The Impact of Public R & D Subsidies on Private R & D Expenditure and Its Innovation Performance—An Empirical Study Based on Industrial Enterprises in Guangdong Province

Based on the panel data of Guangdong industrial enterprises from 2006 to 2013, this paper empirically studies the impact of public R & D subsidies on private R & D expenditure and the impact of the two on the innovation performance of enterprises by using random effects model and fixed effects model. The results show that public R & D subsidies can promote enterprises to carry out private R & D expenditure, and there is an obvious time lag. Using grouping studies, we found that low-level public R & D subsidies can en-courage enterprises to increase their private R & D expenditure, but high-level public subsidies will crowd out part of private R & D expenditure. Finally, this paper divides the innovation performance of enterprises into the stage of knowledge output and the stage of marketization. In the knowledge output stage, public R & D subsidies and R & D staff input can significantly improve the innovation performance of enterprises; in the marketization stage, public R & D subsidies cannot promote the increase of new product income, while the private R & D expenditure of enterprises can significantly increase the income of new products.


Introduction
Today, with economic globalization and increasingly fierce technological competition, the impact of technological innovation on economic growth is increa-How to cite this paper: Xia Table 1 shows the meaning, abbreviations, and units of variables.

Model Setting and Data Processing
In addition, i γ is a random error term that reflects the industry difference, and , i t  reflects the mixed difference between the industry and time.

Model 2
In the existing literature on measuring innovation performance, the knowledge production function is usually established with the extended Cobb-Douglas knowledge production function as the basic model. The input of this function is expressed by R & D capital and personnel, and the innovation performance is expressed by the number of patents and the sales revenue of new products, the form is as follows: Among them, Y represents the innovation output of the enterprise, A is a constant, K represents the R & D investment of the enterprise, L represents the input of the enterprise R & D personnel, X represents other control variables, and β and γ are random error terms, which represent the output elasticity of R & D capital and personnel, and indicate the degree of influence of each control variable on innovation performance. Based on formula (2), this paper divides R & D capital investment into government input (GRD) and enterprise input (ERD), and performs logarithm processing. Finally, the following formula is obtained: Table 2 shows the meaning of the variables.

Data Collection and Descriptive Statistics
The data in this paper comes from the industry data of large and medium-sized  Table 3 gives descriptive statistical characteristics of the explained, explanatory, and control variables. The overall quality of the data is good and there are no extreme abnormal values, which is suitable for the next analysis.
The commonly used estimation methods for panel data are least squares estimation, fixed effect model and random effect model. The data in this paper are short panel data. The least squares estimation, fixed effect regression and ran-dom effect regression are performed on the above models by using Stata12.0 software. Table 4 gives the results of the model by using Stata12.0 software.

Comparative Analysis of FE and RE
The results of fixed-effect analysis using Stata12.0 software showed that F test that all u_i = 0, F(19, 136) =10.17, Prob > F = 0.0000, significantly rejecting the null hypothesis that each sample does not have its own intercept term, so it can be preliminarily judged that each individual has different intercept terms, and the fixed effect model is superior to the ordinary least squares regression model.

Hausman Test
The Hausman test was used to compare and analyze the fixed effects model and the random effects model. The null hypothesis of the Hausman test is to use a random effects model. The results of the Hausman test: Prob > chi2 = 0.0000, P value is much lower than 0.05 (see the figure below), rejecting the null hypothesis that the random effects model should be used, so the model is more reasonable to use a fixed effects model.

Analysis of Regression Results of Fixed Effect Models
In summary, this paper should construct a fixed effects model to analyze the regression relationship between public R & D subsidies and private R & D expenditure. The fixed effects model estimation results show that the model has an F value of 474.46 and a significant P value of 0.0000, the model is very significant overall. The R-square within group of the model is 0.9331 (within = 0.9331), that is, the change ratio explained in the unit is 93.31%; the R-square between group of the model is 0.9831 (between = 0.9831); the overall R-square of the model is 0.9683 (overall = 0.9683), indicating that the explanatory power of the model is very strong.

Hysteresis Analysis
Among the above four R & D external funding sources, only the results of the two subsidies are significant. To test whether the incentive effect of subsidies is in time difference, this paper makes a lag analysis on the above variables.  In other words, the profit factor is indeed considered when the company con-

Group Static Estimation
To public R&D subsidies public R&D subsidies rate 100% All R&D investment of the enterprise = × According to the Stata group processing, the government R & D subsidies rate is 3.7% higher than the high subsidies rate, 2.5% or more and less than 3.7% is the medium-high subsidies rate, 1.4% or more and less than 2.5% is the medium-low subsidies rate, less than 1.4% is low subsidies rate. Fixed-effect regression analysis was performed on the following four sets of data.
, , , Table 6 shows the largest multiplier comes from the sample of the low public subsidies rate group, which is 40.24, but the coefficient is not significant, indicating that the lower public R & D subsidies can promote industrial enterprises to carry out more private R & D expenditure. The sample of the low-and medium-high public subsidies rate group showed significant high multiplier effects,

"U-Shaped" Relationship Test
To   Table 6, it can be found that the coefficient θ is significantly positive at the level of 1%, which indicates that the Guangdong Provincial Government's R & D subsidies for industrial enterprises has an influence on the "U-shaped" curve of the enterprise's private R & D expenditure. It differs from the "inverted U-shaped" curve that most literatures have drawn. In the group regression of the previous part, we conclude that the lower public subsidies rate has the largest multiplier effect on the private R & D expenditure of enterprises, and the high public subsidies level has the least promotion effect on the private R & D expenditure of enterprises. Think of the impact of public R & D subsidies on the private R & D expenditure from 2006 to 2013 as a whole, which is located on the left side of the inflection point of the "U" curve and is close to the inflection point. Therefore, the low subsidies rate's multiplier effect is the biggest. On the contrary, the high subsidies rate's multiplier effect is the weakest. This phenomenon may be caused by the fact that the Guangdong public R & D subsidies for industrial enterprises is too small.

The Impact of Public R & D Subsidies on Corporate Innovation Performance
Enterprises invest in R & D resources for innovative production. The production process can be divided into two stages-the stage of knowledge output and the stage of marketization. In line with this, innovative outputs can be divided into two types-knowledge output and market return. This paper comprehensively considers the two stages of enterprise technology innovation, and uses the number of patent applications and new product sales revenue to measure the technological innovation performance of enterprises. The number of patent applications is the explained variable of the knowledge output stage, which can reflect the output of enterprise technology innovation to a certain extent; the new product sales income is the explained variable in the marketization stage, which can reflect the commercialization level and market value of enterprise technology innovation.

Knowledge Output Stage
The knowledge output of enterprise innovation, also called intermediate output,  Table 7 shows the results of regression analysis on the above model.

Comparative Analysis of FE and RE
The fixed effects analysis results using the stata12.0 software showed (F test that Table 7. Impact of public R & D subsidies on innovation performance of enterprises in the stage of knowledge output. all u_i = 0: F(14, 61) = 9.55, Prob > F = 0.0000), which significantly rejected the null hypothesis that each sample did not have its own intercept term. Therefore, we can preliminarily judge that every individual has different intercept terms, and the fixed effects model is superior to the ordinary least squares regression model.

Hausman Test
In this paper, the Hausman test is used to compare the fixed effect model and the random effects model. The null hypothesis of the Hausman test is to use a random effects model. The results of the Hausman test: P value (Prob > chi2 = 0.7639) accepts the null hypothesis that the random effects model should be used, so it is more reasonable to use the random effects model.
Since the results of RE and OLS are the same, the maximum likelihood method is used to estimate the random effects model, and the result is the MLE of Table 7.

Analysis of Regression Results
According to the random effects regression results of maximum likelihood esti-

Marketization Stage
The   Table 8 shows the results of regression analysis on the above model.

Comparative Analysis of FE and RE
The results of fixed effects analysis using stata12.0 software showed (F test that all u_i = 0: F(14, 61) =7.77 Prob > F = 0.0000), which significantly rejected the null hypothesis that each sample does not have its own intercept term, so we can preliminarily judge that every individual has different intercept terms, and the fixed effects model is better than the ordinary least squares regression model.

Hausman Test
The Hausman test is used to compare the fixed effects model and the random effects model. The null hypothesis of the Hausman test is to use a random effects model. The results of the Hausman test show that the P value (Prob > chi2 = 0.3746) is significantly greater than 0.05, and the null hypothesis that the random effects model should be used is accepted, so it is more reasonable to use the random effects model.

Analysis of Regression Results
In

Conclusions
Through empirical analysis, the paper draws the following conclusions:

Suggestions
Therefore most precious harvest of my postgraduate stage. In the future, I hope that everyone will help each other and work together.