Since rating agencies sequence the corporate credit ratings in order, we use sequencing logic model (Ordered Logit) in our empirical analysis. Based on previous research on credit rating (Elton et al., 2004; Perraudin and Taylor, 2004) [10] [11] , our model to test Hypotheses 1 is as follows:

When credit rating agencies are rating a corporate’s credit, they also give a rating of the bond issued by the corporate. But in the practice of credit rating in China, the rating standards of short-term financing bond are different from other types of bond, thus research often uses the corporate credit rating directly, namely the credit rating for issuing corporates. This article also directly uses the corporate credit rating, and in order to make the results more robust, we also use the bond rating¹ as the robustness test by eliminating the sample of short-term financing bond.

Credit means the corporate credit rating, because the sample of this article is listed company, the rating data does not cover all areas. The corporate credit rating in this article has 6 grades, which are AAA, AA+, AA, AA−, A+ and A. AAA is the best grade. In order to do regression analysis, we assign AAA with value of 6, AA+ with value of 5, etc.².

DisQuality means disclosure quality of corporate information. There are many ways to measure the disclosure quality: Botosan (1997) [12] measured it by categorizing and evaluating five aspects of voluntary disclosed information which are useful for investors to make decisions, including background information, historical information, the key non-financial information, forecasting information and management information. Kim and Verrecchia (2001) [13] built an information disclosure quality model by using the corporate bond yielding rate and trading volume, this model can help avoid problems caused by corporate earnings manipulation. Francis et al. (2003) [14] used the listed companies’ accounting earnings quality to evaluate the quality of information disclosure. Chinese researchers mostly use the information disclosure quality evaluation results published by Shenzhen Stock Exchange since 2001 (Zeng et al., 2006; Wang et al., 2008) [15] [16] . The Shenzhen Stock Exchange’s evaluation results have 4 grades³, which are “excellent”, “good”, “qualified” and “fail” (in some years also represented as ABCD). The main content of the Shenzhen Stock Exchange evaluation includes: 1) The authenticity, accuracy, completeness, timeliness, legal compliance and fairness of the information disclosure of listed companies; 2) Penalties, sanctions and other regulatory measures for listed companies case; 3) Listed companies’ willing to cooperate with the Shenzhen Stock Exchange; 4) Management of the listed companies’ information disclosure affairs; 5) Other situations defined by Shenzhen Stock Exchange. As the Shenzhen Stock Exchange’s data is quite reliable, we also use it to proxy the disclosure quality. On the other hand, considering the Shenzhen Stock Exchange’s data is simply divided into four levels, it does not distinguish between voluntary disclosure and mandatory disclosure, and the evaluation index system is also not announced. More importantly, the data doesn’t cover companies listed on the Shanghai Stock Exchange, so there might be some problems. In order to make the results more robust, we also use the Chinese listed companies information disclosure index (Gao Minghua, Beijing Normal University, 2010, 2012) [17] [18] as the indicator in the robustness test. Gao’s index measured the integrity of the information disclosure (including mandatory and voluntary), although it only has two years data, it covered more listed companies.

According to existing research results, we choose accounting information quality, basic business financial condition, macroeconomic, corporate nature and bond characteristic as control variables of our model.

AccQuality is the variable for the accounting information quality. De Angelo (1981) [19] found a positive correlation between the audit agency’s size and audit quality, as small audit agencies often fail to maintain their independence. Ball and Shivakumar (2005) [20] , Zhang (2008) [21] used accounting conservatism to measure the accounting information quality. Earnings volatility is also commonly used to measure the quality of accounting information. Schipper and Vincent’s (2003) [22] study shows that the volatility of corporate earnings will affect the quality of accounting information. Therefore, according to previous studies, we use audit quality, accounting conservatism and earnings volatility to characterize the quality of accounting information. Big4 (audit quality) is a dummy variable, equals 1 when the company accepts the “Big4” audit. Conserv (accounting conservatism) is calculated based on the cumulative accruals, the higher cumulative accruals, the lower conservatism, the lower accounting information quality. VarRoe (earnings volatility) is measured by the past three years Roe variance, the greater the variance, the greater the earnings volatility, the lower the accounting information quality.

Credit rating can be largely influenced by the corporate financial condition, so we control some relevant micro-financial variables in the model. Roe is return on equity, equals net profit/average shareholders’ equity. CFO is corporate cash flow, equals operating cash flow/total assets. Lev is the asset-liability ratio, equals total liabilities/total assets. Size is the size of scale, equals to the natural logarithm of total assets. Some studies show that domestic macroeconomic development will have a significant impact on the corporate credit rating. Therefore, this article controls the macro-economic factor GDPgrow which is the real GDP growth of the year. In addition, for examining the differences between SOEs and non-SOEs, we use the State variable, a dummy variable that equals 1 when the company is state owned.

Some bond features⁴ such as bond duration and guarantee will also affect the credit risk; we use Duration and Guarantee variables. Guarantee is a dummy variable which equals 1 if there’s guarantee. This article includes a sample of short-term financing bonds, medium-term notes, enterprise bonds and corporate bonds. Type is the bond type dummy variables (including Type S, Type M, Type E, Type C), Type S equals 1 means it is short- term financing bonds; TypeM equals 1 means it is medium-term notes; Type E equals 1 means it is enterprise bonds; Type C equals 1 means it is corporate bonds. In order to reduce the endogenous problem between the dependent and independent variables, the independent variables in this article will all use the previous year data of the bond issuing.

With bond interest rates issued at different times can be affected by economic environment of that time, and the rates also diversify from bond duration due to liquidity difference. Therefore, drawing Zhang (2008) [21] and Wittenberg-Moerman’s (2008) [23] measurement of debt financing cost, our model⁵ to test Hypotheses 2 is as follows:

Spread means bond financing cost, calculated by the bond interest rate minus the bank lending rate of the same period. The same period means the same time and same duration. Other variables are the same with previous.

Bond-related data in this article comes from the Wind financial terminal database, the time period is from 2007 to 2013, including short-term financing bonds, medium-term notes, enterprise bonds, corporate bonds. Due to the current data is based on the listed companies; we eliminate the non-listed companies from the bond-related data⁶. The GDP data is collected from the China Statistical Yearbook, the bank lending and deposit rates are collected from the People’s Bank of China website. We also eliminate the sample from financial sector. We winsorized all the continuous financial data at 1% level to avoid outliers. The sample selection process is shown in the Table 1.

Table 2 shows the descriptive statistics⁷. Quality of information disclosure’s (DisSz, Shenzhen Stock Exchange score) mean is 3.0932, median is 3, the minimum value is 2, indicating information disclosure quality of listed companies which issued bonds in the bond market is higher than the overall average. The interest rate spread’s (Spread) mean is −0.3215, median is −0.5, showing bond financing cost is lower than bank loan. The mean of bond issuing corporate credit rating (Credit) is 3.9492, and the median is 4, showing the average credit rating in the sample is AA. Rating is the bond rating (due to various rating standards, this data does not include short-term bond), its mean is 3.4358, median is 3, indicating the average bond rating is AA.

AccQuality is characterized by three variables: Big 4, Conserv, VarRoe. We noticed only 11.30% companies in the sample choose Big 4 to do the financial report audit. Conserv has a mean with 0.0420 and median with −0.0152, which means the reports are robust in general, but the 0.2601 deviation shows huge differences among companies. VarRoe, earnings volatility, is relatively small; its standard deviation also shows big differences. We can also see big difference among samples in Roe, Size, Lev, CFO, CR, QR, and LLR. About the company type, the 48.02% of the samples were state-owned. From the bond features, 23.73% of the sample has guarantee. The average bond duration is 3.9442 years, the median is 0.4932, and standard deviation is 2.2444, which shows huge differences in bond durations.

Table 3 and Figure 1 show the average information disclosure quality and spread in different corporate credit rating grades. As the quality data is collected from the previous year of the rating year, we can see the higher information disclosure quality, the higher corporate credit rating, which is in line with the Hypothesis 1.

With the increase of the corporate credit rating, the spread becomes lower, indicating the higher information disclosure quality, the lower bond financial cost, which supports the Hypothesis 2. In order to verify the results from multiple angles, the article also examines the average information disclosure quality and spread in different bond credit rating grades, as shown in Table 4 and Figure 2. The results are the same.