154 M. Baker et al.
Ex post misvaluation. A second option is to use the information in future returns. The
idea is that if stock prices routinely decline after a corporate event, one might infer that
they were inflated at the time of the event. However, as detailed inFama (1998)and
Mitchell and Stafford (2000), this approach is also subject to several critiques.
The most basic critique is the joint hypothesis problem: a predictable “abnormal”
return might mean there was misvaluationex ante, or simply that the definition of “nor-
mal” expected return (e.g., CAPM) is wrong. Perhaps the corporate event systematically
coincides with changes in risk, and hence the return required in an efficient capital
market. Another simple but important critique regards economic significance. Market
value-weighting or focusing on NYSE/AMEX firms may reduce abnormal returns or
cause them to disappear altogether.
There are also statistical issues. For instance, corporate events are often clustered in
time and by industry—IPOs are an example considered inBrav (2000)—and thus ab-
normal returns may not be independent.Barber and Lyon (1997)andLyon, Barber, and
Tsai (1999)show that inference with buy-and-hold returns (for each event) is challeng-
ing. Calendar-time portfolios, which consist of an equal- or value-weighted average of
all firms making a given decision, have fewer problems here, but the changing com-
position of these portfolios adds another complication to standard tests.Loughran and
Ritter (2000)also argue that such an approach is a less powerful test of mispricing, since
the clustered events have the worst subsequent performance. A final statistical problem
is that many studies cover only a short sample period.Schultz (2003)shows that this
can lead to a small sample bias if managers engage in “pseudo”-market timing, making
decisions in response to past rather than future price changes.
Analyzing aggregate time series resolves some of these problems. Like the calen-
dar time portfolios, time series returns are more independent. There are also estab-
lished time-series techniques, e.g.,Stambaugh (1999), to deal with small-sample biases.
Nonetheless, the joint hypothesis problem remains, since rationally required returns
may vary over time.
But even when these econometric issues can be solved, interpretational issues may
remain. For instance, suppose investors have a tendency to overprice firms that have
genuinely good growth opportunities. If so, even investment that is followed by low
returns need not beex anteinefficient. Investment may have been responding to omitted
measures of investment opportunities, not to the misvaluation itself.
Cross-sectional interactions. Another identification strategy is to exploit the finer
cross-sectional predictions of the theory. In this spirit,Baker, Stein, and Wurgler (2003)
consider the prediction that iffeis positive, mispricing should be more relevant for
financially constrained firms. More generally, managerial horizons or the fundamental
costs of catering to sentiment may vary across firms in a measurable way. Of course,
even in this approach, one still has to proxy for mispricing with anex anteorex post
method. To the extent that the hypothesized cross-sectional pattern appears strongly in
the data, however, objections about the measure of mispricing lose some steam.