404 A. Ljungqvist
riod (1972–1977) were actually lower than between 1923 and 1930. Evidence based on
the enactment of the 1933 Securities Act is thus inconclusive.
Tinic also finds that more experienced underwriters were associated with lower un-
derpricing in the post-1933 sample but not before. This is consistent with his prediction
that greater due diligence skills reduce the need for underpricing as a form of protection
against lawsuits. On the other hand, simple certification arguments yield the same pre-
diction, so as a test of the legal insurance hypothesis, the relation between underwriter
experience and underpricing has little power. Moreover, as discussed in Section3.1,this
relation appears to have changed sign in the 1990s (Beatty and Welch, 1996). However,
it is not impossible to rationalize a positive relation within the legal insurance hypoth-
esis: more prestigious underwriters may have deeper pockets and so are more worth
suing, leading them to rely more heavily on underpricing. Evidence based on the rela-
tion between underpricing and underwriter experience thus also appears inconclusive.
A potentially more promising research avenue is to investigate the predicted neg-
ative link between underpricing and the probability of litigation, and to do so cross-
sectionally.Drake and Vetsuypens (1993)study a sample of 93 IPO firms that were
sued and compare them to a sample of 93 IPOs that were not sued, matched on IPO
year, offer size, and underwriter prestige. Sued firms are just as underpriced as the con-
trol sample, and underpriced firms are sued more often than overpriced firms. Drake and
Vetsuypens interpret these findings as inconsistent with the legal insurance hypothesis.
Lowry and Shu (2002)argue that such anex postcomparison misses the point because
it does not truly consider theprobabilityof being sued. Empirical analysis of the link
between underpricing and the probability of litigation needs to be careful about the
following simultaneity problem: firms choose a certain level of underpricing to reduce
the probability of litigation, but the level of underpricing they choose depends on the
probability of being sued. Put differently, greater underpricing reduces litigation risk,
but greater litigation risk requires more underpricing.
Due to this simultaneity problem, ordinary least squares estimates are likely biased.
Lowry and Shu propose a two-stage least squares approach. As identifying variables,
they use prior market-index returns in the underpricing equation and the IPO firm’s
expected stock turnover in the litigation equation. The authors motivate these choices
on the basis of prior work and economic common sense, but do not test whether they are
valid^13 or strong^14 identifying variables statistically.Loughran and Ritter (2002)found
a positive relation between lagged index returns and underpricing, but there is no reason
(^13) A necessary and sufficient condition for instrument validity is that the system satisfy the order and rank
conditions. The order condition is easy to check. It requires that the variable be correlated with the endogenous
variable of the first-stage regression, but not with the endogenous variable of the second-stage regression. A
variety of formal tests are available. Stock turnover appears to fail the order condition (seeLowry and Shu,
2002 , Table 5, p. 329).
(^14) Weak instruments may aggravate the effect of simultaneity bias, rather than solving it. To be considered
strong, an instrument needs to be highly correlated with the first-stage endogenous variable.Staiger and Stock
(1997)recommend a cut-off ofF=10. On this basis, Lowry and Shu’s instruments would appear to be weak.