teacher steve
fargm
ChinaIPOunderpricing.pdf
An Empirical Investigation of Underpricing
in Chinese IPOs
Dongwei Su and Belton M. Fleisher ⁄
June 13, 1997
JEL classiflcation: G10; G30; O53
Keywords: Initial public ofierings; Seasoned equity ofierings;
Asymmetric information; Signaling equilibrium; Privatization
⁄Su: Department of Economics, The Ohio State University, Columbus, OH 43210,
U.S.A.. (Tel) 614-292-5461, (E-mail) [email protected]. Fleisher: Department of Economics,
The Ohio State University, Columbus, OH 43210, U.S.A.. (Tel) 614-292-6429, (E-mail)
°[email protected]. We thank Tasneem Chipty, Gary Jefierson, Pok-sang Lam, Wei Li, Audrey
Light, Howard Marvel, Barry Naughton, James Peck, Bruce Weinberg and seminar participants
at the American Economics Association in New Orleans and the Ohio State University for help-
ful comments and discussions. Portions of this paper are developed in the doctoral dissertation
of Su [11].
Abstract
We study the underpricing of Chinese IPOs, using data of 308 flrm-commitment new
issues. We flnd that underpricing can be explained in terms of a separating equilibrium
under asymmetric information in which underpricing is a strategy for flrms to signal
their value to investors. We show that bribery is an unlikely cause of the high IPO un-
derpricing observed in the Chinese data. We also investigate the hypothesis that various
lottery mechanisms for allocation of IPO shares have exacerbated underpricing. Finally,
we flnd that difierences in initial returns between A and B shares can be explained by the
difierences in domestic and foreign investor’s investment opportunities and investment
sentiments.
An Empirical Investigation of Underpricing in
Chinese IPOs
1 Introduction
Initial public ofiering (IPO) underpricing, or high IPO initial return, is a phenomenon
common to most stock markets|both in developed and emerging economies [7]. A
common perception is that underpricing of IPOs is a contradiction to market e–ciency
and may hurt emerging flrms trying to raise capital for expansion. Therefore it has
spawned an extensive literature attempting to explain this apparent flnancial anomaly.
A number of theories of IPO underpricing have been put forth and tested against the
data of various stock markets.
This paper focuses on exlaining the cross-sectional difierences in underpricing of Chi-
nese IPOs using data compiled for 308 flrm-commitment new issues between December
1986 and January 1996. The Chinese case is of interest primarily because of the extreme
magnitudes that have been observed since market trading of stocks began in late 1990.
Deflne IPO initial return as
IPORETN = P1 ¡P0 P0
»= ln P1 ¡ ln P0:
The mean IPO initial return is a relationship between a share’s ofiering price P0 and
its price on the flrst day of market trading P1 averaged over the entire sample. The
IPORETN averaged over a sample of 308 flrms that went public before January 1,
1996 is 948.59%! In other words, the flrst-day market closing price is on average almost
eleven times as high as the initial price ofiered to domestic Chinese investors.
When we present the striking statistics for initial returns of Chinese IPOs to col-
leagues, a frequent and understandable reaction is that the extradorinarily low IPO
prices relative to flrst-day market prices must represent either irrational behavior or,
more likely, a give-away (presumably bribes) to o–cials who allocate the rights to go
public. We believe, however, that the bribery, while it may well occur, does not explain
underpricing for two reasons. First, underpricing an entire public ofiering in order to
1
allocate some of the underpriced shares to o–cials in return for favors is very costly rela-
tive to other means paying illicit bribes. Second, underpricing of shares can be explained
as rational behavior under asymmetric information. We believe that the conditions for
equilibrium underpricing exist in China, and that high underpricing is caused by basic
problems derived from the microeconomic uncertainty and information asymmetry.
Allen and Faulhaber [1], Chemmanur [2], Grinblatt and Hwang [5], and Welch [13]
have proposed a class of signaling models in which IPO underpricing is an equilibrium
outcome where issuers possess superior information than investors. As long as the reve-
lation probability for the issuers’ quality is neither too large nor too small, there exists
a separating equilibrium where high-value issuers signal their quality by retaining a por-
tion of shares and underpricing initial ofierings and low-value issuers sell all of their
shares and do not underprice. In a pooling equilibrium, no signalling occurs, although
high-quality flrms will be observed, after the fact, to underprice. Underpricing, averaged
over all flrms in the pool, will be zero.
It has been shown in Allen and Faulhaber [1] that undrpricing in a separating equi-
librium implies a shortage of IPO shares, which must be allocated by the issuer or
underwriter in some fashion. Granting favored public o–cials the right to purchase
IPO shares may be one of the allocation methods chosen, just as an underwriter may
favor long-term customers with purchase rights. In this case, bribery is a byproduct of
underpricing, not a cause. We believe that our econometric results support this view.
The rest of the paper proceeds as follows. In Section 2, we present institutional
details of the new-issue ofiering process in China and describe variables used in our
empirical work. In Section 3 we test various hypotheses derived from the signalling
models against data of Chinese IPOs and subsequent market behavior. In Section 4, we
consider the hypothesis that IPO underpricing in China is primarily explainable as a
means of bribing public o–cials, and we examine an assertion of the World Bank that
allocating IPO shares by lottery has been a cause of extreme underpricing.
In Section 5, we extend the above approach to examine the difierence in initial returns
between A shares (available only to Chinese investors) and B shares (available only to
foreign investors).
We summarize the flndings and propose future research in this area in Section 6.
2
2 Institutional Background and Data
2.1 Characteristics of Chinese IPOs
We consider the following characteristics of the new-issue and ofiering process that dis-
tinguish the Chinese markets from those in other countries:
1. The aggregate amount of new shares issued each year is determined by a quota set
by the security regulatory authorities, namely, the State Planning Commission,
the People’s Bank of China (the central bank), and the China Securities Regula-
tory Committee (CSRC). The quota is then distributed to individual provinces.
The stated criteria used for allocation of new issues among provinces re°ect the
central security regulatory authorities’ perceived regional development needs and
provincial difierences in production structure and industrial base. Within each re-
gional quota, the local security regulatory authorities invite enterprises to request
a listing and make a selection based upon enterprise performance and sectoral de-
velopment objectives. Infrastructure enterprises, especially those specializing in
electricity and water supply, are given priority.
2. Difierent share types have been introduced by the government in order to allow
ownership of state-owned enterprises to be dispersed among the government itself,
other state-owned enterprises, flrms’ own employees, domestic private investors,
and foreign investors. There are currently flve types of shares: (1) government
shares, which are retained in the state institutions and government departments
and are non-tradable; (2) legal entity shares, or C shares, which can only be held
by other state-owned enterprises. C shares can not be listed in the two o–cial
exchanges (Shanghai and Shenzhen Security Exchange), but a very small number
are traded on the Security Trading and Automatic Quote System (STAQS) and
National Electronic Trading System (NETS); (3) employee shares, which are non-
tradable until the flrm allows their convertibility; (4) ordinary domestic individual
shares, or A shares, which can only be purchased and traded by private Chinese
citizens in the two o–cial exchanges in China; (5) foreign individual shares, which
3
can only be purchased and traded by the foreign investors in security exchanges
in China (B shares), in Hong Kong (H shares) or in NYSE (N shares)1.
Sample statistics of the proportion of each kind of share in new-share ofierings are
shown in table 1. Note that most stock sales are partial sales. The government
maintains control in varying degree over many flrms. The size of government
ownership at the time of IPO ranged from none to 74.27% with an average of
10.05%. Only 89 out of 308 issuers going public between December 1986 and
January 1996 reported no retained government shares, and even these enterprises
reported IPO sizes that did not exceed 50% of flrm’s intrinsic value, indicating
that a large portion of shares were controlled by other state-owned enterprises.
3. It is also noteworthy in terms of the signalling hypotheses tested in this paper
that initial ofiering size relative to flrm’s intrinsic value is on average 15.02%,
which is small relative to subsequent seasoned equity ofierings (49.93% on average).
Seasoned equity ofierings (SEOs) are very frequently observed. About 91% of
the Chinese flrms that went public before June 30, 1994 have issued subsequent
equities. By comparison, in the United States, between 1980 and 1986, only about
20% of flrms going public subsequently issued seasoned ofierings. (See Jegadeesh,
Weinstein and Welch (JWW) [6].)
4. The ofiering mechanism adopted by most Chinese flrms is quite difierent from those
observed in mature stock markets and has undergone several substantial changes
over time. The ofier price is chosen months before the market trading starts, and
in the great majority of ofierings there is no feedback mechanism through market 1An issuer of B shares must, besides satisfying requirements stated in the securities regulations, meet
the following conditions: (1) It must have obtained approval from the relevant authorities for its use of
foreign investment or for its conversion into a foreign-funded enterprise. (2) It must have a stable source
of adequate foreign exchange income and the total amount of its annual foreign exchange income must
be su–cient to pay the annual dividend. (3) The proportion of B shares to the total number of shares
must not exceed the ceiling determined by the relevant authority. The aggregate amount of shares is
flxed in each year and the total number of flrms allowed to issue foreign shares is also limited. An issuer
of H or N shares is not subject to the quota restriction, but is subject to case-by-case approval.
4
demand that allows adjustment in the ofier price. The lottery mechanism, which
remains the primary method of share allocation, has undergone several substantial
changes. Before 1992, the security regulatory authorities designed a lottery system
based on a pre-announced flxed number of application forms. Each retail investor
was allowed to purchase a limited number of lottery forms from the central bank
and its subsidiaries. Lottery winners were entitled to a certain number of shares
per winning form. With the number of lottery forms pre-determined, the odds
of winning the lottery was known to investors. In 1993, the security regulatory
authorities introduced two new lottery mechanisms: One mechanism was based
on unlimited number of application forms. The central bank sold as many lottery
forms as investors were willing to buy. Therefore, the odds of winning the lottery
was unknown to investors at the time of lottery. The other lottery mechanism was
based on savings deposit certiflcates. Investors were required to deposit a certain
quantity of funds into a special saving account when submitting application for
shares, which could not be withdrawn until the lottery was completed. These
special saving accounts were given relatively low interest.
Under the lottery mechanisms, the IPO prices were flxed for all investors. In the
early stage of development in Chinese stock markets, some initial issues were even
ofiered at the shares’ face values. Companies that went public before January
1991, such as Shanghai Vacuum Electronics, Jinbei Automotive, Phoenix Chem-
ical, China Textile Machinery, Shenzhen Vanke Co., Gintian Industry, Shenzhen
Zhenye Co., and Shenyang Materials Development, all ofiered shares at Renminbi
1 yuan.
In 1994, two kinds of auction mechanisms were introduced. Under the flrst auction
mechanism, an issuer set an initial price and investors were required to bid for the
price and quantity. The flnal ofier price was set at the level where the accumulative
quantities demanded by investors equaled the total number of new shares avail-
able. Under the second auction mechanism, the IPO price was flxed and investors
were invited to bid for the quantity of shares. In case of oversubscription, all in-
vestors were guaranteed a certain amount of shares and the remaining shares were
5
distributed on a pro-rata basis. Lottery mechanaisms as described above remained
as options, however, so flrms could choose either a lottery method of distributing
IPO shares or an auction mechanism. The proceeds from the lottery belong to the
state banks that sell shares for the flrms.
5. The average time elapsed between the announcement of IPO and the flrst day
market trading is considerable greater in China than in other countries|260 days
for A shares and 72 days for B shares2.
6. Chinese A-share IPOs are sold to relatively unsophisticated retail investors while
B shares are usually sold to international institutional investors such as foreign
mutual funds.
2.2 The Chinese Data
We use data of all the flrm-commitment IPOs of A-share common stocks occurring
between December 1986 and January 1996. A detailed description is contained in Ap-
pendix A. In order to study the efiects of subsequent equity ofierings on IPO underpric-
ing, we also extract a sub-sample of flrms that went public between December 1986 and
June 1994. The June 1994 cutofi allows 548 days for a flrm to issue SEOs.
Variables used in our empirical work include:
IPOSZ = the monetary price of portion of the flrm initially ofierred to the
investors, or the proceed from IPO in U.S. dollars.
PROFIT = past year’s proflt at the time of IPO in U.S. dollars.
AGE = age of the flrm at the date of its IPO.
TIMEIPO = number of days elapsed between the announcement of an IPO 2There are a number of steps a flrm must take after it is selected for initial public ofiering and before
the market trading begins. Some typical steps include: (1) publication of a prospectus in newspapers
and selection of underwriters; (2) purchase of application forms by prospective investors; (3) a lottery
to determine which individual and institutional investors will be allowed to purchase new issues at the
IPO price; (4) delivery of shares to the lottery winners after payments are made.
6
and the flrst-day market trading.
MKTCAP = the sum of IPO proceeds from A- and B-share ofierings,
government shares, shares purchased by other state-owned
enterprises, and proceeds from all SEOs in U.S. dollars.
AFTRETN1 = after-market return between the flrst day of trading
and the end of the second week of trading.
AFTRETN2 = after-market return between the beginning of the third week
of trading and the end of the fourth week.
STD = standard deviation of after-market returns estimated over a
the 100-day period after inception of market trading.
EXD = stock exchange dummy, EXD = 1 for a company that is
listed in Shanghai Securities Exchange, EXD = 0 for a
company that is listed in Shenzhen Securities Exchange. IPOSZ
MKTCAP = IPO proceeds as a fraction of flrm’s intrinsic value (MKTCAP is
used as a proxy for flrm’s intrinsic value). SEOSZ
MKTCAP = total SEO proceeds as a fraction of flrm’s intrinsic value.
SIC(k) = six industry dummies: durable goods (SIC1), non-durable
goods (SIC2), transportation and public utilities (SIC3),
flnance, insurance and real estate (SIC4), services including
restaurants, department stores and hotels (SIC5) and domestic
and foreign trade (SIC6).
Y EAR(t) = IPO year dummies, t = 1 if a flrm went public before January 1,
1991; t = 2 ¢ ¢ ¢5 for going public in the years 1992 through 1995.
Descriptive statistics for the above variables are presented in table 2. The correlation
matrix for some of the variables is presented in table 3.
7
3 Signaling Hypotheses of IPO Underpricing: Do
Chinese Issuers Leave A Good Taste for the In-
vestors?
3.1 Empirical Implications from the Signaling Models
Allen and Faulhaber [1], Grinblatt and Hwang [5], Welch [13] and Chemmanur [2] have
proposed a class of signaling models of IPO underpricing in which issuers have superior
information than investors about the instrinsic value of the flrms. The discussion in the
remainder of this section draws heavily upon these papers. In the cited models, an issuer
maximizes the value of the flrm through IPO and subsequent SEOs. Because they do
not have complete information, in the absence of a signal, investors cannot distinguish
between \high value" and \low value" flrms. In a separating equilibrium, a \high value"
issuer signals its quality by underpricing. It can afiord to underprice its IPO because
it expects to capture larger revenues through SEOs. In contrast, a \low value" issuer
does not signal because it does not expect to recoup its investment in underpricing
through after-market SEOs. The best a low-value issuer can do is to \take the money
and run" when its stock is initally ofiered. When a separating equilibrium occurs, the
average risk-adjusted IPO return over all new issues will be positive, the quantity of
shares demanded for underpriced issues will exceed quantity supplied, and shares will
be rationed by a mechanism other than the ofier price.
The papers mentioned above have somewhat difierent setups that lead to similar, but
not identical, testable hypotheses. In Allen and Faulhaber’s framework [1], as long as the
probability that a high-quality flrm will remain good after it implements its production
technique, product, or whatever it does is not \very large", the high-value flrm will flnd
it advantageous to signal by underpricing its IPO while a low-value flrm will not flnd it
worthwhile to underprice. Therefore, underpricing occurs in a separating equilibrium.
In Grinblatt and Hwang [5], if the expected value and the variance of future cash °ow
for a risk-averse flrm is unknown to the investors, then, corresponding to the revelation
probability of the flrm’s intrinsic value (between 0 and 1), there exists an equilibrium
8
signaling schedule relating two signals, fractional holding and IPO underpricing. The
flrm signals its value to the market by choosing the optimal combination of underpricing
and IPO size. In Welch [13], low-value flrms must invest in imitation expenses to appear
to be high-value flrms, and with non-zero probability this imitation will be discovered
between the dates of IPO and SEO. Underpricing by high-value flrms at the time of
IPO can add su–cient signaling costs to these imitation expenses to induce low-value
flrm to separate and thus reveal their value voluntarily. In Chemmanur [2], flrm insiders
sell equity both in the IPO market and the secondary market, have private information
about their flrm’s prospects, and outsiders will produce information if the expected
net payofi from information production is nonnegative. IPO underpricing results from
insiders inducing information production in order to obtain a more precise valuation of
their flrms in the secondary market.
The testable implications from the signaling models with separating equilibrium in-
clude:
(1) The correlation between the degree of IPO underpricing and IPO proceeds is neg-
ative. In the signaling model of Grinblatt and Hwang [5], there exists a monotonic
correspondence between ofiering price at which the issue is being sold and the de-
gree of IPO underpricing in a separating equilibrium, holding constant the flrm’s
fractional shareholding (or, equivalently, holding constant the fractional size of the
IPO). That is, if investors observe an issuer’s fractional ownership and can conjec-
ture the value of a benchmark lowest-variance issuer by looking the relative size
of its IPO, they can also infer the degree of IPO underpricing for any arbitrary
issuer given the issuer’s initial ofiering price and relative IPO size.
(2) There is an optimal signaling schedule relating the flrm’s intrinsic value, the degree
of underpricing, and relative IPO size. In particular, the degree of underpricing and
the issuer’s intrinsic value are positively correlated, holding constant the issuer’s
fractional ownership and the variance of returns; the degree of underpricing and the
project variance are positively correlated, holding constant the issuer’s fractional
ownership and the variance of returns; the degree of underpricing and the fractional
holding are positively related, holding constant the variance of returns and the
9
flrm’s intrinsic value.
(3) Issuers with a larger degree of IPO underpricing are more likely to return to the
secondary market and ofier larger amount of SEOs, more quickly than issuers with
lower IPO underpricing.
The are various \stories" rationalizing this hypothesis. One linkage between the
underpricing signal and SEO behavior is that an issuer gives out \free samples"
to the public by underpricing and induces the public to learn more about the
issuer. The learning process leads to a higher price on the flrst day of market
trading than would otherwise occur|but for high-value issuer only. This efiect on
the market price allows a high-value issuer to quickly return to the market with
SEOs and thereby reap the return from underpricing its IPO. In another version,
investors are more passive and \learning" occurs through exogenous revelation of
information after implementation of the flrm’s innovation. The IPO is underpriced
to induce investors to furnish su–cient startup funds to enable implementation of
the innovation. Firms are willing to underprice because they expect a higher-than-
normal return on their investment.
The alternative to a separating equilibrium is a pooling equilibrium where IPO prices
are a weighted average of the present value of high-value and low-value issuers. When
information is revealed after market trading starts, price difierentiation occurs, but the
flrst-day market price cannot be predicted from the IPO price and there is no risk-
adjusted \excess" IPO initial return between the ofier date and the flrst-day market
trading for a large sample of flrms.
In a pooling equilibrium, high-value flrms do issue SEOs, but only in response to
market-provided information about their value. That is, the market possesses superior
information than the issuers. If the after-market returns are high, then flrms will issue
seasoned equities. If they are low, then there are no seasoned issues. No equilibrium
signaling schedule exists. This is the so called market feedback hypothesis.
10
3.2 Empirical Results
3.2.1 Correlation Between IPO Underpricing and IPO Proceed
The flrst set of empirical implications in section 3.1 indicates a monotonic negative
correspondence between the price of initial ofiering and IPORETN, holding constant
the relative size of the IPO in a separating equilibrium under the signaling hypothesis.
In a pooling equilibrium, since IPO price is a weighted average of the intrinsic value of
low- and high-quality flrms, there is no implied correspondence between IPO proceed
and realized IPO initial return.
To test this hypothesis empirically, we need to consider that if the flrst-day market
price and the price for initial ofiering are observed with error, there will be a spurious
negative correlation between the the observed IPO price and the observed IPORETN,
even under a pooling equilibrium. Another source of possible bias also exists for the
relative IPO size variable. We construe relative IPO size to be the fraction of the intrinsic
value of the flrm that is initially ofiered to the public. We measure flrm’s intrinsic
value with the variable MKTCAP which is a function of the flrm’s realized IPO and
SEO behavior. Therefore simultaneous-equation bias may occur if the \raw" measure
of relative IPO size is used as a regressor. To correct for the errors-in-variable (EIV)
simultaneity problems, we adopt a two-stage approach to derive unabiased estimates of
the relationship between the logarithm of IPO return, IPO price, and the relative size
of IPO (the complement of the flrm’s fractional shareholding). In the flrst stage, we
regress ln IPOSZ and IPOSZ MKTCAP
against the exogenous variables ln AGE, ln PROFIT ,
ln TIMEIPO, EXD, SIC(k) and Y EAR(t), and obtain the instruments dln IPOSZ
and dIPOSZ MKTCAP
. In the second stage, we estimate the following regression3:
ln IPORETNi = fl0 + fl1 dln IPOSZ + fl2 dIPOSZ
MKTCAP + †i (1)
The estimation results reported in table 4 show that: (1) holding constant the relative
size of initial ofierings, a one percent decline in the price of initial ofiering is associated 3We interpret Grinblatt and Huang [5] to imply that equation (1) holds for issuers of given initial
wealth. We assume that the issuer is the Chinese government, and therefore the initial wealth in the
cross section of issuers is roughly constant and can thus safely be omitted from the regression.
11
with approximately a 0.8 percent increase in IPO initial return and that (2) a one
percentage-point increase in relative IPO size is associated with a 11 percent decline in
IPO return. Both coe–cients are highly signiflcant.
3.2.2 Correlations Among IPO Underpricing and Issuer’s Intrinsic Value,
Fractional Ownership and Project Variance
The second set of empirical implications in section 3.1 suggests a positive relationship
between an issuing flrm’s IPO return, relative IPO size, variance of future returns, and
intrinsic value (under a separating equilibrium). Testing these implications is compli-
cated by the necessity of measuring the variance of future returns and lack of an error-free
measure of the issuing flrm’s intrinsic value. As a proxy for intrinsic value, we use the
MKTCAP variable described in section 2.2, and we proxy the variance of future returns
with variance of A-share returns over the flrst 100-day period of market trading. More-
over, complement of the fraction of the flrm’s value retained by the issuer, the relative
size of IPO is determined endogenously. Again, to correct for these EIV and endogeneity
problems we estimate the following regression by 2SLS using the same instruments as
in equation (1) to obtain the fltted values of the regressors:
ln IPORETNi = fl0 + fl1 dln MKTCAP + fl2 dln IPOSZ + fl3 dSTD + †i (2)
The estimation results reported in table 5 show that: (1) holding constant the flrm’s
intrinsic value and the project variance, a one percent decrease in the size of the initial
ofiering is associated with approximately a two percent increase in the degree of IPO
underpricing; (2) holding constant the size the initial ofiering and the project variance, a
one percent increase in the flrm’s intrinsic value is associated with approximately a one
percent increase in the degree of IPO underpricing4; and (3) holding constant the flrm’s
intrinsic value and the size of the initial ofiering, an increase of one standard deviation 4We have calculated two measures of a flrm’s intrinsic value. One measure is the sum of the total
proceeds from A-share IPO, B-share IPO (if any), and all SEOs. The other measure is the outstanding
market capitalization as of December 29, 1995. We flnd that both measures of flrm’s intrinsic value are
highly correlated and yield almost the same results in all of our estimations. Therefore, we only report
the estimation results using the flrst measure.
12
in the variation of future returns is associated with almost an eleven percent increase in
the degree of IPO underpricing, which implies an elasticity of approximately 0.6 at the
mean value of the regressor. All of the coe–cients are statistically signiflcant, even at
the 1% level.
Therefore, our empirical results in this section strongly support a separating equilib-
rium under signaling hypothesis.
3.2.3 IPO Underpricing and SEOs
All the signalling models discussed in section 3.1 imply that high-quality issuers under-
price their IPOs so that they can subsequently issue seasoned equities at more favorable
prices than they would otherwise have received. Issuers that do not plan to sell seasoned
equities subsequent to the IPOs do not discount their initial sales. In reality, not all
IPOs that appear to be underpriced are followed by SEOs. About 91% of Chinese flrms
going public before June 30, 1994 issued subsequent equities before January 1, 1996. As
JWW [6] point out, some of the flrms that underprice their IPOs with the intention of
issuing subsequent equities may fail to do so because of unexpected economic shocks.
However, such shocks are less likely to prevent flrms of very high quality to stick to their
original plans. In addition, some underpricing may be unintentional or unexpected.
Therefore we view the relationship between IPO underpricing and subsequent ofierings
probabilistically, that (1) issuers with larger IPO underpricing are more likely to issue
subsequent equities than issuers with lower expected IPO underpricing. Additional hy-
potheses related to IPO-SEO behavior are: (2) Issuers with larger IPO underpricing will
issue larger amounts of SEOs relative to their intrinsic value, and (3) Issuers with larger
IPO underpricing will issue SEOs more quickly after the initial sales5.
Empirical tests need to distinguish between the implications of a separating and a
pooling equilibrium, where flrms to do not signal their value to the investors through
underpricing. A flrm’s speciflc value is revealed only after market trading begins. This
suggests the \market-feedback hypothesis," in which high post-IPO returns lead flrms 5JWW [6] test the signaling models using U.S. data and flnd weak evidence that flrms that underprice
their IPOs are more likely to issue seasoned equities and on average have larger subsequent ofierings.
13
to issue seasoned equities and low post-IPO returns will discourage issues of additional
shares.
To test the hypothesis that issuers with larger expected IPO underpricing are more
likely to issue SEOs, we follow the spirit of JWW [6] and estimate the following logit
model,
ln ˆ
PSEOi 1 ¡PSEOi
!
= `0 + `1IPORETNi + `2AFTRETN1i + `3AFTRETN2i (3)
where PSEOi is the probability that the ith issuer will issue subsequent equity ofierings
after the initial sale. The independent variables are the observed degree of IPO under-
pricing (IPORETN), the after-market return for the flrst two weeks (AFTRETN1)
and the after-market return for the third and fourth weeks (AFTRETN2)6.
The logit regression estimates are presented in table 6. The slope coe–cient for
IPORETN (t-statistic) is 0.001326 (2.1301), which indicates a positive relationship
between the degree of IPO underpricing and the probability for a flrm to issue SEOs.
To assess the relative explanatory power of the signalling versus market-feedback hy-
pothesis, we calculate ‡ @PSEO=@IPORETN
· ¢dIPORETN and
‡ @PSEO=@AFTRETN1
· ¢
dAFTRETN1 and evaluate these two expressions using the standard deviation of the 6JWW [6] include a variable representing IPO size in their equation. Their rationale for including
IPO size is \Since flrms that raise relatively small amounts of capital at the IPO may be more likely to
return with a seasoned equity ofiering, we include the natural logarithm of IPO size as an additional
explanatory variable." We believe that inclusion of IPO size as a regressor is inappropriate when the
hypothesis being tested is that underpricing reduces the size of IPOs relative to SEOs. IPO size becomes
endogenous and may well \pick up" the efiect of underpricing we are trying to estimate. In regressions
not reported here, we have included both the log of IPO size and the log of total market capitalization
as alternative regressors. In both cases, our econometric results are similar to those reported here, but
with log IPO size as a regressor, the results imply somewhat smaller relative explanatory power for the
seprating-equilibrium hypothesis.
We are also aware of the EIV problem when we use the observed IPO initial returns instead of
a proper instrument for the expected IPO initial returns. Unfortunately, we are unable to flnd any
observable exogenous variables that are unknown to the investors under the asymmetric information
assumption. Thus, following JWW [6], we do not use an instrumental-variable or 2SLS approach in
estimating equation (3).
14
respective regressors, obtaining 4.2 and 0.80, respectively, implying that a one standard-
deviation increase in IPO return is associated with an increase in the magnitude of the
probability of an SEO that is more than flve times larger than a one standard-deviation
increase in after-market return over the flrst two weeks of market trading after the IPO
date. Since the estimated coe–cient for AFTRETN2 is much smaller and not signifl-
cantly difierent from zero, we conclude that the empirical power of the market signalling
hypothesis is greater than that of the market feedback hypothesis.
To test the imlication that flrms with larger IPO underpricing tend to issue a larger
amount of seasoned equities, we estimate a tobit regression that relates IPO underpricing
and after-market returns to the ratio of the market value of all SEOs to the flrm’s intrinsic
value which is proxied by the sum of the proceeds from IPO and all SEOs7,
µ SEOSZ
MKTCAP
¶
i =
8 >>><
>>>:
–0 + –1IPORETNi + –2AFTRETN1i +–3AFTRETN2i + "i if RHS> 0
0 otherwise
(4)
The coe–cient estimates of equation (4) are contained in table 7. The coe–cient es-
timate for IPORETN (t-statistic) is 0.0000309 (6.127), which indicates that the higher
the IPO underpricing, the larger is the size of seasoned equity issues.
To compare the relative explanatory power of the signalling and market feedback hy-
potheses, we follow the same procedure as with the estimates of equation (3) and calcu-
late ‡ @ ‡
SEOSZ MKTCAP
· =@IPORETN
· ¢dIPORETN and
‡ @ ‡
SEOSZ MKTCAP
· =@AFTRETN1
· ¢
dAFTRETN1 using the standard deviation of the respective regressors. We obtain
0.091 and 0.048, indicating that a one standard-deviation increase in IPO return and
flrst two-week after market return are associated, respectively, with an increase in relative
SEO size of 10% and 5.1%, respectively. This comparison suggests that the signalling
hypothesis has almost twice as much \strength" in explaining SEO relative magnitude
than does the market-feedback hypothesis. As with the probability of an SEO, the
coe–cient estimate for AFTRETN2 is positive, but is statistically insigniflcant. We 7We allow until January 1, 1996 for a flrm to issue its flrst SEO. About 91% of the flrms that
went public between December 1986 and June 1994 issued subsequent equity ofierings before this date.
Because our sample is censored as of the cutofi date of January 1, 1996, a tobit speciflcation is desirable.
15
conclude that while the tobit regression estimates support the signaling models relating
A-share IPO underpricing to the relative size of SEOs, they do not allow us to reject
the alternative hypothesis that the size of SEOs is also partially related to after-market
returns.
Finally, we examine the relationship between IPO underpricing and the time elapsed
between the IPO and the flrst SEO using the following tobit model:
TIMESEOi =
8 >>><
>>>:
°0 + °1IPORETNi + °2AFTRETN1i +°3AFTRETN2i + "i if RHS> 0
0 otherwise
(5)
where TIMESEO is the number of days elapsed between the IPO date and the flrst
SEO date.
The tobit regression estimates are presented in table 8. The slope coe–cient (t-
statistic) for IPORETN is -0.00222 (-0.5245), indicating that flrms with higher degree
of IPO underpricing tend to return to the secondary market and issue seasoned equity
ofierings more quickly than flrms with lower degree of IPO underpricing and larger IPO
sizes, although the estimated coe–cient of IPO return is not signiflcant at conventional
levels.
In comparing the relative explanatory power of the market-signalling and market-
feedback hypotheses, we note that the median time elapsed between IPO and SEO dates
is almost twenty weeks, far longer than the period over which we measure after-market
returns. Thus, we predict a negative coe–cient on the two after-market return variables
as evidence supporting the market-feedback hypothesis. However, the estimated coe–-
cients (t-statistic) for AFTRETN1 and AFTRETN2 are 1.0864 (2.2203) and 0.7852
(0.9593), respectively, which we interpret as contradictory to the market-feedback hy-
pothesis.
To summarize, we flnd strong evidence from the Chinese A-share data that supports
the signaling models that link IPO underpricing to SEOs. We flnd that Chinese is-
suers who underprice their A-share IPOs more heavily are more likely to return to the
secondary market and issue larger amounts of after-market equities. The alternative,
namely, the market feedback hypothesis, does not fare as well in explaining the high
16
underpricing in the data.
4 Bribery and Lottery Hypotheses of IPO Under-
pricing
4.1 Bribery Hypothesis: Underpricing of Chinese IPOs are
gifts to public o–cialsor other favored purchasers.
Allen and Faulhaber[1] show that underpricing implies an excess demand for IPO shares.
It follows that either the issuing flrm or its agent (e.g. the underwriter) has power to al-
locate valuable assets to favored associates in return for their loyalty, favors, or whatever.
In this context it seems natural to consider that issuers may allocate underpriced shares
to politicians or bureaucrats who have substantial control over some of their planned
resources. For example, under the quota system in China, a flrm is not unlikely to
allocate underpriced shares to bureaucrats in order to gain permission to issue shares8
We reject a priori the likelihood that bribery by underpicing under the alternative
frameworks of full information or a pooling equilibrium with asymmetric information
between issuers and investors accounts for observed IPO returns in China. Why should
a flrm underprice its entire IPO in order to create an opportunity to bribe a few o–cials?
This seems very costly indeed, given that other means are available. It would be much
cheaper for the flrms to transfer funds directly to bureaucrats in order to gain favors.
Admittedly, there may be a few cases involving very small IPOs where the extra cost of
using underpricing as a bribe relative to alternative means is minimal, but we believe
that such occurrences, if they exist, do not shape the broad pattern of underpricing
observed in the data.
As a test of the importance of bribery in the IPO process, we conjecture, that with
bribery, many bribe recipients would want to sell their underpriced shares as quickly as 8Note that bribery itself can be construed as a signal of flrm value in that only a flrm that expects
to recoup its investment in bribery will bribe. It is questionable, however, whether a bribe can be a
signal to the average investor, who presumably is not privy to the illicit transaction.
17
they can, or as soon as the stock prices go up after market trading begins, leading to
signiflcant depression of returns in the secondary market. Moreover, high-quality flrms
that allocate underpriced shares to gain permission to go public will be more likely to
delay their SEOs to avoid depressing stock prices further and give the politicians or
bureaucrats a chance to cash in.
Evidence pertaining to the price-depression efiect noted above can be adduced from
calculating alternative measures of IPO underpricing, using the after-market price from
the flrst day, flrst week, second week, and subsequent weeks’ closing prices. We flnd that
compared to the flrst-day IPO return, which has a mean of 948.59% (with a minimum of
-18.58% and maximum of 38300%), the average degree of IPO underpricing using flrst-
week, second-week, third-week and fourth-week market closing prices remains approxi-
mately the same, 950.98%, 955.9%, 950.49%, and 898.39%, respectively. (The minimums
are -17.64%, -27.95%, -38.72%, -42.05% and the maximums 48920%, 50410%, 52210%,
and 52470%, respectively.) We take this as evidence that bribes (if they occured) were
not \cashed in" during the flrst four weeks of market trading. Since IPO underpricing
was the largest at the early stage of development of stock markets in China when reg-
ulations governing new issues were not yet well developed, it is natural to suspect that
this was a fertile period for bribery. Therefore we calculate difierent measurement of the
average degree of IPO underpricing for 29 flrms that went public before the emergence
of formal stock markets. However, we do not flnd any evidence of a large after-market
price drop for these issues, either9.
We also note that the estimation results from the tobit regression 5, ofiers no evidence
that a higher IPO return is associated with a delay in ofiering SEOs. In addition, the
average time elapsed between the flrst-day market trading and SEO dates for those 29
flrms going public before the establishment of formal stock markets is 254 days, which is 9The average degree of underpricing for 29 new issues that went public before the emergence of
formal stock markets using the flrst-day market price is 5664.29%. (The minimum is 748% and the
maximum is 38300%.) The average degree of IPO underpricing using flrst-week, second-week, third-
week and fourth-week market closing prices for these 29 new issues is 6053.4%, 6181.39%, 6230.83%,
and 5964.11%, respectively. (The minimum is 782%, 728%, 679%, 610%, and the maximum is 48920%,
50410%, 52210%, and 52470%, respectively.)
18
less than the overall average of 284 days for the entire sub-sample of flrms going public
before July 1994. We take this as additional evidence against the bribery hypothesis.
To sum up, while bribery may in principal be a byproduct of undepricing in a sepa-
rating equlibrium under asymmetric information, direct evidence is by the nature of the
process di–cult to flnd. We flnd little if any indirect evidence that bribery ocurred in
a large enough proportion of cases to cause to observable efiects on aftermarket returns
and/or the timing of SEOs. We conclude that the signaling hypothesis characterized by
a separating equilibrium seems to be capable of explaining the a substantial portion of
underpricing behavior of the Chinese flrms.
4.2 Lottery Hypothesis: Lottery mechanism in share allocation
contributes to high IPO underpricing.
Five difierent ofiering mechanisms have been used in allocating A - shares in China. They
are: a lottery mechanism with a flxed number of application forms (OD(1)), a lottery
mechanism with an unlimited number of application forms (OD(2)), a lottery mecha-
nism based on certiflcate of deposit receipts (OD(3)), an auction mechanism with only
quantity bids and flxed IPO prices (OD(4)), and an auction mechanism with quantity
and price bids (OD(5)). A team of World Bank specialists argued that ofiering mecha-
nisms afiected the degree of underpricing (\China: The Emerging Capital Markets" Vol.
II, p. 96).
. . . the allocation mechanism adopted for the new share issue afiects the de-
gree of underpricing. Non-discretionary allocation of shares, by mechanisms
such as a lottery, exacerbate the tendency to underprice.
We disagree with their assertion. It was hypothesized and empirially supported that
a lower ofier price and smaller IPO size leads to higher IPO underpricing. As mentioned
in previous sections, IPO underpricing implies an excess demand for shares initially
and necessitates some form of nonprice rationing. All types of lottery mechanisms and
auction mechanism with only quantity bids are best viewed as simply means to allocate
oversubscribed shares, and do not determine the after-market demand for shares and/or
19
the inital supply of shares. Therefore, these mechanism themselves do not cause IPO
underpicing. An alternative to the lottery, an auction in which both price and quantity
bids are submitted, does create an ofier price that is not predetermined by the issuing
flrm. Investors are invited to submit a single bid for the price they are willing to pay and
the amount of shares they are willing to purchase. The flnal ofier price is set at a level
where the accumulative quantity demanded equal the amount of new shares available.
In case of oversubscription, shares are allocated on a pro rata basis. Eighteen flrms used
this mechanism in issuing and allocating new shares prior to January, 1996. The average
degree of IPO underpricing is only 1.04% for these 18 issues, which is negligible. It is
critically important to note that these flrms chose the auction method when a lottery
mechanims was available. With the choice of share-allocation mechanism endogenous,
it would be di–cult to conclude that the the auction led to virtually zero underpricing.
Rather, the opposite may be inferred, namely, that these flrms chose not to signal their
quality by underpricing and used the auction mechanism as a means to assure they
received the largest possible revenue at the IPO, because they had little nor no intention
of returning to the secondary market with SEOs.
Even though we reject a priori the notion that the IPO allocation mechanisms in
China have caused the large magnitudes of underpricing observed, we nevertheless flnd
it interesting to investigate the relationship between ofiering mechanisms and IPO initial
return cross-sectionally, holding constant the variables associated with IPO return under
the hypothesis of asymmetric information with a separating equilibrum. To do this, we
reforumalate equation (1) as follows, which we estimate using 2SLS:
ln IPORETNi = ·0 + ·1 dln IPOSZi + ·2 dln MKTCAPi + ·3 dSTDi + ·4LDi + ”i (6)
where
LD = dummy variable that takes value 1 if a flrm uses a lottery mechanism or
auction mechanism with only quantity bids in allocating shares and 0 if
it uses auction mechanism with both quantity and price bids.
20
The estimation results presented in table 9 show that the coe–cient for LD vari-
able is highly positively signiflcant, indicating that underpricing is present under lottery
mechanisms and auction mechanism with only quantity bids. Under the auction mech-
anism with both price and quantity bids, which is designed to eliminate excess demand
for shares at the flrst place, underpricing is on average negligible.
We also test the hypothesis that the mean IPO underpricing difiers under difierent
ofiering mechanisms by estimating the following expansion of equation (1) using 2SLS:
ln IPORETNi = µ0 + µ1 dln IPOSZi + µ2 dln MKTCAPi + µ3 dSTDi + µlOD(l)i + ”i (7)
The estimation results in table 9 show that underpricing is on average larger under
the various lottery mechanisms than under the auction mechanisms. In particular, IPO
underpricing is the largest under the lottery with flxed number of application forms, is
the second largest under the lottery with unlimited number of application forms, and is
the smallest under auction mechanisms. To interpret these results, we emphasize that
when more than one mechanism was available, as they were in 1993 through 1995, choice
of mechanism was endogenous. We reiterate that under a asymmetric information with
a separating equilibrum, underpricing is a signal for flrm’s true value, there will be excess
demand for shares and some type of ofiering mechanism has to be used to eliminate the
excess demand. Ofiering mechanisms themselves do not cause underpricing.
5 Underpricing of B-share Ofierings
We now examine the underpricing of B-share initial ofierings using data for 57 flrms that
issued both A and B shares between February 1992 and January 1996. Among these 57
B share issues, 23 of them were ofiered after their corresponding A shares were listed
and traded in the secondary markets. Therefore, these 23 new B share ofierings should
be treated as SEOs. Nevertheless, since the participants in the B-share markets are
primarily international underwriters and investors, we are still interested in examining
the difierence in the initial returns for A- and B-share ofierings
Table 10 presents the sample statistics for flrms issuing both A and B shares. As
shown in the table, that the average initial B-share returns is 37.13% which is consid-
21
erably smaller than the 838.91% observed for the corresponding 57 A shares. A further
comparison of the sample statistics in tables 2 and 10 shows that, on average, flrms
issuing both A- and B shares experienced less A-share IPO underpricing than flrms tha
raised equity capital only in China. The difierence in average initial return between
these two classes of shares raises two questions: First, why would a flrm issue B shares?
Second, what determines the difierences in underpricing of these two classes of shares?
Two likely reasons for Chinese flrms to issue foreign-owned shares are: (1) to obtain
foreign capital that is otherwise di–cult to get under the government’s tight foreign
currency control regime; (2) to enhance the reputation of the flrm. At the same time,
access to international capital markets is doubtless enhanced for flrms that have well-
established domestic histories that can be viewed by international investors as evidence
of strong future prospects. In other words, we believe that information asymmetry is
likely to be a greater hindrance to raising funds in the international market place than
domestically. A comparison of the sample statistics in table 2 with those in table 10
supports this conjecture. The data show that on average, Chinese flrms that ofier B
shares are considerably larger in terms of intrinsic value; have higher proflt before the
IPOs; have experienced smaller degree of A-share IPO underpricing; and have higher
after-market returns on their A shares than the flrms that have not ofiered B shares.
A convenient way to sort out the characteristics distinguishing Chinese flrms that
ofier B shares from those of flrms that ofier only A shares is to estimate the following
logit model, which we apply to all flrms issuing A shares.
ln ˆ
PBi 1 ¡PBi
!
= `0 + `1IPORETNi + `2 ln IPOSZi + `3 ln MKTCAPi (8)
+`4 ln PROFITi + `5 ln TIMEIPOi + `6AFTRETN1i
+`7AFTRETN2i + `kSIC(k)i
where PBi is the probability that the ith flrm will issue B shares after it has ofiered A
shares. PBi = 1 if flrm i issues B shares and 0 otherwise.
The logit regression estimates presented in table 11 indicate that: (1) holding con-
stant the flrm’s intrinsic value, the smaller the size of A-share IPO, the more likely the
flrm will issue B shares; (2) holding constant the A-share IPO size, the larger the flrm’s
22
intrinsic value, the more likely the flrm will issue B shares; (3) the larger the proflt prior
to the IPO, the more likely it will issue B shares; (4) the smaller the degree of IPO un-
derpricing, or the higher the after-market return on its A shares, the more likely it will
ofier B shares; (5) the shorter the time elapsed between the announcement of A-share
IPO and flrst-day market trading, or the more e–cient the flrm’s A-share IPO process,
the more likely it will ofier B shares.
One of the most interesting features distinguishing the ofiering process for B shares
from that of A shares is that lottery mechanism has never been used in allocating B
shares. Moreover, foreign securities flrms such as Sassoon and J. P. Morgan are allowed
to participate in the B-share underwriting process. The ofier prices for B shares are
announced approximately one month prior to the target market trading date and foreign
investors are invited to bid for the quantity of shares they wish to purchase. The absence
of lottery mechanisms for B shares suggests that excess demand for B shares may be
smaller than for A shares, or even nonexistent. This is consistent with our conjecture that
a separating equilibrium with signalling is not an optimal strategy for issuers in the B-
share market. Indeed, when we apply the empirical tests for the implications of signaling
models to the B-share data, we flnd that the signaling hypothesis does not appear to
be capable of explaining the positive initial returns on B-share ofierings. Moreover,
we note that the mean initial return on B shares issued as SEOs is not signiflcantly
difierent from that on B shares issued prior to or together with A-share IPOs, although
the mean A-share IPO underpricing for flrms that issue B shares as SEOs is double that
for those that ofier both A- and B-share IPOs. These results suggest to us that the initial
returns on B shares are not used as signals of flrm value to international investors, in
constrast to the use of A-share initial returns as signals to domestic investors. We also
observe that the degree of A-share underpricing is less for Chinese flrms participating in
international markets, presumably because available information reduces the degree of
information asymmetry.
23
6 Conclusion
We have tested several hypotheses that may explain the extraordinary high IPO un-
derpricing in China. In particular, we contrast implications of an hypothesis of a sepa-
rating equilibrium where flrms signal their quality by retaining shares and underpricing
against the null hypothesis of a pooling equilibrium where no signaling schedule ex-
ists and \underpricing" is purely random. We flnd that signaling hypothesis explains
the underpricing behavior for Chinese issuers rather well, although the market-feedback
hypotheses has some explanatory power and cannot be entirely rejected. On both the-
oretical and empirical grounds we reject bribery and the use of alternative IPO-share
allocation mechanisms as causes or explanations of the observed underpricing behavior.
Our major flndings in support of the signaling hypothesis with separating equilibrium
are:
(1) The correlation between the degree of IPO underpricing and ofier price is negative.
(2) The degree of IPO underpricing is positively related to the intrinsic value of the
issuer, the variance of future returns, and the issuer’s fractional ownership.
(3) Issuers with larger IPO underpricing are more likely to raise larger amounts of
capital through SEOs and to do so more quickly than issuers with smaller degree
of IPO underpricing, although the latter relationship is weak.
When we apply our analysis to the B-share data, we flnd that initial returns on B
shares are much lower than those in A shares and that traditional signaling equilib-
rium models do not explain the positive initial returns on B shares. We take this as
evidence that international investors in Chinese equities rely more on prior acquisition
of information than do Chinese domestic investors. Perhaps this is because the current
sample of international investors in Chinese equities is heavily weighted with experi-
enced, professional investors, while the vast majority of Chinese investors in A shares
are inexperienced in equity markets.
24
A Data Description
The data used to analyze the distribution of initial returns consists of all the flrm-
commitment IPOs of common stocks occurring between December 1986 and January
199610. There are 308 A-share IPOs in the full sample. A sub-sample of 268 IPOs
between December 1986 and January 1996 is extracted from the full sample to allow
studies on IPO underpricing and SEOs. Another sub-sample of 57 flrms that issued both
A and B shares are also taken out to allow studies on the difierences in underpricing
among A and B shares.
Returns are measured as Pt¡Pop Pop
, where Pt is the market price at time t and Pop is
the ofiering price. Table A1 presents the descriptive statistics for variables representing
after-market one-day returns (IPORETN, deflned as the difierence between the flrst-
day market closing price and the IPO price divided by the IPO price), after-market
one-week returns (IPORETN1), after-market two-week returns (IPORETN2) and
after-market four-week returns (IPORETN3). The results in table A.1 show that IPO
returns remain high and are invariant to the way they are measured.
Figure A.1 is the histogram of the cross-sectional distribution of the after-market one-
day returns for 308 flrm-commitment IPOs. As shown in the flgure, about 6 IPOs have
initial returns above 5000%. Figures A.2 and A.3 are the histograms of the cross-sectional
distribution of the after-market one-day returns and after-market four-week returns after
eliminating the outliers. A comparison of both flgures indicates that the distributions
of after-market one-day IPO returns and after-market four-week IPO returns are very
similar.
All of the data come from Shanghai Shenyin-International Securities, Xiamen Branch,
from the Chinese Stocks and Futures Encyclopedia published by Shanghai Xian Zi Infor-
mation Co., Ltd. and from various copies of annual reports of Shanghai and Shenzhen
securities exchanges.
10Since some companies used private placement in issuing stocks before \Corporate Law" was imple-
mented on December 29, 1993, We discard those stocks from the sample.
25
References
[1] Allen, Franklin and Gerald R. Faulhaber, 1989, \Signaling by Underpricing in the
IPO Market", Journal of Financial Economics 23, 303-323.
[2] Chemmanur, Thomas J., 1993, \The Pricing of Initial Public Ofierings: A Dynamic
Model with Information Production", The Journal of Finance 48, 285-304.
[3] Chowdhry, Bhagwan and Ann Sherman, 1994, \International Difierences in Over-
subscription and Underpricing of IPOs", UCLA working paper.
[4] Dewenter, Kathryn L. and Paul H. Malatesta, 1996, \Public Ofierings of State-
Owned and Privately-Owned Enterprises: An International Comparison", Univer-
sity of Washington working paper.
[5] Grinblatt, Mark and Chaun Y. Hwang, 1989, \Signaling and the Pricing of Unsea-
soned New Issue", The Journal of Finance 44, 393-420.
[6] Jegadeesh, Narasimhan, Mark Weinstein and Ivo Welch, 1993, \An Empirical In-
vestigation of IPO Returns and Subsequent Equity Ofierings", Journal of Financial
Economics 34, 153-175.
[7] Loughran, Tim, Jay R. Ritter and Kristian Rydquist, 1994, Initial public ofierings:
International insights, Paciflc-Basin Finance Journal 2, 165-199.
[8] Perotti, Enrico C., 1995, \Credible Privatization", American Economic Review 85,
847-859.
[9] Ritter, Jay R., 1984, \The ‘Hot Issue’ Market of 1980", Journal of Business 57,
215-240.
[10] Ritter, Jay R., 1991, \The Long-run Performance of Initial Public Ofierings", Jour-
nal of Finance 43, 789-822.
[11] Su, Dongwei, 1997, \Studies on the Behavior of Chinese Stock Markets", Ph.D.
Dissertation, The Ohio State University.
26
[12] Su, Dongwei and Belton M. Fleisher, 1996, \Risk, Return and Regulation in Chinese
stock markets", The Ohio State University working paper.
[13] Welch, Ivo, 1989, \Seasoned Ofierings, Imitation Costs and the Underpricing of
Initial Public Ofierings", The Journal of Finance 44, 421-449.
[14] World Bank, 1995, \China: The Emerging Capital Market", Volume I and II.
27
Variable Mean Median Minimum Maximum Number of Observations
Full sample of 308 flrm-commitment IPOs between December 1986 and January 1996
(in million U.S. dollar)
A Share IPOs 16.8289 11.988 0.2892 249.3976 308
Government shares 15.9073 3.6618 0 486.0919 219
Other State Enterprise Shares 7.1049 2.7123 0 289.1277 264
Employee Shares 0.8022 0.3614 0 22.8916 294
Seasoned Equity Ofierings 114.4246 33.6014 0 8513.8276 268
B Share Ofierings 4.0618 2.5398 0.8239 23.0747 57
(as a fraction of flrm’s intrinsic valuea, %)
A Share IPOs 15.02 12.6 0.07 81.25 308
Government shares 10.05 4.41 0 74.27 219
Other State Enterprise Shares 5.81 2.67 0 33.85 264
Employee Shares 0.87 0.39 0 11.09 294
Seasoned Equity Ofierings 49.93 41.58 0 99.87 268
aFirm’s intrinsic value is proxied by the value of the sum of all types of shares issued at the time of
IPO and all subsequent equity ofierings
Table 1: Sample Statistics for Share Ofierings
28
Variable Description Mean Median Std. dev. Minimum Maximum
Full sample of 308 flrm-commitment IPOs between December 1986 and January 1996
IPORETN IPO initial return 948.59% 231.25% 2967.7% -18.58% 38300%
AFTRETN1 flrst two-week return -2.6422% -5.3967% 24.2801% -92.6836% 170.3947%
after market trading starts
AFTRETN2 next two-week return -2.7498% -4.5075% 18.4111% -91.7275% 94.4444%
after market trading starts
IPOSZ IPO proceed (in million US$) 16.8289 11.988 22.5626 0.2892 249.3976
PROFIT past year’s proflt before 466.5848 221.988 98.1307 2.7735 3788.269
IPO (in million US$)
ln AGE logarithm of flrm’s age 2.4849 2.7701 0.8929 0.6931 4.4773
STD standard deviation of 5.8895% 4.7398% 5.1073% 0.6512% 51.2719%
flrst 100-day returns
MKTCAP flrm’s intrinsic value 161.2268 86.0064 237.4339 2.6145 1856.433
(in million US$)
TIMEIPO time elapsed between 260 135 341 3 1868
ofier and trade dates IPOSZ
MKTCAP IPO proceed as a 15.02% 12.6% 11.22% 0.07% 81.25%
fraction of flrm’s value
(Continue on the next page)
Table 2: Descriptive Statistics for Variables to Explain IPO Initial Returns
29
(Continued)
Variable Description Mean Median Std. dev. Minimum Maximum
Sub-sample of 268 flrm-commitment IPOs between December 1986 and June 1994
IPORETN IPO initial return 1043.1% 271.24% 3166.3% -10% 38300%
AFTRETN1 flrst two-week return -2.1446% -5.1389% 25.1854% -92.6836% 170.3947%
after market trading starts
AFTRETN2 next two-week return -3.7509% -4.9375% 16.8386% -91.7275% 69.3227%
after market trading starts
IPOSZ IPO proceed (in million US$) 16.8937 11.653 23.6056 0.2892 249.3976
PROFIT past year’s proflt before 430.6412 211.1253 83.6418 2.7735 3788.269
IPO (in million US$)
ln AGE logarithm of flrm’s age 2.7743 2.6391 0.9016 0.6931 4.4773
STD standard deviation of 6.0078% 4.7356% 5.42% 0.6512% 51.2719%
flrst 100-day returns
MKTCAP flrm’s intrinsic value 160.0253 88.586 236.7245 2.6145 1856.433
(in million US$)
TIMEIPO time elapsed between 251 142 306 3 1831
ofier and trade dates SEOSZ MKTCAP
proceeds from all SEOs as 49.93% 43.85% 28.95% 0 99.87%
a fraction of flrm’s value
TIMESEO time elapsed between 284 220 217 0 1217
flrst-day market trading
and flrst SEO date
Table 2: Descriptive Statistics for Variables to Explain IPO Initial Returns
30
Industry and year dummies are omitted for brevity.
A-share full sample, between December 1986 and January 1996
ln IPOSZ ln MKTCAP IPORETN AFTRETN1 STD ln PROFIT
ln IPOSZ 1
ln MKTCAP 0.4679 1
IPORETN -0.2891 0.02 1
AFTRETN1 -0.0412 0.0331 0.0487 1
STD 0.0464 -0.0168 0.0023 -0.0144 1
ln PROFIT 0.5485 0.6501 -0.0202 -0.0578 0.0134 1
A-share sub-sample, between December 1986 and June 1994
ln IPOSZ ln MKTCAP IPORETN AFTRETN1 STD ln PROFIT
ln IPOSZ 1
ln MKTCAP 0.4602 1
IPORETN -0.2907 0.0211 1
AFTRETN1 -0.0558 0.0313 0.0504 1
STD 0.0403 -0.0188 -0.0002 -0.0271 1
ln PROFIT 0.55 0.6539 -0.0044 -0.0601 0.0249 1
Table 3: Correlation Matrix for Variables to Explain IPO Initial Return
31
The dependent variable is the logarithm of IPO initial return (ln IPORETN). The independent vari-
ables are the instruments for the logarithm of IPO proceed ( dln IPOSZ) and the relative size of IPO
( dIPOSZ MKTCAP
). The instruments are obtained by regressing ln IPOSZ and IPOSZ MKTCAP
on exogenous vari-
ables including ln AGE, ln PROFIT , ln TIMEIPO, stock-exchange dummy (EXD), industry dum-
mies (SIC(k)) and IPO year dummies (Y EAR(t)). Figures in parentheses are t-statistics. ⁄, y denote
5% and 10% level of signiflcance, respectively.
Variable Full Sample Sub-Sample
Constant 15.1907⁄ 13.9522⁄
(23.2712) (20.1465)
dln IPOSZ -0.8612⁄ -0.7021⁄
(-11.3981) (-8.5542) dIPOSZ
MKTCAP -11.2312⁄ -12.3773⁄
(-10.4522) (-10.7187)
„R2 0.5396 0.5469
Table 4: 2SLS Regression Estimates for the Correlation Between IPO Underpricing and
IPO Price
32
The dependent variable is the logarithm of IPO initial return (ln IPORETN). The independent vari-
ables are the instrument for the logarithm of the size of initial ofierings ( dln IPOSZ), instrument for the
logarithm of flrm’s intrinsic value ( dln MKTCAP ), and instrument for the project variance ( dSTD). The
instruments are obtained by regressing ln IPOSZ, ln MKTCAP , and STD on exogenous variables
including ln AGE, ln PROFIT , ln TIMEIPO, stock-exchange dummy (EXD), industry dummies
(SIC(k)) and IPO year dummies (Y EAR(t)). Figures in parentheses are t-statistics. ⁄, y denote 5%
and 10% level of signiflcance, respectively.
Variable Full Sample Sub-Sample
Constant -5.1423⁄ -6.2608⁄
(-2.9377) (-3.1859) dln IPOSZ -1.8383⁄ -1.7676⁄
(-19.9304) (-17.411) dln MKTCAP 1.1406⁄ 1.1646⁄
(12.0549) (10.9617) dSTD 10.9575⁄ 10.2704⁄
(4.0374) (3.7736)
„R2 0.6216 0.5967
Table 5: 2SLS Regression Estimates for the Correlation Among IPO Underpricing and
Issuer’s Intrinsic Value, Fractional Ownership and Project Variance
33
The dependent variable is the probability for the ith flrm to issue SEOs (PSEOi ). The dependent variable
takes value 1 if SEOs are observed and 0 otherwise. The independent variables are the observed IPO
initial return (IPORETN), after-market flrst two-week return (AFTRETN1) and after-market next
two-week return (AFTRETN2). We allow 548 days for a flrm to issue SEOs, therefore our sample only
consists of flrms who went public between December 1986 and June 1994. There are 268 flrms in the
sample. Figures in parentheses are asymptotic t-statistics. ⁄, y denote 5% and 10% level of signiflcance,
respectively.
Variable Coe–cient t-statistic
Constant 2.0242⁄ (6.5531)
IPORETN 0.001326⁄ (2.1301)
AFTRETN1 0.031597⁄ (2.1831)
AFTRETN2 0.014543 (0.8885)
Table 6: Logit Regression Estimates for the Relationship Among IPO Underpricing,
After-Market Returns and Probability of SEOs
34
The dependent variable is the ratio of total seasoned equity ofierings as a fraction of the intrinsic
value for the ith flrm ¡
SEOSZ MKTCAP
¢ i . The independent variables are the observed IPO initial return
(IPORETN), after-market flrst two-week return (AFTRETN1) and after-market next two-week re-
turn (AFTRETN2). We allow 548 days for a flrm to issue SEOs, therefore our sample only consists
of flrms who went public between December 1986 and June 1994. There are 268 flrms in the sam-
ple. Figures in parentheses are asymptotic t-statistics. ⁄, y denote 5% and 10% level of signiflcance,
respectively.
Variable Coe–cient t-statistic
Constant 0.4312⁄ (25.8477)
IPORETN 0.0000309⁄ (6.127)
AFTRETN1 0.002015⁄ (2.6024)
AFTRETN2 0.000483 (0.5501)
Table 7: Tobit Regression Estimates for the Relationship Among IPO Underpricing,
After-Market Returns and Size of SEOs
35
The dependent variable is the number of days between the IPO date and the flrst SEO date for the ith
flrm TIMESEO. The independent variables are the observed IPO initial return (IPORETN), after-
market flrst two-week return (AFTRETN1) and after-market next two-week return (AFTRETN2).
We allow 548 days for a flrm to issue SEOs, therefore our sample only consists of flrms who went public
between December 1986 and June 1994. There are 268 flrms in the sample. Figures in parentheses are
asymptotic t-statistics. ⁄, y denote 5% and 10% level of signiflcance, respectively.
Variable Coe–cient t-statistic
Constant 291.0871⁄ (22.3196)
IPORETN -0.00222 (-0.5245)
AFTRETN1 1.0864⁄ (2.2203)
AFTRETN2 0.7852 (0.9593)
Table 8: Tobit Regression Estimates for the Relationship Between IPO Underpricing,
After-Market Returns and Time Elapsed Between IPO and First SEO
36
The dependent variable is the logarithm of IPO initial return (ln IPORETN). The independent variables are the instru-
ment for the logarithm of the size of initial ofierings ( dln IPOSZ), instrument for the logarithm of flrm’s intrinsic value
( dln MKTCAP), instrument for the project variance ( dSTD) and lottery dummy (LD) or ofiering mechanism dummies
(OD(l)). LD takes value one if a flrm uses lottery mechanism or auction mechanism with only quantity bids in allocating
new shares and 0 if a flrm uses auction mechanism with both price and quantity bids in allocating new shares. OD(l) is
a set of flve dummy variables representing lottery mechanism based on flxed amount of application forms, lottery mecha-
nism based on unlimited amount of application forms, lottery mechanism based on certiflcate of deposit receipts, auction
mechanism with only quantity bids and auction mechanism with both quantity and price bids. Figures in parentheses are
t-statistics and ⁄, y denote 5% and 10% level of signiflcance, respectively.
Full sample Sub-sample
CONSTANT -3.6962⁄ -2.4563⁄ -4.5103⁄ -3.1875⁄
(-2.624) (-1.9631) (-2.8844) (-2.2574) dln IPOSZ -1.4253⁄ -0.9214⁄ -1.3736⁄ -0.927⁄
(-17.6848) (-11.0507) (-15.8996) (-10.6195) dln MKTCAP 0.8819⁄ 0.6002⁄ 0.8947⁄ 0.6327⁄
(11.2355) (8.2209) (10.2935) (7.8407)
dSTD 8.8957⁄ 4.2494⁄ 8.7891⁄ 4.3763⁄
(4.075) (2.2415) (4.0686) (2.2835)
LD 1.3404⁄ 1.4428⁄
(12.9301) (12.5193)
OD1 2.942⁄ 2.9669⁄
(16.1233) (14.6956)
OD2 1.9509⁄ 1.9991⁄
(11.6627) (10.5858)
OD3 1.2655⁄ 1.2849⁄
(6.7202) (6.0654)
OD4 0.8824⁄ 0.8319⁄
(4.6508) (3.7219)
„R2 0.7565 0.8319 0.7467 0.8198
Table 9: 2SLS Regression Estimates for the Relationship Between IPO Underpricing
and Ofiering Mechanisms
37
Variable Description Mean Median Std. dev. Minimum Maximum
57 B-share IPOs
IPORETN IPO initial return 37.13% 21.43% 47.55% -21.14% 236.45%
AFTRETN1 flrst two-week return -5.0245% -4.5455% 14.2491% -54.2373% 24%
after market trading starts
AFTRETN2 next two-week return 3.9321% 0 21.5556% -27.5304% 109.2593%
after market trading starts
ln P0 logarithm of IPO price -0.7009 -0.7351 0.4333 -1.8123 0.3833
IPOSZ IPO size (in million US$) 33.7131 21.08 38.8529 6.8384 191.52
PROFIT past year’s proflt before 859.6096 514.3502 88.8802 137.2619 3788.269
IPO (in million US$)
LNAGE logarithm of flrm’s age 2.6791 2.3979 0.9309 1.0986 4.3307
MKTCAP flrm’s intrinsic value 297.9084 192.5356 318.699 60.4435 1856.433
(in million US$)
TIMEIPO time elapsed between 71.807 44 81.98 6 348
ofier and trade dates
57 Corresponding A-share IPOs
IPORETN IPO initial return 838.9131% 270.5882% 1226.995% 4.5652% 4514.286%
AFTRETN1 flrst two-week return 1.7196% -2.214% 30.6146% -92.574% 133.5%
after market trading starts
AFTRETN2 next two-week return -1.797% -6.307% 22.4403% -85.348% 94.4444%
after market trading starts
ln P0 logarithm of IPO price -0.5691 -0.5147 0.433 -1.7492 0.2249
IPOSZ IPO size (in million US$) 14.1521 11.7398 11.3231 0.5446 56.2316
TIMEIPO time elapsed between 208.54 142 225.13 3 980
ofier and trade dates
Table 10: Descriptive Statistics for Variables to Explain B-share Initial Returns
38
The dependent variable is the probability for the ith flrm to issue B shares after it completes ofier-
ing A shares (PBi ). The independent variables are the IPO initial return (IPORETN), logarithm of
IPO size (ln IPOSZ), logarithm of flrm’s intrinsic value (ln MKTCAP ), after-market flrst two-week re-
turn (AFTRETN1), after-market next two-week return (AFTRETN2), logarithm of past year’s proflt
prior to IPO (ln PROFIT ), logarithm of time elapsed between the ofier date and the flrst trading date
(ln TIMEIPO). We also allow the probability of issuing B shares to vary across industries (SIC(k)).
Figures in parentheses are asymptotic t-statistics. ⁄, y denote 5% and 10% level of signiflcance, respec-
tively.
Variable Coe–cient t-statistic
CONSTANT -33.7464⁄ (-5.7972)
IPORETN -0.000145 (-1.2212)
AFTRETN1 0.01324y (1.8072)
AFTRETN2 0.01079 (1.1565)
ln IPOSZ -1.5739⁄ (-5.1046)
ln PROFIT 0.4223y (1.7704)
ln MKTCAP 1.8664⁄ (5.7864)
ln TIMEIPO -0.3693y (-1.814)
SIC1 0.9968 (1.4674)
SIC2 1.4209⁄ (1.936)
SIC3 0.4717 (0.5927)
SIC4 -0.2068 (-0.2322)
SIC5 0.478 (0.5382)
Table 11: Logit Regression Estimates for the Probability A Firm Will Issue B Shares
39
Variable Description Mean Median Std. dev. Min. Max.
Full Sample, between December 1986 and January 1996
IPORETN after-market one-day return 9.4859 2.3125 29.6766 -0.1858 383
IPORETN1 after-market one-week return 9.5098 2.2199 34.9891 -0.1764 489.2
IPORETN2 after-market two-week return 9.559 2.2 35.7948 -0.2795 504.1
IPORETN3 after-market four-week return 8.9839 2.0386 36.6929 -0.4205 524.7
Sub-sample, between December 1986 and June 1994
IPORETN after-market one-day return 10.431 2.7124 31.6625 -0.1 383
IPORETN1 after-market one-week return 10.5117 2.5368 37.3834 -0.1744 489.2
IPORETN2 after-market two-week return 10.586 2.5159 38.2476 -0.2795 504.1
IPORETN3 after-market four-week return 9.9394 2.2821 39.2312 -0.4205 524.7
Sub-sample, 57 flrms that issue both A and B shares
IPORETN (B) after-market one-day return 0.3713 0.2143 0.4755 -0.2114 2.3645
IPORETN1 (B) after-market one-week return 0.3319 0.1976 0.4869 -0.2577 2.2243
IPORETN2 (B) after-market two-week return 0.31 0.1919 0.502 -0.5187 1.8855
IPORETN3 (B) after-market four-week return 0.356 0.1852 0.5824 -0.4538 1.9434
Sub-sample, 57 flrms that issue both A and B shares
IPORETN (A) after-market one-day return 8.3891 2.7059 12.27 0.0457 45.1429
IPORETN1 (A) after-market one-week return 7.5239 2.76 11.372 0.0217 52
IPORETN2 (A) after-market two-week return 7.9227 2.775 11.6729 -0.0196 48.1
IPORETN3 (A) after-market four-week return 7.3428 2.53 10.6262 -0.087 45.375
Table A1: Descriptive Statistics for Variables Chosen to Represent IPO Returns
40
