Sample, data, and methodology

K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 143 bondholders are somewhat aligned through the protection extended to the bondhold- ers. Cook and Easterwood 1994 provide evidence for this “mutual interest hypothe- sis.” They measure the wealth effect of issuing debt with and without poison put covenants on the outstanding debt and equity. Their sample of bonds with poison put covenants consists of 63 bonds issued in the period of 1988 to 1989. They find that the presence of poison puts affects current stockholders negatively and existing bond- holders positively. They conclude that the issuance of bonds with poison put covenants protects managers from hostile takeovers and bondholders from event risk, at the expense of stockholders. Their findings, however, are in contrast to those reached by Bae et al. 1994. These authors use 83 E-rated bonds issued from 1982 to 1990 to examine the effect of poison put provisions on stockholder wealth. Similar to Cook and Easterwood, they utilize an event-study methodology but find positive stock price response to the announcement of an issuance of bonds with event-risk protection. Their regression analysis indicates that the presence of event-risk covenants increases shareholder wealth primarily by reducing the firm’s agency costs of debt. Their findings basically support the argument offered by Kahan and Klausner 1993. Although they maintain that managerial entrenchment is the primary motivation behind the issuance of poison bonds, Kahan and Klausner argue that to the extent that bondholders pay for the protection by accepting a lower interest rate, the increased value accrues to the benefit of shareholders. In sum, event-risk covenants can be designed to entrench managers andor provide protection to the bondholders. However, the yield effect, at the time of the issue, depends on the bondholder’s perception of the structure of the covenant. We expect to see a positive reaction in the bond market when the option to put the bond back to the firm is at the discretion of bondholders. Having this attribute, we hypothesize that bonds with super and simple poison puts are priced to yield lower return compared to other comparable bonds without such covenants. We suggest a different scenario for bonds with poison call provisions. These cove- nants are under the control of managers and are viewed as a mechanism to force the prospective bidders to negotiate directly with them. These covenants, in return for some payoff for management, can be easily negated without bondholders’ recourse. For this, we hypothesize that bonds with poison call provisions convey negative infor- mation to the bond market and tend to increase borrowing costs to the firm.

3. Sample, data, and methodology

The sample contains complete data on 1015 new industrial bonds issued by 657 firms between January 1, 1986 and December 31, 1990. Of the 657 firms, 468 firms issued only one bond each; 113 firms issued two bonds each; 36 firms issued three bonds each; and the remaining 40 firms issued four or more bonds each. In 42 cases the same firm simultaneously issued two bonds, each with different characteristics. The sample of poison securities includes 200 bonds consisting of 21 poison calls, 100 poison puts, 48 poison callsputs, and 31 super poison puts. The remaining 815 securities are straight bonds. The sample of poison securities is obtained from a data bank 144 K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 provided by the Securities Data Company. Other sources include SP’s CreditWeek, Moody’s Bond Survey, and Moody’s Industrial and OTC manuals. The straight bonds are collected from the SP Bond Guide. Bonds are included in the sample if they satisfy the following criteria: 1 the bond has maturity of seven years or longer; 2 the bond has a fixed coupon rate; and 3 the bond is rated by both SP and Moody’s. Excluded from the sample are asset- backed securities, deep or zero discount bonds, and bonds issued by a firm’s unit or subsidiary. Issue-specific data i.e., offer yield, maturity, size, type are obtained from the SP Bond Guide and confirmed with Moody’s Bond Survey. Bonds with conflicting data points except ratings are excluded from the sample. Other data i.e., insider ownership, board of directors’ composition, leverage are collected from proxy state- ments, 10-K reports, and Moody’s Industrial or OTC manuals. The variables used are defined as: RY : Reoffering yield differential. Calculated as the difference between the bond’s yield and the similar maturity treasury bond at the date of issue. PRIN : Principal value of the issue. TR : Average daily yield of ten years and longer treasury bonds on the day of the issue. VOL : Interest rate volatility. Calculated as the standard deviation of the TR for the 40 business days before the issue. CALL : Bond callability. A dummy variable that takes a value of one if the bond is callable; zero otherwise. SF : Sinking fund provision. A dummy variable that takes a value of one if the bond has a sinking fund provision; zero otherwise. MAT : Maturity of the bond in years. Aaa to Caa: Moody’s ratings. Measured as dummy variables. AAA to CCC2: SP ratings. Measured as dummy variables. E 1 to E5: SP event-risk ratings. Measured as dummy variables. CVT : Convertibility. A dummy variable that takes a value of one if the bond is convertible; zero otherwise. PCAL : Poison call. A dummy variable that takes a value of one if the bond has a poison call provision; zero otherwise. PPUT : Poison put. A dummy variable that takes a value of one if the bond has a poison put provision; zero otherwise. SPP : Super poison put. A dummy variable that takes a value of one if the bond has a super poison put provision; zero otherwise. LEV : The firm’s leverage before the issuance of the new bond. PDIR : Proportion of outside directors in the board. INOWN : Proportion of insider ownership of the firm. Table 1 presents descriptive statistics for the sample. The average size of each issue is about 147.19 million with an average maturity of about 18 years. Interestingly, the ratings assigned by Moody’s and SP are almost identical. For example, although K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 145 Table 1 Descriptive statistics for 1015 bonds issued between 1986 and 1990 Panel A: Variables with real values Mean values Standard deviation RY 1.3841 2.1822 TR 8.1002 0.6853 VOL 0.1382 0.0771 PRIN 147.1934 160.3374 MAT years 18.3873 7.9421 LEV 50.9019 18.8707 PDIR 69.5220 14.6513 INOWN 15.4473 22.4671 Panel B: Dummy variables No. of observations Proportion of total CALL 797 0.4100 CVT 291 0.4521 SF 550 0.4981 Aaa 13 0.1123 Aa 1 6 0.0766 Aa 2 32 0.1741 Aa 3 35 0.1828 A 1 95 0.2917 A 2 127 0.3315 A 3 87 0.2802 Baa 1 58 0.2324 Baa 2 64 0.2436 Baa 3 41 0.1960 Ba 1 15 0.1207 Ba 2 23 0.1481 Ba 3 62 0.2398 B 1 87 0.2803 B 2 168 0.3710 B 3 93 0.2884 Caa 9 0.0933 AAA 12 0.1081 AA1 6 0.0762 AA 46 0.2086 AA2 49 0.2146 A1 74 0.2600 A 127 0.3318 A2 79 0.2682 BBB1 69 0.2516 BBB 54 0.2245 BBB2 51 0.2189 BB1 22 0.1455 BB 6 0.0767 continued 146 K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 Table 1 Continued Panel B: Dummy variables No. of observations Proportion of total BB2 38 0.0374 B1 64 0.0630 B 87 0.0857 B2 164 0.1616 CCC1 39 0.0384 CCC 22 0.0217 CCC2 6 0.0059 E 1 1 0.0010 E 2 3 0.0029 E 3 32 0.0313 E 4 10 0.0098 E 5 15 0.0148 PCALL 69 0.0680 PPUT 179 0.1763 SPP 31 0.0305 not directly observable from Table 1, of the total 1015 bonds studied during the 1986 through 1990, 54.98t received investment grade ratings by Moody’s as compared to 54.68 by SP. In addition, 78.52 or 797 issues of the bonds were callable, 28.67 or 291 issues were convertible, while 54.19 of them or 550 issues contained sinking fund provisions. In this study, a self-selection model is employed to deal with the endogeneity of the decision to include a poison covenant. The model uses a two-stage procedure. In the first stage, the covenant inclusion equation is estimated using a probit model. Probit model, in its simplest form, is an equation where the dependent variable, C i , takes a value of one when the covenant is present and zero otherwise. This categorical variable is essentially a proxy for the benefits of including a covenant i.e., the depen- dent variable takes a value of one when the benefits are high and the covenant is included and a value of zero when the benefits are low and the covenant is not included. The model is: C i 5 a X i 1 e i 1 where X i is the vector of independent variables that determine the benefits of the covenants and e i is assumed to be N 0, s e 2 . The complications surrounding the decision to include a poison covenant prevent us from using the binomial probit model. First, the management of the bond issuing firm is facing the following poison covenant alternatives. It has to decide whether to include a poison call PCALL, a simple poison put PPUT, a super poison put SPP, or a combination of the three. The choice will be the alternative that yields the highest level of utility or profit for the management. Furthermore, there is no a K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 147 Table 2 Values of the dependent variable C i a C i PCALL PPUT SPP 1 1 2 1 3 1 1 4 1 1 a Another conceivable combination is 1,1,1 where the same covenant contains both a super poison put and a poison call. We found a few cases satisfying this combination, but they failed to clearly meet our sample selection criteria. The remaining combinations of 0,0,1 and 1,0,1 are not possible since we cannot have a super poison put without a simple poison put. priori reason to expect that the utility or profitability ranking of these alternatives would be identical across the firms i.e., there would be no single ordering. Second, an examination of these three types of covenants reveals that they are not independent. For example, the super poison put is a reinforced version of the simple poison put, and as such it displays all the characteristics, and has all the effects, of the simple put. In other words, a bond cannot include a super poison put covenant without a simple put. This implies that the effects of the three types of covenants cannot be examined in isolation; they have to be estimated simultaneously. The first complication can be resolved by utilizing either an unordered multinomial probit or a logit model. However, the second complication can only be resolved by the use of unordered multinomial probit, since the logit model requires “independence of the irrelevant alternatives” see Greene, 1993, p. 671. The unordered multinomial probit, which is employed in the first stage of the estimation process, yields a set of probabilities that correspond to the alternative choices faced by the decision maker. The dependent variable C i now takes values 0 through 4 according to the combination of the covenants included. Table 2 shows how the values of the dependent variable are defined. The multinomial probit model is estimated using a version of the Smooth Simulated Maximum Likelihood Multinomial Probit Model SSMLLP. 6 The estimated values of the probit equation enter the second stage using a multiple regression model, with the reoffering yield differentials RY i as the dependent vari- ables as set forth in Eq. 2: RY i 5 b Y i 1 s v {fC ˆ i FC ˆ i } 1 n i 2 where Y i is a vector of the independent variables, C ˆ i is the fitted value of the probit Eq. 1, f C ˆ i is the frequency function, and F C ˆ i is the cumulative distribution function of C ˆ i . The ratio {fC ˆ i FC ˆ i }is the selectivity variable and its presence instead of the 148 K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 Table 3 Results of the multinomial probit model of the poison covenant inclusion decision