VIEWS: 5 PAGES: 38 POSTED ON: 11/4/2011 Public Domain
IPO Information Aggregation and Underwriter Quality Wei Wang Leeds School of Business University of Colorado Wei.Wang@colorado.edu Chris Yung Leeds School of Business University of Colorado Chris.Yung@colorado.edu First Draft: July, 2006 Current Draft: Sept, 2007 Abstract A key distinction between some models of IPO pricing (e.g., auctions and bookbuilding) and others (e.g., fixed-priced models) is whether price discovery occurs in the primary market or the secondary market. We show that higher investment bank reputation is associated with 1) more active filing price revisions and 2) reduced secondary market return variability, both of which are consistent with primary market price discovery. Reputable underwriters exhibit partial adjustment to private information but virtually complete adjustment to information implicit in the returns of public comparables. Non- reputable underwriters partially adjust to all types of information. Taken together, this evidence suggests that theoretical models of primary (secondary) market information aggregation are better suited for reputable (non-reputable) underwriters. 1 Introduction IPO auction and bookbuilding models emphasize the economic importance of information flows from investors to the issuer and/or underwriter during the pre-IPO phase.1 The importance of these flows is motivated as follows. Information asymmetry among investors creates a winner’s curse, which prevents the seller from obtaining a high price (Rock, 1986). A rational response, therefore, is to design a mechanism to elicit this information before the price is set, thereby mitigating the winner’s curse. Any such mechanism entails costs. The question then becomes which is cheaper: designing a pre-IPO mechanism to extract investors’ information (Benveniste and Spindt, 1989) or simply bearing the adverse selection costs (Rock, 1986)? Or, equivalently: is price discovery is more efficient in the primary market or in the secondary market? We study primary market information aggregation by looking at price revisions, defined as the percentage change from the midpoint of the original filing range to the final IPO offer price. Because soliciting investor opinions before the initial filing range is set is prohibited by the Securities Act of 1933, the period between the filing and offering dates fully captures the impact of the underwriter’s bookbuilding efforts. The private information revealed during this period may be either good or bad news. Information aggregation thus leads to diverse movements in prices, i.e., cross-sectional volatility. Our primary methodology is to compare cross-sectional variation in price movement during this phase2 to that of the secondary market. By doing so, we hope to establish a measure of how much information is aggregated in primary markets rather than secondary market. Our first observation is that reputable underwriters revise their prices much more actively in the pre-IPO phase. This is a very robust and dramatic feature of our data. For underwriters ranked 8.5 or above on the Carter-Manaster scale, the standard deviation of price revisions is 28% of the midpoint of the initial filing price range, while for underwriters ranked below 6.75 it is only 17% of the filing price midpoint. This evidence 1 Much of the rest of the literature emphasizes information flows in the opposite direction. This category includes signaling models (e.g., Allen and Faulhaber (1989), Grinblatt and Hwang (1989) and Welch (1989)), certification models (e.g., Carter and Manaster (1990)) and models based upon moral hazard in selling effort (e.g., Baron (1982) and Baron and Holmstrom (1982)). 2 Public information is revealed during the bookbuilding period as well. In Section 4, we discuss how this public component is factored out in order to focus on private information. 1 is consistent with prestigious underwriters acquiring information much more actively during the bookbuilding phase and incorporating that information into the price. Price discovery in the primary market should substitute for price discovery in the secondary market. Hence, given the strong difference in primary market pattern mentioned earlier, one might expect reduced aftermarket volatility for IPOs taken public by reputable underwriters. In the full sample, the univariate evidence for this hypothesis is surprisingly weak. We do find lower secondary market cross-sectional return variability for reputable underwriters for every holding period horizon and in every sub- period we consider except 1996-2000. Yet this one sub-period rejection is sufficiently dramatic as to cause the hypothesis to fail in the full sample. This univariate evidence is of limited power, however, since reputable and non- reputable underwriters issue very different types of IPOs. Offerings by reputable underwriters are much more likely to be high-tech and venture capital backed, among other differences. We control for these differences in two ways. We employ the standard firm characteristic controls (size, age, VC backing, etc.) in our regressions. These controls are insufficient, however, if the asymmetric information profile of firms approaching high and low quality underwriters is different. Recall also that we are primarily interested in the relative return variability in the primary and secondary markets; that is, not how much learning is done in total, but rather when this learning tends to be done. We therefore form the following ratio: cross-sectional variability of price movements in the primary market (i.e., percentage change from the filing range midpoint to the offer price) divided by cross-sectional variability of secondary market returns. This ratio functions as an additional control for unobservable differences, because heightened ex-ante asymmetric information inflates both the numerator and the denominator. We find that reputable underwriters have lower variance ratios in every subperiod except 1996-2000. This lone reversal owes to the crash of tech stocks (disproportionately underwritten by reputable underwriters) near the end of this subperiod. Despite this outlier, reputable underwriters have much lower variance ratios (p-value < .001 for all holding periods studied) in the full sample, both unconditionally and controlling for firm characteristics. 2 1.1 Partial Adjustment to Private Information Benveniste and Spindt (1989) show that primary market information acquisition necessitates partial adjustment to private information. Briefly, the intuition is as follows. Investors know that divulging good information leads to higher prices. Naturally, this makes it difficult to extract favorable information. To counteract the incentive to hide good information, the underwriter commits to adjusting prices by less than warranted when good information is revealed. Equilibrium underadjustment leads to a positive correlation between file price revisions and initial returns. In the context of our aforementioned results – high-quality underwriters are more active in the information aggregation role – this paradigm predicts more pronounced partial adjustment for reputable underwriters.3 To isolate the component of the price revision due to private information, we admit two control variables used in previous literature. For each IPO we measure the average return of public firms in the same Fama-French 48 industry classification between the IPO filing date and the offer date. In addition, we define the “heat” of the IPO market as the average initial return of all IPOs in the 30 day window preceding the offer date. Private information is then orthogonalized against these two pieces of public information. Partial adjustment to private information is observed throughout our sample, but the effect is much stronger (both economically and statistically) for high-quality underwriters. A one-standard-deviation increase in price revision leads to a .43 standard deviation increase in initial returns for high-quality underwriters, but only a .20 standard deviation increase for low-quality underwriters. Again this evidence suggests the information acquisition paradigm is a better fit for reputable underwriters. 1.2 Partial Adjustment to Public Information Recent studies find that partial adjustment of IPO prices to public information is incomplete as well (Bradley and Jordan, 2002). There is disagreement over how to interpret these results, which are unanticipated by classical models. It may point to the 3 The aforementioned results concerning primary vs. secondary market information aggregation are not mechanism-specific, i.e. they do not assume bookbuilding. In contrast, the partial adjustment hypotheses follow directly from Benveniste and Spindt (1989). 3 importance of factors other than information asymmetry in determining average underpricing. Loughran and Ritter (2002), Ljungqvist and Wilhelm (2005) and Ince (2007) take it as evidence of an agency problem between issuers and underwriters in which there is bargaining over prices. An exogenous increase in wealth leads issuers to bargain less aggressively and accept more underpricing. Edelen and Kadlec (2005) develop a tradeoff argument that does not rely on agency costs. In their model, issuers attempt to go public but face the possibility of inefficient failure; in particular, they may be forced to withdrawal from the market. Underpricing reduces the likelihood of this failure. Higher valuations indicate a higher relative “cost” of failure, in which case issuers price their offerings more conservatively, leading to partial adjustment. We depart from previous literature in arguing that partial adjustment to the two proxies for public information (return on public comparables and IPO market heat) merit different interpretation. In classical models, an exogenous increase in the value of public comparables ought to be fully incorporated into the offer price. By contrast, high underpricing may simply indicate an environment of heightened adverse selection. If market conditions are sticky, then adverse selection and high underpricing will persist. Statisticians then observe autocorrelated underpricing, which could be erroneously interpreted as the incomplete adjustment of prices to past valuation surprises.4 With respect to public comparables, our full sample demonstrates a moderate amount of partial adjustment, replicating Edelen and Kadlec’s result. We find, however, that this effect is almost exclusively limited to low-quality underwriters. A one-standard deviation shock to public comparables’ returns leads to a .25 (.04) standard deviation increase in underpricing for low-quality (high-quality) underwriters. This finding is consistent with Loughran and Ritter’s agency theory if reputable underwriters are better able to commit to acting in the interests of issuers. It is also consistent with Edelen and Kadlec’s tradeoff model if inefficient withdrawal – the key piece of their analysis – is more likely for low-quality underwriters, which seems like a reasonable assumption. The results with respect to IPO market heat are reversed. While the full sample exhibits strong partial adjustment, the effect is stronger for reputable underwriters. A 4 In the model of Yung, Colak and Wang (2007) it is shown that positive shocks to investment opportunities lead to heightened adverse selection. Hence, high underpricing follows good news. 4 one-standard deviation shock to IPOheat leads to a .42 (.24) standard deviation increase in underpricing for low-quality (high-quality) underwriters. From the point of view of the agency or tradeoff theories, this reversal is puzzling: it is not clear why increases in wealth coming from public markets and from the IPO market should have differential effects for high and low-quality underwriters. Alternatively, this result may simply indicate more autocorrelation in the asymmetric information profile of firms brought public (i.e., more clustering) by high-quality underwriters for whatever reason. Still other explanations may emerge as the literature on partial adjustment to public information matures. Resolving this puzzle is outside the scope of the current analysis, which instead focuses on the extent to which private information is incorporated in the primary market rather than the secondary market. In other words, partial adjustment to public information acts primarily as a control variable in our analysis rather than a variable of direct interest. The plan of the paper is as follows. In Section 2, we describe the data and the methodology. Section 3 presents the main tests, comparing the cross-sectional volatility of price revisions with that of the secondary market. Section 4 examines how partial adjustment to private information varies with underwriter quality. Section 6 concludes. 2 Data and Methodology The data for this study was drawn from the Thompson SDC database and consists of initial public offerings of equity for the period 1980 through 2006. We exclude unit offers, ADRs, REITs, limited partnerships, closed-end funds and IPOs with an offer price lower than five dollars. Information for each IPO was collected regarding the initial filing range, the offer price, the number of shares sold, the identity of the underwriter, whether the firm was backed by a venture capitalist or not, and whether the firm operates in a high-tech industry or not. The above sample is supplemented with a hand-collected dataset of IPOs from 1980 to 1984, available from Jay Ritter’s website. In the case of overlapping observations (e.g., an IPO in both datasets), the Ritter’s data enable us to backfill missing variables in SDC. Whenever there is disagreement on a variable, we use Ritter’s value. 5 We supplement the dataset with the Carter-Manaster reputation rank of each underwriter and age of the firm, both of which are obtained from Jay Ritter’s website. Carter-Manaster ranks range from 1 (lowest quality) to 9 (highest quality), and Ritter’s website evaluates these ranks over separate time periods. When an IPO is underwritten by multiple lead underwriters, we average the reputation of all involved banks, though rank differences are typically small. In what follows, we use “high-reputation” and “high- quality” interexchangeably. We order our IPO sample by (average) underwriter rank. In what follows we group IPOs into one of three categories: average rank 8.5 or higher, rank between 6.75 and 8.5 and rank below 6.75. These breakpoints divide our sample into three nearly equal subsamples. Our results are qualitatively similar with other divisions. We merge this database with CRSP in order to compute monthly returns, excluding those IPOs which lack data in six months or more during the first 12 months following the offering. When time-series volatility is studied, we use daily prices instead, also extracted from CRSP. Our final sample consists of 7,124 completed offerings. In addition, we extract the deal characteristics (including underwriter identification) for 1,700 withdrawn IPOs for the same period. This inclusion is important in our study because we focus on the information revealed during bookbuilding. Clearly this information is correlated with the decision to continue or withdraw, and so our completed offering sample suffers from a censoring problem. Section 4 discusses Heckman’s (1976, 1979) two-step procedure employed to address this censoring. 2.1 Pre-IPO Price Revisions For each IPO, we calculate the offer price’s deviation from the midpoint of the filing range as follows Offer Pr ice − MidFile PREV = (1) MidFile where MidFile represents the midpoint of the initial filing range, respectively. This is the standard definition of pre-IPO price revision (e.g., Hanley, 1993). An alternative definition uses the dollar-value filing price range width as the denominator in (1) to account for the possibility that the filing range spread is associated with ex-ante level of 6 uncertainty (Cornelli & Goldreich 2003). Our results under this measure are significantly stronger than our reported results.5 Obviously, part of this price revision is due to public information. Following Edelen and Kadlec (2005) we use the average return of firms in the same Fama-French 48 industry (COMPS) between the filing date and the offer date as a proxy for publicly observable changes in valuation. We also employ the average underpricing of all IPOs in the 30 days preceding the offer date as a measure of overall IPO activity (IPOHeat) following Bradley and Jordan (2002). IPOHeat is orthogonalized against COMPS in order to single out the information coming from the IPO market specifically rather than the overall health of the equity markets. Finally, PREV is orthogonalized against both COMPS and IPOHeat to obtain a measure (REV) that isolates the private information revealed by bookbuilding. Earlier drafts of this paper used raw price revisions PREV rather than orthogonalized revisions REV, with similar results. 2.2 Holding Period Horizons Our methodology requires us to specify a horizon over which prices incorporate private information; see Figure 1. There are several institutional factors that may contribute to the inefficiency of immediate aftermarket prices. First, lead underwriters engage in price support during the first month or so of trading (Aggarwal, 2000), which truncates the distribution of returns and therefore reduces the informativeness of prices. Second, IPOs are difficult to short- sell in the immediate aftermarket (Loughran and Ritter, 1995) which further impedes the price discovery role of the market. Third, underwriters employ a system known as the Depository Trust Company’s IPO tracking system. This system identifies shares that are “flipped”, that is, trading immediately for quick profit – a strategy which is said to be opposed to underwriters’ interests. If a syndicate member’s customers are disproportionately identified as “flippers”, the lead underwriter reserves the right to recall 5 We view this difference as an artifact of low-quality underwriters’ lower offer prices, rather than a genuine economic effect. Note that Offer − Mid = Offer − Mid Mid , and the last fraction is smaller for Hi − Lo Mid Hi − Lo low-quality underwriters. Low-quality underwriters therefore have even smaller revisions under this alternative measure. 7 underwriting fees earned by that syndicate member (a “penalty bid”). This recall does occasionally occur in practice, and the tracking system is often monitored for thirty days or more (Aggarwal, 2000). IPO investors, sensitive to their standing with underwriter, may refrain from trading as aggressively as they would in the absence of such a system. Taken together, all of these institutional factors suggest that in seeking an appropriate proxy for true value of an IPO, one should require at least a month (if not more) of seasoning. Long horizons, however, introduce their own problems. First, estimate of long- run abnormal returns are extremely noisy and biased. Kothari and Warner (1997) conclude that inferences from long-run studies “require extreme caution.” Second, much of the variation occurring at very long horizons is unrelated to ex-ante private information of IPO investors, which is the information of interest in the current study. It is our view that most of this private information will be resolved through the trading of investors within the first several months after the IPO. To balance these tradeoffs, we focus on buy-and-hold returns at 3-month horizons. We also follow the stocks for up to twelve months after the IPOs, in part as a robustness check. Differential patterns at the six to twelve month horizons may be of independent interest, however, since they may shed light on the private information of insiders (since most lockups expire at six months) rather than that of secondary market investors. 2.2 Cross-Sectional Variance in Aftermarket Returns For our cross-sectional results, we use monthly CRSP returns. The abnormal return for stock i at month T is defined as T T BHARiT = ∏ (1 + rit ) − ∏ (1 + mit ) (2) t =1 t =1 where rit denotes the return on stock i in month t, and mit denotes the return on the market (proxied by the CRSP stock file value-weighted market index) at time t. Here the implicit assumption is that the market serves as a reasonable measure of expected return for stocks in our sample. Our cross-sectional results are insensitive to the choice of benchmark, however. We note that the benchmark choice is more critical for event 8 studies that focus on the mean rather than (as in our case) the variance, because nearly all of our cross-sectional variance is firm-specific rather than driven by the benchmark. 2.4 Time Series Variance in Aftermarket Returns To examine patterns in secondary market returns in more depth, we use CRSP to obtain daily prices in the first year after the IPO for all the 7,124 completed IPO firms in our data set. For each firm, we run a regressison of the daily returns on the Fama-French factors and the momentum factor to obtain daily abnormal returns. The mean and standard deviation in time-series variances are compared across the subsets of IPOs underwritten by high-quality and by low-quality banks. 2.5 Testable Implications We test the following hypotheses, motivated in the introduction: H1: Cross-sectional dispersion in price revisions is higher for more reputable underwriters. Hypothesis H1 is consistent with a scenario in which low quality underwriters do not extract information from investors; rather, they serve in an uninformative distributional role. This hypothesis has another implication. If no uncertainty is resolved during the bookbuilding phase for low-quality underwriters, this means that all of the uncertainty must be resolved in the secondary market, leading to the following claim. H2: Information aggregation during the bookbuilding period leads to reduced subsequent uncertainty. Therefore high-quality underwriters should be associated with lower (cross-sectional) aftermarket return variability. To test Hypothesis H2 we calculate the cross-sectional dispersion of buy-and-hold returns (BHARs), identifying reduced uncertainty with reduced dispersion in BHARs. We also consider the analogous time-series version of this hypothesis. 9 H3: Firms taken public by more reputable underwriters exhibit lower time-series return volatility. Again, the intuition is that more active price discovery in the pre-market leads to less need for price discovery in the secondary market. The aforementioned tests are univariate. It may be the case, however, that high- quality underwriters are associated with systematically different types of issuers. It is important therefore to control for firm characteristics. We do this in two ways. First, as is standard in the literature, we introduce firm characteristics as additional independent variables in our regressions. However, controlling for firm characteristics is a highly imperfect control if the asymmetric information profile of firms is different across underwriters, which seems likely here. As a second control, we therefore consider the following variant of H1 and H2: H4: Of the total (post-filing) uncertainty to be resolved, for high-quality underwriters a smaller proportion is resolved in the secondary market than for the low-quality underwriters. To test Hypothesis H4 we form the ratio of cross-section dispersion in abnormal aftermarket returns to the cross-sectional dispersion in filing price revisions. This fraction is termed a variance ratio. To the extent that either category of underwriter is associated with higher ex-ante risk, both the numerator and denominator of that ratio will be inflated. In effect, the denominator “factors out” the level of unobservable risk so that investigating Hypothesis H4 is equivalent to investigating whether the proportion of uncertainty resolved in the bookbuilding period varies with underwriter quality. Hypotheses H1 through H4 are not mechanism-specific. That is, they assume that information is aggregated in the primary markets but do not specify how. In contrast, our final hypothesis depends on bookbuilding models in particular. As mentioned earlier, these models predict a positive correlation between underpricing and private information 10 revealed.6 Under the maintained hypothesis that the information gathering paradigm better models the behavior of high-quality underwriters, we predict: H5: Partial adjustment to private information is more pronounced for reputable underwriters. Finally, we turn to partial adjustment to public information. Three possible explanations for this phenomenon were summarized in the introduction: agency costs, the trade-off theory and autocorrelated information asymmetry. We ask now whether the strength effects should vary with underwriter quality. The relevant hypotheses are summarized in the following table and discussed below. Model Agency/ Autocorrelated Trade-off Theory Bargaining Theory Information Asymmetry Partial Adjustment to Return of Yes Yes No Public Comparables? Partial Adjustment to IPOHeat? Yes Yes Yes Effects Stronger for Low Quality Yes Yes No prediction Underwriters? Both the agency and tradeoff theory rely upon valuation shocks to insiders’ personal wealth. Theoretically, it should not matter which channel the news comes from: publicly traded comparables or the returns of contemporaneous IPOs. In contrast, the autocorrelated information asymmetry argument is based upon the amount of adverse selection in the IPO market itself. Controlling for IPO market heat, the movements of public comparables should contain no information regarding the amount of adverse selection for current offerings. The arguments behind the agency and tradeoffs theories are particularly strong for low-quality underwriters. Reputable underwriters tend to mitigate agency problems. Issuers able to attract high-quality underwriters tend to be larger and more well-known, 6 This prediction would not hold in an auction, for example, in which case prices would be weak-form efficient. 11 leading to better outside opportunities. Inefficient withdrawal may also be less likely for IPOs with a high quality underwriter. All of these factors suggest stronger partial adjustment to public information for low-quality underwriters in both the agency theory and the trade-off theory. In contrast, the autocorrelated information asymmetry theory makes no prediction regarding underwriter quality: a priori, either type of underwriter could exhibit stronger intertemporal clustering in the asymmetric information profile of IPOs. 3. Volatility in the Primary and Secondary Markets Table 1 presents descriptive statistics for the full sample, and for the three terciles of IPO firms sorted by underwriter’s reputation ranking in two panels, respectively. High-quality underwriters are more likely to take public firms that are venture capital backed and in high-tech industries. Their offerings are older on average and tend to raise larger proceeds than firms taken public by median or low prestige underwriters. Taken together, these variables have an ambiguous impact on the severity of asymmetric information across terciles: firm seasoning should mitigate market imperfections whereas high-tech industry affiliation may increase them. Price revisions relative to the original filing price range, both in raw form (PREV) and industry-return-adjusted form (REV), are approximately symmetrically distributed around zero in the full sample. On average, they are negative for the low-reputation tercile, near zero for the medium-reputation tercile, and positive for the high-reputation tercile. However, central tendency differences are not large; the median PREV is exactly zero for all three quality terciles. The absolute value of price revision increases monotonically with bank reputation. The mean absolute value of price revision is 0.168 for firms taken public by high-quality underwriters, but only 0.104 for those by low-quality underwriters. This preliminary evidence is consistent with the notion that high-quality underwriters are more active in aggregating information in the marketing phase. Consistent with Beatty and Welch (1996), Kumar McGee and Womack (1998), Habib and Ljungqvist (2002), and Cooney, Singh, Carter and Dark (1999), we find that prestigious underwriters are associated with higher average underpricing than low-quality 12 underwriters (29.8% vs. 14.3%). For all three terciles, the underpricing variable is highly skewed: in all subsamples, the mean underpricing is at least double the median underpricing. Long-run IPO returns in our sample are disappointing. The average 12-month buy-and-hold return in our full sample is –6.0%. Also consistent with existing literature, much of this poor return is concentrated in the offerings of low prestige underwriters (Brav and Gompers 1997). The average 12-month BHAR is –14.2% in the bottom quality tercile, while it is –0.6% in the top quality tercile. However, over shorter horizons there is little or no underperformance: the average 3-month BHAR is 5.6% in the top tercile and -1.0% in the bottom tercile. 3.1 REV and BHAR Variability Table 2 presents the cross-sectional standard deviations of our key variables. Over the whole sample, dispersion in pre-IPO price revision (REV) shows a strongly positive monotonic trend ( σ =28.1% vs. σ =17.0%; top vs. bottom tercile). BHAR dispersion displays a milder, if still identifiable, trend. Buy-and-hold dispersion is greater for high prestige underwriters, but these differences are small. For instance, in the case of 12- month BHARs, the cross-sectional standard deviation for the top tercile is σ =79%, whereas for the bottom tercile it is σ =72%. In contrast to this full sample result, the subperiod analysis indicates that throughout most of our sample high-quality underwriters have been associated with lower secondary market return variability. For example, in the 1980s, the standard deviation of 3-month BHARs was 33% for the bottom tercile and only 22% for the top tercile, and the difference is statistically significant at the 1% level. This pattern persists for all holding time horizons and subperiods 1980s, 1990-1995 and 2001-2006. The only rejection of Hypothesis H2 was during the internet bubble period (1996-2000) in which high-quality underwriters were associated with higher volatility. This difference between subperiod results and full sample results is partly explained by the Simpson’s paradox effect. To see why, note that cross-sectional variation was dramatically higher during hot market of 1996-2000, consistent with Yung, Colak and Wang’s (2007) argument that hot markets are associated with heightened 13 adverse selection. For example, for high-quality underwriters, the standard deviation of 3-month BHARs is 70% during the 1996-2000 period but only 22% during the 1980s. On the other hand, there was a strong substitution towards high-quality underwriters during this market. Hence, even though high-quality underwriters are associated with lower variance in most subperiods, it appears in the full sample as if they are associated with higher variance. To summarize, hypothesis H1 is supported throughout the sample and hypothesis H2 is supported in all subperiods except 1996-2000. 3.1 Variance Ratios The aforementioned descriptive statistics include dispersion in each of our key variables. Recall, however that our analysis is confounded by the fact that different profiles of asymmetric information associated with each quality tercile. To account for this unobservable factor, our main hypothesis is that information aggregation should increase the dispersion of REV relative to that of BHAR, i.e., relatively more information is incorporated in the primary market. Perhaps the simplest way to measure the relative size of these two variables is to form their ratio. Our methodology is as follows. We treat each bank (rather than each IPO) as an observation. For a given bank, we measure the buy-and-hold return on each of its offerings and the cross-sectional variance of these returns. We then compute the cross- sectional variance of the pre-IPO price revisions of these same offerings. Finally, we define the “variance ratio” of this bank (VR) to be the ratio of these two cross-sectional variances (with the variance of BHAR in the numerator since the variance of REV could be very close to zero for small investment banks that have underwritten only a few IPOs). To ensure reasonably stable estimates of second moments, we keep in our sample only underwriters that has made more than 6 offerings (using different cutoffs does not change our results.) Using the same Carter-Manaster tercile cutoffs of 6.75 and 8.5 described in Section 2, we are left with 25 high-quality banks, 41 medium-quality and 145 low-quality banks. Finally, we compute the descriptive statistics of variance ratios within each tercile. 14 Untabulated examination shows the variance ratios thus obtained are highly skewed. In all three terciles, the skewness of variance ratios is an order of magnitude larger than mean, median or standard deviation. Therefore, subsequent statistical analysis employs the natural logarithm of the variance ratio for each bank. Table 3, Panel A summarizes the statistical significance of the differences in log variance ratios for the full sample. Low-quality underwriters have higher ratios than either other terciles, and this difference is significant at all horizons (p-value < 0.01 in three cases and equals to 0.011 in the other), which is consistent with hypothesis H4. The abnormal pattern in aftermarket returns during 1996-2000 shown in Table 2 suggests that closer examination in warranted here as well. In Table 3, Panel B, we isolate the bubble period and find log variance ratios are of roughly identical levels across underwriter groups. Thus the lone failure for H2 applies to H4 as well. In the rest of time, low-quality underwriters exhibit significantly higher ratios than medium-quality and high-quality underwriters, as shown in Table 3, Panel C. The multivariate version of this result tells the same story; see Table 4. We regress log variance ratios on underwriter reputation, controlling for firm characteristics, including the amount filed, firm age, whether the firm is venture capital-backed or in a high-tech industry. We also include in the regression the IPO market spillover information measure, IPOHeat, to control market condition around the IPO time. For an underwriter firm characteristics and IPOHeat are the average of all firms it takes public. Since different underwriters might attract different types of issuers, we orthogonalize mean firm characteristics against underwriter reputation. Columns 2-9 display the estimated coefficients on the reputation variable, all of which are negative, and 7 out of 8 coefficients are statistically significant. In untabulated results, we also consider the possibility that underwriter reputation is the endogenous result of issuer firm characteristics and use a two-stage-least-square style procedure to estimate the model. Again we obtain significant, negative coefficients for bank reputation. Overall, we conclude that prestigious underwriters are associated with increased dispersion of pre-IPO price revisions relative to that of aftermarket buy-and-hold returns, and this result holds after controlling for different firm characteristics. Hypothesis H4 is 15 thus supported. This evidence is consistent with these underwriters taking a more active information aggregation role during the bookbuilding period than do low-quality underwriters. 3.2 Factoring Out the Predictable Component of Underpricing In Section 3.1 we investigate the dispersion of REV relative to that of BHARs, the aftermarket returns, and find that high-reputation banks exhibit higher variance ratios than their low-reputation peers, indicating that a larger proportion of information asymmetry is resolved in the bookbuilding stage by a high-reputation bank than by a low- reputation bank. By design, those tests avoid the underpricing stage (UP) in Figure 1. The reason for this omission is that UP is a contaminated measure. It incorporates both (1) a predictable component resulting from REV, because of the partial adjustment effect and (2) some early aftermarket price revelation. It therefore contains information from the two stages that we want to measure separately. In this section we define an “implied value” from the empirical relationship between pre-IPO price revision and underpricing. This implied value is, effectively, the expected true value conditional on information revealed by REV. The specific method is to run the regressions in Section 4, defining the “implicit offer price” as the real offer price times the sum of one and the predicted underpricing. Doing so identifies the predicted underpricing as the investment bank’s intended underpricing. The residuals are price changes resulted from new information from investors and/or noise during the first day trading. Secondary market price discovery is then defined by deviations from this implied value, rather than from the (biased) offer price. With this new baseline price, we redefine aftermarket price changes, denoted as PChange3, Pchange6, Pchange9 and Pchange12, respectively for 3-month, 6-month, 9-month and 12-month time horizons. The difference between Pchange and BHAR lies in the baseline price: BHAR uses the closing price on the first trading day as the baseline, while Pchange uses the implied value. Table 5 summarizes the comparison of variance ratios derived across firm groups. The average variance ratio is higher for low-reputation underwriters and the differences between the high-reputation group and the low-reputation group are significant at about 16 the 1 percent level. The regression results are displayed in Table 6. Again, the estimated coefficients of reputation are consistently negative and mostly significant, indicating firms taken public by a higher quality investment bank exhibit lower aftermarket return dispersion relative to that of pre-offering price revisions, compared to those taken public by a lower quality bank. The results are qualitatively identical to those in Table 3 and Table 4 when BHARs is used as the measure of aftermarket returns. 3.3 Time-series volatilities In this subsection, we repeat the analysis of Section 3.1, using time-series volatility rather than a cross-sectional measure. Again the argument is that primary market price discovery should reduce secondary market risk. We employ daily return observations and find abnormal daily returns as the residuals of the non-intercept Fama-French-Carhart four-factor model. The market, size, value and momentum factor are all from Kenneth R. French’s website. We then compute the standard deviations of abnormal daily returns for each month and each quarter in the first 12 months following the IPO. The first trading day is excluded. Table 7 displays the average of aftermarket return volatility across the high- reputation-bank-underwritten and low-reputation-bank-underwritten groups. In the whole period of 1980-2006 (Panel A), the results are mixed: for the 3-month and 6-month horizons, the high-reputation group has significantly return volatility, while the pattern switches for the 9-month and 12-month horizons. However, this full sample result masks the relationship seen in the subperiods. Mirroring our cross-sectional results, the variance is lower firms underwritten by reputable banks, for all the four time horizons, in every subperiod except 1996-2000 (Panels B-E). Table 8 summarizes the multivariate version of this test. We regress aftermarket volatility on bank reputation and a variety of firm characteristics. The results are as expected. VC-backed and high-tech firms are riskier. Larger and older firms are less risky. Consistent with Yung, Colak and Wang (2007), initial public offers that occur during hot markets are riskier, ceteris parabus. As Hypothesis 3 predicts, the coefficients of bank reputation for 6-, 9-, and 12-month horizons are all negative. 17 4. Partial Adjustment as a Function of Underwriter Quality Our fifth hypothesis is that the partial adjustment to private information is more pronounced for prestigious underwriters. This prediction will hold provided that they act in the costly information extraction role suggested by Benveniste and Spindt (1989), while low-quality underwriters act in the uninformed, distributional role suggested by Rock (1986). Figure 2 offers preliminary evidence on this question. Raw price revisions are plotted against underpricing. Along the top panel (the lowest quality) there appears to be no relation, and in the middle panel (medium quality) there is a mildly identifiable association. For the highest quality, there is an obvious positive correlation. Note also that this figure re-iterates our core finding of Section 3: higher quality underwriters exhibit much more dispersion in REV. To test Hypothesis H5 formally, we look into whether REV’s predictive power for IPO initial returns differ across underwriter reputation terciles by regressing underpricing on REV. As emphasized by Edelen and Kadlec (2005), this type of regression (though standard in the literature) suffers from a censoring problem since only firms that were received favorably (and therefore didn’t withdraw) are typically included in the sample. To address this issue, we employ a Heckman (1976, 1979) two-stage model. In the first stage, we use the full sample (including both completed and withdrawn IPOs) to run the following Probit regression, I w = α 0 + α 1COMPS + + α 2 COMPS − + α 3WRate + α 4 Re pu + α 5 AmtFile + ε (3) where Iw equals 1 for withdrawn offerings and 0 for completed offerings. COMPS + equals COMPS, the equally-weighted return of listed firms in the same Fama- French 48 industry during the bookbuilding period, if COMPS >0, and 0 otherwise; COMPS − equals COMPS if COMPS < 0, and 0 otherwise. COMPS + and COMPS − together capture the market movement and account for potential asymmetric impact of positive and negative market movements on the IPO withdrawal/completion decision. The latest IPO withdrawal rate, wrate, is computed as the number of IPO withdrawals 18 within 106 days (the median length of registration period in our sample) of filing, divided by the number of active IPOs under registration during the 30 days preceding the offer or withdrawal date. We orthogonalize wrate with respect to COMPS and use the residual WRate in order to isolate the spillover information from activity in the withdrawn IPO market. Repu and AmtFile are lead underwriters’ (average) reputation ranking and the issuing firm’s amount filed. The above specification that uses public information to explain withdrawal/completion decision is in essence similar to that in Edelen & Kadlec (2005). The estimation results of (3) are reported in the Appendix. Using estimated coefficients, we calculate the inverse Mills ratio (sometimes called “selection hazard”), λ. In the second step, we run an OLS regression (using the sample of completed offerings) in which the dependent variable is IPO underpricing. The set of independent variables includes the inverse Mills ratio from the first step, private information REV, market movement variable COMPS, IPO market spillover information, IPOHeat, and risk proxies identified in the underpricing literature. In a separate specification, asymmetric adjustment to both public and private information is allowed. Likewise, we define REV+ as equal to REV for positive REV and 0 otherwise, and REV- as equal to REV for negative REV and 0 otherwise. Partial adjustment will be indicated by positive and significant coefficients on REV, or REV+ and REV- , in the two specifications respectively. We report the results of this regression in Table 9. The estimated coefficient for REV is monotonic in bank reputation, rising from 0.338 for the bottom reputation tercile in to 1.166 for the top reputation tercile in specification (1). Thus, partial adjustment to private information is a much more pronounced phenomenon for reputable underwriters. In addition, this partial adjustment shows significant asymmetry (as indicated by higher coefficients on REV+ than on REV-) for high-quality underwriters, again consistent with payment for information aggregation in the Benveniste and Spindt (1989) framework. For low-quality underwriters, the partial adjustment is nonexistent given positive private information, but big given negative private information, contrary to Benveniste and Spindt’s prediction. The coefficient on COMPS decreases monotonically with reputation. Thus reputable underwriters more fully adjust prices to the information implied by the return of 19 public comparables. In contrast, the coefficient on IPOHeat increases monotonically with reputation. This differential response to two different types of public information is somewhat puzzling from the point of view of the agency/bargaining theory and the tradeoff theory. Both theories emphasize the role of the issuer’s personal wealth, and it is unclear why shocks to wealth coming from the public equity markets and the IPO market should have differential effects. An alternative view discussed in Section 2.5 is that partial adjustment to IPOHeat simply reflects autocorrelation in the severity of frictions in the IPO market. The larger coefficient on IPOHeat for reputable underwriters would then indicate more intertemporal clustering of firms types (i.e., states in which particularly risky IPOs are underwritten are more persistent). Another interesting observation from Table 9 is the differences in R² across underwriter groups. With specification (1), this goodness-of-fit statistic increases from 0.108 for the low-reputation underwriter tercile to 0.466 for the high-reputation underwriter tercile. The differences in R² are statistically significantly, as shown in the bottom panel. With specification (2) and an untabulated univariate specification with REV being the only explanatory variable, the same pattern in R² is observed. This pattern, indicating pre-offering price revisions made by high-reputation underwriters are more informative about underpricing, buttresses our argument above that the Beveniste- Spindt dynamic bookbuilding framework describes the behavior of high-reputation underwriters better. The economic impact of information flows during the bookbuilding period on underpricing is displayed in Table 10. A one-standard-deviation shock to REV causes the underpricing to change by 0.43 standard deviation if the underwriter is a high-reputation one, but only by 0.20 standard deviation if the underwriter is a low-reputation one. Shock to comparable firm return barely alters the underpricing for a high-reputation underwriter, but does cause change to the underpricing for a low-reputation underwriter. In summary, high-reputation underwriters exhibit more pronounced partial adjustment and asymmetric adjustment to private information, both consistent with the bookbuilding model’s prediction; The bookbuilding model, however, fails to describe the behavior of low-reputation underwriters. 20 5. Conclusions Our evidence suggests that reputable banks are more active in primary market information aggregation than non-reputable banks. High-quality banks have more active price revisions: the standard deviation of pre-offering price revisions of high-quality banks is nearly twice that of low-quality banks. This information aggregation apparently substitutes for secondary market price discovery. Both cross-sectional aftermarket return variance and time-series volatility are lower for high-quality banks in all time periods except the bubble period, 1996-2000. Thus, reputable banks have a higher relative proportion of information asymmetry resolved during the bookbuilding period rather than in the secondary market. We also confirm that a central prediction of bookbuilding models – partial adjustment to private information – has much stronger support in the high reputation subsample. Again this is consistent with a stronger price discovery role for reputable underwriters. Finally, we turn to partial adjustment to public information. We show that partial adjustment to two separate pieces of public information (IPO market spillover information and returns of comparable public firms) show reversed patterns for low and high-quality underwriters. These differential effects do not seem to have an obvious basis in either the tradeoff theory or the agency/bargaining theory. It is hypothesized that this result may indicate differences in the intertemporal clustering in types of firms underwriters by different underwriters, but a full investigation along these lines is beyond the scope of this paper. 21 Appendix: Heckman first-step regression The table below presents the probit estimation in the first step of the Heckman (1976, 1979) two- step procedure that is used to correct selection bias. The dependent variable is WITHDRAW, which equals 1 for withdrawn offering and 0 for completed offerings. Independent variables include a series of public information. COMPS + equals COMPS if COMPS > 0, and 0 otherwise; COMPS − equals COMPS if COMPS < 0, and 0 otherwise. the IPO withdrawal rate, wrate, is computed as the number of offerings withdrawn within 106 days of their initial filing date divided by the number of active registrants during the 30 days preceding the offer or withdrawal date. REPU is the underwriters’ average Carter-Manaster reputation ranking. AmtFile is the filing size in million dollars. Estimate Std error Chi-Sq Pr > Chisq COMPS + -0.644 0.0619 108.29 <.0001 COMPS- 5.276 0.2972 315.09 <.0001 wrate -1.004 0.0598 281.44 <.0001 repu 0.162 0.0081 398.23 <.0001 AmtFile -0.166 0.0169 97.27 <.0001 22 References: Aggarwal, R. 2000, “Stabilization activities by underwriters after initial public offerings. Journal of Finance 55, 1075-1103. Beatty, R., and J. Ritter, 1986, Investment Banking, Reputation and the Underpricing of Initial Public Offerings, Journal of Financial Economics, 15, 213 – 232. Beatty, R. P. and I. Welch, 1996, Issuer expenses and legal liability in initial public offerings, Journal of Law and Economics, 39, 545-602. Benveniste, L.M. and P.A. Spindt, 1989, How investment banks determine the offer price and allocation of new issues, Journal of Financial Economics, 24, 343-362. Bradley, D. J., and B. D. Jordan, 2002, Partial adjustment to public information and IPO underpricing, Journal of Financial and Quantitative Analysis, 37, 595-616. Brav, A. and P. A. Gompers, 1997, Myth of Reality? The Long-run Underperformance of Initial Public Offerings: Evidence from Venture and Nonventure Capital-backed Companies, Journal of Finance, 52, 1791-1821. Carter, R.B. and S. Manaster, 1990, Initial Public Offerings and Underwriter Reputation, Journal of Finance, 45, 1045 – 1068. Cooney, J. W., A. K. Singh, R. B. Carter and F. H. Dark, 2001, IPO initial returns and underwriter reputation: Has the inverse relationship flipped in the 1900s? Unpublished working paper, Iowa State University. Edelen, R. and G. Kadlec, 2005, Issuer surplus and the partial adjustment of IPO prices to public information, Journal of Financial Economics 77, 347-373. Fama, E. and K. French, 1997, Industry costs of equity, Journal of Financial Economics 43, 153-193. Habib, M., and Alexander L., 2001, Underpricing and entrepreneurial wealth losses in IPOs: Theory and evidence, Review of Financial Studies 14, 433-458. Hanley, K. W., 1993, The underpricing of initial public offerings and the partial adjustment phenomenon, Journal of financial Economics 34, 231-250. Heckman, J. J., 1976, The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models, Annals of Economic and Social Measurement, 5(4): 475-492. Heckman, J. J., 1979, Sample selection bias as a specification error, Econometrica 47, 153-162. 23 Ince, O. S., 2007, The partial adjustment of IPO offer prices is not due to dynamic information acquisition, working paper, University of Florida. Kothari, S. P. and J. Warner, 1997, Measuring long-horizon security price performance, Journal of Financial Economics 43, 301-339. Krigman, L., W.H. Shaw and K. L. Womack, 2001. Why do firms switch underwriters? Journal of Financial Economics 60, 245-284. Kumar, A., V. McGee and K. L. Womack, 1998. Underwriter value added in IPOs , Unpublished working paper, Dartmouth College. Ljunqvist, A., and W.J. Wilhelm, 2003, IPO pricing in the Dot-Com bubble, Journal of Finance 58, 723-752. Loughran, T. and J. Ritter, 1995. The new issues puzzle, Journal of Finance 50, 23-51. Lowry, M. and G. W. Schwert, 2002, IPO market cycles: Bubbles or sequential learning? Journal of Finance 57, 1171-1200. Ritter, J., and I. Welch, 2002, IPO Activity, Pricing and Allocations, Journal of Finance, 57, 1795-1828. Rock, K., 1986, Why New Issues are Underpriced, Journal of Financial Economics, 15, 187 – 212. Yung, C., G. Colak and W. Wang, 2007, Cycles in the IPO Market, Journal of Financial Economics, forthcoming. 24 Table 1: Descriptive Statistics Panel A: Full Sample (N=7,124) Mean Median 10th percentile 90th percentile Filing Amount ($m, 1980) 14.26 6.63 1.63 25.65 Age (years) 15.86 8 2 43 High Tech dummy 0.45 0 0 1 Venture Capital dummy 0.37 0 0 1 Bank Reputation 7.06 8 3 9 PREV 0.001 0 -0.241 0.217 REV 0.000 -0.001 -0.235 0.209 Underpricing 0.206 0.091 -0.028 0.500 3-month BHAR 0.025 -0.029 -0.348 0.404 6-month BHAR 0.004 -0.080 -0.527 0.563 9-month BHAR -0.024 -0.125 -0.666 0.647 12-month BHAR -0.060 -0.189 -0.775 0.734 Panel B: SubSamples Low Reputation Medium Reputation High Reputation Mean Median Mean Median Mean Median Filing Amount ($m, 1980) 4.09 2.41 9.66 6.78 27.51 12.77 Age (years) 7.42 7 9.01 9 10.34 9 High Tech dummy 0.36 0 0.49 0 0.48 0 Venture Capital dummy 0.23 0 0.42 0 0.43 0 Bank Reputation 4.08 4.81 7.69 8 8.93 9 PREV -0.033 0 -0.012 0 0.043 0 REV -0.035 -0.010 -0.013 -0.006 0.043 0.018 abs(REV) 0.104 0.066 0.151 0.113 0.168 0.115 Underpricing 0.143 0.071 0.180 0.093 0.298 0.111 3-month BHAR -0.010 -0.060 0.028 -0.021 0.056 -0.009 6-month BHAR -0.043 -0.132 0.009 -0.066 0.039 -0.051 9-month BHAR -0.088 -0.204 -0.010 -0.115 0.018 -0.072 12-month BHAR -0.142 -0.279 -0.043 -0.194 -0.006 -0.115 1 Table 2: Cross-Sectional Variation in Pre-IPO Price Revision and Aftermarket Returns This table presents the mean and standard deviation of the pre-IPO price revision and aftermarket buy-and- hold returns for 3-, 6-, 9-, and 12-month periods. REV is the price revision from the midpoint of the original filing range to the offer price, scaled by the midpoint of the offer range and then orthogonalized against industry return during the bookbuilding period. Our sample of IPOs is partitioned according to Carter-Manaster rank of the lead underwriters involved, or average quality if the issue is co-underwritten. IPOs are in the top tercile of our sample if their average underwriter rank is 8.5 or higher, and in the bottom tercile if it is below 6.75. Panel A: 1980-2006 Bottom Tercile Medium Tercile Top Tercile Bottom vs. Top (n=2116) (n=2511) (n=2497) Pr > F Mean -0.019 -0.013 0.029 REV 0.000 Std Dev 0.170 0.194 0.281 Mean -0.010 0.029 0.056 3-month BHAR 0.000 Std Dev 0.365 0.381 0.468 Mean -0.043 0.009 0.039 6-month BHAR 0.000 Std Dev 0.534 0.548 0.620 Mean -0.088 -0.010 0.018 9-month BHAR 0.000 Std Dev 0.631 0.694 0.695 Mean -0.142 -0.043 -0.006 12-month BHAR 0.000 Std Dev 0.723 0.811 0.793 Panel B: 1980s Bottom Tercile Medium Tercile Top Tercile Bottom vs. Top (n=879) (n=712) (n=628) Pr > F Mean 0.016 -0.005 0.009 REV 0.000 Std Dev 0.111 0.139 0.137 Mean 0.007 0.001 0.012 3-month BHAR 0.000 Std Dev 0.332 0.246 0.221 Mean -0.009 -0.016 0.003 6-month BHAR 0.000 Std Dev 0.522 0.377 0.332 Mean -0.057 -0.027 0.000 9-month BHAR 0.000 Std Dev 0.638 0.522 0.436 Mean -0.111 -0.044 -0.021 12-month BHAR 0.000 Std Dev 0.766 0.589 0.526 1 Panel C: 1990-1995 Bottom Tercile Medium Tercile Top Tercile Bottom vs. Top (n=610) (n=766) (n=644) Pr > F Mean -0.017 0.013 0.039 REV 0.000 Std Dev 0.152 0.198 0.198 Mean 0.007 0.056 0.062 3-month BHAR 0.061 Std Dev 0.327 0.299 0.303 Mean -0.008 0.078 0.083 6-month BHAR 0.022 Std Dev 0.495 0.472 0.452 Mean -0.046 0.066 0.069 9-month BHAR 0.000 Std Dev 0.595 0.577 0.507 Mean -0.084 0.050 0.045 12-month BHAR 0.006 Std Dev 0.673 0.743 0.603 Panel D: 1996-2000 Bottom Tercile Medium Tercile Top Tercile Bottom vs. Top (n=511) (n=842) (n=864) Pr > F Mean -0.073 -0.027 0.045 REV 0.000 Std Dev 0.239 0.225 0.410 Mean -0.077 0.028 0.093 3-month BHAR 0.000 Std Dev 0.435 0.526 0.700 Mean -0.158 -0.033 0.037 6-month BHAR 0.000 Std Dev 0.593 0.731 0.903 Mean -0.222 -0.076 -0.017 9-month BHAR 0.000 Std Dev 0.647 0.906 0.982 Mean -0.306 -0.141 -0.058 12-month BHAR 0.000 Std Dev 0.676 1.028 1.091 Panel E: 2001-2006 Bottom Tercile Medium Tercile Top Tercile Bottom vs. Top (n=116) (n=191) (n=361) Pr > F Mean -0.066 -0.040 0.010 REV 0.920 Std Dev 0.195 0.189 0.197 Mean 0.077 0.017 0.030 3-month BHAR 0.000 Std Dev 0.423 0.308 0.267 Mean 0.035 0.003 0.022 6-month BHAR 0.000 Std Dev 0.462 0.354 0.343 Mean 0.064 0.052 0.050 9-month BHAR 0.000 Std Dev 0.598 0.544 0.423 Mean 0.082 0.048 0.066 12-month BHAR 0.000 Std Dev 0.718 0.583 0.517 2 Table 3: Variance Ratios Across Underwriter Quality Terciles This table presents the folowing statistic. For each underwriter in our sample with at least seven IPOs, we compute the cross-sectional variance of secondary market returns of IPOs underwritten by that bank, employing 3-,6-,9- and 12-month holding periods. We also compute cross-sectional variance of pre-IPO price revisions, again for IPOs underwritten by that bank. The variance ratio is defined as the ratio of these two statistics. The high-rep subsample consists of underwriters with Carter-Manaster rank 8.5 or higher, and the low-rep subsample consists of underwriters with Carter-Manaster rank below 6.75. Panel A: 1980-2005 Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 145 25 1.676 1.199 Low vs. High 0.005 6-month VR 145 25 2.475 1.815 Low vs. High 0.000 9-month VR 145 25 2.844 2.300 Low vs. High 0.003 12-month VR 145 25 3.051 2.587 Low vs. High 0.011 Panel B: 1996-2000 Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 41 13 1.413 1.451 Low vs. High 0.912 6-month VR 41 13 1.921 1.958 Low vs. High 0.922 9-month VR 41 13 2.361 2.253 Low vs. High 0.723 12-month VR 41 13 2.455 2.576 Low vs. High 0.772 Panel C: Other time Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 123 23 1.688 0.997 Low vs. High 0.000 6-month VR 120 23 2.557 1.666 Low vs. High 0.000 9-month VR 120 23 2.934 2.210 Low vs. High 0.000 12-month VR 120 23 3.220 2.538 Low vs. High 0.000 1 Table 4: The Determinants of Underwriters' Variance Ratios This table presents the the coefficient estimates, and heteroscedasticity-consistent standard errors in parentheses, of OLS regressions. The dependent variable is the observation for each underwriter (with at least 5 IPOs in our sample) of the following statistic: the log of the cross- sectional dispersion in secondary market returns divided by the cross-sectional standard deviation of price revisions. Statistical significance at the 1%, 5% and 10% level are denoted with superscripts ***,** and *, respectively. Dependent Variables (N=211) Independent Variables VR3 VR6 VR9 VR12 VR3 VR6 VR9 VR12 -0.158*** -0.210*** -0.207*** -0.166*** -0.119** -0.146** -0.169** -0.101 Bank Reputation (0.042) (0.043) (0.048) (0.048) (0.059) (0.065) (0.074) (0.072) -2.870 -2.158 -3.213 -1.628 IPOHeat (2.660) (2.894) (2.938) (3.158) -0.109 -0.339** -0.224 -0.309 Filing Amount (0.152) (0.148) (0.145) (0.141) -0.108 0.005 -0.099 0.078 Age (0.199) (0.202) (0.214) (0.206) -0.698 -0.568 -0.801 -0.614 VC-backed (0.667) (0.653) (0.670) (0.626) 1.273* 1.130* 0.871 0.588 Hi-Tech (0.661) (0.604) (0.640) (0.631) R² 0.081 0.127 0.110 0.072 0.154 0.205 0.178 0.126 1 Table 5: Variance Ratios Across Underwriter Quality Terciles Using Pchange This table presents the folowing statistic. For each underwriter in our sample with at least seven IPOs, we compute the cross-sectional variance of secondary market returns of IPOs underwritten by that bank, employing 3-,6-,9- and 12-month holding periods, and using the estimated "implicit offer price" as the benchmark price. We also compute cross-sectional variance of pre-IPO price revisions, again for IPOs underwritten by that bank. The variance ratio is defined as the ratio of these two statistics. The high-rep subsample consists of underwriters with Carter-Manaster rank 8.5 or higher, and the low-rank subsample consists of underwriters with Carter-Manaster rank below 6.75. Panel A: 1980-2005 Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 145 25 2.178 1.725 Low vs. High 0.011 6-month VR 145 25 2.734 2.075 Low vs. High 0.000 9-month VR 145 25 2.881 2.412 Low vs. High 0.008 12-month VR 145 25 2.953 2.534 Low vs. High 0.012 Panel B: 1996-2000 Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 41 13 1.782 1.983 Low vs. High 0.503 6-month VR 41 13 2.122 2.190 Low vs. High 0.800 9-month VR 41 13 2.446 2.255 Low vs. High 0.477 12-month VR 41 13 2.527 2.410 Low vs. High 0.713 Panel C: Other time Number of Banks Average Log Variance Ratio Low-Rep High-Rep Low-Rep Banks High-Rep Banks Pr > t 3-month VR 120 23 2.233 1.500 Low vs. High 0.000 6-month VR 120 23 2.855 1.940 Low vs. High 0.000 9-month VR 120 23 2.994 2.391 Low vs. High 0.002 12-month VR 120 23 3.123 2.549 Low vs. High 0.001 1 Table 6: The Determinants of Underwriters' Variance Ratios using Pchange This table presents the the coefficient estimates, and heteroscedasticity-consistent standard errors in the parentheses, of OLS regressions. The dependent variable is the observation for each underwriter (with at least 5 IPOs in our sample) of the following statistic: the log of the cross-sectional dispersion in secondary market returns relative to the estimated "implicit offer price", divided by the cross-sectional standard deviation of price revisions. White's correction is employed to estimate standard errors, which are enclosed in parentheses. Statistical significance at the 1%, 5% and 10% level are denoted with superscripts ***,** and *, respectively. Independent Variables (N=205) Dependent Variables VR3 VR6 VR9 VR12 VR3 VR6 VR9 VR12 -0.155*** -0.183*** -0.187*** -0.141*** -0.107** -0.093* -0.144** -0.041 Bank Reputation (0.042) (0.038) (0.044) (0.043) (0.052) (0.052) (0.063) (0.058) -5.556** -4.032* -3.896 -1.892 IPOHeat (2.283) (2.266) (2.393) (2.782) -0.226 -0.317** -0.168 -0.292** Filing Amount (0.145) (0.132) (0.140) (0.128) -0.212 0.064 -0.188 0.155 Age (0.174) (0.163) (0.200) (0.196) -0.543 -0.573 -1.022** -0.755 VC-backed (0.561) (0.494) (0.492) (0.531) 0.732 0.912* 0.847* 0.885* Hi-Tech (0.560) (0.499) (0.511) (0.515) R² 0.084 0.127 0.095 0.059 0.178 0.223 0.172 0.127 2 Table 7: Time-Series Volatility Comparison This table presents the mean aftermarket daily return volatility of IPO firms. Firms are classified into terciles according to the Manaster-Carter reputation ranking of their underwriters and comparisons are made between those underwritten by low-reputation banks and those by high-reputation banks. The last column shows the T-test results indicating whether the differences in the mean volatility are statistically significant. Panel A: 1980-2006 Low-Reputation High-Reputation Comparison (n=2116) (n=2497) Pr > t 3 months 0.0017 0.0020 0.000 6 months 0.0019 0.0020 0.021 9 months 0.0021 0.0020 0.209 12 months 0.0023 0.0020 0.000 Panel B: 1980s Low-Reputation High-Reputation Comparison (n=879) (n=628) Pr > t 3 months 0.0010 0.0008 0.001 6 months 0.0011 0.0009 0.002 9 months 0.0012 0.0009 0.000 12 months 0.0013 0.0010 0.000 Panel C: 1990-95 Low-Reputation High-Reputation Comparison (n=610) (n=644) Pr > t 3 months 0.0018 0.0011 0.000 6 months 0.0020 0.0012 0.000 9 months 0.0022 0.0012 0.000 12 months 0.0025 0.0012 0.000 Panel D: 1996-2000 Low-Reputation High-Reputation Comparison (n=511) (n=864) Pr > t 3 months 0.0029 0.0038 0.000 6 months 0.0032 0.0037 0.002 9 months 0.0037 0.0036 0.885 12 months 0.0039 0.0036 0.142 Panel E: 2001-06 Low-Reputation High-Reputation Comparison (n=116) (n=361) Pr > t 3 months 0.0015 0.0011 0.079 6 months 0.0015 0.0012 0.082 9 months 0.0015 0.0011 0.058 12 months 0.0016 0.0011 0.036 1 Table 8: Explaining Time-Series Return Volatilities The table shows the results of regressing a firm's aftermarket daily return volatility on firm characteristics, underwriter reputation and time period indicators. The bubble dummy is equal to 0.001 if an IPO is conducted in 1999 or 2000, and 0 otherwise. All other explanatory variables are scaled down by a factor of 1000 to secure proper magnitudes for estimated coefficients. Enclosed in parenthesis are standard errors of estimated coefficents above. Statistical significance at the 1%, 5% and 10% level are denoted with superscripts ***,** and *, respectively. 3-month 6-month 9-month 12-month Bank Reputation -0.005 -0.040*** -0.080*** -0.109*** (0.014) (0.013) (0.014) (0.015) IPOHeat 0.079*** 0.074*** 0.076*** 0.080*** (0.021) (0.019) (0.020) (0.022) Filing Amount -0.222*** -0.213*** -0.215*** -0.237*** (0.030) (0.027) (0.029) (0.031) Age -0.156*** -0.131*** -0.146*** -0.166*** (0.023) (0.021) (0.022) (0.024) VC-Backed 0.333*** 0.413 0.385*** 0.383*** (0.055) (0.051) (0.053) (0.057) Hi-Tech 0.697*** 0.662*** 0.625*** 0.620*** (0.054) (0.049) (0.052) (0.056) Bubble dummy 4.113*** 3.844*** 3.723*** 3.684*** (0.076) (0.070) (0.074) (0.079) Adj. R² 0.411 0.426 0.385 0.355 1 Table 9: Partial Adjustment and Underwriter Quality This table presents the coefficient estimates, and heteroscedasticity-consistent standard errors in parentheses, of OLS regressions in which the dependent variable is IPO underpricing. Statistical significance at the 1%, 5% and 10% level are denoted with superscripts ***,** and *, respectively. REV + equals REV if REV > 0, and 0 otherwise; REV - equals REV is REV <0, and 0 otherwise. COMPS + equals COMPS if COMPS >0, and 0 otherwise; COMPS - equals COMPS if COMPS <0, and 0 otherwise. In the lower panel of the table, Cramer's test is used to test the equality of R2 for the long specification. Full Sample Bottom Tercile Medium Tercile Top Tercile (1) (2) (1) (2) (1) (2) (1) (2) REV 0.845*** 0.338*** 0.606*** 1.166*** (0.068) (0.102) (0.045) (0.097) REV+ 1.035*** 0.185 0.909*** 1.318*** (0.155) (0.134) (0.084) (0.163) REV- 0.559*** 0.516*** 0.281*** 0.840*** (0.076) (0.093) (0.080) (0.113) COMPS 0.165*** 0.212*** 0.180*** 0.175 (0.047) (0.057) (0.054) (0.109) COMPS+ 0.162*** 0.160** 0.213*** 0.196 (0.060) (0.074) (0.066) (0.139) COMPS- 0.131 0.686*** -0.101 -0.051 (0.154) (0.209) (0.216) (0.299) IPOHeat 0.844*** 0.810*** 0.376*** 0.408*** 0.758*** 0.725*** 0.979*** 0.941*** (0.048) (0.046) (0.088) (0.086) (0.066) (0.063) (0.075) (0.074) Filing Amount -0.023*** -0.025*** -0.021*** -0.017** -0.014** -0.014** -0.024** -0.023** (0.006) (0.006) (0.008) (0.008) (0.006) (0.006) (0.011) (0.011) Age -0.023*** -0.022*** -0.020*** -0.020*** -0.005 -0.003 -0.031*** -0.030*** (0.004) (0.004) (0.005) (0.005) (0.005) (0.004) (0.008) (0.008) VC-Backed 0.039*** 0.033*** 0.019 0.025 -0.002 -0.012 0.073*** 0.068*** (0.011) (0.012) (0.016) 90.016) (0.015) (0.015) (0.025) (0.025) Hi-tech 0.034*** 0.028*** 0.015 0.021 0.023* 0.015 0.051** 0.044** (0.009) (0.009) (0.014) (0.014) (0.013) (0.013) (0.020) (0.021) λ 0.066** 0.047 0.021 0.072 0.091** 0.028 0.060 -0.004 (0.030) (0.035) (0.048) (0.057) (0.038) (0.042) (0.065) (0.066) Bank Reputation 0.010*** 0.009*** (0.003) (0.003) Adjusted R2 0.388 0.393 0.108 0.116 0.417 0.432 0.466 0.469 Low vs. Medium Medium vs. High Low vs. High Equality of R2 Z -94.85 -15.91 -107.24 for specification (1) Pr > |Z| 0.000 0.000 0.000 1 Table 10: Impact of Pre-Offering Information on Underpricing This table shows how many standard deviations' change is caused by a one-standard-deviation shock in pre-offering price revision (REV ), comparable firms return (COMPS ) and mean underpricing during the 30 days prior to an IPO (IPOHeat ). Panel A doesn't differentiate positive and negative information; Panel B look at the impact of positive (REV + ) and negative (REV - ) price revisions, positive (COMPS + ) and negative (COMPS - ) industry return, and mean past underpricing (IPOHeat ) on underpricing. Panel A: Not Differentiating positive and negative information causes change in underpricing ( # of standard deviations) Full Sample Low Tercile Med Tercile High Tercile 1 st. dev. shock to REV 0.38 0.20 0.36 0.43 1 st. dev. shock to RInd 0.08 0.25 0.08 0.04 1 st. dev. shock to Spillover 0.42 0.24 0.47 0.42 Panel B: Differentiating positive and negative information causes change in underpricing ( # of standard deviations) Full Sample Low Tercile Med Tercile High Tercile 1 st. dev. shock to REV_P 0.32 0.07 0.33 0.36 1 st. dev. shock to REV_N 0.14 0.20 0.10 0.15 1 st. dev. shock to RInd_P 0.08 0.19 0.09 0.04 1 st. dev. shock to RInd_N 0.01 0.08 -0.01 0.00 1 st. dev. shock to Spillover 0.40 0.26 0.45 0.40 1 1