Amid the growing global emphasis on sustainable development, green innovation has attracted significant scholarly and practical attention in recent years. However, existing studies have largely emphasised external institutional and policy drivers, leaving a limited understanding of how internal strategic orientations endogenously foster green innovation. Grounded in dynamic capabilities theory, this study examines whether market orientation serves as an internal strategic driver of green innovation. Analysing panel data on Chinese A-share listed firms from 2012 to 2024, we apply a double machine learning framework to estimate the effects of the two core components of market orientation (customer orientation and competitor orientation) on green innovation. We find that both components significantly promote green innovation, with these findings remaining robust across several rigorous tests. Furthermore, government green subsidies positively moderate the relationships of market orientation dimensions with green innovation. Additional heterogeneity analysis indicates that these relationships are particularly pronounced in non-state-owned and high-tech firms. Overall, this study extends the green innovation literature by foregrounding internal strategic orientations and advances methodological practice in this field by applying double machine learning. It offers valuable implications for future research on green innovation and for policymakers and managers committed to promoting sustainable development.
Current economic development and technological progress have caused significant climate change. Environmental disruptions stemming from human activities, coupled with escalating natural resource scarcity, have engendered growing global concerns (Chen, 2008). As such, society has gradually recognised the imperative for sustainable development, which arises from the critical need to reconcile the inherent tension between economic expansion and environmental degradation. In recent years, as China’s economy transitions from rapid to high-quality development, the Chinese government has implemented diverse initiatives and actions aligned with sustainable development as a strategic response aligned with the Paris Agreement (Yu et al., 2021). Firms are also not immune to ecological responsibilities. They face increasingly mounting pressures from governmental regulations, societal expectations, and customer demands (Yang et al., 2024). Notably, sustainable development and high-quality economic growth can be realised by implementing green innovation (GI) (Jiang et al., 2025; Chang, 2011).
GI refers to developing and applying novel technologies related to environmentally sustainable products or processes (Chen, 2008) that concurrently deliver economic value and environmental benefits. As a conceptual extension of conventional innovation paradigms, GI operationalises a dual-axis value proposition by maintaining socioeconomic advancement and systematically internalising sustainable development imperatives. Consequently, GI represents a strategic business opportunity that enhances operational efficiency and confers superior competitive advantage in the marketplace (Du et al., 2018; Albort-Morant et al., 2016).
However, GI faces significant constraints owing to dual externalities (Zhao et al., 2022): Societal returns and competitive spillovers substantially inhibit firms’ research and development (R&D) investments. Moreover, as a long-term process, GI involves high uncertainty, enormous capital investment and an extended payback period (Jiang et al., 2025; Yang et al., 2024). Thus, despite the support of policy, firms still exhibit insufficient GI incentive and low level of GI overall in China (Han et al., 2024; Irfan et al., 2022). Prior research has examined external drivers such as environmental regulations, institutional pressures, and green finance (Cui, 2022; Qi et al., 2021; Berrone et al., 2013; Yu et al., 2021). However, the literature has largely overlooked internal strategic factors. Furthermore, the sustainability revolution has created novel market demands, which may compel firms to intensify their strategic orientation on market dynamics to advance GI outcomes.
In management scholarship, market orientation (MO) can be analysed using two theoretical lenses: behavioural (Kohli & Jaworski, 1990) and cultural (Narver & Slater, 1990). Notwithstanding the conceptual differences, both perspectives converge on dynamic adaptation to the external market changes, particularly in terms of customers and competitors, and value creation through effective intelligence management (Slater & Narver, 1995). Such a synthesis informs our definition as follows: MO refers to a strategic orientation aimed at achieving superior customer value creation and competitive advantage through culturally driven customer–competitor behavioural integration or systematic intelligence management (Jaworski & Kohli, 1996; Morgan & Strong, 1998; Andreou et al., 2020). It is a critical enabler of firm innovation (Martínez-Azúa et al., 2025; Broekhuizen & Zhu, 2021). Additionally, high-MO firms typically maintain long-term strategic perspectives (Kumar et al., 1998). However, studies seldom explore the direct relationship between MO and GI. Thus, this study addresses the core question: How do MO’s key components—customer orientation (CUMO) and competitor orientation (COMO)—influence GI?
Our theoretical framework is grounded in dynamic capabilities theory (DCT). DCT posits that firms must continuously sense external shifts (e.g. market trends, customer preferences, and regulatory changes), seize opportunities through strategic resource allocation, and transform internal structures to maintain environmental alignment (Teece, 2007). MO functions as a critical manifestation of dynamic capabilities, as its inherent mechanisms enable organisational adaptation to environmental shifts (Kohli et al., 1993). Meanwhile, dynamic capabilities prove essential for GI development (Fan et al., 2024). In GI contexts, dynamic capabilities manifest as organisational capacities to identify environmental demands, integrate green technological resources, and reconfigure processes. From this perspective, MO transcends mere strategic choice, emerging as an organisational meta capability that systematically drives resource reconfiguration and capability iteration within GI ecosystems.
We employ panel data on Chinese A-share listed firms spanning 2012–2024 to investigate the CUMO/COMO-GI relationship. We first operationalise CUMO and COMO using textual analysis methodologies (Andreou et al., 2020). We then apply double machine learning (DML) techniques (Chernozhukov et al., 2018) to test the causal effects. Furthermore, we examine the moderating role of green subsidies (GS) to evaluate governmental support mechanisms while analysing heterogeneity effects across firm ownership structures and industry attributes.
This study offers three key contributions. First, although some scholars (Leal-Rodríguez et al., 2018) suggested that MO positively influences organisational performance through GI mediation, the direct MO–GI relationship remains underexplored. We advance this discourse by deconstructing MO into CUMO and COMO components and examining their distinct effects on GI amidst evolving green market demands. Second, grounded in DCT, we identify an internal strategic orientation that actively fosters GI, addressing China’s current lack of GI impetus within dynamic markets. Additionally, recognising GI’s market failure characteristics, we introduce GS as a moderator to comprehensively analyse how governmental interventions shape the MO–GI relationship. Third, responding to Helfat et al.'s (2023) call for innovative approaches to studying intangible resources, we employ machine learning and textual analysis, thereby pioneering novel methodological pathways for strategic orientation research.
The remainder of the article is organised as follows. Section 2 reviews the relevant literature and develops the hypotheses. Section 3 details the data and empirical methods. Section 4 reports the data analysis results. Section 5 discusses the findings and implications. Section 6 concludes with the limitations and future research directions.
Literature review and hypotheses developmentOur theoretical foundation rests upon DCT (Teece et al., 1997). Here, dynamic capabilities represent the highest organisational capability hierarchy. These capabilities function by governing and reconfiguring the lower-order operational and functional capabilities, which are regarded as the microfoundations. This orchestration enables firms to effectively adapt, integrate, and reconfigure internal and external resources in response to rapidly changing environments (Teece, 2007; Eisenhardt & Martin, 2000). Furthermore, these highest-order dynamic capabilities can be categorised into three distinct yet interrelated clusters (Teece, 2018). The sensing/shaping capability pertains to the identification, development, and evaluation of emerging technological opportunities, evolving customer needs, and strategic threats. The seizing capability involves the effective decision making, mobilisation, and reconfiguration of resources necessary to exploit identified opportunities and capture the associated value. Finally, the transforming capability refers to the continuous renewal and realignment of organisational structures, processes, and assets to maintain long-term adaptability and strategic relevance.
Meanwhile, DCT has recently gained substantial traction across diverse research domains, including digital transformation (Warner & Wäger, 2019), innovation management (Farzaneh et al., 2022), and business model (Seo et al., 2021). Dynamic capabilities are widely recognised as essential mechanisms for firms striving to achieve and sustain competitive advantage and superior performance (Wilden et al., 2013). This perspective highlights the necessity for firms to evolve dynamically, akin to living organisms, within turbulent environments. Consequently, our study is firmly grounded in DCT to theorise the causal relationships under investigation and develop the research hypotheses.
Green innovationConventional innovation has long been recognised as a catalyst for economic growth and societal progress. However, the growing sustainability crisis necessitates a paradigm shift towards GI. As a distinct subset of general innovation, GI integrates environmental sustainability goals into the traditional objectives of technological advancement and economic growth. Specifically, although conventional innovation focuses on enhancing economic efficiency and achieving technological breakthroughs, GI simultaneously addresses environmental concerns. Importantly, GI does not emerge in isolation but is technologically dependent on conventional innovation’s foundational achievements. For instance, clean energy technology development often requires progress in materials science, data driven algorithms, and other core scientific domains. Moreover, both conventional innovation and GI are subject to policy driven incentives, such as R&D subsidies, tax credits, and regulatory frameworks; these are designed to stimulate technological progress across multiple dimensions (Acemoglu et al., 2012). These policy instruments not only foster general innovation capacity but also serve as pivotal levers in advancing environmentally sustainable technologies.
Under advancing carbon neutrality frameworks, countries around the globe have positively been exploring and promoting environmental voluntary agreements as complementary mechanisms to traditional command-and-control regulations (Delmas & Montes‐Sancho, 2010). However, despite rising corporate participation, GI remains hindered by weak intrinsic motivation. This is largely due to its dual externalities of environmental benefits and knowledge spillovers, which generate significant market failures (Zhao et al., 2022; Yuan & Cao, 2022). These failures erode private sector incentives, further intensified by high R&D costs, delayed returns horizons, and systemic constraints such as technological lock-ins (Jiang et al., 2025; Zhang et al., 2025) and ecosystem fragmentation (Adner, 2016). This can question the adequacy of traditional Schumpeterian innovation frameworks in addressing sustainability driven technological transitions.
Moreover, GI drivers are multifaceted. However, the literature predominantly emphasise external factors, such as environmental regulation (Cui, 2022), institutional pressures (Berrone et al., 2013; Qi et al., 2021), green finance (Irfan et al., 2022; Yu et al., 2021), institutional investors (Zhao et al., 2023), and government subsidies (Xiang et al., 2022) as the primary antecedents. Internal drivers include green knowledge management (Abbas & Khan, 2022), CEO experience (Quan et al., 2023), stakeholder pressure (Singh et al., 2022), and environmental, social, and governance ratings (Wang et al., 2023). Table 1 shows an overview of the literature on drivers of GI. Notably, the role and mechanisms of firms’ strategic orientations in shaping GI remain comparatively underexplored. Given that firms are the main drivers of GI, market-oriented mechanisms are essential for sustaining a continuous stream of innovation incentives (Shang et al., 2025). Neglecting these endogenous factors risk obscuring a more comprehensive understanding of how firms proactively engage in sustainable innovation. In particular, dynamic capabilities are the critical enablers of GI that demand the concurrent mastery of conventional innovation methodologies and emergent eco-technologies (Fan et al., 2024). Firms exhibiting enhanced dynamic capabilities manifest a pronounced innovative orientation (Teece, 2007). However, the econometric literature predominantly emphasises external drivers and recent machine learning work ranks the predictors of GI. Meanwhile, our study foregrounds MO and decomposes it into CUMO and COMO. Additionally, the reviewed studies predominantly rely on ordinary least squares/Poisson/negative binomial or prediction-oriented machine learning. Conversely, we implement DML to estimate MO’s causal effect on GI while flexibly controlling for high-dimensional confounders.
Comparison with prior empirical studies on GI drivers.
| Source | Methodology | Representative drivers | Representative finding (topic & citation) |
|---|---|---|---|
| Existing studies | Econometric (difference-in-difference) | Environmental regulation; Green finance | Environmental regulation positively affects GI (Cui, 2022); Green finance is positively associated with GI (Irfan et al., 2022). |
| Econometric (Negative binomial) | Institutional pressure | Institutional environmental pressure is positively related to GI (Berrone et al., 2013). | |
| Econometric (OLS) | Institutional investor; ESG ratings | The GI effect of institutional investors is contingent on financial and social benefits (Zhao et al., 2023); Higher ESG ratings positively affect GI (Wang et al., 2023). | |
| Econometric (Poisson regression) | Government subsidies | Government subsidies enhance GI by alleviating financing constraints (Xiang et al., 2022). | |
| Structure equation modelling (PLS-SEM) | Stakeholder pressure | Stakeholder pressure indirectly affects GI via green dynamic capability (Singh et al., 2022). | |
| Structure equation modelling (SEM) | Green knowledge management | Green knowledge management is positively related to GI (Abbas & Khan, 2022). | |
| Machine Learning (ML) | CEO, firm, and industry characteristics | ML identifies and ranks potential GI drivers, and ESG ratings and internationalization rank among the strongest predictors (Liu et al., 2024). | |
| This paper | Double Machine Learning (DML) | MO | Both CUMO and COMO have positive and significant effects on GI. |
Note: The drivers and studies listed are illustrative rather than exhaustive.
Our research approaches MO from a cultural perspective, viewing it as a unifying organisational mindset that guides individual market-oriented thoughts and actions, thereby notably facilitating firm-level innovation (Büschgens et al., 2013; Martínez-Azúa et al., 2025). With its dual nature, MO can be understood both as corporate culture, reflecting internal values and codes of conduct, and corporate strategy, shaping external action planning and resource allocation. Market-oriented firms are committed to achieving superior customer value while treating competitors as strategic reference points. Notably, market-oriented firms concentrate on fulfilling expressed or aware customer needs and reacting to existing competitive dynamics. This illustrates their adaptation to the prevailing industry norms dominated by others. They also target latent or unarticulated needs and actively shape the competitive landscape. Here, the aim is to generate customer value through novel market insights and anticipatory strategic actions (Slater & Narver, 1998).
MO comprises two core elements: CUMO and COMO (Kotler & Armstrong, 2018; Newman et al., 2016; Haon et al., 2023). CUMO focuses on fulfilling customer needs. Meanwhile, COMO emphasises competing for demands (Lumpkin & Dess, 1996). CUMO entails placing customer interests at the forefront of strategic decision making, fostering close relationships with clients and proactively developing customer value propositions. This orientation supports long-term planning, resource allocation, and the consistent delivery of distinctive customer value, ultimately enhancing customer satisfaction and loyalty (Newman et al., 2016). CUMO involves systematically gathering and interpreting competitor intelligence. Firms monitor rivals’ actions, strategies, and market performance within the industry. Further, they dynamically adjust their own strategic positioning based on comparative assessments of short-term strengths and weaknesses to secure long-term competitive advantage (Haon et al., 2023; Newman et al., 2016). Overall, market-oriented firms are expected to coordinate CUMO and COMO as part of their strategic posture.
Additionally, MO’s conceptual demarcation requires rigorous theoretical differentiation. First, it fundamentally differs from marketing orientation by emphasising the strategic integration of organisational core competences for customer value creation. Meanwhile, marketing is confined to the technical operations of marketing functions (Slater & Narver, 1998). Second, CUMO demands decoding customers’ evolving explicit and implicit needs, including sustainability aspirations, focusing on building competitive advantage through long-term value co-creation. Conversely, customer-led approaches merely address superficial demands driven by short-term profit myopia (Slater & Narver, 1998; Connor, 1999), equivalent with partial CUMO. Such a customer-led strategy may render leading firms vulnerable to disruption, as an excessive focus on high-margin clients can cause the neglect of emerging technologies and latent market demands (Hult & Ketchen, 2001). Taking the smartphone market as an example, Nokia once held a dominant position in the traditional mobile phone sector but ignored the emerging demand shift and technological changes. This ultimately led to its failure in subsequent competition. The aforementioned conceptual discriminations are critical for establishing causal validity in our research design.
However, studies mostly focus on the impact of MO on business performance (Sampaio et al., 2019; Cano et al., 2004). Meanwhile, scholars have started examining how MO plays a role in non-economic goals, such as corporate social responsibility (Yuan & Cao, 2022) and environment performance (Chen et al., 2015). Viewed through a dynamic capabilities lens, MO is more accurately understood as a capability embedded in a socially complex and firm-level system of processes, routines, and learning (Menguc & Auh, 2006), rather than a purely information processing mechanism.
Market orientation and green innovationGI inherently constitutes a dynamic complex system of internal–external synergy, relying on continuous interaction between organisations and their environment. Goal 12 of the United Nations Sustainable Development Goals framework1 has promoted consumer environmental awareness to emerge and transform into actionable practices, driving the sustained expansion of green demand (Ng et al., 2024). Similarly, in China, public consciousness of environmental issues and sustainable purchasing intentions have markedly increased. In this context, MO enables firms to discern and transform green market signals into concrete innovation practices. However, CUMO and COMO perform differentiated and specialised functions in firms’ strategic decision-making processes, as they generate and utilise distinct types of market intelligence to serve different managerial purposes (Eibe Sørensen, 2009). Thus, drawing on DCT (Teece, 2007, 2020), we disentangle the effects on GI using the two MO components of CUMO and COMO, which operate through firms’ enhanced sensing, seizing, and transforming capabilities. Each component possesses a distinct focus and entails a unique set of activities.
Customer orientation and green innovationFirms with high CUMO typically pursue broad market knowledge (Qu & Mardani, 2023). This enhances their abilities to proficiently and effectively discover customer sustainability preferences, and guide customers beyond established cognitive frames (Lukas & Ferrell, 2000). Together, this can enable firms to systematically identify implicit green needs. Moreover, CUMO instils firm-wide responsibility and enthusiasm, particularly encouraging frontline employees to generate and collect novel GI ideas (Broekhuizen & Zhu, 2021). Firms’ target customers are not limited to existing consumers but also encompass the green market, an emerging segment representing potential new customers. Thus, customer-oriented firms are better able to identify and evaluate opportunities and threats within this developing market. Consequently, CUMO strengthens the sensing capabilities of firms that can better anticipate the shifts towards GI consciousness.
Stronger-CUMO firms are more attuned to customers’ environmental concerns, and thus face stronger market pressure to pursue green initiatives. This pressure drives them to make swift and resolute green decisions, thereby reallocating resources towards eco-friendly processes and products. By integrating customer intelligence into green planning and implementation, such firms markedly enhance the effectiveness of their GI efforts (Yuan & Cao, 2022). Their superior absorptive capacity enables them to transform external market knowledge into tangible innovation outcomes (Qu & Mardani, 2023). Thus, CUMO equips firms to make informed decisions and enhance strategic agility in seizing GI opportunities.
Firms with strong CUMO continuously realign their assets, processes, and structures towards green initiatives to deliver superior value to customers as environmental awareness rises. The advantages gained from sensing and seizing capabilities further enhance their ability to reconfigure organisational resources (Wilden & Gudergan, 2015; Winter, 2003) and adapt to evolving environmental demands. When CUMO is deeply embedded as a corporate culture, it establishes feedback mechanisms that foster proactive, rather than reactive, strategic adjustments, thereby maintaining market relevance. Moreover, high CUMO levels facilitate timely information flows and coordinated decision-making, which together support effective resource reallocation towards green R&D and technological innovation (Zhou & Li, 2010). Therefore, CUMO enables firms to sustain GI through enhanced transforming capabilities. Based on the discussion above, we propose:
Hypothesis 1 CUMO positively influences GI.
Firms with strong COMO are better positioned to collect and interpret intelligence regarding competitors’ green strategies and technological trajectories. By systematically tracking rivals’ actions, they can identify emerging green opportunities and evaluate potential strategic threats, such as those triggered by carbon tariff policies. Engineers and R&D professionals in such firms, guided by COMO, recognise the strategic value of competitor intelligence and leverage their expertise to analyse rivals’ technological trajectories, assess alternative innovation routes, and anticipate potential breakthroughs. Consequently, COMO strengthens firms’ sensing capabilities that can better anticipate the technological shifts and competitive landscape in GI.
Meanwhile, firms with COMO monitors rivals’ actions to identify emerging directions for GI and translate market intelligence into concrete tactics (Voss & Voss, 2000). By learning from their own initiatives and competitors’ successes or failures, these firms enhance adaptive capacity and reduce uncertainty, which is amplified by environmental complexity (Martínez-Azúa et al., 2025). Guided by competitive pressure and the urgency to maintain speed advantages, firms with strong COMO can swiftly imitate or upgrade competitors’ green solutions while strategically leveraging GI as a differentiation tool (Chang, 2011; Dong et al., 2024). Thus, COMO drives firms to respond decisively to competitive signals, efficiently reallocate resources, and capture value from emerging GI opportunities.
Once green opportunities have been sensed and seized, firms with a strong COMO systematically integrate external insights to reconfigure operations, renew processes, and realign assets to sustain competitive advantage. These firms often establish dedicated ‘GI response units’ and adjust organisational structures in real time to address market and competitor dynamics. Additionally, firms with strong COMO engage in strategic renewal to align with their business ecological niche, continuously adapting strategies in response to competitors’ environmental actions and potential retaliatory moves. Transforming capabilities are inherently strategic rather than solely reliant on routine processes. Here, COMO ensures long-term strategic relevance. Thus, COMO enables firms to maintain superior transforming capabilities that facilitate successful GI. Accordingly, we hypothesise the following:
Hypothesis 2 COMO positively influences GI.
Governments and customers increasingly demonstrate clear expectations for firms to conduct GI in greater depth. Firms bear a responsibility to implement environmental protection measures to reduce or alleviate the damage caused by human activities to the natural environment (Leal-Rodríguez et al., 2018). Under increasing societal pressures and evolving regulatory policies, firms must constantly balance short-term operational costs against long-term sustainability objectives. GI practices are increasingly seen as a crucial element for building competitive advantage and ensuring sustainable development (Du et al., 2018). However, firms face significant obstacles primarily owing to the market failures inherent in GI. GI is characterised by enormous financial investments coupled with extended payback periods (Jiang et al., 2025; Yang et al., 2024). Meanwhile, customers often remain unwilling to pay substantial premium prices for green products in many markets (Dangelico & Pujari, 2010). Furthermore, the ambiguous reward conditions associated with GI, combined with its inherent technological and market risks, understandably reduce firms’ confidence, particularly those with limited asset scales and financial resilience.(Fig. 1)
Government intervention plays a key role in solving these problems of high uncertainty and resource constraints (Han et al., 2024; Dong et al., 2024). GS represents one of the most direct and effective measures of such government intervention (Xie et al., 2019). First, GS directly alleviates the financial pressure on firms by providing essential capital or resource support, thus reducing the binding constraints on GI activities. This financial relief accelerates market-oriented firms’ ability to transform valuable market intelligence into concrete GI outputs. Moreover, GS delivers a powerful positive signal regarding green initiatives’ strategic importance and societal value. This signal can strongly support firms, effectively amplifying MO’s impact. GS not only further stimulates green market demands but also guides the strategic direction of firm innovation efforts towards societal priorities. Finally, GS’ provision signifies that the government offers to share a portion of the inherent risks with firms engaging in GI, significantly motivating firms to proactively respond to the identified green market signals and opportunities. Accordingly, we hypothesise the following:
Hypothesis 3 GS positively moderates the relationship between CUMO and GI.
Hypothesis 4 GS positively moderates the relationship between COMO and GI.
Our dataset comprises Chinese A-share listed firms from 2012 to 2024, reflecting the most recent data available at the time of analysis, and thus enhancing our empirical evidence’s relevance and robustness. To arrive at our final sample, we first exclude firms designated as under special treatment (ST and *ST) and all entities in the financial sector. Next, continuous variables are winsorised at the 1 % and 99 % levels to mitigate the influence of extreme outliers. We then drop any observations with missing values in key variables. All data are sourced from the China Stock Market & Accounting Research (CSMAR) database, and Chinese Research Data Services (CNRDS) platform. This yields a final dataset with 44,847 firm-year observations covering 5472 companies.
MeasurementsDependent variable: GIConsistent with extant studies (Javed et al., 2023; Quan et al., 2023), we measure GI using the annual count of green patent applications. The raw data are drawn from the Green Patent Research Database (GPRD) on the CNRDS platform. GPRD identifies green patents in full accordance with World Intellectual Property Organization standards and is constructed from a comprehensive consolidation and screening of records from the China National Intellectual Property Administration (CNIPA) and Google Patents. For each listed firm-year, the database documents the application and grant statuses of green patents. To address skewness, we apply a natural logarithmic transformation, using ln(1+greenpatentapplications). To address the potential limitation of using green patent applications as a proxy for GI, wherein some applications may never be granted, we also use granted green patents to verify robustness.
Independent variable: MOFollowing Helfat et al. (2023), who advocate textual analysis for measuring intangible resources and capabilities, we use firm-level text to quantify MO along two dimensions: CUMO and COMO. A recent study illustrates textual analysis’ use in empirical research on strategic orientation and firm innovation (Pang et al., 2025). Building on constructs validated in prior literature (Andreou et al., 2020; Noble et al., 2002; Zachary et al., 2011), we develop a keyword dictionary tailored to Chinese listed firms (see Appendix A). Using Python 3.11, we parse each firm’s annual report, focusing on the Management Discussion and Analysis (MD&A) section, to count the occurrences of CUMO and COMO related terms. Terms preceded by negative modifiers (such as ‘non-’, ‘not’, ‘no’, ‘will not’, and ‘dislike’) are detected and excluded. We aggregate these counts to the firm-year level and use the natural logarithm of one plus the number of counts. The resulting continuous measures serve as our indicators of each firm’s CUMO and COMO in the current year.
Moderator variable: GSOur moderator, GS, is operationalised as a binary indicator reflecting whether a firm received government support specifically earmarked for environmentally focused projects. We extract detailed subsidy records from the CSMAR database and filter these by project descriptions and subsidy categories to only isolate those disbursements labelled ‘green’, ‘environmental’, ‘environment’, ‘sustainable development’, and ‘clean’, ‘energy’. Firms with no identified green project subsidies in a given year receive a GS value of 0, whereas those with one or more qualifying subsidies receive a GS value of 1. This dichotomous construction allows us to test how external policy incentives interact with a firm’s MO to influence its GI output.
ControlsFinally, we include a comprehensive set of controls drawn from extant research (Andreou et al., 2020; Liu et al., 2024). Firm characteristics include firm age (Age), measured as ln(1+yearssinceestablishment); ownership type (SOE), a dummy for state-owned enterprises (SOE); firm size (Size), proxied by ln(totalassets); leverage (Lev), defined as total liabilities/total assets; liquidity (Liquid), the current ratio (current assets/current liabilities); and managerial structure (Dual), a dummy for CEO-chair duality. We further control for financial performance using return on assets (Roa) and Tobin’s Q (TobinQ), with the latter proxied by the market value to assets ratio. Recognising the crucial role of R&D investment, we include R&D intensity (RDI), measured as R&D expenditure/sales. Industry conditions are captured by a high-tech sector indicator (HighTech), while market concentration is measured by the Herfindahl–Hirschman Index (HHI). Finally, we incorporate year and industry fixed effects (dummies) to absorb unobserved time- and industry-specific heterogeneities, respectively.
DML construction processBuilding on extant studies (Bach et al., 2024; Chernozhukov et al., 2018; Yang et al., 2020), we implement DML with a partial linear regression (PLR) specification.2 Our causal question concerns MO’s effect on GI, separately measured by CUMO and COMO. We declare the outcome Yit as GI, the (continuous) treatment Dit as either CUMO or COMO, and high-dimensional covariate vector Xit as the set of controls (including firm characteristics, financials, and fixed-effects dummies for year and industry). Because the treatment is continuous, we adopt the PLR rather than the interactive regression model (IRM). The PLR decomposes the outcome into a linear treatment effect and nonparametric component in the controls:
where g0(·) and m0(·) are unknown nuisance functions, and ζitand vit are disturbances. The target parameter θ0 captures the causal effect of CUMO (or COMO) on GI.
We use the orthogonalised (Neyman-orthogonal) score with cross-fitting to mitigate overfitting bias and achieve valid inference with a high-dimensional Xit. Implementation is in Python 3.11 using the DoubleML package. For both nuisance functions g0(·) and m0(·), we employ Lasso learners from scikit-learn, and hyperparameters are tuned via five-fold cross validation. Cross fitting partitions the data into K=5 folds, estimating the nuisance functions on held-out folds and then plugging the predictions into the orthogonal score to obtain θ^. This procedure reduces regularisation bias and enhances robustness to flexible model selection.
The DML-PLR framework partials out the (potentially) complex influence of Xit on both Yit and Dit, thereby isolating θ0 as the marginal effect of MO on GI. We report standard errors that are asymptotically valid under cross fitting and present extensive sensitivity analyses (learner choice and omitted variable robustness) in subsequent sections.
Data analysis and resultsDescriptive statisticsTable 2 reports the summary statistics. GI’s mean is 0.438 (SD = 0.837), CUMO averages 5.226 (SD = 0.564), and COMO averages 5.085 (SD = 0.564). The descriptive statistics for the control variables follow similar patterns and are omitted here for brevity. As shown in Table 3, CUMO and COMO exhibit statistically significant positive pairwise correlations with GI, with coefficients r=0.061 and r=0.122, respectively. Multicollinearity appears limited, as the mean variance inflation factor (VIF) is 1.65 and maximum VIF is 2.95, with all values being well below conventional concern thresholds.
Descriptive analysis.
N = 44,847.
Correlation statistics.
| Age | HHI | HighTech | RDI | GS | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Age | 1 | ||||||||||
| HHI | −0.003 | 1 | |||||||||
| HighTech | −0.118*** | −0.377*** | 1 | ||||||||
| RDI | −0.133*** | −0.245*** | 0.402*** | 1 | |||||||
| GS | −0.001 | −0.001 | 0.003 | −0.115*** | 1 | ||||||
*p < 0.1, **p < 0.05, ***p < 0.01.
Table 4 reports the baseline DML-PLR estimates. Our primary specification employs Lasso (lasso) for the nuisance functions, with robustness checks using three alternative learners, including Random Forest (forest), LightGBM (lgbm), and XGBoost (xgboost). Across all four learners, CUMO (θ^=0.034 (lasso, p<0.01), 0.039 (forest, p<0.01), 0.041 (lgbm, p<0.01), and 0.044 (xgboost, p<0.01)) and COMO exhibit positive statistically significant associations with GI (θ^=0.054 (lasso, p<0.01), 0.052 (forest, p<0.01), 0.061 (lgbm, p<0.01), and 0.059 (xgboost, p<0.01)). Fig. 2 illustrates the comparison across learners. Thus, Hypotheses 1 and 2 are supported, indicating that CUMO and COMO significantly enhance GI.
Baseline results among DML models.
standard error in parentheses,.
* p < 0.1, ** p < 0.05, *** p < 0.01.
Comparation of baseline regression among ML methods.
Notes: This figure shows the coefficient estimates and 95 % confidence intervals from partially linear regressions of CUMO (blue circles) and COMO (orange squares) using four machine learning methods (Lasso, Random Forest, LightGBM, and XGBoost). Vertical error bars denote the 95 % CIs.
Next, we conduct several robustness checks. First, we strengthen the baseline specification by adding higher dimensional controls and reporting firm clustered results. Second, to address endogeneity concerns stemming from omitted variables and potential reverse causality, we perform sensitivity analyses and implement instrumental variable (IV) procedures within the DML framework.
Higher dimensional controls within cluster and GI replacementTo assess the MO–GI relationship’s stability, we extend the baseline in three ways. First, we introduce high dimensional interactions among the control variables and re-estimate the model using different learners. Table 5 shows that CUMO remains positively and significantly associated with GI across learners (θ^=0.026 (lasso, p<0.01), 0.040 (forest, p<0.01), 0.042 (lgbm, p<0.01), and 0.036 (xgboost, p<0.01)). The COMO–GI pathway likewise yields robust positive estimates (θ^=0.049, 0.052, 0.060, and 0.056, respectively (all p<0.01)). Second, following prior work (Chiang et al., 2022), we implement a firm-level clustered DML estimator to account for within-firm dependence. In the Lasso specification, CUMO (θ^=0.033, p<0.01) and COMO (θ^=0.053, p<0.01) continue to exhibit strong positive effects on GI when standard errors are clustered by firm. The results for the alternative learners under the clustered framework are reported in Panel A in Appendix B. Third, we replace the dependent variable with the natural logarithm of green patent grants (rather than applications) and re-estimate all learners. As reported in Panel B in Appendix B, the CUMO-GI and COMO-GI effects remain positive and statistically significant across specifications.
Higher dimensional controls regression results.
standard error in parentheses,.
* p < 0.1, ** p < 0.05, *** p < 0.01.
To gauge omitted variable bias risk, we conduct a sensitivity analysis following the framework of Cinelli and Hazlett (2020). This approach quantifies how strongly an unobserved confounder needs to relate to both the treatment (CUMO/COMO) and outcome (GI), in terms of partial R2, to attenuate our estimated effects. Here, we report the results for the CUMO–GI relationship, and analogous results for COMO-GI appear in Appendix C.
We benchmark against RDI, one of the most salient observed controls. Setting the confounder’s relative strength to ρ=1 (i.e. assuming the omitted variable is as influential as RDI with respect to both CUMO and GI), we obtain the robustness values of 0.007 (for statistical insignificance) and 0.014 (for a zero effect). Interpreted as a residual partial R2, these correspond to an unobserved confounder which would need to explain at least 1.01 % of the residual variance in both CUMO and GI to render the CUMO coefficient statistically insignificant, and at least 1.79 % to drive the effect to zero.
As shown in Fig. 3, the benchmark point (assuming an omitted confounder as strong as RDI) lies well above the zero-effect contour, indicating that even a confounder comparable in strength to RDI would not overturn the positive CUMO-GI estimate. We also benchmark using Size and Lev. Both yield lower robustness values than RDI. However, in each case, an omitted variable as influential as either Size or Lev still falls short of eliminating statistical significance. Overall, our main findings are unlikely to be driven by unobserved confounding.
CUMO sensitivity analysis results.
Notes: This sensitivity contour plot illustrates how the estimated effect of CUMO on GI would be biased by an unmeasured confounder whose association strengths with CUMO (x-axis) and GI (y-axis) vary. Red markers denote the empirically observed confounding strengths for specific controls or scenarios: Size, Lev, the unadjusted model, and a chosen Scenario based on RDI.
To guard against chance correlations, we perform a placebo test by randomly permuting the CUMO scores across firms, leaving all other variables intact and re-estimating the DML model 500 times. As shown in Fig. 4, the distribution of placebo coefficients is tightly clustered around zero and remains within the 95 % confidence bounds. Conversely, the true coefficient of CUMO (0.034) lies well beyond the 97.5th percentile of the permutation distribution. This placebo exercise shows that our observed CUMO-GI relationship is not an artifact of random variation but reflects a genuine effect. The COMO-GI placebo test presents similar results, illustrated in the Appendix D.
CUMO placebo test diagram.
Notes: By shuffling the CUMO values 500 times, we obtained the distribution of coefficients shown in the figure. The scatter plot represents the p-values of the coefficients, while the density curve (KDE) shows the distribution of the coefficients. These coefficients are mostly distributed near zero, and most of the p-values are higher than 0.05, suggesting the real effect is not coincidental happened.
Following the partially linear IV (PLIV) approach in Bach et al. (2024), we estimate a DoubleML-PLIV model in which a one-year lag of CUMO/COMO serves as an instrument for the current treatment. This specification targets concerns about reverse causality by leveraging within-firm temporal variation under standard IV assumptions (relevance and exogeneity conditional on controls). Again, CUMO remains positive and highly significant (θ^=0.041), as does lagged COMO (θ^=0.062), with both p<0.01. These IV estimates reinforce the baseline findings and further mitigate endogeneity concerns, supporting a causal interpretation of the MO–GI relationship.
Moderation effect analysisGroup average treatment effect analysis in DMLWe examine whether GS moderate the MO-GI relationship using group average treatment effects (GATE) within the DML framework. As reported in columns (1) and (2) in Table 6, the CUMO effect on GI is 0.015 (p=0.149) among non-subsidised firms and rises to 0.080 (p<0.01) among subsidised firms. Columns (3) and (4) show a similar pattern for COMO: 0.020 (p<0.05) without GS versus 0.130 (p<0.01) with GS. The between-group differences are 0.065 for CUMO-GI and 0.110 for COMO-GI. Thus, GS amplifies MO’s positive causal effect on GI. Fig. 5 visualises these moderation effects. The GATE estimates with 95 % confidence intervals demonstrate a clear upward shift in the treatment effect under GS. To alleviate reverse causality concerns in the moderation effect, we conduct a robustness check by treating one-year lagged GS (L.GS) as a pre-determined moderator. Columns (5)–(8) show that the GATE differences remain positive and statistically significant, reinforcing the conclusion that policy support strengthens the MO–GI linkage. To aid interpretation, Fig. 6 plots the linearised GATE curves. In both CUMO–GI and COMO–GI, the GS = 1 lines are markedly steeper, validating GS as a positive moderator.
Moderation effects results using GATE.
standard error in parentheses,.
* p < 0.1, ** p < 0.05, *** p < 0.01.
Combined Error-Bar Plot of GATE by GS.
Notes: Points plot the GATE of CUMO and COMO on GI, and the vertical bars are 95 % confidence intervals. Groups are defined by GS and lagged subsidies (L.GS). Effects are small and statistically weak without subsidies, but substantially larger and statistically significant when GS=1 and when L.GS=1, indicating that both GS and L.GS positively moderate the MO→GI effect.
Predicted GI from MO by GS (Linearized GATE).
Notes: Panels plot predicted GI under a linearized GATE approximation for CUMO (left) and COMO (right). For each GS group (blue = No GS, red = Yes GS), the line uses the group-specific GATE slope anchored at the group median of the x-axis and the group mean of GI, and the shaded area is a 95 % CI constructed from the standard error of the GATE slope. The markedly steeper red lines indicate that GS positively moderate the causal effect of MO on GI.
To further validate the moderating role of GS, we re-estimate the moderation effect in a conventional OLS framework. First, we run separate regressions by GS status for CUMO–GI and COMO–GI. Columns (1)–(4) in Table 7 show that the group specific coefficients closely mirror the DML-GATE estimates. Further, formal tests of coefficient differences across groups indicate statistically significant gaps. Thus, GS’ moderation effect is not an artifact of the DML procedure. Second, we augment the baseline OLS model with interaction terms between MO and GS. Columns (5) and (6) show that the interaction terms are positive and statistically significant in the CUMO–GI and COMO–GI specifications, corroborating Hypotheses 3 and 4. Together, the between group OLS estimates and interaction confirm that GS amplifies MO’s positive effect on GI, reinforcing the DML-based conclusions.
Robustness analysis for moderation effects using OLS.
standard error in parentheses.
* p < 0.1, ** p < 0.05, *** p < 0.01.
Following recent recommendations for visualising heterogeneous treatment effects (McShane et al., 2024), we examine CATE by firm ownership and industry attributes, proxied by SOE and HighTech. As shown in Fig. 7, CUMO and COMO exhibit stronger effects on GI in non-SOEs than in SOEs and high-tech firms relative to non-high-tech peers. This pattern indicates meaningful contextual heterogeneity in the MO–GI relationship. A plausible mechanism is that non-SOEs and high-tech firms possess greater strategic discretion and absorptive capacity, benefit more from knowledge spillovers among technological peers, and can more rapidly translate market-oriented insights into innovation output.
CATE by SOE and HighTech.
Notes:This bar plot presents the conditional average treatment effect (CATE) of CUMO/COMO on GI for four subgroups defined by state ownership (Soe = 0 non‐state, 1 state) and high‐technology status (HighTech = 0 low‐tech, 1 high‐tech). Each bar shows the subgroup's point estimate and the vertical whiskers denote the 95 % confidence interval. The CATE is largest for non-state, high-tech firms (Soe = 0, HighTech = 1), indicating notable heterogeneity in CUMO/COMO's effect across these organizational contexts.
Within China’s regulatory context, environmental competition has often manifested as “races to the bottom”, whereby firms undercut one another on environmental standards (Feng et al., 2021). Here, market pull forces remain critical GI drivers. Using panel data on Chinese A-share listed firms from 2012 to 2024 and a DML-PLR framework, we examine how MO, operationalised as CUMO and COMO, shapes GI. We find that CUMO and COMO are positively and significantly associated with GI, highlighting internal strategic orientations’ role in fostering sustainable innovation. Moreover, GS strengthens the MO–GI linkage. Thus, GS’ existence enables market-oriented firms to more effectively initiate and implement green projects. Finally, the CUMO–GI and COMO–GI effects are stronger in non-SOEs and high-tech industries.
Theoretical implicationsFirst, this study enriches the literature on GI antecedents by foregrounding internal strategic drivers in an emerging economy context. Although extant research has emphasised external stimuli, such as environmental regulations (Cui, 2022), institutional pressures (Berrone et al., 2013; Qi et al., 2021), and green finance (Irfan et al., 2022; Yu et al., 2021), we show that enduring market‐oriented capabilities are equally vital for sustained green development. GI is inherently a long‐term pursuit. We show that CUMO and COMO significantly directly affect firms’ GI. By establishing these links, we build on Leal-Rodríguez et al.’s (2018), who identified an indirect GI pathway, and extend the theoretical narrative by connecting corporate market culture to proactive capabilities for GI.
Second, our research extends the MO literature by elucidating its impact on GI. Studies widely document the positive link between MO and conventional innovation (Broekhuizen & Zhu, 2021; Büschgens et al., 2013; Martínez-Azúa et al., 2025); however, few investigate MO’s direct and specific effects on GI. Given increasing global concerns over climate change, understanding how a firm’s market‐oriented strategy fosters green development is imperative. Consistent with Grinstein (2008)), we show that CUMO and COMO significantly drive GI. Moreover, by employing panel data analysis, we capture the dynamic evolution of these effects over time, offering methodological rigor and novel insights into the temporal mechanisms through which MO influences sustainable innovation. Additionally, the textual analysis and DML method jointly show a novel path for investigating such an intangible firm resource as MO.
Third, our findings illuminate a critical context for the MO–GI relationship by highlighting GS’ moderating role. CUMO and COMO bolster firms’ abilities to sense and seize opportunities and reconfigure resources for GI. However, the substantial capital and technological investments required for GI can inhibit especially smaller or resource‐constrained firms from acting on market signals. By showing that GS’ existence significantly amplify the positive effects of CUMO and COMO on GI, we extend the contingency perspective within strategic orientation research. External government support mechanisms can alleviate resource‐based constraints and enhance the efficacy of internal market‐oriented capabilities in driving sustainable innovation. This theoretical integration of public policy and firm‐level strategy advances our understanding of how internal and external forces jointly shape GI outcomes.
Fourth, the CATE results indicate that non-SOEs and high-tech firms exhibit the largest and most statistically significant MO-GI effects for CUMO and COMO. Research suggests that the innovation returns to strategic orientations vary by ownership and industry context (Pang et al., 2025). The GI literature often reveals that non-SOEs are more effective innovators (Wang et al., 2023). This may be because non-SOEs operate with stronger market incentives and fewer administrative constraints, enabling a more effective “outside-in” process of sensing customer and competitor cues, and translating them into GI initiatives. Meanwhile, SOEs typically face stricter government oversight and planning requirements, which can dampen the speed and flexibility needed to convert market signals into GI. Additionally, high-tech firms usually possess deeper knowledge stocks and higher R&D intensity; these capabilities are associated with superior innovation outcomes (Abbas & Khan, 2022; Abbas & Sağsan, 2019; Liu et al., 2024). Once green opportunities are sensed, these firms are better equipped to mobilise resources and implement GI. Together, the combination of lighter governance constraints (in non-SOEs) and richer knowledge capital (in high-tech firms) helps explain why these groups realise stronger MO-driven gains in GI.
Practical implicationsFirst, from a managerial standpoint, firms must recognise the urgency of climate challenges and embed GI into their core strategy. By adopting a MO that combines customer and competitor insights, firms can develop forward‐looking capabilities that align long‐term sustainability goals with operational realities. This strategic posture sharpens firms’ dynamic capabilities, enhancing their ability to sense emerging green opportunities, seize them through targeted investment, and reconfigure resources to support complex innovation processes. Managers should calibrate the balance between CUMO (to understand evolving environmental preferences) and COMO (to monitor industry best practices). This can ensure demand-driven and competitively informed resource allocations for green initiatives, thereby translating strategic intent into consistent sustainable practices.
Second, given GI’s unpredictable risks and substantial capital requirements, firms should proactively leverage government support policies, particularly GS, to mitigate financial constraints and de‐risk their sustainability initiatives. Firms with a strong MO are better positioned to strategically deploy these funds, thereby unlocking pilot projects, accelerating the scale‐up of eco‐friendly processes, and attracting complementary private financing. Furthermore, the endorsement implicit in GS reinforces corporate legitimacy, elevates customer environmental awareness, and stimulates green purchasing behaviours, thus creating positive feedback for firms’ market‐oriented strategies. To maximise these benefits, managers should directly incorporate subsidy planning into their innovation roadmaps, aligning application timelines with R&D milestones, ensuring rigorous environmental performance reporting, and seeking collaborative opportunities with industry peers. However, GS alone may be insufficient to drive effective GI. These firms often lack the internal mechanisms, such as organisational learning capacity, technological R&D capability, and resource integration competence, required to absorb and transform external support into meaningful outcomes. Hence, strengthening GS’ marginal effectiveness requires enhancing firms’ internal capabilities and aligning strategic orientations. Overall, green technology commercialisation can be effectively accelerated through the synergistic effect of firm-level strategies and policy interventions.
Third, GS amplifies the positive effect of MO on GI. Moreover, only market-oriented firms receiving such subsidies can significantly enhance their GI outcomes. Thus, policymakers should prioritise market-oriented firms and use fiscal support to alleviate their resource and innovation constraints, thereby ensuring precise and effective subsidy allocation. In practice, policymakers should move beyond conventional subsidy allocation criteria and instead adopt a MO-based screening mechanism to enhance policy precision, particularly for non-SOEs and high-tech firms. Structured evaluation procedures can be embedded within the application or renewal processes for GS or innovation funds to quantitatively assess firms’ MO level. Such assessments may include indicators reflecting the robustness of customer insight mechanisms and effectiveness of competitive intelligence systems.
Nevertheless, several potential challenges or unintended effects may arise during implementation. Ensuring that MO’s assessment remains objective and impartial is inherently difficult. Meanwhile, overly complex evaluation standards may increase administrative costs and create opportunities for rent-seeking. Moreover, concentrating fiscal resources on firms already possessing strong MO can widen the innovation gap. Excessive dependence on financial incentives may undermine firms’ intrinsic motivation for self-driven innovation. Policymakers should strike an appropriate balance between incentives and constraints, ensuring that GS serve as catalysts rather than substitutes for firms’ internal strategic efforts, ultimately fostering MO as GI’s endogenous driver.
ConclusionThis study examines the causal relationship between MO and GI, finding that CUMO and COMO are positively associated with GI. Crucially, we offer an “outside-in” perspective, showing how market forces pull firms towards innovation and the development of green capabilities. Moreover, this effect strengthens when firms receive government support for green development. However, this study has limitations. First, our sample comprises only Chinese A‐share listed firms, which may constrain our findings’ external validity and generalisability. Scholars should extend this analysis to more diverse industries and geographic regions, particularly contrasting emerging and developed economies, and include small and medium‐sized enterprises to test our framework’s robustness and universality.
Second, we do not consider how varying market environments, such as market turbulence, competitive intensity, and technological turbulence, which may condition MO’s effectiveness on GI. Future research should incorporate these contextual factors to empirically assess how different market atmospheres shape the MO–GI relationship. By integrating boundary conditions such as market dynamism and competitive pressure, scholars can deepen the explanatory framework, and enhance the persuasiveness and generalisability of conclusions across diverse settings.
Third, although we demonstrate how GS moderate the MO–GI relationship, we treat MO in isolation and do not consider that firms typically pursue multiple strategic orientations concurrently (e.g. technology, entrepreneurial or learning orientations). These orientations may synergistically interact, enhancing a firm’s ability to identify and exploit green opportunities beyond any single orientation’s effect. Future research should examine whether complementarities among strategic orientations further amplify GI outcomes. Testing these interaction effects can yield a more nuanced understanding of how firms can configure strategic portfolios to maximise sustainable outcomes.
CRediT authorship contribution statementYanqing Guo: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Chang Zhao: Writing – review & editing, Writing – original draft, Visualization, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Ming Pang: Writing – review & editing, Writing – original draft, Visualization, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Di He: Writing – review & editing, Visualization, Software, Investigation, Funding acquisition, Data curation, Conceptualization, Supervision.
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
This work was supported by The National Social Science Fund of China [grant number: 24BGL033].
Notes: The selected keywords were based on previous studies (Andreou et al., 2020; Noble et al., 2002; Zachary et al., 2011). The keywords are presented in Chinese, with English translations provided in parentheses.
In the Appendix B, Panel A shows the cluster DML results and Panel B shows the GI replacement results.
standard error in parentheses,.
*p < 0.1, ** p < 0.05, *** p < 0.01.
This Appendix illustrates how COMO’s estimated effect on GI can be biased by an unmeasured confounder whose association strengths with the two variables. Contour lines are labelled by the resulting bias in the COMO coefficient. This means that even an omitted variable is as influential as Size, RDI, or Lev, the coefficient is not leading to zero contour line.
articles