keyword falling into one of two major categories, a value of 2 will be assigned. Additionally, recognizing the substantial variation in the extent of digital transformation across industries, we modified the measurement of DODT. Instead of calculating the ratio of a company’s digital transformation keywords to the total number of digital transformation keywords across all companies in a given year, we computed the ratio of a company’s digital transformation keywords to the total number of digital transformation keywords within its respective industry. The empirical findings subsequent to the substitution of variables are presented in Table 8. In panel A, the coefficients of digital transformation and IOCG exhibit a significant negative relationship, while the coefficients of BODT and OOCG demonstrate a significant positive association. The coefficients of DODT, although positive, do not reach statistical significance. In panel B, both ASR and USR exert a significant negative regulatory influence on the relationship between BODT/DODT and IOCG. In panel C, slack resources exhibit a significant negative moderating effect on the relationship between BODT/DODT and OOCG. Specifically, USR demonstrates a significant negative moderating effect on the relationship between BODT and OOCG, but does not exhibit a significant negative moderating effect on the relationship between DODT and OOCG. Overall, the empirical results subsequent to the substitution of explanatory variables are basically consistent with the above hypothesis testing results.

To mitigate the abnormal impact of the COVID-19 outbreak that commenced in 2020, we excluded the observation data from that year. The empirical findings based on the revised sample are presented in Table 9. In panel A, the coefficient representing the relationship between digital transformation and IOCG is significantly negative at the 1% level, while the coefficient indicating their influence on OOCG is positively significant. In Panel B, slack resources exert a significant negative regulatory effect on the relationship between digital transformation and IOCG. In panel C, both ASR and USR exhibit a significant negative regulatory effect on the relationship between BODT and OOCG. Furthermore, ASR demonstrates a significant negative moderating effect on the relationship between DODT and OOCG, whereas USR does not exhibit a significant moderating effect on the relationship between DODT and OOCG. Overall, the empirical results following the adjustment of the sample align closely with the aforementioned hypothesis testing outcomes. Additionally, to further validate the robustness of the findings, we conducted an additional analysis by excluding the data from 2015, considering the financial crisis during that period, and on this basis, we also removed the data of 2020 for repeated verification. The results from both analyses remain basically consistent with the previous findings.

To address the issue of endogeneity, we followed the approach outlined in refer to Breuer et al. (2018) and utilized agglomeration-level data, specifically the mean values of BODT and DODT for the same year in the same province, as instrumental variables (IV) in a 2SLS regression. The agglomeration layer data exhibits a significant correlation with the explanatory variables, but not a direct correlation with the explained variables, satisfying the instrumental variable requirements. The empirical findings subsequent to the use of instrumental variables are presented in Table 10. In panel A, both the coefficients of digital transformation and IOCG are significantly negative, while the coefficients of digital transformation and OOCG are significantly positive. In panel B, both ASR and USR demonstrate a significant negative regulatory effect on the relationship between BODT/DODT and IOCG. In panel C, both ASR and USR exhibit a significant negative moderating effect on the relationship between BODT/DODT and OOCG. Additionally, the Cragg-Donald Wald F statistic is considerably larger than 10, rejecting the null hypothesis of weak instrumental variables. Therefore, even after addressing the endogeneity issue using instrumental variables, the results remain robust.