Ch. 2: Self-Selection Models in Corporate Finance 65
the conflicts of interest hypothesis predicts the morepositivecoefficient.^22 Gande et al.
(1997), andGande, Puri and Saunders (1999)report similar findings for debt issues of-
fered after the 1989 relaxation of the Glass–Steagall Act. Underwritings by commercial
banks convey positive information that improves the prices at which debt offerings can
be sold.^23
8.2. Underwriting syndicate structure:Song (2004)
Song (2004)analyzes debt offerings as inPuri (1996), Gande et al. (1997), andGande,
Puri and Saunders (1999)but there are some important differences in her specifications.
Song uses a switching regression instead of the Heckman model. Second, she focuses
on the effect of the syndicate structure rather than the commercial/investment banking
dichotomy on debt issue spreads.
In Song’s model, commercial banks could enter as lead underwriters or be part of
a hybrid syndicate with investment banks. Alternatively, issues could be underwritten
by a pure investment bank syndicate. For each outcome, we observe the yield of the
debt offering, which is modeled as a function of public information and (implicitly) the
private information conveyed in the firm’s choice of a syndicate structure. The resulting
specification is a variant of the switching regression model of Section3.1, and can be
written as
Ai=1if(−ZAiγA+ηAi)> 0 , (48)
Bi=1if(−ZBiγB+ηBi)> 0 , (49)
Ci=1if(−ZCiγC+ηCi)> 0 , (50)
Y 1 i=X 1 iβ 1 +η 1 i, (51)
Y 2 i=X 2 iβ 2 +η 2 i, (52)
Y 3 i=X 3 iβ 3 +η 3 i, (53)where we have adapted Song’s notation for consistency with the rest of this chapter.^24
In equations(48)–(50), the counterfactuals areA=0,B=0, andC=0, respectively.
In Song’s modelAi =1 if a lead investment bank invites a commercial bank to
participate in the syndicate.Bi =1 if the commercial bank joins the syndicate, and
zero otherwise.Ci =1 if a commercial bank led syndicate is chosen andCi = 0
(^22) Of course, it is possible that investors paid more for bank underwritten issues but were fooled into doing
so.Puri (1994)rules out this hypothesis by showing that bank underwritten offerings defaulted less than
non-bank issues.
(^23) Chiappori and Salanie (2000)use similar methods to analyze the role of private information in insurance
markets.Liu and Malatesta (2006)is a recent application of self-selection models to seasoned equity offerings.
They analyze the availability of a credit rating on the underpricing and announcement effects of SEOs.
(^24) Song’s usage of signs for coefficients and error terms illustrates some confusing notation in the limited
dependent variable literature. Her notation followsMaddala (1983)where the selection criterion is often
written asZγ−η>0, while the more modern textbook convention is to useZγ+η>0.