Note that we have an estimate of the nonsystematic risk of firm i from the estimation
period in the event study. The estimation period from t=-110 to t=-10 yielded an estimation σ2
which we can now use in a Generalized Least Square (GLS) framework to obtain more efficient
estimates of (3). The procedure used is to weigh all dependent and independent variables in (3)
with the inverse of σi, as outlined in Judge et al. (1988, page 359). Thus, in effect we estimate the
following equation:
where ui = εi∕σi. The variance of the error term ui is not assumed to be scaled to 1 by this
procedure, but is estimated in the regression. We will now turn to the results and their interpreta-
tion.
CARi1
σi
= β0 *2 + β1 ⅛ + β2 *ΞS +
01 2
σ η σi
β3*^umi ÷ u,
σi i
(4)
Regression analysis results
To answer the question whether shareholders react differently to the issuance of CBs in relation to
WBs, we estimate equation (3) using GLS. The results of this regression are given in table 3.
[Insert Table 3]
The variables Snew and SB are expected to give negative coefficients. Contrary to our
expectations this turns out not to be the case for both variables. In both cases positive, but not
significant, values are found for these variables6. Therefore these results do not confirm the
Pecking Order Theory.
Finally the dummy variable tests whether there is a difference between CB issues and WB
issues after accounting for structural differences between individual issues. This appears to be the
case. The coefficient of 0.0168 indicates that the cumulative abnormal return is 0.0168 (1.68%)
higher for a WB than for a CB. However, the explanatory power of the regression is very small.
This may be caused by the fact that the announcement effects for a great deal are caused by the
announcements of other firm specific news. It would probably be more interesting to run these
regressions for the samples of Long and Sefcik (1990) and Billingsley et. al. (1990), since the
announcement effects studied by them are not, or in any case less, contaminated by other firm
specific news than the announcement effects in our sample.
6 We have also run regressions for other variables on which CBs and WBs differ. These
regressions included either the average or the maximum maturities of the bonds and the warrants
(conversion rights). None of these variables appeared to have a significant effect.