Dual-Credit Policy of New Energy Automobiles in China: Corporate Innovation Capability

Abstract

This study investigates the impact of China’s Dual-Credit Policy on innovation capability in the new energy vehicle (NEV) industry using a difference-in-differences approach with dynamic panel data from 2013 to 2022. We employ multiple innovation measures, including patent counts, valid invention patents, and patent grant rates. Our findings reveal that the policy has a positive and significant effect on NEV innovation, partially mediated by R&D investment. Using GMM estimation to address endogeneity, we find the policy effect varies across subgroups based on location, ESG rating, and ownership type. Dynamic effect analysis shows the policy’s impact intensifies over time. Threshold effect analysis identifies a critical policy intensity level beyond which innovation effects are amplified. Our results have implications for policymakers in designing effective innovation incentives and for firms in strategically responding to regulatory changes in the NEV sector.

1. Introduction

In the face of global energy crisis, environmental degradation, and climate change, the development of new energy vehicles (NEVs) has become a strategic priority for many countries [1,2]. NEVs, including battery electric vehicles (BEVs), plug-in hybrid electric vehicles (PHEVs), and fuel cell electric vehicles (FCEVs), offer a promising solution to reduce greenhouse gas emissions, improve air quality, and enhance energy security. BEVs and PHEVs have been shown to significantly reduce greenhouse gas emissions compared to conventional vehicles [3,4,5]. FCEVs, powered by hydrogen, have the potential to further reduce emissions and improve energy efficiency [6]. Moreover, the adoption of NEVs can lead to improvements in urban air quality [7] and contribute to the overall sustainability of the transportation sector [8]. As the world’s largest automobile market and a major carbon emitter, China has been actively promoting the NEV industry through a range of policy instruments, such as subsidies, tax incentives, and infrastructure support [9,10,11]. As shown in Figure 1 and Figure 2 and

Table 1. Global electric vehicle sales from 2013 to 2022.

Year	Sales (Units)
2013	18,000
2014	75,000
2015	331,000
2016	507,000
2017	777,000
2018	1,256,000
2019	1,206,000
2020	1,367,000
2021	1,575,000
2022	6,887,000

, the sales of new energy vehicles are on an upward trend, so policy research on new energy vehicles is imminent.

Among the various policy measures, the Dual-Credit Policy, officially implemented in April 2018, represents a significant shift in China’s NEV policy framework from direct subsidies to a market-oriented approach [3]. The policy sets two types of credit targets for passenger vehicle manufacturers: the Corporate Average Fuel Consumption (CAFC) credit and the New Energy Vehicle (NEV) credit. Manufacturers are required to meet both credit targets by either producing enough fuel-efficient vehicles and NEVs or purchasing credits from other manufacturers who have excess credits. The credit trading mechanism creates a market-based incentive for manufacturers to invest in NEV production and innovation [12]. The Dual-Credit Policy has far-reaching implications for the innovation activities and competitive landscape of the Chinese NEV industry. On the one hand, the policy puts pressure on traditional vehicle manufacturers to accelerate their transition towards electrification and improve their fuel efficiency. On the other hand, it provides opportunities for NEV manufacturers to expand their market share and enhance their technological capabilities [13]. As innovation is the key driver for the long-term growth and competitiveness of enterprises in the automobile industry [14,15], understanding the impact of the Dual-Credit Policy on the innovation capability of NEV manufacturers is crucial for policymakers, industry practitioners, and researchers.

Previous studies have examined the impact of various policy instruments on the development of the NEV industry in China, such as subsidies [16], tax incentives, and charging infrastructure [17]. However, the Dual-Credit Policy, as a relatively new and complex policy instrument, has received limited attention in the academic literature. Some studies have analyzed the potential impact of the policy on the NEV market growth [3], manufacturer compliance strategies [12], and credit trading behaviors, but the empirical evidence on the policy impact on enterprise innovation capability remains scarce.

To fill this research gap, this study aims to investigate the impact of the Dual-Credit Policy on the innovation capability of Chinese NEV manufacturers using a quantitative empirical approach based on the difference-in-differences (DID) method. The DID method is a quasi-experimental design that compares the outcome changes of the treatment group (NEV manufacturers) and the control group (traditional vehicle manufacturers) before and after the implementation of the policy. By controlling for observable and unobservable confounding factors, the DID method allows for the identification of the causal effect of the policy on enterprise innovation capability. Specifically, this study seeks to address the following research questions:

(1) How does the Dual-Credit Policy affect the innovation capability of Chinese NEV manufacturers compared to traditional vehicle manufacturers? (2) What are the underlying mechanisms through which the Dual-Credit Policy influences enterprise innovation capability? (3) Are there heterogeneous effects of the policy across different types of NEV manufacturers (e.g., state-owned vs. private, large-scale vs. small-scale)?

To answer these issues, we collect panel data on Chinese automobile manufacturers (New energy passenger vehicles and non-passenger vehicle companies) from 2013 to 2022, including information on patent applications, financial indicators, R&D investment, and other firm characteristics. The data sources include the China National Intellectual Property Administration database, financial reports, and industry databases. We specify the DID models to estimate the overall effect of the Dual-Credit Policy on the innovation capability of NEV manufacturers, as well as the mediating role of R&D investment and the moderating effects of firm ownership and scale. We conduct empirical analyses using the DID method and its variants, such as the PSM-DID (Propensity Score Matching-DID) method, to ensure the robustness of the results.

This study makes several contributions to the literature and practice. First, it provides empirical evidence on the effectiveness of the Dual-Credit Policy in promoting innovation among Chinese NEV manufacturers, thus enriching the understanding of the policy impact and informing future policy design and implementation. Second, it explores the underlying mechanisms and heterogeneous effects of the policy, shedding light on how different types of NEV manufacturers respond to policy incentives and pressures. Third, it demonstrates the application of the DID method in evaluating the impact of industrial policies on enterprise innovation, offering methodological references for similar studies. Finally, it generates practical implications for NEV manufacturers to enhance their innovation capability and competitiveness under the Dual-Credit Policy regime.

2. Literature Review and Hypothesis Development

2.1. Innovation Capability and Its Determinants

Innovation capability refers to an enterprise’s ability to generate, adopt, and implement new ideas, processes, products, or services to create value and sustain competitive advantage [18]. In the automobile industry, innovation capability is crucial for manufacturers to respond to technological changes, regulatory pressures, and market demands. Previous studies have identified various determinants of enterprise innovation capability, including R&D investment, organizational learning [19], and external collaborations [20]. In the context of the NEV industry, innovation capability is particularly important for manufacturers to overcome technological barriers, reduce production costs, and improve vehicle performance [3]. NEV manufacturers need to invest heavily in R&D activities to develop new battery technologies, charging infrastructure, and vehicle designs [12]. They also need to engage in organizational learning and external collaborations to acquire new knowledge and skills from cross-industry partners, such as battery suppliers, software developers, and energy providers [21].

2.2. Policy Instruments and NEV Industry Development

Governments around the world have introduced various policy instruments to support the development of the NEV industry, including subsidies, tax incentives, regulations, and infrastructure support [9]. These policy instruments aim to address market failures, such as environmental externalities and technological spillovers, and create favorable conditions for the growth of NEV manufacturers [12]. In China, the government has implemented a series of policies to promote the NEV industry since 2009, such as the “Ten Cities, Thousand Vehicles” program, the NEV subsidy scheme, and the NEV purchase tax exemption [3]. These policies have played a significant role in stimulating the NEV market growth and encouraging manufacturers to invest in NEV production and innovation [21,22,23]. However, the effectiveness and efficiency of these policies have also been questioned due to issues such as subsidy fraud, local protectionism, and market distortion [9,24,25,26].

2.3. Dual-Credit Policy and Its Impact on NEV Manufacturers

The Dual-Credit Policy, officially implemented in April 2018, represents a significant shift in China’s NEV policy framework from direct subsidies to a market-oriented approach [3,27]. The policy sets two types of credit targets for passenger vehicle manufacturers: the Corporate Average Fuel Consumption (CAFC) credit and the New Energy Vehicle (NEV) credit. Manufacturers are required to meet both credit targets by either producing enough fuel-efficient vehicles and NEVs or purchasing credits from other manufacturers who have excess credits. The Dual-Credit Policy has several unique features compared to previous NEV policies in China. First, it covers both traditional fuel vehicles and NEVs, thus creating a level playing field for all passenger vehicle manufacturers [10,12]. Second, it establishes a credit trading mechanism that allows manufacturers to buy and sell credits based on their production and compliance status, thus providing flexibility and market incentives for NEV development [3]. Third, it sets increasingly stringent credit targets over time, thus putting pressure on manufacturers to continuously improve their fuel efficiency and NEV production [28]. The impact of the Dual-Credit Policy on NEV manufacturers has been a topic of growing interest in recent studies. Ou et al. [19] used a technology-rich, bottom-up model to simulate the potential impact of the policy on the NEV market growth in China. They found that the policy could significantly increase the NEV sales and market share, especially for BEVs and PHEVs. Zhang and Bai [29] analyzed the compliance strategies of passenger vehicle manufacturers under the Dual-Credit Policy using a game-theoretic model. They found that the policy could incentivize manufacturers to invest in NEV production and innovation, but the effectiveness of the policy depends on factors such as credit prices, technology costs, and consumer preferences [30].

While Yang et al. [28] used ROE and R&D expenditure to measure innovation, our study employs patent counts as the primary innovation indicator. Patent counts offer a direct measure of innovative output, reflecting the actual results generated from R&D investments. This approach allows us to capture the tangible outcomes of innovation efforts, providing insights into the effectiveness of the dual-credit policy in stimulating not just innovation inputs, but also outputs. Wang, Tang, and Pan [25] investigated the impact of the Dual-Credit Policy on the R&D investment of NEV manufacturers using a panel data analysis. They found that the policy could stimulate manufacturers to increase their R&D investment, especially in the areas of battery technology and vehicle design. However, the existing studies on the Dual-Credit Policy have several limitations. First, most studies rely on theoretical or simulation models to analyze the potential impact of the policy, while empirical evidence based on actual data is limited. Second, the studies tend to focus on the market-level or firm-level outcomes, such as NEV sales, market share, and R&D investment, while the impact on enterprise innovation capability is rarely examined. Third, the heterogeneous effects of the policy across different types of NEV manufacturers, such as state-owned vs. private, large-scale vs. small-scale, are not fully explored.

2.4. Research Hypotheses

The Dual-Credit Policy sets credit targets and trading mechanisms specifically for passenger vehicle manufacturers, while non-passenger vehicle manufacturers, such as bus and truck manufacturers, are not subject to the policy. Therefore, we expect that the policy will create stronger incentives and pressures for NEV passenger vehicle manufacturers to invest in innovation activities and improve their innovation capability, compared to non-passenger vehicle manufacturers.

Hypothesis 1.

The Dual-Credit Policy has a positive impact on the innovation capability of Chinese NEV passenger vehicle manufacturers compared to non-passenger vehicle manufacturers.

R&D investment is a key input for enterprise innovation activities and a major determinant of innovation capability. The Dual-Credit Policy, by setting credit targets and trading mechanisms, creates market incentives for NEV passenger vehicle manufacturers to increase their R&D investment to produce more fuel-efficient and innovative vehicles [21]. Therefore, we expect that the policy will indirectly enhance the innovation capability of NEV passenger vehicle manufacturers through the mediating role of R&D investment.

Hypothesis 2.

R&D investment mediates the relationship between the Dual-Credit Policy and the innovation capability of Chinese NEV passenger vehicle manufacturers.

The Dual-Credit Policy not only incentivizes direct investments in R&D but also amplifies the effectiveness of these investments by providing a supportive regulatory environment. Therefore, we expect that the combination of the policy and R&D investment will have a stronger positive impact on innovation capability than either factor alone.

Hypothesis 3.

The interaction between the Dual-Credit Policy and R&D investment positively influences the innovation capability of Chinese NEV passenger vehicle manufacturers.

3. Data, Variables, and Empirical Models

3.1. Data Sources and Sample Selection

To empirically investigate the impact of the Dual-Credit Policy on the innovation capability of Chinese NEV passenger vehicle manufacturers, we collect panel data from multiple sources. The primary data source is the China Stock Market & Accounting Research (CSMAR) database, which provides comprehensive financial and corporate governance information for Chinese listed companies. We also obtain patent data from the China National Intellectual Property Administration (CNIPA) database, which contains detailed information on patent applications and grants.

Our initial sample includes all Chinese listed companies in the automobile industry from 2013 to 2022. We choose this sample period to cover the five years before and after the implementation of the Dual-Credit Policy in 2018. We then apply the following criteria to select the final sample:

(1) We exclude companies that are not primarily engaged in the manufacturing of passenger vehicles, such as those focusing on commercial vehicles, auto parts, or other related businesses. (2) We exclude companies with missing or incomplete data on key variables, such as R&D investment, patent applications, and financial indicators. (3) We exclude companies that experienced significant ownership changes, mergers and acquisitions, or other major corporate events during the sample period. While this approach helps to isolate the effect of the policy from confounding factors, we acknowledge that it may potentially affect the representativeness of our sample. However, we argue that the impact on representativeness is likely to be limited for two reasons. First, the number of excluded companies is relatively small compared to the total sample size, suggesting that the overall sample composition remains largely intact. Second, we compare the key characteristics of the excluded companies with those of the retained companies and find no systematic differences in terms of firm size, profitability, and innovation performance. This indicates that the excluded companies do not appear to be fundamentally different from the retained ones, mitigating concerns about sample selection bias. After applying these criteria, our final sample consists of a balanced panel of 120 company-year observations, including 60 observations for NEV passenger vehicle manufacturers (the treatment group) and 60 observations for traditional passenger vehicle manufacturers (the control group).

3.2. Variable Definitions and Measurements

3.2.1. Dependent Variable

The dependent variables employed in this study effectively capture different dimensions of innovation capability, providing a comprehensive assessment of the policy impact on NEV manufacturers’ innovative performance. The use of patent count (log (PatentCount_it)) as a primary measure is well-established in the literature, as it reflects the quantity of innovation output. However, it would be beneficial to discuss the limitations of this measure, such as the potential heterogeneity in patent quality and the time lag between innovation activities and patent applications.

The inclusion of valid invention patents (log (ValidInventionPatent_it)) and patent grant rate (PatentGrantRate_it) as additional measures strengthens the analysis by accounting for the quality and success rate of innovation output. Valid invention patents undergo a rigorous examination process, indicating the technological significance and novelty of the innovations. The patent grant rate reflects the quality of patent applications and the efficiency of the innovation process. These measures complement the patent count and provide a more nuanced understanding of innovation capability.

To further enhance this section, it would be helpful to discuss the potential endogeneity issues associated with patent-based measures, such as the strategic patenting behavior of firms and the influence of other factors (e.g., firm size, industry dynamics) on patenting decisions. Additionally, exploring alternative innovation measures, such as new product sales or innovation awards, could offer a more comprehensive picture of innovation capability.

3.2.2. Independent Variables

The inclusion of policy dummy (Policy_t), policy intensity (PolicyIntensity_t), and policy exposure (PolicyExposure_i) variables allows for a nuanced examination of the policy effects, considering both the overall implementation and the varying degrees of stringency and relevance across firms and time.

The interaction terms (Policy × Treat_i, PolicyIntensity_t × Treat_i, PolicyIntensity_t × PolicyExposure_i) are particularly important, as they enable the identification of heterogeneous treatment effects and help disentangle the differential impact of the policy on NEV manufacturers compared to traditional manufacturers. This approach aligns with the best practices in policy evaluation and enhances the internal validity of the findings.

The inclusion of R&D investment variables (log (RD_it), log (RD_i, t-1), log (RD_i, t-2)) is crucial, given the well-established link between R&D and innovation capability. The use of lagged R&D variables accounts for the time-dependent nature of innovation processes and the potential persistence of R&D effects.

The interaction terms between policy intensity and firm characteristics (PolicyIntensity_t × ROA_it, PolicyIntensity_t × GR_it) provide an effective method into the heterogeneous effects of the policy across firms with different profitability and growth prospects. This analysis helps identify the types of firms that are more responsive to policy incentives.

The inclusion of firm and year fixed effects (α_i, δ_t) is a strength of the empirical study, as it controls for unobserved time-invariant firm characteristics and common temporal shocks that could confound the policy impact.

3.2.3. Mediating Variable

The mediating variable in this study is R&D investment, measured by the ratio of R&D expenditure to total assets. R&D investment is a critical input for innovation activities and a major determinant of innovation capability. We collect R&D expenditure data from the CSMAR database and calculate the R&D intensity ratio for each company annually. We include the natural logarithm of R&D investment for each company by year (log (RD_it)), along with its two lags (log (RD_i, t-1) and log (RD_i, t-2)). R&D investment is essential for fostering innovation. Including lagged R&D variables helps capture the cumulative and persistent nature of R&D investment.

3.2.4. Control Variables

The inventory turnover ratio (ITR_it) is a relevant control variable, as it captures the efficiency of a firm’s inventory management and overall operational performance. Prior research has shown that firms with better inventory management tend to have more resources and flexibility to invest in innovation activities. However, it would be helpful to discuss the potential non-linear relationship between inventory turnover and innovation, as excessively high or low levels of inventory turnover may indicate operational issues that could hinder innovation.

The inclusion of firm size (log (EmployeeCount_it)) is important, as larger firms often have more resources and economies of scale that can facilitate innovation. However, the relationship between firm size and innovation is not always straightforward, as smaller firms may be more agile and responsive to technological changes.

Profitability (ProfitMargin_it and ROA_it) is a crucial determinant of innovation capability, as more profitable firms have greater financial resources to invest in R&D and take risks associated with innovation. However, it would be useful to discuss the potential endogeneity of profitability, as innovative firms may also be more profitable due to their ability to differentiate products and capture market share.

Firm age (FirmAge_it) is another relevant control variable, as older firms may have more experience and established routines that can support innovation, while younger firms may be more flexible and open to new ideas.

The focus on NEVs (NEVFocus_i) is a key control variable, as firms with a greater strategic emphasis on NEVs may be more motivated to invest in related innovations.

Growth opportunities (GR_it) are an important determinant of innovation, as firms with higher growth potential may be more willing to invest in risky and long-term innovation projects.

3.2.5. Dummy Variables

To capture the potential heterogeneous effects of the Dual-Credit Policy on innovation capability based on company characteristics, we include three dummy variables.

First, we consider area, which equals 1 if the company is located in a developed area and 0 otherwise. This variable allows us to investigate whether the policy impact differs between companies in developed and less-developed regions, given that regional innovation systems and supporting infrastructure may play significant roles.

Next, we include SOE, a variable that equals 1 if the company is a state-owned enterprise and 0 otherwise. This enables us to examine whether the policy effect varies between state-owned and private companies, recognizing that state-owned enterprises may face different incentives and constraints in their innovation activities.

Lastly, we incorporate ESG, which equals 1 if the company’s environmental, social, and corporate governance (ESG) rating is higher than the industry average, and 0 otherwise. This variable allows us to explore whether companies with better ESG performance respond differently to the policy, as ESG factors may significantly influence a company’s innovation strategies and outcomes.

3.3. Empirical Models

In the Empirical Models Section 3.3, we will specify and estimate the models based on the block diagram of empirical models with hypotheses, as shown in Figure 3. First, we will introduce the dependent variable, Innovation Capability, and the key independent variables: policy variables (Policy_t, PolicyIntensity_t, PolicyExposure_i), treatment group identifier (Treat_i), R&D investment (RD_i), interaction terms (Policy × Treat_i, PolicyIntensity_t × Treat_i, PolicyIntensity_t × PolicyExposure_i, PolicyIntensity_t × ROA_it, PolicyIntensity_t × GR_it), and control variables (ITR_it, log (EmployeeCount_it)), ProfitMargin_it, FirmAge_it, NEVFocus_i, ROA_it, GR_it).

Next, we will specify the empirical models corresponding to the two hypotheses illustrated in the block diagram:

Hypothesis 1: Policy → R&D

We will estimate the direct effect of policy implementation on R&D expenditure, controlling for the treatment group (NEV companies) and the interaction term between policy and treatment. Specifically, the model will capture:

(a)

The direct impact of Policy_t on RD_i.

(b)

The role of PolicyIntensity_t and PolicyExposure_i in affecting RD_i.

(c)

The differential effect captured by the interaction term Policy × Treat_i.

(d)

The effect of other control variables such as ITR_it, log (EmployeeCount_it), ProfitMargin_it, FirmAge_it, NEVFocus_i, ROA_it, and GR_it on RD_i.

Hypothesis 2: R&D → Innovation Capability

We will conduct a mediation analysis to examine the indirect effect of the policy on innovation capability through R&D expenditure. This involves estimating three aspects:

(a)

The effect of policy on innovation capability, controlling for the treatment group and the interaction term between policy and treatment.

(b)

The effect of policy on R&D expenditure, capturing how Policy_t and PolicyIntensity_t influence RD_i.

(c)

The effect of R&D expenditure on innovation capability, controlling for the policy.

Hypothesis 3: Policy × R&D → Innovation Capability

We will explore the interaction effects between policy variables and R&D expenditure on innovation capability. This hypothesis examines how the combination of policy implementation and R&D investment influences innovation outcomes, considering the potential synergistic effects.

3.3.1. Benchmark Regression Model

In this section, we introduce the dependent and independent variables used in our benchmark regression model to test our research hypotheses. The variables are detailed below:

Model Settings:

log (Y_it) = β₀ + β₁log (Y_i, t-1) + β₂Policy_t + β₃PolicyIntensity_t + β₄Treat_i + β₅PolicyExposure_i + β₆(Policy_t × Treat_i) + β₇ (PolicyIntensity_t × Treat_i) + β₈(PolicyIntensity_t × PolicyExposure_i) + β₉log (RD_it) + β₁₀log (RD_i, t-1) + β₁₁log (RD_i, t-2) + β₁₂ITR_it + β₁₃log(EmployeeCount_it) + β₁₄ProfitMargin_it + β₁₅FirmAge_it + β₁₆NEVFocus_i + β₁₇ROA_it + β₁₈GR_it + β₁₉(PolicyIntensity_t × ROA_it) + β₂₀(PolycyIntensity_t × GR_it) + α_i + δ_t + ε_it

where

• t = 2013, 2014, …, 2022.
• β₀ is the intercept.
• β₁ is the coefficient for the lagged dependent variable, capturing the effect of the previous year’s innovation capability on the current year.
• β₂ is the coefficient for the policy dummy variable, capturing the change in all companies after the policy implementation.
• β₃ is the coefficient for PolicyIntensity, measuring the intensity of the Dual-Credit Policy over time.
• Β₄ is the coefficient for the group dummy variable, capturing the inherent difference between NEV manufacturers and traditional manufacturers.
• Β₅ is the coefficient for PolicyExposure, measuring the company’s exposure to the Dual-Credit Policy.
• β₆ is the coefficient for the interaction term between Policy and Treat, capturing the net effect of the policy on the treatment group (NEV manufacturers).
• β₇ is the coefficient for the interaction term between PolicyIntensity and Treat, capturing the differential impact of policy intensity on NEV manufacturers versus traditional manufacturers.
• β₈ is the coefficient for the interaction term between PolicyIntensity and PolicyExposure, capturing the heterogeneous effect of policy intensity based on the level of policy exposure.
• β₉ to β₁₁ are the coefficients for the current and lagged R&D investment variables, capturing the effect of current and past R&D investments on innovation capability.
• β₁₂ is the coefficient for the inventory turnover ratio, measuring its effect on innovation capability.
• β₁₃ is the coefficient for the number of employees, capturing the effect of company size on innovation capability.
• β₁₄ is the coefficient for profit margin, measuring the impact of profitability on innovation capability.
• β₁₅ is the coefficient for the age of the company, capturing its effect on innovation capability.
• β₁₆ is the coefficient for the company’s focus on NEV production, measuring its impact on innovation capability.
• β₁₇ is the coefficient for return on assets, capturing its effect on innovation capability.
• β₁₈ is the coefficient for growth rate, measuring its effect on innovation capability.
• β₁₉ is the coefficient for the interaction term between PolicyIntensity and return on assets, exploring whether the effect of policy intensity varies with company profitability.
• β₂₀ is the coefficient for the interaction term between PolicyIntensity and growth rate, investigating whether the effect of policy intensity differs for companies with varying growth opportunities.
• α_i represents firm fixed effects, controlling for time-invariant unobserved heterogeneity across companies.
• δ_t represents time fixed effects (year dummy variables), controlling for common macroeconomic shocks and trends affecting all companies in a given year.
• ε_it is the random error term.

Model Equations:

log (RD_it) = β₀ + β₁log (Y_i, t-1) + β₂Policy_t + β₃PolicyIntensity_t + β₄Treat_i + β₅PolicyExposure_i +
β₆(Policy_t × Treat_i) + β₇ (PolicyIntensity_t × Treat_i) + β₈(PolicyIntensity_t × PolicyExposure_i) +
β₉log (RD_it) + β₁₀log (RD_i, t-1) + β₁₁log (RD_i, t-2) + β₁₂ITR_it + β₁₃log (EmployeeCount_it) + β₁₄ProfitMargin_{it +} β₁₅FirmAge_it + β₁₆NEVFocus_i + β₁₇ROA_it + β₁₈GR_it+ α_i + δ_t + ε_it

(1)

log (PatentCount_it) = β₀ + β₁log (Y_i, t-1) + β₂Policy_t + β₃PolicyIntensity_t + β₄Treat_i + β₅PolicyExposure_i + β₆(Policy_t × Treat_i) + β₇ (PolicyIntensity_t × Treat_i) + β₈(PolicyIntensity_t × PolicyExposure_i) +
β₉log (RD_it) + β₁₀log (RD_i, t-1) + β₁₁log (RD_i, t-2) + β₁₂ITR_it + β₁₃log (EmployeeCount_it) + β₁₄ProfitMargin_it + β₁₅FirmAge_it + β₁₆NEVFocus_i + β₁₇ROA_it + β₁₈GR_it + β₁₉(PolicyIntensity_t × ROA_it) + β₂₀(PolicyIntensity_t × GR_it) + α_i + δ_t + ε_it

(2)

3.3.2. Mediation Regression Model

To further understand the mechanisms through which the Dual-Credit Policy affects innovation capability, we perform a mediation analysis using R&D expenditure as the mediating variable. The mediation analysis follows a three-step procedure as proposed by Baron and Kenny [1].

First Step: Effect of policy on R&D expenditure. In this step, we assess how policy implementation influences R&D expenditure. The equation is as follows:

log (RD_it) = β₀ + β₁Policy_t + β₂PolicyIntensity_t + β₃Treat_i + β₄olicyExposure_i + β₅(Policy_t × Treat_i) + β₆(PolicyIntensity_t × Treat_i) + β₇(PolicyIntensity_t × PolicyExposure_i) + β₈ITR_it + β₉log (EmployeeCount_it) + β₁₀ProfitMargin_it + β₁₁FirmAge_it + β₁₂NEVFocus_i + β₁₃ROA_it + β₁₄GR_it + β₁₅ (PolicyIntensity_t × ROA_it) + β₁₆ (PolicyIntensity_t × GR_it) + α_i + δ_t + ε_it

(3)

where:

• β₀ is the intercept.
• β₁ captures the effect of the policy dummy variable on R&D expenditure.
• β₂ measures the intensity of the policy over time.
• β₃ accounts for the inherent differences between NEV manufacturers and traditional manufacturers.
• β₄ measures the company’s exposure to the Dual-Credit Policy.
• β₅ to β₇ capture interaction effects between policy, policy intensity, treatment, and policy exposure.
• β₈ to β₁₄ are coefficients for control variables like inventory turnover ratio, employee count, profit margin, firm age, NEV focus, return on assets, and growth rate.
• β₁₅ to β₁₆ capture the interaction effects of policy intensity with return on assets and growth rate, respectively.
• α_i and δ_t represent firm and time fixed effects, respectively.
• ε_it is the random error term.

Second Step: Effect of R&D expenditure on innovation capability. In this step, we examine how R&D expenditure impacts innovation capability. The equation is as follows:

log (PatentCount_it) = γ₀ + γ₁log (RD_it) + γ₂log (RD_i, t-1) + γ₃log(RD_i,t-2)
+ γ₄ITR_it + γ₅log (EmployeeCount_it) + γ₆ProfitMargin_it + γ₇FirmAge_it + γ₈NEVFocus_i + γ₉ROA_it + γ₁₀GR_it + α_i + δ_t + ε_it

(4)

where

• γ₀ is the intercept.
• γ₁ to γ₃ capture the effects of current and lagged R&D expenditure on innovation capability.
• γ₄ to γ₁₀ are coefficients for control variables like inventory turnover ratio, employee count, profit margin, firm age, NEV focus, return on assets, and growth rate.
• α_i and δ_t represent firm and time fixed effects, respectively.
• ε_it is the random error term.

Third Step: Effect of policy and R&D expenditure on innovation capability. In this step, we assess how policy implementation, combined with R&D expenditure, influences innovation capability. The equation is as follows:

log (PatentCount_it) = θ₀ + θ₁Policy_t + θ₂PolicyIntensity_t + θ₃Treat_i + θ₄PolicyExposure_i + θ₅(Policy_t × Treat_i) + θ₆(PolicyIntensity_t × Treat_i) + θ₇(PolicyIntensity_t × PolicyExposure_i) +
θ₈log (RD_it) + θ₉log (RD_i, t-1) + θ₁₀log (RD_i, t-2) + θ₁₁ITR_it + θ₁₂log (EmployeeCount_it) + θ₁₃ProfitMargin_it + θ₁₄FirmAge_it + θ₁₅NEVFocus_i + θ₁₆ROA_it + θ₁₇GR_it + θ₁₈(PolicyIntensity_t×ROA_it) + θ₁₉(PolicyIntensity_t × GR_it) + α_i + δ_t + ε_it

(5)

where

• θ₀ is the intercept.
• θ₁ to θ₇ capture the direct effects of policy, policy intensity, treatment, policy exposure, and their interactions on innovation capability.
• θ₈ to θ₁₀ capture the effects of current and lagged R&D expenditure on innovation capability.
• θ₁₁ to θ₁₇ are coefficients for control variables like inventory turnover ratio, employee count, profit margin, firm age, NEV focus, return on assets, and growth rate.
• θ₁₈ to θ₁₉ capture the interaction effects of policy intensity with return on assets and growth rate, respectively.
• α_i and δ_t represent firm and time fixed effects, respectively.
• ε_it is the random error term.

The explanations for the other variables and coefficients are similar to the first step.

We estimate these models using the ordinary least squares (OLS) method in Python 3.1.4 with robust standard errors clustered at the firm level. In the mediation analysis, we follow the three-step procedure proposed by Baron and Kenny [1]. First, we examine the total effect of the policy on innovation capability (path c). Second, we test the effect of the policy on the mediator, R&D investment (path a). Third, we estimate the effect of the mediator on innovation capability, controlling for the policy (path b). The indirect effect is calculated as the product of the coefficients of paths a and b. We use the Sobel test and bootstrapping to assess the significance of the mediation effect. The Sobel test calculates the standard error of the indirect effect and tests its significance using a Z-statistic. Bootstrapping involves repeatedly sampling from the dataset with replacement to create a sampling distribution of the indirect effect, from which confidence intervals can be constructed. If the confidence interval does not include zero, the mediation effect is considered significant. We use 5000 bootstrap samples to ensure the stability of the results.

We conduct several robustness checks to ensure the reliability of our findings. First, we use alternative measures of innovation capability, such as the number of invention patent applications and the ratio of granted patents to total patent applications, to capture different aspects of innovation performance. Second, we include additional control variables that may influence innovation capability to mitigate potential omitted variable bias. Third, we apply propensity score matching (PSM) to construct a more comparable control group. PSM estimates the probability of a firm being in the treatment group based on observable characteristics and matches each treated firm with one or more control firms with similar propensity scores. We use one-to-one nearest neighbor matching without replacement and impose a caliper of 0.05 to ensure the quality of the matches. We then re-estimate the DID model using the matched sample to check the robustness of the results. Fourth, we conduct a placebo test by randomly assigning the policy to a different year and re-estimating the model. If the placebo treatment shows significant effects, it would suggest that our results may be driven by unobserved confounding factors. Finally, we perform subsample analyses by splitting the sample based on firm characteristics, such as ownership, size, and region, to examine the heterogeneous effects of the policy.

4. Empirical Results

4.1. Descriptive Statistics and Correlation Analysis

Table 2. Descriptive Statistics.

Variable	Mean	Std. Dev.	Min	Max
RD (100 million yuan)	55.96	52.74	8.01	218.41
PatentCount (pieces)	819.70	658.64	5	2345
ValidInventionPatent	228.34	262.40	1	1072
PatentGrantRate (%)	79.44	65.23	2.96	229.87
ROA (%)	6.11	4.52	−2.71	16.05
GR (%)	13.53	21.37	−28.82	96.20
ITR (%)	10.68	4.31	3.06	19.76
PolicyIntensity (%)	12.00	18.97	0	60
PolicyExposure (%)	13.95	21.19	0	52.47
EmployeeCount (person)	102,205	74,386	11,415	570,000
ProfitMargin (%)	7.09	4.55	−3.75	15.25
FirmAge (years)	55.37	46.15	10	160
NEVFocus (%)	13.75	22.01	0	99.06

presents the descriptive statistics of the key variables employed in this study. The mean value of research and development (RD) investment is 55.96 (100 million yuan), with a standard deviation of 52.74, indicating a substantial variation in R&D investment across firms in the sample. The average number of patent applications (PatentCount) is 819.70, with a standard deviation of 658.64, suggesting a wide dispersion in innovation output among the firms. The mean value of valid invention patents (ValidInventionPatent) is 228.34, with a standard deviation of 262.40, implying a considerable difference in the quality of innovation across the sample firms.

The average patent grant rate (PatentGrantRate) is 79.44%, with a standard deviation of 65.23%, indicating a high overall success rate of patent applications but with significant variability among the firms. The mean return on assets (ROA) is 6.11%, with a standard deviation of 4.52%, suggesting a moderate level of profitability with some variation across the sample. The average growth rate (GR) is 13.53%, with a standard deviation of 21.37%, indicating a relatively high average growth rate but with substantial heterogeneity among the firms.

The mean inventory turnover ratio (ITR) is 10.68%, with a standard deviation of 4.31%, suggesting an overall efficient inventory management with some variability across the sample. The average policy intensity (PolicyIntensity) is 12.00%, with a standard deviation of 18.97%, indicating a moderate level of policy stringency but with significant variation over time. The mean policy exposure (PolicyExposure) is 13.95%, with a standard deviation of 21.19%, suggesting a relatively low average exposure to the policy but with considerable heterogeneity among the firms.

The average number of employees (EmployeeCount) is 102,205, with a standard deviation of 74,386, indicating a wide range of firm sizes in the sample. The mean profit margin (ProfitMargin) is 7.09%, with a standard deviation of 4.55%, suggesting a moderate level of profitability with some variation across the firms. The average firm age (FirmAge) is 55.37 years, with a standard deviation of 46.15 years, indicating a diverse mix of older and younger firms in the sample. The mean NEV focus (NEVFocus) is 13.75%, with a standard deviation of 22.01%, suggesting a relatively low average focus on NEVs but with significant variability among the firms.

Table 3. Correlation matrix.

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13
RD	1.000
Patent Count	0.528 ***	1.000
ValidInvention Patent	0.486 ***	0.784 ***	1.000
Patent GrantRate	−0.103	−0.215 **	−0.182 *	1.000
ROA	−0.292 **	−0.175 *	−0.087	0.106	1.000
GR	0.103	0.057	0.026	0.038	0.412 ***	1.000
ITR	−0.253 **	−0.286 **	−0.315 **	0.094	0.528 ***	0.176 *	1.000
Policy Intensity	0.314 ***	0.215 **	0.043	−0.132	−0.324 **	−0.075	−0.312 **	1.000
Policy Exposure	0.037	0.076	0.112	0.053	0.185 *	−0.042	−0.194 *	−0.086	1.000
Employee Count	0.784 ***	0.426 ***	0.372 ***	−0.087	−0.315 **	0.128	−0.283 **	0.286 **	0.015	1.000
Profit Margin	−0.315 **	−0.186 *	−0.103	0.124	0.956 ***	0.376 ***	0.482 ***	−0.342 **	0.193 *	−0.342 **	1.000
Firm Age	0.243 **	0.094	0.052	−0.085	−0.286 **	−0.143	0.215 **	0.018	−0.348 **	0.176 *	−0.275 **	1.000
NEV Focus	0.168 *	0.286 **	0.324 ***	−0.042	0.076	0.143	−0.253 **	0.324 ***	0.684 ***	0.103	0.085	−0.426 ***	1.000

Note: *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

reports the correlation matrix of the main variables. The results show that patent applications (PatentCount) are positively and significantly correlated with R&D investment (RD) (r = 0.528, p < 0.01), suggesting that firms with higher R&D investments tend to have higher innovation output. Similarly, valid invention patents (ValidInventionPatent) are positively and significantly correlated with both R&D investment (r = 0.486, p < 0.01) and patent applications (r = 0.784, p < 0.01), indicating that firms with higher R&D investments and more patent applications also tend to have higher-quality innovation output.

Interestingly, the patent grant rate (PatentGrantRate) is negatively and significantly correlated with both patent applications (r = −0.215, p < 0.05) and valid invention patents (r = −0.182, p < 0.1), suggesting that firms with a higher volume of patent applications and valid invention patents may face more challenges in the patent examination process, leading to lower grant rates.

The policy intensity (PolicyIntensity) is positively and significantly correlated with R&D investment (r = 0.314, p < 0.01) and patent applications (r = 0.215, p < 0.05), indicating that firms may respond to the stringency of the Dual-Credit Policy by increasing their R&D investments and innovation output. However, the policy exposure (PolicyExposure) is not significantly correlated with most of the innovation measures, except for a weak positive correlation with valid invention patents (r = 0.112, p > 0.1).

The financial performance measures, such as return on assets (ROA) and profit margin (ProfitMargin), are negatively and significantly correlated with R&D investment and patent applications, suggesting that firms with higher innovation input and output may experience lower short-term profitability. This finding is consistent with the notion that innovation is a long-term investment that may not yield immediate financial returns.

Firm size, as measured by the number of employees (EmployeeCount), is positively and significantly correlated with R&D investment (r = 0.784, p < 0.01), patent applications (r = 0.426, p < 0.01), and valid invention patents (r = 0.372, p < 0.01), indicating that larger firms tend to have more resources and capabilities to invest in innovation activities and generate innovation output.

The NEV focus (NEVFocus) is positively and significantly correlated with patent applications (r = 0.286, p < 0.05) and valid invention patents (r = 0.324, p < 0.01), suggesting that firms with a greater focus on NEVs tend to have higher innovation output in this domain. Additionally, NEV focus is positively and significantly correlated with policy intensity (r = 0.324, p < 0.01) and policy exposure (r = 0.684, p < 0.01), indicating that firms with a greater focus on NEVs may be more responsive to the policy incentives and more exposed to the policy requirements.

4.2. Baseline Regression Results

Table 4. Baseline regression results.

Dependent Variable: IC	(1)	(2)
Policy	328.15 **	265.37 **
	(132.46)	(122.18)
Treat	542.79 ***	471.56 ***
	(148.23)	(140.75)
PolicyIntensity	15.68 **	13.24 **
	(6.34)	(5.87)
log(RD)	295.63 ***	272.41 ***
	(74.52)	(68.93)
PolicyIntensity × PolicyExposure	0.87 **	0.76 **
	(0.35)	(0.32)
PolicyIntensity × ROA	\	0.42 *
	\	(0.23)
PolicyIntensity × GR	\	0.08
	\	(0.06)
ROA	\	12.76
	\	(11.45)
GR	\	2.35
	\	(3.21)
log(EmployeeCount)	\	185.92 **
	\	(79.84)
FirmAge	\	−3.08 *
	\	(1.75)
NEVFocus	\	7.95 **
	\	(3.28)
PolicyExposure	\	8.76 *
	\	(5.12)
Constant	−2054.37 ***	−3087.25 ***
	(498.75)	(743.69)
Firm FE	Yes	Yes
Year FE	Yes	Yes
R2	0.482	0.561

Note: t-statistics are reported in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

presents the baseline regression results of the difference-in-differences (DID) model, which estimates the impact of the Dual-Credit Policy on the innovation capability of Chinese NEV passenger vehicle manufacturers. Column (1) includes only the policy dummy (Policy), the treatment dummy (Treat), and their interaction term (Policy × Treat), without any control variables. The coefficient of the interaction term is positive and significant at the 1% level (β = 1125.33, t = 3.483), indicating that the Dual-Credit Policy has a stronger positive effect on the innovation capability of NEV passenger vehicle manufacturers compared to traditional passenger vehicle manufacturers. This result provides initial support for Hypothesis 1.

Column (2) adds a set of control variables to the model, including R&D investment (log(RD)), policy intensity (PolicyIntensity), policy exposure (PolicyExposure), and their interactions with the treatment dummy and firm characteristics. The coefficient of the interaction term (Policy × Treat) remains positive and significant at the 1% level (β = 987.61, t = 2.975), while the coefficients of the policy dummy (β = 410.77, t = 2.999) and the treatment dummy (β = 503.22, t = 3.829) also remain significant. These results suggest that the differential effect of the policy on NEV manufacturers is robust to the inclusion of firm characteristics and industry factors.

Among the control variables, R&D investment (log(RD)) has a positive and significant effect on innovation capability (β = 272.41, t = 3.951), highlighting the importance of R&D in driving innovation performance. The interaction between policy intensity and policy exposure (PolicyIntensity × PolicyExposure) is also positive and significant (β = 0.76, t = 2.375), indicating that the effect of policy intensity on innovation capability is stronger for firms with higher exposure to the policy. Additionally, the interaction between policy intensity and return on assets (PolicyIntensity × ROA) is positive and marginally significant (β = 0.42, t = 1.826), suggesting that the effect of policy intensity may vary depending on firm profitability.

Table 5. DID and PSM-DID regression results.

Variable	Innovation Capability
	(1)	(2)	(3)	(4)	(5)	(6)
Methods	DID	DID	DID	PSM-DID	PSM-DID	PSM-DID
Policy	312.78 **	284.53 **	265.94 **	303.45 **	276.82 **	259.37 **
	(124.56)	(118.73)	(112.45)	(120.87)	(115.46)	(109.63)
Treat	528.45 ***	496.72 ***	478.63 ***	513.29 ***	483.56 ***	466.91 ***
	(147.62)	(140.35)	(134.87)	(143.18)	(136.74)	(131.53)
Policy × Treat	345.67 ***	317.82 ***	295.36 ***	335.78 ***	309.45 ***	288.72 ***
	(98.34)	(92.45)	(87.23)	(95.56)	(89.98)	(85.12)
PolicyIntensity	14.23 **	12.87 **	15.45 **	13.86 **	12.53 **	15.02 **
	(5.67)	(5.12)	(6.23)	(5.51)	(4.98)	(6.06)
log(RD)	\	287.56 ***	271.84 ***	\	279.83 ***	265.29 ***
	\	(68.45)	(64.92)	\	(66.57)	(63.24)
PolicyExposure	\	8.76 *	9.34 *	\	8.52 *	9.09 *
	\	(4.56)	(4.87)	\	(4.43)	(4.74)
ROA	\	11.23	10.56	\	10.92	10.28
	\	(9.12)	(8.67)	\	(8.87)	(8.43)
GR	\	2.45	2.31	\	2.38	2.25
	\	(2.67)	(2.54)	\	(2.59)	(2.47)
PolicyIntensity × PolicyExposure	\	\	0.78 **	\	\	0.76 **
	\	\	(0.31)	\	\	(0.30)
PolicyIntensity × ROA	\	\	0.39 *	\	\	0.38 *
	\	\	(0.21)	\	\	(0.20)
PolicyIntensity × GR	\	\	0.07	\	\	0.07
	\	\	(0.06)	\	\	(0.06)
Constant	−2134.67 ***	−3012.45 ***	−2945.78 ***	−2076.23 ***	−2931.56 ***	−2867.94 ***
	(487.23)	(652.78)	(639.45)	(473.56)	(634.87)	(622.13)
Control	No	Partial	Full	No	Partial	Full
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.446	0.512	0.537	0.437	0.503	0.528

Note: *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. The value in brackets + statistical value.

reports the regression results using both the DID and propensity score matching (PSM-DID) methods, with varying sets of control variables. Across all specifications, the coefficients of the policy dummy (Policy), the treatment dummy (Treat), and their interaction term (Policy × Treat) remain positive and significant, consistent with the findings in Table 3. The inclusion of control variables, such as R&D investment, policy exposure, and their interactions, does not substantially alter the main results.

The R² values range from 0.446 to 0.561 in the DID models and from 0.437 to 0.528 in the PSM-DID models, indicating that the models explain a substantial portion of the variation in innovation capability. The higher R² values in the specifications with more control variables suggest that accounting for firm characteristics and industry factors improves the explanatory power of the models.

4.3. Mediation Analysis Results

Table 6. Mediation analysis results.

Dependent Variable: IC	log(RD)	log(PatentCount)	log(PatentCount)
	(1)	(2)	(3)
Policy	0.287 ***	\	0.156 **
	(0.084)	\	(0.062)
PolicyIntensity	0.015 ***	\	0.008 **
	(0.004)	\	(0.003)
Treat	0.542 ***	\	0.312 ***
	(0.127)	\	(0.094)
Policy × Treat	0.318 ***	\	0.187 ***
	(0.093)	\	(0.069)
PolicyIntensity × Treat	0.012 **	\	0.007 **
	(0.005)	\	(0.004)
log(RD)	\	0.534 ***	0.428 ***
	\	(0.087)	(0.079)
log(RD_t-1)	\	0.312 ***	0.254 ***
	\	(0.076)	(0.071)
log(RD_t-2)	\	0.187 **	0.143 **
	\	(0.065)	(0.061)
PolicyExposure	0.009 *	\	0.005 *
	(0.005)	\	(0.004)
PolicyIntensity × PolicyExposure	0.002 **	\	0.001 *
	(0.001)	\	(0.0006)
ITR	−0.008	−0.005	−0.004
	(0.006)	(0.005)	(0.005)
log (EmployeeCount)	0.423 ***	0.287 ***	0.246 ***
	(0.098)	(0.076)	(0.072)
ProfitMargin	0.015 **	0.009 *	0.008 *
	(0.006)	(0.005)	(0.004)
FirmAge	−0.004	−0.003	−0.002
	(0.003)	(0.002)	(0.002)
NEVFocus	0.018 ***	0.012 **	0.010 **
	(0.005)	(0.004)	(0.004)
ROA	0.021 **	0.014 *	0.012 *
	(0.008)	(0.007)	(0.006)
GR	0.003	0.002	0.002
	(0.002)	(0.002)	(0.002)
PolicyIntensity × ROA	0.001 *	\	0.0006
	(0.0006)	\	(0.0004)
PolicyIntensity × GR	0.0002	\	0.0001
	(0.0002)	\	(0.0001)
Constant	−3.876 ***	−2.543 ***	−2.187 ***
	(0.987)	(0.765)	(0.712)
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
R2	0.683	0.712	0.745

Note: t-statistics are reported in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

displays the direct effects of the Dual-Credit Policy on innovation capability and R&D investment. The coefficient for the policy variable is positive and significant, indicating that the Dual-Credit Policy has a strong positive impact on innovation capability. Specifically, the coefficients for Policy (0.287, p < 0.01) and Policy × Treat (0.318, p < 0.01) suggest that the policy significantly enhances the innovation output of NEV manufacturers relative to traditional manufacturers.

The mediation analysis follows a three-step approach. First, the effect of the policy on R&D investment is examined. The coefficient for Policy × Treat in column (2) of Table 5 is positive and significant (0.009, p < 0.1), demonstrating that the Dual-Credit Policy positively influences R&D investment. This meets the first condition for mediation, where the independent variable (policy) must significantly affect the mediator (R&D investment).

Next, the effect of R&D investment on innovation capability is analyzed. Column (3) of Table 5 shows that R&D investment (log (RD)) positively and significantly impacts innovation capability (0.534, p < 0.01). This fulfills the second condition for mediation, confirming that the mediator (R&D investment) significantly affects the dependent variable (innovation capability).

Finally, when both the policy and R&D investment are included in the model (column (3) of Table 5), the coefficient for Policy × Treat remains significant but decreases in magnitude (0.187, p < 0.01), indicating a partial mediation effect. This suggests that R&D investment partially mediates the relationship between the Dual-Credit Policy and innovation capability.

In

Table 7. Stepwise regression results of the mediating effect of DID and PSM-DID.

Variable	IC	RD	IC	IC	IC	RD	IC	IC
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Methods	DID	DID	DID	DID	PSM-DID	PSM-DID	PSM-DID	PSM-DID
Policy	0.312 ***	0.287 ***	\	0.156 **	0.303 **	0.279 ***	\	0.148 **
	(0.084)	(0.084)	\	(0.062)	(0.087)	(0.086)	\	(0.064)
Treat	0.528 ***	0.542 ***	\	0.312 ***	0.513 ***	0.528 ***	\	0.302 ***
	(0.127)	(0.127)	\	(0.094)	(0.131)	(0.130)	\	(0.097)
Policy × Treat	0.345 ***	0.318 ***	\	0.187 ***	0.335 ***	0.309 ***	\	0.180 ***
	(0.093)	(0.093)	\	(0.069)	(0.096)	(0.095)	\	(0.071)
log(RD)	\	\	0.534 ***	0.428 ***	\	\	0.521 ***	0.416 ***
	\	\	(0.087)	(0.079)	\	\	(0.089)	(0.081)
log(RD_t-1)	\	\	0.312 ***	0.254 ***	\	\	0.304 ***	0.247 ***
	\	\	(0.076)	(0.071)	\	\	(0.078)	(0.073)
log(RD_t-2)	\	\	0.187 **	0.143 **	\	\	0.182 **	0.139 **
	\	\	(0.065)	(0.061)	\	\	(0.067)	(0.063)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.512	0.683	0.712	0.745	0.503	0.674	0.703	0.736

Note: **, and *** represent significance at the 5%, and 1% levels, respectively. The value in brackets + statistical value.

, the coefficients for Policy and Policy × Treat are significant across all specifications, indicating that both the implementation of the Dual-Credit Policy and the interaction with the treatment group (NEV manufacturers) positively influence innovation capability (IC). In models (1) and (4), the policy variables significantly affect IC, with coefficients of 0.312 and 0.303, respectively, suggesting a robust positive impact of the policy on innovation capability.

Model (2) shows that the Dual-Credit Policy significantly increases R&D investment (coefficient = 0.287, p < 0.01), demonstrating that the policy encourages firms to invest more in R&D activities. This is further supported by model (6) under the PSM-DID method, where the coefficient for Policy is 0.279 (p < 0.01), indicating a similar positive effect on R&D investment.

When R&D investment is included in the models predicting IC (models (3), (4), (7), and (8)), the coefficients for log (RD) and its lags are significant, confirming that higher R&D investment leads to greater innovation capability. Specifically, in model (3), the coefficient for log (RD) is 0.534 (p < 0.01), and in model (7), it is 0.521 (p < 0.01), highlighting the substantial role of R&D in driving innovation.

The interaction term Policy × Treat remains significant in all models, suggesting that the Dual-Credit Policy’s effect on innovation capability is stronger for NEV manufacturers compared to traditional manufacturers. This interaction is positive and significant in models (1), (3), (4), (5), and (8), with coefficients ranging from 0.180 to 0.345, indicating a consistently positive differential impact of the policy on NEV manufacturers.

The R² values across the models indicate that a substantial proportion of the variance in innovation capability is explained by the included variables. For example, in model (1), the R² value is 0.512, meaning that 51.2% of the variance in innovation capability is explained by the model. This increases to 0.745 in model (4) when additional controls are included, showing that the model explains 74.5% of the variance in innovation capability. Similar patterns are observed in the PSM-DID models, with R² values indicating strong explanatory power.

To confirm the mediation effect, the Sobel test and bootstrapping methods are employed [31].

Table 8. DID and PSM—DID Sobel test results.

Method	Sobel Test	Bootstrapping	Proportion of Mediating Effect	Control Variable	Firm Fixed Effect	Year Fixed Effect
Index	Indirect Effect (a × b)	Indirect Effect (a × b)	Unit: %	Yes or no
	Sobel Test Statistic (z)	95% Confidence Interval
	p-value	Significance
DID	0.136 ***	0.136 ***	74.5	Yes	Yes	Yes
	3.092	[0.0502, 0.2218]
	0.00198	Significance
PSM-DID	0.129 ***	0.129 ***	73.6	Yes	Yes	Yes
	2.957	[0.0443, 0.2137]
	0.00311	Significance

Note: *** represent significance at the 1% levels, respectively. The value in brackets + statistical value.

presents the results of these tests, confirming the significance of the mediation effect.

The Sobel test results indicate that the indirect effect of the policy on innovation capability through R&D investment is significant for both DID and PSM-DID methods. The Sobel test statistic for DID is 3.092 (p = 0.00198), and for PSM-DID, it is 2.957 (p = 0.00311), both showing high significance levels. The bootstrapping results further confirm the significance of the mediation effect, with the 95% confidence intervals for the indirect effect not including zero. This indicates that R&D investment significantly mediates the relationship between the Dual-Credit Policy and innovation capability.

4.4. Heterogeneity Analysis Results

Table 9. Heterogeneity analysis: whether located in a developed region.

Variable	DID (Area = 1)	DID (Area = 0)	PSM-DID (Area = 1)	PSM-DID (Area = 0)
Policy × Treat	423.67 **	215.33	411.22 **	209.45
	(184.29)	(178.56)	(179.13)	(173.62)
p-value	0.022	0.228	0.022	0.228
Constant	578.33 **	312.67	562.11 **	304.29
	(298.45)	(256.78)	(290.22)	(249.59)
p-value	0.044	0.224	0.044	0.223
Control Variables	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
R2	0.5123	0.4287	0.5082	0.4169

Note: ** p < 0.05.

presents the heterogeneity analysis results based on whether the firms are located in developed regions. The coefficient for Policy × Treat is positive and significant for firms located in developed regions (423.67, p < 0.05), indicating a strong positive impact of the Dual-Credit Policy on innovation capability for these firms. Conversely, the coefficient for firms in less-developed regions is positive but not statistically significant (215.33, p = 0.228).

This suggests that the Dual-Credit Policy is more effective in promoting innovation among firms in developed regions. The advanced infrastructure, better access to resources, and supportive innovation ecosystems in developed regions likely enhance the policy’s impact. In contrast, firms in less-developed regions may face structural challenges that limit the effectiveness of the policy.

Table 10. Heterogeneity analysis: whether ESG.

Variable	DID (ESG = 1)	DID (ESG = 0)	PSM-DID (ESG = 1)	PSM-DID (ESG = 0)
Policy × Treat	415.75 **	223.25	403.28 **	216.55
	(182.64)	(179.87)	(177.16)	(174.47)
p-value	0.023	0.215	0.023	0.215
_cons	592.50 *	298.50	574.73 *	289.55
	(302.18)	(253.05)	(293.11)	(245.46)
p-value	0.051	0.239	0.051	0.239
Control Variables	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
R2	0.5235	0.4175	0.5078	0.4050

Note: * p < 0.1, ** p < 0.05.

analyzes the heterogeneity of the policy impact based on firms’ ESG performance. For firms with higher ESG ratings, the Policy × Treat coefficient is positive and significant (415.75, p < 0.05), indicating that the Dual-Credit Policy significantly boosts innovation capability for these firms. However, for firms with lower ESG ratings, the coefficient is positive but not significant (223.25, p = 0.215).

The significant impact for high ESG firms may be attributed to their better management practices, higher transparency, and stronger commitment to sustainable practices, which align well with the goals of the Dual-Credit Policy. These firms are potentially more adept at leveraging policy incentives to enhance their innovation efforts. Firms with lower ESG ratings might not fully capitalize on the policy benefits due to weaker organizational capabilities and less alignment with sustainability goals.

Table 11. Heterogeneity analysis: whether SOE.

Variable	DID (SOE = 1)	DID (SOE = 0)	PSM-DID (SOE = 1)	PSM-DID (SOE = 0)
Policy × Treat	357.50 *	281.50	346.78 *	273.06
	(199.23)	(192.37)	(193.25)	(186.60)
p-value	0.074	0.144	0.074	0.144
_cons	521.75 *	369.25	506.10 *	358.17
	(285.61)	(265.18)	(277.04)	(257.22)
p-value	0.068	0.165	0.068	0.165
Control Variables	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
R2	0.4912	0.4498	0.4765	0.4363

Note: * p < 0.1.

presents the results based on whether the firms are state-owned enterprises (SOEs). The Policy × Treat coefficient is positive and significant for SOEs (357.50, p < 0.1), suggesting that the Dual-Credit Policy effectively enhances innovation capability among SOEs. For non-SOEs, the coefficient is also positive but not significant (281.50, p = 0.144).

SOEs may benefit more from the Dual-Credit Policy due to their larger scale, better access to government resources, and greater strategic alignment with national policy objectives. These advantages likely enable SOEs to respond more effectively to policy incentives. Non-SOEs, while potentially more flexible and market-driven, may lack the same level of access to resources and support, which could diminish the policy’s impact on their innovation activities.

5. Dynamic Effect Analysis

This section presents the dynamic effects of the Dual-Credit Policy on the innovation capability of NEV manufacturers over time, as illustrated by the results in

Table 12. Dynamic effect analysis results.

Year	Coefficient	Standard Error	p-Value
2018	0.089 *	0.047	0.058
2019	0.152 **	0.061	0.013
2020	0.231 ***	0.072	0.001
2021	0.305 ***	0.083	0.000
2022	0.389 ***	0.095	0.000
Control Variables	Yes
Firm FE	Yes
Year FE	Yes
R2	0.7124

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

. The analysis covers the period from 2018 to 2022, capturing the longitudinal impact of the policy implementation. For the dynamic effect analysis, our model to incorporate these variables is as follows:

log (Y_it) = β₀ + β₁Treat_i + β₂PolicyIntensity_t + β₃PolicyExposure_i + Σ_t γ_t(Treat_i × PolicyIntensity_t) + δX_it + α_i + ε_it

(6)

where

X_it includes control variables like log (RD_it), ITR_it, log (EmployeeCount_it), ProfitMargin_it, FirmAge_it, NEVFocus_i, ROA_it, and GR_it.

The dynamic effect analysis reveals a progressively increasing impact of the Dual-Credit Policy on innovation capability over the analyzed period. In 2018, the first year of policy implementation, the coefficient (γ_t) is 0.089, with a p-value of 0.058, indicating a marginally significant positive effect at the 10% level. This initial impact suggests that the policy began to influence firms’ innovation activities shortly after its introduction.

In 2019, the effect strengthens, with a coefficient of 0.152 and a p-value of 0.013, significant at the 5% level. This indicates that as firms began to adapt to the policy requirements, their innovation capabilities improved more noticeably. The increasing significance and magnitude of the coefficient highlight the policy’s growing influence.

The trend continues in 2020, where the coefficient rises to 0.231 with a p-value of 0.001, reflecting a highly significant impact at the 1% level. This substantial increase suggests that the policy mechanisms were well integrated into firms’ operational strategies, leading to enhanced innovation outputs.

By 2021, the coefficient further increases to 0.305, maintaining its significance at the 1% level (p-value = 0.000). The continued upward trajectory indicates that the policy’s effects were becoming more pronounced, possibly due to cumulative investments in R&D and improved regulatory compliance.

In the final year of the analysis, 2022, the coefficient reaches 0.389, with a p-value of 0.000, reaffirming the highly significant and robust impact of the Dual-Credit Policy on innovation capability. The consistent year-over-year growth underscores the effectiveness of the policy in fostering a sustained innovation environment within the NEV sector.

The R² value of 0.7124 indicates that approximately 71.24% of the variance in innovation capability is explained by the model, demonstrating a high level of explanatory power. This suggests that the model is well-specified and that the included variables effectively capture the dynamics influencing innovation capability.

6. Threshold Effect Analysis

This section examines the threshold effects of the Dual-Credit Policy intensity on the innovation output of NEV manufacturers using a panel threshold regression model, as illustrated by the results in

Table 13. Threshold effect analysis.

Regime	Coefficient	Std. Error	t-Statistic	p-Value
PolicyIntensity ≤ 18.5	0.103 **	0.042	2.405	0.015
Policyintensity > 18.5	0.189 ***	0.043	3.714	0.000
Threshold Variable	PolicyIntensity
Threshold Estimate	18.5
95% Confidence Interval	[16.2, 20.2]

Note: ** p < 0.05, *** p < 0.01; F-statistic: 15.27; p-value: 0.002.

. The model is specified as follows:

log (Y_it) = μ_i + β′₁x_it I (q_it ≤ γ) + β′₂x_it I (q_it > γ) + e_it

(7)

where

• Y_it is the innovation output.
• x_it is a vector of explanatory variables.
• q_it is the threshold variable (PolicyIntensity).
• γ is the threshold value.
• e_it is the indicator function.

The threshold effect analysis reveals a significant non-linear impact of policy intensity on innovation output, with a distinct threshold value identified at 18.5. The results indicate two different regimes based on whether the policy intensity is below or above this threshold.

(a)

Regime 1: PolicyIntensity ≤ 18.5

In the first regime, where the policy intensity is less than or equal to 18.5, the coefficient for PolicyIntensity is 0.103, significant at the 5% level (p = 0.015). This positive and significant coefficient suggests that in this regime, moderate policy intensity effectively enhances innovation output, but the impact is relatively modest.

(b)

Regime 2: PolicyIntensity > 18.5

In the second regime, where the policy intensity exceeds 18.5, the coefficient for PolicyIntensity increases substantially to 0.189, significant at the 1% level (p = 0.000). This indicates that higher levels of policy intensity have a much stronger positive effect on innovation output. The marked increase in the coefficient in this regime highlights the enhanced responsiveness of innovation activities to more intensive policy measures.

The threshold estimates of 18.5, with a 95% confidence interval ranging from 16.2 to 20.2, indicates a critical level of policy intensity beyond which the effects on innovation output are significantly amplified. The F-statistic for the threshold effect test is 15.27, with a corresponding p-value of 0.002, confirming the presence of a statistically significant threshold effect.

7. GMM Analysis (As an Additional Robustness Check and Methodological Extension)

7.1. Combined Time Series Model (Dynamic Panel GMM)

To capture the dynamic effects and control for potential endogeneity, we estimate the following model:

Δlog(Y_it) = β₁Δlog(Y_i,t-1) + β₂ΔPolicy_t + β₃ΔPolicyIntensity_t + β₄Δ(Policy_t × Treat_i) + β₅Δ(PolicyIntensity_t × Treat_i) + β₆Δlog(RD_it) + ΔδX_it + Δε_it

(8)

where

Y_it represents our measure of innovation capability, X_it includes control variables, and Δ denotes first differences to eliminate individual effects. The results are presented in

Table 14. Combined time series analysis.

Variable	Coefficient	Std. Error	z-Statistic	p-Value
log (Yi,t1)	0.312 ***	0.073	4.274	0.000
Policyt	0.178 **	0.069	2.580	0.010
PolicyIntensityt	0.026 ***	0.008	3.250	0.001
Policyt × Treati	0.203 ***	0.061	3.328	0.001
PolicyIntensityt × Treati	0.035 ***	0.011	3.182	0.001
log (RDit)	0.295 ***	0.068	4.338	0.000

Note: ** p < 0.05, *** p < 0.01.

.

The results from the GMM estimation provide robust evidence supporting the dynamic relationship between policy measures and innovation capability. The lagged dependent variable log (Y_i, t-1) is highly significant (coefficient = 0.312, p < 0.01), justifying the use of a dynamic model.

The coefficients for Policy_t (0.178, p < 0.05) and PolicyIntensity_t (0.026, p < 0.01) are both positive and significant, indicating that the Dual-Credit Policy and its intensity significantly enhance innovation capability. The interaction terms, Policy_t × Treat_i (0.203, p < 0.01) and PolicyIntensity_t × Treat_i (0.035, p < 0.01), are also significant, suggesting that the policy has a stronger effect on treated firms, which are likely NEV manufacturers.

R&D investment (log (RD_it)) remains a significant positive determinant of innovation (coefficient = 0.295, p < 0.01), reinforcing the critical role of R&D in fostering innovation.

7.2. Post-Estimation Tests

To ensure the validity of the GMM estimation, we conducted several post-estimation tests:

Arellano-Bond Test for AR (1) in First Differences: The z-statistic is −3.42 with a p-value of 0.001, indicating first-order autocorrelation in the differenced errors, which is expected.

Arellano-Bond Test for AR (2) in First Differences: The z-statistic is −1.08 with a p-value of 0.280, suggesting no second-order autocorrelation. The absence of second-order autocorrelation is necessary for the consistency of the GMM estimator.

Sargan Test of Overidentifying Restrictions: The chi-square statistic is 52.36 with a p-value of 0.273. This test does not reject the null hypothesis of valid overidentifying restrictions, indicating that the instruments used in the GMM estimation are appropriate and valid. To facilitate comparison, we present the key coefficients from our baseline DID, PSM-DID, and GMM models in

Table 15. Comparison of DID, PSM-DID, and GMM Results.

Variable	DID	PSM-DID	GMM
Policyt	0.265 **	0.259 **	0.178 **
	(0.122)	(0.110)	(0.069)
Policyt × Treati	0.295 ***	0.289 ***	0.203 ***
	(0.087)	(0.085)	(0.061)
log(RDit)	0.272 ***	0.265 ***	0.295 ***
	(0.069)	(0.063)	(0.068)

Note: ** p < 0.05, *** p < 0.01.

.

The comparison across DID, PSM-DID, and GMM models reveals several key insights:

The Policy Effect (Policy_t): The coefficient for Policy_t is positive and statistically significant across all three models, confirming that the Dual-Credit Policy has a positive effect on innovation capability. However, the GMM estimate (0.178, p < 0.05) is slightly lower than the DID (0.265, p < 0.05) and PSM-DID (0.259, p < 0.05) estimates. This reduction in magnitude suggests that the baseline models may overestimate the policy effect due to potential endogeneity issues, which the GMM approach addresses more rigorously.

Interaction Effect (Policy_t × Treat_i): The interaction term between the policy and the treatment group is also consistently positive and highly significant in all models. The GMM estimate (0.203, p < 0.01) is slightly lower than the DID (0.295, p < 0.01) and PSM-DID (0.289, p < 0.01) estimates, indicating a robust effect of the policy on the treated firms, albeit with some attenuation when endogeneity is controlled for.

R&D Investment (log (RD_it)): The effect of R&D investment on innovation capability is consistently positive and significant across all models. The GMM estimate (0.295, p < 0.01) is slightly higher than the DID (0.272, p < 0.01) and PSM-DID (0.265, p < 0.01) estimates, reinforcing the critical role of R&D investment in enhancing innovation capability.

8. Robustness Test

8.1. Parallel Trend Test

The parallel trend test examines whether the treatment and control groups have similar trends in the outcome variable before the policy implementation. Figure 4 presents the results of the parallel trend test. The graph plots the coefficients and confidence intervals of the interaction terms between the treatment dummy (experimental group) and the year dummies. The coefficients before the policy implementation year (2018) are not statistically significant, as indicated by the confidence intervals crossing the zero line. This suggests that the innovation capability of NEV passenger vehicle manufacturers (experimental group) and traditional passenger vehicle manufacturers (control group) followed similar trends prior to the Dual-Credit Policy, supporting the parallel trend assumption. In addition, Figure 5 shows the time trend of innovation capability (measured by the average innovation capacity) for the experimental group and the control group from 2013 to 2022. The two groups show similar trends before the implementation of the Dual-Credit Policy in 2018. After 2018, the innovation capability of NEV passenger vehicle manufacturers (experimental group) increases at a faster rate compared to traditional passenger vehicle manufacturers (control group), consistent with our main findings.

8.2. Placebo Test Results

The placebo test is a robustness check that further validates the main findings of the study by artificially assigning the policy intervention to a different year. The purpose of this test is to ensure that the observed treatment effect is indeed caused by the policy implementation and not by random chance or unobserved confounding factors.

In this study, the placebo test is conducted by randomly assigning the Dual-Credit Policy implementation year to a year other than the actual implementation year of 2018. The difference-in-differences (DID) model is then re-estimated using the placebo policy year. This process is repeated multiple times to create a distribution of placebo coefficients, which represents the range of estimated treatment effects under the assumption that the policy was implemented in a different year.

Figure 6 presents the distribution of the estimated coefficients from the placebo tests, along with the actual estimated coefficient from the main model, represented by the red dashed line. The placebo coefficients are centered around zero, indicating that when the policy is artificially assigned to a different year, the estimated treatment effect is close to zero on average. This suggests that the observed treatment effect in the main model is unlikely to be driven by random chance or unobserved confounding factors.

Moreover, the actual estimated coefficient from the main model lies in the far-right tail of the placebo coefficient distribution, far from the mass of placebo estimates. This positioning of the actual coefficient provides strong evidence that the observed treatment effect is statistically significant and robust to the placebo test. If the actual coefficient were to fall within the range of placebo coefficients, it would cast doubt on the validity of the main findings, as it would suggest that the observed effect could be attributed to factors other than the policy intervention.

Figure 7 provides a combined visualization of the actual coefficient estimate and the distribution of placebo coefficients. The histogram represents the distribution of placebo coefficients, while the blue line represents the kernel density curve of the placebo distribution. The red dashed line indicates the actual coefficient estimate from the main model.

The kernel density curve is a non-parametric estimation of the probability density function of the placebo coefficients. It provides a smooth approximation of the distribution, allowing for easier visualization and interpretation of the results. The fact that the actual coefficient lies far in the right tail of both the histogram and the kernel density curve further reinforces the robustness of the main findings.

9. Discussion, Implications, Limitations and Future Research

9.1. Discussion

The study’s results consistently demonstrate that the Dual-Credit Policy significantly enhances innovation capability among NEV manufacturers. This conclusion is supported by various analytical methods, each reinforcing the robustness and reliability of the findings. The inclusion of firm fixed effects and year fixed effects across all models ensures that the estimated effects are not confounded by unobserved heterogeneity or time-specific shocks.

The results from the DID and PSM-DID analyses indicate a significant positive effect of the policy on innovation capability, as evidenced by the significant coefficients for Policy_t and Policy_t × Treat_i. The GMM results, which address potential endogeneity issues, confirm these findings but with slightly lower coefficient magnitudes. This suggests that while the policy’s impact is robust, baseline models may slightly overestimate the effect due to endogeneity. The role of R&D investment is particularly noteworthy. Across all models, the coefficients for log (RD_it) are positive and significant, underscoring the crucial role of R&D investment in driving innovation. This finding highlights the importance of continued and enhanced R&D activities to sustain and amplify the innovation gains facilitated by the policy.

The dynamic effect analysis, capturing the temporal impact of the Dual-Credit Policy, reveals an increasing trend in innovation capability from 2018 to 2022. The progressively increasing coefficients for Policy_t over the years suggest that the policy’s impact intensifies as firms adapt and integrate the policy measures into their operational strategies. This dynamic aspect underscores the policy’s sustained and growing influence on innovation over time.

The heterogeneity analysis provides deeper insights into how the policy’s impact varies across different firm characteristics. Firms located in developed regions, those with higher ESG ratings, and state-owned enterprises (SOEs) exhibit a more pronounced response to the policy. This variation highlights the importance of contextual factors in shaping the effectiveness of policy interventions. Firms in developed regions and those with strong ESG practices are better positioned to leverage the policy incentives, leading to greater innovation outputs. Similarly, SOEs benefit more from the policy due to their larger scale and better access to resources.

9.2. Implications

The findings of this study have several significant implications for policymakers, industry stakeholders, and researchers focused on innovation in the NEV sector and beyond. The results indicate that higher policy intensity has a more substantial impact on innovation capability. Policymakers should consider increasing the intensity of support measures within the Dual-Credit Policy framework to amplify its effectiveness. This could involve increasing the credits awarded for innovations, enhancing penalties for non-compliance, or providing additional financial incentives for R&D activities. Given the pivotal role of R&D investment in driving innovation, policies should prioritize and facilitate greater investment in R&D. This can be achieved through tax incentives, grants, and subsidies specifically aimed at R&D projects. Additionally, establishing public–private partnerships could help leverage private sector innovation capabilities with public sector resources.

The heterogeneity analysis underscores the differential impact of the policy based on firm characteristics such as location, ESG performance, and ownership type. Policymakers should tailor interventions to the specific needs and conditions of different firms. For instance, firms in less-developed regions might benefit from additional infrastructural support and capacity-building initiatives, while SOEs could be encouraged to engage more with innovative startups and private firms to foster a more dynamic innovation ecosystem.

Firms should recognize the significant benefits of investing in R&D, as evidenced by the positive impact on innovation capability. Strategic allocation of resources towards R&D can yield substantial long-term benefits. Firms should also consider collaborative R&D efforts to share costs and risks while enhancing innovation potential. The positive impact of ESG performance on the effectiveness of the Dual-Credit Policy suggests that firms with strong ESG practices are better positioned to leverage policy incentives. Firms should integrate ESG considerations into their business strategies, not only for compliance but also to enhance their innovation capabilities and competitiveness.

Furthermore, firms, particularly NEV manufacturers, should actively engage with and leverage policy measures to enhance their innovation outputs. This involves staying informed about policy changes, actively participating in policy dialogues, and aligning business strategies with policy objectives to maximize benefits.

The study highlights the importance of addressing endogeneity issues in policy evaluation. Researchers should employ robust methodological approaches, such as GMM, to obtain accurate estimates of policy impacts. This enhances the reliability of the findings and provides a clearer understanding of the causal relationships. Future research should focus on longitudinal studies to capture the dynamic effects of policies over time. Such studies can provide deeper insights into how policies evolve and their long-term impacts on innovation and other economic outcomes. Comparative studies across different regions, industries, and policy frameworks can provide valuable insights into best practices and the contextual factors that influence policy effectiveness. This can inform the design of more effective and context-sensitive policies.

9.3. Limitations and Future Research

Despite the robust findings of this study, several limitations should be acknowledged. First, the data used in the analysis are primarily derived from Chinese NEV manufacturers, which may limit the generalizability of the results to other countries or industries. Different regulatory environments, market dynamics, and levels of technological advancement could lead to varying impacts of similar policies in other contexts. Future research should aim to replicate this study in different geographical and industrial settings to validate the findings. Another limitation is the potential for measurement errors in the variables used, particularly in self-reported data such as R&D investment and innovation outputs. While efforts were made to ensure data accuracy and reliability, inherent biases in self-reported measures could affect the results. Additionally, the study primarily uses quantitative methods to evaluate the policy impact, which, although powerful, may overlook qualitative aspects such as managerial perceptions, firm strategies, and external factors influencing innovation that are not easily quantifiable.

It is important to consider how an “output” measure might perform better than an “input” measure in answering the research question. Output measures, such as the number of patents granted or new products developed, directly reflect the tangible results of innovation efforts. These measures provide a clearer indication of the policy’s effectiveness in generating innovation, as they capture the actual marketable innovations rather than just the resources allocated to R&D activities. On the other hand, input measures, like R&D spending, indicate the commitment of resources but do not guarantee successful innovation outcomes. Future research could benefit from incorporating more comprehensive output measures to assess the direct impact of policies on innovation achievements. Moreover, the issue of time lag between “input” and “output” remains a significant limitation. While the policy might increase the inputs of innovation, such as R&D investments (direct effect), it does not necessarily lead to an immediate increase in innovation outputs. This time lag means that the effects of increased R&D spending may not be observable within the short timeframe of this study. Other factors, such as market conditions, technological readiness, and organizational capabilities, also play crucial roles in translating R&D inputs into successful innovation outputs. To tackle this issue, future research should extend the observation period and employ longitudinal data to capture the delayed effects of R&D investments on innovation outputs more accurately.

The dynamic nature of policy impacts, as evidenced by the increasing effects over time, suggests that a longer observation period could provide a more comprehensive understanding of the policy’s long-term effects. The current study spans five years, but future research could benefit from extending this period to capture more extended trends and potential lagged effects of policy changes.

While the GMM approach helps address endogeneity issues, it is not without its limitations. The choice of instruments in GMM is critical, and inappropriate instruments can lead to biased estimates. Although the Sargan test indicates that the instruments used in this study are valid, the results should still be interpreted with caution. Further research could explore alternative instruments or employ different methodological approaches to confirm the robustness of the findings.

Lastly, the study’s focus on the Dual-Credit Policy means that other concurrent policies and external factors influencing innovation in the NEV sector may not be accounted for. The NEV industry in China has been subject to various supportive measures and rapid technological advancements, which might confound the observed effects of the Dual-Credit Policy. Future studies should consider a more holistic approach, incorporating a broader range of policies and external influences to provide a more nuanced understanding of the drivers of innovation in this sector.