Ch. 6: Security Offerings 353
regression is then
rI−rM=(αI−αM)+(β 1 I−β 1 M)F 1 +, (12)where=(β 2 I−β 2 M)F 2 +u, whereuis a white noise error term. The definition
of a “good match” is thatβIis close toβM. For example, if the size and B/M match-
ing often used in the literature in fact produces a good match, then you expect the
“issuer–match” regression to have both a small alpha and values of beta close to zero.
Alternatively, if the matching technique fails to control for important risk factors, then
the zero-investment “issuer–match” portfolio will contain significant factor loadings.
Eckbo, Masulis, and Norli (2000)(SEOs and debt offerings),Eckbo and Norli (2005)
(IPOs), and this survey (all issue categories) all lead to the conclusion that the zero-
investment portfolio exhibit significant factor loadings in the extended Fama–French
model, but that the alpha of this portfolio is not significantly different from zero. This
is consistent with the proposition that the technique of matching on size and B/Mis
insufficient to control for important risk exposures of the issuing firms.Lyandres, Sun,
and Zhang (2005)reach a similar conclusion for their sample of SEOs after performing
a three-way sort of size, B/M and investment intensity.
Alternative factor structures. Eckbo, Masulis, and Norli (2000)use a model with six
prespecified macro factors: the value-weighted CRSP market index, and factor mimick-
ing portfolios for the return spread between Treasury bonds with 20-year and one-year
maturity, the return spread between 90-day and 30-day Treasury bills, the seasonally
adjusted percent change in real per capita consumption of nondurable goods, the dif-
ference in the monthly yield change on BAA-rated and AAA-rated corporate bonds,
and unexpected inflation.^56 This six-factor model produce regressionR^2 similar to the
Fama–French model, and the alphas are uniformly indistinguishable from zero.
Eckbo, Masulis, and Norli (2000)also report alpha estimates when the time series of
the demeaned, raw macroeconomic factors is used rather than factor-mimicking portfo-
lios. Raw macro factor shocks are interesting in part because they are not affected by
stock market mispricing (if any). Also, factor-mimicking portfolios contain measure-
ment error vis-à-vis the true risk factors, which raw factors avoid. On the other hand,
there is measurement error induced by the demeaned raw macroeconomic factors them-
selves. It is difficult to determine a priori which of the two sources of measurement
error is most severe (and thus whether factor mimicking is superior).^57 In any event, the
alpha estimates remain insignificantly different from zero, though somewhat larger in
absolute value than those for regressions based on factor-mimicking portfolios.^58
(^56) These factors also appear in,Ferson and Harvey (1991), Evans (1994), Ferson and Korajczyk (1995),
Ferson and Schadt (1996),andEckbo and Smith (1998)among others.
(^57) Factor mimicking portfolios are required when estimating risk premiums (denominated in returns).
(^58) Eckbo, Masulis, and Norli (2000)report that a similar conclusion emerges when alpha is estimated using
factors extracted from the covariance matrix of returns using the principal components approach ofConnor
and Korajczyk (1988). Although principal component factors do not have intuitive economic interpretations,
they provide yet another factor structure useful for sensitivity analysis.