Download What Determines the Survival of Internet IPOs?

What Determines the Survival of Internet IPOs?
Michiel Botman
University of Amsterdam
Tjalling van der Goot*
University of Amsterdam
Noud van Giersbergen
University of Amsterdam
刪除:
An earlier version of this paper has been designated for the outstanding Empirical Research award of the
2004 Southern Finance Association conference at Naples (Florida, USA). We would like to thank
participants of the Southern Finance Association meeting, in particular Bharat Jain, for their helpful
comments. Furthermore, we thank an anonymous referee for his/her helpful suggestions and constructive
comments. The usual disclaimer applies.
* Corresponding author. Correspondence address: Department of Business Studies, University of
Amsterdam, Roetersstraat 11, 1018 WB Amsterdam, The Netherlands. Phone: +31 20 525 4171, Fax: +31
20 525 5281, E-mail: [email protected].
刪除: are grateful to
刪除: The usual disclaimer
applies.
What Determines the Survival of Internet IPOs?
Abstract
This paper examines whether the variables that are significant in non-internet IPOs play a
similar role for internet IPOs. To this end, we analyze the determinants of survival of internet firms
that have gone public at the NASDAQ stock exchange from December 1996 through February 2001.
Financial and non-financial data published in the IPO prospectuses are examined.
For our analysis, we use the semiparametric Cox proportional hazard model and estimate the
effect of these variables on the trading (survival) time using the parametric Log-logistic survival
model. It appears that the average operating history of internet IPOs is remarkably small compared to
non-internet IPOs, namely 2.4 and 10 years, respectively. Furthermore, we find that the average
number of risk factors for internet IPOs is four times higher than the number reported for non-internet
IPOs. The econometric results show that six of the twelve explanatory variables in our model are
significant. The results of the parametric analysis correspond to the semiparametric results.
Consistent with studies of non-internet IPOs, our empirical results provide evidence for the
significant effects of high prestigious underwriters and retained ownership. None of the investigated
variables stemming from accrual accounting are significant, whereas the financial ratio based on cash
accounting is.
Our findings hold under a number of different model specifications and robustness checks.
The sensitivity analysis in the Log-logistic model reveals that the greatest positive effect on the
survival time comes from investor demand followed by operational cash flow over liabilities. The
expected survival time is shortened mostly by IPO market level, followed by valuation uncertainty.
Keywords:
JEL codes:
internet firms, financial and market information, initial public offering
G1, M4
2
刪除: , which
刪除:
刪除: 3
刪除: situation
What Determines the Survival of Internet IPOs?
刪除: is quite different with
respect to
刪除: Following the increase of
the rate of corporate failure which
can be observed during the last
decade, Demers and Joos (2006)
report that academic research on
this topic is growing also.
Whereas many papers focus on
the prediction of bankruptcy of
already listed firms o
刪除: study
1.
刪除: papers
Introduction
刪除: by
Numerous studies have investigated which variables play an important role in determining the value
of IPOs (see Section 2 hereafter for references). This paper, however, focuses on internet IPOs and
刪除: Hensler, Rutherford, and
Springer (1997); hereafter Hensler
et al.
刪除: no
刪除: are in their sample
刪除: of
their survivability. Our paper follows the article by Hensler, Rutherford, and Springer (1997, hereafter
Hensler et al.), which is one of the few studies that examines the survivability of IPO firms. Because
刪除: do comprise
刪除: However, t
刪除: T
the data from Hensler et al. are from the years 1975-1984, their sample does not contain internet firms.
刪除: examine about 4,200 IPOs
of which
By examining IPO failure risk during the years 1980-2000, Demers and Joos (2006) have both
刪除: the
internet and non-internet IPOs in their sample. However, the latter authors conclude that internet IPOs
刪除: form a relatively small part
of their study
刪除: . T
are very different from non-internet IPOs and therefore they omit the internet IPOs from their sample
刪除: ,
刪除: for a number of tests
for a number of tests.
刪除: of
Given the high failure risk of the internet sector and their different nature, it is interesting to
刪除:
刪除: Given the
examine whether variables, which are used in earlier studies are also important for internet IPOs.
刪除: importance
Therefore, this paper studies internet firms that have gone public at the NASDAQ stock exchange
刪除: high failure risk of the
internet sector and their different
nature
during the period December 1996 through February 2001. Due to the relatively high market values of
刪除: for the economy today and
in the future
internet firms during the internet bubble years 1998-1999 and the sharp decline of their market
刪除: , it is interesting to
valuations from early 2000 onwards, prior research on internet firms has focused on explaining pre
and post downturn market values and the internet stock downturn phenomenon itself. The main aim of
this paper, however, is to investigate the rate of survival of internet firms directly. Here, survivors
refer to firms that remain listed on the exchange from the time of IPO (initial public offering)
刪除: see
刪除: examine whether new
variables, which are characteristic
... [1]
刪除: both
刪除: earlier studies
刪除: aforementioned
刪除: are also important with... [2]
刪除: .
3
throughout the studied period. Non-survivors denote firms that are delisted from the exchange for
刪除: for instance, in line with
Dimovski and Brooks (2003)
negative reasons, such as not meeting the minimum bid price requirements or bankruptcy.
刪除: (see
Our research design is specifically chosen to focus on the key issue of this study: the variables
that predict the rate of success and failure of firms in the internet industry. We use both financial and
non-financial information that is available to investors from the IPO prospectuses. The effect of
刪除: , for instance,
刪除: among others
刪除:
刪除: Dimovski and Brooks,
2003)
刪除: .
financial information is measured through several financial multiples, for instance net sales over
assets and operating cash flows over liabilities. In addition to financial information, we examine firmspecific non-financial information, such as the number of risk factors mentioned in the IPO prospectus,
刪除: in
刪除: his study on the factors that
刪除: for internet start-ups, even
though his dependent variable is
different from ours
刪除: affect
lead investment bank reputation, and insider ownership retention; see also Grilli (2005) for the use
刪除: by
non-financial information, even though his dependent variable is the amount of bank loans for internet
刪除:
start-ups. Although many authors analyze the value of internet firms (Bartov et al., 2002; Cassia et al.,
刪除: Although m
刪除: As aforementioned,
刪除: M
2004; Demers and Lev, 2001; Demers and Lewellen, 2003; Dimovski and Brooks, 2003; Johnston and
Madura, 2002; Knauff et al., 2003; Ljungqvist and Wilhelm, 2003; Loughran and Ritter, 2002 and
2004), to the best of our knowledge, this paper is the first that focuses exclusively on the analysis of
刪除: papers
刪除: have analyzed
刪除: value
刪除: for instance
刪除:
survival determinants for internet IPOs. Furthermore, our model incorporates a number of explanatory
variables that are analyzed simultaneously instead of separately.
Due to the structure of the data involved, i.e. having a duration dependent variable where
some survival periods are censored (see Section 4 hereafter), we use the Cox (1972) proportional
刪除: 3
刪除: . Also, studies on the
demand elasticities for internet
services (Goel et al., 2006), and
the financing of internet firms
(Grilli, 2005) have
刪除: s
刪除: been published.
hazard method to investigate which variables determine the rate of success and failure of internet
firms. In addition to comparing mean differences, which can be classified as a two-sample analysis,
we account for all cross-correlations between the explanatory variables in the hazard analysis.
刪除: to
刪除: T
刪除: firms
刪除: Also
刪除: that
Furthermore, we use the parametric Log-logistic survival model to estimate the effect of individual
variables on the expected survival time. Except for firm risk, both the Cox and Log-logistic models
provide qualitatively similar results.
The remainder of the paper is organized as follows. Section two discusses our hypotheses on
internet firm survival. Section three describes our data sample and the descriptive statistics. The
刪除: Our empirical results
provide evidence for the
significant effects of high
prestigious underwriters and
retained ownership on the survival
rate of an internet IPO. This is
consistent with the outcome from
studies on non-internet IPOs. Also,
our findings show that internet
shares bought during a stock
market level peak have a lower
survival rate. Finally, none of...the
[3]
刪除: two
刪除: ir
刪除: Section three discusses...our
[4]
4
fourth section presents the hazard methodology. Section five contains the empirical results, robustness
checks, and sensitivity analysis. The final section summarizes and concludes the paper.
2.
刪除: Section five contains the
empirical results, robustness
checks, and economic
significance
刪除: .
Hypotheses
In this section, we discuss the hypotheses with respect to the hazard rate; see Formula (1) in Section 4
刪除: .
for a formal definition. The higher the hazard rate, the higher the probability that a firm will be
刪除: Recall that t
delisted and the shorter the trading (survival) time.
A common way to quantify firm specific risk is the use of the standard deviation of
aftermarket returns. Before the offering, however, investors do not know aftermarket returns.
Therefore, we follow Beatty and Welch (1996), Hensler et al. (1997) and Van der Goot (2003) by
using the number of risk factors reported in the IPO prospectus as a direct measure for ex ante firmspecific risk. Hence, the first hypothesis is:
H1:
There is a positive relation between the number of risk factors and the hazard rate of a firm.
刪除: capital
Investment banks, also referred to as underwriters, depend on their reputation to generate new
business. The investment banks with the highest reputation are considered to be more successful at
performing activities, such as taking firms public. This creates an incentive to be selective at picking
firms for offering, to monitor firms after the offering, and to create interest for the firms’ stocks by
distributing analyst reports (see Hansen and Torregrosa, 1992; Macklin et al., 1992; Jain and Kini,
1999a and Aggarwal, 2000). All authors assign higher scores to higher reputation underwriters, and
vice versa. We assign a value of 1 to offerings taken public by investment banks with reputations of 8
or 9; a value of 0 is assigned to an IPO if the reputation of the lead manager is lower.
H2:
A high reputation of the lead manager is inversely related to a firm’s hazard rate.
5
In general, larger firms have better access to the capital market and are better equipped to
withstand rough market conditions and adverse outcomes of investment decisions. Many studies
document a higher risk of failure for smaller firms (Schultz, 1993; Jain and Kini, 1999b). Therefore,
we expect that firm size is inversely related to a firm’s hazard rate. We use the logarithm of the dollar
amount of the gross offering proceeds as a measure of size. The hypothesis is as follows:
H3:
There is a negative relationship between firm size and its hazard rate.
Before an IPO takes place, an investment bank sets a preliminary price range and conducts
road shows to market the offering. During the road show, the investment bank receives signals about
investors’ interest in the offering. Guided by these signals, an investment bank may decide to adjust
the initial price range, where upward adjustments indicate high investor demand. Following Hanley
(1993), the measure of this investor demand is the price change of the final offer price as percentage
of the average value of the initial price range, where we also correct for changes in the number of
shares offered. A higher demand from professional investors can be seen as a positive signal with
regard to an internet IPO.
H4:
A firm’s hazard rate is a decreasing function of investor demand.
Investment banks start the process of marketing an IPO by setting an initial price range as
required by the regulatory authorities. At that time, there are only limited indications regarding
investors’ interest in the firm. Consequently, investment banks are depending on their own judgment
for setting the preliminary price range. High uncertainty about firm valuation, for instance, due to a
limited operating history or the infant industry argument, is likely to result in relatively large spreads
of the initial price range. Jain and Kini (1999b) relate this valuation uncertainty to uncertainty about a
firm’s performance after the offering. They measure valuation uncertainty as the dollar spread of the
6
initial price range divided by the average value of the initial price range. The authors find a negative
relationship between valuation uncertainty and firm survival. Therefore, the hypothesis is:
H5:
A firm’s hazard rate is an increasing function of valuation uncertainty.
Leland and Pyle (1977) argue that firm owners can signal quality in equity markets by
retaining equity. Consistent with their signaling theory, a high percentage of insider ownership
retention at IPOs serves as a certification that managerial decisions will coincide with the outside
shareholders’ interest, which results in better firm performance after the offering (Jensen and
Meckling, 1976), and a longer survival period (Hensler et al., 1997). Insider ownership retention is
measured as the number of shares held by the original owners as percentage of the total number of
shares outstanding after the offering.
H6:
A firm's hazard rate is inversely related to the percentage of insider ownership retention at the
offering.
Several authors (for instance, Ritter, 1991; Jegadeesh et al., 1993; Hensler et al., 1997)
document the relation between a firm’s age at the time of offering and its risk. The rationale behind
this finding is twofold. First, older firms are more mature and in a more steady state and, therefore,
have less uncertainty about future operating performance. Secondly, the managers of mature firms are
in a better position to reduce information asymmetry at the time of offering when a firm has a longer
operating history. We examine the operating history of a firm’s internet activities at the time of
offering to capture the fact that several non-internet firms have transformed their business model to
become internet firms. We conjecture that an internet firm’s hazard rate is a decreasing function of its
age.
7
H7:
A firm’s hazard rate is inversely related to the duration of its internet activities at the time of
offering.
Managers recognize periods, in which equity market levels are relatively high, e.g. due to
over-optimism about the earnings potential of young growth companies. They take advantage of these
‘windows of opportunity’ by taking their firms public, which results in IPO volume peaks (Lowry and
Schwert, 2002). Firms, which have gone public during these peaks, underperform relative to other
offerings. This can be attributed to lower quality issuers being unable to withstand periods of
economic downturn (Ritter, 1991). This finding is confirmed by Hensler et al. (1997), who report a
negative relation between market level at the IPO and firm survival. Market level is measured as the
average value of the NASDAQ Composite Index during the month of offering.
H8:
There is a positive relationship between market level at the time of offering and a firm’s
hazard rate.
Further, we examine the relation between a firm’s hazard rate and its offer-to-book ratio. The
firm’s offer-to-book ratio can be seen as an indication of an internet IPO's growth opportunities (Kim
and Ritter, 1999; Kogan, 2004). The larger its offer-to-book ratio the higher the growth opportunity of
an internet IPO. In addition, when the offering price is high the firm receives a larger amount of
money in its IPO with which it can survive longer at a given “burn rate” of cash. Hence, we
hypothesize that the firm's hazard rate is a decreasing function of its offer-to-book ratio.
H9:
There is a negative relation between a firm’s hazard rate and its offer-to-book ratio.
Finally, we test the hypothesis that an internet firm’s accounting performance prior to the
刪除: , and
offering affects its survival (for instance, Jain and Kini, 1994; Jain and Kini, 1999b; Bhabra and
刪除: 2
Pettway, 2003). Therefore, we have included in our model financial ratios: i) two variables based on
8
accrual accounting, namely net sales over assets and receivables over total assets, and ii) one based on
cash accounting, operating cash flow over liabilities. When an internet firm uses its total assets
measured by its net sales over assets more efficiently, we expect that its rate of survival will be greater.
刪除:
Furthermore, we expect a positive relationship between an internet firm’s survival and its short-term
liquidity measured by operating cash flow over liabilities and receivables over total assets,
respectively.
H10:
There is a negative relation between a firm’s hazard rate and its net sales over assets,
operating cash flow over liabilities, and receivables over total assets, respectively.
3.
Data and Descriptive Statistics
3.1
Sample Selection
In this paper, the survivability of U.S. internet IPOs that went public during the period December
刪除: 0
1996 through February 2001 is studied. Following Hand (2003), internet firms are defined as firms
obtaining more than fifty percent of their revenues through or because of the internet. Since this
definition does not rule out arbitrariness, we have compiled our sample from two sources of internet
刪除: 3
IPOs. Our first source is a list of 527 internet-related offerings used by Loughran and Ritter (2004),
who obtained their data by merging and amending internet identifications of Thomson Financial
Securities Data, Dealogic and IPOmonitor.com. Next, we matched this list against the firms marked
as internet-related by our second source, www.edgar-online.com. IPOs documented by both sources
are included in our initial sample, which contains 382 firms.
In order to be included in the final sample, firms have to meet two criteria. First, firms must
be listed at the NASDAQ stock exchange to create equal delisting norms throughout the sample.
Secondly, the final prospectus must be available at www.sec.gov, including annual accounts covering
9
a full fiscal year. Furthermore, unit offerings and financial institutions are excluded from the sample
as the characteristics of these IPOs differ significantly from other offerings.
From our initial sample, 13 firms were excluded because they were issued at an exchange
other than the NASDAQ. Two firms are left out, as their final prospectuses were not available. For 31
firms the annual accounts accompanying the prospectus do not cover a full year. A further eight firms
are financial institutions and two more firms are left out of the sample as, on closer inspection, the
company description in the prospectus does not justify the status of internet related IPO according to
the aforementioned definition. After the exclusion of those 56 firms, our final sample consists of 326
刪除: ,
internet offerings. Figure 1 illustrates the time distribution of these internet IPOs with peaks in 1999
and the beginning of 2000.
<Insert Figure 1 about here>
Many variables used in our analysis (for instance, the number of risk factors, underwriter
reputation, valuation uncertainty, and the financial ratios) have been hand-collected from the final
offering prospectuses of the issuing firms. The date of the first trading day has been obtained from
www.edgar-online.com.
We have segmented our sample into two different aftermarket states: survivors and nonsurvivors. These states of the internet firms in the aftermarket are determined as of March 21, 2005,
using data from www.nasdaq.com and www.sec.gov. This period encompasses the bubble and
subsequent burst of the internet industry. Survivors are internet IPOs that remain listed on the
NASDAQ stock exchange from the time of issue throughout the studied period ending at March 21,
2005. Non-survivors are classified as internet firms delisted from the NASDAQ stock exchange due
to negative reasons. Hence, in this paper the terms non-surviving and delisted both refer to the same
aftermarket state. With respect to our sample, negative reasons include both failing firms and firms
moving to other exchanges with less strict listing criteria, such as the NASDAQ Small Caps market or
10
the Bulletin Board (OTCBB); see Table 1 for an overview of the reasons for delisting. Of the 326
internet firms, 115 firms have been acquired or merged. These firms were excluded from the analysis.
< Insert Table 1 about here>
Figure 2 illustrates the time allocation of delistings of the firms studied in this paper.
Although there are concentrations of delisted firms in several quarters, the burst of the internet bubble
刪除: statistics
of spring 2000 is not evident from these numbers.
< Insert Figure 2 about here>
3.2
Summary statistics of the independent variables
In this section we describe the independent variables that were used to explain the hazard rate of the
Internet IPOs. Table 2 lists these variables and provides a description of each of them. Some of the
variables were transformed to get approximately normally distributed data and to reduce the impact of
extreme (untransformed) observations. All transformations were chosen to preserve the sign of the
observations, i.e. a positive (negative) untransformed observation corresponds to a positive (negative)
transformed observation. Table 3 presents the summary statistics of the explanatory variables.
<Insert Table 2 about here>
As can be seen in Table 3, the IPO prospectuses mention 32.0 (32) risk factors on average
(median). The average number of risk factors is much higher than the 14 reported in Beatty and
Welch (1996) and the 8 reported in Hensler et al. (1997). The fraction of underwriters with a high
prestige is 89 percent. The measure we use for investment bank reputation is based on the relative
11
placement of underwriters in tombstone advertisements, as originally developed by Carter and
Manaster (1990) and Carter, Dark and Singh (1998) and later updated and amended by Loughran and
刪除: 3
Ritter (2004) for the period 1980 through 2000. The non-logged average (median) firm size is USD
94.4 (67.5) million. Hensler et al. (1997) report an average firm size of USD 11.3 million. Investor
demand has an average (median) value of 28.1 (25) percent. Insider ownership retention is on average
(median) 81.3 (82.3) percent. The percentage of 81.3 is relatively high compared to Ljungquist and
刪除: for all IPO firms
Wilhelm (2003), who report for all IPO firms a percentage of 63.9 in 1996 declining to 51.8 percent in
刪除: percent
2000. The operating history of an IPO firm's internet activities is on average (median) 28.8 (27.9)
months or 2.40 (2.33) years. The latter figure is small compared to Ritter (1991) and Hensler et al.
(1997), who find an average IPO firm age of slightly more than 10 and 8.6 years respectively. As can
be inferred from Table 3, the coefficient of variation of the internet IPOs’ age is small (0.54)
compared to the figure of 1.25 reported by Hensler et al. (1997).
<Insert Table 3 about here>
The Pearson correlation matrix of the explanatory variables is shown in Table 4. All
刪除: for
correlations are below 0.5, except the one between two financial ratios.
刪除: of the
<Insert Table 4 about here>
Table 5 presents the mean differences of the explanatory variables for the two aftermarket
states examined. Three out of nine of the mean differences regarding the offering and firm
characteristics are significant, whereas only one of the three financial ratios is insignificant. Surviving
firms have a higher percentage of underwriters from the high reputation bracket (96.6 percent).
Further, the average values of investor demand and insider ownership retention of survivors are
刪除:
significantly larger than those of the non-survivors.
12
As can be seen in Table 5, the mean differences of firm size are not only insignificant, but
also from Table 3 can be inferred that the coefficient of variation is much lower than that of Hensler et
al. (1997), namely 0.04. The latter authors report a coefficient of variation for firm size of 1.32.
Otherwise stated, the internet IPOs sample is more homogeneous than that of Hensler et al. As noted
earlier, a similar reasoning regarding the internet IPOs’ age holds. Although less significant, survivors
have better values with respect to operational cash flow over liabilities and receivables over total
assets.
<Insert Table 5 about here>
4.
Methodology
This section describes the methods used to analyze our data. We begin by illustrating the
characteristics of our dependent variable. Thereafter, we discuss nonparametric, semiparametric and
刪除: function
parametric estimation of the hazard rate.
4.1
Characteristics of the Dependent Variable
The dependent variable in this study is the survival time or duration of internet firms on the
NASDAQ stock exchange. The duration of non-survivors is measured as the number of months from
the IPO through the date of delisting. The duration of survivors is measured by the number of months
from the IPO through the end of the measurement period, March 21, 2005, as their duration has not
been completed up to that point in time.
Duration is completed when one of the events mentioned in Table 1 has occurred. As these
events are still to occur for surviving firms, the completed durations for these firms cannot be
measured, and instead, only uncompleted durations are observed. These uncompleted durations are
13
called right-censored observations1. Models can be estimated with or without censored data, although
models including censored data are considered more efficient.
We use the Kaplan-Meier methodology (without explanatory variables) for nonparametric
estimation of the hazard function. The semiparametric Cox proportional hazard model is used to
incorporate explanatory variables. Finally, the effect of the individual variables on the expected
survival times is estimated by a parametric Log-logistic survival model. For an extensive theoretical
discussion of hazard models, we refer to Kalbfleisch and Prentice (1980), Kiefer (1988) and Lancaster
(1990). We refer to delistings as an event for the remainder of this section.
4.2
Nonparametric Estimation of the Hazard Function
As a preliminary analysis, we estimate the shape of the hazard function by the nonparametric KaplanMeier (product limit) estimator; see Kaplan and Meier (1958). Let T be a random variable measuring
the duration of an offering at the NASDAQ stock exchange, with duration distribution function2 F(t)
= Pr (T < t) and density function f(t) = dF(t)/dt. When the distribution function of T is specified
alternatively, we obtain the survival function S(t) = 1 – F(t) = Pr(T ≥ t), being the probability that
刪除: will equal or exceed
random variable T is at least the value t. Now, our duration data can be characterized in terms of the
hazard function, i.e. the probability that an event occurs during an infinitesimal small time interval dt,
given that it has survived up to t:
λ (t ) =
f (t ) Pr (t ≤ T < t + dt | T ≥ t )
.
≈
S (t )
dt
(1)
1
Left-censoring occurs when durations have started before the beginning of the measurement period, which is
not applicable to our data.
2
Note that this definition is slightly different from the ordinary statistical terminology, i.e. cumulative
distribution function G(t) = Pr (T ≤ t).
14
刪除: sample
Figure 3 presents a smoothed plot of the estimated nonparametric Kaplan-Meier hazard rate for non刪除: illustrates
survivors; see the Appendix for the formulae. The figure shows an increasing hazard rate from the
offering through around the third year of listing, after which the hazard rate decreases. The peak
hazard rate is around 2¾ percent per month or around 33 percent annualized.
<Insert Figure 3 about here>
4.3
Semiparametric Estimation of the Hazard Function
As we do not have strong a priori reasons for imposing a particular functional form on the dependence
of a firm’s hazard rate on time, we model this particular relationship nonparametrically. In contrast,
the effects of the explanatory variables on the hazard rate are modeled parametrically. The
semiparametric Cox proportional hazard model can be written as:
λ(ti ) = λ 0 (ti ) exp(β ' xi ),
(2)
where xi denotes a vector of explanatory variables for the i-th firm with unknown coefficients β and
λ 0 (ti ) is the baseline hazard function.
Cox (1972) has shown that β can be estimated by maximum likelihood; see the Appendix for
an expression of the log-likelihood function. Although the method is semiparametric in nature, it is
still able to deliver accurate estimates, i.e. the loss of efficiency is small in comparison to a fully
parametric hazard model (see, for instance, Efron, 1977).
The significance of the explanatory variables in the model can be measured by the likelihood
ratio statistic, given by –2 (log(L0) – log(Ln)), where log(L0) denotes the maximum log-likelihood
value under the restriction β = 0, and log(Ln) denotes the maximum log-likelihood of the unrestricted
15
estimated model. The likelihood ratio statistic is asymptotically χ 2 (k ) distributed, where k denotes
the dimension of β.
4.4
Parametric Estimation of the Hazard Function
The nonparametric approach is especially appropriate for identifying the variables that determine the
hazard rate without making a distributional assumption on the hazard rate. However, the Cox
approach cannot be used to predict survival times. Hence, we also include a parametric analysis in this
paper. We closely follow Hensler et al. (1997), to make our results comparable to theirs.
Looking at the nonparametric estimate of the hazard function (see Figure 3), it appears
reasonable to assume a nonmonotonic function for the hazard, i.e. the hazard rate steadily increases to
a maximum followed by a slowly decreasing rate. The Log-logistic survival model is able to provide
such a non-monotonic hazard. In contrast to the Cox model, which is specified in the proportional
hazard metric, the Log-logistic model is specified in the accelerated time metric. In this latter metric,
the model is of the form
log(ti ) = β ' xi + log(τi ),
(3)
where τi has a particular distribution. For the Log-logistic survival model, τi = exp(−β ' xi )ti is
assumed to be distributed as Log-logistic; see the Appendix for more details.
Just like the Cox model, the Log-logistic model can be estimated by maximum likelihood.
Although both approaches can be used to model hazard rates, they differ in several respects. First, the
Cox model has no intercept since it is unidentifiable from the data. Secondly, the sign of the
coefficients flips, since the hazard rate is scaled by exp(β ' xi ) in the proportional hazard model, while
the time acceleration parameter is proportional to exp(−β ' xi ) .
The corresponding hazard function is equal to
16
刪除: A7
λ(ti ) =
1/ γ−1
δ / γ (δti )
1 + (δti )1/ γ
(4)
,
where δ = exp(−β0 − β ' xi ) . The parameter δ is called the acceleration parameter. If this parameter is
1, then time passes at its "normal" rate, while if δ > 1 , then time passes more quickly for the internet
firm and delisting would be expected to occur sooner. If δ < 1 , then time passes more slowly and
delisting would be expected to occur later.
5.
Empirical Results
In Section 5.1 we will discuss estimates obtained by the semiparametric Cox model. A sensitivity
analysis based on the parametric Log-logistic model is presented in Section 5.2. In Section 5.3 we
report a number of robustness checks.
5.1
刪除: A discussion about the
economic significance of our
findings based on the parametric
Log-logistic model is presented in
Section 5.2. In Section 5.3 we
report a number of additional
analyses
Cox Proportional Hazard Model
Table 6 presents the results for the Cox proportional hazard regression models of trading (survival)
durations of internet IPO firms. The column labeled Hazard Ratio in Table 6 provides the impact on
the hazard rate, given a unit change of the explanatory variable3. As can be seen in this table, the
likelihood ratio (LR) statistic is significant at the one percent level. All coefficients have the predicted
direction, except for two of the three variables indicating firm-specific risk, namely firm size and the
duration of internet activities. However, these two risk variables are not significant. Note that we have
made the size effect dependent on sub-sectors. On closer inspection of the data, it appears that the size
of delisted Internet Service Providers and IT-infrastructure network firms in our sample is
3
For instance, when looking at the Cox regression results, an increase of Firm Risk with one risk factor affects
the hazard rate with a factor 1.025 (or a 2.5 percent increase). Similarly, a one unit increase of Investor Demand
affects the hazard rate with a factor 0.272 (equal to a decrease of 72.8 percent).
17
刪除: since, without this
subdivision, the size effect is
significantly positive
significantly larger than that of other internet IPOs4. Therefore, we have run regressions with two size
variables: one capturing the size of Internet Service Providers and IT-infrastructure IPO firms as a
control variable, the other capturing the size of the remaining internet IPOs. Since the classification
cannot be made unambiguously and due to sample size limitations, we have not included internet
subsections in our model. The third variable for firm-specific risk, the number of risk factors listed in
the prospectus, is significant at the 10 percent level only.
Regarding the financial ratios only one of the three financial ratios, namely operating cash
flow over liabilities, is significant and in line with our hypothesis H10. The other two variables, net
sales over assets and receivables over total assets, are not significant. The outcome supports the notion
刪除: Anecdotal evidence has
learned that it is not possible to
unambiguously classify the subsections of internet firms. In
addition, given
刪除:
刪除:
刪除:
刪除: Because the relation
between the various internet
activities it is not clear
刪除: what the benefits could be
from including
刪除: internet subsections
that a variable from cash accounting can be more useful than one from accrual accounting for
predicting the survivability of an internet IPO; see, for instance, Teoh, Welch and Wong (1998). In
addition, the insignificance of the other two financial ratios is consistent with the findings of Trueman
刪除: various studies (for
instance,
刪除: ,
et al. (2000). Moreover, the different nature of internet firms is documented by many authors (for
instance, Hand, 2003 and Knauff et al., 2003), namely that their market value is a decreasing function
of their earnings.
刪除: (
刪除: In addition
刪除: ;
刪除: ,
刪除: 0
刪除: ;
<Insert Table 6 about here >
刪除: ,
刪除: that the internet firms’
financial ratios prior to the IPO
are of limited significance to firm
survival
刪除: i
刪除: ii
刪除: iii
刪除: iv
刪除: v
刪除: vi
刪除: vii
4
All IPOs were classified into one of the following ten sectors: (I) Internet Service Providers (II) Content /
portals (III) E-commerce – products (IV) E-commerce – services (V) IT-infrastructure – network (VI) ITinfrastructure – equipment (VII) Internet software – licensed (VIII) Internet software – server (IX) Professional
internet services (X) Internet advertising services.
18
刪除: viii
刪除: ix
刪除: x
5.2
Parametric Log-logistic Survival Model
After the semiparametric analysis, we now discuss the results based on the parametric Log-logistic
model using the same independent variables. The results are qualitatively similar to those of the Cox
proportional hazard model; see Table 6. Again, all coefficients but those for firm size and duration of
internet activities have the predicted sign (see the remarks at the end of Section 4.4 for an explanation
for the flipping of the signs when turning from the Cox to the Log-logistic model). Except for firm
risk, the significance of all other variables that were significant in the Cox model has increased or
remained the same. This is consistent with the fact that the coefficients will be estimated more
precisely if more information is taken into account in the estimation method, i.e. the parametric Loglogistic specification leads to lower standard errors. The three variables for firm-specific risk are not
significant.
As aforementioned, the Cox proportional hazard model does not provide information about
the extent to which the survival time of an internet IPO is affected by a specific explanatory variable.
Table 7 shows the effects on the survival time when a significant explanatory variable is changed one
and two standard deviations (σˆ ) from its mean while holding the remaining variables constant.
Overall, the effect on the expected survival time is slightly asymmetric, i.e. a positive shock to an
刪除: n
explanatory variable has a different effect than a negative shock. As can be seen in Table 7, the
greatest positive effect on the survival time comes from investor demand followed by operational cash
flow over liabilities: an increase in investor demand by one σ̂ leads to an increase in the expected
survival time of 22.9 months, while an increase in operational cash flow over liabilities of one σ̂
lengthens the survival time by 11.1 months. The expected survival time is shortened mostly by IPO
market level ( +σˆ leads to a decrease of 7.8 months), followed by valuation uncertainty ( +σˆ leads to
a decrease of 5.3 months).
<Insert Table 7 about here >
19
刪除:
刪除: t
刪除: is different from
Figure 4 shows the estimated survival function based on the Log-logistic analysis for IPOs
with low versus high prestigious underwriters. For IPOs with low prestigious underwriters, the
quartiles of survival times (25, 50 and 75 percentiles) are 23, 36 and 55 months, while for IPOs with
high prestigious underwriters, these quartiles are 33, 51 and 79 months, respectively. Further analysis
of the estimated survival function shows that, on average (also with respect to the variable underwriter
reputation), 25 percent of the internet IPOs is expected to be delisted within 32 months, 50 percent
within 49 months, and 75 percent within 76 months. Note that these quartiles are substantially lower
than the numbers reported by Hensler et al. (1997), namely 59, 104, and 184 months. Of course, this
is due to the fact that we study internet IPOs in particular, whereas Hensler et al. (1997) study IPOs in
general.
<Insert Figure 4 about here >
5.3
Robustness
After controlling for outliers, i.e. removing the largest and lowest five observations with respect to a
particular variable, the mean differences from Table 5 are qualitatively similar. Furthermore, in order
to check the robustness of our hazard models we have run the regressions presented in Table 6 using
the significant explanatory variables only. Again, the results of the regression were qualitatively
刪除: of
similar to the results shown in Table 6, which indicates that multicollinearity is not a problem.
In addition to the analyses aforementioned, we have run a number of Cox regressions using
alternative explanatory variables (not reported). With respect to underwriter reputation, we varied the
reputation indicators to be included in our dummy variable. Also, we have run regressions using a
number of alternative variables for firm size, such as the log of total offer value and the log of the
number of employees. As an alternative for insider ownership retention, we have modeled
management ownership retention. Finally, in addition to market level at the time of offering, we have
20
刪除: .
controlled for the year and quarter of the year of the offering by including dummies. All those
alternative model specifications provided qualitatively similar results.
刪除:
6.
Conclusions
刪除: , which
This paper examines whether the variables that have been documented as significant in
studies on non-internet IPOs play a similar role for internet IPOs. To this end, we analyze the
determinants of survival of internet firms that have gone public at the NASDAQ stock exchange from
December 1996 through February 2001. Financial and non-financial data published in the IPO
prospectuses are examined.
The contribution to the existing literature of this paper is threefold. First, encompassing the
years 1996-2005, the paper focuses on the analysis of survival determinants for internet IPOs only.
Secondly, our model incorporates a number of explanatory variables, which are analyzed
simultaneously instead of separately. By taking the correlations between the variables into account,
our analysis is safeguarded against spurious relationships. Thirdly, our research design is specifically
chosen to answer the question which explanatory variables predict the rate of success and failure of
firms in the internet industry. In addition, the effect of the independent variables on the survival time
is estimated.
The descriptive statistics reveal that the average operating history of an internet IPO is
remarkably small, namely 2.4 years, compared to non-internet IPOs, which have an average age of
slightly more than 10 (Ritter, 1991) and 8.6 years (Hensler et al., 1997). Given the short operating
history, the firms in the sample are far from being in a steady state. Furthermore, the average number
of risk factors (32) is four times higher than the number reported by Hensler et al. (1997). Finally,
with respect to firm size and operating history, the sample of our internet IPOs is much more
homogeneous than that of non-internet IPOs studied by Hensler et al. (1997).
21
刪除: The model examined in this
paper enlightens a number of
different aspects with regard to an
internet IPO. These aspects refer
to market conditions (investor
demand and IPO market level),
managerial decisions (retained
inside ownership), a firm specific
characteristic that can be seen as
relatively objective from an
accounting point of view
(operating cash flow over
liabilities), and the quality of the
intermediary used (underwriter
reputation). Our results show that
all these aspects play a significant
role in determining the
survivability of internet IPOs.
This paper examines whether the
variables, which are significant in
non-internet IPOs play a similar
role for internet IPOs.
Our regression results show that the IPO size and the duration of internet activities are not
significant determinants of the survivability of internet IPOs. In addition, the number of risk factors
listed in the prospectus is not or only marginally significant depending on the estimation model. Since
刪除: s
these variables measure firm-specific risk, it appears that during the internet bubble this category of
risk played a minor role in contrast with studies of non-internet IPOs. Furthermore, the econometric
analysis reveals that six explanatory variables are significant in our model. The significant variables in
刪除: ,
刪除: firm-specific
刪除: ‘ordinary’
刪除: internet IPOs
刪除: e
both the semiparametric Cox proportional hazard model and the parametric Log-logistic survival
刪除:
model are the same except for firm risk, which is only significant at a 10 percent level in the Cox
model. Hence, the restrictions imposed by the parametric Log-logistic model do not appear to
influence the statistical conclusions. Our findings show that surviving firms can be distinguished in a
number of ways from non-surviving firms. The survival time of an internet IPO is an increasing
function of underwriter reputation, investor demand, inside ownership retention and operational cash
flow over liabilities. The survival time is found to be a decreasing function of valuation uncertainty
and IPO market level. Our findings hold under a number of different model specifications.
Furthermore, our empirical results are qualitatively similar when outliers are removed.
The sensitivity analysis in the Log-logistic model reveals that the greatest positive effect on
the survival time comes from investor demand followed by operational cash flow over liabilities. The
expected survival time is shortened mostly by IPO market level, followed by valuation uncertainty.
Furthermore, the survival times based on the estimated Log-logistic survival curve for the internet
IPOs are about half as long as for those examined by Hensler et al. (1997): for internet IPOs 50
percent are delisted within 49 months versus 104 months for non-internet IPOs.
Our empirical results provide evidence for the important roles of high prestigious
underwriters and retained ownership. The significant and positive effect of both variables on the
刪除: effects of both variables on
the survival rate of an internet
IPO is
survival rate of an internet IPO is consistent with the outcome from studies on non-internet IPOs. A
刪除: moment
‘hot’ IPO market is not a good time for investing: internet shares bought during a stock market level
刪除: ex
peak have a lower survival rate. Finally, whereas none of the accrual accounting variables considered
is significant, the financial ratio based on cash accounting is.
22
刪除: amined
刪除: T
Overall, the model examined in this paper enlightens a number of different aspects with
regard to an internet IPO. These aspects refer to market conditions (investor demand and IPO market
level), managerial decisions (retained inside ownership), a firm specific characteristic that can be seen
as relatively objective from an accounting point of view (operating cash flow over liabilities), and the
quality of the intermediary used (underwriter reputation). Our results show that all these aspects play
a significant role in determining the survivability of internet IPOs.
23
刪除: In sum
Appendix: Econometric Details
Nonparametric Estimation of the Hazard Function
Let t1, t2, t3, ..., tn denote either the completed or censored durations of the n firms in our sample.
Suppose we order the completed durations from smallest to largest, t(1) < t(2) < ... < t(K), and let hj be
the number of events after t(j) months, while mj is the number of firms censored between months t(j)
and t(j+1). If n j =
∑ i ≥t
t( K )
( j)
(mi + hi ), the nonparametric Kaplan-Meier estimator for the firms’ hazard
rate per month is given by
λˆ (t ) =
hj
∆t j n j
,
(A1)
for t(j) ≤ t < t(j+1), where ∆t j = t( j +1) − t( j ) . Equation (A1) provides an unconditional estimate of the
刪除:
firms’ hazard rate at time t(j), which is shown in Figure 3.
Semiparametric Estimation of the Hazard Function
When accounting for censoring, it can be shown that the log-likelihood function for Cox proportional
hazard model shown in equation (2) is given by (for instance, see Kalbfleisch and Prentice, 1980)
⎧
⎫
n
⎪
⎪
log L(β) = ∑ Ii ⎨β ' xi − log ∑ exp (β ' xl ) ⎬ ,
i =1 ⎩
l∈R (t( j ) )
⎪
⎭⎪
(A2)
where Ιi is the censoring indicator, which has a value of 0 for censored durations and 1 for completed
durations, and R(t( j ) ) is the set of all firms that have remained listed and are uncensored just prior to
24
t(j). To account for ties5, we have used the Breslow (1974) approximation. Maximum likelihood
estimates of β can be obtained through numerical methods.
Parametric Estimation of the Hazard Function
For the Log-logistic survival model, τi = exp(−β ' xi )ti is assumed to be distributed as Log-logistic
with parameters (β0 , γ ) , i.e. the cumulative distribution function is given by
−1
F (τi ) = 1 − ⎡1 + {exp( −β0 )τi }1/ γ ⎤ .
⎣
⎦
(A3)
In the Log-logistic model, it can be shown that
log(ti ) = β0 + β ' xi + εi ,
(A4)
where εi follows a logistic distribution with mean 0 and standard deviation πγ / 3 , so that
E[log(ti ) | xi ] = β0 + β ' xi . The survivor function is given by
−1
S (ti | xi ) = ⎡1 + {exp( −β0 − β ' xi )ti }1/ γ ⎤ .
⎣
⎦
5
(A5)
A tie is the occurrence of more than one event, in this case a delisting or acquisition, within a single time
interval t.
25
References
Aggarwal, R. (2000) “Stabilization activities by underwriters after initial public offerings”, Journal of Finance
55, 1075-1103.
Bartov, E., P. Mohanram and C. Seethamraju (2002) “Valuation of internet stocks – An IPO perspective”,
Journal of Accounting Research 40, 321-346.
Beatty, R. and I. Welch (1996) “Legal liability and issuer expenses in initial public offerings”, Journal of Law
and Economics 39, 545-603.
Bhabra, H.S. and R.H. Pettway (2003) “IPO prospectus information and subsequent performance”, The
Financial Review 38, 369-397.
Breslow, N.E. (1974) “Covariance analysis of censored survival data”, Biometrics 30, 89-100.
刪除: 2
刪除: unpublished Concordia
University working paper
Carter, R.B. and S. Manaster (1990) “Initial public offerings and underwriter reputation”, Journal of Finance 45,
1045-1067.
Carter, R.B., F.H. Dark and A.K. Singh (1998) “Underwriter reputation, initial returns, and the long-run
performance of IPO stocks”, Journal of Finance 53, 285-311.
Cassia, L, G. Giudici, S. Paleari and R. Redondi (2004) "IPO underpricing in Italy", Applied Financial
Economics 14, 179-194.
Cox, D.R. (1972) “Regression models and life Tables” (with discussion), Journal of the Royal Statistical Society
34, 187-220.
Demers, E.A., and P.P.M. Joos (2006) “IPO Failure Risk”, INSEAD and Tilburg University, working paper.
Demers, E.A., and B. Lev (2001) “A rude awakening: Internet Shakeout in 2000”, Review of Accounting Studies
6, 331-359.
Demers, E.A. and K. Lewellen (2003) “The marketing role of IPOs”, Journal of Financial Economics 68, 413437.
Dimovski, W. and R. Brooks (2003) “Financial characteristics of Australian initial public offerings from 1994 to
1999”, Applied Economics 35, 1599-1607.
Efron, B. (1977), “The Efficiency of Cox's Likelihood Function for Censored Data”, Journal of the American
Statistical Association 72, 557-565.
Grilli, L. (2005), “Internet start-ups access to the bank loan market: evidence from Italy”, Applied Economics 37,
293-305.
Hand, J.R.M. (2003) “Profits, losses and the non-linear pricing of internet stocks”, in Intangible Assets: Values,
Measures and Risks, eds. J.R.M. Hand and B. Lev (Oxford University Press).
Hanley, K.W. (1993) “The underpricing of initial public offerings and the partial adjustment phenomenon”,
Journal of Financial Economics 34, 231-250.
刪除: Goel, R.K., E.T. Hsieh,
M.A. Nelson, and R.Ram (2006),
“Demand elasticities for Internet
services”, Applied Economics 38,
975-980.
刪除: 0
刪除: ,
刪除: unpublished
Hansen, R.S. and P. Torregrosa (1992) “Underwriter compensation and corporate monitoring”, Journal of
Finance 47, 1537-1555.
Hensler, D.A., R.C. Rutherford and T.M. Springer (1997) “The survival of initial public offerings in the
aftermarket”, Journal of Financial Research 20, 93-110.
26
刪除: University of North
Carolina, working paper
Jain, B.A. and O. Kini (1994) “The post-issue operating performance of IPO firms”, Journal of Finance 49,
1699-1726.
Jain, B.A. and O. Kini (1999a) “On investment banker monitoring in the new issues market”, Journal of
Banking and Finance 23, 49-84.
Jain, B.A. and O. Kini (1999b) “The life cycle of initial public offering firms”, Journal of Business Finance &
Accounting 26, 1281-1307.
Jegadeesh, N.M., M. Weinstein and I. Welch (1993) “Initial public offerings and subsequent equity offerings”,
Journal of Financial Economics 34, 153-175.
Jensen, M.C. and W.H. Meckling (1976) “Theory of the firm: Managerial behavior, agency costs and ownership
structure”, Journal of Financial Economics 3, 305-360.
Johnston, J. and J. Madura (2002) “The performance of internet firms following their initial public offering”,
The Financial Review 37, 525-550.
Kalbfleisch, J.D. and R.L. Prentice (1980) “The statistical analysis of failure time data”, John Wiley & Sons,
New York.
Kaplan, E.L. and P. Meier (1958), “Nonparametric estimation from incomplete observations”, Journal of the
American Statistical Association 53, 457-481.
Kiefer, N.M. (1988) “Economic duration data and hazard functions”, Journal of Economic Literature 26, 646679.
Kim, M. and J.R. Ritter (1999), “Valuing new issues”, Journal of Financial Economics 53, 409-437.
Knauff, P., P. Roosenboom and T. van der Goot (2003) “Is accounting information relevant to valuing European
internet IPOs?”, in New Venture Investment: Choices and Consequences, Elsevier Publishers, ISBN: 0444-51239-X.
刪除: , 1-301
Kogan, L. (2004) “Asset pricing and real investment”, Journal of Financial Economics 73, 411-431.
Lancaster, T. (1990) “The econometric analysis of transition data”, Econometric Society Monographs No. 17,
Cambridge University Press, Cambridge.
Leland H.E. and D.H. Pyle (1977) “Informational asymmetries, financial structure, and financial
intermediation”, Journal of Finance 32, 371-387.
Loughran, T. and J.R. Ritter (2002) “Why don’t issuers get upset about leaving money on the Table in IPOs?”,
Review of Financial Studies 15, 413-443.
Loughran, T. and J.R. Ritter (2004) “Why has IPO underpricing changed over time?”, Financial management
33, 5-37.
刪除: 3
Lowry, M. and G.W. Schwert (2002) “IPO market cycles: Bubbles or sequential learning?”, Journal of Finance
57, 1171-1200.
刪除: unpublished University of
Notre Dame working paper
Ljungqvist, A. and W. Wilhelm (2003) “IPO pricing in the dot-com bubble”, Journal of Finance 58, 723-752.
Macklin, G.S., A.M. Adlerman and K.Y. Hao (1992) “Going public and the NASDAQ market”, in The
NASDAQ Handbook, Probus Publishing Company, Chicago.
Ritter, J.R. (1991) “The long-run performance of initial public offerings”, Journal of Finance 46, 3-28.
Schultz, P. (1993) “Unit initial public offerings: A form of staged financing”, Journal of Financial Economics
34, 199-229.
27
刪除: increased
Teoh, S.H., I. Welch and T.J. Wong, 1998, “Earnings Management and the Long-Run Market Performance of
Initial Public Offerings”, Journal of Finance 53, 1935-1974.
Trueman, B., M.H.F. Wong and X. Zhang (2000) “The eyeballs have it: Searching for the value in internet
stocks”, Journal of Accounting Research 38, 137-162.
Van der Goot, T. (2003) “Risk, the quality of intermediaries and legal liability in the Netherlands IPO market”,
International Review of Law and Economics 23, 121-140.
28
Distribution over time of internet IPOs by month.
Figure 1
30
Number of firms
25
20
15
10
5
0
Jul 1996
Jan 1997
Jul 1997
Jan 1998
Jul 1998
Jan 1999
M onth of IPO
29
Jul 1999
Jan 2000
Jul 2000
Jan 2001
Distribution over time of delisted internet firms by quarter.
Figure 2
25
Num ber of firm s
20
15
10
5
0
Q1
1999
Q3
1999
Q1
2000
Q3
2000
Q1
2001
Q3
2001
Q1
2002
Q3
2002
Quarte r of de listing
30
Q1
2003
Q3
2003
Q1
2004
Q3
2004
Q1
2005
Smoothed hazard function using Kaplan-Meier estimate. The smoothed hazard
rate has its maximum at about 2.5 years (31 months).
Figure 3
The estimated nonparametric hazard curve (the thin line) is highly volatile due to the limited number of
completed durations (i.e. delistings) in our sample. Therefore, the curve has been smoothed using a
Epanechnikov kernel with a data-driven bandwidth (the thick line).
Monthly Hazard Rate
5.0%
4.0%
3.0%
2.0%
1.0%
0.0%
0
10
20
30
40
Time (months)
31
50
60
70
Estimated survival curves for high (black line) and low (gray line)
underwriter reputation.
Figure 4
As can be seen in this figure, 25 percent IPOs with high prestigious underwriters are expected
to be delisted within 33 months; 50 percent within 51 months; and 75 percent within 79
months. For IPOs with low prestigious underwriters, the quartiles change to 23, 36 and 55
months.
1
Survival Probability
0.75
0.5
0.25
0
0
50
100
Time in Months
32
150
200
Table 1
Causes of delisting.
Cause of Delisting
Number of Firms
Minimum bid price
Bankruptcy
Net tangible assets
Reporting requirements
Fraud (leading to bankruptcy)
Equity requirements
Liquidation
Minimum market value
Minimum public float
89
16
5
4
2
2
2
1
1
73.0%
13.1%
4.1%
3.3%
1.6%
1.6%
1.6%
0.8%
0.8%
Total
122
100.0%
33
Table 2
Description of the independent variables.
Independent variables
Description
Offering and Firm Characteristics
Firm risk
Number of risk factors listed in prospectus
Underwriter reputation
Binary: 1 if underwriter reputation is 8 or 9; 0 otherwise
Firm size
Log of gross dollar offering proceeds at the offering price
Investor demand
(Offer price - average value price range) / average value price range
Valuation uncertainty
(Price range spread / average value of price range) x 100
Insider ownership retention
Percentage of firm ownership retained by insiders after the offering
Age of internet activities
Number of months from start of internet activities to its IPO
IPO market level
Average NASDAQ Composite Index in offering month
Offer-to-book ratio
Firm value at offer price / Book value of equity
Financial Ratios
Net sales over assets
Net sales / total assets
Operating cash flow over liabilities
Operating cash flow / total liabilities
Receivables over total assets
Accounts receivable / total assets
Transformations
Offer-to-book ratio (log)*
= log(OBR) if OBR≥ 0, –log(–OBR) if OBR < 0
Net sales over assets (log)
= log(1+NSOA)
Operating cash flow over liabilities (log)
= –log(1–OCFOL)
Receivables over total assets (log)
= log(1+ROTA)
*
Since no observations for the untransformed offer-to-book ratio is contained in [–1,1], the
transformation is 1-to-1, i.e. monotonic increasing.
34
Table 3
Summary statistics of the explanatory variables.
Independent variables (N = 211)
Mean
Std. Dev. Median Minimum Maximum
Offering and Firm Characteristics
Firm risk
Underwriter reputation*
Firm size
Investor demand
Valuation uncertainty
Insider ownership retention
Age of internet activities
IPO market level (in thousands)
Offer-to-book ratio (log)
32.024
0.891
18.092
0.281
17.410
0.813
28.804
3.199
2.273
6.483
32.000
11.000
49.000
0.633
0.392
3.682
0.072
15.619
0.864
3.747
18.068
0.250
18.182
0.823
27.880
2.808
3.465
16.524
-0.493
0.000
0.538
0.329
1.225
–9.016
21.372
2.000
35.294
0.946
115.134
4.803
11.990
0.474
–0.628
0.149
0.401
0.573
0.130
0.421
–0.612
0.107
0.000
–2.013
0.000
1.986
0.824
0.531
Financial Ratios
Net sales over assets (log)
Operating cash flow over liabilities (log)
Receivables over total assets (log)
* Because Underwriter reputation is a binary variable we present the value of its mean only.
35
Underwriter reputation
0.173
Firm size
0.197 0.271
Investor demand
0.089 0.109 0.404
Valuation uncertainty
0.020 0.014 -0.214 0.043
Insider ownership retention
Age of internet activities
Oper. cash flow / liabilities (log)
Net sales over assets (log)
Offer-to-book ratio (log)
IPO market level (in thousands)
Age of internet activities
Insider ownership retention
Valuation uncertainty
Investor demand
Firm size
Underwriter reputation
Pearson correlation matrix of the explanatory variables.
Firm risk
Table 4
0.130 0.435 0.302 0.213 -0.099
-0.145 -0.072 -0.050 -0.086 -0.073 -0.134
IPO market level
0.143 0.074 0.244 0.106 0.022 0.055 0.029
Offer-to-book ratio (log)
0.109 0.137 0.103 0.142 0.036 0.252 -0.056 0.084
Net sales over assets (log)
-0.360 -0.235 -0.233 -0.073 -0.062 -0.108 0.269 -0.115 -0.179
Oper. cash flow / liabilities (log)
-0.331 -0.105 -0.002 -0.075 -0.127 -0.020 0.230 -0.007 -0.173 0.466
Receivables over total assets (log)
-0.307 -0.114 -0.164 0.063 -0.049 0.019 0.136 0.040 -0.079 0.705 0.368
36
Table 5
Mean differences of the explanatory variables.
Non-survivors
Survivors
(89 firms)
Independent variables
(122 firms)
Mean
Mean
Survivors vs.
Non-survivors
T-test of
difference
Offering and Firm Characteristics
Firm risk
Underwriter reputation
Firm size
Investor demand
Valuation uncertainty
Insider ownership retention
Age of internet activities
31.517
32.393
0.966
0.836
18.139
18.058
0.397
0.197
17.006
17.704
0.833
0.799
(–0.970)
(3.359) ***
(0.919)
(3.762) ***
(–1.362)
(3.452) ***
28.247
29.210
(–0.441)
IPO market level (in thousands)
3.184
3.210
(–0.212)
Offer-to-book ratio (log)
2.668
1.985
(1.309)
0.507
0.450
(1.026)
–0.528
–0.701
0.169
0.134
Financial Ratios
Net sales over assets (log)
Operating cash flow over liabilities (log)
Receivables over total assets (log)
(2.182) **
(1.962) *
The t-statistic in parentheses is calculated as the ratio of the coefficient estimate to its standard error; all t-tests are two-tailed.
* Significant at the 10% level, ** Significant at the 5% level, *** Significant at the 1% level.
37
Table 6
Regression results of Cox hazard model and Log-logistic survival model on the
duration of IPOs.
Nonparametric Cox
Hazard Model
Independent variables
Coefficient
Estimate
TStatistic
0.025 *
-0.657 **
0.107
(1.612)
(-2.101)
(0.607)
刪除:
Parametric Log-Logistic
Model
Hazard
Ratio
Coefficient
Estimate
TStatistic
1.025
-0.011
(-1.219)
0.518
1.113
0.359 **
-0.080
(-0.745)
(-0.557)
Offering and Firm Characteristics
Firm risk
Underwriter reputation
Firm size (ISPs+IT-infrastr. network)
(0.361)
1.068
-0.062
(-4.430)
0.272
Valuation uncertainty
-1.300 ***
0.047 **
(1.718)
1.048
0.776 ***
-0.024 **
(-1.675)
Insider ownership retention
Age of internet activities
-1.964 *
0.001
(-1.339)
0.140
(0.157)
1.001
1.409 **
-0.001
(-0.257)
Firm size (other)
Investor demand
IPO market level (in thousands)
Offer-to-book ratio (log)
0.066
(2.110)
0.172 *
(4.739)
(1.711)
(1.435)
1.188
(-0.873)
0.978
-0.149 ***
0.018
(-2.361)
-0.022
0.088
(0.218)
1.092
-0.012
(-0.060)
(2.827)
0.564
(-0.129)
0.860
(1.248)
Financial Ratios
Net sales over assets (log)
Operating cash flow over liabilities (log)
Receivables over total assets (log)
-0.572 ***
-0.151
Constant
59.17 ***
Likelihood Ratio Statistic
Number of Firms
211
0.279 ***
0.254
(2.598)
(0.434)
4.758 **
(2.394)
68.63 ***
211
The t-statistic in parentheses is calculated as the ratio of the coefficient estimate to its standard error. Since most of the hypotheses H1H10 are one-sided, the significance of all t-tests related to one-sided hypotheses is based on one-sided p-values. The p-values for
hypothesis H3 and H7, and the constant are two-sided.
* Significant at the 10% level, ** Significant at the 5% level, *** Significant at the 1% level.
38
刪除: L
Table 7
Sensitivity analysis of the significant explanatory variables for the Log-logistic
survival model.
−2σˆ
Underwriter reputation (=0, left / =1, right)
Expected survival time (# of months)
Absolute change
Relative change
Investor demand
Expected survival time (# of months)
Absolute change
Relative change
Valuation uncertainty
Expected survival time (# of months)
Absolute change
Relative change
Insider ownership retention
Expected survival time (# of months)
Absolute change
Relative change
IPO market level (in thousands)
Expected survival time (# of months)
Absolute change
Relative change
Operating cash flow over liabilities (log)
Expected survival time (# of months)
Absolute change
Relative change
−σˆ
+σˆ
46.8
-17.6
-27.4%
+2σˆ
67.0
2.6
4.0%
35.0
-29.4
-45.6%
47.5
-16.9
-26.2%
87.3
22.9
35.6%
118.4
54.0
83.8%
76.6
12.2
18.9%
70.2
5.8
9.1%
59.0
-5.3
-8.3%
54.1
-10.2
-15.9%
52.5
-11.8
-18.4%
58.2
-6.2
-9.7%
71.3
6.9
10.7%
78.9
14.5
22.5%
83.3
18.9
29.4%
73.2
8.9
13.8%
56.6
-7.8
-12.1%
49.8
-14.6
-22.7%
46.8
-17.6
-27.3%
54.9
-9.5
-14.8%
75.5
11.1
17.3%
88.6
24.2
37.6%
刪除: L
When all independent variables are set to their mean values, the expected survival time for an internet IPO based
on the estimated Log-logistic model is 64.4 months.
"Expected survival time" = denotes the expected firm's survival time in months when an independent variable is
changed 1 and 2 estimated standard deviations (σˆ ) from its mean while holding the
remaining variables constant.
"Absolute change"
= "Expected survival time" – 64.4.
"Relative change"
= "Absolute change" / 64.4.
39
第 3 頁: [1] 刪除
nvg
2006/7/20 12:00:00 PM
examine whether new variables, which are characteristic for an internet IPO and those
used in
第 3 頁: [2] 刪除
nvg
2006/7/20 12:00:00 PM
are also important with respect to internet IPOs
第 4 頁: [3] 刪除
nvg
2006/7/19 12:00:00 PM
Our empirical results provide evidence for the significant effects of high prestigious
underwriters and retained ownership on the survival rate of an internet IPO. This is consistent
with the outcome from studies on non-internet IPOs. Also, our findings show that internet
shares bought during a stock market level peak have a lower survival rate. Finally, none of the
accrual accounting variables examined is significant.
第 4 頁: [4] 刪除
LRTnederlands
Section three discusses our hypotheses on internet firm survival.
2006/7/15 10:36:00 AM