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
Variables C 5
1 C 5
2 C 5
3 C 5
4 Constant
2 9.545
2 9.315
2 8.953
2 8.815
0.241 0.312
0.215 0.229
LEV 2
0.506 0.326
2 0.525
2 0.004
0.556 0.417
0.552 0.616
PDIR 1.651
2.070 2.618
1.993 0.729
0.631 0.760
0.892 INOWN
0.012 0.002
0.010 2
0.007 0.005
0.004 0.004
0.011 LNPRIN
2 2.893
2 3.547
2 4.227
2 4.199
0.727 0.747
0.068 0.081
Likelihood Ratio Index 5 0.445 The term in parentheses are standard errors.
Significant at the .01 level. Significant at the .05 level.
dummy variable removes the bias of the parameter estimators. In the presence of the selectivity variable, the error term n
i
is N0, s
n 2
.
4. Results
4.1. The poison covenant inclusion equation As discussed earlier, firms with high probability of takeover are most likely to
include poison covenants in their bond indentures. Therefore, the variables that affect the probability of a takeover are also the variables that are expected to determine
the types of covenants to be included. Consequently, the probit model is estimated in conjunction with various combinations of poison covenants specified in Table 2.
According to Harris and Raviv 1988, Stulz 1988, Israel 1991, Hendershott 1991, among others, the main variables that seem to affect the probability of a firm
receiving a takeover bid are: a lower level of the existing debt; a larger percentage of the outsiders in the board of directors; and a smaller block of the common shares
held by the insiders i.e., management. The probit model estimated is:
C
i
5 a
1 a
1
LEV
i
1 a
2
PDIR
i
1 a
3
INOWN
i
1 a
4
LNPRIN
i
1 e
i
3 The dependent variables, C
i
, take values from 1 through 4, with C 5 0 the case of straight bond serving as default. Variables LEV, PDIR, and INOWN are defined in
the previous section, and LNPRIN is the logarithm of the principal of the new bond issue. Although the size of the issue may not have any bearing on the probability of
a takeover, its interrelationship with the inclusion of poison covenants is of interest. That is, firms issuing large amounts of debt might be required to include some types
of poison covenants in their bond indenture irrespective of their status of being takeover candidates.
Table 3 reports the results of the estimation of the multinomial probit model. The
K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 149
likelihood ratio index, a measure approximately equivalent to the R
2
, shows that the model explains about 45 of the variation in the dependent variables.
The proportion of outside directors and the logarithm of the new bond’s principal turn out to be the variables that display the most consistent statistical significance.
The results indicate that investors assign higher probability of takeover to firms with higher proportion of outside directors and, hence, demand some kind of protective
covenants. In contrast to our expectation, the logarithm of the new bond’s principal has a consistently negative effect, indicating that poison covenants are encountered
more frequently in the smaller issues. The existing level of leverage, however, has no statistically significant effect on the inclusion of poison covenants. Interestingly, the
effect of the proportion of insider ownership is positive and statistically significant when the inclusion of a poison call is considered. This result suggests that when insiders
control a large block of common shares, they intend to issue bonds with poison calls in an attempt to force the prospective bidders to negotiate directly with them. This
is consistent with the Stulz’s 1988 argument of management entrenchment at higher levels of managerial share ownership.
4.2. The reoffering yield equation In the second stage of the estimation process a linear regression model is used to
determine the effects of various poison covenants upon the cost of the firm’s bond issue. The cost is measured as the reoffering yield differential between the bond’s
yield to maturity with the yield on a treasury bond having the same time to maturity. The following Eq. 4 is estimated using the Ordinary Least Squares method:
RY
i
5 d
1 d
1
LNPRIN
i
1 d
2
TR
i
1 d
3
VOL
i
1 d
4
CALL
i
1 d
5
SF
i
1 d
6
MAT
i
1 d
7
RATG
i
1 d
8
CVT
i
1 d
9
Z
i
1 n
i
4 where
RATG
i
: a vector of dummy variables describing the ratings assigned to the bonds by either Moody’s or SP. The highest rating triple A is the default
variable. Z
i
: a vector of the four selectivity variables calculated in Eq. 3. n
i
: the error term assumed to have zero mean and constant variance. Tables 4 and 5 present the coefficient estimates of the Z-variables using Moody’s
and SP ratings, respectively. In each table, two sets of estimates are reported: one using the full sample and the other excluding the convertible bonds, reducing the
sample to 724 bonds. Regression diagnostics variance inflation factors show that the data do not suffer from the multicollinearity problem. The reported standard errors
of the coefficients are corrected for heteroskedasticity using White’s 1980 method to yield consistent estimates.
All four estimates yield approximately similar results. Each one explains about 82 of the overall variation of the reoffering yield differential. The exclusion of the
convertible bonds leads to weakening of the significance of some coefficients and sign
150 K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156
Table 4 Reoffering yield estimation results: Moody’s ratings
Full sample Convertibles excluded
Coefficient Standard
Coefficient Standard
Variables of observations 5 1015
error of observations 5 724
error Constant
3.530 0.592
2.150 0.626
LNPRIN 0.141
0.053 0.054
0.061 TR
2 0.378
0.043 2
0.334 0.044
VOL 0.122
0.375 2
0.012 0.380
CALL 0.054
0.078 2
0.080 0.065
CVT 2
4.701 0.122
— —
SF 0.330
0.111 0.128
0.112 MAT
2 0.011
0.005 0.010
0.005 AA
1 0.164
0.131 0.015
0.162 Aa
2 0.475
0.134 0.433
0.118 Aa
3 0.441
0.153 0.434
0.111 A
1 0.524
0.124 0.571
0.107 A
2 0.677
0.119 0.690
0.104 A
3 0.929
0.130 0.846
0.113 Baa
1 1.286
0.158 1.069
0.117 Baa
1 1.261
0.142 1.175
0.122 Baa
3 1.748
0.235 1.014
0.248 Ba
1 2.377
0.477 2.282
0.800 Ba
2 2.909
0.264 3.351
0.568 Ba
3 3.105
0.176 3.531
0.188 B
1 3.304
0.191 3.808
0.238 B
2 3.980
0.171 4.721
0.166 B
3 4.305
0.246 5.105
0.308 Caa
4.624 0.691
3.990 1.241
Z
1
0.230 0.212
0.813 0.323
Z
2
2 0.434
0.171 0.177
0.217 Z
3
2 0.343
0.126 2
0.215 0.129
Z
4
2 0.151
0.112 2
0.100 0.092
R
2
0.824 0.823
ADJ. R
2
0.819 0.816
F -value
170.534 123.671
SEE 0.924
0.797 Significant at the .01 level.
Significant at the .05 level. Significant at the .10 level.
reversals. Among the control variables variables other than Zs the exclusion of the convertible bonds renders the effects of LNPRIN size of the new issue, SF sinking
fund, and MAT maturity statistically insignificant, while the direction of the effects of VOL market rate volatility, CALL call provision, and MAT is reversed. The
coefficients of TR long-term treasury rates and almost all of the ratings variables are strongly significant and have the expected signs. The coefficient of the long-term
K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 151
Table 5 Reoffering yield estimation results: SP ratings
Full sample Convertibles excluded
Coefficient Standard
Coefficient Standard
Variables of observations 5 1015
error of observations 5 724
error Constant
3.746 0.622
2.654 0.676
LNPRIN 0.103
0.052 0.027
0.062 TR
2 0.309
0.043 2
0.297 0.044
VOL 0.339
0.382 2
0.058 0.351
CALL 0.039
0.079 2
0.095 0.068
CVT 2
4.647 0.114
— —
SF 0.428
0.110 0.195
0.115 MAT
2 0.016
0.005 0.006
0.005 Aa1
0.164 0.154
0.004 0.204
AA 0.152
0.107 0.231
0.101 AA2
0.451 0.119
0.409 0.098
A1 0.309
0.111 0.400
0.105 A
0.498 0.094
0.538 0.093
A2 0.682
0.108 0.672
0.095 BBB1
1.012 0.132
0.934 0.136
BBB 0.987
0.147 0.843
0.144 BBB2
1.730 0.185
1.151 0.156
BB1 2.635
0.189 2.886
0.273 BB
3.718 0.665
4.269 0.776
BB2 2.933
0.240 3.474
0.411 B1
3.111 0.195
3.609 0.300
B 3.115
0.179 4.044
0.266 B2
3.583 0.154
4.381 0.151
CCC1 4.173
0.212 4.991
0.192 CCC
4.176 0.501
4.882 0.849
CCC2 5.314
0.303 4.778
0.109 E
1 2
1.712 0.219
2 0.562
0.219 E
2 0.587
0.218 0.331
0.118 E
3 2
0.357 0.262
2 0.169
0.273 E
4 0.076
0.288 2
0.007 0.265
E 5
2 0.185
0.221 2
0.228 0.265
Z
1
0.363 0.212
0.662 0.301
Z
2
2 0.578
0.178 0.112
0.240 Z
3
2 0.510
0.133 2
0.305 0.110
Z
4
2 0.362
0.220 2
0.312 0.255
R
2
0.824 0.822
ADJ. R
2
0.818 0.813
F -value
134.718 95.540
SEE 0.926
0.804 Significant at the .01 level.
Significant at the .05 level. Significant at the .10 level.
152 K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156
Table 6 The effects of poison provisions on the reoffering yields of corporate bonds in terms of basis points,
1986 through 1990 Moody’s ratings
SP ratings Convertibles
Convertibles Convertibles
Convertibles Poison covenants
included excluded
included excluded
Z
1
: PCALL 0.267
0.542 0.421
0.359 Z
2
: PPUT 2
0.586 0.230
2 0.780
0.146 Z
3
: PCALL, PPUT 2
0.391 2
0.245 2
0.582 2
0.347 Z
4
: PPUT, SPP 2
0.173 2
0.116 2
0.416 2
0.361 Significant at the .01 level.
Significant at the .05 level. Significant at the .10 level.
treasury rate implies that the yield differential shrinks by about 31 to 38 basis points every time the treasury rate’s yield increases by 1. The coefficients of the ratings
variables can be interpreted as the yield differential, in basis points, between a bond and AAA or Aaa bond. For example, the convertibility feature reduces the yield
differential by about 470 basis points as reported in Table 4. Interestingly, the results indicate that issuers of the speculative bonds could enjoy the yield differential of
almost an AAA bond by making their bonds convertible.
It should be noted that the E-rating bonds are included in the regressions using the SP ratings, while they are excluded under Moody’s. As pointed out before, the
E1 rating indicates the strongest event-risk protection while the E5 indicates the weakest. The signs, the size, and the statistical significance of the coefficients of these
ratings are erratic at best. While the rating of E1 has a negative sign and is statistically significant, the rating of E2 has a counterintuitive positive and significant effect on
the yield differential.
7
Similarly, the rating of E4 in the full sample has a positive, albeit statistically insignificant, effect. The remaining ratings have the expected nega-
tive signs, but they are not statistically significant. The coefficients of the analysis variables Z
1
to Z
4
are somewhat difficult to interpret in their present forms. A convenient way is to transform the coefficients into familiar
basis points and evaluate the effects of the analysis variables at their corresponding means. Table 6 reports the results.
The relative structures of the effects of poison covenants remain the same under both rating systems. As expected, the presence of a single poison call provision has
a positive effect on the yield differential. This positive effect is about 27 to 42 basis points in the full sample and becomes larger, about 36 to 54 basis points, and strongly
significant when the convertible bonds are excluded. The effect of a single simple poison put in the full sample is negative and significant, reducing the yield differential
by 58 to 78 basis points. When the convertible bonds are excluded, the effect becomes positive, but insignificant.
When a bond carries both a poison put and a poison call, a strong negative effect
K.M. Torabzadeh et al. International Review of Economics and Finance 9 2000 139–156 153
on the yield differential will emerge. The effect ranges from 225 to 258 basis points and remains significant even after purging the effects of convertible bonds. The effect
of super poison put is negative and insignificant under Moody’s ratings. The effect, however, becomes marginally significant under the SP rating system, reducing the
yield by 42 basis points.
5. Summary and conclusion
