4.1. Variables and Models
To verify hypothesis 1, we use the OLS model. First, to examine the impact of global value chain embedding on corporate risk taking, the model is designed as follows:
In model (1), the dependent variable in this article is risk taking. A company with a high level of risk taking implies that the company tends to choose investment projects with high risk and high return. By summarizing the existing literature, the indicators commonly used include surplus volatility [
14,
40], the volatility of stock returns, etc. [
41,
42]. Due to the great volatility of the Chinese stock market, the level of risk assumption of Chinese enterprises is widely measured by Roa volatility. This paper also uses the degree of Roa volatility of enterprises during the observation period to measure the level of corporate risk taking. The greater the volatility of the Roa, the higher the level of corporate risk taking. The Roa is measured by the EBIT divided by the total assets at the end of the year. According to John et al.’s [
30] research, in order to mitigate the impact of the industry and the cycle, one should subtract the company’s Annual Industry mean from the corporate’s Roa to obtain the Adj_Roa. Specifically, with every five years (t − 4 to t years) as an observation period, the observations of the Roa without 5 consecutive years are excluded, and then the standard deviation and range of the industry-adjusted Roa (Adj_Roa) are calculated on a rolling basis. Referring to Facio et al. [
43], we multiply the result by 100 to obtain the level of risk-taking.
The formula is as follows:
The independent variable in this article is GVC embedding, which represents the degree of enterprises’ global value chain embedment. Since the traditional method is based on the macro estimation method of the World Input–Output Database (WIOD), this type of method is represented by the KWW method of Koopman et al. [
16,
17] and the WWZ method of Wang et al. [
18]. But this method can only carry out industry-level analysis and is unsuitable for micro-enterprise-level analysis. With the availability of Chinese enterprise-level customs trade data, it is possible to estimate the FVAR of enterprise exports directly at the micro-level. This paper is based on the matching data of the CSMAR database and the China Customs import and export database, referring to Upward et al. [
16], Zhang et al. [
26], Lv et al. [
7], and other that studies used enterprise micro-data to measure the GVC-embedding degree of 1102 listed companies in China from 2000 to 2016. The specific calculation method is as follows:
represents the real import value of intermediate goods, represents the real export value, represents the real processing trade import value, represents the real general trade import value, and represents the real import value of general trade. D represents the domestic sales value.
Drawing on the existing literature, we have controlled for other factors that affect the risk assumption of the enterprises in our models. The firm age is the operating life of the enterprise; the longer the operating life of the enterprise, the higher the level of risk assumption, and its regression coefficient is expected to be significantly positive [
44]. Cap is a cash payment made by a company for constructing fixed assets, intangible assets, and other long-term assets, measured by the ratio of capital expenditure to total assets at the end of the period [
45]. The higher the risk-taking level of the enterprise, the more capital expenditures formed by its long-term investment in fixed assets, intangible assets, and other long-term assets; thus, the regression coefficient is expected to be significantly positive [
46]. PPE stands for fixed asset ratio, which is equal to the ratio of fixed assets to the total assets at the end of the period [
41,
47]. The regression coefficient is expected to be significantly positive. Growth is the growth of the enterprise, which is expressed by the growth rate of the sales revenue of the enterprise [
41,
45,
47]. The faster the company’s sales revenue grows, the stronger the company’s profitability, and the weaker the company’s motivation to obtain income through venture capital. The estimated coefficient is expected to be significantly negative [
46]. Ownership is the degree of equity concentration, which is equal to the sum of the shareholding ratios of the top five shareholders [
46]. Major shareholders may choose more stable investment projects because of the pursuit of private income [
28]; so, the regression coefficient is expected to be significantly negative. Indratio is the proportion of independent directors, which is equal to the ratio of the number of independent directors to the total number of board members [
48]. The larger the proportion of independent directors, the stronger the independence of the board of directors, the more effective the exercise of rights to supervise the management, the more restrictive the management’s risk-averse behavior, and the higher the level of risk taking of the enterprise. Therefore, the regression coefficient is expected to be significantly positive.
To verify Hypothesis 2, we also use the OLS regression model:
Model (2) adds R&D to model (1), which represents the company’s R&D investment, measured by the proportion of R&D expenditure to the primary business income. Through global value chain embedding, companies invest more in R&D activities, thereby increasing the level of corporate risk taking. If β2 is positive, it means that R&D investment plays an intermediary role in promoting corporate risk assumptions in global value chains. Based on assumption 2, we expect β2 to be significantly positive. The definition of the remaining variables in model (2) is the same as in model (1).
Model (3) adds the variable SA and its interaction term with the GVP to model (1). Among them, SA is the level of corporate-financing constraints. Hadlock and Pierce divided the types of corporate-financing constraints based on corporate financial reports, and constructed a financing constraint index using the size and age of enterprises, calculated as: SA = 0.737 × size + 0.043 × size2 − 0.04age. Compared with other financing constraint composite indices, the two variables used in the SA index, the size of the enterprise and the age of the enterprise, do not change much over time, are highly exogenous, and have good characteristics for the financing constraints of Chinese enterprises, so this paper uses the SA index to measure the financing constraints of enterprises. The higher the financing constraints of the enterprise, the more difficult it is for the enterprise to obtain funds, which is not conducive to the research and development activities of the enterprise. If the coefficient of GVC ∗ SA is significantly negative: the higher the financing constraint, the weaker the promotion effect of GVC ∗ SA on the risk assumption of enterprises. Based on assumption 3, we expect β3 to be significantly negative. The definition of the remaining variables in model (3) is the same as in model (1).
4.2. Sample Selection and Descriptive Statistics
The independent variable GVC embedding is derived from the CSMAR China Listed Company Database and China Customs Import and Export Data. After matching the company name, the enterprise’s legal person, and further matching the post code and company phone number, we eliminated unreasonable matching data and missing values. Finally, the panel data of 1102 listed companies from 2000 to 2016, a total of 5292 valid samples, were sorted out.
It should be noted that the processing method of database matching in this paper refers to the related research of Zhang et al. [
7,
26]. The data-matching process is listed as follows: first of all, this paper matches the enterprise name and the enterprise legal person, which is not easy to change in the short term. Secondly, this paper further uses the postal code of the company’s location and the company’s phone number to match the companies that were not successfully matched the first time. The consolidation of the listed company information with the income statement, balance sheet, etc., is calculated based on the year-end value of the parent company’s statement. For a situation wherein the company name changes during the year, we determine the individual company based on the company’s security code that will not change. The consolidated enterprise data also include the financial information and import and export information of the enterprise. Finally, this paper selects the panel data of listed companies that meet the research requirement.
The data on the dependent, moderate, and controlled variables were derived from the CSMAR Database, and the mediated variable R&D was obtained from the China Wonder Database (Wind) for the initial data. For the above data, we have carried out the following processing scheme: (1) exclude enterprises in the financial industry, (2) exclude companies with relevant financial data, (3) eliminate missing data values, and (4) winsorize the processing of 1% for continuity variables. Consequently, we obtained a total of 8485 samples from 1111 companies.
Table 1 reports the descriptive statistical results for the main variables. Judging from the statistical results of the entire sample description, the average value of RiskT1 is 2.871, the maximum value is 33.34, and the minimum value is 0.131. The average value of RiskT2 is 7.064, the maximum value is 82.28, and the minimum value is 0.321, indicating that there is a significant difference in the risk-taking level of enterprises. The average value of the GVC is 0.185, the maximum value is 1, and the minimum value is 0, indicating that China as a whole is still in the downstream position of the global value chain. The remaining variables are within the normal range and have no extreme values.
Table 2 reports the correlation coefficients between the variables. Among them, the level of corporate risk commitment (RiskT1 and RiskT2) is significantly positively correlated with the GVC, indicating that as enterprises become more deeply involved in global value chains, their risk level also gradually increases, which is consistent with assumption 1. In addition, the level of corporate risk taking (RiskT1 and RiskT2) is also significantly positively correlated with the R&D level, which confirms hypothesis 2 to some extent. The correlation coefficients between the relevant control variables in this paper are all lower than 0.6, indicating that there is no serious multicollinearity problem between the related control variables.