ISSUER OPERATING PERFORMANCE AND IPO PRICE FORMATION. Michael Willenborg University of Connecticut

Transcription

1 ISSUER OPERATING PERFORMANCE AND IPO PRICE FORMATION Michael Willenborg University of Connecticut Biyu Wu University of Connecticut March 14, 2014

2 ISSUER OPERATING PERFORMANCE AND IPO PRICE FORMATION ABSTRACT We extend the study of the association between public information and IPO price formation by examining Brau and Fawcett s (2006) chief financial officer survey response that having strong historical earnings is the most important signal of quality an IPO issuer can convey to investors. For a sample of IPOs, we find measures of pre-ipo operating performance explain a substantial portion of the variation in the revision from the mid-point of the initial price range to the IPO offer price. Moreover, for these recent IPOs, the partial adjustment phenomenon concentrates among issues with strong performance; whereas for those with weak performance, the downward price adjustment is nearly full. As for why issuers with strong performance seem to acquiesce to a partial increase in offer price, these IPOs have the largest wealth gains from shares sold / retained upon going public. Overall, our results are consistent with Loughran and Ritter s (2002) agency / prospect theory explanation for IPO partial adjustment and suggest an important role for historical accounting information in the pricing of book-built IPOs. Keywords: Operating performance, initial public offerings, partial adjustment, underpricing Data Availability: Public sources

3 I. INTRODUCTION When a company files to go public in a firm commitment initial public offering (IPO), in either its initial registration statement or a subsequent amendment, it must provide an initial price range within which it expects to sell its shares. Following this disclosure, representatives of the company and its lead underwriters meet with select investors to obtain feedback and indications of interest to price and pre-sell the offering. Subsequent to this road-show period, the offer price decision is made the day before the shares begin to trade. Then, as is well known, on average, the offer price is lower than the first-day closing price. This price formation of IPOs, from initial price range to offer price to closing price on the first trading day, is a topic of longstanding interest. While the latter return (underpricing) garners the majority of attention, understanding the former (price revision) is arguably the key. For example, Ritter and Welch (2002, p. 1803) conclude the solution to the underpricing puzzle has to lie in focusing on the setting of the offer price, where the normal interplay of supply and demand is suppressed by the underwriter. In an influential paper that links setting the offer price with underpricing, Benveniste and Spindt (1989) propose a book-building theory of IPO pricing and allocation by underwriters. In their model, to induce regular (e.g., road-show) investors to divulge indications of strong demand, underwriters only partially revise the range upward to arrive at the offer price. As a result, underpricing of IPO shares, along with discretionary allocation of them, compensates investors for truthful revelation of favorable private information. Consistent with this, Hanley (1993) was the first to document the strong, positive association between the price revision and first-day underpricing (i.e., the partial-adjustment phenomenon). More recently, Loughran and Ritter (2002) posit an agency / prospect theory explanation for the partial adjustment phenomenon. They argue that if issuers anchor on the midpoint of filing range and offset the loss of company proceeds from underpricing on primary shares with their wealth revaluation on retained / secondary shares, they will acquiesce to under-adjustment of the offer price in response to strong demand. Given this emphasis on the covariance between money left on the table and unexpected wealth changes, Loughran and Ritter (2002, p. 438)

4 assert issuers bargain hard over the offer price in a bad state of the world, whereas they are pushovers in bargaining over the offer price in a good state of the world. Importantly, their explanation makes no distinction between private and public information. In support of this, they show that market returns during the 15 days pre-ipo are positively associated with both the price revision and initial returns and that this association is considerably stronger when returns are negative (i.e., they document partial (nearly full) adjustment to favorable (unfavorable) public information). In his discussion of Loughran and Ritter (2002), Daniel (2002, p. 453) poses [s]ome questions that merit further investigation include [t]o what extent are revisions predictable using information available as of the preliminary prospectus date? From an empirical standpoint, one way to address Daniel s (2002) question, and further our understanding of IPO price formation, is to focus on the price revision and advance proxies that plausibly surrogate for favorable or unfavorable public information. In Brau and Fawcett s (2006) survey of chief financial officers (CFOs), all categories of respondents chose having strong historical earnings as the most important positive signal issuers convey to investors regarding the value of a firm going public. In this paper, we study the relation between pre-ipo operating performance and IPO price formation, with a focus on the revision from the mid-point of the initial price range to the offer price. We report evidence of a strong, positive association between pre-ipo operating performance and the price revision; a relation apparent in descriptive statistics, robust to multiple regression and consistent with an identification strategy wherein we substitute stale versions of our performance measures. Overall, these findings are supportive of an important pricing role for the historical accounting information in the IPO prospectus. We study firm commitment, share IPOs by domestic companies from January 2001 to December At average underpricing of 13.2%, these IPOs sustain a comparable fraction of money left on the table as the pre-bubble samples of Loughran and Ritter (2002) and Lowry and Schwert (2004). Using the most-recent financial statements in the final prospectus, we compute three measures of pre-ipo operating performance: operating income to average assets; net income to average assets; and operating cash flow to average assets. Descriptively each of 2

5 these measures is strongly positively associated with both the price revision and underpricing, particularly in the tails of the distribution (e.g., whereas IPOs in the lowest decile of operating income, net income or operating cash flow have () price revisions of 22% ( 21%) and () underpricing of 5% (1%); those in the highest decile have () price revisions of +5% (+8%) and () underpricing of +22% (+19%)). This suggests an asymmetric association between operating performance and IPO offer price formation, in that downward (upward) price adjustments for IPOs with weak (strong) operating performance are almost full (partial). Within Loughran and Ritter s (2002) framework, this is consistent with the view that weak (strong) pre-ipo operating performance represents unfavorable (favorable) public information that engenders weak (strong) demand and motivates issuers to bargain aggressively (passively) with the underwriter over the IPO offer price at the pricing meeting. The positive association between pre-ipo operating performance and offer price revision is robust to controlling for other determinants (e.g., book-building market returns, ownership retention, selling shareholders and whether the issuer is in a high technology industry). Moreover, specification of each of our operating performance variable of interest substantially increases the explained variation in price revision over-and-above that of a baseline estimation. We also adopt an identification strategy to assess whether our results weaken upon substituting older versions of our variables of interest. To do this, we restrict the sample for our offer price revision regressions to those IPOs with both year-end and interim financial statements in their final prospectus. These are issuers for which the year-end statements have gone stale. This occurs, per Securities and Exchange Commission (SEC) rules, when the time between the date of the year-end financial statements and the effective date of the registration statement exceeds 134 days (e.g., for a December 31 year-end issuer, this occurs at the close of business on May 14 th ). The coefficients on our operating performance variables of interest are smaller and the regression explanatory power is lower when we use older, year-end financial statements in place of more-recent, interim financial statements to calculate our variables of interest. 3

6 To refine our test of Brau and Fawcett s (2006) CFO survey response, which specifically references strong historical earnings as a signal of quality, we also parse our continuous pre- IPO operating performance variables into a series of indicator variables based on decile ranking. Upon regressing price revision on these indicators, we find a near-monotonic increase in the coefficients on these indicator variables of interest as they ascend decile ranks. The association between pre-ipo operating performance and initial returns, while also positive and significant, is much weaker than with the price revision. For example, when we supplement a benchmark underpricing regression, which controls for the price revision, with operating income to assets the adjusted R-squared increases from 39.2% to 40.0%. We conclude that to the extent pre-ipo operating performance influences IPO price formation, it is primarily with respect to the revision from mid-point of initial price range to offer price. To examine why issuers with strong performance seem to acquiesce to a partial upward adjustment of offer price, we follow Loughran and Ritter (2002) and compute the change in pre- IPO shareholder wealth. Two components comprise this revaluation: the change from mid-point of the initial price range to offer price for the shares insiders sell at the IPO; and the change from mid-point of the initial price range to closing price on the first trading day for the shares that shareholders retain. For issuers in the highest decile of pre-ipo performance, the revaluation in pre-ipo shareholders wealth exceeds money left on the table by several orders of magnitude. Therefore, one explanation for why issuers with strong performance agree to a partial adjustment of the initial price range in response to strong demand is they anchor on the mid-point of the price range and offset the company s loss of proceeds with their gain in wealth from revaluation. Combining with Brau and Fawcett s (2006) CFO survey with Loughran and Ritter s (2002) agency / prospect theory of partial adjustment, our findings are consistent with strong (weak) pre-ipo operating performance signaling favorable (unfavorable) public information regarding firm value. In the favorable scenario, underwriters, who prefer to market underpriced shares (Baron 1982; Loughran and Ritter 2002), exploit issuers satisfaction with their positive wealth revaluation by partially adjusting the IPO offer price. Issuers, happy with the increase 4

7 vis-à-vis the mid-point of the initial range, acquiesce to this partial adjustment at the pricing meeting. In contrast to this, the unfavorable scenario, issuers bargain aggressively at the pricing meeting and, as such, little money is left on the table. Overall, our paper contributes by providing evidence suggestive of an important role for historical accounting information in the price formation of book-built IPOs. Previous papers focus on the valuation of IPOs, oftentimes by studying subsets of the population, and generally find historical accounting information to be of relatively little importance (e.g., Kim and Ritter 1999; Bartov, Mohanram and Seethamraju 2002; Berger 2002). We extend the study of the importance of accounting information in the context of IPOs by focusing on the price formation, from mid-point of the initial range to offer price to first closing price. II. BACKGROUND AND MOTIVATION In this section, we discuss the literature on partial price adjustment and underpricing of book-built IPOs. Two primary streams comprise this literature: one emphasizes informational issues and book building and another emphasizes agency issues and bargaining incentives. We then discuss the role of accounting information in the valuation and price formation of IPOs. Investors, underwriters and private information One stream focuses on investors and underwriters and concludes that book building can improve pricing accuracy by facilitating the revelation or acquisition of private information. Benveniste and Spindt (1989) model the pre-market interaction between underwriters and regular (e.g., road show) investors as a single-price auction wherein investors bid by indicating non-binding interest at different prices which underwriters use to construct a demand curve to set the price. In contrast to indications of weak demand, which should require little inducement, underwriters keep prices intentionally low to reward investors for truthful revelation of strong demand. Following this, partial offer price adjustment to favorable demand, in concert with preferential allocation of shares, compensates investors for providing private information that 5

8 allows the underwriter to more-accurately price the issue. Underpricing therefore results from the partial offer price adjustment necessary to ensure the revelation of strong demand from regular investors is incentive compatible. Among Benveniste and Spindt s (1989, p. 353) implications is that [u]nderpricing is directly related to the level of interest in the premarket. 1 Several empirical papers provide results consistent with this dynamic, information-based view of IPO pricing. Hanley (1993) reports a strong, positive relation between the IPO price revision (from mid-point of the initial price range to offer price) and first-day returns. Based on a small, yet detailed sample of international equity issues, Cornelli and Goldreich (2001) find that investors providing bids with both share and price information receive favorable allocations from underwriters. Ljungqvist and Wilhelm (2002) study a large, worldwide sample for which they have share allocation data and conclude the data more strongly support the view that allocations promote price discovery, as opposed to representing discriminatory practices. Issuers, underwriters and public information A second stream focuses on agency issues between issuers and underwriters, such as the former s bargaining incentives or non-price aspects of their objective function. Loughran and Ritter (2002) apply a prospect theory framework (Kahneman and Tversky 1979) and argue that if issuers anchor on the filing range and offset the loss of company proceeds from underpricing primary shares with the positive wealth revaluation on retained or secondary shares, they will acquiesce at the pricing meeting to partial adjustment of the offer price in response to strong demand. Following this focus on the covariance between money left on the table and unexpected wealth changes, if bookbuilding demand is favorable (unfavorable), due to either private or public information, issuers acquiesce (negotiate) at the pricing meeting. In support of this, they show that public information in the form of market returns during the 15 1 In contrast to the revelation of private information Chemmanur (1993) models the issuer s decision to underprice the IPO in order to compensate investors for information production and several papers (Benveniste and Wilhelm (1990); Sherman (2000); Sherman and Titman (2002) model book-building / underpricing as way to incentivize investors to incur the costs of becoming informed (see also Yung 2005). 6

9 days pre-ipo are positively associated with both the price revision and initial returns and that this association is stronger when market returns are negative. 2 Loughran and Ritter (2002) interpret the asymmetric association between book-building market returns and price revision as consistent with a theory of bargaining, as they assert that issuers will aggressively (passively) bargain over the offer price when roadshow demand is unfavorable (favorable). Habib and Ljungqvist (2001) and Ljungqvist and Wilhelm (2003) posit that underpricing is partly a function of the extent to which owners care about it. These papers focus on the relation between proxies for issuer incentives to bargain for a higher offer price and IPO pricing. For example, Ljungqvist and Wilhelm (2003) examine a sample of domestic IPOs from 1996 to 2001 and conclude that decreases in CEO ownership and insider selling along with an increase in directed-share programs (i.e., allocation of shares to friends and family ) explain a large portion of the variation in underpricing during the internet bubble. For a sample of domestic IPOs from , Loughran and Ritter (2004) test three non-mutually exclusive explanations for the time-series variation in underpricing: 1) a change in issuer risk composition (Ritter 1984); 2) realignment of issuer bargaining incentives (Ljungqvist and Wilhelm 2003); and 3) a change in issuer objective function away from maximizing IPO proceeds. They conclude the latter, specifically issuer desire to garner coverage from top stock analysts ( analyst lust ) and underwriter allocations of hot IPOs to executives of yet-to-go-public companies ( spinning ), best explains the underpricing during Of note, subsequent legal and regulatory intervention limits both of these drivers of bubble-period underpricing. Lowry and Schwert (2004) conclude that neither the mid-point of the initial price range nor the IPO price fully reflect public information, which they suggest may stem from an implicit contract between underwriter and issuer to limit revisions. Consistent with prior literature, they document an asymmetric association between book-building market returns and price revision; as the downward price revision to negative returns is almost one-for-one, but the upward price 2 Bradley and Jordan (2002) also show that initial returns are associated with the initial returns of previous IPOs. 7

10 revision to positive returns is much less than this. They also question the significance of publicly available variables prior to disclosure of the initial price range to explain the price revision and conclude the IPO market is almost efficient with respect to public information. Because IPO failure is costly, Edelen and Kadlec (2005) conclude partial adjustment to public information, and its asymmetric response to good and bad news, is partly an artifact of the selection bias inherent in using samples of successful IPOs. They argue that firms proceed with IPOs when positive market responses occur but, to ensure a successful offering and the issuer surplus it entails, only partially adjust price upward; in contrast, when negative market responses occur, because of the costs of withdrawal, firms more-fully adjust price downward. Accounting information and the pricing of IPOs Several papers examine the relation between accounting information in the prospectus and the pricing or valuation of IPOs. In general, this literature documents little role for historical accounting measures of operating performance. Klein (1996) studies the valuation usefulness of prospectus financial statement variables for a sample of 193 IPOs from 1980 to 1991 with positive pre-ipo income. She documents a positive association between the offer price and the market price one week after the offering and both pre-ipo earnings per share and book value per share. Kim and Ritter (1999) study the association between issuer and industry multiples (e.g., price-earnings ratios) for a sample of 190 IPOs from with positive pre-ipo income. They report positive, albeit statistically weak, associations between issuer and industry multiples (e.g., upon regressing issuer P/E on industry- P/E, the adjusted-r 2 is just 5.0%). Kim and Ritter (1999, page 424) conclude historical accounting numbers are of limited value for understanding IPO valuation and that the performance of the comparable firms approach is surprisingly weak. Bartov, Mohanram and Seethamrau (2002) study of the association between IPO prices and earnings, operating cash flows and sales for samples of 98 Internet IPOs and 98 matched non-internet IPOs during Their specification partitions earnings and operating cash 8

11 flows by whether they are positive or negative. For Internet IPOs, their findings provide no support for an association between earnings and IPO prices but rather strong evidence that negative operating cash flows are associated with higher offer prices. For non-internet IPOs, they find offer prices are positively related to earnings and cash flows, but only for issuers for which these variables are non-negative. Akin to Kim and Ritter s (1999) conclusion, Berger (2002, page 348) summarizes Bartov et al. s (2002) findings as [t]he results point to a very limited pricing role for the financial statement data contained in the IPO prospectus. A recent paper by Brau and Fawcett (2006) motivates re-visiting the relation between accounting measures of operating performance and IPO pricing. They survey three categories of chief financial officers, those with companies: that went public; that began to go public but withdrew; large enough to go public but have not done so. Among their questions is [w]hat type of signal do the following actions convey to investors regarding the value of a firm going public? (italics in original) All groups of respondents chose having strong historical earnings as the most important positive signal issuers convey to investors regarding the value of a firm going public. We extend the study of the association between public information and IPO price formation, as Loughran and Ritter (2002) theorize, by examining this CFO survey response. III. SAMPLE, VARIABLE SPECIFICATION AND DESCRIPTIVE STATISTICS Sample Panel A of Table 1 presents our sample. Using Thompson Financial s SDC database, we begin by identifying 1,826 firm-commitment IPOs by stand-alone (non-carve out) US companies from January 1, 2001 through December 31, This period covers a dozen, post-ipo bubble years and, for the most part, begins where those of Loughran and Ritter (2002), Ljungqvist and Wilhelm (2003) and Lowry and Schwert (2004) end. 3 We focus on share IPOs by non-financial companies by eliminating 790 IPOs by financial companies (i.e., SDC SIC 6xxx; most of which 3 Loughran and Ritter s (2002) sample is ; Ljungqvist and Wilhelm s (2003) sample is ; and Lowry and Schwert s (2004) primary sample is though, for certain analyses, it extends through As such, our sample is least comparable to Ljungqvist and Wilhelm s (2003) IPO bubble period sample. 9

12 are unit investment trusts), and 113 unit IPOs, because they are mostly small offerings by small companies. We eliminate 71 IPOs without necessary financial statement information in their registration statement, most of which are issuers that do not provide two balance sheets to calculate average assets for the year before going public. Since our interest is price formation of book-built IPOs, we eliminate 15 IPOs sold via auction (Degeorge, Derrien and Womack 2010). 4 We also eliminate 7 IPOs for which the time between the date of the registration statement with the initial price range and the IPO date exceeds one year (Edelen and Kadlec 2005) and 6 IPOs with a mid-point of their initial price range of $5 or less. Our final sample consists of 824 IPOs. Variable specification and descriptive statistics Panel B of Table 1 presents descriptive statistics and Table 2 presents correlations for all dependent and independent variables. Mean () PriceRevision (change from the midpoint of the initial price range to the IPO offer price), in percentage terms, is 5.2% (0.0%). This average is more negative than the 1.4% Lowry and Schwert s (2004) report for their pre-bubble period sample, and much less than the +5.8% Ljungqvist and Wilhelm (2003) report for their bubble period sample. 5 Mean () InitialReturn (change from the IPO offer price to the closing price on the first trading day), in percentage terms, is 13.2% (8.3%). This average is similar to Loughran and Ritter s (2002) 14.1% and Lowry and Schwert s (2004) 12.3% but, not surprisingly, far below Ljungqvist and Wilhelm s (2003) average underpricing of 35.7%. Our three operating performance variables are: OI/Assets (annualized operating income, per the most-recent pre-ipo income statement, divided by average assets), NI/Assets (annualized net income, per the most-recent pre-ipo income statement, divided by average assets) and OCF/Assets (annualized operating cash flow, per the most-recent pre-ipo statement of cash flows, divided by average assets). All of these variables have negative s and positive s. As might be expected, because of the effects of special items, interest expense and 4 See W.R. Hambrecht s website at 5 Loughran and Ritter (2002) do not report descriptive statistics for price revision. However, per their Table 2, because the fraction of their sample with offer prices below the minimum of the initial range exceeds that above the maximum (27.3% and 24.3%, respectively), the average price revision for their sample seems likely negative. 10

13 income taxes, average () OI/Assets is less negative (more positive) than average () NI/Assets. Of note, all three of these measures have a considerable amount of dispersion; for example, NI/Assets has a standard deviation of around a of Not surprisingly, this dispersion is particularly evident at the negative tail of each variable s distribution. 6 In terms of covariates, we specify two spillover variables for public information during book building (Ljungqvist and Wilhelm 2003). The first is MktReturn, the equally-weighted return on all companies in CRSP residing in the issuer s Fama and French (1997) industry from the date of the registration statement with the initial price range to IPO date. When specifying our regressions, we allow for the likelihood that positive secondary market returns affect the price revision differently than negative market returns (Loughran and Ritter 2002; Lowry and Schwert 2004; Edelen and Kadlec 2005). The second is IPOReturn, the average initial return for all IPOs during the period from the date of the first registration statement to IPO date. 7 We also control for ownership retention and insider selling. The average () issuer retains 71.0% (73.5%) ownership (i.e., Retain%, one minus the number of shares sold in the IPO divided by the number of post-ipo shares outstanding). While the majority of our IPOs do not have secondary shares, the average SellingShr% (number of selling shareholder shares divided by number of total shares in the IPO) is 16.7%. Leland and Pyle (1977) theorize that higher values of Retain% and SellingShr% are positive and negative signals, respectively, of firm value to potential investors. In addition, these two variables also proxy for IPO issuer incentives to bargain at the IPO pricing meeting (Loughran and Ritter 2002; Ljungqvist and Wilhelm 2003). Following this, a positive (negative) association between Retain% and PriceRevision is consistent with a signaling (agency) story and a negative (positive) association between SellingShr% and PriceRevision is consistent with a signaling (agency) story. 6 For example, for Ventrus Biosciences Inc. (a company that went public on December 16, 2010), OI/Assets, NI/Assets and OCF/Assets are 5.802, and , respectively. 7 In specifying IPOReturn, to retain observations, we compile initial returns starting with the date of an issuer s first prospectus. If we were to start from the date of an issuer s prospectus that discloses the initial price range, we would lose observations because, in numerous instances, no companies went public during this shorter time frame. 11

14 The motivation for our other covariates (e.g., UW, VC, Proceeds, Age, HighTech, Assets) is extant IPO papers on price revision and initial returns, notably Ljungqvist and Wilhelm (2003) and Lowry and Schwert (2004). Following Lowry and Schwert (2004), we adjust Proceeds and Assets for inflation, converting them to 1983 dollars with the Consumer Price Index. Table 2 shows correlations, Pearson (Spearman) below (above) the diagonal. Each operating performance measure is positively correlated with PriceRevision and InitialReturn. The measure with the highest Pearson (Spearman) correlation with PriceRevision is OI/Assets (OCF/Assets). Of note, consistent with the dispersion in these measures, Spearman correlations between dependent variables (PriceRevision or InitialReturn) and independent variables of interest (OI/Assets, NI/Assets, OCF/Assets) are more positive than their Pearson counterparts (e.g., the Pearson (Spearman) correlation between PriceRevision and NI/Assets is (0.309)). Because of extreme values, particularly in the left tail of the distribution, we winsorize OI/Assets, NI/Assets and OCF/Assets at + / 1% for our regression estimations. Table 3 provides descriptive sorts of our sample by: OI/Assets (panel A), NI/Assets (panel B) and OCF/Assets (panel C). For each panel, we sort the variable of interest by decile and show and values of PriceRevision and InitialReturn. It is interesting to compare the extremes of these sorts. For issuers in Decile 1, IPOs with very negative operating income, net income or operating cash flow, () PriceRevision ranges from 23.2% to 21.2% ( 21.8% to 20.0%) and () InitialReturn ranges from +4.2% to +6.4% (+0.5% to +1.1%). In contrast, for issuers in Decile 10, IPOs with very positive operating income, net income or operating cash flows (i.e., per Brau and Fawcett s survey, with strong historical earnings), () PriceRevision ranges from +4.6% to +6.1% (+6.7% to +9.7%) and () InitialReturn ranges from +21.8% to +24.2% (+17.6% to +21.7%). Within the context of Loughran and Ritter (2002), these findings are consistent with the view that weak (strong) pre-ipo operating performance represents unfavorable (favorable) public information that engenders weak (strong) roadshow demand and motivates issuers to bargain aggressively (passively) with the underwriter at the pricing meeting. 12

15 Taken together, these findings suggest the association between pre-ipo operating performance and IPO price formation is asymmetric. That is, for those issuers with very weak (very strong) performance that go public, the downward (upward) price adjustment is almost full (partial). This asymmetry is akin to that between book-building market returns and the price revision (Loughran and Ritter 2002; Lowry and Schwert 2004; Edelen and Kadlec 2005) in that bad ( good ) news is nearly fully (partially) associated with downward (upward) price revision. IV. EMPIRICAL ANALYSIS Issuer operating performance and price revision We begin by regressing each measure of pre-ipo operating performance on the IPO price revision. 8 IPO year. 9 These estimations supplement the Table 2 correlations by clustering standard errors by Given the presence of extreme values among these measures, we impart a winsor of +1% and 1%. Because successful IPOs comprise our sample, the coefficients we report are contingent upon the offering being completed (Loughran and Ritter 2002; Ljungqvist and Wilhelm 2003; and Lowry and Schwert 2004) PriceRevision = β0 + β1oi/assets + ε PriceRevision = β0 + β1ni/assets + ε PriceRevision = β0 + β1ocf/assets + ε (1a) (1b) (1c) Where: PriceRevision OIAssets NI/Assets OCF/Assets = (IPO price mid-point of initial filing range) mid-point of initial filing range = Annualized operating income per most-recent pre-ipo financial statements Average assets = Annualized net income per most-recent pre-ipo financial statements Average assets = Annualized operating cash flow per most-recent pre-ipo financial statements Average assets 8 If we take a step back from studying the PriceRevision, and regress the mid-point of the initial price range on our measures of pre-ipo operating performance, we find the coefficient on each of them is strongly positive. When we then split these measures into separate variables depending on whether they are positive or negative, we find that the positive association stems from IPOs with negative OI/Assets, NI/Assets or OCF/Assets (i.e., issuers with poor pre- IPO operating performance have lower mid-points of their initial price range). Following Brau and Fawcett s (2006) survey response, our analysis focuses on investor reaction to pre-ipo operating performance as opposed to examining the determinants of the initial price range. 9 We do not assume companies that go public during a given time period have pricing residuals that are independent in cross-section. Because of this, for all regression estimations, we cluster standard errors by IPO year. 13

16 Columns 1 3 of Table 4 panel A report the results of estimating equations (1a), (1b) and (1c). The coefficient for each operating performance measure is positive and highly significant and their specification explains a substantial portion of the variation in PriceRevision. To specify a benchmark to assess the effect of issuer operating performance, we estimate equation (2) by regressing the PriceRevision on determinants from the literature. To control for public information that arises during book building, we follow Ljungqvist and Wilhelm (2003) and Edelin and Kadlec (2005) and specify MktReturn and IPOReturn as spillover, from secondary and primary markets, respectively. With regard to the former, because Loughran and Ritter (2002) and others document an asymmetric revision to positive versus negative market returns, we also specify MktReturn+, which equals MktReturn when it is positive, and zero otherwise. To control for the positive signal issuers convey by ownership retention (Leland and Pyle 1977; Brau and Fawcett 2006) or, alternatively, for lower issuer incentives to bargain for a higher IPO price (Loughran and Ritter 2002), we specify Retain%. To control for the negative signal issuers convey by selling secondary shares or, alternatively, for higher issuer incentives to bargain to increase the offer price, we specify SellingShr% (Ljungqvist and Wilhelm 2003). Following Lowry and Schwert (2004), we specify Ln(Proceeds) and NYSEAMEX as transaction characteristics. To control for effects associated with professional advisors / intermediaries, we specify UW, BigN and VC (Ljungqvist and Wilhelm 2003; Lowry and Schwert 2004; Edelin and Kadlec 2005; and Brau and Fawcett 2006). Lastly, we also specify Ln(Age), HighTech and Ln(Assets) as issuer characteristics (Lowry and Schwert 2004; Ljungqvist and Wilhelm 2003). Overall, equation (2) provides a stringent benchmark against which to compare the inclusion of our operating performance variables of interest in equations (3a), (3b) and (3c). PriceRevision = β0 + β1mktreturn + β2mktreturn + + β3iporeturn + β4retain% + β5sellingshr% + β6uw + β7bign + β8vc + β9ln(proceeds) + β10nyse/amex + β11ln(age) + β12hightech + β13ln(assets) + εi,t (2) Where: PriceRevision = (IPO price mid-point of initial filing range) mid-point of initial filing range 14

17 MktReturn = Average return on all companies in CRSP in issuer s Fama-French (1997) industry for the period from the date of issuer s prospectus with the initial price range to IPO date MktReturn+ = MktReturn when it is positive, and zero otherwise IPOReturn = Average initial return of IPOs from the date of issuer s first prospectus to IPO date Retain% = One less (number of shares sold in IPO number of post-ipo shares outstanding) SellingShr% = Number of shares sold by selling shareholders number of shares sold in IPO UW = Underwriter rank (Carter, et al., 1998; and Loughran and Ritter 2004) BigN = One if IPO issuer has a BigN audit firm, and zero otherwise VC = One if IPO issuer has venture capital backing, and zero otherwise Ln(Proceeds) = Natural logarithm of IPO proceeds (inflation adjusted to 1983 dollard) per issuer s prospectus with the initial price range (i.e., midpoint of initial price range times number of shares filed) NYSE/AMEX = One if IPO is listed on the NYSE or AMEX, and zero otherwise Ln(Age) = Natural logarithm of one plus the number of years from year of company founding or incorporation, if founding date is unavailable, to IPO year HighTech = One if IPO issuer is a high technology company per SDC, and zero otherwise Ln(Assets) = Natural logarithm of issuer s pre-ipo total assets, in millions (inflation adjusted) Column 4 of Table 4 panel A reports the results of estimating equation (2). The explanatory power, an adjusted R 2 of 12.7%, is between the 11% for Lowry and Schwert s (2004) sample and the 22% for Ljungqvist and Wilhelm s (2003) sample. As for covariates, consistent with the extant literature, the price revision to book-building market returns is more complete when they are negative. MktReturn s coefficient of implies that a market return during bookbuilding of 10% corresponds to a price revision of 9.18%. In contrast, the coefficient on MktReturn+ of suggests a book-building market return of +10% corresponds to a price revision of +1.66% ( ). 10 Consistent with Habib and Ljungqvist (2001) and Ljungqvist and Wilhelm (2003), the price revision to public information from the primary market and for issuer incentives to bargain for a higher IPO price (IPOReturn and SellingSh%, respectively), are both positive and significant. The coefficient on Retain% is positive and significant, consistent with the view that the equity that entrepreneurs retain signals their private information regarding firm value (Leland and Pyle 1977). As with Ljungqvist and Wilhelm (2003) and Lowry and Schwert (2004), the coefficient on HighTech is positive and significant. Lastly, consistent with its Table 2 correlation, the coefficient on Ln(Assets) is positive and significant, though this association is not robust to specifying our operating performance variables of interest in equations (3a), (3b) and (3c). 10 Edelen and Kadlec (2005) argue that this asymmetry, particularly the strength of the relation between negative market returns and price revision, is largely due to selection bias pertaining to an issuer s option to withdraw. 15

18 We then augment equation (2) by specifying our three pre-ipo operating performance variables in equations (3a), (3b) and (3c). PriceRevision = β0 + β1oi/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + εi,t PriceRevision = β0 + β1ni/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + εi,t PriceRevision = β0 + β1ocf/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + εi,t (3a) (3b) (3c) Columns 5 7 of Table 4 panel A present the results of estimating equations (3a), (3b) and (3c). In each case, the coefficient on the pre-ipo operating performance variable remains positive and significant and their specification substantially increases the explanatory power above that of equation (2) (e.g., specifying OCF/Assets in equation (3c) increases the adjusted R 2 from 12.7% for equation (2) to 19.5%). 11. As with the univariate regressions, the coefficient on OCF/Assets is the largest and most significant; and, at 0.147, implies a one standard deviation change in OCF/Assets is associated with a 6.8% increase in PriceRevision (0.147 * ). 13 We also adopt an identification strategy to assess whether our results weaken upon substituting older versions of our variables of interest. To do this, we restrict the sample for our price revision regressions to the majority (668 of 824, 81%) of IPOs with both year-end and interim financial statements in their final prospectus. These are issuers with year-end financial statements that have gone stale, ing the time period between the effective date of the registration statement and financial statements exceeds 134 days (e.g., for a registrant with a 11 We also estimate equations (1a), (1b), (1c), (2), (3a), (3b) and (3c) via ordered probit using, in place of the continuous variable PriceRevision, the five categories of price revision we show in Table 7. The results of these estimations are very similar to those we provide in Table In contrast to the in Table 1 panel B, the standard deviation for winsorized OCF/Assets is The findings in Table 3 and the positive coefficients on our pre-ipo operating performance variables contrasts with Roosenboom (2007), which finds French IPOs with higher ratios of forecasted earnings before interest and taxes to sales are associated with lower initial IPO price discounts. 16

19 December 31 year-end, this occurs at the close of business on May 14 th ). Given the strength of OCF/Assets in Table 4, we table the results of estimating equations (1c), (3c) and (1d), (3d); the latter two specify OCF/Assets Stale, which we compute using the older, year-end statements. PriceRevision = β0 + β1ocf/assets Stale + ε PriceRevision = β0 + β1ocf/assets Stale + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + εi,t (1d) (3d) Table 4 panel B reports the results. For both univariate and multivariate estimations, the coefficient on OCF/Assets Stale is smaller than that on OCF/Assets. For the multivariate regressions, the coefficient on OCF/Assets is (column four) whereas on OCF/Assets Stale it is (column five); a decrease of 22%. In addition, the regression s R 2 decreases from 18.9% to 15.7% upon substituting OCF/Assets Stale in place of OCF/Assets; a decrease of 16%. A Vuong (1989) likelihood ratio test to assess the incremental R 2 of Equation (3c) over that of (3d) yields a z-statistic of 1.89, with a p-value of These weaker results from specifying a stale version of pre-ipo operating performance in place of the most-recent version provide additional assurance of the positive association between operating performance and the revision from mid-point of the initial price range to IPO offer price. An alternative test of the association between pre-ipo operating performance and price revision is to specify a series of indicator variables using the Table 2 (OI/Assets, NI/Assets and OCF/Assets) decile in which a given IPO resides. In equations (4a), (4b) and (4c), we substitute each variable of interest with a series of nine indicators (the intercept captures Decile 1). This approach combines the descriptive statistics in Table 2 with the regressions in Table 4. PriceRevision = β0 + β1oi/assets Decile2 + β2oi/assets Decile3 + β3oi/assets Decile4 + β4oi/assets Decile5 + β5oi/assets Decile6 + β6oi/assets Decile7 + β7oi/assets Decile8 + β8oi/assets Decile9 + β9oi/assets Decile10 + β10mktreturn + β11mktreturn + + β12iporeturn + β13retain% + β14sellingshr% + β15uw + β16bign + β17vc + β18ln(proceeds) + β19nyse/amex + β20ln(age) + β21hightech + β22ln(assets) + εi,t (4a) 17

20 PriceRevision = β0 + β1ni/assets Decile2 + β2ni/assets Decile3 + β3ni/assets Decile4 + β4ni/assets Decile5 + β5ni/assets Decile6 + β6ni/assets Decile7 + β7ni/assets Decile8 + β8ni/assets Decile9 + β9ni/assets Decile10 + β10mktreturn + β11mktreturn + + β12iporeturn + β13retain% + β14sellingshr% + β15uw + β16bign + β17vc + β18ln(proceeds) + β19nyse/amex + β20ln(age) + β21hightech + β22ln(assets) + εi,t (4b) PriceRevision = β0 + β1ocf/assets Decile2 + β2ocf/assets Decile3 + β3ocf/assets Decile4 + β4ocf/assets Decile5 + β5ocf/assets Decile6 + β6ocf/assets Decile7 + β7ocfi/assets Decile8 + β8ocf/assets Decile9 + β9ocf/assets Decile10 + β10mktreturn + β11mktreturn + + β12iporeturn + β13retain% + β14sellingshr% + β15uw + β16bign + β17vc + β18ln(proceeds) + β19nyse/amex + β20ln(age) + β21hightech + β22ln(assets) + εi,t (4c) Table 5 reports the results of estimating equation (4a) in panel A, equation (4b) in panel B and equation (4c) in panel C. For presentation purposes, we show the results of the Table 4 panel A regression with the continuous version of each variable of interest (i.e., equations 3a, 3b and 3c). We suppress the control variables results, though they are consistent with Table 4. In general, across all three panels, the coefficients on the operating performance indicator variables increase with decile rank. In addition, consistent with Brau and Fawcett s (2006) CFO survey response that strong historical earnings is the most important signal of quality an IPO issuer can convey, for all estimations the coefficient on OCF/Assets Decile10 is the most positive. Overall, the results in Table 5 reinforce those in Tables 2, 3 and 4, and provide insight regarding the linearity of the relation between pre-ipo operating performance and the IPO price revision. Issuer operating performance and initial returns In this section, we examine the relation between issuer pre-ipo operating performance and initial returns (i.e., underpricing). As with the price revision regressions, we begin by regressing winsorized versions of each pre-ipo operating performance measure on initial returns. InitialReturn = β0 + β1oi/assets + ε InitialReturn = β0 + β1ni/assets + ε (5a) (5b) InitalReturn = β0 + β1ocf/assets + ε (5c) 18

21 Where: InitialReturn = (Closing price on the first day of trading IPO price) IPO price Columns 1 3 of Table 6 reports the results of estimating equations (5a), (5b) and (5c). As with the PriceRevision regressions in Table 4 Panel A, he coefficient for each measure of issuer pre-ipo operating performance is positive and significant. However, in contrast to the price revision estimations, the explanatory power of these underpricing regressions is considerably lower. For example, whereas the adjusted R-squared from regressing OCF/Assets on PriceRevision is 12.3%, it is just 4.3% upon regressing OCF/Assets on InitialReturn. To specify a benchmark to assess the effect of issuer pre-ipo operating performance, we estimate equation (6) by regressing the InitialReturn on the covariates in equation (2) plus the price revision. As with regard to the relation between secondary market returns during book building and the price revision, following Ljungqvist and Wilhelm (2003), we allow for an asymmetric relation between the IPO price revision and initial returns. InitialReturn = β0 + β1mktreturn + β2mktreturn + + β3iporeturn + β4retain% + β5sellingshr% + β6uw + β7bign + β8vc + β9ln(proceeds) + β10nyse/amex + β11ln(age) + β12hightech + β13ln(assets) + β14pricerevision + β15pricerevision + + εi,t (6) Where: PriceUpdate PriceUpdate+ = (IPO price mid-point of initial filing range) mid-point of initial filing range = PriceUpdate when it is positive, and zero otherwise Column 4 of Table 6 reports the results of estimating equation (6). The explanatory power, an adjusted R 2 of 39.3%, is consistent with literature that finds IPO underpricing is predictable (e.g., Bradley and Jordan 2002). The positive coefficient on Retain% is consistent with Leland and Pyle s (1977) signaling story and Loughran and Ritter s (2002) prediction that IPOs that sell a smaller percentage ownership should be more underpriced. 14 The positive coefficients on PriceRevision and PriceRevision+, and that the latter is larger than the former, is consistent with Ljungqvist and Wilhelm (2003). Whereas PriceRevision s coefficient of implies an IPO price adjustment of 10% corresponds to underpricing of 1.54%; the coefficient 14 The results in Table 6 are quantitatively very similar if we substitute a measure of overhang (i.e., the ratio of shares retained to shares offered) in place of %Retain (see Bradley and Jordan, 2002). 19

22 on PriceRevision+ of implies an IPO price adjustment of +10% corresponds to underpricing of +9.22% ( ). We then augment equation (6) by specifying our three pre-ipo operating performance variables of interest in equations (7a), (7b) and (7c). InitialReturn = β0 + β1oi/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + β15pricerevision + β16pricerevision + + εi,t InitialReturn = β0 + β1ni/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + β15pricerevision + β16pricerevision + + εi,t InitialReturn = β0 + β1ocf/assets + β2mktreturn + β3mktreturn + + β4iporeturn + β5retain% + β6sellingshr% + β7uw + β8bign + β9vc + β10ln(proceeds) + β11nyse/amex + β12ln(age) + β13hightech + β14ln(assets) + β15pricerevision + β16pricerevision + + εi,t (7a) (7b) (7c) Columns 5 7 of Table 6 present the results of estimating equations (7a), (7b) and (7c). While the coefficients on OI/Assets, NI/Assets and OCF/Assets remain positive and significant, specifying these variables results in small increases in explanatory power versus the equation (2) benchmark (e.g., specifying OI/Assets in equation (7a) increases the adjusted R-squared from 39.3% for equation (6) to 40.0%). We conclude that to the extent pre-ipo performance influences IPO price formation, it is primarily with respect to the change from the mid-point of the initial price range to IPO offer price and not from IPO offer price to the first closing price. Why might issuers with strong operating performance go along with a partial adjustment? While our findings are consistent with Brau and Fawcett s (2006) CFO survey response that having strong historical earnings is an important signal of IPO quality to investors, they beg the question of why such issuers seem to acquiesce to a partial upward adjustment of offer price. To examine this, we follow Loughran and Ritter (2002) and compare the amount of money left-on-the-table with the change in wealth for pre-ipo shareholders. Two components 20