340 B.E. Eckbo et al.
based on security offerings made over the 1980–2001 sample period, and compare these
to the extant literature.
5.3.1. Sample selection
The choice of sample period generally affects the statistical significance of reported
abnormal return estimates.^49 Shorter sample periods reduce statistical power, while
different sample periods have varying exposure to the problem of cross-correlation
of overlapping holding-period returns (discussed extensively byKothari and Warner
(2007)inChapter 1of this volume). The literature uses security offer samples from as
early as 1961 (Mitchell and Stafford, 2000) and as late as 2003 (Lyandres, Sun, and
Zhang, 2005), with the bulk of the existing studies sampling from the 1980s and the
early 1990s. The primary data source after 1980 is SDC, while earlier samples typically
are found by searching the Wall Street Journal for issue announcements or relying on
the SECs now defunct Registered Offerings of Securities database. Stock returns are
almost always drawn from CRSP Daily Stock Price and Returns database.
Some authors exclude issues by public utilities on the grounds that the regulatory
agencies make utility issues relatively predictable. Utility issues occurred on relatively
frequent basis in the 1970s, and again as a result of deregulations in the late 1990s
(Eckbo, Masulis, and Norli, 2000). As discussed above, the market reaction to SEOs
is significantly smaller for utility issuers than for industrial issuers. Thus, it matters
whether the utility issues are pooled in the long-run performance analysis. It is also
customary to exclude issuers with stock price less than $5, as well as unit offerings
and simultaneous offerings of other securities. Issues by foreign corporations, closed-
end funds, unit investment trusts, and real estate investment trusts are also customarily
excluded. Moreover, most studies require data on book value of equity, taken from Com-
pustat, which further reduces sample size.
Our sample selection for the long-run analysis below is as follows. We start with
the overall sample of 80,627 security issues from Section2.3above. Recall that this
sample already ensures that the issuing firm is found on the CRSP tape for the relevant
period. We then exclude the following issues using information from SDC: (1) ADRs
and GDRs, (2) simultaneous offerings of debt and equity, (3) simultaneous offerings of
international issues, (4) unit offerings, (5) offers with missing SDC information on of-
fering proceeds, and (6) offerings after year 2000. The last restriction ensures five years
of post-issue stock return data. These six criteria reduces the total sample to 54,283.
We then apply restrictions specific to CRSP: (7) CRSP share code must be either 10 or
11 (ordinary common shares), (8) the issuer must be listed on NYSE/AMEX/Nasdaq,
(^49) Figure 3 inEckbo and Norli (2005)presents a striking illustration of the impact of sample period on the
average holding period return. Due to the slump in the stock market in the mid-1970s, a study of long-run
returns following IPOs (which starts with the first Nasdaq IPOs in 1973), will easily conclude that the IPO
portfolio underperform therisk-freerate if the sample period ends prior to the mid-1980s.