summary

ARTICLE IN PRESS

0304-405X/$ - s

doi:10.1016/j.jfi

$ We wish to

Schwert (the e

University, Un

Virginia (Darde

(2003) and esp

Council for fin � Correspond

E-mail addr

Journal of Financial Economics 87 (2008) 610–635

www.elsevier.com/locate/jfec

Second time lucky? Withdrawn IPOs that return to the market $

Craig G. Dunbar � , Stephen R. Foerster

Richard Ivey School of Business, University of Western Ontario, London, Ontario, Canada, N6A 3K7

Received 3 March 2006; received in revised form 7 July 2006; accepted 11 August 2006

Available online 19 August 2007

Abstract

We investigate issuers withdrawing an IPO (after security regulation filings) that return later for a successful offering.

Venture capital backing and reputation of the lead underwriter are key factors in predicting successful return. The

possibility of returning has a significant impact on the decision to withdraw and the pricing of offerings that succeed. Our

sample of returning IPOs also provides a unique setting to investigate underwriter switching after a withdrawal but before a

successful IPO. We find that switching occurs in response to poor bank performance and when switching firms ‘‘graduate’’

to banks that have high industry market shares.

r 2007 Elsevier B.V. All rights reserved.

JEL classification: G14; G24; G32

Keywords: IPOs; Withdrawals; Return performance; Investment bank reputation; Switching

1. Introduction

The most significant event in the life of a corporation is arguably its transition from a private to public company through the initial public offering (IPO) process. The IPO provides a major source of capital and allows the existing owners to have a liquid market for their shares. Firms rely on the IPO for either their survival or their ability to take advantage of growth opportunities.

Yet not all firms are successful in making the transition from a private to public company. In fact, after an IPO process has been initiated with the support of an investment bank, a surprisingly large number of proposed IPOs are withdrawn from the market before being completed. An emerging literature examines the prevalence of proposed IPOs that are registered but withdrawn before issue. For example, Dunbar (1998) and Busaba, Benveniste, and Guo (2001, BBG hereafter) show that between the mid-1980s and mid-1990s almost

ee front matter r 2007 Elsevier B.V. All rights reserved.

neco.2006.08.007

thank Mark Huson, Kathy Kahle, Kai Li, Michelle Lowry (the referee), Greg Nachtwey, Gordon Roberts, Tim Simin, Bill

ditor), Lee Ann Woo, Chad Zutter, two anonymous referees, seminar participants at the Bank of Canada, Queens

iversity of Arkansas, University of British Columbia, University of Hawaii, the University of Pittsburgh, University of

n), York University, the Northern Finance Association Meetings (2002), the Financial Management Association Meetings

ecially Colette Southam for excellent research assistance. We also thank the Social Sciences and Humanities Research

ancial support.

ing author.

ess: [email protected] (C.G. Dunbar).

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 611

one in five IPOs was withdrawn. Evidence from more recent periods, as uncovered in this paper, suggests that this fraction has increased to over one in two in some years. Several studies have attempted to explain the choice to withdraw an IPO. BBG argue that the decision to withdraw an IPO should depend on the issuer’s reservation value for the offering relative to possible investor valuations. Welch (1992) argues that negative information ‘‘cascades’’ can result in investor valuations falling below a level deemed reasonable by issuers, resulting in withdrawal. Dunbar (1998), however, finds that issuers withdrawing IPOs are unlikely to return for a successful public equity offering. In a related literature, Mikkelson and Partch (1988) and Clarke, Dunbar, and Kahle (2001) examine withdrawn seasoned equity offerings. If withdrawals are in response to temporary market misvaluations, it is surprising that so few firms return. The choice to withdraw, therefore, remains puzzling since it can significantly restrict a firm’s access to the liquid and relatively inexpensive public capital markets.

To gain insights into the choice of withdrawal, we examine a sample of firms that withdraw an IPO but are able to return eventually to the public equity markets for a successful IPO. We first attempt to identify the factors that affect a withdrawn issuer’s likelihood of being able to return successfully for an IPO. We find that firms initially brought forward by more reputable investment banks, and those having venture capital backing, are more likely to return. Issues withdrawn in more active IPO markets, when interest rates are high and when market returns are low, are also more likely to be able to return.

Since the likelihood of returning is predictable, we next examine whether this likelihood affects the firm’s choice to withdraw. We find that the probability of withdrawal is positively related to the likelihood of successful return. Issuers that face the choice to withdraw but do not expect a second chance are more likely to try to push forward and complete their IPO. The likelihood of withdrawal and the possibility of return should also impact the pricing of successful IPOs. In order to ensure success, firms expected to withdraw with a low chance of returning should be more likely to cut prices during the bookbuilding process. Controlling for commonly used variables in the literature, we find that price adjustments are more negative for these firms, leading to higher first-day returns.

Overall, the evidence indicates that firms consider the costs of withdrawal when attempting to decide whether or not to proceed with an IPO. Our study makes a number of additional contributions to the literature on IPO withdrawals. First, we extend the analysis in Dunbar (1998), BBG, and Benveniste, Ljungqvist, Wilhelm, and Yu (2003) on the determinants of offering withdrawal. Dunbar examines 3,540 withdrawn and successful IPOs from 1984 to 1993 and relates the choice to withdraw to a short list of four observable variables. BBG consider a larger number of variables obtained directly from IPO prospectuses but only study 536 IPO filings from 1990 to 1992. Benveniste, Ljungqvist, Wilhelm, and Yu look at a longer time period (1985–2000) but focus on a number of market measures proxying for ‘‘information spillovers.’’ Like Benveniste Ljungqvist, Wilhelm, and Yu, we look at a longer time period but also examine a wider range of market and firm-specific variables, including prospectus-level information from the SEC’s Edgar system. Some new variables emerge as very important in explaining IPO withdrawals. The most significant variable in our probit model, economically and statistically, is the industry market share of the investment bank in the IPO. Issuers brought forward by banks having a significant presence in the industry of the issuer are more likely to be successful. Other significant new explanatory variables include corporate bond yield spreads and the industry average book-to-market ratio. As yield spreads increase and access to borrowing becomes more difficult, firms are less likely to withdraw. Firms with lower book-to-market ratios, or overvalued firms, are also more likely to withdraw.

An interesting feature of the sample of returning IPOs is that in approximately 75% of the cases, the investment bank leading the successful IPO is different than the bank used in the initial unsuccessful attempt. Withdrawn IPOs that subsequently return to the market, therefore, provide a unique setting to explore underwriter switching compared with the existing literature examining switching after an IPO. Firms could switch because of dissatisfaction with the investment bank’s efforts in the original failed IPO process (the performance hypothesis) or because they can now obtain the services of a more reputable underwriter (the graduation hypothesis). Conversely, firms could choose not to switch after an unsuccessful IPO if they have confidence in the underwriter and view the previous failed IPO as related to other external and uncontrollable factors such as an unfriendly market environment. We find evidence supportive of both the graduation and performance explanations.

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635612

The remainder of the paper is organized as follows. In Section 2 we describe the data used in our analysis. We develop hypotheses and present evidence on the factors affecting the choice to withdraw an IPO in Section 3. Evidence on factors affecting the successful return to the IPO market after withdrawal is presented in Section 4. We examine the underwriter switching choice for withdrawn IPOs that return to the market in Section 5. The effect of underwriter switching and the possibility of returning on the choice to withdraw is examined in Section 6. The effect of the possibility of withdrawal on the pricing of successful IPOs is examined in Section 7. Finally, we present conclusions in Section 8.

2. Data

Our study examines all US firms that file documents to raise capital through a firm commitment initial public offering of equity between 1985 and 2000 (Ritter, 1987; Cho, 1992; Dunbar, 1998; also examine withdrawals but within the context of best�efforts offering methods). Our primary data source is Thomson Financial Securities Data’s (TFSD) New Issues Database. We begin our analysis in 1985, as TFSD’s coverage of withdrawn IPOs begins in 1984 but is complete only beginning in January 1985. We consider all IPOs filed over that period but, following the existing literature (e.g., BBG), we screen offerings on a number of criteria. Specifically, we exclude unit offerings (combinations of equity and warrants), REITs, ADRs, and closed-end mutual funds, although unlike BBG, we do not screen out firms in certain industries such as financials or service firms. For each offering, we gather data from TFSD on firm characteristics (e.g., data from past financial statements) and offering characteristics including offering size, price, and information to estimate investment bank reputation variables. Data on market returns around the proposed offerings are collected from the Center for Research in Security Prices (CRSP) database. Data on market interest rates around the proposed offerings are obtained from the Federal Reserve Bank of St. Louis web site (http:// research.stlouisfed.org/fred2/). For many withdrawn offerings, TFSD data are incomplete. Additional information on venture capital backing is obtained from VentureXpert. We also obtain initial prospectuses from the SEC’s Edgar system for all withdrawn IPOs starting in 1996 (electronic filing only began in the mid-1990s). Offering characteristics (proposed price and size) and past financial information are then obtained for these offerings.

TFSD data allow us to identify all successful IPOs as well as all withdrawn IPOs. It is somewhat more challenging to identify which successful IPOs were previously withdrawn and then returned to the market. We use a number of approaches to identify these returners. The first step in identifying matches is to examine CUSIPs as well as unique company identifier numbers assigned to issuers by TFSD. Company identifier numbers and CUSIPs from TFSD’s withdrawn IPO dataset are matched to TFSD’s database of successful IPOs. To ensure the data are as complete as possible, we use a number of other approaches to identify returning IPOs. TFSD provides a contact name for each issuer in its database. We look for common names in the two databases. TFSD also provides information on business location, which we use as a check. In other cases we look for name matches (using parts of names). As a last step, where possible, we check our matches of withdrawn and successful offerings using actual filing documents from Edgar to ensure that the matches are correct. In spite of our best efforts, we recognize that it is likely that we have missed some returning issuers.

In Table 1 we report the number of observations in our initial database, broken down by filing year and ultimate outcome (completed or withdrawn offering, and for withdrawn offerings we report the number of cases where the firm returns for a successful IPO). Overall we have 7,442 firms in our database, 1,473 of which were withdrawn (approximately 20%). Of those firms withdrawing an IPO, only 138 (or a little over 9%) ever return for a successful offering. The number of filings varies considerably over time from a low of 154 in 1989 to a high of 824 in 1996. The percentage of withdrawn IPOs ranges from 8.88% in 1991 to a staggering 55.29% in 2000. The percentage of successful returns also varies considerably over time from 0.34% in 2000 to 17.31% in 1992. Ignoring the more recent two years (since many of those firms have not had time to return), the lowest rate of successful returns is 2.70% in 1985. The correlation between the annual number of filings and the annual percent of withdrawals is 0.07. The correlation between the annual percentage of withdrawals and the percentage of those withdrawals that return is 0.08 (if the last two years are excluded, the correlations are �0.04 and 0.28, respectively).

ARTICLE IN PRESS

Table 1

IPO filings

Number of IPO filings from 1985 to 2000. The sample is obtained from the Thomson Financial Securities Data (TFSD) database. Issues

that are unit offerings, REITs, ADRs, or closed-end funds are excluded.

Year of filing Number of IPOs

filed in year

Number filed in

year that are

withdrawn

Number filed in year

that are withdrawn

then return for

successful IPO

Percentage of filings

that are withdrawn

Percentage of

withdrawn offerings

that return for a

successful IPO

1985 298 37 1 12.42 2.70

1986 665 92 7 13.83 7.61

1987 448 95 14 21.21 14.74

1988 185 35 3 18.92 8.57

1989 154 15 1 9.74 6.67

1990 171 35 5 20.47 14.29

1991 394 35 4 8.88 11.43

1992 507 104 18 20.51 17.31

1993 623 83 14 13.32 16.87

1994 504 116 16 23.02 13.79

1995 564 54 8 9.57 14.81

1996 824 128 17 15.53 13.28

1997 569 113 6 19.86 5.31

1998 405 131 21 32.35 16.03

1999 592 102 2 17.23 1.96

2000 539 298 1 55.29 0.34

Total 7442 1473 138 19.79 9.37

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 613

3. Determinants of the choice to withdraw an IPO

While a number of studies have empirically investigated withdrawals, we provide a more thorough investigation by examining an extensive set of variables that we categorize into four areas: issuer and issue characteristics, investment bank characteristics, market conditions at the time of filing, and market conditions after the filing. Variables in the first two categories are used by Dunbar (1998, 2000) or BBG, while most of the variables in the last two categories are new. We describe predicted effects of each of these variables on withdrawals in some detail because we rely on similar arguments in later analyses that examine the likelihood that withdrawn firms will return.

First, we consider variables related to deal riskiness. Benveniste and Spindt (1989) present a model where investment banks precommit to allocation and pricing schemes that induce investors to truthfully reveal information regarding the value of securities being issued prior to final pricing. When information revealed is sufficiently negative, offerings can be withdrawn. They argue that negative information is more likely to arise in offerings by firms whose value, ex ante, is more uncertain. This suggests that offerings by firms with greater ex ante valuation uncertainty, and hence risk, are more likely to be unsuccessful. Four issuer and issue characteristic variables related to deal riskiness are included in our empirical analyses. Filing size is the average filing price (the average of the low and high price indicated in the initial prospectus) multiplied by the number of shares (in millions) to be offered as indicated in the initial prospectus. In order to control for differences in filing dates, filing sizes are measured in January 2000 dollars using the CPI as a deflator. Firms with lower filing sizes (and lower prices) tend to be riskier (see Seguin and Smoller, 1997). The technology dummy is set equal to one when the issuer is from Fama and French (1997) industries 34 (business services) and 36 (chips). Withdrawal can cause bad publicity, and this is particularly true for technology firms where information asymmetries are likely to be most significant. Employees and suppliers are also likely to have job-specific skills making withdrawal very costly (see Titman and Wessels, 1988). Thus, we would expect the likelihood of withdrawal to be negatively related to our industry dummy. The venture capital backing dummy, a proxy for a firm’s access to capital, is set equal to one if the issuing firm has venture capital backing prior to the filing date. The debt retirement dummy, another measure of access to capital, is set equal to one if the primary use of

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635614

proceeds in the IPO is to retire debt. Firms planning to retire debt and those with venture capital backing presumably have greater access to capital and, therefore, would be less dependent on an IPO. This would suggest that the likelihood of withdrawal should be positively related to the debt retirement dummy and the venture capital dummy. Alternatively, issues with venture capital backing could be argued to have greater certification. In this case, the probability of withdrawal should be lower. Given this ambiguity, we leave it to the data to determine which effect dominates.

Second, we consider variables investment bank reputation variables. Potential IPO investors face the classic lemons problem (Akerlof, 1970): since insiders have better information regarding the true value of their firm, they have an incentive to offer securities when they are overvalued by investors. Booth and Smith (1986) argue that this problem can be ameliorated if insiders credibly certify that they are not selling overpriced securities by hiring an investment bank, which relies on its reputation to win future business, to manage the offering. Other certification mechanisms, such as insider retention and venture capital backing, are examined by Grinblatt and Hwang (1989) and Lerner (1994). Three investment bank reputation and certification variables are included in our empirical analyses. Carter-Manaster rank is obtained from Carter and Manaster (1990) as updated by Carter, Dark, and Singh (1998) and more recently by Loughran and Ritter (2004). These rankings are on a 0 to 9 scale, with 9 being the most reputable underwriter. Bank market share is measured for the bank taking the firm public. For each IPO we examine all IPOs in the year leading up to the offer (including the IPO). We compute the sum of gross proceeds (on global shares excluding overallotments) for which the underwriter is also the book manager. To account for mergers in the investment banking industry, we gather data from TFSD on all combinations during the period. If the book manager recently merged, the gross proceeds of all offerings by any precedent bank are added together. In cases with multiple book managers, equal credit is given to each bank. Market share is then defined as the sum of gross proceeds for the bank, divided by the sum of gross proceeds for all IPOs over the sample period. Bank industry market share is measured as the sum of gross proceeds for the bank over the year prior to the IPO of all offerings in the same Fama-French industry as the issuer (where the book manager is the same as the one in the current deal) divided by the sum of gross proceeds on all industry IPOs over the same period. Offerings brought forward by banks with higher Carter-Manaster ranks, overall market shares, and industry market shares have greater certification. We would, therefore, expect that the likelihood of withdrawal is lower for those issuers. Welch’s (1992) cascades model results in a similar prediction for investment bank ranking. Bates and Dunbar (2002) note that market share could also be capturing a bank’s market power. If banks use this power to ensure that deals are completed, the relation between market share and withdrawal should still be negative.

Third, we consider five variables reflecting market conditions at the time of the filing. As a measure of the intensity of the IPO market, we include the number of IPO filings over the two months prior to the IPO’s filing date, number of filings in prior two months (Dunbar, 1998; BBG; Booth and Chua, 1996; and Benveniste, Ljungqvist, Wilhelm, and Yu, 2003). Market intensity can have two effects on withdrawals. In markets with more filings, information spillovers become more significant, resulting in enhanced precision of valuation, suggesting that withdrawals are less likely

1 . Alternatively, the pool of available capital could be limited,

suggesting that withdrawals are more likely. A second measure of market intensity is the number of filings over the two months prior to an IPO in the same Fama-French industry as the issuer (number of industry filings in prior two months). If information spillovers are more effective in reducing valuation uncertainty at the industry level, this variable would do a better job of picking up this effect. Access to capital could remain important at the industry level, however, so we again leave it to the data to determine which relation dominates. We include two interest rate variables to provide information about market conditions at the time of the filing. BAA-AAA yield spread at filing is defined as the difference between average rates on BAA-rated corporate bonds (by Moody’s) and AAA-rated bonds. This yield spread is often used as an indicator of default probabilities in the economy. In periods when the spread is large, default probabilities are expected to be higher. If negative firm information is more likely to arise in this market environment, we would expect withdrawals to be more

1 Booth and Chua (1996) present a model of the IPO market where information-gathering costs are reduced when offerings are clustered.

This process results in a greater precision of IPO valuation by investment banks, increasing the probability of offering success. Number of

filings could also be positively associated with the probability of success if filings are proxying for general economic conditions (i.e., more

business starts during expansionary periods) or market irrationality (more firms file when valuations are positively biased).

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 615

likely when spreads are higher. As an alternative view, access to capital is often limited when spreads are large. Firms attempting to raise capital in high-spread environments could have few alternatives and are, therefore, less likely to cancel an IPO. The relation between yield spreads and the probability of withdrawal is ultimately, therefore, an empirical question. The second interest rate variable is the yield on ten-year Treasury bonds (ten-year Treasury yield). In periods when interest rates are high, alternative sources of capital should be either scarce or expensive. An alternative view is that long-term rates tend to increase during expansionary periods. Again, the relation with the probability becomes an empirical question. As a measure of relative valuation we include the industry average book-to-market ratio, measured one year prior to the filing

2 . If book-to-market

captures growth opportunities, then the relation should be positive (firms with more growth having lower book-to-market ratios are less likely to withdraw; the argument leading to this prediction is similar to that made for the technology dummy variable). On the other hand, if book-to-market captures market misvaluation, firms with low book-to-market ratios (overvalued firms) should be more likely to withdraw (it is more likely that firms will be detected as overvalued during the bookbuilding process).

All of the empirical measures in the three categories discussed thus far are observable at the time of the initial IPO filing. A probit model estimated just with these variables, therefore, provides insights into the ex ante likelihood of offering success. Information received after the filing date should also affect a firm’s decision to withdrawal, however. We therefore examine a fourth category that includes six variables to proxy for market conditions after filing. Number of filings two months after filing and number of industry filing two months after filing capture the IPO market intensity during bookbuilding. Changes to the interest rate environment are captured by the change in BAA-AAA yield spread two months after filing and the change in ten-year Treasury yield two months after filing. Changes to the stock market environment are captured by the return on the Nasdaq composite index over two months after filing and the change in industry BM (book-to-market) over year of filing. Predictions regarding the relation between these variables and withdrawals are similar to those for similar variables before withdrawals. Where there are ambiguous arguments regarding the effect of variables, it is possible to have different relations before and after filing (one argument could dominate before and the other after).

As a preliminary univariate investigation, we report descriptive statistics for the data items, broken down according to the ultimate success of the offering in Table 2. Note that sample sizes change depending on the variable examined, reflecting the fact that TFSD coverage of data items is extremely limited in some cases (especially for withdrawn issues). Most of the differences between successful and withdrawn IPOs are significant and as expected. Withdrawn offerings have significantly lower average initial filing size compared to completed offerings. A greater percentage of technology firms are successful than withdrawn. Firms with venture capital backing are more likely to succeed. This is consistent with venture capital backing acting as a certification for the offering (but not with the capital constraints conjecture). Also, Gompers (1996) notes that venture capitalists have an incentive to bring firms early to the IPO market to capitalize their claims. This argument would suggest that venture capitalists would lobby hard for IPO completion. Withdrawn offerings are also more likely to have been targeting for debt retirement. Successful offerings are more likely to be taken public by banks with greater reputations, proxied by Carter-Manaster rankings, market share, and industry market share.

Withdrawn IPOs are filed after periods with a greater average number of filings over the prior two months than completed offers, suggesting that companies are more likely to withdraw when more companies are competing for limited capital. In addition, withdrawn IPOs are filed after periods with greater numbers of industry filings, suggesting a limited pool of capital. BAA-AAA yield spreads are higher for successful offerings, suggesting that yield spreads are more likely to be capturing access to capital than default probabilities. Withdrawn offerings are more likely to be from industries with lower book-to-market ratios, consistent with predictions from the misvaluation theory (firms in lower book-to-market industries are more likely to be detected as overvalued).

The effect of changes in yield spreads is opposite to that detected pre-filing (the change is more positive for withdrawn offerings). The Nasdaq composite return post-filing is more positive for successful offerings, suggesting a better investment climate. Finally, the changes to the industry book-to-market ratio are less

2 Obtained from Ken French’s web site (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html).

ARTICLE IN PRESS

Table 2

Descriptive statistics—successful and withdrawn IPOs

This table reports sample means and number of observations for different variables broken down by whether the IPO filing is successful or

withdrawn. Issuer and issue characteristic variables are defined as follows. Average filing price is the average of the high and low price indicated

in the initial filing. Filing size equals the average filing price multiplied by the number of shares to be sold as indicated in the initial filing

(reported in January 2000 dollars using the CPI as a deflator). Technology dummy takes the value one if the issuer is in Fama-French industries

34 (business services) or 36 (chips) and zero otherwise (see Fama and French, 1997). Venture capital backing dummy takes the value one if the

issuing firm has received venture capital investments prior to filing and zero otherwise. Debt retirement dummy takes the value one if the

primary stated use of proceeds is retirement of debt. Investment bank characteristic variables are defined as follows. Carter-Manaster rank is

the Carter-Manaster (1990) ranking on a 0–9 scale for the book manager of the IPO (the maximum rank if there is more than one book

manager). Bank market share is the sum of gross proceeds (not including the overallotment option) over the year prior to the IPO of all

offerings where the IPO book manager is the book manager (equal credit is given if there is more than one manager) divided by the sum of

gross proceeds on all IPOs over the same period. Bank industry market share is the sum of gross proceeds over the year prior to the IPO of all

IPOs in the same Fama-French industry where the IPO book manager is the book manager divided by the sum of gross proceeds on all

industry IPOs over the same period. Market condition variables at the time of the filing are defined as follows. Number of filings in prior two

months is the number of IPOs filed with the SEC during the two months prior to the filing date for the IPO. Number of industry filings in prior

two months is the number of IPOs in the same Fama-French industry filed with the SEC during the two months prior to the filing date for the

IPO. BAA-AAA yield spread at filing is the spread between BAA and AAA corporate bonds (from Moody’s) on the day of the filing. Ten-year

Treasury yield at filing is the average yield on US Treasury bonds having ten years to maturity measured on the day of the filing. Industry

average book-to-market ratio pre-filing is the book-to-market ratio for firms in the IPO issuer’s Fama-French industry at the end of the year

prior to filing (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). Market condition variables after the offering are

defined as follows. Number of filings two months after filing is the number of IPOs filed with the SEC during the two months after the filing date

for the IPO. Number of industry filings two months after filing is the number of IPOs in the same Fama-French industry filed with the SEC

during the two months after the filing date for the IPO. Change in BAA-AAA yield spread two months after filing is the BAA-AAA yield spread

two months after the filing date less the yield spread on the filing date. Change in ten-year Treasury yield two months after filing is the ten-year

Treasury yield two months after the filing date less the ten-year Treasury yield on the filing date. Return on Nasdaq Composite Index over two

months after filing is the compound return on the index over the two months beginning on the filing date. Change in industry BM over year of

filing is the industry average book-to-market ratio at the end of the filing year less the average ratio at the beginning of the filing year.

Successful Withdrawn p-values (from t-test)

successful vs. withdrawn

Mean Obs Mean Obs

Issuer and issue characteristics

Filing size 69.510 5538 59.118 1076 0.025

Technology dummy 0.276 5538 0.241 1076 0.014

Venture capital backing dummy 0.388 5538 0.148 1075 0.000

Debt retirement dummy 0.317 5338 0.428 306 0.000

Investment bank characteristics

Carter-Manaster rank 7.069 5538 6.625 1076 0.000

Bank market share 4.330 5538 1.881 1076 0.000

Bank industry market share 14.453 5538 2.715 1076 0.000

Market conditions at time of filing

Number of filings in prior two months 100.650 5538 106.550 1076 0.000

Number of industry filings in prior two months 10.948 5538 12.204 1076 0.053

BAA-AAA yield spread at filing 0.836 5538 0.810 1076 0.001

Ten-year Treasury yield at filing 6.872 5538 6.843 1076 0.475

Industry average book-to-market ratio 0.474 5538 0.451 1076 0.007

Market conditions after the filing

Number of filings two months after filing 102.750 5538 100.490 1076 0.125

Number of industry filings two months after filing 11.265 5538 11.765 1076 0.454

Change in BAA-AAA yield spread two month

after filing

�0.003 5538 0.013 1076 0.000

Change in ten-year Treasury yield two months after filing �0.028 5538 �0.048 1076 0.195

Return on Nasdaq Composite Index over two months after filing 0.028 5538 �0.012 1076 0.000

Change in industry BM over year of filing �0.045 5538 �0.028 1076 0.000

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635616

negative for withdrawn offers, consistent with book-to-market capturing growth opportunities. This appears to be inconsistent with the evidence for the book-to-market ratio at the time of filing, which is more consistent with this ratio capturing firm overvaluation. Since both overvaluation and growth opportunity explanations

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 617

could play a role in explaining withdrawals, it is possible that the book-to-market variables measured at different times are simply picking up the different effects.

We formalize our univariate analysis of the determinants of IPO withdrawal in Table 3 using a probit analysis. The dependent variable in the analysis takes the value one if the IPO filing is withdrawn and zero otherwise. We consider as independent variables all of the measures noted above. Table 3 reports our probit model coefficient estimates and associated t-statistics. We also report the marginal effect for each variable

Table 3

Probit analysis of the decision to withdraw an IPO for IPO filings between 1985 and 2000

The dependent variable equals one for IPO filings that are withdrawn and zero for completed offerings. Issuer and issue characteristic

variables are defined as follows. Logarithm of the filing size equals the natural logarithm of the average filing price (average of the high and

low price indicated in the initial filing) multiplied by the number of shares to be sold as indicated in the initial filing (reported in January

2000 dollars using the CPI as a deflator). Technology dummy takes the value one if the issuer is in Fama-French industries 34 (business

services) or 36 (chips) and zero otherwise (see Fama and French, 1997). Venture capital backing dummy takes the value one if the issuing

firm has received venture capital investments prior to filing and zero otherwise. Debt retirement dummy takes the value one if the primary

stated use of proceeds is retirement of debt. Investment bank characteristic variables are defined as follows. Carter-Manaster rank is the

Carter and Manaster (1990) ranking on a 0-9 scale for the book manager of the IPO (the maximum rank if there is more than one book

manager). Bank market share is the sum of gross proceeds (not including the overallotment option) over the year prior to the IPO of all

offerings where the IPO book manager is the book manager (equal credit is given if there is more than one manager) divided by the sum of

gross proceeds on all IPOs over the same period. Bank industry market share is the sum of gross proceeds over the year prior to the IPO of

all IPOs in the same Fama-French industry where the IPO book manager is the book manager divided by the sum of gross proceeds on all

industry IPOs over the same period. Market condition variables at the time of the filing are defined as follows. Number of filings in prior

two months is the number of IPOs filed with the SEC during the two months prior to the filing date for the IPO. Number of industry filings

in prior two months is the number of IPOs in the same Fama-French industry filed with the SEC during the two months prior to the filing

date for the IPO. BAA-AAA yield spread at filing is the spread between BAA and AAA corporate bonds (from Moody’s) on the day of the

filing. Ten-year Treasury yield at filing is the average yield on US Treasury bonds having ten years to maturity measured on the day of the

filing. Industry average book-to-market ratio pre-filing is the book-to-market ratio for firms in the IPO issuer’s Fama-French industry at the

end of the year prior to filing (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). Market condition variables

after the offering are defined as follows. Number of filings two months after filing is the number of IPOs filed with the SEC during the two

months after the filing date for the IPO. Number of industry filings two months after filing is the number of IPOs in the same Fama-French

industry filed with the SEC during the two months after the filing date for the IPO. Change in BAA-AAA yield spread two months after

filing is the BAA-AAA yield spread two months after the filing date less the yield spread on the filing date. Change in ten-year Treasury

yield two months after filing is the ten-year Treasury yield two months after the filing date less the ten-year Treasury yield on the filing date.

Return on Nasdaq Composite Index over two months after filing is the compound return on the index over the two months beginning on the

filing date. Change in industry BM over year of filing is the industry average book-to-market ratio at the end of the filing year less the

average ratio at the beginning of the filing year. Marginal effect is defined as f(bx)�b�sx where f() is the standard normal probability density function, b is the coefficient estimate, x is the mean of the independent variable for the sample, and sx is one standard deviation for the independent variable (sx is set to 1 for dummy variables). Pseudo R

2 is defined as 1 less the log likelihood for the estimated model

divided by the log-likelihood for a model with only an intercept as an independent variable.

Coefficient Marginal

effect

t-stat Coefficient Marginal

effect

t-stat

Intercept �0.715 �0.221 �3.47 2.586 0.036 5.03

Issuer and issue characteristics

Logarithm of the filing size 0.209 0.071 7.07 0.346 0.070 6.49

Technology dummy �0.350 �0.139 �5.30 �0.398 �0.158 �3.34

Venture capital backing dummy �0.769 �0.296 �14.27 �0.745 �0.286 �8.38

Debt retirement dummy 0.282 0.112 3.97

Investment bank characteristics

Carter-Manaster rank 0.014 0.013 1.12 �0.005 �0.004 �0.23

Bank market share �0.032 �0.074 �5.49 �0.048 �0.111 �4.39

Bank industry market share �0.056 �0.347 �18.23 �0.058 �0.345 �10.32

Market conditions at time of filing

Number of filings in prior two months 0.001 0.020 1.63 0.005 0.075 4.38

Number of industry filings in prior two months 0.001 0.009 0.55 �0.005 �0.031 �1.08

BAA-AAA yield spread at filing 0.080 0.008 0.63 �1.884 �0.057 �4.55

Ten-year Treasury yield at filing 0.012 0.006 0.43 �0.438 �0.002 �6.72

Industry average book-to-market ratio �0.389 �0.041 �3.69 �2.033 �0.138 �6.90

ARTICLE IN PRESS

Table 3 (continued )

Coefficient Marginal

effect

t-stat Coefficient Marginal

effect

t-stat

Market conditions after the filing

Number of filings two months after filing �0.004 �0.057 �5.07 �0.003 �0.045 �2.33

Number of industry filings two months after filing 0.005 0.031 1.73 0.002 0.012 0.39

Change in BAA-AAA yield spread two month after

filing

0.387 0.015 1.63 �1.838 �0.070 �3.67

Change in ten-year Treasury yield two months

after filing

�0.123 �0.023 �2.39 �0.260 �0.047 �2.46

Return on Nasdaq Composite Index over two

months after filing

�2.840 �0.102 �11.54 �0.675 �0.023 �1.61

Change in industry BM over year of filing 1.010 0.040 3.71 �0.042 �0.002 �0.06

Pseudo R 2

0.214 0.316

Number of observations 6613 5644

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635618

which captures the change in the probability of withdrawal given a one standard deviation change in the independent variable.

Our first probit model includes all the variables noted above except the debt retirement variable (the sample size drops when this variable is included so we consider that sample separately). We measure offering size using the natural logarithm of the average filing size (logarithm of the filing size) to account for scale effects. Our analysis largely confirms the findings from the univariate analysis. The probability of withdrawal is significantly positively related to the number of filings and industry filings over the two months post-filing (suggesting competition for capital) and to the change in the industry book-to-market ratio over the year of filing. The probability of withdrawal is negatively related to the technology dummy (withdrawing can cause bad publicity), the venture capital backing dummy (greater certification), the investment bank overall market share (reputation), the industry market share (reputation), the average industry book-to-market ratio pre- filing (misvaluation), the number of filings post-filing (reduced information spillovers), the change in the ten- year Treasury yield two months after filing (economic conditions), and the return on the Nasdaq composite index post-filing (market conditions). Note that the coefficient signs on the two variables measuring the number of filings (pre- and post-filing) are different. Since there are different possible predictions based on information spillover and competition for capital, our results suggests that both explanations help to explain withdrawals. Industry market share is the most significant variable, economically, with a one standard deviation increase in industry market share resulting in a 35% reduction in the probability of withdrawal.

Our probit analysis indicates that logarithm of the filing size is significantly positively related to the probability of withdrawal, reversing the negative relation found in the univariate analysis. All else equal, firms attempting to raise more capital might have more alternative sources of capital or are simply more likely to be ‘‘caught’’ attempting to raise too much money given market misvaluation (detected in the bookbuilding process).

The second probit model reported in Table 3 includes the debt retirement variable. The sample size drops due to data limitations. The economic and statistical significance of most variables previously considered are not altered in this smaller sample, with a few exceptions. For example, the ten-year Treasury yield at filing, the yield spread at filing, and the change in yield spread have a significant negative effect on the likelihood of withdrawal. The Nasdaq composite return post-filing now does not have a significant effect on the probability of withdrawals, nor does the change in the industry book-to-market ratio. The new variable, debt retirement, has a significantly positive effect on the probability of withdrawal, consistent with the univariate analysis.

4. Withdrawn IPOs that return to the market

In this section we examine the sample of 1,485 withdrawn IPOs from 1985 to 2000. As noted previously, 138 of these firms (or approximately 9%) return to the market for a successful IPO. For the sample of successfully

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 619

returning firms, we first examine the average length of time between the first unsuccessful withdrawal date and the ultimately successful issue date. Results are presented in Table 4. From Panel A, the mean (median) time is 819 (663) days or 2.24 (1.82) years. The minimum is 77 days or 0.21 years, while the maximum is 3,523 days or 9.65 years. We report the full distribution of time to return in Panel B of Table 4. Of those that return for a successful IPO, 95% take more than 142 days to return. Only 25% take more than 3.24 years to return. If ‘‘market conditions,’’ often a stated reason for the postponement of new issues, are a primary cause of the initial withdrawal, conditions should have improved within a few years at the latest. However, it could well be that some firms cannot survive while waiting for market conditions to improve.

We next examine which factors most affect the probability of successful return. As discussed previously, firms are less likely to withdraw an IPO if cancellation can cause a ‘‘lemons’’ problem for the issuing firm (Akerlof, 1970). Riskier firms that withdraw should be less likely to be able to return successfully in the future. Empirical measures considered previously as proxies for issuer and issue riskiness should also be considered here (filing size, technology dummy, and venture capital backing dummy). However, as noted in the previous section, there is a high cost for technology firms to withdraw, so firms that choose to withdraw could be the ones that are more likely to return, resulting in a positive relation between the technology dummy and the probability of successful return. Since the technology variable could have two different effects, we let the data determine. Firms canceling an IPO because they have access to other good sources of capital (proxied by the venture capital backing variable) should be more likely to return (since ‘‘questions’’ about firm quality should less significant).

Table 4

Time from initial withdrawal to successful reissue for 138 IPOs from 1985 to 2000

This table reports the length of time between a withdrawn issue and successful re-issue. Panel A reports sample statistics and Panel B

reports the entire distribution of time to return (e.g., 95% of successful returners take more than 142 days to return).

Panel A

Day Years

Mean 818.6 2.24

Median 663.0 1.82

Minimum 77 0.21

Maximum 3,523 9.65

Standard deviation 630.1 1.73

Panel B

Fraction of Successful Returners (%) Days to return exceed Years to return exceed

95 142 0.39

90 216 0.59

85 270 0.74

80 307 0.84

75 329 0.9

70 378 1.04

65 427 1.17

60 517 1.42

55 604 1.65

50 663 1.82

45 777 2.13

40 828 2.27

35 928 2.54

30 963 2.64

25 1,181 3.24

20 1,291 3.54

15 1,398 3.83

10 1,628 4.46

5 2,223 6.09

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635620

Firms with greater certification (taken forward initially by more reputable investment banks or having venture capital backing) should face lower ‘‘lemons’’ concerns, increasing the likelihood of successful return

3 .

As in the previous section, we utilize the three investment bank characteristics as measures of reputation: Carter-Manaster rank, bank market share, and bank industry market share.

Market conditions at the time of withdrawal should also impact a firm’s ability to return. Market intensity prior to withdrawal, measured by number of filings in prior two months and number of industry filings in prior two months, is likely to affect the likelihood of return. Firms that are unsuccessful because resources are scarce could be more likely to return when competition for capital declines in the future. Alternatively, firms canceling an IPO in intense markets when information spillovers are more significant should be less likely to return since these are the firms detected to be undesirable investments. As discussed earlier, interest rate- related variables (BAA-AAA yield spread at withdrawal and ten-year Treasury yield at withdrawal) could proxy for a firm’s access to alternative sources of capital. Firms withdrawing when yield spreads and interest rates are low should be more likely to return (they find interim capital and face less significant lemons concerns). Alternatively, interest rate variables could be capturing macroeconomic conditions. Issuers withdrawing when rates and spreads are high could be more likely to return since the economy is expanding. Other market condition variables, industry average book-to-market pre-withdrawal and change in industry BM over year pre- withdrawal, reflect changing market sentiment, resulting in changing valuations. Firms withdrawing an IPO after a market decline as captured by return on Nasdaq Composite Index from filing to withdrawal (or declining industry valuation ratios) are less likely to be able to return.

Finally, variables capturing market conditions after the withdrawal should affect the likelihood of successful return. IPO market intensity after withdrawal (number of filings 12 months after withdrawal and number of industry filings 12 months after withdrawal) should impact the likelihood of successful return. In the previous section we focused on withdrawals and as such focused on a two-month window after filing since this is the typical registration period for successful offerings. In this section we consider a longer 12-month period after withdrawals in order to more closely match the typical time period between initial filing and possible return (see Table 4). The impact of market intensity after withdrawal on the likelihood of withdrawal is difficult to predict. Under the scarce resource view, failed IPOs are less likely to be able to return if market intensity increases after withdrawal since competition for scarce capital does not subside. Under the information spillover view, greater intensity could result in more positive valuation spillovers, increasing the likelihood of return. Variables proxying for access to alternative sources of capital should also be related to the probability of successful return. As sources of capital dry up, those canceling IPOs should turn back to the market. Thus, as yield spreads, as measured by change in BAA-AAA yield spread 12 months after withdrawal, and Treasury yields, as measured by change in ten-year Treasury yield 12 months after withdrawal, increase, the likelihood of return should increase. The macroeconomic conditions view of interest rate variables makes similar predictions (likelihood of return increases in expansions). If valuations improve after withdrawal (Nasdaq Composite increases, as measured by return on Nasdaq Composite Index 12 months after withdrawal), firms should be more likely to emerge successfully.

As a preliminary univariate investigation, Table 5 reports descriptive statistics for the data items, broken down by whether the failed issuer is able to return for a successful IPO. Returners are significantly more likely to be venture capital backed (greater certification). The Carter-Manaster ranking is significantly higher for filers that return for a successful offering than for filers that never return (greater certification). The number of industry filings two months prior to withdrawal is significantly lower for returning issuers, consistent with the scarce resource view of market intensity. The number of filings 12 months after withdrawal is greater for

3 While firms backed by a venture capitalist have greater certification, making it easier for them to return to the IPO market, there is

another consideration related to venture capital backing. Venture capitalists are more likely to be more diversified and hence are likely to

be less risk averse than firm managers. Consequently, venture capitalists might have a preference for trying (again) to take the firm public

in an IPO, as an IPO tends to have a higher payoff than an acquisition. In contrast, firm managers might have a preference for the more

certain payoff from an acquisition if that alternative becomes available. Given that venture capitalists almost always have board seats, they

might push firms towards returning to the IPO market. However, it is not clear whether mere board representation will result in venture

capitalists carrying the day. We thank the referee for this observation. The debt retirement variable can also have an alternative

interpretation. Firms wishing to repay debt could be close to their debt capacity. In this case, these firms could have more limited

alternative sources of capital.

ARTICLE IN PRESS

Table 5

Descriptive Statistics—withdrawn IPOs broken down by eventual public status

This table reports sample means and number of observations for different variables broken down by whether the withdrawn filing

eventually returns for a successful IPO or not. Issuer and issue characteristic variables are defined as follows. Average filing price is the

average of the high and low price indicated in the initial filing. Filing size equals the average filing price multiplied by the number of shares

to be sold as indicated in the initial filing (reported in January 2000 dollars using the CPI as a deflator). Technology dummy takes the value

one if the issuer is in Fama-French industries 34 (business services) or 36 (chips) and zero otherwise (see Fama and French, 1997). Venture

capital backing dummy takes the value one if the issuing firm has received venture capital investments prior to filing and zero otherwise.

Investment bank characteristic variables are defined as follows. Carter-Manaster rank is the Carter-Manaster (1990) ranking on a 0–9 scale

for the book manager of the IPO (the maximum rank if there is more than one book manager). Bank market share is the sum of gross

proceeds (not including the overallotment option) over the year prior to the IPO of all offerings where the IPO book manager is the book

manager (equal credit is given if there is more than one manager) divided by the sum of gross proceeds on all IPOs over the same period.

Bank industry market share is the sum of gross proceeds over the year prior to the IPO of all IPOs in the same Fama-French industry where

the IPO book manager is the book manager divided by the sum of gross proceeds on all industry IPOs over the same period. Market

condition variables at the time of the withdrawal are defined as follows. Number of filings in prior two months is the number of IPOs filed

with the SEC during the two months prior to the withdrawal date for the IPO. Number of industry filings in prior two months is the number

of IPOs in the same Fama-French industry filed with the SEC during the two months prior to the withdrawal date for the IPO. BAA-AAA

yield spread at withdrawal is the spread between BAA and AAA corporate bonds (from Moody’s) on the day of the filing. Ten-year

Treasury yield at withdrawal is the average yield on US Treasury bonds having ten years to maturity measured on the day of the

withdrawal. Industry average book-to-market ratio pre-withdrawal is the book-to-market ratio for firms in the IPO issuer’s Fama-French

industry at the end of the year prior to filing (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). Return on

Nasdaq Composite Index from filing to withdrawal is the compound return on the index from the filing date to the withdrawal date. Market

condition variables after the withdrawal are defined as follows. Number of filings 12 months after withdrawal is the number of IPOs filed

with the SEC during the 12 months after the withdrawal date for the IPO. Number of industry filings 12 months after withdrawal is the

number of IPOs in the same Fama-French industry filed with the SEC during the 12 months after the withdrawal date for the IPO. Change

in BAA-AAA yield spread 12 months after withdrawal is the BAA-AAA yield spread 12 months after the withdrawal date less the yield

spread on the withdrawal date. Change in ten-year Treasury yield 12 months after withdrawal is the ten-year Treasury yield 12 months after

the withdrawal date less the ten-year Treasury yield on the withdrawal date. Return on Nasdaq Composite Index 12 months after withdrawal

is the compound return on the index over the 12 months beginning the withdrawal date. Change in industry BM over year of withdrawal is

the change in the IPO industry book-to-market ratio over the calendar year ending after the withdrawal.

Initial filing of withdrawn

IPOs that later return for

successful offer

Initial filing of withdrawn

IPOs that never return for

successful offer

p-values (from

t-test)—not

return vs.

return

Mean Obs Mean Obs

Issuer and issue characteristics

Filing size 60.473 120 58.948 956 0.856

Technology dummy 0.250 120 0.240 956 0.804

Venture capital backing dummy 0.250 120 0.135 955 0.006

Investment bank characteristics

Carter-Manaster rank 7.445 120 6.522 956 0.000

Bank market share 2.055 120 1.860 956 0.633

Bank industry market share 1.160 120 2.911 956 0.000

Market conditions at time of withdrawal

Number of filings in prior two months 88.242 120 86.796 956 0.681

Number of industry filings in prior two months 6.833 120 9.719 956 0.002

BAA-AAA yield spread at withdrawal 0.817 120 0.816 956 0.963

Ten-year Treasury yield at withdrawal 6.895 120 6.693 956 0.089

Industry average book-to-market ratio 0.445 120 0.423 956 0.416

Return on Nasdaq composite index from filing to withdrawal 0.023 120 0.033 956 0.491

Market conditions after the withdrawal

Number of filings 12 months after withdrawal 501.390 120 452.760 956 0.002

Number of industry filings 12 months after withdrawal 39.000 120 39.171 956 0.974

Change in BAA-AAA yield spread 12 months after withdrawal �0.061 120 �0.003 956 0.000

Change in ten-year Treasury yield 12 months after withdrawal �0.126 120 �0.302 956 0.094

Return on Nasdaq composite index 12 months after withdrawal 0.243 120 0.153 956 0.000

Change in industry BM over year of withdrawal �0.016 120 �0.004 956 0.057

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 621

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635622

returning firms, consistent with the spillover view of market intensity. Alternatively, a higher number of filings might suggest that it is simply a better time to go public due to stronger economic conditions. The return on the Nasdaq composite index 12 months after withdrawal is significantly higher for returning issuers, consistent with predictions.

Two results are not consistent with predictions. First, industry market share is significantly lower for firms that eventually return. There are two possible explanations for this finding. As we will see in upcoming analyses, returning firms often switch to banks with greater industry market shares that are then more likely to successfully return. Further, industry market share could be considered to be more a measure of market power than reputation (see Bates and Dunbar, 2002). Issuers unsuccessful in their first attempt to go public even though they are taken forward by banks with more market power are more likely to be ‘‘flawed,’’ and therefore less likely to return. The second confounding finding is that the change in the BAA-AAA yield spread post-withdrawal is significantly lower for returning issuers, although this finding does not hold up in the multivariate analysis presented next.

We formalize our analysis of the determinants of IPO return for withdrawn IPOs in Table 6 using a multivariate probit analysis (We estimate, but do not report, probit models using a subsample of data where debt retirement is observable; the findings are similar to those reported here for the variables considered and the debt retirement variable is not significant). The dependent variable in the analysis takes the value one if the withdrawn IPO filing eventually returns for a successful offering and zero otherwise. This variable is biased for recent withdrawers since some will likely return at some point. The bias is probably small since most successful returners come back to market within two to three years of withdrawal (see Table 4). To check the importance of this bias more formally, we reestimate our models using a dependent variable that takes the value one if the firm returns within three years of withdrawal; the results are qualitatively similar. Our first model in Table 6 excludes variables measured after withdrawal (capturing the ex ante probability of returning as of the time of withdrawal). Our analysis largely confirms the findings from the univariate analysis. The technology dummy (suggesting self-selection), venture capital backing (suggesting greater certification), Carter-Manaster ranking (suggesting greater certification), the number of filings two months prior to withdrawal (suggesting information spillovers are important), and the ten-year Treasury yield at withdrawal have a positive effect on the likelihood of successful return. Bank industry market share (suggesting a ‘‘flawed’’ attempt) and the number of industry filings two months prior to withdrawal (suggesting competition for capital is important) have a negative impact on the likelihood of returning. One variable that is not significant in the univariate analysis becomes significant once controlling for other factors. The ten-year yield at withdrawal has a positive effect on the likelihood of a successful return, consistent with interest rates capturing macroeconomic conditions. As noted earlier, the coefficient signs on the different variables measuring the number of filings are different. Since there are different possible predictions based on information spillover and competition for capital, our results suggests that both explanations help to predict successful returning.

The second model estimated in Table 6 includes market condition variables measured after withdrawal. The signs, statistical significance, and economic significance (captured by marginal effects) of most ex ante variables are unchanged when post-withdrawal variables are included. The technology dummy and the number of filings two months prior to withdrawal are no longer statistically significant. The BAA-AAA yield spread at withdrawal (suggesting access to interim capital) and the return on the Nasdaq Composite Index from filing to withdrawal (not supporting the improved market conditions argument) become significantly negative.

Only two of the post-withdrawal variables are significant in the expanded model. The number of filings 12 months after withdrawal (suggesting information spillovers are important) and the change in the ten-year Treasury yield after withdrawal (suggesting better macroeconomic conditions) have a positive effect on the likelihood of successful return.

The most significant variables in this analysis, economically, are venture capital backing, investment bank industry market share, and the change in the ten-year Treasury yields one year after withdrawal. Issuers backed by venture capital are 15–19% more likely to successfully return. A one standard deviation increase in industry market share results in a 12–14% reduction in the probability of a successful return. Finally, a one standard deviation increase in ten-year Treasury yields post-withdrawal results in a 13% increase in the probability of a successful return.

ARTICLE IN PRESS

Table 6

Probit analysis of successful returns for IPO filings that were withdrawn between 1985 and 2000

The dependent variable equals one for IPO filings that are withdrawn but eventually return for a successful offering and zero for

withdrawn offerings that never return. Issuer and issue characteristic independent variables are defined as follows. Average filing price is

the average of the high and low price indicated in the initial filing. Logarithm of filing size equals the natural logarithm average the average

filing size, which is the average filing price multiplied by the number of shares to be sold as indicated in the initial filing (reported in

January 2000 dollars using the CPI as a deflator). Technology dummy takes the value one if the issuer is in Fama-French industries 34

(business services) or 36 (chips) and zero otherwise (see Fama and French, 1997). Venture capital backing dummy takes the value one if the

issuing firm has received venture capital investments prior to filing and zero otherwise. Investment bank characteristic variables are defined

as follows. Carter-Manaster rank is the Carter-Manaster (1990) ranking on a 0–9 scale for the book manager of the IPO (the maximum

rank if there is more than one book manager). Bank market share is the sum of gross proceeds (not including the overallotment option)

over the year prior to the IPO of all offerings where the IPO book manager is the book manager (equal credit is given if there is more than

one manager) divided by the sum of gross proceeds on all IPOs over the same period. Bank industry market share is the sum of gross

proceeds over the year prior to the IPO of all IPOs in the same Fama-French industry where the IPO book manager is the book manager

divided by the sum of gross proceeds on all industry IPOs over the same period. Market condition variables at the time of the withdrawal

are defined as follows. Number of filings in prior two months is the number of IPOs filed with the SEC during the two months prior to the

withdrawal date for the IPO. Number of industry filings in prior two months is the number of IPOs in the same Fama-French industry filed

with the SEC during the two months prior to the withdrawal date for the IPO. BAA-AAA yield spread at withdrawal is the spread between

BAA and AAA corporate bonds (from Moody’s) on the day of the filing. Ten-year Treasury yield at withdrawal is the average yield on US

Treasury bonds having ten years to maturity measured on the day of the withdrawal. Industry average book-to-market ratio pre-withdrawal

is the book-to-market ratio for firms in the IPO issuer’s Fama-French industry at the end of the year prior to filing (http://

mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). Return on Nasdaq Composite Index from filing to withdrawal is the

compound return on the index from the filing date to the withdrawal date. Market condition variables after the withdrawal are defined as

follows. Number of filings 12 months after withdrawal is the number of IPOs filed with the SEC during the 12 months after the withdrawal

date for the IPO. Number of industry filings 12 months after withdrawal is the number of IPOs in the same Fama-French industry filed with

the SEC during the 12 months after the withdrawal date for the IPO. Change in BAA-AAA yield spread 12 months after withdrawal is the

BAA-AAA yield spread 12 months after the withdrawal date less the yield spread on the withdrawal date. Change in ten-year Treasury

yield 12 months after withdrawal is the ten-year Treasury yield 12 months after the withdrawal date less the ten-year Treasury yield on the

withdrawal date. Return on Nasdaq Composite Index 12 months after withdrawal is the compound return on the index over the 12 months

beginning the withdrawal date. Change in industry BM over year of withdrawal is the change in the IPO industry book-to-market ratio over

the calendar year ending after the withdrawal. Marginal effect is defined as f(bx)�b�sx where f() is the standard normal probability density function, b is the coefficient estimate, x is the mean of the independent variable for the sample, and sx is one standard deviation for the independent variable (sx is set to 1 for dummy variables). Pseudo R

2 is defined as 1 less the log likelihood for the estimated model

divided by the log-likelihood for a model with only an intercept as an independent variable.

Coefficient Marginal

effect

t-stat Coefficient Marginal

effect

t-stat

Intercept �2.989 �0.014 �5.92 �4.982 0.000 �5.76

Issuer and issue characteristics

Filing size �0.020 �0.010 �0.29 0.039 0.020 0.55

Technology dummy 0.411 0.163 2.54 0.257 0.102 1.36

Venture capital backing dummy 0.397 0.158 2.85 0.488 0.194 3.30

Investment bank characteristics

Carter-Manaster rank 0.101 0.090 3.11 0.098 0.089 2.94

Bank market share 0.003 0.004 0.19 0.007 0.011 0.49

Bank industry market share �0.037 �0.123 �2.64 �0.041 �0.135 �2.77

Market conditions at time of withdrawal

Number of filings in prior two months 0.005 0.058 2.48 0.003 0.042 1.39

Number of industry filings in prior two months �0.023 �0.128 �3.53 �0.020 �0.112 �1.79

BAA-AAA yield spread at withdrawal �0.085 �0.007 �0.27 �0.939 �0.061 �1.94

Ten-year Treasury yield at withdrawal 0.128 0.042 2.21 0.427 0.003 4.62

Industry average book-to-market ratio �0.101 �0.008 �0.31 �0.236 �0.022 �0.84

Return on Nasdaq composite index from filing to withdrawal 0.059 0.006 0.22 �1.135 �0.086 �2.57

Market conditions after the withdrawal

Number of filings 12 months after withdrawal 0.001 0.077 1.98

Number of industry filings 12 months after withdrawal 0.002 0.057 1.12

Change in BAA-AAA yield spread 12 months after withdrawal �0.634 �0.049 �1.22

Change in ten-year Treasury yield 12 months after withdrawal 0.311 0.128 3.94

Return on Nasdaq composite index 12 months after withdrawal 0.402 0.049 1.40

Change in industry BM over year of withdrawal �0.889 �0.029 �1.00

Pseudo R 2

0.076 0.138

Number of observations 1075 1075

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 623

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635624

5. Investment bank switching for returning issuers

An interesting feature of the sample of successful reissuers is that in many (but not all) cases, the issuing firm switches investment banks from the initial attempt to the final success. Two notable studies examine the decision of a firm to switch investment banks from its IPO to a seasoned equity offering. James (1992) examines the underwriter switching decision in the context of relationship-specific assets. He argues that given high setup costs (learning), firms would tend not to switch banks (so they can amortize those costs over multiple offerings) unless performance by the bank on the IPO was poor. We interpret Daniel’s (2002) discussion of Loughran and Ritter (2002) as suggesting that a perception of deliberate underpricing by a particular underwriter could be an alternative explanation of why a firm might withdraw from the IPO and later return successfully with a different underwriter and with less underpricing. Consistent with this, James finds that pricing errors at the time of the IPO are significantly associated with bank switches. Krigman, Shaw, and Womack (2001) also examine the choice to switch underwriters. In addition to the possibility that firms switch due to IPO mispricing, Krigman, Shaw, and Womack consider a number of other explanations including poor share placement (resulting in high flipping), limited market-making activity, limited research coverage, and graduation (simply moving to a bank with greater reputation). They find evidence most consistent with the limited research coverage and graduation explanations.

In the context of a previously withdrawn IPO, many of the explanations proposed by Krigman, Shaw, and Womack simply cannot apply, including IPO mispricing, poor share placement, limited market-making, and limited research coverage. All of those explanations require that the firm become public. We consider the two remaining possibilities in our analysis. First, firms might switch investment banks because a bank with a greater reputation is willing to underwrite the offering (the graduation hypothesis). Second, firms might switch underwriters due to concerns about the initial investment bank’s performance in the IPO process (the underwriter performance hypothesis).

The graduation hypothesis is examined in Table 7. For each firm that switches, we examine measures of bank reputation at the time of the successful offering both for the bank used on the initial attempt and for the bank to which the issuer switched. The graduation hypothesis predicts that switches occur in order to move up to banks with greater reputation. In Panel A of Table 7 we examine a sample of 98 firms that returned for a successful IPO with a different investment bank. At the time of the successful offering, the original bank has an average Carter-Manaster rank of 7.3 whereas the replacement bank has an average Carter-Manaster of 7.6. The difference between these two means is not significant (median measures for the two groups of banks are also not significantly different). At the time of the successful offering, the original bank has an average market share of 3.6% whereas the replacement bank has a market share of 4.9%; again, this difference is not statistically significant (although the difference in medians is significant). While these findings are not consistent with graduation, the industry market share evidence is more supportive. At the time of the successful offering, the original bank has an average industry market share of 2.3% whereas the replacement bank has an industry market share of 15.7%. The difference is significant at the 1% level (the difference in medians is also highly significant). The importance of industry market share supports evidence presented previously on factors affecting withdrawal. In that analysis, industry market share is by far the most significant (economically) factor affecting offering success. After experiencing an IPO failure, issuers appear to attempt to change banks to maximize their chance of success.

While not explored previously, several firms attempt to go public a second time and yet fail. In Panel B of Table 7, we examine a small sample of 15 issuers that attempt to go public a second time with a new bank and fail. The Carter-Manaster ranking for the bank originally attempting to take the firm public is 7.9 compared to 6.8 for the new bank. The difference is significant at the 10% level. This suggests that the firms’ inability to obtain a more reputable bank on the second attempt hurt its ability to successfully complete an IPO. Industry market share is also lower for the new bank, although differences are not statistically significant. Market share increases after the switch (the new bank arguably does not have a large average market share, however). Overall, this analysis suggests that the ability to attract a more reputable bank on the second attempt has an impact on offering success for reissuers.

In Table 8, we focus on successful returners to examine implications of the underwriter performance hypothesis. The measure of performance we consider is the amount of capital to be raised (the filing size).

ARTICLE IN PRESS

Table 7

Tests of the graduation hypothesis

This table presents descriptive statistics on variables for a subsample of IPOs that are withdrawn and later return for a successful or

failed IPO using a different investment bank from the first attempt. IPO market share is the sum of gross proceeds (not including the

overallotment option) over the year prior to the IPO of all offerings where the IPO book manager is the book manager (equal credit given

if there is more than one manager) divided by the sum of gross proceeds on all IPOs over the same period. Market share is defined as of the

date of the filing leading up to the second attempt. IPO industry market share is defined similarly to overall IPO market share where only

issues in the IPO firm’s Fama-French industry (see Fama and French, 1997) are considered. IPO industry market share is defined as of the

date of the filing leading up to the second attempt. Carter-Manaster rank is the Carter-Manaster (1990) ranking on a 0–9 scale for the

book manager of the IPO (the maximum rank if there is more than one book manager).

Initial bank on first failed

filing (at time of second

attempt)

New bank on

subsequent filing

Change p-value from 2-tailed t-test

or Wilcoxon (H0: initial

bank ¼ second bank)

Panel A-All withdrawn IPOs that subsequently return for a successful offer with a different bank (98 observations)

IPO market share

Mean 3.608 4.863 1.254 0.120

Median 0.846 2.823 0.901 0.004

IPO industry market share

Mean 2.327 15.690 13.363 0.001

Median 0.000 7.212 5.528 0.000

Carter Manaster ranking

Mean 7.288 7.575 0.287 0.193

Median 8.100 8.100 0.000 0.127

Panel B-All withdrawn IPOs that subsequently return for an unsuccessful offer with a different bank (15 observations)

IPO market share

Mean 1.024 2.641 1.617 0.062

Median 0.141 1.269 0.536 0.064

IPO industry market share

Mean 3.649 1.552 �2.097 0.333

Median 0.325 0.000 0.000 0.313

Carter Manaster ranking

Mean 7.963 6.833 �1.130 0.077

Median 8.100 7.100 �1.000 0.094

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 625

Better-performing banks should be able to successfully raise more capital for an issuer. For issuers switching due to bank performance concerns, we should observe significant increases in the attempted issue size (i.e., the issuer believes the initial bank should have been successful given the issue size and switches to another bank it believes would be more capable of successfully raising more capital). For issuers not switching, we should see less significant changes in the attempted issue size.

Consistent with these predictions, successful returners not changing banks attempt to raise about the same amount on the second attempt (a 6% mean increase and 0% median increase) whereas successful returners switching banks attempt to raise substantially more capital on the second attempt (a 140% mean increase and a 28% median increase). Differences in the rate of change are significant at the 5% level (differences in median percent change in filing size are significant at approximately the 1% level).

Percent changes in filing sizes could lead to confounding conclusions if switchers and non-switchers are clustered in periods where ‘‘normal’’ changes to filing sizes differ. To account for this possibility, we define ‘‘abnormal’’ filing size as the difference between the filing size on this attempt and the average filing size on all filings over the six months leading up to this attempt (similar findings emerge when we consider other measures based on industry filings). The abnormal percent change in filing size is then defined as the abnormal filing size on the second attempt divided by the abnormal filing size on the first attempt, subtract one.

ARTICLE IN PRESS

Table 8

Tests of the underwriter performance hypothesis

This table presents descriptive statistics on variables for a subsample of IPOs that are withdrawn and later return for a successful second

attempt. Subsamples are examined based on whether the issuer changes investment banks from the first failed attempt to the eventual

successful offering. The variables examined are defined as follows. IPO filing size—initial unsuccessful filing equals the average filing price

(average of the high and low price indicated in the initial filing) multiplied by the number of shares to be sold as indicated in the initial

(unsuccessful) filing, reported in millions of dollars. IPO filing size – successful IPO equals the average filing price (average of the high and

low price indicated in the initial filing) multiplied by the number of shares to be sold as indicated in the initial filing for the successful IPO.

Percent change in filing size is defined as the filing size in the successful attempt divided by the filing size in the initial attempt, less one.

Abnormal filing size is defined as the filing size less the average filing size in IPOs from the same Fama-French industry over the six months

leading up to the filing (if there are fewer than four industry IPOs, the average filing size for all IPOs is used). Abnormal percent change in

filing size is defined as the abnormal filing size in the successful attempt divided by the abnormal filing size in the initial attempt, less one.

All dollar amounts are converted to January 2000 dollars using the CPI.

All withdrawn IPOs

that subsequently

return for a successful

offer

Successful returning

IPOs where the

investment bank is

switched from initial

unsuccessful filing

Successful returning

IPOs where the

investment bank is not

switched

p-value from 2-tailed

t-test or Wilcoxon

(H0: switchers ¼ non-

switchers)

Mean percent change in filing size 107.184 139.730 6.180 0.051

p-value from 2-tailed t-test (H0:

%t change ¼ 0)

0.038 0.040 0.553

Median percent change in filing size 10.577 28.122 0.000 0.011

percentage positive 67.2 72.2 50.0

Mean abnormal percent change in

filing size

68.263 96.196 �18.427 0.078

p-value from 2-tailed t-test (H0:

%t change ¼ 0)

0.164 0.112 0.132

Median abnormal percent change in

filing size

1.229 6.723 �19.316 0.097

percentage positive 51.3 55.6 37.9

Number of observations 119 90 29

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635626

Accounting for ‘‘normal’’ changes in filing size reduces the significance (economically) of unadjusted findings but our conclusions remain unchanged. The mean abnormal percent change in filing size for successful returners having no bank change is �18% whereas the mean abnormal percentage change in filing size for successful returners with a change in bank is 96% (differences in medians are similar economically and statistically).

We interpret this evidence as being consistent with the underwriter performance hypothesis. We conjecture that replaced banks attempt to raise ‘‘too little’’ on the failed attempt. The issuer recognizes this as poor performance and decides to switch to a bank that is able to raise more money for the firm. Banks for non- switchers initially attempt to raise an appropriate amount of money and on return they are successful at that level. The issuer undoubtedly interprets the first failure as the result of factors beyond the bank’s control, such as poor market conditions.

An alternative view is that the evidence in Table 8 indirectly supports the graduation hypothesis. Firms that perform well after the initial withdrawal are likely to raise more capital. Since they are stronger firms, they are able to switch to a more highly ranked bank. This possibility is examined more directly in Table 9. This table replicates the analysis in Table 8 for successful returners that changed banks, breaking down the evidence on filing size changes based on whether the new bank has higher or lower measures of reputation. We consider there to be a positive ‘‘graduation’’ if the new bank has a higher market share or Carter-Manaster rank. Most industry market share changes are positive, however, so we define a positive graduation to have occurred if the new bank has a 3% higher industry market share than the old bank (other arbitrary cutoffs yield similar results). Generally, the mean and median percentage change in filing size (and abnormal change in filing size) is larger when the replacement bank has a greater measure of reputation. For example, switching to a bank

ARTICLE IN PRESS

Table 9

Interactions between the graduation and underwriter performance hypotheses

This table presents descriptive statistics on variables for subsamples of IPOs that are withdrawn and later return for a second successful

attempt using a different investment bank. The independent variables examined are defined as follows. Mean percentage change in IPO

filing size is the IPO filing size (average of the high and low price indicated in the initial filing multiplied by the number of shares to be sold

as indicated in the initial filing) for the second successful filing divided by the IPO filing size on the first unsuccessful attempt, less one.

Abnormal filing size is defined as the filing size less the average filing size in IPOs from the same Fame-French industry over the six months

leading up to the filing (if there are fewer than four industry IPOs, the average filing size for all IPOs is used). Abnormal percent change in

IPO filing size is defined as the abnormal filing size in the second attempt divided by the abnormal filing size in the initial attempt, subtract

one. All dollar amounts are converted to January 2000 dollars using the CPI. Subsamples are examined based on changes in investment

bank reputation measured upon change in underwriter. Reputation measures are defined as follows. IPO market share is the sum of gross

proceeds (not including the overallotment option) over the year prior to the IPO of all offerings where the IPO book manager is the book

manager (equal credit given if there is more than one manager) divided by the sum of gross proceeds on all IPOs over the same period.

Market share is defined as of the date of the filing leading up to the second attempt. IPO industry market share is defined similarly to

overall IPO market share where only issues in the IPO firm’s Fama-French industry (see Fama and French, 1997) are considered. IPO

industry market share is defined as of the date of the filing leading up to the second attempt. Carter-Manaster rank is the Carter-Manaster

(1990) ranking on a 0–9 scale for the book manager of the IPO (the maximum rank if there is more than one book manager). a, b, c denote

significance of mean and median observations (under the hypothesis that change in filing size equals 0) at the 1%, 5%, and 10% level

(based on a two-tailed t-test for means and Wilcoxon test for medians).

Median

percentage

change in IPO

filing size

Mean

percentage

change in IPO

filing size

Median

abnormal

percentage

change in IPO

filing size

Mean abnormal

percentage

change in IPO

filing size

Number of

obser-

vations

IPO market share increases 50.000 b

181.376 b

14.035 127.407 63

IPO market share does not increase 4.167 42.557 �1.153 23.372 27

– p-value from 2-tailed t-test or median test (H0:

group with market share increase ¼ group with

market share decrease)

0.001 0.065 0.358 0.306

IPO industry market share increases 3% or more 0.462 180.779 b

22.169 133.137 c

67

IPO industry market share increases less than 3% 2.171 20.154 c

�12.322 �11.414 23

– p-value from 2-tailed t-test or median test (H0:

group with large positive industry market share

change ¼ group with less positive change)

0.146 0.080 0.146 0.063

Carter-Manaster rank increases 55.708 b

242.784 c

23.038 170.944 42

Carter-Manaster rank does not increase 8.760 49.558 b

0.038 30.792 48

– p-value from 2-tailed t-test or median test (H0:

group with Carter-Manaster increase ¼ group with

no Carter-Manaster increase)

0.001 0.180 0.291 0.340

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 627

with a higher market share is associated with a percentage change in filing size of 181%, whereas switching to a bank with a lower market share leads to a percentage change in filing size of only 43%. The difference is significant at the 10% level. This is consistent with the view that better-performing firms that can increase offering sizes tend to seek out stronger banks. It is also consistent with the graduation and performance hypotheses being complements. When dissatisfied with the performance of their lead bank on the first attempt, successful returners switch to banks with greater industry presence who are able (due to their reputation and market power) to raise significantly more capital on the second attempt. While changes in filing sizes are more positive for firms changing to more reputable banks, they are also generally positive for firms not moving to more reputable banks (although the mean and median abnormal filing size changes are never statistically significantly positive). For example, the mean percentage change in filing size for firms moving to banks with equal or lower Carter-Manaster ranks is a significantly positive 50%. This suggests that some concern regarding bank performance likely plays a role in the decision to switch banks.

Overall, we conclude that the evidence in this section indicates that underwriter performance and graduation are complementary explanations for investment bank switching. Even a modest importance of underwriter performance is not consistent with the Krigman, Shaw, and Womack’s (2001) evidence for a sample of firms switching banks after a successful IPO. We believe our findings complement Krigman, Shaw,

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635628

and Womack’s to give a broader impression of the importance of different roles played by investment banks in the IPO process.

6. Likelihood of return, investment bank switching and the decision to withdraw an IPO

As documented previously, approximately 20% of all IPO filings are withdrawn prior to completion. Of those withdrawals, only 9% manage to return for a successful offering. If going public through an IPO is an important part of a company’s strategy for long-term success, it is reasonable to expect that most firms canceling an IPO hope to return. All else equal, we would therefore expect that the probability of successful return should be an important factor in a firm’s choice to withdraw its IPO. Issuers more likely to return should be more likely to withdraw.

To examine this hypothesis, we generate the predicted probability of successful return for all issuers using the first model estimates from Table 6 (i.e., we use an ex ante model of successful returning where all variables are known when the firm faces the decision to withdraw or proceed with the IPO). Specifically, the probability of return is F(bx), where x is a vector of independent variables, b is a vector of probit estimates, and F(b) is the standard cumulative normal distribution function. This variable is then added to the list of independent variables considered previously in Table 3 to predict withdrawal. Again, since we are trying to estimate a probability model for withdrawal as of the time of withdrawal, market condition variables are defined at that point.

The results are in Table 10. The effects of various variables previously considered on withdrawals are similar (in sign and in economic and statistic significance) to those reported previously. The probability of successful return variable is significantly positive, as expected. A one standard deviation increase in the likelihood of successful return results in a 21% increase in the probability of withdrawal, making this variable among the most economically significant variables in the model. The market condition variables at time of withdrawal are not comparable to findings in Table 3 since the time periods over which variables are measured are not the same. These variables are measured post-filing, however, so they are more comparable to the market conditions after filing variables considered previously. These comparisons suggest that the effects of variables are similar in the two analyses.

We formally examine the relation between withdrawal on a second attempt and investment bank switching by estimating a probit model of withdrawal for second-time filers. Given the relatively small sample size, we attempt to develop a parsimonious base model using independent variables considered previously. In Table 10 we report a model using variables that are most significant (economically or statistically). The effects of various independent variables on withdrawal are similar to that for the full sample of first-time filers, although statistical significance is reduced in most cases. Two new variables are added to the model to capture the effects of investment bank switching. The first is a dummy variable equal to one if there is a bank switch. The second is a variable equal to the change in industry market share (measured at the time of the filing on the second attempt) from the first to the second bank (if there is no bank change, this variable is set to zero, so it can be considered an interactive variable with the bank change dummy).

The change in bank dummy variable has a significantly positive effect on withdrawals. The change in industry market share variable has a significantly negative effect. Second-time unsuccessful filers are more likely to switch but (as noted previously) tend to switch to banks with less significant industry presence. This suggests that these are firms that are rejected by their initial bank and unsuccessful in their attempt to find a more prominent bank to take them public. Successful second-time filers switch banks only if they can attract a bank with significantly greater industry stature, which increases their chance of success.

7. Withdrawals, returners, and the pricing of IPOs

In this section we examine the pricing of successful IPOs from 1985 to 2000. Specifically, we examine price adjustments made during the bookbuilding process and first-day returns (underpricing) for successful IPOs. We examine whether the likelihood of withdrawal and the possibility of a successful return affect pricing behavior for successful offerings. BBG present a model where issuers can use the threat of withdrawal as leverage with investors in the bookbuilding process to illicit truthful information. In their model, firms more

ARTICLE IN PRESS

Table 10

The probability of return, bank switching, and the choice to withdraw an IPO for filings between 1985 and 2000

The dependent variable equals one for IPO filings that are withdrawn and zero for completed offerings. Issuer and issue characteristic

independent variables are defined as follows. Logarithm of the filing size equals the natural logarithm of the average filing price multiplied

by the number of shares to be sold as indicated in the initial filing (reported in January 2000 dollars using the CPI as a deflator).

Technology dummy takes the value one if the issuer is in Fama-French industries 34 (business services) or 36 (chips) and zero otherwise (see

Fama and French, 1997). Venture capital backing dummy takes the value one if the issuing firm has received venture capital investments

prior to filing and zero otherwise. Investment bank characteristic independent variables are defined as follows. Carter-Manaster rank is the

Carter-Manaster (1990) ranking on a 0-9 scale for the book manager of the IPO (the maximum rank if there is more than one book

manager). Bank market share is the sum of gross proceeds (not including the overallotment option) over the year prior to the IPO of all

offerings where the IPO book manager is the book manager (equal credit given if there is more than one manager) divided by the sum of

gross proceeds on all IPOs over the same period. Bank industry market share is the sum of gross proceeds over the year prior to the IPO of

all IPOs in the same Fama-French industry where the IPO book manager is the book manager divided by the sum of gross proceeds on all

industry IPOs over the same period. Market condition independent variables at the time of the issue/withdrawal are defined as follows.

Number of filings in prior two months is the number of IPOs filed with the SEC during the two months prior to the issue/withdrawal date for

the IPO. Number of industry filings in prior two months is the number of IPOs in the same Fama-French industry filed with the SEC during

the two months prior to the issue/withdrawal date for the IPO. BAA-AAA yield spread at issue/withdrawal is the spread between BAA and

AAA corporate bonds (from Moody’s) on the day of the issue/withdrawal. Ten-year Treasury yield at issue/withdrawal is the average yield

on US Treasury bonds having ten years to maturity measured on the day of the issue/withdrawal. Return on Nasdaq Composite Index from

filing to issue/withdrawal is the compound return on the index from filing date to issue/withdrawal date. Industry average book-to-market

pre-issue/withdrawal is the book-to-market ratio for firms in the IPO issuer’s Fama-French industry at the end of the year prior to issue/

withdrawal (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). Firm and bank characteristic variables at issue/

withdrawal are defined as follows. Probability of successful return if withdrawn is estimated using the first probit model from Table 6 (i.e.,

no variables are included that are measured after withdrawal). Investment bank change dummy equals one if the offering is a second attempt

and the bank used on the second attempt is different than that used on the first attempt. Change in bank industry market share if bank is

changed is equal to the difference between the industry market share of the banks on the second and first attempts in cases where the bank

has changed and the offering is a second attempt. Marginal effect is defined as f(bx)�b�sx where f() is the standard normal probability density function, b is the coefficient estimate, x is the mean of the independent variable for the sample, and sx is one standard deviation for the independent variable (sx is set to 1 for dummy variables). Pseudo R

2 is defined as 1 less the log likelihood for the estimated model

divided by the log-likelihood for a model with only an intercept as an independent variable.

All IPO filers Second time filers

Coefficient Marginal

effect

t-stat Coefficient Marginal

effect

t-stat

Intercept �1.125 �0.238 �5.21 3.203 0.008 1.92

Issuer and issue characteristics

Logarithm of the filing size 0.049 0.021 1.57 �0.677 �0.009 �1.54

Technology dummy �0.376 �0.149 �5.16 �2.149 �0.726 �2.30

Venture capital backing dummy �0.496 �0.195 �8.52 �0.874 �0.314 �1.62

Investment bank characteristics

Carter-Manaster rank �0.004 �0.004 �0.34

Bank market share �0.012 �0.028 �1.98

Bank industry market share �0.033 �0.241 �10.66 �0.044 �0.329 �1.00

Market conditions at the time of issue/withdrawal

Number of filings in prior two months �0.006 �0.068 �7.76 �0.020 �0.039 �2.19

Number of industry filings in prior two months 0.011 0.056 4.27 0.058 0.280 2.30

BAA-AAA yield spread at issue/withdrawal �0.075 �0.007 �0.60

Ten-year Treasury yield at issue/ withdrawal 0.094 0.037 3.70

Return on Nasdaq composite index from filing to issue/withdrawal �2.038 �0.103 �11.84 �7.151 �0.419 �3.06

Industry average book-to-market pre-issue/withdrawal �0.060 �0.006 �0.54

Firm and bank characteristics at issue /withdrawal

Probability of successful return if withdrawn 5.568 0.209 21.08 3.833 0.162 1.76

Investment bank change dummy 1.599 0.143 2.70

Change in bank industry market share if bank is changed �2.642 �0.168 �3.49

Pseudo R 2

0.290 0.628

Number of observations 6613 134

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 629

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635630

likely to withdraw need not underprice as severely in response to positive information. BBG do not explicitly address the effect of withdrawal likelihood on price adjustments. If the threat of withdrawal allows issuers to reduce required underpricing, they should be able to more fully adjust prices during bookbuilding in response to positive information. This suggests that price adjustments should be positively related to the likelihood of withdrawal (more accurately, price adjustments should be non negatively related to the likelihood of withdrawal as it is possible that the negative impact of withdrawal likelihood on underpricing is reflected in a higher initial filing price). In contrast, we posit that firms more likely to withdraw that also face a lower probability of successful return are more likely to cut prices in an attempt to assure success on the first attempt. Price adjustments, therefore, should be more negative, all else equal, for these issuers. Relatedly, first- day returns could be more positive for these firms if price cutting is deeper than necessary to ensure a successful offering.

To test these predictions, we first estimate ordinary least square models of price adjustments. Hanley (1993) presents the first analysis of price adjustments made between the initial filing of regulatory documents with the SEC and the approval of the offering when the bank sets a final price. She argues that price adjustments proxy for information acquired during bookbuilding. When information revealed is positive (negative), price revisions are positive (negative).

Benveniste, Ljungqvist, Wilhelm, and Yu (2003) examine price adjustments and argue that price adjustments should be related to proxies for information spillovers (from both market factors and more industry-related measures). They also argue that price adjustments should be related to deal and market riskiness (information is more likely to be uncovered in more speculative offerings and in riskier periods). Our model of price revision, therefore, includes fairly standard measures of deal riskiness including venture capital backing (a variable taking the value one if the IPO is venture capital backed as indicated by TFSD and zero otherwise; see Barry, Muscarella, Peavy, and Vetsuypens, 1990; Megginson and Weiss, 1991), the exchange listing (NYSE, which is a dummy variable taking the value one if the IPO lists on the New York Stock Exchange, and AMEX, which equals one if the IPO lists on the American Stock Exchange; see Lowry and Schwert, 2004), firm standard deviation (the standard deviation of stock returns from days +21 to +50 relative to the IPO; see Johnson and Miller, 1988; Carter, Dark, and Singh, 1998; Lowry and Schwert, 2004) and overhang (the number of shares outstanding after the IPO net of the number of shares offered in the IPO, all divided by the number of shares offered in the IPO where the number of shares in the IPO is obtained from TFSD and includes global tranches but exclude the overallotment option). Overhang can be thought of as a liquidity measure but also captures insider retention. Both are related to deal riskiness (see Bradley and Jordan, 2002). We also consider the market standard deviation (the standard deviation of daily returns from day -50 to -2 relative to the IPO on the CRSP value-weighted index) as a measure of market risk in our models but this variable is not significant. Consistent with the previous research we include Carter-Manaster ranking as a measure of investment bank reputation.

We control for variables likely to be associated with information spillovers in our regressions. Market return, defined as the compound return from day �50 to day �2 relative to the IPO on the CRSP value- weighted index, is included as an independent variable in our analysis. The specification for the market return in this regression is chosen to match that used by Lowry and Schwert (2004). To allow for non linearities in the relation between market returns and price adjustment, we also include market return

+ as an independent

variable where market return + takes the same value as market return whenever it is positive, and zero otherwise

(see Loughran and Ritter, 2004; Lowry and Schwert, 2004). As an additional measure of pre-IPO market activity (and information spillovers) we include in our regressions number of prior IPOs, the number of IPOs from days �60 to �1 relative to the offering, and number of prior industry IPOs, the number of industry IPOs over the same period (see Booth and Chua, 1996; Benveniste, Ljungqvist, Wilhelm, and Yu, 2003).

To examine the impact of the likelihood of withdrawal and return on price adjustments, we construct a variable that reflects the ex ante probability of withdrawing and not returning. Specifically, predicted probabilities of withdrawal and returning are estimated using the ex ante probit models in Section 6. The variable probability HW-LR (i.e., high probability of withdrawal and low probability of return) is equal to the predicted probability of withdrawal multiplied by one minus the predicted probability of returning.

The ordinary least square estimates are reported in Table 11. The dependent variable, price adjustment, is defined as the final offering price minus the average of the high and low initial filing prices all divided by the

ARTICLE IN PRESS

Table 11

Price adjustments and initial returns for IPOs between 1985 and 2000

The dependent variables are defined as follows. Price adjustment is the IPO offer price divided by the average of the high and low initial

filing price. Initial return is defined as 100�(P1–P0)/P0 where P1 is the first-day closing stock price or bid-ask average (from CRSP) and P0 is the IPO offer price. Independent control variables are defined as follows. Overhang is (S1–S)/S where S1 is the shares outstanding after

the IPO and S is the shares offered in the IPO. Venture capital backing dummy equals one if the issue is venture capital-backed and zero

otherwise. NYSE equals one if the IPO lists on the New York Stock Exchange and zero otherwise. AMEX equals one if the IPO lists on the

American Stock Exchange and zero otherwise. Firm std. deviation equals the standard deviation of daily stock returns for the issuing firm

from days 21 to 50 relative to the IPO. Market return equals the buy and hold CRSP value-weighted index return from days �50 to �2

relative to the IPO. Market return + equals the buy and hold CRSP value-weighted index return from days �50 to �2 relative to the IPO if

positive, and zero otherwise. Lagged avg. underpricing is the average initial return for issues on days�60 to�1 relative to the IPO. Number

of prior IPOs is the number of issues from days �60 to �1 relative to the IPO. Number of prior industry IPOs is the number of issues in the

same Fama-French industry over the year prior to the IPO. Carter-Manaster rank is the Carter-Manaster (1990) ranking on a 0–9 scale for

the book manager of the IPO (the maximum rank if there is more than one book manager). Independent variables indicating effects of

price adjustments and withdrawals are defined as follows. The variable probability HW-LR is generated using predictions from ex ante

probit models of withdrawal and returning. Probability of successful return if withdrawn is estimated using the first probit model from

Table 6 (i.e., no variables are included that are measured after withdrawal). Probability of withdrawal is estimated using the first probit

model from Table 3 (i.e., excluding the debt retirement variable). The variable probability HW-LR in this regression equals the probability

of withdrawal multiplied by one minus the probability of return. In the initial return regression, price adjustment, as defined above, is also

an independent variable. The variable price adjustment + is the price adjustment variable when that variable is positive and zero otherwise.

Independent variables indicating whether the IPO was previously withdrawn are defined as follows. Previously withdrawn IPO dummy

takes the value one if the IPO was from a firm that previously attempted and withdrew an offering. Previously withdrawn IPO with change

in bank takes the value one if the IPO was from a firm that previously attempted and withdrew an offering and is using a different

investment bank on a successful offering. Previously withdrawn IPO with no change in bank takes the value one if the IPO was from a firm

that previously attempted and withdrew an offering and is using the same investment bank on a successful offering.

Price adjustment Initial return

Coefficient t-stat Coefficient t-stat Coefficient t-stat Coefficient t-stat

Control variables

Intercept �0.110 �8.12 �0.110 �8.13 �2.619 �1.58 �2.622 �1.59

Overhang 0.015 11.26 0.015 11.23 1.974 9.69 1.970 9.67

Venture capital backing dummy 0.010 1.56 0.011 1.59

NYSE 0.017 1.75 0.017 1.70 �2.942 �1.99 �3.053 �2.06

AMEX �0.068 �4.39 �0.068 �4.38 �4.426 �1.86 �4.416 �1.86

Firm std. deviation 0.015 10.60 0.015 10.59 2.258 9.81 2.269 9.85

Market return 0.012 6.29 0.012 6.27

Market return +

�0.005 �2.37 �0.005 �2.36

Lagged avg. underpricing 0.407 15.92 0.405 15.83

Number of prior IPOs �0.001 �6.41 �0.001 �6.35

Number of prior industry IPOs 0.001 9.27 0.001 9.27

Carter-Manaster rank 0.004 2.80 0.004 2.79 �0.984 �4.94 �0.982 �4.93

Variables indicating effects of price adjustments and withdrawals

probability HW-LR �0.156 �3.30 �0.155 �3.29 �2.313 �0.26 �2.179 �0.25

Price adjustment 20.336 3.88 20.240 3.86

Price adjustment +

116.785 16.42 116.829 16.43

Price adjustment � probability HW-LR �100.771 �1.91 �99.992 �1.90 Price adjustment

+ � probability HW-LR �489.148 �6.57 �487.577 �6.55 Variables indicating whether IPO was previously withdrawn

Previously withdrawn IPO dummy �0.024 �1.30 1.966 0.69

Previously withdrawn IPO with no change in bank 0.017 0.48 12.155 2.20

Previously withdrawn IPO with change in bank �0.039 �1.81 �1.671 �0.50

Adjusted R 2

0.131 0.131 0.463 0.463

Number of observations 5564 5564 5564 5564

C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 631

average of the high and low initial filing prices. Overhang is positively related to price adjustments, as found by Bradley and Jordan (2002). Consistent with prior research, offerings backed by venture capital and those underwritten by banks with higher Carter-Manaster ranks tend to have more positive price adjustments

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635632

during the bookbuilding process. Price adjustments are also more positive for riskier offerings (proxied by the firm’s standard deviation of returns) and when market and industry information spillovers are likely to be positive (proxied by the market return and the number of prior industry IPOs).

The variable probability HW-LR, reflecting the likelihood of withdrawal and return, is negatively related to price adjustments. All else equal, issuers are more likely to cut their offering price if they believe (as of the time of issue or withdrawal) that they have a strong chance of failing and little chance of getting a second opportunity. This finding is consistent with our prediction regarding the effect of withdrawal and return likelihood on price adjustments, although it is not consistent with BBG whose theory only makes predictions regarding the effect of withdrawal probability. As a robustness check, therefore, we include the probabilities of withdrawal and return as separate variables. In this specification both probabilities are significantly negative. The negative relation between withdrawal probability and price adjustments is not consistent with BBG.

To examine the impact of withdrawal and return probabilities on IPO initial returns, we estimate ordinary least squares initial return regressions. The dependent variable, IPO initial return, is defined as 100�(P1st day close–Poffer)/Poffer, where P1st day close is the closing price at the end of the first day of trading and Poffer is the offering price from TFSD. Control variables are motivated based on the existing literature. All the independent variables used to explain price adjustments are also included here. As an additional measure of pre-IPO market activity we include lagged average underpricing, the mean first-day returns on all IPOs on days �60 to �1 relative to the offering, as an independent variable (see Loughran and Ritter, 2004; Lowry and Schwert, 2004; Bradley and Jordan, 2002). Consistent with prior research we also include variables reflecting price adjustments in our underpricing models. Specifically we include price adjustment, as defined previously, and price adjustment

+ . The latter variable equals the actual price adjustment if it is positive and zero otherwise.

This specification recognizes the differential impact of good and bad news on underpricing. We also include variables to reflect predictions regarding the effect of withdrawal and return probability on

initial returns. BBG argue that the threat of withdrawal allows issuers to reduce expected underpricing. We therefore include probability HW-LR as an independent variable. BBG predict that this variable is negatively related to initial returns. More accurately, BBG only predict a negative relation between initial returns and the probability of withdrawal. In unreported results, we include the probabilities of withdrawal and return as separate variables and find that the sign and significance of the probability HW-LR variable is similar to that for the probability of withdrawal; we report only the more parsimonious specification. BBG also argue that the impact of withdrawal likelihood on initial returns should be more pronounced in high demand states and they include a variable that interacts the probability of withdrawal and positive price adjustments. Consistent with this argument, we include the interactive variable price adjustment

+ � probability HW-LR in our model. BBG would predict that this variable is negatively related to initial returns (issuers more likely to withdraw in heavy demand states can more fully adjust prices given positive information).

In contrast to BBG, our paper focuses on the response of issuers to negative information. As discussed previously, firms more likely to withdraw (and not return) that receive negative information during bookbuilding are more likely to deeply discount prices to ensure offering success. This price cutting should result in a higher initial return if issuers reduce prices more than necessary. To capture this argument, we include the interactive variable price adjustment � probability HW-LR in our initial return regression. Since we have a separate variable to capture the impact of positive price adjustments, the price adjustment variable reflects the impact of negative price adjustments. We expect this interactive variable to also be negatively related to initial returns (as probability HW-LR increases, we expect deepening price cutting to result in more positive initial returns).

The initial return regressions are reported in Table 11. It should be noted that price adjustment � probability HW-LR is the one new variable relative to what has been considered in the literature. The relation between control variables and initial returns is consistent with the prior literature. Overhang, firm standard deviation, and lagged average underpricing have a positive impact on initial returns. NYSE and AMEX listing and Carter-Manaster ranking have a negative impact on initial returns. The effect of expected price adjustments on initial returns is consistent with the literature. Both price adjustment and price adjustment

+ are positively

related to initial returns. The coefficient on probability HW-LR is insignificantly negative. The coefficient on price adjustment

+ � probability HW-LR is significantly negative, consistent with BBG. The coefficient on price adjustment �

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 633

probability HW-LR is also significantly negative, consistent with our prediction. Overall our results suggest that both BBG’s arguments and our arguments help explain the pricing of initial public offerings. Issuers that are more likely to withdraw can use that leverage to lower expected initial returns when good news is revealed. When bad news is revealed, however, issuers choose to deeply discount prices to ensure offering success, resulting in higher initial returns.

The sample of successful IPOs examined in Table 11 includes first-time issuers and some that are successful on their second attempt. We add additional variables to our regressions to see whether IPO pricing is different for second-time issuers. In the context of previously withdrawn IPOs, we would expect price adjustment behavior to be different than that observed for first-time issuers. Specifically, we expect price adjustments to be less positive, all else equal, for two reasons. First, issuers would be more averse to failure and, therefore, willing to leave more ‘‘money on the table’’ to ensure success. In the face of positive valuation information revealed in the pre-market, issuers would not respond by adjusting price as significantly as first-time issuers. Second, return issuers would be perceived as riskier by investors (given the ‘‘lemons’’ problem referred to previously). Benveniste and Spindt (1989) argue that price adjustments should be lower (and first-day returns higher) for speculative offers since information is more costly to acquire (and more valuable) so investors need greater returns to induce truth telling.

In the first price adjustment regression of Table 11, we add a dummy variable that takes the value one if the IPO was previously withdrawn. This variable is negative, consistent with expectations, but not significant. In the second price adjustment regression of Table 11, we add two dummy variables. The first takes the value one if the offering was previously withdrawn and brought forward by the same bank that helped the firm make the initial attempt. The second takes the value one if the offering was previously withdrawn and brought forward by a new bank. We have no strong prior beliefs about which group should have more negative price adjustments. If a firm uses the same bank, that bank could be very nervous about failing twice and therefore would be more likely to aggressively cut prices. Also, those firms attracting new banks might benefit from the renewed certification that comes from being able to attract a new intermediary. This suggests that price cutting could be deeper if there is no change in bank. On the other hand, the evidence reported previously indicates that the initial filing size for second-time issuers experiences a larger jump than observed for issuers that do not change banks. The new banks might have been overly aggressive in their attempt to win the business and, therefore, are more likely to have to reduce prices during the offering process.

The evidence in Table 11 is more consistent with the second story. Price adjustments are significantly negative (at the 10% level) if the offering was previously withdrawn and then brought forward by a different bank. Price adjustments are not significantly unusual controlling for other factors, however, if there is no bank change.

We also examine initial returns for second-time issuers in Table 11. Firms with previously withdrawn IPOs might be viewed as riskier by investors, controlling for other observable characteristics, since the withdrawal creates a ‘‘lemons’’ problem. In the first initial return regression of Table 11 we include the dummy variable that takes the value one if the IPO was previously withdrawn. This variable is positive, consistent with the risk factor story, but not significant. In the second initial return regression, we include two dummy variables. The first takes the value one if the offering was previously withdrawn and brought forward by the same bank that helped the firm make the initial attempt. The second takes the value one if the offering was previously withdrawn and brought forward by a new bank. One benefit of attracting a new bank is that it creates incremental certification for the issue, reducing lemons concerns. The evidence is consistent with this view. The coefficient on the dummy variable that takes the value one if the offering was previously withdrawn and brought forward by a new bank is insignificantly negative. The coefficient on the dummy variable taking the value one if the IPO was previously withdrawn and brought forward by the same bank is significantly positive.

8. Conclusions

This study adds to our understanding of the decision to withdraw an IPO. This decision is puzzling since so few firms ever return to the market for a successful IPO. We study returning issuers and find that firms initially brought forward by more reputable investment banks, and those having venture capital backing, are more likely to return. In other words, issues that have more ex ante certification have a better chance of surviving

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635634

the negative event of a withdrawal. Market conditions at the time of the withdrawal and afterwards also have an impact on an issuer’s ability to return. Issues withdrawn in more active IPO markets, when interest rates are high and when market returns are low, are more likely to be able to return.

Since the likelihood of returning is predictable, we examine whether this likelihood affects the firm’s decision to withdraw. We find that the probability of withdrawal is positively related to the likelihood of a successful return. Issuers that face the decision to withdraw but do not expect a second chance are more likely to try to push forward and complete their IPO. The likelihood of withdrawal and the possibility of return also have an impact on the pricing of successful IPOs. In order to ensure success, firms expected to withdraw with a low chance of returning cut prices during the bookbuilding process. This price cutting results in higher first- day returns, suggesting that firms with a low probability of a second chance are willing to leave more money on the table to ensure success.

Overall, the evidence in this paper indicates that firms consider the costs of withdrawal when attempting to decide whether or not to proceed with an IPO. Firms not likely to get a second chance are more likely to push forward, even though this might require that the issuer cut prices more substantially than would be expected.

Our sample of previously withdrawn IPOs also provides a unique context in which to investigate underwriter switching after a withdrawal but before a successful IPO, complementing the existing literature on switching after a successful IPO but before a subsequent equity offering. We find that issuers ‘‘graduate’’ to banks with larger industry market shares when they deem that their initial bank did not perform well on the first attempt (they attempted a small offering and did not succeed). This evidence provides a more balanced view of the role of graduation and performance than previous research, which diminished the importance of the performance story but examined cases where performance was arguably good. This study, therefore, also provides insights into the role played by investment banks in the process of raising capital through an IPO.

References

Akerlof, G., 1970. The market for ‘lemons’: quality and the market mechanism. Quarterly Journal of Economics 84, 488–500.

Barry, C., Muscarella, C., Peavy, J., Vetsuypens, M., 1990. The role of venture capital in the creation of public companies: evidence from

the going public process. Journal of Financial Economics 27, 447–471.

Bates, T., Dunbar, C., 2002. Investment bank reputation, market power, and the pricing and performance of IPOs. Unpublished working

paper. University of Arizona and University of Western Ontario.

Benveniste, L., Ljungqvist, A., Wilhelm, W., Yu, X., 2003. Evidence of information spillovers in the production of investment banking

services. Journal of Finance 58, 577–608.

Benveniste, L., Spindt, P., 1989. How investment bankers determine the offer price and allocation of new issues. Journal of Financial

Economics 24, 343–361.

Booth, J., Chua, L., 1996. Ownership dispersion, costly information, and IPO underpricing. Journal of Financial Economics 41, 291–310.

Booth, J., Smith, R., 1986. Capital raising, underwriting and the certification process. Journal of Financial Economics 15, 261–281.

Bradley, D., Jordan, B., 2002. Partial adjustment to public information and IPO underpricing. Journal of Financial and Quantitative

Analysis 37, 595–617.

Busaba, W., Benveniste, L., Guo, R.-J., 2001. The option to withdraw IPOs during the premarket: empirical analysis. Journal of Financial

Economics 60, 73–102.

Carter, R., Dark, F., Singh, A., 1998. Underwriter reputation, initial returns, and the long-run performance of IPO stocks. Journal of

Finance 53, 285–311.

Carter, R., Manaster, S., 1990. Initial public offerings and underwriter reputation. Journal of Finance 45, 1045–1067.

Cho, S.-I., 1992. The possibility of failure and the pricing of best-efforts initial public offerings. Quarterly Review of Economics and

Finance 31, 30–45.

Clarke, J., Dunbar, C., Kahle, K., 2001. Long-run performance and insider trading in completed and canceled seasoned equity offerings.

Journal of Financial and Quantitative Analysis 36, 415–430.

Daniel, K., 2002. Discussion of ‘‘Why don’t issuers get upset about leaving money on the table in IPOs?’’. Review of Financial Studies 15,

445–454.

Dunbar, C., 1998. The choice between firm-commitment and best-efforts offering methods in IPOs: the effect of unsuccessful offers.

Journal of Financial Intermediation 7, 60–90.

Dunbar, C., 2000. Factors affecting investment bank initial public offering market share. Journal of Financial Economics 55, 3–41.

Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics 43, 153–193.

Gompers, P., 1996. Grandstanding in the venture capital industry. Journal of Financial Economics 42, 133–156.

Grinblatt, M., Hwang, C.-Y., 1989. Signalling and the pricing of new issues. Journal of Finance 44, 393–420.

ARTICLE IN PRESS C.G. Dunbar, S.R. Foerster / Journal of Financial Economics 87 (2008) 610–635 635

Hanley, K., 1993. The underpricing of initial public offerings and the partial adjustment phenomenon. Journal of Financial Economics 34,

231–250.

James, C., 1992. Relationship-specific assets and the pricing of underwriter services. Journal of Finance 47, 1865–1885.

Johnson, J., Miller, R., 1988. Investment banker prestige and the underpricing of initial public offerings. Financial Management 17, 19–29.

Krigman, L., Shaw, W., Womack, K., 2001. Why do firms switch underwriters? Journal of Financial Economics 60, 245–284.

Lerner, J., 1994. Venture capitalists and the decision to go public. Journal of Financial Economics 35, 293–316.

Loughran, T., Ritter, J., 2002. Why don’t issuers get upset about leaving money on the table in IPOs? Review of Financial Studies 15,

413–443.

Loughran, T., Ritter, J., 2004. Why has IPO underpricing changed over time? Financial Management 33, 5–37.

Lowry, M., Schwert, G.W., 2004. Is the IPO Pricing Process Efficient? Journal of Financial Economics 71, 3–26.

Megginson, W., Weiss, K., 1991. Venture capitalist certification in initial public offerings. Journal of Finance 46, 879–903.

Mikkelson, W., Partch, M., 1988. Withdrawn security offerings. Journal of Financial and Quantitative Analysis 23, 119–133.

Ritter, J., 1987. The costs of going public. Journal of Financial Economics 19, 269–281.

Seguin, P., Smoller, M., 1997. Share price and mortality: an empirical evaluation of newly listed Nasdaq stocks. Journal of Financial

Economics 45, 333–363.

Titman, S., Wessels, R., 1988. The determinants of capital structure choice. Journal of Finance 53, 1935–1974.

Welch, I., 1992. Sequential sales, learning, and cascades. Journal of Finance 47, 695–732.