*3.3. Analysis Model and Variables*

Based on the above theoretical analysis, this study first constructs a model to calculate the ideal bank loans (BL). The bank loans estimation model is mainly inspired by Berger and Udell (2006) [88]. The authors propose a complete conceptual framework for bank credit determination. In that framework, the main factors affecting credit availability are divided into eight lending technologies: financial statement lending, small business credit scoring, asset-based lending, factoring, fixed-asset lending, leasing, relationship lending, and trade credit. Among these technologies, some are suitable for banks' general lending decisions. As for financial statement lending, bank's lending decision is based on a firm's financial ratios reflecting its financial condition in financial statements. Credit scoring is a transaction technology based primarily on hard information about the firm's owner as well as the firm. Asset-based lending is a lending technology which banks make lending decisions with by focusing on a subset of the firm's assets. This technology provides working capital financing secured primarily by accounts receivable and inventory. Fixed-asset lending technologies involve lending against assets that are long-lived and are not sold in the normal course of business. Accordingly, firm's liquidity, profitability, growth, ownership, are common determinants in bank lending. These indicators are closely associated with firm's short-term or long-term ability of repayment [89,90].

The bank loan estimation model is as follows:

$$\begin{array}{c} \text{Loan}\_{\text{fit}} = \mathbf{a}\_0 + \mathbf{a}\_1 \mathbf{Cash}\_{\text{lt}-1} + \mathbf{a}\_2 \mathbf{Size}\_{\text{lt}-1} + \mathbf{a}\_3 \mathbf{Lew}\_{\text{lt}-1} + \mathbf{a}\_4 \mathbf{Liquid}\_{\text{lt}-1} + \mathbf{a}\_5 \mathbf{Z}\_{\text{lt}-1} + \mathbf{a}\_6 \mathbf{Roe}\_{\text{lt}-1} + \mathbf{a}\_7 \mathbf{Growth}\_{\text{lt}-1} + \\\ \mathbf{a}\_8 \mathbf{Top1}\_{\text{lt}-1} + \mathbf{a}\_9 \mathbf{Hirth}ah \mathbf{h}\_{\text{lt}} \mathbf{3}\_{\text{lt}-1} + \sum \mathbf{I} \mathbf{nd} + \sum \mathbf{Year} + \mathbf{c}\_{\text{ft}} \end{array} (1)$$

As described above, overborrowing (ΔBL) corresponds to extra bank loans the firm actually obtains that exceeds the estimation value. Referring to the measurement of overinvestment [24,25], we measure firm-level overborrowing as the statistically significant discrepancy between the actual bank loan the firm obtains and the estimated bank loan calculated by the model (1).

Secondly, we employ the following model to test the factors affecting overborrowing:

Overloanit = η<sup>0</sup> + η1Stateit + η2Politicit + η3Occupyit + η4Marketit + η5Financebackit + ∑Ind + ∑Year + εit (2)

Thirdly, to verify the mediating effect of overborrowing on ownership type and R&D expenditure, this study sets up the following regression model:

#### R&D\_ratioit = σ<sup>0</sup> + σ1Stateit + σ2Overloanit + σ3Governsciit + +σ4Stock\_incentiveit + σ5Independentit+ σ6Assignit + σ7Herfindahl\_3it + ∑ Ind + ∑ year + εit (3)

Here, R&D \_ratio denotes the R&D spending intensity. A number of researchers have used R&D spending divided by firm sales as a measure of R&D intensity [91,92]. The innovation intensity indicator has advantages in the consistency of dimensions and is directly related to a firm's financing, so it is suitable to reflect a firm's innovation performance. To ensure the stability of this indicator, we also use R&D expenditure scaled by the total assets as an alternative measure for a firm's innovation performance in the robustness test section. All missing values for R&D intensity are replaced with zero, and the upper bound for R&D intensity is set at 1.

From the standpoint of corporate governance, Li et al. (2019) pointed out that the key feature of current Chinese SOEs governance is the coexistence of administrative governance and economic governance [93]. The administrative governance factors, such as implicit as well as explicit guarantees for SOEs' financial crisis and the prevalence of soft-budget, are thought to be the main causes of firms' misconduct in investment [24,94]. Therefore, many state-owned banks can hardly treat firms equally in loan granting, but show some degree of discrimination. Previous studies have shown that SOEs generally obtain more loans than non-SOEs, and the gap increases significantly during periods of austerity and recession [95]. The state ownership type (State) is a dummy variable where one represents those firms ultimately owned by the government, and zero represents those not. Following Petersen and Rajan (1994), La Porta et al. (1999), Faccio and Lang (2002), Anderson et al. (2004), Bigelli and Mengoli (2004), Armstrong and Vashishtha (2012), Ben-Nasr et al. (2015), and Grosman et al. (2016) [96–103], we control for main financial and governance variables in our analysis. The main variables involved in the above models are defined in Table 1.

**Table 1.** Variables and Definitions in Models (1) to (3).



**Table 1.** *Cont.*

\* According to the China Securities Regulatory Commission "China-listed companies Industry Classification guidelines (2012)", we exclude the financial industry and divide the manufacturing industry into two-digit industry categories. In addition, other industries are subdivided into single-digit industry categories.

### **4. Results**

#### *4.1. Descriptive Statistics*

Table A1 shows the descriptive statistics of the main variables. All continuous variables are winsorized by 1% to mitigate the impact of outliers. Before examining the overborrowing of the sample firms, it would be useful to generally discuss the financial structure of firms in China. The leverage ratio of Chinese firms is higher than reported by Fan et al. (2008) [49], which indicates there is no fundamental change in the financial structure during the past decade. Moreover, the average percentage of long-term bank loans to total assets of listed companies is 0.04 and its standard deviation is 0.08. The average proportion of long-term loans to total assets for SOEs is 0.07, whereas the proportion for non-SOEs is only 0.03, significantly lower than SOEs. The average overborrowing for listed companies is 0.05, accounting for 56% of the average loan for the same firms with overborrowing (whose average bank loan of total assets is 0.09), indicating that the overborrowing for listed companies in China is significant. In the table, about 19% of all firms have political connections. This indicator is lower in SOEs, thus showing that non-SOEs in China are more inclined to seek political connections, giving evidence that more efficient firms are more likely to build connections to secure their access to finance. The average administrative level of the chairman or general manager is 76.6. Furthermore, the average administrative level for SOEs is much higher. In summary, it provides consistent evidence that the ban on officials' holding concurrent posts in firms issued by the government in 2014 has not been strictly enforced with lower administrative level officials. Table A2 shows that the average overborrowing levels from 2012 to 2018 are 0.0547, 0.0547, 0.0497, 0.0481, 0.0475, 0.0463 and 0.0441, respectively, which means that the overall overborrowing for Chinese listed companies shows a gradual downward trend. In addition, the average percentage of overborrowing for SOEs is 0.07, while it is 0.04 for non-SOEs, which indicates

that SOEs' overborrowing is significantly higher than that of non-SOEs. These results lend additional support to our conjecture that overborrowing is associated with ownership attributes and closer with lower political connections relative to firms' economic factors. Moreover, multicollinearity was diagnosed by examining the variance inflation factors (VIFs) for the predictors (See Table A3). The VIF values for the predictors, all substantially lower than the rule-of-thumb cutoff of 10 [104], revealed that multicollinearity is not a problem in this study.

### *4.2. Regression Analysis*

The empirical test of this study comprises three steps. The first step is to calculate the annual overborrowing level for each firm by the residual of Model (1). The result of the regression analyses regarding the bank loans is shown in Table 2. Regarding the explanatory capability of model (1), the chi-squared test of fitting degree is 0.402, which indicates a suitable model specification. In terms of the impact of independent variables, the level of share concentration (Herfindahl\_3) is positively related to a firm's long-term bank loan (Loan). In other words, instead of a diversified ownership structure, high ownership concentration works for firms to obtain bank loans under the administrative-economic governance mode. Of course, this does not necessarily mean that the mixed ownership reform associated with ownership decentralization is ineffective.


**Table 2.** Regression results of Model (1).

**Notes:** Standard errors are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05.

The long-term trend (Figure 1) suggests that Chinese firms gradually improve the degree of diversification in ownership structure over the period of 2012–2018. Evidence also indicates that ownership decentralization does contribute to mitigating the potential expropriation by large investors of other investors and stakeholders in the firm [102]. However, existing large investors of SOEs can still expropriate substantial gains from the firms, resulting in severe agency costs. In addition, the coefficient for profitability and a firm's long-term bank loan is positive but not significantly (coefficient = 0.002); the growth capability is negatively related to the long-term bank loans of firms also. We can thus

deduce that banks have not fully considered the profitability and growth of firms in their loan-granting decisions.

**Figure 1.** Ownership Structure of SOEs and non-SOEs (China, 2012–2018).

Secondly, according to econometric theory, the residual in Model (1) represents the long-term bank loan a firm obtains that cannot be explained by common economic factors. Therefore, it can be regarded as a measure of a firm's overborrowing. Considering the nature of overborrowing, only when the residual is greater than zero does firm overborrowing exist. Otherwise there is no overborrowing and Overloan is converted to zero correspondingly. Thirdly, the causal steps approach [105] is adopted to verify the mediating effect of overborrowing.

In the context of the current administrative-economic governance of state-owned commercial banks, firms may easily obtain access to a higher level of overborrowing owing to connections arising from being controlled by the same governmental stockholder, thus shrinking their innovation expenditure. If these four conditions described by Baron and Kenny (1986) are met, we can conclude that a mediation effect occurs. Additionally, we use the Sobel (1982) test to test the indirect effects of overborrowing on firm R&D intensity [106]. The Sobel test of significance assumes that the indirect effect of the independent variable is normally distributed, an assumption that may make this a conservative test [107]. The indirect effect is signified to be significant when the Sobel test Z value is significant (>1.96 or <−1.96) [108].

Hypothesis 1 suggests overborrowing mediates the relationship between state ownership and firm R&D expenditure. For the specification of the mediation link, we follow Baron and Kenny's (1986) procedure and find that all four steps are fulfilled (Table 3). A mediation effect exists if the coefficient of the direct path between the independent variable (state ownership) and the dependent variable (firm R&D intensity) is reduced when the indirect path via the mediator (overborrowing) is introduced in the model. In Step 4, the coefficient of state ownership is reduced and still significantly negative at the 1% level, suggesting a partial mediation role of overborrowing on the firm R&D intensity. The indirect effect (η1\*σ2) the mediating variable overborrowing assumes approximately accounts for 5.12% of the inherent innovation efficiency loss. It means overborrowing essentially weakens the innovation capability of Chinese SOEs and would eventually defer their sustainable innovation. The results of the Sobel test in Table 3 also provide significant evidence of the existence of an indirect effect (as the Sobel Z values are significant: Z < −1.96) for the above model.


**Table 3.** The causal steps approach of overborrowing's mediating effect.

**Notes:** Standard errors are in parentheses. \*\*\* *p* < 0.01.

#### *4.3. Political Connections Moderating Mechanism*

The specific empirical proxy for the political connection level is the administrative order of politically connected managers in the firm. By referring to past studies on firm political connections and based on Chinese administrative-economic governance characteristics [60,109], the chairman or CEO in listed companies is usually nominated by the larger shareholders and thus has relatively bigger decision-making power on the board, making their position suitable for testing the speculation of firm political connections. The information related to this variable was collected from the CSMAR database. The administrative order divides into eighteen categories according to the administrative level classification standard of CSMAR (See Appendix B).

The political connections mechanism test is carried out in the form of group regression. First, the samples are encoded into two groups according to the administrative level of the firm's political connection. If a firm's chairman or general manager has a political connection at or above the provincial level, it is classified as a "high" administrative level firm; otherwise, it is classified as a "low" level one. Then, the above causal steps approach regression is repeated for each group, respectively. From the regression results in Table 4, overborrowing only passes the mediating effect test in the low political connection level group. In other words, while the firm is at a lower political level, which corresponds to much stronger promotion incentives and weaker administrative supervision, the local government is more inclined to intervene in the firm's access to bank loans through its impact on local commercial banks.


**Notes:** Standard errors are in parentheses.

 \*\*\* *p* < 0.01, \*\* *p* < 0.05.

To further test the equality of coefficients between two linear regressions across groups, the Chow Test [110] is performed. The Chow Test is to determine whether changes occurred between two regressions. More specifically, step 1 and step 2 are re-regressed with an inclusion of the interaction of independent variable State and moderating variable Dummy\_pc (Dummy\_pc equals to one if a firm's chairman or general manager has a political connection at or above the provincial level; otherwise, it equals to zero). Step 3 is re-regressed with an inclusion of the interaction of mediating variable Overloan and moderating variable Dummy\_pc. Step 4 is re-regressed with an inclusion of the above two interactions. If the test determines that the coefficients of the interaction terms are significantly not equal to zero, this means there is significant evidence that a heterogeneous effect exists above and below that administrative break point. Observably, based on the interaction model, Table 5 Step 2 shows a significant effect of Dummy\_pc on the relationship between State and Overloan (the coefficient of interaction Dummy\_pc ∗ State is negative and statistically significant at the 1% level). Thus, SOEs with a higher level of political connections tend to restrain the overall level of overborrowing. In particular, Table 5 Step 3 shows a higher level of political connection inversely moderates the negative effect of overborrowing on R&D expenditure. In other words, a positive political connection's moderating role is mainly reflected in the path of state ownership (X) to firm overborrowing (M).


**Table 5.** Chow test of political connections' moderating role.

**Notes:** Standard errors are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05.
