Key Investors in IPOs

Transcription

1 Key Investors in IPOs David C. Brown Sergei Kovbasyuk March 15, 2017 Abstract We statistically identify institutional investors who persistently hold US IPOs with high initial returns. As a group, these key investors holdings are strongly related to initial returns and offer price revisions, more so than any other variables. Key investors are better informed than other investors; their trades predict future returns and their participation more strongly relates to initial returns when they specialize in the IPO firm s industry. Instrumenting key investor participation with pre-ipo information we show that key investors industry specializations, and not their underwriter relationships, predict initial returns. JEL Classifications: G23, G24, G32 Keywords: IPO Underpricing, Institutional Investors, Underwriters We would like to thank Bruno Biais, Andrew Ellul, Croci Ettore, Steve Foerster, Jerry Hoberg, Marco Pagano, Hong Ru, Ann Sherman, Mitch Towner, Aazam Virani, Bill Wilhelm, and seminar participants at the University of Colorado, University of Arizona, Arizona State University, Einaudi Institute for Economics and Finance, University of Naples Federico II, University of Texas at San Antonio, 2015 Northern Finance Association, Paris December 2015 Finance Meetings, and 2016 China International Conference in Finance for their helpful insights and suggestions. We would also like to thank Jay Ritter for making his data available. The Online Appendix is available at Eller College of Management, University of Arizona, dcbrown@ .arizona.edu Einaudi Institute for Economics and Finance, skovbasyuk@gmail.com. 1

2 1 Introduction In recent decades and across the globe, initial public offerings (IPOs) have experienced significant first-day price increases, averaging upwards of 15% (Ritter and Welch, 2002; Loughran and Ritter, 2002; Ljungqvist, 2007; Loughran, Ritter and Rydqvist, 1994). The significance, persistence and wide-spread nature of these initial returns have puzzled researchers and spawned a large theoretical and empirical literature. The existing literature has mainly focused on mechanisms that explain initial returns, but has not made a serious attempt to identify the investors that benefit from them. Our paper fills this gap. We first identify a group of institutional investors, termed key investors, that are persistently associated with high-initial-return IPOs. Our methodology uses institutional investors 13F filings to proxy for IPO participation, and for a given year identifies investors who are associated with statistically-significant abnormal initial returns. An institutional investor is classified as key in a given year if the average initial return on the investors 13F holdings is in the top 1% of the bootstrapped distribution of average initial returns for that year. On average, each year we classify 11% of institutional investors as key investors. A priori, different investors could be classified as key each year, as would be the case if investors association to high initial returns was due to random luck. However, our measure is persistent and key investors continue to be associated with high initial returns in the future: 39% of key investors in a given year are classified as key investors in the following year. Furthermore, the measure is persistent for 10 years, suggesting that key investors employ certain strategies or have certain traits that lead to frequent and repeated participation in high-initial-return IPOs. To learn about the primary drivers of initial returns, we attempt to distinguish what makes the investors key. For instance, key investors could be buying IPOs of firms with characteristics associated with high initial returns, such as being high-tech or VC-backed. Alternatively, key investors may be institutions that are better able to utilize other public information, as identified by Field and Lowry (2009), or they may posses superior non-public information, consistent with many models of bookbuilding. Finally, key investors may be deliberately favored by underwriters, consistent with theory and evidence regarding agency concerns in the underwriting process. We show that key investors association to IPOs with abnormal initial returns can t be explained by their being attracted to IPOs observable characteristics. First, we document a strong relation between the number of key investors holding shares after a given IPO and 2

3 (a) Key Investors (b) Non-Key Investors Figure 1: Key investors participation is positively related to initial returns. the IPO s initial returns. 1 Figure 1(a) shows this finding: the number of key investors is positively related to initial returns. In univariate regressions, the number of key investors explains 25% of the variation in initial returns, more than any other variable. If key investors are simply attracted to firm characteristics, then the positive relation should disappear once we control for common explanatory variables from the IPO literature. Including control variables, a one-standard-deviation increase in key investors participation increases initial return 11% (from the average initial return of 19% to 30%). This suggests that at least some purchases by key investors are not driven by observable common factors. In contrast, Figure 1(b) shows non-key institutional investors participation is not related to initial returns. Many existing theories are consistent with our finding that key investors tend to persistently participate in IPOs with abnormal initial returns. For example, some investors may receive more underpriced shares due to informational advantages (Rock, 1986; Benveniste and Spindt, 1989; Sherman and Titman, 2002), their abilities to add value to firms (Stoughton and Zechner, 1998), or favoritism by investment banks (Aggarwal, Krigman and Womack, 2002; Loughran and Ritter, 2004; Hao, 2007). Many different theories have, at least to some extent, received empirical support. 2 Such a diversity of theories with supporting evidence suggest that a number of factors can contribute to significant initial returns. As examples, studies have documented evidence of favoritism in IPOs primarily during the tech bubble (Reuter, 2006; Ritter and Zhang, 2007), and evidence that allocations reward 1 The results are robust to using the percentage of shares held by key investors instead of the number of key investors. In a horse race between alternative measures, the number of key investors dominates. See Table 11 for details. 2 Ritter and Welch (2002), Ljungqvist (2007) and Ritter (2011) review the empirical and theoretical underpricing literature. 3

4 information has originated from small, proprietary data sets (Cornelli and Goldreich, 2001; Aggarwal, Prabhala and Puri, 2002). Using IPOs in the US between 1985 and 2014 allows us to analyze key investors in both a large cross section and large time series without favoring any particular IPO theory. Key investors characteristics suggest that information-based, and favoritism-based mechanisms may be at play. Overall, key investors portfolios are more concentrated in IPO firms industries and key investors tend to participate in a larger portion of underwriters past offerings. Interestingly, high industry concentration is more pronounced for smaller (belowmedian AUM) key investors, while the strength of the underwriter relationship is stronger for larger key investors. Smaller key investors are also more likely to be hedge funds, consistent with their being more informed. To gauge the importance of key investors industry specializations and underwriter relationships, we relate initial returns to the participation of specialist and relationship investors. We classify an investor as a specialist if his portfolio shows a higher-than-average weighting in the IPO firm s industry, and we classify an investor as related to the underwriter if the investor participated in at least two of the underwriter s last 10 offerings. We find that initial returns are lower when more underwriter-related key investors report holdings, while initial returns are higher when more specialist key investors report holdings. This evidence suggests that key investors information is more likely to be driving the relation between their participation and initial returns. Four additional tests support information-based theories of underpricing. First, we find that the relation between the number of key investors and the initial returns is non-linear, which is also suggested by the convex shape of Figure 1(a). This is consistent with Sherman and Titman (2002) s idea of underwriters using extreme underpricing to compensate key investors for information. Second, the relation between number of key investors participating and initial returns is stronger for harder-to-value firms, in which information from investors is likely more valuable. Third, we show that the relation between initial returns and participation is significant for both large and small key investors, and the point estimate is higher for small key investors. In addition to supporting an information-based explanation, this evidence is inconsistent with large investors bargaining power driving our results. Finally, we show a strong positive relation between key investors participation and offer price revisions, consistent with underwriters adjusting offer prices conditional on key investors information. This effect is also stronger for smaller key investors, suggesting that small specialist investors may be affecting price revisions. Collectively, these results provide positive support that key 4

5 investors information is a determinant of initial returns. We find no evidence that underwriters desires to generate kickbacks for some investors may lead to key investors purchasing underpriced shares. Kickback-based explanations suggest that the number of shares allocated to key investors, not the number of key investors participating, should relate to initial returns. In a horse-race between the two, only the number of key investors participating is significantly related to initial returns, which is unusual if one believes that the underwriter wants to generate rents for favored investors. As a final test of initial returns, we predict the number of investors and key investors in each IPO and use the predicted values as instruments in an instrumental variables analysis. We run a probit regression of investors participation in an IPO using detailed information about investors known at the date of the IPO, including investor s portfolio concentration in the industry of the IPO firm, churn ratio, past relationships with the underwriter conducting the IPO, etc. The estimates show that investors that tend to overweight the IPO firm s industry are more likely to buy the shares, as are investors that had interacted with the underwriter in the past. The fact that 13F holdings reported by funds after the IPO can be predicted with information available before the IPO suggests that 13F holdings do contain information about investors IPO participation and their demands for shares in IPOs. We then instrument for the actual number of key and other institutional investors reporting 13F holdings in IPOs using the aggregated probit estimates of individual investors participation and the standard control variables from our initial return regressions. We regress initial returns on these instruments and standard controls and find that the instrumented number of total institutional investors positively predicts initial returns, while the instrumented number of key investors is not related to initial returns. This finding supports the classic Rock (1986) model of underpricing: high expected participation of institutional investors creates the winner s curse problem for other investors. As a result, the underwriters deliberately lower the offer price in order not to lose other investors, leading to positive initial returns. The fact that predicted participation of key investors is not related to initial returns, suggests that key investors purchases of high-initial-return IPOs stem from their private information, which is not observable prior to the IPO. 3 The lack of a relation to initial returns also indicates that key investors association with abnormal initial returns is unlikely to come from their relationships with underwriters, as underwriter relationships are known prior to the IPO, and controlled for in our probit model. 3 Note, that number of key investors is only around 10% of the number of institutional investors reporting holdings of IPO shares in 13Fs each year. 5

6 Because we rely on holdings data, our results could be due to either key investors involvement in the IPO process or unrelated post-ipo buying. While 13F holdings have been used to proxy for allocations in prior studies (Reuter, 2006; Binay, Gatchev and Pirinsky, 2007), we analyze IPO timing to further justify their use. We find that key investors holdings do not systematically differ from other investors holdings based on when an IPO occurs within a quarter. If post-ipo buying were driving key investors holdings, it is likely that IPOs occuring earlier in the quarter would have relatively more key investor participation. The lack of any difference suggests that post-ipo buying is not biasing our measure. Furthermore, the evidence from our analysis of offer price revisions suggests that if post-ipo buying by key investors were driving our results, those investors would need to condition their buying decisions on the portion of the offer price revision that is orthogonal to initial returns. While possible, we view this explanation as unlikely. Our paper makes a number of contributions to the literature. First, we develop a methodology which identifies a group of investors who persistently participate in IPOs with abnormally high initial returns. Most closely related to our work, Liu (2014) studies the persistence of institutional investors performance using Chinese IPOs, finding that institutional investors that performed well in the past tend to perform well in the future. We are the first to document persistent performance in US IPOs for a group of key institutional investors. We also contribute to the literature relating initial returns to distinct groups of investors classified based on characteristics not related to their IPO performance. Hanley and Wilhelm (1995), Aggarwal, Prabhala and Puri (2002), Field and Lowry (2009) and Chemmanur, Hu and Huang (2010) all provide evidence of institutional investors importance to the IPO process. Reuter (2006) and Ritter and Zhang (2007) study funds with close ties to underwriters. In a similar vain, Jenkinson and Jones (2004) uses proprietary data from an underwriter and finds that this underwriter tends to tilt allocations of underpriced shares to long-term investors. While these studies provide insights into various channels contributing to IPO pricing and initial returns, using reported holdings allows us to leverage a larger sample, identify key investors most related to underpricing, and study the most economically significant drivers of initial returns. For instance, if one does not control for participation of key investors, the participation of institutional investors appears related to initial returns, yet this relationship disappears once participation by key investors is controlled for. Several studies have found that investors that provide information are important in the IPO process. Liu et al. (2015) relates investors attention, e.g. attending a road show and forming an opinion, to IPO underpricing. Similarly, Cornelli and Goldreich (2001) using 6

7 propriatary data from an underwriter, documents that this underwriter tends to allocate underpriced shares to investors that submit information bids. Our findings are consistent with these papers; underwriters may be seeking key investors attention and subsequently their information and opinions. The rest of the paper is organized as follows. Section 2 describes the data and Section 3 describes how we identify key investors. We present our main results in Section 4 and consider information-based explanations in Section 5. Section 6 discusses possible alternative explanations, and Section 7 concludes. 2 Data and Sample We identify IPOs using the Thomson Securities Data Corporation (SDC) Platinum Global New Issues database. The sample includes IPOs of U.S. firms common stocks completed between 1985 and As is common in the literature we exclude unit offerings, spinoffs, real estate investment trusts, rights issues, closed-end funds and trusts, and IPOs with an offer price less than five dollars. To be included in the sample, we require that a firm be in the Center for Research in Security Prices (CRSP) database and that at least one institution reports owning shares in the first quarter after the IPO. Holdings data are from Thomson- Reuters 13F Institutional Holdings database. We supplement data from the SDC, CRSP and 13F databases from several sources. Consumer Price Index (CPI) data from the Bureau of Labor Statistics is used to adjust dollar values to year 2000 dollars. Founding dates, monthly underpricing and issuance activity, and underwriter rankings are taken from Jay Ritter s website. 4 The resulting sample includes 4,928 IPOs. Lacking direct data on participation of investors in IPOs, we follow Binay, Gatchev and Pirinsky (2007) and Reuter (2006) and proxy for participation using the first reported institutional holdings data after issuance. This strategy helps to overcome a common limitation in the IPO literature; a lack of data on allocations in IPOs. 5 data allows for alternative interpretations of our results. However, using quarter-end The 13F data noisily identifies investors that were interested and participated in IPOs, yet some holdings originating from post-ipo buying may be a significant factor. In fact, a priori, the role of investors buy- 4 The data are available at 5 Jenkinson and Jones (2004) and Cornelli and Goldreich (2001) overcome this limitation by using detailed, proprietary underwriters data about bids and allocations. In both cases, the data are from a single underwriter. However, the papers find mixed results, possibly due to differences between the underwriters that supplied the data. 7

8 ing shares after the IPO may be as important for determining the offer price as the role of investors receiving allocations, as it is possible that investors buying after the IPO have unsuccessfully attempted to buy shares in the IPO. In other words, investors that bought shares after the IPO may have acquired the information about the firm before or during the IPO, and possibly communicated their willingness to buy shares to the underwriter. While we attempt to distinguish whether 13F holdings are more driven by allocations or post-ipo buying, we acknowledge that our measure cannot definitively separate the two. While using 13F holdings data to proxy for investors participation in IPOs has the above mentioned shortcomings several studies provide evidence that this proxy is highly correlated with actual IPO allocations. Using proprietary data on a sample of 38 IPOs managed by a single underwriter, Hanley and Wilhelm (1995) finds that the correlation between 13F holdings data and actual allocations is 91%. 3 Identifying Key Investors In our definition, key investors are the funds that are associated with abnormal initial returns. To determine these funds, we begin by constructing, on a quarterly basis, a measure reflecting the average adjusted initial returns for the funds recently reported holdings. For each quarter, we consider IPOs over the past 12 months, excluding any funds that did not report holdings in at least 4 IPOs. For each IPO, we adjust realized initial returns by subtracting the month s average initial returns: AdjInitialReturn i = InitialReturn i J(i) j=1 InitialReturn j J(i) (1) where J(i) is the set of IPOs completed in the same month as IPO i. A fund s average adjusted initial return is the average of the adjusted initial returns for the IPOs, over the last 12 months, for which the fund reported holdings: AvgAdjInitialReturn k = Ii AdjInitialReturn i 1 i,k Ii 1 i,k (2) where k indexes funds and 1 i,k equals 1 if fund k reported holding shares in IPO i and I is the set of IPOs over the past year. We rely on statistical methods to determine which funds received abnormal initial returns. In a nutshell, each quarter we identify funds that reported IPOs with initial returns in the 8

9 Figure 2: Scatter-plot of key funds and other funds in 1994, as an example. The solid line represents the threshold at which we are 99% confident (generated from 100,000 random sample portfolios) that the average adjusted initial return is significantly different from zero. top 1% of the bootstraped distribution of average IPO initial returns over previous four quarters. For each quarter, and for each possible number of IPOs received by a fund, we bootstrap distributions of average adjusted initial return. For example, to benchmark a fund that received 10 IPOs in 1994, we would sample, with replacement, 10 IPOs from those that occurred in We then calculate average adjusted initial return for that random sample. We repeat this process 100,000 times for each date and for each number of potential IPOs reported. Finally, we compare each realized value of average abnormal initial returns to the fund s corresponding distribution of randomly generated values. We define key investors (KeyInvestor = 1) as the funds having realized values greater than at least 99,000 of the random draws, equivalent to a statistical threshold of 1% (p-value of 0.01). 11% of fund-year observations meet the 1% threshold. Figure 2 shows realized values at the start of 1994, as an example. The x-axis displays average abnormal initial return, while the y-axis displays the percentage of the IPOs the fund reported holding. Note that even negative abnormal initial return of 10% on the figure can still imply positive average initial return for a fund once one adds the mean monthly initial 9

10 return of 14% for The solid line represents the bootstrapped threshold, while the Xs, which lie to the right of the threshold, represent key investors. Those investors have average abnormal initial return which is statistically greater than zero at the 1% confidence level. The triangles represent other investors, and all lie to the left of the threshold. Table 1 shows summary statistics of key and non-key institutional investors reporting 13F forms. Key investors are larger and older than non-key investors. Key investors more actively churn their portfolios, but key and non-key investors tend to hold IPOs similar lengths of time. While hedge funds are slightly under-represented in the key investor population, the difference is not significant. 6 That key investors are distinct from other institutional investors is our first evidence suggesting that key investors participation in IPOs with abnormal initial returns is not a result of pure luck, but is rather a result of their expertise, ability, or of some other advantage over ordinary institutional investors Table 2 summarizes the most common key investors, showing that a broad range of fund types and sizes are represented. For example, Essex Investment Management Company, a hedge fund and the most frequent key investor, manages a little over $1 billion in assets, while Fidelity and Vanguard are ranked in the top 25 and manage over $400 billion. 7 general, the funds represented are heterogeneous, including the largest and most prominent funds and banks, as well as insurance companies and many smaller and lesser-known funds. Motivated by the heterogeneity of the top key investors, we examine the characteristics of large and small key investors separately. Each year, we split key investors into above and below median based on fund size. The right-most columns of Table 1 show the differences between the two groups. Unsurprisingly, large key investors are much bigger than small key investors. However, the difference in size is striking. Large funds are almost twenty times bigger on average. The larger funds are also substantially older. Smaller funds appear to be more active. They have higher portfolio turnover (Churn) and hold their reported IPO holdings for less time. Furthermore, smaller funds appear to be more specialized in certain industries: on average, their portfolios have one-half of a standard deviation more concentration in the IPO firms industries (for details see Section 5.1). Finally, large funds tend to have relationships with underwriters and interact more with them (for details see Section 4.3). While this may suggest stronger ties between underwriters and large funds, it is also to be expected as large key investors average almost twice as many reported IPO holdings per year. In subsequent analysis, we test whether industry specialization or underwriter 6 We use the hedge fund classifications introduced in Agarwal, Fos and Jiang (2013) and Agarwal et al. (2013). 7 Dollar figures are based on reported 13F holdings. In 10

11 relationships can explain key investors participation in IPOs with high initial returns. 4 Key Investors Are Neither Lucky Nor Favored Having identified key investors for each year, we first check if our classification is a result of pure lack. Indeed, if different institutional investors get lucky and happen to buy IPOs with abnormal initial returns each year, then different investors will be classified as key each year. We show that this is not the case: key investors are persistent, and it is unlikely that their purchases of IPOs with abnormal initial returns stems from pure luck. 4.1 Key investors are persistent and not simply lucky. We construct yearly key investor measures, to avoid overlapping sample periods and mechanical correlations, and find that our key investor measures are persistent, suggesting that purchases of IPOs with high initial returns is not a result of luck. The first row of Table 3 shows that our key investor measure is persistent. 39% of key investors at the beginning of one year are classified as key investors at the beginning of the following year. This is a significant portion as random assignment would suggest only 11% overlap. Furthermore, this persistence continues for 10 years. In each subsequent year, the proportion (relative to random assignment) classified as key investors remains significant. The next three rows show that alternative measures of key investors (using either a 5% threshold, medians rather than means, or both) show persistence as well. The fifth and sixth rows show that using money left on the table (Shares Of f erp rice InitialReturn) in place of initial return yields apparently stronger results. Moving forward, we will use our primary measure of key investors, based on the mean abnormal initial returns over the prior 12 months. While the persistence results are stronger using investors money left on the table, which incorporate the number of shares each investors reports, results which we discuss in Section 6.5 show that incorporating investors shares reported leads to a weaker measure. Supplemental results are also provided in the Online Appendix. Significant persistence of key investors suggests that our key investor measure (i) is not a simple a statistical artifact, and (ii) identifies a group of investors with particular features or abilities. 11

12 4.2 Key investors do not simply react to observable IPO attributes. The prior section establishes the existence of key investors who persistently report holdings of IPOs with high initial returns. We now go a step further and test if publicly observable characteristics of IPO firms enable key investors to choose the deals with high initial returns. For instance, it is well established that VC-backed or tech IPOs are more likely to experience significant initial returns than other IPOs, and key investors may be focusing on these types of IPOs. To establish a baseline, we begin by regressing initial returns on common control variables from the IPO literature and year fixed effects. Column (1) of Table 4 provides the results. As is common in the literature, initial returns are positively related to positive price revisions (P osp ricerevision), the percentage of shares retained by pre-ipo owners (Retention), initial returns of concurrent IPOs (ConcurrentU nderpricing), the underwriter s reputation for initial returns (U W premium) and the price of the offering (InvP rice). Initial returns are negatively related to firm size (LogSize) and age (LogAge), and the percentage of primary shares issued in the offering relative to shares outstanding after the IPO (Expansion). If key investors decisions to buy IPO shares are based on the information that we control for in the above regressions, then if we introduce a variable associated with key investors participation in an IPO in the regression for initial returns, the coefficient associated with this variable should not be significantly different from zero. We introduce two alternative measures of key investors participation in an IPO. First, we define TotalKeyInvShares as the total number of shares reported held by key investors at the end of the quarter following the IPO divided by the number of shares sold in the IPO. T otalkeyinvshares = k K Shares k SharesSold where K is the set of investors who hold shares at the end of the first quarter following the IPO. Second, NumKeyInvestors, counts the number of key investors who hold the firm s stock at the end of the first quarter following the IPO: (3) NumKeyInvestors = k K KeyInvestor k. (4) A number of theories could lead to either of these being more strongly related to initial returns. For example, if key investors are favored by underwriters due to their ability to earn kickbacks, the percentage of shares held by key investors may better capture that 12

13 motivation. However, information-based theories, such as Sherman and Titman (2002), suggest that the number of key investors may be important, as the underwriter is likely to elicit information depending on the value of that information to the pricing process and the firm. Accordingly, we use both measures separately and together to determine which better measures the relation between key investors participation and initial returns. Columns (2) and (3) of Table 4 illustrate that key investors participation in IPOs is positively related to initial returns in the presence of common controls. The coefficients on both T otalkeyinvshares and N umkeyinvestors are both strongly significant (t-stats of approximately 8 and 11). Furthermore, in column (3), NumKeyInvestors is the variable most significantly related to initial return, and the economic magnitude of the effect is strong (a one-standard-deviation increase in key investors participation is associated with a 11% increase in initial return). 8 Including controls allows us to test reject the hypothesis that key investors are simply attracted to offer characteristics (e.g. firm size, VC-backed, primary versus secondary shares sold, etc.) that are associated with high initial returns. This suggest that IPO participation by key investors is driven by other factors, possibly not commonly known at the time of the IPO. We consider our two measure of key investors participation concurrently in Column (4). When combined in the same regression, NumKeyInvestors remains significantly related to initial return, and the economic magnitude of the relation is nearly unchanged. However, TotalKeyInvShares is no longer significantly related to initial return. This is our first evidence that suggests informational motivations may be the reason for key investor participation in high-initial-return IPOs. 4.3 Key investors relation to initial returns is not due to underwriter relationships Relationships between underwriters and funds may explain strong association between key investors IPO participation and initial returns. Underwriters may allocate the offerings that are expected to experience high initial returns to favored clients. To analyze this possibility, we track underwriter-fund relationships over time. We consider a fund and underwriter to be related if the fund has reported holdings in at least 8 This result is robust to separate analysis of the periods , and Alternatively, using the ratio of the number of key investors to the number of total investors gives qualitatively similar results. 13

14 2 of the underwriter s last 10 IPOs (within the last 5 years). 9 For each IPO, we define NumUW RelatedInv as the number of investors who report holdings and are related to the offering s underwriter. It is possible that the majority of key investors are also related investors, so we construct N umkeyu W RelatedInv, which counts key investors who report holdings and are related to the underwriter. Interestingly, more related investors report holdings in IPOs than key investors. For the average IPO, 10.8 related investors report holdings and 6.3 key investors report holdings, of which, 3.9 are also related to the underwriter. Table 5 shows that initial returns are positively related to key investors participation but less so to related key investors participation. Column (1) shows our baseline results for comparison, while Column (2) shows that compared to N uminstinvestors, N umu W RelatedInv is positively related to initial return. In other words for ordinary, non-key institutional investors, relationships with underwriters positively relate to initial returns. The opposite holds for key investors. Column (2) shows that incrementally, N umkeyu W RelatedInv is negatively related to initial return and N umkeyinvestors is positively related to initial return. That is, the number of key investors who also have a relationship with the underwriter is less strongly related to initial return than the number of key investors unrelated to the underwriter. Related key investors frequent participation in an underwriter s offerings may allow that underwriter to lower initial return as part of a repeated game as discussed in Benveniste and Spindt (1989). Altogether, these test suggest that underwriter-fund relationships are not driving the positive relation between key investors participation and initial returns. 5 Key Investors Appear To Be Informed One common explanation for the strong relationship between key investors IPO participation and high initial returns is that key investors are better-informed investors or are investors who add value to firms. Both of these explanations have rich theoretical backgrounds. Beginning with Rock (1986) and Benveniste and Spindt (1989), many models have linked information asymmetry to high initial returns. Similarly, many models have linked various value-adding activities to high initial returns. As examples, Mello and Parsons (1998) and Stoughton and Zechner (1998) propose investors add value through monitoring, Holmstrom and Tirole (1993) and Brown (2015) consider investors impacts on firm value by increasing price informativeness, and Banerjee, Hansen and Hrnjić (2009) focuses on investors long- 9 A similar measure is used in Gondat-Larralde and James (2008). 14

15 term holding. All of these theories commonly predict that key investors participation in IPOs is positively correlated with initial return, and importantly, most of the value-add theories rely on investors being informed to some degree. As a result, most of our tests are not able to distinguish explicitly among the information and value-add theories. To simplify exposition going forward, we refer to tests of these many hypotheses as tests of information theories. 5.1 Key investors industry specializations relate to initial returns. If funds are providing information in the IPO pricing process, it seems likely that they do so in industries in which they specialize. 10 Therefore, if information is driving the relation between key investors participation and initial return, then this relation is likely stronger when those key investors are more specialized in the IPO firm s industry. To test this hypothesis, we construct a measure of fund specialization based on reported 13F holdings. In each quarter, funds reported holdings are divided into Fama-French 48 industries, and each fund is classified as a specialist in an industry if it holds more that the mean percentage of assets in the industry. 11 Thus, for each IPO we can identify the IPO firm s industry, and compute the number of investors and the number of key investors specializing in this industry that report holdings of the firm s shares: N umspecialistinv and N umkeyspecialistinv. We introduce these two measures as additional controls in the initial returns regressions. Table 5 displays results consistent with key investors specializations relating to initial return. Column (3) shows that initial return is higher when more institutional investors specializing in the IPO firm s industry participate in the IPO. More specialist key investors are also related to higher initial returns, with a coefficient almost three times larger than that for specialist investors. Column (4) also shows the effect is stronger for key investors than non-key investors, and does so while controlling for the underwriter relationships tested in the prior section. These findings are consistent with key investors industry expertise and likely superior information about IPO firms driving initial returns. 5.2 Key investors matter more for hard-to-value firms. If key investors are providing valuable information in the IPO, it is likely that this information is more valuable in some IPOs than others. If this is the case, key investors presence should 10 Kacperczyk, Sialm and Zheng (2005) shows that funds concentrate their holdings in industries in which they have informational advantages. 11 Classification data are available at 15

16 matter relatively more in IPOs with more uncertain valuations. In general, growth options are more difficult to value than assets-in-place, so we measure firms based on the percentage of their value attributable to growth options. We predict that the relation between key investors participation and initial return will be stronger for those firms whose values are more predominantly driven by growth options. We follow Benveniste et al. (2003) in using the present value of growth options, P V GO, as a measure of valuation uncertainty. P V GO = E[P ] EP S/R E[P ] (5) where E[P ] is the midpoint of the offer price filing range and EP S/R is the present value of the issuing firm s current earnings at the time of the IPO discounted at the industry cost of capital. The lower the value of P V GO, the less speculative the offering. In our sample, the mean(median) P V GO is 0.75(0.93), so 75% of the average company s offer price reflects future growth-option value. We test whether key investors are more important for pricing in hard-to-value IPOs by interacting key investors participation with P V GO and a dummy variable indicating whether the firm had negative earnings prior to the IPO (which is consistent with higher growth-option value). We expect the interaction to be positive for P V GO and N egearnings. Columns (2) and (3) in Table 6 display the results. As predicted, firms having more of their value in growth options display a stronger relation between key investors participation and initial return, as do firms with negative earnings. This evidence is again consistent with key investors having superior information about IPOs and them bringing that information to the IPO process. 5.3 Large numbers of key investors are associated with extreme initial returns. Another test of key investors informational role is based on the bookbuilding model of Sherman and Titman (2002). In their model, an underwriter compensates investors for costly information production by lowering the offer price and generating high initial return (a practice called underpricing ). To maximize mechanism efficiency, the underwriter concentrates underpricing in offerings where virtually all investors report good information. This results in a skewed distribution of underpricing and initial returns, with a few hot issues having enor- 16

17 mous price jumps. 12 This suggest a non-linear relation between key investor participation and initial returns, with extreme initial returns occurring when large numbers of key investors report holding shares. Given these observations, we expect that NumKeyInvestors 2 will be positively related to initial returns. Column (1) of Table 6 shows that the coefficient on NumKeyInvestors 2 is significantly positive, confirming the non-linear relation between key investors participation and initial returns. Figure 1(a) is also consistent with this finding, showing a convex shape and average initial returns over 80% for IPOs with the most key investor participation. While the non-linear relation does not rule out other theories, it provides support for key investors being informed. 5.4 Offer price revisions are related to key investors participation. Offer price revisions provide additional support for key investors being informed rather than receiving kickbacks from underwriters. In a typical information theory of book-building, the underwriter collects information from informed investors and revises (adjusts) the offer price accordingly (e.g., Benveniste and Spindt (1989)). If investors provide positive information and order many shares, the offer price is revised upwards. If key investors are informed, this theory predicts a positive relationship between the offer price revision and the participation of key investors. Table 7 shows that N umkeyinvestors is positively related to offer price revisions. By itself, Column (1) shows that N umkeyinvestors explains a significant fraction of variation in offer price revisions. Column (2) provides a baseline specification with controls, and Column (3) shows that N umkeyinvestors maintains as an important explanatory variable in the presence of controls. Finally, Column (4) shows that the number of small key investors, with below median value of assets under management in the sample, has a stronger impact on offer price revision than the number of large key investors. Overall, our results for offer price revisions are consistent with information theories for key investors post-ipo holdings. 13 Were initial returns entirely motivated by underwriters favoring some investors, it is likely that key investors would experience less positive or even negative revisions as underwriters would set offer prices lower to transfer more rents to those investors (and subsequently recapture those rents through other lines of business). While that broadly applies to kickbackrelated explanations, it is important to note that laddering (i.e., illegal price support) can 12 Sherman and Titman (2002) pg. 16. Liu et al. (2015) generates similar predictions. 13 Bubna and Prabhala (2011) and Ljungqvist and Wilhelm (2002) also provide evidence consistent with investors receiving rewards for information revelation during book-building, while Chiang, Qian and Sherman (2010) show sophisticated investors earn better returns in auctioned IPOs. 17

18 generate the opposite prediction. As shown by Hao (2007), laddering can be associated with positive offer price revisions and positive initial returns. 5.5 Industry specialization predicts IPO participation. As an additional test of information theories we estimate the probability that investors report holdings in each IPO using various measures. Using a probit analysis, we relate investors reported holdings to investors characteristics, the underwriter s characteristics and the IPO company s characteristics. 14 We consider many different predicting variables: underwriterfund relationships; fund characteristics (e.g. size, age, and holdings-based measures); and funds industry specializations. We also include KeyInvestor and interaction terms with KeyInvestor. Table 8 show the results of the estimations. As expected, key investors are more likely to report holdings of IPO firms. While a number of additional characteristics influence whether funds report holdings, underwriter relationships and fund specialization (whether the fund is above median in holdings of the IPO firm s industry) are both strong drivers of reported holdings. An additional point is worth making. If key investors are conveying information to underwriters during book-building, then it is likely that those investors are receiving allocations. For key investors to affect pricing, this need not be the case. For example, expressing interest to an underwriter may convey information, but underwriters do not reward all who express interest with allocations. This expression of interest by an investor reflects active participation in the IPO regardless of actual allocation. However, it is unlikely that funds would continue to provide information were they not awarded with allocations from time to time. 5.6 Expected total investor participation, and not expected key investor participation, predicts initial returns. We use the coefficient estimates in Table 8 to estimate the probability of each investor s reporting holdings in each IPO. We then aggregate these probabilities across the key and non-key investors in each IPO, and we use these aggregate probabilities together with other IPO characteristics known at the IPO date to instrument for the actual number of investors and key investors who will report holdings in each IPO. 14 We follow Brown (2015) in estimating probit models of funds end-of-quarter holdings. 18

19 Our instrument also allows us to indirectly test the theory of Rock (1986). If a group of investors are more likely to be informed, then a higher expected number of those investors in an IPO should lead to more underpricing and higher initial returns, due to a more severe winner s curse for the non-informed investors (whose participation is required to complete the offering). While the bookbuilding setting is not strongly linked to the model assumptions in Rock (1986), to the extent that some regular investors are uninformed, we would expect more informed investors to lead to a stronger winner s curse. While it is possible that our key investors impose a winner s curse on other institutional investors, it is also possible that institutional investors collectively impose a winner s curse on retail investors who make up approximately 30% of allocations. Table 9 reports the results of our instrumental-variables analysis. The first two columns present the first-stage estimation of our instruments, and both instruments meet the relevance criterion. The third column shows the second-stage estimates, revealing that total investors instrumented participation is positively related to initial returns, but key investors instrumented participation is not related to initial returns. This is consistent with all investors participation imposing a winner s curse on retail investors, as initial returns are higher when a larger population of the more-informed, more-frequently-participating investors are expected to participate. However, the lack of a positive coefficient on key investors participation suggests that a strong winner s curse is not anticipated by non-key institutional investors. Furthermore, the lack of relation suggests that key investors purchases of high-initial-return IPOs stem from their private information, which is not observable prior to the IPO. Perhaps the expected initial return for institutional investors is sufficiently high that additional compensation for asymmetric information within institutional investors is unnecessary. Finally, the lack of a relation to initial returns also indicates that key investors association with abnormal initial returns is unlikely to come from their relationships with underwriters, as underwriter relationships are known prior to the IPO, and controlled for in our probit model. 6 Alternative Explanations and Robustness We analyze and discuss several alternative explanations for our findings. While we cannot rule them out conclusively, a lack of support for alternatives indirectly supports our main results. We also consider alternative key investor measures, showing that our primary measure is most significantly related to initial returns. 19

20 6.1 Agency-Based Explanations Several of our prior tests suggest that key investors participation in abnormally underpriced IPOs is unlikely to be motivated by their being favored by underwriters in exchange for future business. First, we have shown that investors past relationships with an underwriter are negatively related to underpricing once key investors participation is controlled for, suggesting that favoritism is not driving our results. Second, kickbacks-based arguments suggest that underwriters would favor a small number of trusted key investors in IPOs, particularly those with high underpricing. However, this is not consistent with our finding that extreme underpricing corresponds with large numbers of participating key investors. Third, it would be easier for underwriters to transfer rents by giving many shares to favored investors rather than ration a large number of investors. Our results suggest the opposite: underpricing is strongly related to the number of key investors participating, but not the number of shares they hold. Finally, one would expect that underwriters would revise offer prices moderately in order to transfer more rent to favored key investors, which we do not find. Underwriters desires to lessen their own costs of price support is an agency-based motivation for key investors participation in IPOs that does not directly contradict our findings. Underwriters typically attempt to keep the price of the firm above the opening price for several weeks or more after the IPO. To ease this process, underwriters can allocate shares to investors who implicitly commit to buying additional shares in the secondary market, an illegal process called laddering. Hao (2007) builds a theoretical model which shows that allocating more shares to laddering investors increases underpricing, arguing that it is possible that key investors are those most likely to ladder and aid in the underwriter s price support activities. However, as in theories related to kickbacks, the shares held by key investors and not the number of key investors should be related to underpricing. Nevertheless, we conducted additional tests of the laddering hypothesis. In addition to overall key investors participation, we separately included a measure of the participation of key investors who frequently sell their holdings in the quarter after the IPO, a behavior that is consistent with laddering. We found no significant difference between regular or frequently-selling key investors participation and further attempts to relate frequently-selling key investors to laddering did not yield robust results. This suggests that laddering or price support by key investors is not driving our main results. 20