Method for evaluating innovation efficiency of automobile manufacturing enterprises in different stages based on industrial chain

Abstract

The application discloses an evaluation method for staged innovation efficiency of automobile manufacturing enterprises based on an industrial chain, relates to the technical field of industrial technology innovation evaluation, and comprises the following steps: S1, collecting innovation input, technical output, operating benefit, environmental performance and policy attribute data of a target sample enterprise, and marking core component manufacturing, passenger vehicle manufacturing and commercial vehicle manufacturing three types of industrial chain role labels according to the main business scope and industrial chain orientation of the enterprise; the application is adapted to the innovation resource allocation characteristics of enterprises at different industrial chain links through the industrial chain role label marking and weight dynamic adjustment mechanism, improves the accuracy of two-stage efficiency calculation, makes up for the systematic evaluation deviation caused by the fixed weight, corrects the calculation defects of a single output dimension through a result transformation index system coupling economic and environmental benefits, and comprehensively reflects the comprehensive value of green technology innovation.

Description

Technical Field

[0001] This invention relates to the field of industrial technology innovation assessment technology, specifically a method for assessing the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain. Background Technology

[0002] Automobile manufacturing enterprises are a crucial pillar industry of the economy, covering the entire chain from core component R&D and production to vehicle design, integration, assembly, and after-sales services. Their technological innovation capabilities directly determine the industry's core competitiveness and green transformation progress, playing an irreplaceable supporting role in implementing the dual-carbon development strategy and promoting high-quality development of the manufacturing sector. Phased innovation efficiency assessment breaks down the enterprise's green technology innovation process into two interconnected stages: R&D input-output and value-added transformation. It then calculates the resource utilization efficiency and value transformation capability of each stage. Compared to a single overall efficiency assessment, phased assessment can accurately pinpoint efficiency loss points throughout the innovation chain, providing quantitative evidence for enterprises to optimize resource allocation and for governments to formulate precise industrial policies.

[0003] However, existing innovation efficiency assessments for automobile manufacturers mostly focus on passenger vehicle manufacturers, failing to fully cover key links in the industrial chain such as core components and commercial vehicles, and thus failing to reflect the spillover effects of environmental policies on the industrial chain. Two-stage efficiency calculations generally employ fixed weight thresholds and a single calculation order, which cannot adapt to the innovation resource allocation characteristics of different roles in the industrial chain, easily leading to systematic biases in efficiency calculation results. Furthermore, existing assessment methods fail to couple and quantify green environmental benefits with economic benefits, and lack a diagnostic mechanism linking policy effects and efficiency bottlenecks, making it difficult to support differentiated policy optimization and precise innovation decisions by enterprises. Therefore, developing a phased innovation efficiency assessment method for automobile manufacturers based on the industrial chain is of great significance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a phased innovation efficiency evaluation method for automobile manufacturing enterprises based on the industrial chain. It can solve the problems of incomplete industrial chain coverage, rigid weight setting, single benefit dimension, and lack of policy correlation diagnosis in existing evaluation methods, and achieve innovation efficiency evaluation with full industrial chain coverage, adaptive weight adjustment, multi-benefit coupling calculation and accurate diagnosis of policy effects.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a phased innovation efficiency evaluation method for automobile manufacturing enterprises based on the industry chain, comprising the following steps:

[0006] S1. Collect data on innovation input, technological output, operating efficiency, environmental performance, and policy attributes of target sample enterprises. Based on the enterprise's main business scope and industrial chain positioning, label three types of industrial chain roles: core component manufacturing, passenger vehicle manufacturing, and commercial vehicle manufacturing.

[0007] S2. Construct a two-stage green technology innovation efficiency evaluation index system that includes initial input, intermediate output, additional intermediate input and final output. Based on the enterprise's industrial chain role label and input structure characteristics, dynamically match the minimum weight threshold of the R&D stage and the achievement transformation stage. Use a two-stage network DEA model with weight constraints to solve the sub-stage efficiency and decompose it to obtain the overall green technology innovation efficiency, R&D efficiency and achievement transformation efficiency.

[0008] S3. The implementation of the dual-credit policy is used as a quasi-natural experiment. Automobile manufacturing enterprises are used as the treatment group and other advanced manufacturing enterprises are used as the control group. Enterprise characteristic variables are selected as matching variables to implement propensity score matching (PSM). Based on the matched samples, a two-way fixed-effect difference-in-differences (DID) model is constructed. The PSM-DID evaluation framework is formed by combining propensity score matching and the difference-in-differences model to measure the net impact of the dual-credit policy on three types of efficiency indicators.

[0009] S4. Group regressions are performed according to the role in the industrial chain, enterprise ownership, region and initial efficiency level to identify the multidimensional heterogeneity of policy effects, construct a mediation effect test model to identify policy transmission paths, and measure the spillover intensity of policies between upstream and downstream links in the industrial chain.

[0010] S5. Based on the two-stage efficiency measurement results, identify the efficiency bottleneck links in the entire innovation chain of enterprises, and combine the heterogeneity analysis conclusions to generate innovation resource optimization directions and industrial policy adjustment suggestions for different types of enterprises.

[0011] Furthermore, the step S2, which dynamically matches the minimum weight threshold between the R&D stage and the results transformation stage, includes the following steps:

[0012] We conducted cluster analysis on the input structure of sample enterprises in different roles in the industrial chain, and statistically analyzed the distribution characteristics of the proportion of R&D investment to total innovation investment for enterprises in each role.

[0013] Based on the proportional distribution characteristics obtained from clustering, adaptive threshold calculation formulas are used to set corresponding minimum weight thresholds for the R&D stage and the achievement transformation stage for core component manufacturers, passenger vehicle manufacturers, and commercial vehicle manufacturers. The adaptive threshold calculation formula is as follows: ,in For the first Minimum weight threshold for the R&D stage corresponding to a role in the industry chain. The preset global minimum weight benchmark value, For the first The average R&D investment as a percentage of total innovation investment for sample companies in the industry chain. For the first Information entropy of the input structure of roles in the industry chain For all industry chain role categories, the global minimum weight benchmark value is 0.3. The average R&D investment ratio is obtained from the investment data of sample enterprises in the corresponding category. The investment structure information entropy is calculated through the proportion distribution of various investment indicators. The higher the R&D investment ratio and investment structure entropy value, the lower the minimum weight threshold of the corresponding category in the R&D stage. The minimum weight threshold of the achievement transformation stage is obtained by subtracting the minimum weight threshold of the R&D stage from 1.

[0014] The two minimum weight thresholds that have been set are substituted into the two-stage DEA network model with weight constraints, and used as weight constraints in the efficiency solution calculation.

[0015] Furthermore, step S3, when performing propensity score matching (PSM), includes the following steps:

[0016] The liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale were selected as matching variables, and the propensity score of each sample enterprise was calculated by Logit regression.

[0017] Within the common support domain, a nearest neighbor matching method with calipers was used to match enterprises in the treatment group and the control group one-to-one, with the caliper width set to one-quarter of the propensity score standard deviation.

[0018] A balance test was performed on the matched samples to verify that there were no significant differences in the matched variables between the treatment group and the control group after matching.

[0019] Furthermore, step S4, when calculating the spillover intensity of the policy between upstream and downstream links of the industrial chain, includes the following steps:

[0020] The samples of core component manufacturers and passenger vehicle manufacturers were grouped and regressed separately to obtain the policy effect regression coefficients for the two types of enterprises.

[0021] Retrieve upstream and downstream industry correlation data to calculate the industry correlation weight and technology correlation coefficient between the core component manufacturing link and the passenger vehicle manufacturing link;

[0022] The spillover intensity from passenger vehicle manufacturing to core component manufacturing under the dual-credit policy is calculated using the industrial chain technology spillover intensity calculation formula. The industrial chain technology spillover intensity calculation formula is as follows: ,in For downstream links Upstream links The intensity of policy spillover effects The regression coefficient for the policy effect in the downstream sector. The regression coefficient of the policy effect in the upstream segment This represents the direct input-output consumption coefficient between upstream and downstream links. This is the coefficient representing the degree of technological correlation between upstream and downstream links. The input-output direct consumption coefficient is taken from the corresponding industry data in the National Economic Input-Output Table. The technology correlation coefficient is calculated through upstream and downstream patent citation data, and the value range is 0 to 1. When the calculation result is positive, it represents a positive technology spillover effect. When the calculation result is negative, it represents a reverse siphon effect. The spillover effect intensity is taken as the absolute value of the calculation result.

[0023] Furthermore, in step S2, the two-stage green technology innovation efficiency evaluation index system sets two initial inputs in the R&D stage: R&D funds and R&D personnel; one intermediate output: the number of patent applications; and in the achievement transformation stage, the patent achievements corresponding to the intermediate outputs are set as the core inputs, fixed assets, current assets, and total number of employees as three additional intermediate inputs. Finally, it sets two final outputs: operating revenue and comprehensive environmental performance. The comprehensive environmental performance is calculated by coupling three types of data: carbon emission reduction performance, environmental compliance performance, and points trading revenue collected in step S1.

[0024] Furthermore, the two-stage network DEA model with weighted constraints in step S2 is constructed based on the assumption of variable returns to scale. In the efficiency solution process, the optimal solution for the overall green technology innovation efficiency is first obtained, then the efficiency of the R&D stage is maximized according to a preset priority, and finally, the efficiency value of the technology transfer stage is calculated based on the additive weighted decomposition method. The additive weighted decomposition formula is as follows: ,in For the first The overall green technology innovation efficiency of the company. For the first The efficiency of a company's R&D phase. For the first The efficiency of a company's technology transfer process. As a weighting factor in the research and development phase, The stage weights for technology transfer are calculated based on the proportion of investment in each stage to the total investment, and satisfy the following conditions: In the solution process, non-Archimedean infinitesimal constraints are introduced to limit the lower bound of the weights of each index. The overall efficiency calculation result satisfies... .

[0025] Furthermore, in step S3, the two-way fixed effects difference-in-differences (DID) model incorporates individual fixed effects and time fixed effects. The model uses three types of indicators—overall green technology innovation efficiency, R&D efficiency, and achievement transformation efficiency—as dependent variables, respectively. The interaction term between the policy timing variable and the treatment group variable is used as the core explanatory variable. Liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale are used as control variables. The coefficient values ​​of the core explanatory variables are used as the quantitative results of the net policy impact.

[0026] Furthermore, in step S3, after calculating the net impact of the dual-integral policy on the three types of efficiency indicators using the PSM-DID evaluation framework, four robustness test operations are performed. The first is to adjust the priority of sub-stage efficiency measurement, recalculate the efficiency indicators, and repeat the regression operation. The second is to remove the sample data from the year the policy was implemented and then re-execute the matching and regression operation. The third is to set a virtual policy time point and extract samples from the corresponding time period to perform a placebo test. The fourth is to randomly select samples to construct a fictitious treatment group and repeat the regression operation multiple times to verify the distribution characteristics of the results.

[0027] Furthermore, the mediation effect testing model constructed in step S4 sets up two types of mediation variables: R&D investment intensity and patent output scale. In the testing process, the significance of the policy's impact on the mediation variable is first verified, and then the significance of the mediation variable's impact on the corresponding efficiency indicator is verified. The existence and direction of the mediation effect are determined by stepwise regression, and the specific transmission path of the policy's impact on innovation efficiency is clarified.

[0028] Furthermore, in step S5, when identifying efficiency bottlenecks, the industry average for R&D efficiency and commercialization efficiency of all sample companies in the same industry chain role is first calculated. Then, the R&D efficiency and commercialization efficiency of a single company are compared with the corresponding industry averages. The stage where the value is lower than the corresponding industry average is determined as the efficiency bottleneck of the company. Based on the bottleneck attributes, corresponding innovation resource optimization directions are generated. Combined with the differences in policy effects in different dimensions obtained from heterogeneity analysis, suggestions for industrial policy adjustments in the corresponding industry chain role, ownership, and regional dimensions are generated.

[0029] Compared with existing technologies, this phased innovation efficiency evaluation method for automobile manufacturing enterprises based on the industrial chain has the following beneficial effects:

[0030] This invention, through a mechanism of industry chain role labeling and dynamic weight adjustment, adapts to the innovation resource allocation characteristics of enterprises in different links of the industry chain, improves the accuracy of two-stage efficiency measurement, compensates for the systematic evaluation bias caused by fixed weights, corrects the measurement defects of a single output dimension by coupling an indicator system for the transformation of economic and environmental benefits, comprehensively reflects the comprehensive value of green technology innovation, systematically reveals the impact and mechanism of the dual-integral policy on the entire industry chain through a policy evaluation framework that combines biased scoring and dual difference, and realizes the application of evaluation results through efficiency bottleneck diagnosis and policy adaptation modules, providing reliable quantitative support for enterprise innovation decision-making and precise optimization of industrial policies.

[0031] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0032] Figure 1 A flowchart for a phased innovation efficiency assessment method for automobile manufacturing enterprises based on the industrial chain;

[0033] Figure 2 A flowchart outlining the steps of a phased innovation efficiency assessment method for automobile manufacturing enterprises based on the industrial chain;

[0034] Figure 3 This is a flowchart for dynamically matching the minimum weight threshold between the R&D stage and the commercialization stage. Detailed Implementation

[0035] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0036] The phased innovation efficiency evaluation method for automobile manufacturing enterprises based on the industrial chain provided by the present invention starts from the perspective of the entire new energy automobile industrial chain, disassembles the green technology innovation process into two correlated phases: research and development, and achievement transformation, realizes sample stratification and classification through industrial chain role calibration, adopts an adaptive dynamic weight mechanism to optimize the two-stage network data envelopment analysis model, accurately calculates the overall green technology innovation efficiency and phased efficiency of enterprises, then evaluates the net impact effect of the dual-credit policy by combining the combined framework of propensity score matching and difference-in-differences, reveals the policy transmission rule through multi-dimensional heterogeneity analysis and industrial chain spillover intensity calculation, and finally outputs the enterprise innovation optimization direction and industrial policy adjustment suggestions based on efficiency bottleneck identification. This method can adapt to the innovation characteristics of different industrial chain links, correct the evaluation deviation caused by fixed weights and a single output dimension, and provide quantitative support for the green transformation of the automobile industry and the precise formulation of policies. The detailed description is given below in combination with specific embodiments.

[0037] This embodiment is oriented to the dual-credit policy effect evaluation scenario, selects advanced listed manufacturing enterprises over consecutive years as research samples, which is used to quantitatively evaluate the actual impact of the dual-credit policy on the green technology innovation efficiency of automobile manufacturing enterprises, and identify the efficiency differences and policy response characteristics of different links in the industrial chain. As shown Figure 1 , the overall execution process of this embodiment sequentially includes five core links: data collection and role calibration, two-stage efficiency calculation, policy net effect evaluation, heterogeneity and spillover analysis, and bottleneck diagnosis and scheme output. All links are executed in sequence, and the output result of the previous link serves as the input basis for the next link.

[0038] The first step is to perform data collection and industrial chain role calibration operations. The initial samples in this embodiment cover listed enterprises in the manufacturing field. During the screening process, enterprises with abnormal operating status and enterprises with missing core indicator data for more than one consecutive year are eliminated, and a basic sample library containing more than 800 enterprises is finally formed. Sample screening follows unified standards to ensure that the financial data and innovation data of all sample enterprises are comparable, and avoid abnormal samples from interfering with subsequent evaluation results. The collected data dimensions cover five categories: innovation input, technology output, operating benefit, environmental performance and policy attribute. All data are taken from industry-authoritative financial databases and patent databases, and some missing data are supplemented and calibrated through the public annual reports of enterprises to ensure the completeness and accuracy of the data.

[0039] The innovation investment dimension includes two core data points: R&D funding and the number of R&D personnel. R&D funding is derived from the R&D expenses disclosed in the company's annual report, and the number of R&D personnel is derived from the number of R&D personnel disclosed in the company's annual report. These two data points together reflect the company's level of R&D resource investment. The technology output dimension includes data on the number of green patent applications filed annually. Patent data is taken from publicly available intellectual property databases, and the number of green patent applications is selected according to the International Patent Classification (IPC) and the Green Technology Classification (GCC) standards to represent the technological output results of the company's R&D activities. The operating efficiency dimension includes multiple financial indicators such as fixed asset size, current asset size, total number of employees, operating revenue, liquidity ratio, sales growth rate, inventory turnover rate, and return on assets. All indicators are calculated according to corporate accounting standards and are used to reflect the company's operating status and resource endowment. The environmental performance dimension includes data related to the company's environmental performance rating, carbon emission reduction performance, environmental compliance performance, and credit trading revenue. The environmental performance rating is derived from the company's environmental rating results published by a third-party professional rating agency; carbon emission reduction performance is calculated based on the company's disclosed carbon emission data; environmental compliance performance is based on statistics of the company's environmental penalties and compliance records; and credit trading revenue is derived from the company's disclosed credit trading revenue data. The policy attribute dimension includes data on the enterprise's industry classification, ownership type, region, and annual government subsidy amount, which are used for subsequent group heterogeneity analysis and control variable setting.

[0040] After data collection and cleaning, role labeling was conducted based on the companies' main business scope and industry chain positioning. According to the segmentation of the new energy vehicle industry chain, the sample companies were divided into two main categories: a treatment group and a control group. The treatment group consisted of automobile manufacturing companies, further subdivided into three industry chain roles: core component manufacturing, passenger vehicle manufacturing, and commercial vehicle manufacturing. Core component manufacturing companies encompass those producing power batteries, drive motors, electronic control systems, and key traditional automotive components. These companies are upstream in the industry chain, undertaking core technology R&D and component production functions. Passenger vehicle manufacturing companies mainly engage in the design, integration, and assembly of passenger vehicles, and are directly regulated by the dual-credit policy, occupying a core position in the midstream of the industry chain. Commercial vehicle manufacturing companies mainly engage in the R&D and production of commercial vehicles; they also belong to the automobile manufacturing sector but are less directly affected by the policy than passenger vehicle companies. The control group consisted of other advanced manufacturing companies not involved in automobile manufacturing. These companies are not directly affected by the dual-credit policy and are used for reference and comparison in subsequent policy effect assessments. Figure 2 As shown, the samples after role labeling are entered into the subsequent two-stage efficiency measurement stage, serving as the basic analysis unit for efficiency evaluation.

[0041] The second step is to construct a two-stage green technology innovation efficiency evaluation system and conduct efficiency calculations. This embodiment, based on the innovation value chain theory, divides the entire green technology innovation process into a research and development (R&D) stage and a results transformation stage, correspondingly constructing a multi-level indicator system including initial input, intermediate output, additional intermediate input, and final output. The R&D stage corresponds to the scientific and technological achievement output stage. The core task is to transform R&D resources into technological knowledge achievements. This stage uses R&D funds and R&D personnel as two initial inputs, and the number of green patent applications filed by the enterprise annually as an intermediate output, reflecting the ability of R&D investment to transform into technological achievements. The results transformation stage corresponds to the technology value realization stage. The core task is to transform technological achievements into actual economic benefits and environmental benefits. This stage uses the patent achievements produced in the R&D stage as the core input, while introducing three additional intermediate inputs: fixed assets, current assets, and total number of employees, representing the production and operation resources required for the results transformation process. The final output includes two types of indicators: operating revenue and comprehensive environmental performance. The comprehensive environmental performance is calculated by coupling three types of data: carbon emission reduction performance, environmental compliance performance, and credit trading revenue. The calculation process standardizes the three types of data and then weights and sums them to simultaneously reflect the environmental value and policy benefits brought about by innovation activities. This overcomes the limitations of evaluating a single economic benefit indicator and aligns with the dual attributes of green technology innovation.

[0042] To adapt to the innovation resource allocation characteristics of different roles in the industry chain, this embodiment adopts a dynamic weight matching mechanism to set the minimum weight threshold for both stages, avoiding systematic evaluation bias caused by fixed weights. Traditional two-stage efficiency measurement generally uses a uniform fixed weight threshold, which cannot take into account the differences in innovation focus among different links in the industry chain, and is prone to bias in the efficiency measurement of R&D-driven or commercialization-driven enterprises. The dynamic weight mechanism of this embodiment adaptively adjusts the threshold according to the actual input structure characteristics of various enterprises, achieving a match between weight constraints and industry characteristics. Figure 3 As shown, the execution process of dynamically matching the minimum weight threshold includes three steps: input structure clustering, threshold formula calculation, and threshold substitution constraint, which are executed sequentially.

[0043] First, cluster analysis of input structure was conducted on sample enterprises in different roles within the industry chain. For the three sample categories—core component manufacturing, passenger vehicle manufacturing, and commercial vehicle manufacturing—the distribution characteristics of the proportion of R&D investment to total innovation investment were statistically analyzed, and the information entropy value of the input structure for each category was calculated. The input structure information entropy, calculated based on the proportion distribution of various input indicators, reflects the degree of dispersion of enterprise innovation investment. A higher entropy value indicates a more diversified input structure and a more balanced resource allocation, while a lower entropy value indicates a more concentrated input structure and a more pronounced resource bias. The cluster analysis employed a conventional clustering algorithm, grouping enterprises according to the similarity of their input structures to ensure consistency in the input characteristics of enterprises within the same category, providing a reliable classification basis for subsequent threshold calculations.

[0044] In the specific implementation of this embodiment, an adaptive threshold calculation formula is used to calculate the minimum weight threshold for the R&D stage corresponding to the three types of industry chain roles. The calculation formula is as follows: In the formula, For the first Minimum weight threshold for the R&D stage corresponding to a role in the industry chain. The preset global minimum weight benchmark value, For the first The average R&D investment as a percentage of total innovation investment for sample companies in the industry chain. For the first Information entropy of the input structure of roles in the industry chain This is a collection of all roles and categories across the entire industry chain.

[0045] The global minimum weight baseline value is set at 0.3. This value is based on the common weight boundary setting in existing two-stage network DEA research, and is determined in conjunction with the basic proportion characteristics of the two stages of innovation in the automotive manufacturing industry. This ensures that both stages occupy a reasonable proportion in the overall efficiency evaluation, avoids extremely low weights for a single stage, and ensures that both stages can effectively impact overall efficiency. The average R&D investment proportion is obtained by taking the arithmetic mean of the R&D investment proportion data of all sample enterprises in the corresponding category. It directly reflects the R&D investment intensity characteristics of this type of enterprise; a higher value indicates that the enterprise attaches more importance to R&D investment. The information entropy of the investment structure is calculated through the proportion distribution of various investment indicators. In the calculation process, the proportion value of each type of investment indicator is first statistically analyzed, and then the proportion items are weighted and summed based on the basic formula of information entropy to finally obtain the investment structure entropy value of the corresponding category. A higher entropy value indicates a more balanced investment structure.

[0046] The calculation logic of this formula is as follows: for enterprise categories with higher R&D investment ratios and higher investment structure entropy values, the corresponding minimum weight threshold for the R&D stage is lower. This means that for enterprises with high R&D investment intensity and diversified structures, the lower limit of the weight for the R&D stage is appropriately relaxed, giving greater room for weight fluctuation in the technology transfer stage, which aligns with the characteristics of these enterprises having complete innovation chains and strong transformation capabilities. For enterprise categories with lower R&D investment ratios and lower investment structure entropy values, the corresponding minimum weight threshold for the R&D stage is higher. This means that for enterprises with weak R&D foundations, the minimum weight requirement for the R&D stage is strengthened, avoiding excessive emphasis on the transformation stage in efficiency evaluation, and guiding enterprises to value R&D investment. The minimum weight threshold for the technology transfer stage is obtained by subtracting the minimum weight threshold for the R&D stage from 1, ensuring that the sum of the lower limits of the weights for the two stages is 1, maintaining the balance of the overall weight constraints.

[0047] After calculating the minimum weight thresholds for two categories, these thresholds are substituted into a two-stage network DEA model with weighted constraints, serving as weighted constraints in the efficiency calculation. The two-stage network DEA model used in this embodiment is constructed based on the assumption of variable returns to scale, which better reflects the industry reality of automobile manufacturing enterprises with significant differences in scale and varying returns to scale. During the solution process, non-Archimedean infinitesimal constraints are introduced to limit the lower bounds of the weights of each indicator, avoiding the indicator failure problem caused by weights reaching zero values, and ensuring that all input-output indicators can participate in the efficiency evaluation. The efficiency solution follows a sequence of overall efficiency first, then sub-stage efficiency. First, the optimal solution for the overall green technology innovation efficiency is solved; then, the efficiency of the R&D stage is maximized according to a preset priority; finally, the efficiency of technology transfer is calculated based on the additive weighted decomposition method.

[0048] In the specific implementation of this embodiment, the additive weighted decomposition standard formula is: ,in For the first The overall green technology innovation efficiency of the company. For the first The efficiency of a company's R&D phase. For the first The efficiency of a company's technology transfer process. As a weighting factor in the research and development phase, The stage weights are calculated based on the proportion of investment in each stage to the total investment, and satisfy the following conditions: The entire efficiency calculation process incorporates non-Archimedean infinitesimal constraints to define the lower limits of the weights of each input and output indicator, thus avoiding the calculation failure caused by the indicator weights returning to zero. The overall efficiency calculation result output by the model always satisfies the interval constraints. .

[0049] The weight calculation logic perfectly matches the basic additive decomposition framework of the two-stage network DEA: the total investment is the sum of the initial investment in the R&D stage and the additional intermediate investment in the commercialization stage. The proportion of resource investment in the R&D stage and the proportion of resource investment in the commercialization stage are calculated separately, directly corresponding to the two-stage weights. , This model is built on the assumption of variable returns to scale, and is adapted to the industry characteristics of the automobile manufacturing industry, which has large differences in enterprise size and differentiated returns to scale. The solution priority is fixed as first optimizing the overall global efficiency, then maximizing the efficiency of the R&D sub-stage, and then using the above weighted decomposition to derive the efficiency of the results transformation stage, thus completing the output of the staged efficiency breakdown.

[0050] The third step involves constructing the PSM-DID assessment framework to measure the net impact of the dual-credit policy. This embodiment uses the year the dual-credit policy was officially implemented as the policy impact point, and incorporates the treatment group (automobile manufacturing) and other advanced manufacturing control groups into a quasi-natural experiment analysis framework. Traditional difference-in-differences (DID) methods directly compare two groups of samples, which is easily affected by initial differences in the samples, leading to sample selection bias. This embodiment first performs propensity score matching to construct comparable samples, and then conducts DID analysis based on the matched samples, effectively mitigating sample selection bias and endogeneity issues, and improving the accuracy of policy effect assessment.

[0051] During the matching process, five characteristic variables were selected as matching variables: liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale. These correspond to the company's solvency, growth capacity, operational efficiency, profitability, and policy support level, respectively. All five variables are important factors affecting corporate innovation efficiency and can effectively characterize the company's fundamental features. First, a propensity score was calculated for each sample company using Logit regression. In the regression, whether the company belongs to the treatment group was used as the dependent variable, and the five matching variables were used as explanatory variables. The probability of each company entering the treatment group was output as its propensity score. Then, within the common support domain, a nearest neighbor matching method with calipers was used to match companies in the treatment and control groups one-to-one. The caliper width was set to one-quarter of the standard deviation of the propensity score to control the matching distance and avoid excessive matching bias, ensuring that the feature similarity of the matched samples met the requirements. After matching, a balance test was performed on the samples to verify that there were no significant differences in the matching variables between the treatment and control groups, ensuring that the matching effect met the prerequisite requirements for subsequent difference-in-differences analysis.

[0052] A two-way fixed-effects difference-in-differences model was constructed based on the matched samples. The model incorporates individual fixed effects and time fixed effects, respectively controlling for time-invariant enterprise characteristics and time-trend factors, thus eliminating the interference of time-invariant individual heterogeneity and common time shocks on the estimation results. The model uses three categories of indicators—overall green technology innovation efficiency, R&D efficiency, and technology transfer efficiency—as dependent variables, and the interaction term between the policy timing variable and the treatment group variable as the core explanatory variable. Liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale are also included as control variables. Regression calculations output the coefficients of the core explanatory variables, which represent the net impact of the dual-integral policy on the corresponding efficiency indicators. The sign of the coefficient indicates the direction of the policy's influence, and the magnitude of the coefficient indicates the intensity of the policy's influence.

[0053] To verify the reliability of the baseline regression results, this embodiment performs four robustness checks. The first check adjusts the priority of sub-stage efficiency calculations, prioritizing maximizing the efficiency of the results transformation stage before recalculating the three efficiency indicators, and then repeats the regression calculation to verify the impact of the calculation order on the results and eliminate bias caused by the choice of calculation order. The second check removes sample data from the year of policy implementation and then re-executes the matching and regression calculations to eliminate interference factors during the policy transition period and verify the stability of the policy effect. The third check sets a virtual policy time point and performs a placebo test by selecting samples from the corresponding period. Setting the virtual policy time point to a year before policy implementation proves that the actual policy effect exists, verifying the time specificity of the policy effect. The fourth check randomly selects samples to construct a fictitious treatment group and repeats the regression calculation multiple times to verify the distribution characteristics of the results. If the baseline regression results significantly deviate from the random distribution, it proves that the effect is not caused by random factors, eliminating spurious regression problems caused by random factors. After all four checks pass, the baseline regression results are confirmed to be robust, and the assessment conclusions are credible.

[0054] The fourth step involves heterogeneity analysis and policy transmission path identification. Baseline regression only reflects the average effect of policies and cannot capture the differences in policy responses among different types of enterprises. This embodiment conducts grouped regressions across four dimensions to identify the multidimensional heterogeneity of policy effects, providing a basis for differentiated policy formulation. The first dimension is the industry chain role dimension, performing regressions on three types of samples: core component manufacturing, passenger vehicle manufacturing, and commercial vehicle manufacturing. This compares the differences in the impact of policies on different links in the industry chain, identifies the industry chain spillover characteristics of policies, and verifies whether policy effects are transmitted from the directly regulated vehicle manufacturing segment to the upstream component segment. The second dimension is the enterprise ownership dimension, performing regressions on state-owned and non-state-owned enterprise samples to compare the differences in policy responses among enterprises with different ownership structures and analyze the moderating effect of ownership attributes on policy effects. The third dimension is the regional dimension, performing regressions on enterprise samples from the eastern and central-western regions to compare the differences in policy effects under different regional industrial bases, reflecting the impact of regional industrial ecosystems on policy effects. The fourth dimension is the initial efficiency level dimension. Using the average overall efficiency of the year before the policy implementation as the dividing line, the sample is divided into an initial high efficiency group and an initial low efficiency group. The differences in the effects of the policy on different basic enterprises are compared to verify whether the policy has a Matthew effect.

[0055] To clarify the intrinsic transmission path of policy's impact on innovation efficiency, this embodiment constructs a mediation effect testing model, setting two types of mediating variables: R&D investment intensity and patent output scale. The testing process employs a stepwise regression method. The first step verifies the significance of the policy's impact on the mediating variables; the second step verifies the significance of the mediating variables' impact on the corresponding efficiency indicators; and the third step simultaneously incorporates the policy variables and mediating variables to perform regression, judging the existence and direction of the mediation effect based on coefficient changes. Ultimately, this clarifies the specific transmission path of the dual-integral policy's impact on innovation efficiency, revealing the intrinsic mechanism by which the policy exerts its influence.

[0056] To address the policy spillover effects across the industrial chain, this embodiment employs a quantitative formula to calculate the spillover intensity, focusing on the transmission effect of policies from passenger vehicle manufacturing to core component manufacturing. Traditional studies often qualitatively describe industrial chain spillover effects and lack quantitative measurement methods. This embodiment achieves precise quantification of the effects by constructing a spillover intensity formula.

[0057] In the specific implementation of this embodiment, the formula for calculating the intensity of technology spillover from the industrial chain is as follows: In the formula, For downstream links Upstream links The intensity of policy spillover effects The regression coefficient for the policy effect in the downstream sector. The regression coefficient of the policy effect in the upstream segment. This represents the direct input-output consumption coefficient between upstream and downstream links. This is the coefficient representing the degree of technological correlation between upstream and downstream links. This is the set of all upstream links.

[0058] The direct consumption coefficient of input-output is taken from the corresponding industry data in the industry input-output table. It represents the direct consumption of upstream products by a unit of output in the downstream segment, reflecting the degree of economic linkage between industries. The higher the coefficient, the higher the economic interdependence between upstream and downstream. The technology linkage coefficient is calculated through upstream and downstream patent citation data. It statistically analyzes the proportion of downstream industry patents citing upstream industry patents, reflecting the closeness of technology linkage between industries. The value ranges from 0 to 1, and the higher the coefficient, the stronger the technological linkage between upstream and downstream. The numerator of the formula is the product of the linkage degree of a specific upstream and downstream combination, and the denominator is the sum of the products of the linkage degrees of all upstream and downstream segments. It is used to calculate the weight ratio of the upstream segment in all upstream linkages of the downstream, reflecting the importance of the upstream segment in supporting the downstream.

[0059] A positive result indicates a positive technology spillover effect, representing that downstream policy effects drive efficiency improvements in upstream processes, and downstream innovation demand is transmitted upstream, promoting technological progress. A negative result indicates a reverse siphon effect, representing that downstream policy implementation has a draining effect on upstream innovation resources, with resources concentrating in the midstream and downstream, thus affecting upstream innovation capabilities. The spillover effect strength is taken as the absolute value of the result, used to measure the degree of policy transmission impact; the larger the value, the more significant the spillover effect.

[0060] The fifth step is to perform efficiency bottleneck diagnosis and generate optimization solutions. This embodiment, based on the results of two-stage efficiency measurement, identifies the efficiency bottleneck links in each company's entire innovation chain, providing targeted guidance for precise optimization. The diagnosis process first calculates the industry average for R&D efficiency and the industry average for technology transfer efficiency of all sample companies in the same industry chain role, as efficiency benchmarks. Then, the R&D efficiency and technology transfer efficiency values ​​of a single company are compared with the corresponding industry averages. The stages with values ​​lower than the corresponding industry averages are identified as the efficiency bottleneck links of that company. If the values ​​of both stages are lower than the industry averages, the company is identified as a dual-bottleneck type, requiring optimization of resource allocation in both stages simultaneously.

[0061] For the identified bottlenecks, corresponding optimization directions for enterprise-side innovation resources are generated. For enterprises facing bottlenecks in R&D efficiency, the focus is on providing optimization suggestions regarding adjusting R&D investment structure, optimizing R&D personnel allocation, and guiding the direction of core technology breakthroughs, helping enterprises improve the R&D input-output ratio and solidify their innovative technological foundation. For enterprises facing bottlenecks in technology commercialization efficiency, the focus is on providing optimization suggestions regarding technology industrialization path design, improving market access mechanisms, and expanding channels for realizing environmental benefits, helping enterprises accelerate the commercialization of technological achievements and enhance the overall benefits of innovation.

[0062] Based on the differences in policy effects across different dimensions obtained from heterogeneity analysis, policy adjustment recommendations for the industry are generated. Differentiated incentive rules are designed for different links in the industrial chain to strengthen guidance on technological innovation of core components and promote collaborative innovation across the upstream and downstream of the industrial chain. Supporting policies for enterprises of different ownership types are improved to address the resource shortcomings of non-state-owned enterprises in the technology transfer stage and enhance policy inclusiveness. Tiered policy schemes are formulated for different regions to strengthen policy support for industrial innovation in the central and western regions and narrow the innovation gap between regions. A tiered incentive mechanism is established for enterprises with different initial efficiency levels, balancing policy inclusiveness and precision, ensuring that weaker enterprises receive basic support while incentivizing stronger enterprises to continuously improve their innovation capabilities.

[0063] In summary, this embodiment, through sample coverage and role stratification from a full industry chain perspective, achieves a systematic assessment of the impact of the dual-credit policy on innovation across the entire automotive manufacturing chain, breaking through the limitations of previous studies that only focused on vehicle manufacturers. By employing an adaptive dynamic weighting mechanism and an additive weighted decomposition efficiency measurement model, it effectively adapts to the innovation characteristics of different links in the industry chain, improving the accuracy of phased efficiency measurement and correcting assessment biases caused by fixed weights and redundant measurements. Through an output indicator system that couples economic and environmental benefits with the PSM-DID assessment framework, it comprehensively quantifies the net impact of the policy on the entire innovation process, clarifying the policy transmission path and industry chain spillover patterns. The resulting enterprise-side optimization solutions and industry-side policy recommendations can directly provide quantitative basis for enterprise innovation resource allocation and precise optimization of industrial policies, possessing practical support value for promoting the green transformation and development of the automotive industry.

[0064] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain, characterized in that, The method includes the following steps: S1. Collect data on innovation input, technological output, operating efficiency, environmental performance, and policy attributes of target sample enterprises. Based on the enterprise's main business scope and industrial chain positioning, label three types of industrial chain roles: core component manufacturing, passenger vehicle manufacturing, and commercial vehicle manufacturing. S2. Construct a two-stage green technology innovation efficiency evaluation index system that includes initial input, intermediate output, additional intermediate input and final output. Based on the enterprise's industrial chain role label and input structure characteristics, dynamically match the minimum weight threshold of the R&D stage and the achievement transformation stage. Use a two-stage network DEA model with weight constraints to solve the sub-stage efficiency and decompose it to obtain the overall green technology innovation efficiency, R&D efficiency and achievement transformation efficiency. S3. The implementation of the dual-credit policy is used as a quasi-natural experiment. Automobile manufacturing enterprises are used as the treatment group and other advanced manufacturing enterprises are used as the control group. Enterprise characteristic variables are selected as matching variables to implement propensity score matching (PSM). Based on the matched samples, a two-way fixed-effect difference-in-differences (DID) model is constructed. The PSM-DID evaluation framework is formed by combining propensity score matching and the difference-in-differences model to measure the net impact of the dual-credit policy on three types of efficiency indicators. S4. Group regressions are performed according to the role in the industrial chain, enterprise ownership, region and initial efficiency level to identify the multidimensional heterogeneity of policy effects, construct a mediation effect test model to identify policy transmission paths, and measure the spillover intensity of policies between upstream and downstream links in the industrial chain. S5. Based on the two-stage efficiency measurement results, identify the efficiency bottleneck links in the entire innovation chain of enterprises, and combine the heterogeneity analysis conclusions to generate innovation resource optimization directions and industrial policy adjustment suggestions for different types of enterprises.

2. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, The step S2, which dynamically matches the minimum weight threshold between the R&D stage and the commercialization stage, includes the following steps: We conducted cluster analysis on the input structure of sample enterprises in different roles in the industrial chain, and statistically analyzed the distribution characteristics of the proportion of R&D investment to total innovation investment for enterprises in each role. Based on the proportional distribution characteristics obtained from clustering, adaptive threshold calculation formulas are used to set corresponding minimum weight thresholds for the R&D stage and the achievement transformation stage for core component manufacturers, passenger vehicle manufacturers, and commercial vehicle manufacturers. The adaptive threshold calculation formula is as follows: ,in For the first Minimum weight threshold for the R&D stage corresponding to a role in the industry chain. The preset global minimum weight benchmark value, For the first The average R&D investment as a percentage of total innovation investment for sample companies in the industry chain. For the first Information entropy of the input structure of roles in the industry chain A collection of all roles across the industry chain; The two minimum weight thresholds that have been set are substituted into the two-stage DEA network model with weight constraints, and used as weight constraints in the efficiency solution calculation.

3. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, The propensity score matching (PSM) process in step S3 includes the following steps: The liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale were selected as matching variables, and the propensity score of each sample enterprise was calculated by Logit regression. Within the common support domain, a nearest neighbor matching method with calipers was used to match enterprises in the treatment group and the control group one-to-one, with the caliper width set to one-quarter of the propensity score standard deviation. A balance test was performed on the matched samples to verify that there were no significant differences in the matched variables between the treatment group and the control group after matching.

4. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, The step S4, which involves calculating the spillover effect of the policy between upstream and downstream links of the industrial chain, includes the following steps: The samples of core component manufacturers and passenger vehicle manufacturers were grouped and regressed separately to obtain the policy effect regression coefficients for the two types of enterprises. Retrieve upstream and downstream industry correlation data to calculate the industry correlation weight and technology correlation coefficient between the core component manufacturing link and the passenger vehicle manufacturing link; The spillover intensity from passenger vehicle manufacturing to core component manufacturing under the dual-credit policy is calculated using the industrial chain technology spillover intensity calculation formula. The industrial chain technology spillover intensity calculation formula is as follows: ,in For downstream links Upstream links The intensity of policy spillover effects The regression coefficient for the policy effect in the downstream sector. The regression coefficient of the policy effect in the upstream segment This represents the direct input-output consumption coefficient between upstream and downstream links. This is the coefficient representing the degree of technological correlation between upstream and downstream links. This is the set of all upstream links.

5. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, In step S2, the two-stage green technology innovation efficiency evaluation index system sets two initial inputs in the R&D stage: R&D funds and R&D personnel; one intermediate output: the number of patent applications; and in the technology transfer stage, the patent results corresponding to the intermediate outputs are set as the core inputs, three additional intermediate inputs: fixed assets, current assets, and total number of employees. Finally, it sets two final outputs: operating revenue and comprehensive environmental performance. The comprehensive environmental performance is calculated by coupling three types of data: carbon emission reduction performance, environmental compliance performance, and points trading revenue collected in step S1.

6. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, In step S2, the two-stage network DEA model with weighted constraints is constructed based on the assumption of variable returns to scale. In the efficiency solution process, the optimal solution for the overall green technology innovation efficiency is first obtained, then the efficiency of the R&D stage is maximized according to a preset priority, and finally, the efficiency value of the technology transfer stage is calculated based on the additive weighted decomposition method. The additive weighted decomposition formula is as follows: ,in For the first The overall green technology innovation efficiency of the company. For the first The efficiency of a company's R&D phase. For the first The efficiency of a company's technology transfer process. As a weighting factor in the research and development phase, Weighting for the results transformation stage.

7. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, In step S3, the two-way fixed effects difference-in-differences (DID) model incorporates individual fixed effects and time fixed effects. The model uses three types of indicators—overall green technology innovation efficiency, R&D efficiency, and achievement transformation efficiency—as dependent variables, respectively. The interaction term between the policy timing variable and the treatment group variable is used as the core explanatory variable. Liquidity ratio, sales growth rate, inventory turnover rate, return on assets, and government subsidy scale are used as control variables. The coefficient values ​​of the core explanatory variables are used as the quantitative results of the net policy impact.

8. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, In step S3, after calculating the net impact of the dual-integral policy on the three types of efficiency indicators using the PSM-DID evaluation framework, four robustness test operations are performed. The first is to adjust the priority of sub-stage efficiency measurement, recalculate the efficiency indicators, and repeat the regression operation. The second is to remove the sample data from the year the policy was implemented and then re-execute the matching and regression operation. The third is to set a virtual policy time point and extract samples from the corresponding time period to perform a placebo test. The fourth is to randomly select samples to construct a fictitious treatment group and repeat the regression operation multiple times to verify the distribution characteristics of the results.

9. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain as described in claim 1, characterized in that, The mediation effect testing model constructed in step S4 sets up two types of mediating variables: R&D investment intensity and patent output scale. In the testing process, the significance of the policy's impact on the mediating variables is first verified, and then the significance of the mediating variables' impact on the corresponding efficiency indicators is verified. The existence and direction of the mediation effect are determined by stepwise regression, and the specific transmission path of the policy's impact on innovation efficiency is clarified.

10. The method for evaluating the phased innovation efficiency of automobile manufacturing enterprises based on the industrial chain according to claim 1, characterized in that, In step S5, when identifying efficiency bottlenecks, the industry average for R&D efficiency and commercialization efficiency of all sample companies in the same industry chain role is first calculated. Then, the R&D efficiency and commercialization efficiency of a single company are compared with the corresponding industry averages. The stage where the value is lower than the corresponding industry average is determined as the efficiency bottleneck of the company. Based on the bottleneck attributes, corresponding innovation resource optimization directions are generated. Combined with the differences in policy effects in different dimensions obtained from heterogeneity analysis, suggestions for industrial policy adjustments in the corresponding industry chain role, ownership, and regional dimensions are generated.

Images