When do high stock returns trigger equity issues?∗
Aydoğan Altı
University of Texas at Austin
aydogan.alti@mccombs.utexas.edu
Johan Sulaeman
University of Texas at Austin
johan.sulaeman@phd.mccombs.utexas.edu
MAY 2007 - PRELIMINARY DRAFT
∗
PRELIMINARY AND INCOMPLETE
When do high stock returns trigger equity issues?
Abstract
One of the most prominent stylized facts in corporate finance is that firms are more likely to
issue equity following periods of high stock returns. We document that firms exhibit such timing
behavior only in response to high returns that coincide with strong institutional investor demand
for their stock. When not accompanied by institutional purchases, stock price increases have little
impact on the likelihood of equity issuance. The results suggest that potential issuers pay close
attention to who stands by their stock prices.
Introduction
Equity market timing is widely regarded to be a major objective of corporate financial policy. In
recent years, market timing has also come to be seen as having a first-order impact on capital
structure.1 Under the market timing hypothesis, firms attempt to time their security issues
to exploit mispricing of their equity. There are obvious difficulties with this idea, such as why
investors would be willing to take the losing side of transactions that are understood to be timingmotivated. Despite these difficulties, the market timing view maintains its popularity because it
conforms with empirical evidence. The tendency of firms to issue equity following periods of high
stock returns constitutes one of the most well-known stylized facts in corporate finance and is
often interpreted as direct evidence of timing behavior.
In this paper, we take a closer look at the timing of equity issues and find that issuers do
not respond to stock returns per se. High stock returns trigger equity issues only when they
coincide with strong demand from institutional investors. When not accompanied by institutional
purchases, high stock returns have little impact on the likelihood of equity issuance. In other
words, potential issuers care not only about having high stock prices but also about who stands
behind those prices.
We argue that issuers’ concern about institutional demand is a manifestation of asymmetric
information problems in equity issuance. When firms know more about their intrinsic values than
outsiders do, equity issue announcements are greeted as bad news by the market. In standard
adverse selection models such as Myers and Majluf (1984), this market reaction is deterministic
since investors are assumed to be homogeneously informed. In reality, investors differ substantially
in their information. For example, some investors may possess private information about firm
value that other market participants lack. Part of this private information is likely to find its
way into the stock price during the registration period (i.e., from the announcement of the equity
issue to the offer date). Suppose, for example, that the stock price starts to fall following the
announcement. If they have a positive assessment of the firm, informed investors may step in and
add to their holdings of the stock, in effect “supporting the price” via their demand. Potential
issuers are likely to be highly concerned about whether they can count on such demand by informed
1
See, for example, Baker and Wurgler (2002) and Huang and Ritter (2005).
1
investors. We formalize these ideas with a simple model that we develop in the next section. We
associate informed investors in the model with institutional investors that produce information
for trading purposes in practice.
We then proceed with an empirical analysis of the timing and outcomes of seasoned equity
offerings (SEOs). We highlight four main findings. First and as already discussed above, we find
that high stock returns trigger equity issues only when accompanied by high institutional investor
demand. In particular, firms appear to pay close attention to the number of institutional investors
establishing new positions in their stock. To give a sense for the magnitudes, the unconditional
per-quarter probability that a firm announces an SEO is 1.21%. When the previous quarter’s
excess stock return is in the top quintile but the number of institutions initiating positions in
the stock is in the bottom quintile, the SEO announcement probability is only 0.60%. However,
when both the excess return and the number of institutional initiations are in their respective top
quintiles, the SEO announcement probability jumps to 4.21%. This institutional demand effect
remains robust in multivariate probit regressions that control for various firm characteristics, stock
characteristics, and time fixed effects. Interestingly, firms do not seem to condition on the trading
demand of their existing institutional shareholders in deciding whether to issue equity. It is only
the demand by new institutional shareholders that matters for the likelihood of announcing an
SEO.
Second, we analyze stock returns during the announcement-to-offer period. It is well known
that SEO announcements generate negative stock price reactions on average. We find that this
negative SEO announcement effect is similar for firms with high versus low pre-announcement
institutional demand. However, firms with high (low) pre-announcement institutional demand
experience a full reversal (worsening) of the negative SEO announcement effect during the postannouncement period. We interpret this finding as evidence that institutional purchases support
the stock price during the SEO period.
Third, we examine issuers’ stock returns during the period immediately following the offer.
This is the period during which market participants learn about the outcome of the SEO and in
particular about institutional demand for the issuer’s stock. We find that issuers with high offerquarter institutional demand outperform those with low demand by about 7% in the post-offer
part of the offer quarter and another 6% in the following quarter. A 13% return differential in
2
less than two quarters is quite large; it indicates that the market participants pay a great deal of
attention to the resolution of uncertainty regarding institutional participation in the SEO. This
finding also poses a puzzle: high-demand issuers seem to leave too much money on the table; it
looks like they could issue shares at substantially higher prices by waiting a bit more.
Finally, we analyze the long-run stock returns of issuers. Previous research has documented
that SEOs underperform in the long run. Whether such underperformance is due to mispricing is
a source of ongoing debate in the literature. The results discussed above show that firms strongly
time their equity issues to coincide with increased institutional demand. Furthermore, issuers
with relatively high demand exhibit better stock return performance than other issuers around
the time of their SEOs. Perhaps firms view periods of increased institutional demand for their
shares as windows of opportunity in which they can sell overvalued equity. If so, one would expect
high-demand issuers to underperform low-demand issuers in the long run. We find that this is not
the case. While we confirm the finding in previous studies that SEOs underperform in general,
there is no evidence of stronger underperformance for issuers with high institutional demand. In
particular, the short-run gains of high-demand issuer that we discuss above are not reversed in
the long run.
Taken together, our results point to the difficulty of successfully timing the market. Market
timing is often described from a partial-equilibrium perspective, i.e., as an objective of issuers.
Clearly, firms would like to issue at the highest price possible, but investors are likely to recognize
this timing motive and price the stock of issuers accordingly. Our results show that firms are
concerned about how investors will react to an equity issue even after periods of exceptionally
high stock returns. Potential issuers require not only high but also sustainable stock prices.
The remainder of the paper proceeds as follows. In Section 1 we develop the model. Section
2 describes the empirical setup. Section 3 presents the empirical analysis and results. Section 4
concludes.
1. Equity issuance and institutional investors: a simple model
In this section, we present a simple model that highlights the role of institutional investors in
the equity issuance process. The primary purpose of the model is to help fix ideas and motivate
3
the empirical tests; accordingly, the discussion in this section focuses on the basic aspects of the
problem at hand. We delegate various details to the Appendix.
There are four dates (0, 1, 2, and 3), the discount rate is zero, and all agents are risk-neutral.
There is one firm that is either a ‘good’ or a ‘bad’ type. The firm type is characterized by the
magnitude of a single cash flow V that accrues at date 3. A good-type firm receives V G = 1
whereas a bad-type firm receives VB = 0. The two types are equally likely; therefore, one can
think of the initial (i.e., date-0) stock price to be 1/2.
At date 0, the firm has two kinds of shareholders, an insider and a group of outsiders. The
insider holds a block of shares in the firm and privately knows the firm type. Outside shareholders
have access to public information only. The total number of shares of the firm’s stock is normalized
to one. The firm is all-equity financed.
We motivate the firm’s equity issue decision as stemming from the insider’s liquidity needs.
Therefore, one should think of the SEO in this model as a sale of the insider’s block to new outside
investors. We pursue this modelling strategy to simplify the exposition. Our results remain the
same in an alternative model in which the firm has projects itself and issues primary shares to
raise capital.
We model the insider’s liquidity needs in a simple way. With probability λ, a good-type
insider faces a liquidity shock at date 0. When this happens, the insider gains personal access
to a constant returns-to-scale project outside the firm that pays R ∈ (1, 2) units of output per
unit of input. A bad-type insider never faces a liquidity shock.2 Whether the insider receives a
liquidity shock is not publicly observable. The insider has no cash on hand and cannot borrow
from alternative sources, so she has to sell her equity stake in order to invest in her project (if
she has one). For simplicity we do not allow partial sales; the insider sells either all of her shares
or none. In the rest of this section, we refer to the liquidity-constrained good type as G L , the
unconstrained good type as GU , and the bad type as B . Notice that since R > 1, GL would want
to sell all of her shares and invest in her project if she could obtain a fair price for her shares (i.e.,
V = 1). Also, since R < 2, GL will choose to sell none of her shares if she expects to receive the
date-0 (i.e., average) price of 1/2.
2
As will become clear below, whether the bad type has a project or not does not affect our results in any material
way. The assumption we make is intended to facilitate the interpretation that a bad type has purely opportunistic
motives to sell equity.
4
Dates
0
Type GL
Project
Firm
type
1a
1b
1c
2
3
Case 1: All
investors observe
public signal
Stock traded
at price P1
Firm makes
SEO decision
Stock traded/issued
at price P2
Cash flow V
realized
Case 2: Informed
traders observe
private signal s
Stock traded
at price P1
Firm makes
SEO decision
Stock traded/issued
at price P2
Cash flow V
realized
Type GU
No Project
Type B
No project
Figure 1. Timing of events
The firm’s stock is traded in the market at dates 1 and 2. We describe the details of the trading
process below. After observing the date-1 stock price P1 , the insider decides whether to sell her
stake in an SEO or keep it. If the insider decides to sell her stake, the SEO is conducted at date
2. We do not model the details of the SEO process. The insider sells her stake at the market
price P2 . The game ends at date 3 with the payment of V to the firm’s shareholders.
We analyze two cases: a benchmark case with symmetrically-informed investors, and the main
case of interest with privately-informed investors. Figure 1 summarizes the timing of the events
in these two cases.
Case 1: Symmetrically-informed investors (Myers-Majluf )
Consider the benchmark case where investors (i.e., market participants other than the insider)
are symmetrically informed about firm type. This case essentially replicates Myers and Majluf
(1984). In the static model of Myers and Majluf there is no stock return prior to the issuance
decision. Since we are mainly interested in how potential issuers respond to stock returns, we
allow for some new public information to arrive at date 1 and move the stock price. Specifically,
at the start of date 1 all market participants observe a public signal θ = V + , where is
a standard normal random variable.3 Examples of public information that θ summarizes are
3
The exact specification of does not play any particular role in this model; we choose one (the standard normal
distribution) for completeness.
5
analysts’ earnings forecasts, news about the demand for the firm’s products, etc.
Since information is symmetric among investors, the stock trades at P1 = E(V | θ) at date
1. After observing this stock price, the insider decides whether to sell her shares in an SEO. Let
ISEO = 1 (= 0) denote the decision to sell (not to sell). Clearly, GU always announces ISEO = 0,
since she is better off consuming VG = 1 rather than selling undervalued shares. Now consider B.
If GL is expected to announce ISEO = 1 with positive probability given P1 , B strictly prefers to
announce ISEO = 1 in order to sell overpriced equity. If GL is expected to announce ISEO = 1
with zero probability given P1 , B is indifferent between ISEO = 1 and ISEO = 0, since in either
case she receives VB = 0 (recall that B does not have a project). In such cases we assume without
loss of generality that B announces ISEO = 0.4 Thus, B fully pools with GL for any realization
of P1 .
Our main interest is in what GL decides to do. Suppose that, given a particular realization
of P1 , investors anticipate GL to announce ISEO = 1. Before the announcement, the expected
firm value is P1 . Upon observing ISEO = 1, investors revise their expectation of the firm value
downward:
P2 = E (V | θ, ISEO = 1) =
λP1
< P1 .
λP1 + 1 − P1
(1)
This is the familiar SEO announcement effect. Before the announcement, investors put positive
weights on the firm being one of GU , GL , and B. Since GU never issues equity, the SEO announcement reveals that the firm type is not GU and hence increases the probability that it is
B. Importantly, the SEO announcement is the last informational event in the model; therefore,
the date-2 stock price P2 is also given by the revised expected value in (1). In making the SEO
decision, then, GL compares P2 to her opportunity cost:
ISEO = 1 ⇔
λP1
1
≥ .
λP1 + 1 − P1
R
(2)
Let P 1 denote the value of P1 that satisfies (2) as an equality. If P1 ≥ P 1 , GL announces ISEO = 1.
If P1 < P 1 , GL announces ISEO = 0. Since R < 2, P 1 > 1/2. In words, an equity issue takes
place only after a positive stock return from date 0 to date 1.
4
Of course, it is also a best response to announce ISEO = 1 in such cases. We view announcing ISEO = 0 as the
more robust outcome. For example, announcing ISEO = 0 would be a strict best response under arbitrarily small
fixed costs of equity issuance.
6
Case 2: Privately-informed investors
Consider now the main case of interest in which the stock price is determined by trading activity
among heterogeneously-informed investors. Following standard market microstructure models,
we introduce three types of investors: informed traders, noise traders, and market makers.
Informed traders correspond to professional investors that conduct research and produce information for trading purposes in practice. At the start of date 1, informed traders privately
observe a signal s ∈ {L, H} on firm type. If the firm type is bad, s = L. If the firm type is good,
s = H with probability q and s = L with probability 1 − q. We assume that the insider does not
observe the realization of s.5 Notice that from the perspective of GL , there is some chance (q)
that informed traders recognize the firm type, but there is also some chance (1 − q) that this is
not the case.6
We model informed traders as a continuum of competitive (e.g., price-taking) agents. Informed
traders hold no shares of the firm’s stock at date 0. Suppose that the stock trades at price P t at
some subsequent date t. Then, informed traders’ demand for the stock at date t is given by

 1 − Pt if s = H
ht =
 0
if s = L.
(3)
The demand function ht reflects two assumptions that we make about informed traders. First,
informed traders face a short-selling constraint. When their expected return on the stock is
negative (i.e., when s = L), informed traders do not trade because of the binding short-selling
constraint.7 Second, when s = H, informed traders’ demand is a decreasing function of Pt . This
assumption captures in a reduced-form way various trade-offs informed traders face in selecting
their portfolios. The most straightforward interpretation is that informed traders find it costly
to hold undiversified positions in the stock (due to risk-aversion or institutional diversification
objectives). Given such costs, a higher expected gain 1 − Pt induces larger optimal positions. An5
Of course B infers that s = L by construction.
The assumption that H reveals the firm type simplifies the analysis considerably, but otherwise this assumption
is not essential. Our main results go through as long as H is a sufficiently precise (but not necessarily noise-free)
signal of a good type.
7
In practice, institutional investors such as mutual or pension funds (which the informed traders are modelled
after) are typically restricted from short-selling. For example, Almazan et al. (2004) document that most mutual
funds have either explicit rules or implicit policies against short-selling activity.
6
7
other interpretation is that there are sidelined informed traders who find the stock less attractive
relative to their alternative investment opportunities. Under this interpretation, an increase in
1 − Pt triggers a purchase by the marginal sidelined traders.8
The firm’s stock is publicly traded at dates 1 and 2. Consider date 1 first. Noise traders
demand a random number of shares n1 , which is a draw from the standard normal distribution.
Let x1 denote the number of shares that informed traders choose to trade at the market price P 1 .
A risk-neutral and competitive group of market makers clear the market at price P 1 , which equals
the expected value of V conditional on the aggregate date-1 order flow d 1 ≡ x1 + n1 . Notice that
the informed traders’ demand x1 depends on their private signal s. Therefore, P1 can be written
as a function of the pair (s, n1 ). Essentially, the pricing rule P1 corresponds to a noisy rational
expectations equilibrium.
The market equilibrium at date 2 is similar to date 1. Noise traders demand a random number
of shares n2 , which is a draw from the standard normal distribution. Let x2 denote the number
of shares that informed traders choose to trade at the market price P 2 . The market makers clear
the market at price P2 , which equals the expected value of V conditional on the aggregate date-1
order flow d1 , the aggregate date-2 order flow d2 ≡ x2 + n2 , and the insider’s SEO announcement
ISEO .
Equilibrium in the stock market
We start by characterizing informed traders’ optimal trades x1 and x2 given the stock prices P1
and P2 :

 [1 − P1 , P1 − P2 ] if s = H,
[x1 , x2 ] =
 [0, 0]
if s = L.
(4)
If informed traders observe H, they learn that the firm type is good and add x1 = 1 − P1 shares
to their holdings at date 1. A decrease (increase) in the stock price at date 2 makes informed
traders buy more (sell some of their existing) shares. If informed traders observe L, their optimal
demand is zero and hence they do not trade at either date (recall that the short-selling constraint
8
The reduced-form approach of modelling ht avoids the need to solve informed traders’ dynamic optimization
problem (e.g., maximizing multi-period utility under risk aversion). This simplification is unlikely to have a significant bearing on our qualitative results. The functional form 1 − Pt is chosen for expositional convenience and can
easily be generalized to an arbitrary downward-sloping demand function D(Pt ).
8
binds in this case).
Given the optimal strategy of informed traders, stock prices at date 1 and date 2 are calculated
as conditional expectations of firm value:
P1 = E (V | d1 = x1 + n1 ) ,
(5)
P2 = E (V | d1 = x1 + n1 , d2 = x2 + n2 , ISEO ) .
Notice that since VG = 1 and VB = 0, P1 and P2 correspond to conditional probabilities that the
firm type is good. The equations in (5) characterize P1 and P2 implicitly. For example, P1 is a
function of d1 = x1 + n1 , but informed traders’ demand x1 depends on P1 . We delegate the full
characterizations of P1 and P2 to the Appendix and state here some of their standard properties
without proof:
• The market is semi-strong form efficient: E (P2 |d1 ) = P1 .
• The date-1 stock price P1 is strictly increasing in the date-1 order flow d1 . The same is true
for P2 with respect to d1 and d2 provided that the SEO announcement does not fully reveal
the firm type.
The SEO decision
As in Case 1, GU always announces ISEO = 0 and B always pools with GL . Our main interest is
again in what GL does. Suppose that, given the particular realization of P1 , market participants
anticipate GL to announce ISEO = 1. Before the announcement, the expected firm value is P1 .
Upon observing ISEO = 1, market makers revise their expectation of the firm value downward:
E (V | d1 , ISEO = 1) =
λP1
< P1 .
λP1 + 1 − P1
(6)
Notice that (6) is identical to (1): the SEO announcement reveals that the firm type is not G U
and hence increases the probability that it is B. What is different in this case is that the date-2
stock price P2 is not given by (6). Rather, P2 is to be determined through date-2 trading activity;
that is, as a function of the date-2 aggregate order flow d2 . In making the SEO decision, then,
9
GL compares the expected date-2 price to her opportunity cost:
ISEO = 1 ⇔ EG (P2 | d1 , ISEO = 1) ≥
1
,
R
(7)
where the G-subscript signifies that the expectation is taken by GL . The following result provides
a comparison of the two expectations in (6) and (7):
Proposition 1: For any realization of P1 ,
EG (P2 | d1 , ISEO = 1) > E (P2 | d1 , ISEO = 1) =
λP1
.
λP1 + 1 − P1
(8)
The intuition for Proposition 1 is simple. Upon observing the announcement ISEO = 1, market
makers adjust their estimate of firm value downward. The revised estimate also constitutes market
makers’ expectation of P2 right after the announcement. However, P2 is random and its realization
depends on date-2 aggregate order flow d2 . In predicting d2 , the insider GL has an informational
advantage relative to the market makers. This is because the insider knows her own type (that
is, that the type is good), and hence assigns a higher probability to the event that informed
traders have observed a high signal relative to market makers’ expectations. In essence, the
insider expects informed traders’ demand to positively surprise the market makers and keep the
stock price relatively high (or prevent it from falling too much). To put it differently, the insider
anticipates stronger date-2 aggregate order flow d2 than the market makers do.9
Similar to Case 1, one can show that the expected value in (7) is strictly increasing in P 1 .
Therefore, there exists a cutoff P1∗ at which (7) is satisfied as an equality. If P1 ≥ P1∗ , GL
announces ISEO = 1. If P1 < P1∗ , GL announces ISEO = 0. In the Appendix, we show that
P2 < P1 with probability one; therefore, EG (P2 | d1 , ISEO = 1) < P1 . Since R < 2, it follows that
P1∗ > 1/2. In words, an equity issue takes place only after a positive stock return from date 0 to
date 1.
9
Formally, the distribution of d2 conditional on the firm type being good first-order stochastically dominates the
distribution of d2 conditional on public information.
10
Comparison of Case 1 and Case 2
In Case 1, date-1 stock price P1 is set in response to the public signal θ. In this case, GL (and
hence B, since B fully pools with GL ) decides to sell equity if P1 exceeds the cutoff P 1 . In Case 2,
P1 is set in response to date-1 aggregate order flow d1 which acts as a noisy indicator of informed
traders’ signal. In this case, GL (and again B) decides to sell equity if P1 exceeds the cutoff P1∗ . In
both cases, the cutoff values exceed 1/2, which represents the unconditional expected firm value.
Interpreting 1/2 as the initial stock price in the model, equity issues in both cases follow positive
stock returns.
In which case a given P1 is more likely to trigger an equity issue? From Proposition 1, it follows
that P1∗ < P 1 . Not surprisingly, there are no equity issues following very low prices (P1 < P1∗ ).
Also not surprisingly, sufficiently high prices (P1 > P 1 ) trigger equity issues regardless of how P1
is reached. The interesting region is P1 ∈ (P1∗ , P 1 ). In this region, an equity issue takes place
only if P1 is set in response to informed trading activity. If the same price obtains via the arrival
of public information, firms choose not to issue equity. This prediction constitutes the basis of
our main empirical tests.
Discussion and further issues
Alternative modelling choices and assumptions: Our model highlights the role privatelyinformed investors play in the equity issuance process. To make the comparison to the benchmark
case of symmetrically-informed investors as clear as possible, we present the benchmark case as
a separate model; that is, the two models are not nested. The non-nested model formulation
captures much of the intuition we want to illustrate. Furthermore, one can argue that the two
cases that we analyze correspond to different conditions firms may face at the time their issuance
needs arise (e.g., a recent runup in the stock price of a firm may have been triggered by public
news or by order flow). We have also analyzed a nested version of the model in which stock prices
react to both public news and informed trading activity. Similar to the current model, the nested
model delivers the prediction that a particular realization of P1 is more likely to be followed by
an equity issue announcement if P1 reached through higher aggregate order flow d1 (rather than
a higher public signal θ).
11
Another modelling choice concerns insiders’ information regarding the realization of informed
traders’ signal. Throughout the analysis we make the conservative assumption that the insider
has access to public information only in assessing the likelihood of informed trading. In practice,
firms may also make use of non-public information about demand conditions when making their
issuance decisions. For instance, most top-tier investment banks also have trading desks that
possess private information about their institutional clients’ demand. Anecdotal evidence suggests
that such private information finds its way into firms’ issuance decisions through the assistance of
investment bankers. In the notation of the model, firms may observe not only d1 (as we assume),
but also x1 or a noisy version of x1 . In the interest of brevity we do not formally analyze this
alternative model; we simply state its (not-so-surprising) prediction that good types are more
likely to issue when they learn informed traders’ signal to be high.
The relationship of the model to market timing: Our model describes firms’ financing
choices in an asymmetric information context. Market timing, on the other hand, refers to firms’
tendency to issue equity following high stock returns. In principle, these two phenomena are
distinct from each other. Past stock returns do not play any particular role in the simple static
adverse selection model of Myers and Majluf (1984). Perhaps due to the wide influence of that
model, market timing is typically viewed in the literature as a behavioral phenomenon rather than
a manifestation of asymmetric information. While the behavioral view is of interest on its own,
there is a natural connection between market timing and asymmetric information once the simple
static framework is replaced by a dynamic one. By reducing misvaluation from the viewpoint of
good types, high stock returns alleviate the adverse selection problem and trigger equity issues.
In our model, the adverse selection problem is initially severe, in the sense that good types are
not willing to issue equity at the average price 1/2. Good types decide to issue only following
sufficiently high stock returns, captured by relatively high values of P1 .
Interpretation of date-2 informed demand as gradual diffusion of information: We
model informed traders as receiving their private information at date 0 and trading on it at dates
1 and 2. In particular, if date-2 stock price is low informed traders step in and purchase additional
shares. An alternative interpretation of the model is that information diffuses gradually among
investors. Under this interpretation, the purchases made by early informed traders at date 1
12
signal to the insider that more informed traders will receive their information and purchase the
stock at date 2.
Announcement effects and announcement-to-offer returns: In the model, the registration
period (i.e., the period between the SEO announcement and the actual offer) is compressed into a
single trading date, date 2. Thus, date-2 trading activity should be interpreted as being spread out
throughout the whole registration period. This interpretation has implications for announcement
effects and announcement-to-offer returns, which we now discuss. Essentially, the strength of
demand during the registration period determines how high the offer price P 2 will be relative
to the stock price just prior to the announcement P1 . Part of the return P2 /P1 is realized
around the announcement day (i.e., “the announcement effect”) and part of it is realized as the
announcement-to-offer return. Notice that, unlike in Myers and Majluf (1984), both of these
returns are functions of firm type and random noise trader demand. Good-type SEO announcers
expect higher values of both returns than bad-type SEO announcers do, since good types count
on informed traders to “support the price” through their high demand. Nevertheless, the actual
realizations of the announcement effect and the announcement-to-offer return for a good type
may be negative due to low noise trader demand.
2. The empirical setup
2.1 Data and sample construction
The initial sample of potential SEO announcers consists of all firm-quarter observations in the
intersection of CRSP, COMPUSTAT and CDA/Spectrum Institutional (13F) databases for 84
quarters from 1985:01 to 2005:04. We restrict the sample to exclude financial firms (SIC codes
6000-6999), utilities (SIC codes 4900-4949), and firms that have issued equity within the previous
two quarters. In most of our analysis, we also exclude firms that had less than 5% institutional
ownership in the previous four quarters. Our final sample of potential SEO announcers consists
of 244,475 firm-quarter observations.
We use the CDA/Spectrum database to calculate the institutional ownership and demand
variables. CDA/Spectrum reports quarterly snapshots of institutional investors’ portfolios ex-
13
tracted from 13F reports filed with the SEC.10 The types of institutions covered by this database
are banks, insurance companies, investment companies, independent investment advisors, and
an “others” category that includes mainly pension funds. As in previous studies that use this
data, we approximate institutions’ quarterly trades by calculating the difference in their holdings
between two consecutive reports.
The initial SEO sample consists of all SEO announcements reported by the Securities Data
Corporation (SDC) for firms that are in our final sample of potential SEO announcers. We restrict
the sample to exclude spinoffs, unit offers, and rights offerings. Our final SEO sample consists of
2,961 SEO announcements and 2,784 completed offers.
Throughout the analysis, we calculate excess stock returns by subtracting the returns of benchmark portfolios that are rebalanced quarterly using a double sort on size and book-to-market
(B/M) ratio. To construct the benchmark portfolios used in quarter q, firms are sorted into five
size portfolios using their market equity at the end of q − 2 and the breakpoints corresponding to
NYSE market equity quintiles at the end of q − 2. Within each of these size portfolios, firms are
then sorted into five B/M portfolios using the book equity reported in quarter q − 2 divided by
market equity at the end of q − 2.11
2.2 Definitions and summary statistics of institutional demand variables
Institutional investor demand for a given stock can be measured in several different ways. In
this regard, we make two particular choices. First, we use variables that reflect changes in (as
opposed to levels of) institutional holdings. While clearly informative about institutions’ views,
level variables are affected by other factors that create noise for our purposes. For example, one
would expect certain types of stocks (e.g., larger, more liquid) to have higher levels of institutional
ownership. Change variables track demand shifts more closely; they also provide more meaningful
comparisons to the effects of stock returns. Second, our demand variables are count-based (e.g.,
the number of institutions purchasing the stock). The alternative would be to aggregate trades
10
Institutions are allowed to omit reporting small positions, defined as those that meet both of the following two
criteria: (i) the institution holds less than 10,000 shares of a given issuer; (ii) the aggregate fair market value of the
holdings in the same issuer is less than $200,000.
11
We calculate benchmark portfolios with a two-quarter lag to address the concern that issuers may have high
returns prior to their SEO announcements for idiosyncratic reasons and may not share the characteristics of the
firms in the resulting higher size or lower B/M portfolios. The results are very similar when a one-quarter lag is
used instead.
14
across institutions (e.g., change in institutional ownership). While the two approaches provide
similar qualitative results, count-based measures perform better quantitatively.12
Specifically, we construct two institutional demand variables for each stock i -quarter t pair.
Buy minus Sell is the number of institutions increasing their holdings (including position initiations) minus the number of institutions decreasing their holdings (including position terminations)
in stock i in quarter t. Initiation is the number of institutions initiating new positions in stock i
in quarter t. Both variables are normalized by the number of institutions holding the stock at the
beginning of quarter t. In calculating the institutional demand variables, we restrict attention to
firms that exhibit more than 5% institutional ownership during the past four quarters.
Notice that Initiation is a component of Buy minus Sell by construction. Since Buy minus Sell
provides a more complete representation of institutional demand, we start our analysis in Section
3 using this variable. However, as we show below, all the explanatory power of Buy minus Sell
is due to its inclusion of Initiation. Once we establish this result, we use Initiation as our main
empirical construct in the rest of the analysis.
Table I reports the summary statistics of the institutional demand variables. Along with Buy
minus Sell and Initiation, we also report statistics on Change in ongoing positions, which is
defined as the difference between Buy minus Sell and Initiation.
As Panel A.1 shows, Buy minus Sell has a quite symmetric distribution with a median of zero.
Initiation, which is by construction positive, has a median of 0.111 (i.e., about 11 new institutions
per 100 existing institutional shareholders initiate positions in a typical firm-quarter). Notice
that both demand variables are substantially higher in the most recent quarter prior to the SEO
announcement (Panel A.2) and the offer quarter (Panel A.3). This higher demand appears to be
mainly driven by new institutional shareholders; Change in ongoing positions does not exhibit a
significant increase around the time of the SEO.
Panels B through D of Table I report various correlations and autocorrelations. These statistics are calculated as time-series averages of quarterly cross-sectional correlations. Institutional
demand exhibits moderate positive autocorrelation. It is also positively correlated with past and
contemporaneous (i.e., same quarter) excess stock returns. The positive sign of these correlations
12
Similarly, Sias, Starks, and Titman (2006) show that contemporaneous returns are more strongly related to
changes in the number of institutional shareholders than changes in institutional ownership.
15
is in line with the finding in previous studies that institutional investors tend to buy recent winners
(see, for example, Grinblatt, Titman, and Wermers (1995)). The fact that the contemporaneous
return correlation is considerably higher than the lagged return correlations is indicative of the
price impact of institutions’ trades. That is, stock prices seem to react positively to purchases by
institutional investors. The demand variables are positively correlated with the market-to-book
ratio and negatively correlated with firm age and firm size (measured by book assets). However, none of these correlations are very high. Overall, the institutional demand variables exhibit
substantial variation that is independent of past and contemporaneous stock returns and firm
characteristics.
3. Results
3.1 The equity issuance decision
In this section, we present our main results that link the equity issuance decision to past stock
returns and institutional demand. The variable of interest is the probability that a firm-quarter is
associated with an SEO announcement, which we take as an indicator of intention to issue equity.
For ease of interpretation, we first report SEO announcement probabilities in simple univariate
and bivariate sorts using past stock returns and institutional demand as the sorting variables. We
then check the robustness of the basic patterns in these simple sorts by estimating multivariate
probit regressions that include several control variables.
The unconditional quarterly probability that a firm announces an SEO in our sample is 1.21%.
It is well-known that firms are more likely to issue equity following high stock returns. We start
by verifying this finding in our sample. In each quarter, we sort firms into quintiles based on
previous quarter’s excess stock return. We then calculate the percentage of firms within each
quintile that announce an SEO during the quarter. Panel A in Table II reports the time-series
averages of these percentages.13 As expected, stock returns have a strong impact on the likelihood
of SEO announcements. Firms in the highest return quintile are almost ten times more likely to
announce SEOs than firms in the lowest return quintile. We also report the results separately
13
The t-statistics reported in Table I are computed using Newey-West correction for heteroscedasticity and serial
correlation.
16
for the sub-samples of firms with more and less than 5% institutional ownership.14 Stock returns
positively affect the likelihood of SEO announcements in both sub-samples. Notice however that
firms with low institutional ownership are substantially less likely to announce SEOs in each
return quintile.
Next, we analyze the likelihood of an SEO announcement as a function of both stock returns
and institutional demand. At the beginning of each quarter, firms are sorted independently into
quintiles based on previous quarter’s excess stock return and institutional demand. Within each
return-demand group, we calculate the percentage of firms that announce an SEO during the
quarter. The results are reported in Panels B through D of Table II.
In Panel B, the institutional demand variable is Buy minus Sell. As the panel shows, stock
returns and institutional demand have a strong interaction effect on the likelihood of SEO announcements. When institutional demand is in its lowest quintile, moving from the lowest to
the highest stock return quintile increases the announcement probability by only 1.08%. While
this increase is statistically significant, its magnitude is quite small (recall that the unconditional
SEO announcement probability is 1.21%). When institutional demand is in its highest quintile,
however, we see that stock return has a very strong effect on the SEO announcement probability.
Moving from the lowest to the highest return quintile in this case increases the announcement
probability from 0.39% to 4.06%. Conversely, institutional demand has a weakly positive effect
when the stock return is low or moderate, but a very strong positive effect when the stock return
is high.
The demand variable Buy minus Sell is composed of position initiations as well as changes to
ongoing positions of institutional investors. Position initiations may constitute a relatively less
noisy indicator of investors’ views on the stock. Changes to ongoing positions are likely to be
driven in part by flows and liquidity considerations. Furthermore, issuers may be more concerned
about demand from new investors than demand from their existing shareholders.15 To see which
component of institutional demand matters more for the equity issuance decision, we repeat the
double-sorting exercise described above for each component of Buy minus Sell.
14
Recall that we calculate the institutional demand variables only for firms with more than 5% institutional
ownership during the past four quarters. Accordingly, most of our analysis below focuses on the sample of firms
with more than 5% institutional ownership.
15
Placing the issue with the existing shareholders may require larger price concessions (e.g., as compensation for
the larger undiversified positions these shareholders would have to hold).
17
Panel C of Table II reports the results for Initiation. The interaction effect discussed above
exhibits itself even more strongly in this case. Consider the highest return-lowest institutional
demand cell. The average excess stock return in this case is 37.60% (not reported in the table).
The probability of an SEO announcement, however, is only 0.60%; that is, less than half of the
unconditional probability. Similarly, in the lowest return-highest demand cell, the probability of
an announcement is only 0.70%. When both the return and institutional demand are in their
highest quintiles, however, the probability of an announcement jumps up to 4.21%.
In Panel D of Table II, the institutional demand variable is Change in ongoing positions, which
is the difference between Buy minus Sell and Initiation (i.e., the number of existing institutional
shareholders that increase their holdings minus the number of those that decrease or terminate
their holdings). Notice that institutional demand defined this way has very little impact on the
likelihood of equity issuance. Within any return quintile, moving from the lowest to the highest
demand quintile causes a very small and at best marginally significant increase in the probability
of an SEO announcement. It appears that almost all of the explanatory power of Buy minus Sell
is due to its inclusion of Initiation.
The results in Table II constitute our main findings regarding the impact of institutional
investors on the equity issuance decision. We now discuss some robustness issues. First, one may
be concerned about the fact that institutional demand and stock returns are positively correlated.
From Panel C of Table I, we see that the demand-return correlations are at moderate levels and
hence unlikely to cause multi-collinearity problems. However, even moderate levels of correlation
may obscure the effects of each sorting variable.16 Second, the analysis so far does not control for
other potential determinants of the equity issuance decision.
16
For example, there could be substantial variation in returns within each return quintile, and institutional
demand variables may in part be picking up the effect of such variation on the announcement probability. In
unreported analysis, we find that this concern is negligible. In the highest return quintile, for instance, the difference
in average returns between the highest and the lowest Buy minus Sell quintiles is only 0.56%.
18
We address both concerns by estimating probit regressions of the following form:
Pr(Announce i,t ) = F (c0 + c1 High Demand i,t−1 + c2 Medium Demand i,t−1 + c3 Excess Return i,t−1
+c4 Excess Return i,t−1 × High Demand i,t−1
+c5 Excess Return i,t−1 × Medium Demand i,t−1
(9)
+c6 Xi,t−2 + c7 Quarter fixed effects t ).
In (9), Announce i,t is an indicator variable that takes the value of one if firm i announces an SEO
in quarter t and zero otherwise. High Demand i,t−1 (Medium Demand i,t−1 ) is a dummy variable
that takes the value of one if institutional demand in quarter t − 1 is in the highest quintile
(middle three quintiles) and zero otherwise. The regression includes the excess stock return in
quarter t − 1 and its interactions with the demand dummies.17 The vector Xi,t−2 denotes a set
of control variables measured as of the end of quarter t − 2. The control variables are as follows.
First, we include the excess stock return in quarter t − 2. The concern here is that institutional
demand in quarter t − 1 may be positively correlated with previous quarter’s stock return (see
Table I Panel C). Second, we control for institutional ownership (the fraction of firm’s stock held
by institutional investors) at the end of quarter t − 2. Third, we control for a number of firm
characteristics that previous research has identified to be correlated with equity issues. These
variables are Firm age measured as the logarithm of the number of quarters since IPO, Firm size
measured as the logarithm of book value of assets, Market-to-book ratio, Profitability measured
as EBITDA divided by book assets, and Book leverage. We estimate (9) by including time fixed
effects for each quarter t and clustering observations at the firm level.
Table III reports the results, which replicate the basic patterns in Table II. The stock return
in quarter t − 1 has a significant effect on the SEO announcement probability in quarter t, but
the magnitude of this effect is quite small. An increase of 100% in excess stock return increases
the announcement probability by only 0.79% for firms with low institutional demand. When
institutional demand is measured by Initiation, moving from the lowest to the highest demand
quintile adds 2.26% to the announcement probability. There is also a strong interaction effect; an
17
The magnitude of the interaction effect in nonlinear models (such as probit) does not equal the marginal effect
of the interaction term. We use the methodology described in Ai and Norton (2003) to calculate the magnitude
and standard errors of interaction effects.
19
announcement is substantially more likely following high stock returns that are accompanied by
high institutional demand. Both the level and the interaction effects of institutional demand are
largely reduced when demand is measured by Change in ongoing positions. The control variables
exhibit the expected signs. The probability of announcement increases with the stock return in
quarter t − 2, institutional ownership, firm size, market-to-book ratio, and book leverage, and
decreases with firm age.
To summarize, firms are more likely to issue equity following periods in which both their stock
return and the institutional demand for their stock are relatively high. High stock returns that
are not accompanied by high demand have very little impact on the likelihood of issuance. The
results are primarily driven by demand from new institutional investors. Whether the existing institutional shareholders exhibit positive or negative demand does not appear to affect the issuance
decision in a material way.
3.2 Stock returns in the announcement-to-offer period
By announcing an SEO, a firm indicates its intention to issue equity. The offer itself typically
takes place several weeks after the announcement (in our sample, the median number of trading
days between the announcement and the offer is 21). At the time they announce their SEOs,
firms are primarily concerned about the offer price at which they will be able to sell their shares.
Accordingly, we now analyze stock returns during the announcement-to-offer period. In particular, we ask whether stock returns following an announcement in quarter t can be predicted based
on institutional demand in quarter t − 1. Whether one should expect to find any such return predictability depends on the extent to which quarter t−1 institutional demand is public information
at the time of the quarter t SEO announcement. Part of the information content of institutional
demand is likely to be already reflected in the stock price at the time of the announcement due
to the price impact of trades that the demand variable is composed of. However, there may still
be some residual private information that institutional demand reflects.
In this section and in the remainder of the paper, we focus on Initiation as the institutional
demand variable. Specifically, we assign each SEO announcement into one of five demand groups
using the Initiation quintile breakpoints obtained from all potential issuers in the quarter prior
to the announcement. Since firms in the highest demand group are the most likely to announce
20
as documented in the previous section, there are more SEO announcements in this group than in
other demand groups.
We are interested in comparing announcement-to-offer period stock returns across different
institutional demand groups. However, this comparison is problematic because the length of the
registration period (the period between SEO announcement and the offer) varies across SEOs.
We use two different approaches to deal with this problem. First, we analyze 60 trading-day
post-announcement returns for all SEO announcements (i.e., including those that are not followed by an offer within 60 trading days). This approach provides a clear representation of the
effect of institutional demand on post-announcement returns. However, some of the observations
correspond to delayed or withdrawn offers. As a second approach, we focus on SEOs that are
completed within 60 trading days from the announcement.18 This reduces the possibility that
the announcement-to-offer returns are dominated by large positive/negative returns associated
with SEOs that stay in registration for extremely long periods. While it allows us to calculate
announcement-to-offer returns, this approach is subject to a sample selection bias: firms that
experience negative returns following their announcements are more likely to delay or withdraw
their offers.
Panel A of Table IV reports the results with the two approached described above. In the
first two columns, the sample includes all SEO announcements. The first column reports the
announcement effect, defined as the excess stock return in the (−1, +1) three-day window around
the announcement. The second column reports the excess stock return in the (+2, +60) window
following the announcement. In the next two columns, the sample is restricted to SEO announcements that are followed by an offer within 60 trading days. The third column reports the excess
stock return in the (+2, of f er) window following the announcement; that is, from two days after
the announcement to the last trading day before the offer. The fourth column reports the offer
discount, defined as the offer price divided by the last closing stock price.
As documented by previous studies, the announcement effect is on average negative. In other
words, the market greets an equity issue as bad news. More importantly for our purposes, the
negative reaction to SEO announcements is similar across different pre-announcement institutional
18
About 80% of announced SEOs are completed within 60 trading days. The choice of 60 trading days (about
one quarter) is admittedly arbitrary; however, the results are robust to changing this threshold.
21
demand groups. The offer premium also does not vary in a statistically significant way across
different demand groups. In contrast, stock returns following the announcement period increase
with pre-announcement institutional demand. While firms with relatively low pre-announcement
demand (those in the bottom four demand groups) experience negative returns that add to their
negative announcement effect, firms in the highest demand group experience a positive return of
3.87% that more than offsets their negative announcement effect of -1.89%. As Panel B shows,
these results remain qualitatively similar in regressions that control for past excess returns and
institutional ownership.
The basic result that emerges from Table IV is that issuers with high pre-announcement institutional demand benefit from relatively better stock-price performance in the post-announcement
period. The finding raises two questions. First, does institutional demand predict short-term returns in general or only around SEO announcements? In unreported analysis, we find that there
is very little return predictability based on institutional demand in general. This indicates that
the information content of institutional demand is somewhat unique to either SEO episodes or the
types of firms that resort to SEOs. Second, do issuers have access to privileged information about
institutional demand at the time they make their SEO decisions? Anecdotal evidence suggests
that issuers indeed obtain such information through the connections of their investment bankers.
Furthermore, the results in the previous section show that the issuance decision is highly sensitive
to institutional demand. However, our findings are not sufficient to conclude that issuers have
privileged information about institutional demand. As in the model of Section 1, issuers and market makers may both have access to the same information regarding trading activity (i.e., both
observe aggregate order flow). The sensitivity of the issuance decision to institutional demand
may simply reflect the fact that institutional demand is part of aggregate order flow.
3.3 Stock returns around and following the offer
In this section, we focus on the relationship between offer-quarter institutional demand and the
stock returns in the period immediately following the offer. This is the period during which market
participants learn about the outcome of the offer and in particular about institutional demand
for the issuer’s stock. Our objective is to investigate how the market reacts to this resolution of
uncertainty.
22
We start by sorting all completed SEOs in our sample into five groups based on the Initiation
demand variable calculated in the offer quarter. Panel A of Table V reports the average excess
returns of each demand group over various periods both in the offer quarter and the next quarter.
Not surprisingly, the excess stock returns in the offer quarter are highly correlated with the
institutional demand in the same quarter: SEOs with offer quarter’s demand in the top quintile
experience 21.71% excess returns in the offer quarter, while those in the bottom quintile experience
-2.82% excess returns in the same period. The results are similar in the three-day period around
the offer day: SEOs in the top demand quintile experience excess stock returns that are 1.81%
higher on average than those in the bottom demand quintile in the three-day period. This pattern
is repeated in the post-offer part of the offer quarter. High-demand issuers experience excess stock
returns that are 7.15% higher than low-demand issuers.
Although the magnitude of these return differentials are substantial, this analysis is problematic
because the causality is ambiguous. Since we do not observe the within-quarter institutional
demand, the results we document here are also consistent with institutional feedback trading:
institutions may demand more of SEOs that perform well in the offer quarter, around the offer
day, and following the offer.
In order to alleviate the causality concern, we focus on the quarter immediately following the
offer quarter in the last column of Panel A of Table V. SEOs in the top demand quintile in the
offer quarter outperform those in the bottom quintile by 5.83% in the next quarter. In unreported
analysis, we find that most of this return differential is obtained in the first half (45 days) of the
quarter. This is consistent with other market participants observing the institutional demand
during the offer quarter through the 13F reports that institutional investors have to file within
45 days of the end of the offer quarter. Combining the offer quarter’s post-offer return and the
return in the quarter after the offer results in a 13% return differential between high-demand and
low-demand issuers, which is quite large for a period of less than two quarters. This indicates
that market participants pay a great deal of attention to the resolution of uncertainty regarding
institutional participation in the SEO.
We repeat the same analyses in a multivariate setting and report the results in Panel B of
Table V. Here we find that the effect of offer quarter’s institutional demand on excess returns are
robust to the inclusion of past returns, lagged institutional ownership and quarter fixed effect.
23
The only exception is the effect of institutional demand on the offer premium which becomes
statistically insignificant.
<TO BE COMPLETED>
3.4 Long-run stock returns following the offer
The results so far show that firms strongly time their equity issues to coincide with increased
institutional demand. Furthermore, issuers with relatively high demand exhibit better stock
return performance than other issuers around the time of their SEOs. Perhaps firms view periods
of increased institutional demand for their shares as windows of opportunity in which they can
sell overvalued equity. If so, one would expect high-demand issuers to underperform low-demand
issuers in the long run. Since the previous section already documents the effect of institutional
demand on the stock performance in the quarter immediately following the offer quarter (t + 1),
we focus here on the long-run returns (1/3/5 years) starting from the next quarter (t + 2).
We use two different approaches to detect long-run performance: event-time and calendartime. The event-time approach consists of calculating the average event-time long-run returns,
which are the sample averages of SEO-specific average of monthly excess returns calculated during
one/three/five years following the quarter after the issuance. Since this approach can only be used
in a descriptive fashion for reasons pointed out in Barber and Lyon (1997) and Brav (2000), our
discussion and statistical tests will be focused on the calendar-time approach. This approach
consists of calculating the time-series monthly three-factor alphas of portfolios of firms that have
issued SEOs in the past one/three/five years.
As reported in Panel A of Table VI, SEO firms in our sample experience relatively low stockreturn performance in the five years following the SEO. The long-run underperformance following
SEOs is widely documented in the literature (see, for example, Loughran and Ritter (1995) and
Brav, Geczy, and Gompers (2000)). The question of interest for us is whether this underperformance varies as a function of institutional demand around the time of the SEO. Since long-run
performance tests tend to have low power, we split our SEO sample into two groups instead of the
five groups we use in earlier sections.19 In particular, we sort all SEOs in our sample on Initiation
19
The results are qualitatively similar if we split the sample into five groups.
24
into two demand groups. Panel B (C) of Table VI reports the average SEO performance for
SEO groups sorted on Initiation in the pre-announcement (offer) quarter. While high-demand
SEOs seem to underperform slightly more than low-demand SEOs, the differences between the
two groups are not statistically significant.
<TO BE COMPLETED>
4. Conclusion
It is well-known that firms are more likely to issue equity following periods of high stock returns.
We document that firms do so only if high return periods coincide with strong demand from
institutional investors. Stock price increases that are not accompanied by institutional demand
have little impact on the likelihood of equity issuance. Issuers with strong pre-announcement
institutional demand outperform those with low demand during the announcement-to-offer period;
that is, they are able to issue shares at relatively more attractive prices. Also, issuers with high
institutional demand around the time of the offer experience significant gains in the short run
following the offer. These gains are not reversed in the long run. Overall, our results highlight
the importance of asymmetric information problems in equity issuance. Firms appear to pay
considerable attention to who stands by their stock prices in deciding whether to issue equity.
25
Appendix: Derivations and proofs
Derivation of stock prices in Case 2: Let f denote the probability density function of the
standard normal distribution. First, consider the date-1 stock price:
P1 = E (V | d1 ) =
1
2 qf
(d1 − (1 − P1 )) + 12 (1 − q)f (d1 )
1
.
1
1
2 qf (d1 − (1 − P1 )) + 2 (1 − q) + 2 f (d1 )
(A1)
(A1) implicitly defines P1 as a function of d1 = x1 + n1 . It is easy to show that a solution
P1 (d1 ) always exists. Furthermore, one can show that the solution is unique [PROOF TO BE
COMPLETED]
Next, consider the date-2 stock price. If GL is expected to announce ISEO = 0 in equilibrium
given the realization of P1 , then

 P1 if ISEO = 0,
P2 =
 0 if I
SEO = 1.
(A2)
If GL is expected to announce ISEO = 1 in equilibrium given the realization of P1 , then

 1
if ISEO = 0,
P2 =
 E (V | d , d , I
1 2 SEO = 1) if ISEO = 1,
(A3)
where
E (V | d1 , d2 , ISEO = 1) =
1
2 λqf
(d1 − (1 − P1 )) f (d2 − (P1 − P2 )) + 12 λ(1 − q)f (d1 ) f (d2 )
1
.
1
1
2 λqf (d1 − (1 − P1 )) f (d2 − (P1 − P2 )) + 2 λ(1 − q) + 2 f (d1 ) f (d2 )
(A4)
Again, it is easy to show that a solution P2 (d2 | P1 ) always exists. Unlike for P1 , however, P2
need not be unique. For some values of P1 , there is a range of d2 such that multiple values of P2
solve (A4). In such cases, we select the equilibrium P2 to be the largest solution to (A4). If P1
is such that the equilibrium selection rule is used, then the function P2 (d2 | P1 ) is discontinuous
but strictly-increasing in d2 . One can also show that E (V | d1 , d2 , ISEO = 1) < P1 for any finite
value of d2 and that E (V | d1 , d2 , ISEO = 1) converges to P1 from below as d2 → ∞ [PROOF TO
BE COMPLETED].
Proof of Proposition 1:
26
Suppose that the firm announces ISEO = 1. Then
E (P2 | d1 , ISEO = 1) = E (V | d1 , ISEO = 1)
=
=
1
2 λqf
(A5)
1
2 λ(1
(d1 − (1 − P1 )) +
− q)f (d1 )
1
(d1 − (1 − P1 )) + 2 λ(1 − q) + 12 f (d1 )
λP1
,
λP1 + 1 − P1
1
2 λqf
where the first equality follows from the law of iterated expectations.
In calculating E (P2 | d1 , ISEO = 1), GL and the market makers differ only in their estimates
of the probability that informed traders’ signal equals s = H. Using this fact, we have
EG (P2 | d1 , ISEO = 1) =
λP1
+
λP1 + 1 − P1
[EG (H | d1)−E (H | d1, ISEO = 1) ][E (P2 | d1, ISEO = 1, s = H)−E (P2 | d1, ISEO = 1, s = L) ],
(A6)
where EG (H | d1 ) and E (H | d1 , ISEO = 1) denote GL ’s and the market makers’ conditional
probability estimates that s = H, respectively:
EG (H | d1 ) =
E (H | d1 , ISEO = 1) =
qf (d1 − (1 − P1 ))
.
qf (d1 − (1 − P1 )) + (1 − q) f (d1 )
1
2 λqf
1
2 λqf
(d1 − (1 − P1 ))
(d1 − (1 − P1 )) + 12 λ(1 − q) + 12 f (d1 )
(A7)
(A8)
< EG (H | d1 ) .
Furthermore,
E (P2 | d1 , ISEO = 1, s = H) − E (P2 | d1 , ISEO = 1, s = L) =
∞
n2 =−∞
[E (V | d1 , ISEO = 1, d2 = n2 + P1 − P2 ) − E (V | d1 , ISEO = 1, d2 = n2 )] f (n2 )dn2 > 0.
(A9)
The inequality follows because P2 is a strictly increasing function of d2 and P1 > P2 for any value
27
of n2 . Since (A8) and (A9) show that the two bracketed terms in the second line of (A6) are both
positive, it follows that
EG (P2 | d1 , ISEO = 1) >
λP1
= E (P2 | d1 , ISEO = 1) .
λP1 + 1 − P1
28
(A10)
References
Ai, Chunrong, and Edward C. Norton, 2003, Interaction terms in logit and probit models, Economics Letters 80, 123-129.
Almazan, Andres, Keith C. Brown, Murray Carlson, David A. Chapman, 2004, Why constrain
your mutual fund manager?, Journal of Financial Economics 73, 289-321.
Baker, Malcolm, and Jeffrey Wurgler, 2002, Market timing and capital structure, Journal of
Finance 57, 1-32.
Barber, Brad, and John D. Lyon, 1997, Detecting long-horizon abnormal stock returns: the
empirical power and specification of test statistics, Journal of Financial Economics 43, 341-372.
Brav, Alon, 2000, Inference in long-horizon event studies: a Bayesian approach with application
to initial public offerings, Journal of Finance 55, 1979-2016.
Brav, Alon, Christopher Geczy, and Paul A. Gompers, 2000, Is the abnormal return following
equity issuances anomalous?, Journal of Financial Economics 56, 209-249.
Grinblatt, Mark, Sheridan Titman, and Russ Wermers, 1995, Momentum investment strategies,
portfolio performance and herding: a study of mutual fund behavior, American Economic Review
85, 1088-1105.
Huang, Rongbin, and Jay R. Ritter, 2005, Testing the market timing theory of capital structure,
working paper, University of Florida.
Loughran, Tim, and Jay R. Ritter, 1995, The new issue puzzle, Journal of Finance 50, 23-51.
Myers, Stewart C., and Nicholas S. Majluf, 1984, Corporate financing and investment decisions
when firms have information that investors do not have, Journal of Financial Economics 13,
187-221.
Sias, Richard, Laura T. Starks, and Sheridan Titman, 2006, Changes in institutional ownership
and stock returns: assessment and methodology, Journal of Business 79, 2869-2910.
29
Table I
Summary statistics of institutional demand variables
The table reports the summary statistics of institutional demand variables. Buy minus Sell is the number of institutional
investors increasing their holdings in the stock (including position initiations) minus the number of institutional investors
decreasing their holdings in the stock (including position terminations) in a given quarter. Initiation is the number of
institutional investors initiating holdings in the stock in a given quarter. Change in ongoing positions is the difference
between Buy minus Sell and Initiation. All institutional demand variables are normalized by the number of institutional
investors holding the stock at the beginning of the measurement quarter. The sample is restricted to firm-quarters such
that the firm had more than 5% institutional ownership during the past four quarters. Panel A reports the distributions of
the institutional demand variables. Means and percentiles in Panel A are calculated across all firm-quarters. Panel B
reports the autocorrelations of the institutional demand variables. Panel C reports the correlations of the institutional
demand variables with lag, contemporaneous, and lead excess stock returns. Excess stock returns are calculated relative
to size and book-to-market benchmark portfolio returns. Panel C reports the correlations of the institutional demand
variables with contemporaneous firm characteristics. All the statistics reported in Panels B through D are time-series
averages of quarterly cross-sectional correlations.
Panel A: Distribution
1. All firm-quarters
Buy minus Sell
Initiation
Change in ongoing positions
2. Pre-SEO announcement quarters
Buy minus Sell
Initiation
Change in ongoing positions
3. SEO offer quarters
Buy minus Sell
Initiation
Change in ongoing positions
Mean
10th
25th
Percentile
50th
75th
90th
0.015
0.146
-0.132
-0.324
0.000
-0.423
-0.128
0.051
-0.243
0.000
0.111
-0.097
0.167
0.188
0.000
0.345
0.300
0.125
0.175
0.268
-0.094
-0.152
0.071
-0.364
0.000
0.125
-0.217
0.138
0.206
-0.088
0.318
0.333
0.033
0.522
0.517
0.167
0.355
0.512
-0.157
-0.149
0.152
-0.462
0.015
0.238
-0.332
0.243
0.370
-0.167
0.571
0.628
0.000
1.000
1.000
0.176
Lag 3
0.078
0.103
0.096
Lag 4
0.068
0.086
0.084
Panel B: Autocorrelation (all firm-quarters)
Lag 1
0.141
0.149
0.182
Buy minus Sell
Initiation
Change in ongoing positions
Lag 2
0.104
0.118
0.130
Panel C: Correlation with excess stock returns (all firm-quarters)
Buy minus Sell
Initiation
Change in ongoing positions
Lag 2
0.072
0.101
0.024
Lag 1
0.084
0.141
0.011
Lag 0
0.173
0.246
0.030
Lead 1
0.028
0.023
0.019
Lead 2
0.017
0.011
0.015
Firm
age
-0.051
-0.151
0.048
Book
asset
-0.079
-0.086
-0.027
Panel D: Correlation with other firm characteristics (all firm-quarters)
Buy minus Sell
Initiation
Change in ongoing positions
Market/
book
0.045
0.102
-0.017
Profit
margin
0.006
-0.006
0.009
Book
leverage
-0.031
-0.028
-0.014
Table II
SEO announcement probability: Univariate and bivariate sorts
The table reports quarterly SEO announcement probabilities as a function of previous quarter's excess stock return and
institutional demand. At the beginning of each quarter, firms are independently sorted on previous quarter's excess stock
return and institutional demand. Panel A reports the SEO probability for each return quintile, while Panels B, C and D
report the SEO probability as a bivariate function of returns and institutional demand. Panels B, C, and D exclude firmquarters with less than 5% institutional ownership in at least one of the past four quarters. The institutional demand
variable is Buy minus Sell in Panel B, Initiation in Panel C, and Change in ongoing positions in Panel D. Probabilities
are reported as percentages and are calculated as the time-series averages of quarterly probabilities within each return or
return-demand bin. The reported t-statistics in parentheses are adjusted using Newey-West correction for
heteroscedasticity and serial correlation.
Panel A: Univariate sort on excess stock return
Return quintile
Excess stock return
Low
2
3
4
High
-0.32
-0.13
-0.03
0.08
0.40
SEO announcement prob.
High minus Low
All Observations
0.24
0.51
0.77
1.29
2.26
2.02
(13.21)
Institutional ownership < 5%
0.15
0.33
0.49
0.88
1.27
1.12
(10.53)
Institutional ownership > 5%
0.28
0.56
0.84
1.39
2.68
2.40
(12.29)
Panel B: Bivariate sort on excess return and Buy minus Sell
Return quintile
Low
2
3
4
High
Low Demand
0.16
0.47
0.59
1.09
1.24
1.08
(6.63)
2
0.22
0.49
0.59
1.11
2.10
1.88
(7.00)
3
0.28
0.40
0.91
1.31
1.94
1.65
(10.09)
4
0.55
0.68
0.89
1.37
2.68
2.14
(8.03)
High Demand
0.39
0.84
1.25
1.94
4.06
3.67
(12.38)
High Minus Low
0.23
0.37
0.66
0.85
2.82
(2.16)
(2.66)
(4.36)
(4.08)
(9.81)
High minus Low
(continued on next page)
Table II (continued)
SEO announcement probability: Univariate and bivariate sorts
Panel C: Bivariate sort on excess return and Initiation
Return quintile
Low
2
3
4
High
Low Demand
0.10
0.25
0.29
0.36
0.60
0.50
(4.00)
2
0.23
0.43
0.59
0.83
1.40
1.17
(6.12)
3
0.27
0.60
0.85
1.15
1.52
1.24
(6.20)
4
0.25
0.66
1.07
1.78
2.14
1.90
(8.77)
High Demand
0.70
1.06
1.65
2.48
4.21
3.51
(11.58)
High Minus Low
0.60
0.82
1.36
2.12
3.61
(4.73)
(5.53)
(7.19)
(9.82)
(11.23)
High minus Low
Panel D: Bivariate sort on excess return and Change in ongoing positions
Return quintile
Low
2
3
4
High
Low Demand
0.22
0.59
0.83
1.63
2.73
2.52
(8.76)
2
0.24
0.60
0.72
1.51
2.38
2.14
(9.22)
3
0.38
0.40
0.97
1.32
3.04
2.67
(8.75)
4
0.26
0.39
0.56
0.98
2.25
1.98
(9.51)
High Demand
0.42
0.87
1.10
1.57
3.18
2.76
(9.13)
High Minus Low
0.20
0.28
0.27
-0.07
0.44
(1.91)
(2.17)
(1.72)
(-0.32)
(1.62)
High minus Low
Table III
SEO announcement probability: Probit regressions
The table reports the marginal effects from probit regressions in which the dependent variable is Announcei,t, which is a
dummy variable that takes the value of one if firm i makes an SEO announcement in quarter t, and zero otherwise. The
main independent variables in each regression are two dummy variables corresponding to institutional demand in quarter
t-1: (1) High Demandi,t-1, which takes the value of one if firm i is in the highest quintile of institutional demand in quarter
t-1, and zero otherwise; and (2) Medium Demandi,t-1, which takes the value of one if firm i is in the middle three quintiles
of institutional demand in quarter t-1, and zero otherwise. The demand variable is Buy minus Sell in Column 1, Initiation
in Column 2, and Change in ongoing positions in Column 3. The control variables are Excess Returni,t-2, which is the
excess stock return in quarter t-2, and the following firm characteristics measured at the end of quarter t-2: Institutional
ownership measured as the fraction of firm’s stock held by institutional investors, Firm age measured as the logarithm of
the number of quarters since IPO, Firm size measured as the logarithm of book value of assets, Market-to-book ratio,
Profitability measured as EBITDA divided by book assets, and Book leverage. All regressions are estimated with quarter
fixed effects. Marginal effects are computed at the means of the explanatory variables except for the dummy variables
and are reported in percentage terms. Observations are clustered at the firm level. Robust z-scores are reported in
parentheses.
Dependent Variable: Announcei,t
Demand variable
High Demandi,t-1
Medium Demandi,t-1
Excess Returni,t-1
Excess Returni,t-1 u High Demandi,t-1
Excess Returni,t-1 u Medium Demandi,t-1
Excess Returni,t-2
Institutional ownership i,t-2
Firm age i,t-2
Firm size i,t-2
Market-to-book ratio i,t-2
Profitability i,t-2
Book leverage i,t-2
1.05
2.26
Change in ongoing
positions
0.21
(9.42)
(12.39)
(3.44)
Buy minus Sell
Initiation
0.26
0.64
-0.09
(5.10)
(11.54)
(-1.78)
0.79
0.79
0.84
(6.59)
(5.25)
(7.22)
1.16
1.87
1.01
(4.94)
(6.49)
(5.90)
0.94
1.35
0.46
(7.05)
(8.45)
(3.27)
0.86
0.72
0.94
(13.56)
(12.48)
(14.15)
0.62
0.47
0.61
(6.79)
(5.73)
(6.55)
-0.47
-0.39
-0.50
(-22.36)
(-19.51)
(-23.11)
0.05
0.04
0.05
(3.94)
(2.79)
(3.36)
0.05
0.04
0.05
(7.51)
(6.85)
(7.62)
-0.10
-0.10
-0.02
(-1.04)
(-1.23)
(-0.20)
0.89
0.89
0.92
(10.20)
(10.99)
(10.26)
Quarter fixed effects
Yes
Yes
Yes
2
0.112
230,857
0.121
230,857
0.106
230,857
Pseudo R
Number of firm-quarter observations
Table IV
Stock returns in the announcement-to-offer period
Panel A reports excess stock returns following SEO announcements as a function of institutional demand measured by
Initiation in the quarter prior to the SEO announcement quarter. Announcement effect is the (–1,+1) three-day excess
stock return around the announcement. Announcement (+2,+60) is the excess stock return from two days after the SEO
announcement to 60 days after the announcement for all announced SEOs. Announcement(+2,offer) return is the excess
stock return from two days after the announcement to the last trading day before the offer for announcements followed
by an offer within 60 trading days. Offer discount is the offer price divided by the last closing price before the offer for
announcements followed by an offer within 60 trading days. Panel B reports coefficient estimates from OLS regressions
of Announcement (+2,+60) and Announcement-to-offer return. The independent variables are the demand dummy High
Demandi,t-1, which takes the value of one if firm i is in the highest Initiation quintile in quarter t-1, and zero otherwise,
Excess Returni,t-1, and Institutional ownership i,t-2. The t-statistics of the coefficient estimates are reported in parentheses.
Panel A: SEO announcements sorted on pre-announcement institutional demand
Announcements followed by an offer
All announcements
within 60 trading days
Announcement
Announcement
Announcement
Offer
Initiation Quintile
N
effect
(+2,+60)
(+2,offer)
discount
Low
116
-2.45%
-4.37%
-4.13%
-4.89%
2
308
-1.66%
0.52%
-1.44%
-3.59%
3
434
-1.30%
-1.80%
-1.26%
-3.47%
4
664
-1.74%
2.29%
-0.08%
-2.59%
1,364
-1.89%
3.87%
0.87%
-2.89%
0.56%
8.24%
5.00%
1.99%
(0.84)
(3.03)
(3.20)
(1.18)
-1.65%
0.26%
-0.98%
-3.21%
-0.24%
3.62%
1.85%
0.31%
(-1.02)
(3.86)
(3.50)
(0.54)
High
High minus Low
Bottom 4
High minus Bottom 4
1,522
Panel B: Regression Analysis
Announcement
(+2,+60)
0.039
Announcement
(+2,offer)
0.014
(3.85)
(2.37)
-0.020
0.026
(-1.57)
(3.18)
0.039
0.0303
(1.78)
(2.56)
Yes
Yes
R2
0.042
0.079
N
2,886
2,416
Independent Variables
High Demandi,t-1
Excess Returni,t-1
Institutional ownership i,t-2
Quarter Fixed Effect
Table V
Stock returns around and following the offer
The table reports excess stock returns around and after the offer day as a function of offer-quarter institutional demand as
measured by Initiation in the offer quarter t. All SEO firm-quarter observations are sorted into quintiles based on the
offer quarter’s Initiation. Panel A reports the averages for each quintile, while Panel B reports the coefficients from OLS
regressions which include two dummy variables corresponding to institutional demand in quarter t: (1) High Demandi,t,
which takes the value of one if firm i is in the highest quintile of institutional demand in quarter t, and zero otherwise;
and (2) Medium Demandi,t, which takes the value of one if firm i is in the middle three quintiles of institutional demand
in quarter t, and zero otherwise; Excess Returni,t-1, Institutional ownership i,t-1, and quarter fixed effects. The t-statistics of
the coefficient estimates are reported in parentheses.
Panel A: Sorted by Offer Quarter Initiation
Offer quarter t
Offer day
(-1:1) return
Offer
discount
Offer quarter
post-offer period
Quarter
t+1
-2.82%
-1.08%
-2.14%
0.35%
-1.98%
Initiation Quintile
Low
N
537
2
537
0.49%
-0.89%
-2.21%
0.50%
0.22%
3
539
4.36%
-0.25%
-2.50%
2.56%
-0.19%
4
537
10.85%
0.21%
-2.38%
6.66%
2.19%
High
538
21.71%
0.73%
-2.95%
7.50%
3.85%
24.53%
1.81%
-0.81%
7.15%
5.83%
(13.80)
(4.17)
(-3.25)
(6.87)
(3.68)
High minus Low
Panel B: Regression Analysis
Independent Variables
High Demandi,t
Medium Demandi,t
Excess Returni,t-1
Offer quarter t
Offer day
(-1:1) return
Offer
discount
Offer quarter
post-offer period
Quarter
t+1
0.2739
0.0322
0.0016
0.0742
0.0322
(13.86)
(6.78)
(0.51)
(6.22)
(1.69)
0.0878
0.0129
0.0031
0.0256
0.0158
(5.87)
(3.61)
(1.33)
(2.85)
(1.11)
-0.0096
-0.0049
-0.0030
0.0046
0.0228
(-0.64)
(-1.35)
(-1.28)
(0.51)
(1.65)
Excess Returni,t
-0.0007
(-0.04)
Institutional ownership i,t-1
0.1234
0.0480
0.0262
0.0344
-0.0558
(4.65)
(7.41)
(6.18)
(2.17)
(-2.27)
Yes
Yes
Yes
Yes
Yes
R
0.140
0.053
0.087
0.076
0.071
N
2,622
2,627
2,626
2,569
2,543
Quarter Fixed Effect
2
Table VI
Long-run stock returns following the offer
The table reports long-run returns after SEOs as a function of institutional demand as measured by Initiation. Panel A
reports the long-run performance for the entire sample, while Panel B (C) reports the performance for SEOs with preannouncement (offer) quarter’s Initiation above and below the median. Event-time returns are cross-sectional averages
of SEO-specific average of monthly excess returns calculated during one/three/five years after the end of the first quarter
after the SEO. Calendar-time returns are time-series monthly three-factor alphas of portfolios consisting of firms that
have issued SEOs in the past one/three/five years. High minus Low portfolio is a zero-cost portfolio with long (short)
position in high (low) institutional demand portfolio. Panel D reports the monthly alphas of portfolios consisting of nonSEO firms starting at the end of the quarter immediately following the quarter in which Initiation is measured. The tstatistics reported in parentheses are from standard errors calculated using Newey–West corrections with twelve lags.
Event-Time Returns
1 Year
3 Years
5 Years
All SEOs
Low
High
High minus Low
-0.40%
Panel A: All SEOs
-0.21%
-0.06%
Calendar-Time Returns
1 Year
3 Years
5 Years
-0.72%
-0.64%
-0.48%
(-4.30)
(-3.90)
(-3.54)
-0.48%
-0.72%
-0.24%
-0.41%
-0.49%
-0.08%
(-1.15)
(-0.49)
-0.62%
-0.59%
0.03%
-0.50%
-0.40%
0.10%
(0.14)
(0.56)
Panel B: SEOs Sorted by Pre-Announcement Initiation
-0.27%
-0.15%
-0.04%
-0.56%
-0.57%
-0.27%
-0.08%
-0.88%
-0.30%
-0.12%
-0.04%
-0.32%
(-1.32)
Low
High
High minus Low
Panel C: SEOs Sorted by Offer Quarter Initiation
-0.30%
-0.21%
-0.10%
-0.62%
-0.53%
-0.20%
0.00%
-0.80%
-0.23%
0.01%
0.10%
-0.18%
(-0.66)
Panel D: Non-SEO Firms Sorted by Lagged Initiation (Calendar-Time Returns)
Low
-0.14%
-0.08%
High
-0.21%
-0.16%
High minus Low
-0.07%
-0.08%
(-1.30)
(-1.34)
-0.07%
-0.14%
-0.07%
(-1.29)