Guaranteed Corporate Bonds

Transcription

1 Guaranteed Corporate Bonds Fang Chen, Jing-Zhi Huang, Zhenzhen Sun, Tong Yu March 2014 Chen is at the College of Business, University of New Haven. Huang is at the Smeal College of Business, Pennsylvania State University. Sun is at the School of Business, Siena College. Yu is at the College of Business Administration, University of Rhode Island. s: and All errors are our own. Comments are welcome.

2 Guaranteed Corporate Bonds Abstract This paper investigates the use of guarantee by corporate bond issuers. Guarantees provide a protection to corporate bondholders to cover their losses in the event of default, and have been widely used in the corporate bond market since late 90s. However, this important financing method has received very little attention in the academic literature. Using a large sample of new issues of corporate bonds over the period , we present empirical evidence that guarantees indeed enhance bond credit quality. Specifically, given the same issuer rating, the average rating of bonds issued with guarantees significantly exceed that of bonds not issued with guarantees. Further, this positive impact of guarantees on bond rating concentrates on issuers with BB or BBB ratings. Yet, guarantees do not have a significant impact on bond yields. We also find that firms using guarantees tend to be more financial constrained, subject to more severe agency problem, and have lower firm ratings and higher default probabilities, in comparison to non-guaranteed bond issuers.

3 1 Introduction It is known that many corporate debt covenants are used to protect bondholders from default risk. One recent development in the corporate bond market is the use of guarantees provided by firms themselves when they issue corporate bonds, a new kind of covenants introduced in mid 1990s. For instance, based on the Mergent Corporate Bond Database, nearly 40 percent of corporate bonds (in terms of issuance amount) issued in 2009 are embedded with guarantees. guarantees provide investors a protection through an internal or a thirdparty arrangement which secures the payments when issuers are at default. Owing to its significant role in the financial market during the crisis, the use of guarantees, also known as a common type of credit enhancements, catches a wide attention among academia and practitioners (e.g., Demyanyk and Hemert, 2011; Dou, Liu, Richardson, and Vyas, 2012). However, the spotlight so far has been on the packaging of credit enhancements with commercial loans, mortgage- and asset-backed security sector, and municipal bonds. 1 The use of guarantees in the corporate bond sector has received very little attention in the literature. Using a large sample of corporate bond issuance from 1993 through 2012, we empirically analyze the impacts and determinants of the use of guarantees. Based on our finding, there is a striking firm rating difference between bond issuers using and not using guarantees roughly two thirds of guarantee users have either high speculative ratings (i.e., S&P s BB+, BB, and BB- ratings) or low investment grade grades (S&P s BBB+, BBB, BBB-) while two thirds of issuers not using guarantees receive median (A+, A, A-) or low investment grades. This is consistent with the conventional wisdom that bonds with relatively poor credit worth may be inclined to issue securities with guarantees. Investors of guaranteed bonds are protected due to the presence of the third party to make payments when the issuers default. Such an institutional arrangement of guaranteed bonds gives rise to the first set of questions of the study: do guarantees help to improve 1 On the applications in mortgage backed securities, see Adelson (2003), Ashcraft and Santos, 2009; Griffin and Tang (2012); Arora, Gandhi and Longstaff (2012). For the applications in municipal bonds, see Braswell, Nosari and Browning (1982), Kidwell, Sorensen and Wachowicz (1987), Nanda and Singh (2004). 1

4 the ratings of bond issues? and similarly, do guarantees help reduce bond yields? To address these questions, we perform regression analysis on bond ratings and yields at issuance for a sample including bonds with and without guarantees. In the rating regression, the dependent variable of the regression is the bond rating at issuance. The key explanatory variables is the interaction between a dummy indicating whether the issue is a guaranteed bond and the issuer s rating. The setup allows us to compare the average rating of bonds packed with guarantees with the rating of non guaranteed bonds holding issuers ratings constant. Following the literature (e.g., Ashcraft, Goldsmith-Pinkham, and Vickery, 2009), we additionally control for firm size, tangible asset, debt level and bond characteristics. The result reveals a positive role of guarantees the average rating of guaranteed bonds is 10% higher than that of non-guaranteed bonds in terms of the numerical expression of rating. The further examination shows that the most significant rating enhancement occurs within the firms whose ratings are close to the default, and the second significant effect concentrates on the firms that have ratings close to the investment and junk grade cut-off. Compared with the credit ratings of non-guaranteed bond issuers, guaranteed bond issuers on average are poorer in their corporate financial ratings. One may expect the improved ratings of guaranteed bonds result in lower yields of these bonds. However, this is not true based on our comparison analysis that compares the initial yields between guaranteed and non-guaranteed bonds. Different from the result of rating regressions, we find that guarantee doesn t have any significant impact on yield in general. We further dig into this issue by analyzing the guarantee effect on bond yields conditioned on corporate financial ratings. We interact a bond s guarantee use and the issuer s rating and see how the variable affects bond yields. The result shows guarantee generally has no significant impact on bond yields except when the firm rating is BBB-, a rating at the edge of investment grade. Even worse, when firm ratings are at CC (a default rating) or AAA (the highest credit rating), the yields of guaranteed bonds are higher than yields of nonguaranteed. Investors, or more precisely underwriters, do not take the guarantee as positive signal of the bond s credit prospect. 2

5 Why just a subset of firms using guaranteed bonds? This is our second question. It is known that guarantees are arranged internally either subsidiaries (the super majority case) or parent companies provide guarantee to bond investors. Such an arrangement gives rise to an easily conceived reason for the restrained use of guarantees: non-guaranteed bonds may be issued by firms not having any subsidiary or not associated with a parent firm. This conjecture is however quickly dismissed in our sample, roughly 85 percent of firms have a parent or subsidiaries. An alternative explanation is that financing cost of guaranteed bonds could be high. One must recognize that the direct guarantee fee is trivial given that guarantees are provided internally. However, guarantees occupy the resources of guarantors in two ways. First, guarantees provided by guarantees typically are unconditional. The secured guarantee use various collateral provided by the guarantors. Moreover, the covenants of debt with guarantees normally have restrictions on the guarantors. The covenants may restrict the guarantors to pay dividends, making loans or transferring any of their property to the issuers or between subsidiary guarantors. The third question of the study is what factors drive corporate use of guarantees. We hypothesize that the primary force to be financial constraints faced by corporates. Based on Myers and Majluf (1984) s pecking order theory, internal capital is least costly and a firm has to raise external capital, debt financing is preferred relative to equity financing as debt financing is less expensive. Financial constraints, however, hurt a firm s capacity to raise debt from investors (e.g., Kaplan and Zingales, 1997, Lamont, Polk and Saa-Requejo, 2001, and Baker, Stein and Wurgler, 2002). This will lower firms ability to achieve their optimal investments. guarantees, which offer a guarantee to bondholders when the firm cannot afford their debt payments, improve corporate credit worthiness thus enable these firms to raise debt. Corporates do not have the equal access to investment opportunities. For firms having poor investments opportunities, the purpose of using guarantees could lie on the managerial agency problems. Jensen (1986) argued that the managers may increase the size of the firms for their own perquisite, i.e., the management s empire-building tendency. For instance, 3

6 a firm may adopt a new project or acquire another regardless of the return (Stulz, 1990; Titman, Wei, and Xie, 2004; Cooper, Gulen, and Schill, 2008). As noted in the literature, a consequence of the agency conflict (e.g., in the firms with high free cash flow and low growth opportunities) is the overinvestment (e.g., Opler and Titman, 1993). Aligned with the agency conflict explanation, the use of the subsidiaries/parents guarantee may be a consequence of the empire-building incentive of issuers rather than investing in positive net present value (NPV) investment opportunities. Issuers use the resource of their subsidiaries/parents to achieve this goal. Under this explanation, firms with poorly aligned interest between management and shareholders are more likely to use guarantees. We perform logistic regressions to look into the determinants of bond issuers guarantee uses. Regarding the financial constraints theory, firm ratings are negatively associated with the use of guarantees. Our finding also reveals that firms with less collateral (proxied by cash and tangible assets) have a higher chance to use guarantees. The results shows a strong evidence that firms with financial constraints tend to use guarantee to increase debt capacity. Regarding the influence of agency problems on the use of guarantees, we find that high degree of agency problem (as proxied by high free cash cash/low growth opportunities) are positively associated with the use of guarantees. When we use different proxies for of growth opportunities, we find consistent results. The finding supports the idea that the use of subsidiaries do not align with the growth opportunities of the parents. Instead, the use of guarantees may be a consequence of empire-building of the parents or corporate socialism among the subsidiaries. Moreover, since overinvestment is more likely to occur in firms with low grow opportunities (e.g., Jensen, 1986; Lang and Litzenberger, 1989), the positive relation between the guarantee and high free cash cash/low growth opportunities implies an overinvestment problem in firms using guarantees. It is interesting to note that some issuers alternatively use guaranteed and non-guaranteed bonds. For the purpose of shedding light on this cause, we separate the cases that firms mix their guarantee uses and that firms persistently use guarantees and run a multinomial regression. The dependent variable is a categorical variables taking three values: it is 0 if 4

7 the firm issues non-guaranteed bonds only; it is 1 if the firm issues a mix of non-guaranteed bonds and guaranteed bonds; and it is 2 if the firm issues guaranteed bonds only. We find that the lower firm rating, higher agency problems, less cash, less tangible, larger size, and less debt increase the odds of issuing a mix of non-guaranteed and guaranteed bonds than that of issuing guaranteed bonds only. Finally, as a robustness check on the agency problems explanation, we conduct subsample analysis. Chang and Mayers (1992) and Masulis, Wang and Xie (2009) find that managerial voting power is positively associated with the agency problem. Based on this finding, we partition the sample into three even-sized groups based on managerial voting power. Dummy variable highmgtvote equal one if the issuer is in the highest managerial voting rights group and zero otherwise. The intuition with higher managerial voting rights indicates a weaker corporate governance and therefore stronger agency problems. We are likely to observe a stronger tendency of guarantee uses in the high managerial voting rights group than that in the low managerial voting rights group. We run the logistic regression on the use of guarantee. The coefficient of highmgtvote is positive and significant. The results of the subsamples regression yield further supports to the hypothesis that firms with more agency problem are more likely to use guarantees. The remainder of the paper is organized as follows. Section 2 provides the institutional background of guarantees used in the corporate bond market. In Section 3, we introduce the hypotheses. Section 4 describes our sample and data used in the sample. Section 5 presents empirical findings. Section 6 concludes. 5

8 2 Institutional Background on Credit Guarantee for Corporate Bonds Being a potential means to alleviate credit risk of a corporate bond, in recent years a significant proportion of corporate bonds are issued with guaranties. 2. The ratio of corporate bonds with guaranties increased from 2% in 1990 to 26% in 2010 in terms of dollar value. Overall, corporate bonds have been sold for USD16,711 billion (par value) during the period of 1993 to Of bonds issued in the same period, 17% (USD2,853 billion), by dollar value, were issued with guaranties. Guaranties protect bondholders from defaults through providing the guarantee to the payment of principal and interest payments of the underlying bonds in the event of issuer defaults. Based on the Mergent database, firms employ three major types of credit enhancements: i) guarantee, ii) bond insurance, and iii) letter of credit (LOC). They respectively account for 96%, 3% and 1% of the total credit enhancement over the entire sample period from 1993 to The distinction across a guarantee, bond insurance, and letter of credit are obvious. Parent/subsidiary guarantee is an internal arrangement which requires a filing to SEC, while there is no such requirement for insurance or letter of credits which is treated as an external contract. Because basically all bond insurances have an AAA rating, it is common for the insured bonds to have an AAA rating. By contrast, bonds with parent/subsidiary guarantee have their ratings varying from AAA as the highest to D as the lowest. We focus on guaranties in this analysis because the super majority of guaranties are provided through corporate guaranties (96% among guaranties). There are several distinct features associated with corporate guaranties. First, guaranties are typically arranged internally where guarantors are either parent companies or subsidiaries. In our sample, the majority of the guarantors (90%) in our sample are the subsidiaries of the issuers. When the parent debt is guaranteed, the full consolidated financial statements of the parent company (together with the subsidiaries) are used in the rating 2 A recent suvey shows that the majority of over 1,100 risk managers from major international corproates consider credit risk as one of the most important risk exposures (Bodnar, et al.,

9 determination. See e.g., DBRS Criteria (2010) for the reference. For instance, MGM Mirage issued a bond using all its domestic subsidiaries as the guarantors. The aggregate par value of the bond issuance is USD225,000,000. Second, internally arranged parent/subsidiary guaranties typically involve a low cost. An explicit expense of the transaction is the cost of filing to Securities and Exchange Commissions (SEC), which is not materially high. An additional cost is the guarantee cost. Since it is considered as an arm s-length transaction, subsidiary guarantee fee is typically nominal and seldom paid by parent companies or subsidiaries. Third, there are indirect costs associated with guarantee uses. A parent firm and its subsidiaries are distinct legal entities. Without guarantee, if the subsidiary is at default, its debtholders have not recourse to the parent unless the parent is involved in some wrong-doing (Thomson, 1991). However, with a guarantee from the parent, the subsidiary debt holders have recourse to its parent guarantors should the subsidiary default. In practice, most of the guaranties to the public issuers were conducted jointly by most or all its domestic subsidiaries. By law, the guaranties are senior obligation of the guarantor, ranked equally with all other existing and future senior debt of the guarantors in right of payment. It is also a common practice to contain the covenants in the indenture to limit the payments of the subsidiaries or parent companies (whoever the guarantor is), such as dividend payout, shares repurchase or making early principle payment prior to the schedule, among other arrangements. bonds. Consequently, this arrangement helps to reduce credit risk of guaranteed Fourth, it is possible for a subsidiary to achieve a higher rating than that of a parent or a consolidated family if it is insulated. Generally, the rating of a subsidiary can exceed that of the parent firm by up to three notches (Standard & Poor s, 2000). 3 When credit rating of the guarantor is downgraded, the rating agency will reevaluate the bond and adjust the rating. Finally, there is no evidence that the use of guarantee lowers the yields to maturity at 3 Insolation means parent companies may be prevented or restricted from accessing the resources of the subsidiaries. For specific insulation factors, see Standard & Poor s (2000). 7

10 bond issuance. For example, the Teck Resources Ltd. issued six corporate bonds in 2011 and 2012, three with guaranties (issued in 2012) and three without guaranties (issued in 2011). The yield spreads (the difference in the yields of the corporate bond and the treasury bonds with corresponding maturities) of guaranteed bonds are respectively 195bps, 235bps, 285bps while those with guaranties are 150bps, 165bps, and 190bps. 4 3 Hypotheses We propose the hypotheses in the section. The first empirical question is how guaranties affect bond ratings at issuance? As noted in Section 2, guaranteed bondholders have recourse to the guarantors in the event of the issuers default. As a result, it is likely that guarantee reduces the credit risk of the guaranteed bonds. We postulate the following hypothesis. H1. Guarantee has an positive impact on bond rating. Retaining investment grades is a critical condition for bond issues. There is a significant drop in bond yields when bond ratings rise from non-investment grades to investment grades. Institutional investors typically are subject to regulatory scrutiny when investing in junk bonds (see, e.g, Ellul, Jotikasthira, and Lundblad, 2011). The higher possibility of staying at investment grade or increasing from junk to investment grade, the more potential benefit from guarantee. This reasoning leads to the following hypothesis: H1a. The positive rating effect of guarantee is highest when firm ratings are closer to the cutoffs of investment and noninvestment grades. Firms with financial constraints have difficulty of accessing the capital because of their 4 This does not necessarily indicate that yield spreads of guaranteed bonds are higher as in the example here all guaranteed bonds are issued in 2012, a period that the average yield spread is greater than that of

11 low creditworthiness. These firms either cannot attract enough investors or have to pay a high price. Credit ratings is an important factor in the requirement by several regulations on financial institutions and other intermediaries investments in bonds. For example, regulations restrict banks from investment in speculative-grade bonds since 1936 (e.g., Partnoy, 1999; West, 1973). In 1989, savings and loans were required to completely liquidate their speculative-grade bonds by 1994 (e.g., Kisgen, 2006). Finally, pension fund guidelines often prevent bond investments from speculative-grade bonds (e.g., Boot, Milbourn, and Schmeits, 2003). Since the main benefit of guarantee is to increase the creditworthiness of debt to lower the cost, firms with financial constraints can benefit from the creditworthiness increase, regardless their heterogeneous characteristics. However, firm specific value like rating, size, dividend, collateral and rating reflect the firm s financial constraint situation and have been widely used as the proxy for financial constraints. In general firms with financial constraints are more motivated to use guarantee to reduce the financing cost. We therefore postulate the following hypothesis. H2. Firms with more financial constraints are more likely to use guarantee. The puzzle is what drives these firms with financial constraints to use guarantee to increase their rating and thus debt capacity? Does guarantee align with the best interest of the parent firms? To answer these questions, we offer two possible explanations. The first possible reason is the agency problem. If agency problem is strong, then the firm may take negative NPV project or forfeit positive NPV project. Our paper is on debt issuance, thus the focus is on the former type of agency problem which leads to overinvestment problem. Tobin s Q (the ratio of the market value of assets to the current replacement cost of those assets) has been used in the literature (e.g. Lang and Litzenberger (1989)) as a proxy for investment opportunities, with the notion that the managers of a firm with poor investment opportunities ( low Q) are likely to overinvest or waste their stockholders cash. Using Q as 9

12 a proxy for overinvestment, Lang and Litzenberger (1989) find that returns around dividend change announcements are significantly more positive for firms with Q less than one than for firms with Q greater than one. Investment may be the consequence of the empire-building of parents rather than following the investment opportunities (e.g., Jensen, 1986). Moreover, in the specific parent-subsidiary corporate structure, the agency problem may act through another channel. Scharfstein and Stein (2000) argue that a two-tier agency problem, stemming from misaligned incentives at parents and at divisions, is necessary for corporate socialism in internal capital allocation. Kolasinski (2009) finds the main reason for subsidiary to issue debt is protecting themselves from the corporate socialism and poaching problems in internal capital markets. On other hand, parents use of the subsidiary guarantee may be a consequence of empire-building preference of parents rather than following the investment opportunities (e.g., Jensen, 1986). In both cases, the agency problem of the parent spurs the use of guarantee regardless the low growth opportunities. We therefore postulate the following hypothesis: H3. Firms with more agency problem are more likely to use guarantee. Another possible reason of guarantee use is the tax benefit from increased debt level. An optimal debt structure is to max the marginal tax benefit. While firms are near their debt capacity but do not reach their desired debt structure, they can use external resources like subsidiary guarantee to expand the debt capacity. Therefore, when examining a firm s motivation of using guarantee, it is important to consider not only the firm s ability to access the capital, but also its preferred degree of debt capacity. The more the marginal tax benefit from increased debt capacity, the more likely the firm will use guarantee. This reasoning leads to the following empirical hypothesis: 10

13 4 Sample and Data 4.1 Sample Our bond data comes from the Mergent Fixed Income Securities Database (FISD). FISD provides the information of public bonds by both public and private issuers. We select U.S. corporate bonds including US corporate debenture, US corporate MTN, asset-backed security and other US corporate bonds. We exclude the US corporate convertible bonds and preferred stocks. FISD shows whether the corporate bonds are issued with one of the credit enhancements: guarantee, letter of credit (LOC) or insurance. Our data sample period is from 1993 to In this period, there are US corporate bonds issued with one of the three credit enhancements, including guarantees, 466 insurance, 3 LOC. We further identify the guarantors of the guarantees. FISD provides the guarantor information in two ways. First, FISD directly lists most guarantors as Subsidiaries. Second, FISD list the parent of the issuers. If the parent id matches the guarantor id, the bond is guaranteed by the parent of the issuer. Among guarantees, we find 7196 parent guarantors, 3702 subsidiaries guarantor besides 3509 unknown guarantors. The paper investigates the public issuers only. We limit the data to the issuers that are listed at the time of bond issuance by matching the CUSIP of the issuers in FISD with that in CRSP. Also for the rest of the unknown guarantors, we manually search their SEC filings. The guarantors information are disclosed in the 424B, S-4, 8-k, 10-Q or 10-K. From 1993 to 2012, there are 2278 guaranteed bonds issued by public firms, including 2012 bonds with subsidiaries guarantees, 130 bonds with parents guarantees and 144 bonds with insurance. For the purpose of comparison, the sample includes both bonds with guarantee and those without guarantee. Follow the convention on bond literature, we exclude financial firms (SIC codes ) and regulated utility firms (SIC codes ). From FISD, we obtain the bond issuance information such as bond maturity, initial bond yield, bond ratings, the indicators for callable bonds, putable bonds, and secured bonds, and so on. Then we merge sample bonds with the Compustat by the issuer ID. We obtain firm characteristics data 11

14 from the Compustat including total assets, cash, operating income, tangible assets, total debt, free cash flow, and sales. Firm ratings come from Compusta and are the most recent S&P long term issuer ratings before bond issuance. Bond ratings are provided by FISD. We first use S&P rating. If S&P rating is missing, we use Moody s rating. If both S&P and Moody s ratings are missing, we use Fitch s rating. We delete firms with missing data on total assets, bond ratings and firm ratings. After the screening, the sample includes 8321 corporate bonds issued by 2214 public firms from 1993 to Among them, 794 bonds were offered with guarantee and 7527 bonds without guarantee. Figure 1 provides a detailed depiction of the number of bonds with guarantee issued by publicly listed companies. Since all the guarantees are internal arrangement either by the parents or the subsidiaries, it is natural to examine whether the issuer has the parent or the subsidiaries. We use Capital IQ to search the information manually. Capital IQ collects companies information from their SEC filings and the information includes the parent and the subsidiaries of the companies in its corporate tree section. Among 2214 public firms, we identify 1035 firms with subsidiaries only, 347 firms with parents only, 424 firms with both subsidiaries and parents, 409 firms without both subsidiaries and parents. We first provide a distribution of guaranteed bonds in the sample in Table 1. We find that the percentage of guaranteed bonds in the sample is about 12% in terms of both the number of bonds and the aggregate value of bonds. We divide the sample firms into public firms and private firms. In terms of the number of bonds, the percentage of guaranteed bonds issued by public firms are higher than that issued by private firms (13.02% vs %). The percentage of aggregate value of bonds issued by public firms is on average 8.50% with the range between 1.19% in 1994 and 20.09% in However, the average percentage of the aggregate value of bonds issued by private firms is 15.76% with the range between 0.78% in 1993 and 41.54% in In sum, public firms have more guaranteed bonds but less bond value than private firms. Secondly, we plot the time-series percentage of guaranteed bonds in terms of the aggregate 12

15 value and the total number of issues by private and public firms together and present the plot in Figure 2. In Panel A, the percentage of guaranteed bonds in terms of the aggregate value increases over time and peaks in Panel B shows that the percentage of guaranteed bonds in terms of the number of issues varies and peaks in 2009 as well. The average percentages of guaranteed bonds are about 12% in terms of both the aggregate value and the number of issues. This finding is consistent with that in Table 1. Finally, we divide sample bonds into seven groups based on the issuers ratings. We assign number 1 to group 1 if firms ratings are AAA, AA+, AA, or AA-; number 2 to group 2 if firms ratings are A+, A, or A-; number 3 to group 3 if firms ratings are BBB+, BBB, or BBB-; number 4 to group 4 if firms ratings are BB+, BB, or BB-; number 5 to group 5 if firms ratings are B+, B, or B-; number 6 to group 6 if firms ratings are CCC+, CCC, or CCC-; and number 7 if firms ratings are CC, C, or D. We then compute the percentage of the aggregate value in each group for guaranteed bonds and non-guaranteed bonds respectively. The plot is presented in Figure 3. We find that most of guaranteed bonds are issued by firms with the ratings of BB+, BB, or BB- while most of non-guaranteed bonds are issued by firms with the ratings of A+, A, or A-. It suggests that firms with lower ratings are more likely to issue bonds packed with guarantee. 4.2 Data and Summary Statistics We provide descriptive statistics of guaranteed bonds and non-guaranteed bonds as well as their issuers characteristics in Table 2. The variables are computed as of the year end before the debt offerings. The financial data is from Compustat and stock data is from CRSP. We find a dramatic difference between guaranteed bonds and non-guaranteed bonds in Panel A of Table 2. For instance, non-guaranteed bonds have higher firm ratings than guaranteed bonds (17.79 vs ). In terms of the size, they are similar (8.40 vs. 8.34). The mean (median) market-to-book of firms with guaranteed bonds is 1.53 (1.41) and this value increases to 1.87 (1.53) for firms without guaranteed bonds. The results show that the firms with guarantee have a lower market to book than those without guarantee. Two 13

16 proxies of growth opportunities, P/E and sales growth, have a different result in two samples. the mean and median P/E of the guarantee firms are significantly lower than those of nonguarantee firms. In contrast, sales growth for guarantee firms is significantly higher for the guarantee firms that for the non-guarantee firms. The implication of this difference is that P/E and sales growth may capture a different aspect of growth opportunities. Considering this possibility, we estimate the regression for the high growth and the low growth samples using P/E and sales growth respectively. The S&P long-term domestic debt rating for guarantee firms is lower than that for non-guarantee firms. As shown by Table 2, bond-issuing firms with guarantee tend to have a higher debt level, lower ROA, lower market-to-book, lower P/E, less free cash flow, lower long-term credit rating by S&P, higher sales growth than those without guarantee. Except free cash flow and inventory, all the mean differences are statistically significant. Panel B of Table 2 shows that the correlation matrix of variables used in the analysis. Firm ratings are positively correlated with firms size, operation profits, bonds par value, free cash flow, rating distance, and market to book ratio but negatively correlated with the marginal tax rate, and sales growth. For instance, the correlation between firm ratings and firm size is Empirical Results 5.1 Effect of Guarantee on Bond Ratings In this section, we first examine the impact of guarantee on the bonds rating at issuance. The dependent variable is bond rating at issuance. The regression specifications are specified as follows. Bond rating = α 0 + α 1 Guarantee + α 2 Firm variable + α 3 Bond variable + ε (1) where Guarantee is the indicator variable equal to one if the bond is guaranteed bond and zero otherwise. We include a set of firm variables and bond variables as the control variables. 14

17 Firm variables are firm rating (FirmRat), firm size (Size), the percentage of tangible assets in the total assets (Tangible), profit (Profit) and debt ratio (Debt) (e.g., Kovner and Wei, 2012). The bond variables include bond maturity (Mat), bond issuance amount (Par),the dummy variable for secured bonds (Secure), the dummy variable for callable bonds (Call), and the dummy variable for putable bonds (Put). Finally, we add the industry dummies and control for the fixed year effect. The regression results are reported in Table 3. In column (1), we include Guarantee only. The coefficient on Guarantee is significantly negative, suggesting that the low rating or low quality of Guarantee firms or negative impact of Guarantee. In column (2), we control for the firm rating. The coefficient of guarantee dummy turns to positive and significant at 1% level. We add firm variables in column (3) and include both firm and bond variables in last column. We find that The coefficient of Guarantee is positive and significant at 5% level. The result confirms a positive relation between guarantee and bond rating. On average, guarantee increases bond rating by The coefficients of firm ratings, firm size, profit, par value, and the dummy for secured bonds are all significantly positive. The coefficient of debt ratio and the dummy for putable bonds is significantly negative. Overall, the finding shows a positive impact of guarantee on bond ratings and therefore provide strong support to our hypothesis H1. We next examine whether the impact of guarantee on bond rating is conditional on the firm rating.to answer this question, we perform the regressions of the guaranty and firm ratings on the change in bond ratings relative to firm ratings. The dependant variable is the bond rating at issuance. Then we identify the firm ratings in which there are both guaranteed bonds and non-guaranteed bonds. The firm ratings meeting the criteria are AAA, BBB+, BBB, BBB-, BB+, BB, BB-, B+, B, B-, CCC+, CCC and CC. We interact the guarantee dummy with the firm rating dummies respectively. We also control for the firm rating, other firm variables and bond variables as in the regressions in Table 3. The regression specifications are specified as follows. 15

18 Bond rating = α 0 + α 1 guarantee*firm Rating + α 2 Firm variable + α 3 Bond variable + ε (2) The dependent variable is bond rating at issuance. guarantee*firm Rating is one when the bond is guaranteed and the firm rating is the specified rating. Other independent variables are the same as defined in the regression of bond rating in Table 3. The result is reported in Table 4. The coefficient of firm rating is positive and confirms the positive impact of firm rating on bond rating.when the firm rating is close to the default rating, the coefficient on guarantee*ccc is and the coefficient on guarantee*cc is and significant at 1% level. The coefficient of guarantee*bbb is When the firm rating is close to the investment and non-investment grade cut-off, the coefficient of guarantee*bbbis and the coefficient of guarantee*bb+ is They are all significant at 1% level. While the firm rating is AAA, the coefficient of guarantee*aaa is and significant. The coefficients for the interaction of Guarantee with other firm rating are not significant. In sum, the positive impact of guarantee concentrates on the firms with ratings either close to default or the split of investment and non-investment grade, with the largest positive impact when firm ratings are close to default. There are not positive influence of guarantee on bond ratings at other firm ratings. The result provide evidence that impact of guarantee on bond rating is conditional on the firm rating and support hypothesis H1b. 5.2 Effect of Guarantee on Bond Yield at Issuance In this section, we examine the impact of guarantee on the bonds at issuance. The dependent variable is bond offering yield spread which is the difference in the yields of the corporate bond and the treasury bonds with corresponding maturities.the regression specifications are specified as follows. Yield Spread = α 0 + α 1 guarantee + α 2 Firm variable + α 3 Bond variable + ε (3) The independent variables are the same as defined in the regression of bond rating in Table 3. The results are presented in Table 5. In Column (1), we add the dummy variable 16

19 Guarantee first. The coefficient is not significant. Then we add more firm rating, other firm variables and bond variables into column (2) to (4) and find the coefficient of dummy variable Guarantee is not significant in all specifications. Contract to the result of rating regressions, we find that guarantee doesn t have any significant impact on yield in general. To further explore this issue, we perform the regression conditioned on each firm rating level. Specifically, we add the key dummy variable as in the rating regression to indicate the interaction between a guarantee dummy and the issuer s rating dummy. Similar to the regression of bond rating, the firm ratings AAA, BBB+, BBB, BBB-, BB+, BB, BB-, B+, B, B-, CCC+, CCC and CC are selected because there are both guaranteed bonds and non-guarantee bonds at these firm ratings. The regression specifications are specified as follows. Yield Spread = α 0 + α 1 guarantee*firm rating + α 2 Firm variable + α 3 Bond variable + ε (4) The dependent variable is bond yield spread at issuance. guarantee*firm rating is one when the bond is guaranteed and the firm rating is the specified rating. Other independent variables are the same as defined in the regression of bond rating in Table 3. The result is reported in Table 6. The coefficient of guarantee*bbb- is and sufficient at 1% level. The result shows guarantee lowers the yield spread significantly when the firm rating is at BBB-. While firm rating is at CC, B+, BB, or AAA, the coefficients are positive and significant and thus the guaranty increases the yield spread significantly instead at these firm ratings. For firms at most of the ratings, we do not detect a significant yield spread difference between two types of bonds. The result reveals the fact that investors interprets the information of guarantee on bonds conditioned on issuer s ratings. Specifically, investors, or more precisely underwriters, do not take the guarantee as positive signal of the bond s credit prospect in general. 17

20 5.3 The Determinants of the Use of Guarantees Bond-level Analysis In this section, we examine the determinants of the use of the guarantee. The special arrangement of subsidiary guarantee requires a firm to have subsidiaries to obtain guarantee. As presented in Panel A of Table 2, guarantee firms have the largest number of domestic subsidiaries (39) at rating range ( BBB+, BBB, BBB- ) while non-guarantee firms have the largest number of domestic subsidiaries (31) at rating range ( B+, B, B- ). On average, guaranteed bonds have 45 domestic subsidiaries and non-guarantee firms have 26 domestic subsidiaries. The difference is not significant. The appearing question is why non-guarantee firms do not use subsidiary guarantee while they could, given the increase of bond rating from guarantee? To answer this question, we perform logistic regressions to examine the determinants of firms that consider guaranteed bonds or non-guaranteed bonds (Y). For each bond, firms either consider guarantee (Y=1) or don t consider guarantee (Y=0). We consider a set of factors in a vector x to explain the decision. We model the probability that a firm uses guarantee as a probit function: Pr(Y ) = Φ(β X) (5) where Y* is not observable while we can observe y, Φ(.) denotes the standard normal distribution, X is a set of variables explaining bond issuers propensity to use guarantee (discussed below). The set of parameters β reflects the impact of changes in on the probability. In the setting of probit, we have: 1, if Y > 0; Y = 0, if Y <= 0. X includes Guarantee dummy and two group of variables: financial constraints and agency problem. Guarantee is the dummy variable if the bond is packed with guarantee and zero otherwise. (6) 18

21 The independent variables are in two groups. The first group of variables proxy for financial constraint consists of firm ratings FirmRat, firm size Size, cash ratio Cash, tangible assets Tangible and debt ratio Debt. The second group of variables proxy for agency problem contains High FCF*Low MB, High FCF*Low SG and High FCF*Low PE. We control for Moody s yield spread for corporate bonds, bond maturity, bond issuance amount at issuance, the dummies for callable bonds, putable bonds, and secured bonds. The key measure for financial constraints is the credit quality. In this study, credit quality is measured by firm rating. A firm s rating is its Standard & Poor s long-term issuer rating as recorded in Compustat and reflects a firm s absolute creditworthiness. We expect the higher the firm s rating, the more likely the firm uses guarantee. Therefore, both coefficients are expected to have positive signs. In addition, we control for the collateral of the firm. Bernanke and Campbell (1988) advocate that cash and PPE variables can be used to measure a firm s collateral for external capital. The profitability is also a typical proxy for debt capacity in the corporate finance literature. We expect collateral to decrease the use of guarantee. Specifically, we expect a negative sign of their coefficients. Size is also a standard measure of financial constraints (e.g., Gilchrist and Himmelberg, 1995; Erickson and Whited,2000; Almeida and Campello, 2007; Li and Zhan, 2010). Small firms have more information asymmetry than larger firms and lack collateral to back up their borrowing. Consequently, small firms have limit access to debt market and have more financial constraints (Whited, 1992). Therefore smaller size firms are more likely to use guarantee. However, smaller size firms are less likely to have strong subsidiaries as guarantors. The sign of the coefficient can be either positive or negative. The net effect is a empirical test question. Tangible is proxy for collateral. The agency problem is measured by the indicator of a combination of high free cash flow and low growth opportunities. In the agency literature, it has been well established that the higher free cash and lower growth opportunities increase the agency problem (Jenson, 1986; Opler and Titman, 1993). Following Kolasinski (2009) and Custodio, Ferreira and Laureano (2013), we use market-to-book ratio as a measure for growth opportunities. Alternative, 19

22 McConnell and Servaes (1995) use sales growth and P/E ratio as the measures for growth opportunities. We divide the sample into high and low free cash flow groups and growth opportunities group respectively. For the firms in the high free cash flow and low growth opportunities group, we define it as the firms more likely to have agency problem. As stated in previous section, agency problem may play an important role in the decision of guarantee use. It seems reasonable to assume the sign of the coefficients of number of subsidiaries and agency problems proxies is positive. All data are in the fiscal year before the debt issuance. We use a set of control variables. Par value (Par) is the total offering amount in $ millions. Time to maturity (Mat) is the maturity of each bond in years. Secured dummy (Secure) is equal to one if the bond is secured. Call dummy (Call) and Put dummy (Put) are dummy variables equal to one if the bond has a call or put provision. Moody s yield spread for corporate bond is used to control for the timing of bond issuance by a firm. We use the average monthly yield of Moody s AAA and Baa bonds.the lower the yield of bonds in the market, the more likely the firms use guarantee to increase their debt capacity. We expect a negative sigh for the yield variable. To control for time-varying macroeconomic factors and industry specific factors, we also include the fixed year effect and fixed industry effect. To overcome the potential bias in standard errors from the correlation between firms, we follow Kolasinski s (2009) approach in his research of determinants of subsidiary debt. Specifically, Robust standard errors are clustered at the firm and year level. The results are reported in Table 7. In column (1), we include the major financial constraint proxy firm rating and market yield spread. The coefficient of FirmRat is and significant at 1% level. As a major measure of financial constraint, the result shows that firms with more financial constraints are more likely to use guarantee. In column (2), we include other financial constraints variables Size, Cash, Tangible and Debt and bonds variables. The coefficient of Size) is positive and significant. The finding reveals that the larger firms are more likely use guarantee. The coefficients of (Cash and Tangible are 20

23 negative and Debt is positive and they are all statistically significant. Overall, the results of the financial constraint variables confirm that an increase of financial constraint is associated with an higher possibility of using guarantee. The results support hypothesis H1. In column (3), we include both firm rating, other financial constraints variables and bond variables. The result is consistent with the results in column (1) and column (2). In column (4) - column (6), we add three agency problem proxy variables HighCF*LowMB, HighCF*LowSG, HighCF*LowPE respectively. The coefficient of High FCF*Low MB is and significant at 5% level. When we use the alternative proxy for growth opportunities, the coefficients for High FCF*Low SG is and High FCF*Low PE is and significant at 1% level. The results indicate that the more agency problem a firm has, the more likely it uses guarantee. Our finding supports hypothesis H Multinomial Logistic Regressions on guarantee Uses: Firm-level Analysis For firms using credit enhancements, there are actual two options. One is to issue a mix of guaranteed bonds and non-guaranteed bonds and another one is to issue guaranteed bonds only. In this section, we investigate the determinants of the firms choice between issuing guaranteed bonds only and issuing a mix of guaranteed and non-guaranteed bonds. The probability that uses the issuance strategy is estimated using a multinomial logistic regression model. The multinomial logistic model extended to include fixed effects can be written as follows: Log( P (Y m = 1, 2) P (Y i = 0) ) = α m + K β mk X ik + ε (7) k=1 The three categories of the dependent variable Y are: 0: if the firm issues non-guaranteed bonds only in year k. 1: if the firm issues both non-guaranteed bonds and guaranteed bonds in year k. 2: if the firm issues guaranteed bonds only in year k. 21

24 In this multinomial logistic regression model, the baseline group is Y i = 0. X is a vector of independent variables that includes: subsidiary number, market yield of corporate bonds, firm rating, size, cash, tangible, debt, marginal and the indicator of agency problem High FCF*low MB. The result is reported in Table 8. The second column is the regression results for Y i = 1 in which the dependant variable is the ratio of using mixed strategy to issuing non-guaranteed bonds only and the third column is the regression results for Y i = 2 in which the dependant variable is the ratio of using pure strategy to issuing non-guaranteed bonds only. The coefficients of the firm rating are in the mixed strategy group while in the pure strategy group. In other words, holding other variables at a fixed value, with a one-unit decrease in the firm rating, the log odds of using the mixed strategy (j = 1) over using non-guaranteed bonds (j = 0) increase 0.321, while the log odds of using the pure strategy (j = 2) over using non-guaranteed bonds (j = 0) is The coefficients of the High FCF*low MB variable in the mixed strategy group and in the pure strategy group are not significant. The coefficients of the cash variable are in the mixed strategy group and in the pure strategy group. The coefficients of the tangible variable are in the mixed strategy group and in the pure strategy group. All these coefficients are significant at 1% level. The coefficient of Size is in the mixed strategy group and in the pure strategy group and both are significant at 1%level. In sum, the lower firm rating, less cash, less tangible, larger size increase the odds of using mixed strategy more than that of using pure strategy. 5.4 Subsample Analysis The analysis confirms that agency problem has a positive influence on the use of guaranties. Because the corporate governance is negatively associated with the agency problem, a reasonable expectation is that the influence of agency problem on the use of guarantee will be 22

25 affected by the corporate governance 5. In the this section, we examine the conjecture Effect of Managerial Voting Power Chang and Mayers (1992) examine employee stock ownership plan announcements and find the evidence of more agency problem from increased managerial voting power. Masulis, Wang and Xie (2009) use a example of U.S. dual-class companies to examine how divergence between managerial voting and cash flow rights affects managers private benefit. Their empirical results support that the managers with more voting power are more prone to purse private benefit. We therefore postulate that firms with more managerial voting power are more likely to use guaranties. We examine the impact of the managerial voting power on the use of guaranties. We use the percentage of the voting shares owned by the managers to proxy the managerial voting power. The subsample is an intersection of the whole sample and the data with available percentage of the voting shares from RiskMetrics. There are 5,247 observations in the subsample.the sample is then divided into 3 groups based on the managerial percent control of voting power. We assign bonds issuance in the largest percentage group to high managerial voting power subsample and bonds issuance in the smallest percentage group to the low managerial voting power subsample. The logistic regression runs in the subsample with available managerial voting power data. The dependent variable is the dummy variable for guarantee use. The independent variables are a dummy variable HighMgtVote which equals to one if the firm is in the high managerial voting power subsample and zero otherwise. Other independent variables include financial constraints variables FirmRate, Size, Cash, Tangible and Debt and bond variables Mat, Par, Secure, Call and Put. The results are reported in the Table 9. In the first column, we include dummy variable HighMgtVote and financial constraints variables FirmRate, Size, Cash, Tangible. The coefficient of HighMgtVote is positive and 5 The widely used Gindex for corporate governance is available from 1990 to Our sample is from 1993 to The subsample with GIndex loses two thirds of the observations. Therefore we do not use GIndex in this analysis. 23