Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
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