Why are IPO Investors Net Buyers through Lead Underwriters?

Transcription

1 Why are IPO Investors Net Buyers through Lead Underwriters? JOHN M. GRIFFIN, JEFFREY H. HARRIS, AND SELIM TOPALOGLU * November 12, * Griffin is at the University of Texas at Austin, Harris is at University of Delaware, and Topaloglu is at Queen s University. Griffin can be reached at john.griffin@mccombs.utexas.edu, Harris at harrisj@lerner.udel.edu, and Topaloglu at stopaloglu@business.queensu.ca. We thank the Nasdaq Stock Market for providing essential data and Kelvin Huang, Stoyan Iliev, Jie (Olivia) Lian, Yongjun (Dragon) Tang, Michael Yates, and especially Johan Sulaeman, for excellent research assistance. We are also grateful for helpful discussions with Aydogan Alti, Kirsten Anderson, David Brown, Keith Brown, Daniel Dorn, Sonia Falconieri, Eric Falkenstein, Mark Flannery, Bruce Foerster, James Griffin, Jay Hartzell, Jim Lastoskie, Darius Palia, Bob Parrino, Jay Ritter, Jeff Smith, Matt Spiegel, Laura Starks, René Stulz, Paul Tetlock, Ivo Welch, Donghang Zhang, and seminar participants at Case Western University, the University of Florida, Koc University, University of Missouri Columbia, Queen s University, Rutgers University, the University of Texas at Austin, and Tilburg University.

2 Why are IPO Investors Net Buyers through Lead Underwriters? Abstract In Nasdaq IPOs from 1997 to 2002, clients of the lead underwriter bought shares worth $35.36 billion on the first day of public trading but sold shares worth only $21.45 billion, leading to a net buy imbalance of $13.91 billion or 8.79 percent of the shares issued. We investigate several explanations for this buy imbalance through the lead underwriter. First, contrary to clientele demand-based explanations, brokerage houses with large client buying when the broker is the lead underwriter experience net client selling when the broker is a comanager or another syndicate member. Second, inconsistent with the hypothesis that underwriter clients obtain superior information on certain IPOs, we find that the strong buying activity from lead underwriter clients is present in 85 percent of all IPOs. Third, we find that large block trades receive worse execution through the bookrunner than through non-syndicate market makers, so differential execution quality cannot explain the net buy imbalances. Fourth, inconsistent with initial client buying being driven by long-term shareholders, IPOs with large first-day net buying through the lead underwriter experience more institutional investor selling over the subsequent quarters. Fifth, the strong net buying activity through the lead underwriter is driven by large block trades, widely present in IPOs of various degrees of underpricing, and more prevalent for underwriters issuing multiple IPOs. Overall, our findings are consistent with allocation theories where underwriters extract rents from clients in the form of aftermarket price assistance and increased IPO demand.

3 In 1999 and 2000, shares of initial public offerings (IPOs) soared an average of 71.7 and 56.1 percent on the first day of trading, respectively, transferring more than $65 billion of wealth from issuers to fortunate shareholders through underpricing. 1 While many have focused on the puzzle of why issuing firms leave large sums of money on the table, it is perhaps even more perplexing that sophisticated underwriters also presumably forego the substantial fees associated with this underpricing ($4.6 billion during ). 2 This paper provides evidence consistent with a partial explanation for this puzzle, namely that underwriters may receive benefits from underpricing in the form of quid pro quos, such as strong aftermarket client demand. Using a unique sample of Nasdaq IPO trading by brokerage houses for the period, we find that on the first day of public trading the clients of the bookrunner (lead underwriter) buy an amount equal to percent of shares issued but sell only percent for a net buy imbalance of 8.79 percent. These patterns of net buying by bookrunner clients and small net selling through other syndicate members are widely prevalent across most brokerage houses, time periods, and IPOs with varying degrees of underpricing. We examine predictions of competing explanations for this phenomenon including those based on clientele effects, underwriter reputation, information from underwriters, familiarity, differential execution quality, strategic allocation, and laddering. Lead underwriter or bookrunner clients generally receive large allocations and hence may be most likely to sell for diversification purposes. However, there are many possible reasons why clients of the bookrunner might be net buyers. A popular explanation promoted recently in the press (Pulliam and Smith (2000a, b)) called laddering finds motivation from models by Fulghieri and Spiegel (1993) and Loughran and Ritter (2002) where underwriters are able to extract indirect rents 1 These numbers are taken from the Ritter and Welch (2002) survey of the extensive IPO literature. 2 $4.6 billion is seven percent of the underpricing in 1999 and 2000 based on a flat fee of seven percent of capital raised (Chen and Ritter (2000)). 1

4 from clients in exchange for underpriced shares. In Fulghieri and Spiegel s model, an investment bank allocates underpriced shares to clients who provide business for other parts of the bank. Loughran and Ritter (2002) provide an explanation for the firm s ex ante choice of an underwriter who is likely to underprice the issue ex post. 3 Because underpricing is lucrative for clients of the underwriter who receive IPO allocations, these clients engage in rent-seeking behavior to increase their probability of receiving shares. Underwriters are judged by both the amount of demand that they can generate for the IPO and their ability to stabilize prices in the aftermarket. The working assumption in the IPO industry is that there is a negatively sloped demand curve. Thus, if a brokerage house fails to generate sufficient client demand for an IPO then subsequent issuers may view the underwriter as a poor promoter. In the case of particularly weak demand, the bookrunner may be forced to provide costly aftermarket price support. In terms of both generating sufficient IPO demand and protecting itself from costly aftermarket support, an underwriter could benefit from encouraging its clients to buy aftermarket shares in exchange for current or future share allocations. There are many other reasons to think that the bookrunner s clients will be stronger net buyers than customers from other brokerage houses. For instance, customers with excess demand for IPOs may migrate to brokerage houses that specialize in issuing IPOs and where they can receive larger share allocations. Yet, in contrast to this clientele explanation, we find that clients are consistently large net buyers when their brokerage house is the bookrunner but are typically small net sellers when the same brokerage house is another member of the syndicate. Similarly, brokerage house reputation does not explain the strong bookrunner client buying. Client familiarity with the IPO could explain the aftermarket patterns but clients of the co-manager, which have access to the 3 Ljungqvist and Wilhelm (2004) provide support for the prospect theory predictions of Loughran and Ritter (2002) by focusing on subsequent security issues. 2

5 road show presentations, generally display aftermarket selling even after controlling for underwriter identity. Strong buying through the bookrunner could reflect better or more important information passed on from the bookrunner to clients that can be used by the clients to take long positions only in IPOs that trade at prices lower than their fundamental value. However, we find that bookrunner client net buying activity is not limited to IPOs that offer good aftermarket value, but prevails in over 85 percent of all IPOs in our sample. Moreover, initial buying through the bookrunner does not bear a relation to intermediate and long-run stock return performance. Buying through the lead underwriter could also be due to the bookrunner offering relatively more attractive prices on block trades to encourage buying. Yet, inconsistent with this explanation, we find that in the trade sizes where most net buying occurs, the bookrunner actually offers slightly worse executions compared to non-syndicate brokerage houses. Zhang (2004) argues that underwriters make use of the fact that institutional investors target positions above a certain size. 4 Since the bookrunner has more of an interest in generating long-term shareholders than other members of the syndicate, they may strategically allocate toeholds to stable long-term institutional shareholders to encourage them to establish even more sizeable positions in the aftermarket. A distinguishing prediction of the strategic allocation hypothesis is that those IPOs with initial bookrunner client buying, if it is due to long-term shareholders, should have a more stable shareholder base over time. In contrast, we find that IPOs with higher levels of bookrunner client buying experience more institutional selling over the four subsequent quarters. The features of the data that are inconsistent with clientele effects, underwriter reputation, familiarity, information from underwriters, differential execution quality, and strategic allocation are all consistent with the laddering explanation. Consistent with institutional traders driving the 4 Aggarwal (2003, p. 115) also notes that institutional investors with a small allocation will either flip or increase their position. 3

6 relation, we find that all the bookrunner client net buying is due to trades for 1,000 shares or more and 85 percent of this net buying is due to trades of 5,000 shares or more. In addition, institutional clients offer unconditional aftermarket buying support with laddering arrangements but this buying is clearly more beneficial to underwriters around the offer price. Consistent with this aspect of laddering, we find positive block trade buying through the bookrunner at all temperatures of IPOs and price ranges even at low price ranges. Arguably, laddering arrangements might be more prevalent for investment banks that issue many IPOs since they can credibly threaten access to future IPOs. Consistent with this prediction, we find more buying through the bookrunner for those underwriters who issue many IPOs. Although our results are consistent with laddering and generally inconsistent with other explanations, non-laddering explanations may explain some of the variation in net buying through the bookrunner and we, of course, cannot rule out other explanations. Our paper is related to a growing body of literature that examines the dual nature of underwriting and market making activities by investment banks (e.g., Schultz and Zaman (1994), Aggarwal (2000), Ellis, Michaely, and O Hara (2000), and Ellis (2003)). 5 We focus less on the underwriter s proprietary trading activity and more on the trades executed as agents for their clients. 6 Nimalendran, Ritter, and Zhang (2004) and Reuter (2004) provide evidence that brokerage house clients may have rebated some of underpricing profits back to the underwriter through excessive trading in other liquid stocks. Our paper evaluates another avenue from which underwriters may extract rents from their clients and contributes to a growing literature (e.g., Ljungqvist and Wilhelm 5 Ellis, Michaely, and O Hara (2000), for instance, find that the lead underwriter becomes the dominant market maker and generates positive trading profits that are increasing in the level of underpricing. Schultz and Zaman (1994) show that underwriters stabilize cold IPOs, while Aggarwal (2000) and Ellis, Michaely, and O Hara find that market makers accumulate inventory in cold IPOs to cover initial short positions generated from the use of the overallotment option. 6 Other papers examining client trading activity have focused on flipping. Aggarwal (2003) finds that IPO flippers represent a relatively small amount of trading volume and flipping (as a percentage of the allocation) is much more likely in IPOs with large abnormal first-day returns. Krigman, Shaw and Womack (1999) find a negative relation between a proxy for institutional flipping and long-run price performance. Aggarwal (2003) and Corwin, Harris and Lipson (2004) suggest that this negative relation may result from the variation in total volume and not necessarily institutional flipping. 4

7 (2003)] that seems to support Ritter and Welch s (2002) conjecture that agency explanations will play a bigger role in the future [IPO] research agenda. The remainder of the paper is organized as follows. Section 1 discusses the alternative hypotheses about differential buying and selling activities across members of the syndicate as well as their distinguishing testable predictions. We discuss the data and summarize the characteristics of our Nasdaq IPO sample in Section 2. Section 3 provides empirical evidence on aftermarket client trading imbalances and Section 4 tests predictions of the previously outlined hypotheses. We explore the tradeoff between bookrunner client and bookrunner proprietary trading as well as their relation to short- and long-run returns in Section 5. We discuss accounting issues of how initial shares sold might arise from short selling or flipping in Section 6 and conclude in Section Explanations for Differential Buying Patterns through the Lead Underwriter We detail a number of explanations for net buying activity through the lead underwriter and map out distinguishing testable implications associated with each hypothesis below. A. Laddering There has been a substantial amount of controversies surrounding the turbulent IPO activity during the internet bubble period. One of the major controversies claims that underwriters allocated issues to favored institutional clients with the understanding that the clients would buy more shares of these IPOs in the aftermarket. This activity is labeled laddering in that it may help to push and support prices to higher levels through artificial demand. 7 Laddering involves a quid pro quo relationship between underwriters and their clients where clients receive IPO share allocations with the implicit or explicit understanding that they would purchase additional shares in the 7 Underwriters are also alleged to have allocated shares to those clients who would rebate some of the underpricing profits back to the underwriter through active trading (perhaps at inflated commission rates) in other liquid stocks. There are charges that underwriters gave shares to favored clients, such as CEOs, hoping that they would use the underwriter in a future offering (known as spinning). 5

8 aftermarket. Underwriters enforce these alleged arrangements by withholding future IPO allocations. 8 Concerns about laddering are not new. In 1961, 1984, and 2000, the SEC issued warning statements that laddering violates antifraud and anti-manipulative security laws. 9,10 Jay Ritter notes that such laddering practices have occurred in the past with penny stock underwriters but generally has not been a focus of allegations against prestigious underwriters until recently (Harris (2001)). Nevertheless, Barry Barbash, former director of the SEC's Division of Investment Management, states that proving this kind of conspiracy as a practical matter is very difficult (Loomis (2001)). 11 The lead underwriter plays a key role in the IPO process. It risks more reputational capital than other syndicate members and is judged by its ability to generate post-ipo demand and provide aftermarket price support. While other syndicate members have limited incentives to support aftermarket prices, the lead underwriter is more likely to coerce its clients to assist with aftermarket support. Therefore, for a particular lead underwriter, a necessary but not sufficient distinguishing feature of the laddering hypothesis is that client buying activity exceeds selling activity. H1.1: Client Net Buying U, Bookrunner > 0 where U denotes a particular underwriter. 8 According to the SEC (Litigation Release No , 2003), an from a J.P. Morgan sales representative says that an institutional customer followed up in the aftermarket exactly as promised (every share through us). 9 Security Exchange Act Release No (April 24, 1961); Report of the Security and Exchange Commission Concerning the Hot Issues Market (August, 1984); SEC Staff Legal Bulletin No. 10 (August 25, 2000). 10 Deneen and Hooghuis (2001) note that recent lawsuits charge that underwriters and certain issuers violated Sections 11, 12(a) (2) and 15 of the Securities Act of 1933 and Section 10(b) of the Securities Exchange Act of 1934 and Rule 10b-5. (See SEC Release No. 4358, 1933, and SEC Release No. 6536, 1934.) 11 The Economist states: Seasoned regulators admit that it will be hard to prove that Wall Street firms were systematically up to no good (Getty (2001)). Perhaps this is why the SEC fined J.P. Morgan only $25 million on October 1, 2001 under Rule 101 of Regulation M instead of imposing larger fines that might have been levied under a 10b-5 violation (Plitch and Hennessey (2003)). Rule 101 of Regulation M prohibits underwriters from inducing people to purchase securities in the aftermarket. 6

9 Perhaps because institutional investors have large aftermarket buying capacity, laddering allegations suggest prearranged unconditional buying support by large institutional clients, which should lead to persistent net buying through the bookrunner more in large trades than in small ones. H1.2: Client Net Buying Large Trades U, Bookrunner > Client Net Buying Small Trades U, Bookrunner Although claims of laddering suggest unconditional buying support, creating demand is most valuable to underwriters when an IPO is doing poorly. If net buying influences demand, first-day IPO prices and returns will be endogenously determined. Nevertheless, for an underwriter who engages in laddering, block buying should be present in cold IPOs. H1.3: [Client Net Buying Large Trades U, Bookrunner > 0 Cold IPO] Since the most likely penalty for a client not purchasing shares in the aftermarket is withholding allocations in future IPOs, it would be easier to enforce such agreements for underwriters that issue many IPOs. H1.4: Client Net Buying Underwriter of Many IPOs, Bookrunner > Client Net Buying Underwriter of Few IPOs, Bookrunner However, since large underwriters also enjoy generally higher reputation and have more to lose if they are caught in a laddering arrangement, this last prediction should be examined loosely. We now turn to examining predictions of other hypotheses. B. Clienteles Certain investors prefer to invest in young companies with high volatility and growth potential like IPO firms. These clienteles may migrate their trading to brokerage houses with a propensity to issue IPOs and it may be natural for them to seek additional shares in the aftermarket. Indeed, Cornelli and Goldreich (2001) and Jenkinson and Jones (2004) find that bidders generally receive a larger pro-rata portion of the IPO when their broker is the bookrunner. To investigate whether aftermarket client buying is due to pent-up demand by IPO investors, we examine client trading by brokerage house. The clientele hypothesis predicts that for a particular underwriter, net 7

10 client aftermarket demand may be greater when the brokerage house is not the bookrunner (relative to demand when the broker is the bookrunner). H2: Client Net Buying U, Bookrunner < Client Net Buying U, Non-lead C. Marketing to create familiarity In a similar vein, a familiarity bias could explain differences in net buying propensity across brokerage houses. Investors are more likely to hold securities they are familiar with (see Kang and Stulz (1997), Coval and Moskowitz (2001), and Huberman (2001), among others). If familiarity drives net aftermarket buying patterns, we should see little difference between bookrunner and comanager client net buying since all are exposed to similar information during the road show. H3.1: Client Net Buying U, Bookrunner Client Net Buying U, Co-manager Underwriters that take multiple IPOs to market may also have larger, more effective marketing forces. Consequently, initial client buying should be greatest for underwriters that are involved with relatively more IPOs (compared to less active underwriters). For bookrunners, this is identical to the prediction from laddering (hypothesis H1.4). However, while laddering specifically applies to bookrunners but not co-managers, the marketing hypothesis should apply to all members of the syndicate that market the security. In this regard, a distinguishing prediction of marketing to create familiarity is that co-managers who issue multiple IPOs should generate more aftermarket demand than those that issue fewer IPOs. H3.2: Client Net Buying Underwriter of Many IPOs, Co-Manager > Client Net Buying Underwriter of Few IPOs, Co-manager D. Informational advantages There are also several informational explanations for net buying differences across brokerage houses. For example, clients of the bookrunner may be better informed than other investors about true IPO values. If they either partially reveal their demand in the primary market or are rationed by the IPO allocation process, they are likely to seek additional shares in the secondary market. If so, 8

11 we would expect bookrunner client buying activity to be associated with positive future abnormal returns. Thus, we examine whether net client buying forecasts future returns. However, it is difficult to know the average investment horizon of the aftermarket investors and long-run return tests lack power. Nevertheless, bookrunner client net buying should occur only in IPOs that initially trade below fair value. Since the median IPO underperforms over most horizons, not more than half of IPOs can be underpriced in the aftermarket. 12 H4: Proportion (Client Net Buying U, Bookrunner >0) < 0.50 E. Reputation The bookrunner s reputation could attract more net buying activity. However, controlling for the reputation of the underwriter (which is known to all traders), we would expect to see consistent demand from both clients and non-clients alike. To examine reputational effects, we focus on trading through underwriters with high reputation only and then examine whether client behavior for a given underwriter varies with the role of the underwriter. H5: [Client Net Buying U, Bookrunner = Client Net Buying U, Non-lead high underwriter reputation] F. Execution quality The bookrunner may also generate client buy imbalances by offering superior buy trade execution (i.e. allowing their customers to buy within posted spreads). We investigate whether the bookrunner offers differential execution quality relative to all other non-syndicate brokerage houses. Because execution varies by trade size, we investigate execution quality across multiple trade-size groups. The execution quality hypothesis predicts that bookrunner clients receive better prices than the clients of non-syndicate market makers (MM). H6: [Bookrunner (% of buy trades mid-point) > Non-syndicate MM (% of buy trades midpoint) Trade-size group] 12 We examine the validity of this assumption for our sample. 9

12 Since the bookrunner imbalances could be due to offering poor prices to those wishing to sell shares, we examine sell trade execution quality as well. G. Strategic allocations to long-term shareholders Bookrunners may strategically allocate shares to investors with long-term interests in the stock to encourage them to buy more shares in the aftermarket. Because long-term shareholders are often thought to be stable institutional investors, the strategic allocation hypothesis predicts, like laddering, that buy imbalances are due to large trades (H1.2). Since the lead underwriter bears primary responsibility for fostering long-term shareholders, the strategic allocation hypothesis also predicts, like laddering, that stronger client block buying persists when the brokerage house is the bookrunner (H1.1 and H1.2). A distinguishing feature of the strategic allocation hypothesis is that strong aftermarket buy imbalances should result in more stable long-term institutional holdings. IPO firms with strong aftermarket bookrunner client buying should have fewer shareholders selling their shares in subsequent quarters than IPOs with weak initial bookrunner client buying. H7: Correlation(Client Net Buying, Long-term Institutional Ownership) > 0 In contrast, laddering predicts that the clients that buy aftermarket shares immediately after the IPO may unwind their positions over time. Although we discuss many possible explanations for bookrunner client net buying and distinguishing testable implications, for brevity, we do not compare and contrast all hypotheses. 13 Nevertheless, we discuss some nuances of the hypotheses and their empirical counterparts when taking the hypotheses to the data below. 13 For example, laddering has opposing predictions to H2, H3.1, and H5. While the laddering hypothesis makes no concrete prediction regarding the proportion of IPOs with positive client net buying (H4), one would expect a bookrunner who engages in laddering to have a high proportion. 10

13 2. Data and Summary Statistics Using the Thomson Financial Securities Data Company (SDC) new issues database we identify IPOs from January 1997 to December 2002 and exclude certificates, ADRs, shares of beneficial interest, units, closed-end funds, REITs, companies incorporated outside the U.S., and IPOs not covered by the Center for Research in Securities Prices (CRSP) data. We also collect characteristics of the offerings from SDC, calculate total shares issued by aggregating all the underwriter allocations for an IPO, and gather Carter-Manaster underwriter reputation rankings from Jay Ritter s website. We restrict our attention to Nasdaq IPOs since we use proprietary Nasdaq data. Our final sample consists of 1294 IPOs, with most listing in the first half of our sample period. We collect Nasdaq clearing data that identifies all trades by brokerage house from January 1997 to December 2002 for the first 21 days of trading in each IPO. In addition to the brokerage house identification, the Nasdaq data identifies both the buyer and the seller in each trade, allowing us to avoid misclassification errors that result from tick test rules. Each side of the trade is classified as to whether the market maker is trading for its own account (as a principal) or handling a trade for a brokerage client (as an agent). Trades handled for clients of a particular broker will either be marked with an agency indicator or with a blank counterparty identification. 14 Thus for each brokerage house in the syndicate, we classify trades according to whether the market maker is executing orders for a client or trading from its own inventory (as a principal). Although each trade is reported only once in our final data, some trades are routed multiple times. The Nasdaq data includes both reported and non-reported legs of routed trades. We check for consistency in assigning whether a market maker acted as a principal or an agent for each leg of routed trades and do not classify trades that are inconsistently reported in each leg of the routing report. We are able to classify percent of trades and percent of volume over the first In discussions with Nasdaq officials, we confirm the blank counterparty reporting standard is uniform throughout the data. 11

14 days of trading and a similar proportion (97.19 percent of trades and percent of volume) on the first day of trading. We manually match the underwriter firms from SDC and the market maker identities in the Nasdaq data, accounting and adjusting for market making firms that have merged, changed names, or used multiple market maker codes (for different trading desks, for instance). We are able to match 96.2 percent of shares issued to a syndicate member. 15 Since our focus is on the trading behavior of clients, we separately analyze trades executed through bookrunners, co-managers, other syndicate members, and non-syndicate members. Table 1 presents some summary statistics for our data, first for the whole period and then by year. Not surprisingly, the number of IPOs is highest in 1999 (390) and 1997 (332) and lowest in 2001 (45) and 2002 (38). For the whole sample period ( ), the mean one-day return is percent and the median return is percent. 16 The year-by-year IPO returns show that the large one-day returns are primarily driven by the and percent average returns in 1999 and The average one-week abnormal return is only slightly higher than the one-day return. However, the one-month return is substantially higher at percent mainly due to percent return in For a typical IPO in our sample, there is one bookrunner, two co-managers, and twelve syndicate members. The average bookrunner underwrites percent of the total shares issued, co-managers underwrite percent, and other syndicate members underwrite the remaining percent. The amount of shares allocated by the bookrunner is typically much greater than the number of shares the bookrunner underwrites (Chen and Ritter (2000)). Unfortunately, we do not 15 The unmatched syndicate members are typically small market makers that are neither the lead nor the co-manager and receive a small proportion of the allocation. As shown in Table 2, trading through these other syndicate members is extremely small in comparison to other groups. Nevertheless, we note that what we label as non-syndicate member volume may contain an extremely small proportion of activity from syndicate members with market maker codes that we were unable to match as such. 16 This is in line with estimates for the late 1990s through 2001 reported by Ritter and Welch (2002). 12

15 have data on the actual allocations. The over-allotment allocation is used in 74 percent of our firms with mean and median over-allotment allocations of and percent, respectively. 3. Trading through Lead Underwriter, Co-Manager, and Other Syndicate Members Table 2 documents patterns of buying and selling behavior for clients of lead underwriters, co-managers, syndicate members, and non-syndicate members on the first day of the IPO. Panel A shows that most of the first-day volume is due to bookrunner clients and clients of market makers that are not part of the syndicate. On average, bookrunner clients buy percent of total shares offered and sell percent, resulting in a first-day net buy imbalance of 8.79 percent of the total shares issued. Co-manager clients are responsible for a much smaller proportion of trading and are small net sellers of about 0.63 percent of shares issued. Consistent with Aggarwal and Conroy (2000), trading through the other syndicate members accounts for only a small proportion of volume with similar amounts of buying and selling. Non-syndicate member clients, who do not receive primary shares through non-syndicate brokerage houses, are net sellers, likely reflecting either short-selling activities or attempts to mask sales by flipping through non-syndicate brokerage houses. We discuss and provide more evidence on this in Section 6. Consistent with Ellis, Michaely, and O Hara (2000), the lead underwriter is a very active market maker, buying back percent of shares issued and selling percent, on average. This net buying (just 1.52 percent) is small compared to the 8.79 percent net buying activity by their clients. Conversely, co-managers and other syndicate members experience little change in inventory positions on the first day of trading. Non-syndicate members and their clients sell to offset bookrunner client buying. 13

16 Panel B of Table 2 presents trading as a percentage of total trading volume and Panel C displays total dollar trading. Imbalances scaled by volume are similar in magnitude to those scaled by shares issued. Bookrunner client buying and selling represent and percent of first-day trading volume, respectively. Panel C shows that over the full sample bookrunner clients buy shares worth $35.36 billion on the first day and sell $21.44 billion. In contrast, the bookrunner buys $36.77 billion and sells $37.06 billion for its own inventory. Despite net buying of 1.52 percent, the net bookrunner dollar position is nearly flat because the bookrunner generally sells stocks at higher prices than purchase prices. Panel D of Table 2 presents the buying and selling through the bookrunner for each year separately. The net buying pattern by bookrunner clients is prevalent throughout the period. In 2000 and 2001, the pattern of net buying by bookrunner clients is extremely strong with over twelve percent net buy imbalances. Bookrunner clients bought only 3.47 percent of the shares issued for the 38 IPOs in 2002, but the bookrunner picked up a net buy imbalance of 8.62 percent on the first day of trading. In contrast, in 2000 the bookrunner buy imbalance of 0.50 percent pales in comparison to the percent buy imbalance from bookrunner clients. We also examine the trading dynamics by various investor types over the first 21 trading days. Figure 1 examines trading by bookrunner clients, bookrunner for its own inventory, and changes in co-manager client and other syndicate member client holdings scaled by the initial number of shares issued. Figure 1 shows that most of the buying by bookrunner clients and the bookrunner takes place on the first day of trading. Bookrunner client net buying is 8.79 percent of shares issued on the first trading day, an additional 1.40 percent on the second day, and is positive but small throughout the rest of the first 19 trading days. Consistent with Ellis, Michaely and O Hara (2000) the bookrunner is also a net buyer for the first 15 trading days, buying the most during the 14

17 first day and week of trading. Small but persistent selling activity occurs through the co-manager throughout the first 21 trading days as these clients likely unload some of their initial allocation. 4. Comparison by Brokerage Houses, Trade Size, Execution Quality, Long-term Ownership, and Price Patterns We use various features in the cross-section of IPOs to shed light on the testable hypotheses outlined in Section 1. We first examine trading patterns for each underwriter when they are the bookrunner versus when acting as another syndicate member to examine predictions most closely related to the clientele, informational advantage, familiarity, and reputation hypotheses. We then examine imbalances by trade size to analyze the laddering and strategic allocation predictions. We also examine execution quality across trade-size groups. Finally, we analyze institutional shareholdings and buying activity by price levels to distinguish predictions of strategic allocation from those of laddering. A. Examination of Clientele, Informational Advantage, Reputation, and Marketing Hypotheses The clientele hypothesis (H2) predicts that clients with an affinity for IPOs migrate to brokerage houses that specialize in issuing IPOs and will be more likely to buy IPO shares through the bookrunner in aftermarket trading than other clients. To explore this hypothesis, we examine the client trading of each market maker when they are the lead underwriter compared to when they join the syndicate in another capacity. A. 1. Clienteles Figure 2 plots bookrunner client net buying on the y-axis and net client buying when the firm is a syndicate member (Panel A) or a co-manager (Panel B) on the x-axis as a fraction of total shares issued. Very few lead underwriters face net selling pressure from their clients in the secondary market (or at least are successful in deterring net selling). Only 10 out of 121 underwriters have net 15

18 client selling when acting as a bookrunner, whereas 71 brokers face net client selling when participating as another syndicate member. Those that do face net client selling pressure as a bookrunner tend to be less active underwriters, handling fewer than five IPOs in our sample. Most of the points are in the northwest quadrant (63 of 121), indicating that most brokerage houses experience positive net client buying when they are the bookrunner but net client selling otherwise. 17 A. 2. Marketing to create familiarity The marketing hypothesis posits that clients of the bookrunner might be more familiar with an IPO than a client of a brokerage house who is only a loosely associated syndicate member. Panel B of Figure 2 displays net buying levels for bookrunner clients and co-manager clients parties that are likely to both be familiar with the IPO (and possibly share similar information). Panel B shows that clients are much more likely to be net buyers when their broker is the lead underwriter (92 of 98 underwriters who serve as a bookrunner and a co-manager have positive bookrunner client net buying, but 61 of these same underwriters have clients who are net sellers when they are the comanager). These results do not support the hypothesis (H3.1) that bookrunner clients and comanager clients trade similarly based on familiarity. Without allocation data, we cannot precisely compare the magnitude of bookrunner and comanager trading relative to the underlying allocations. However, we can extrapolate from Boehmer, Boehmer, and Fishe (2004), who report that the lead underwriter allocates 79.4 percent, and Chen and Ritter (2000), who in their example assume that the co-manager allocates percent of shares not allocated by the lead underwriter. Using these estimates, 1.77 percent of shares issued would be 17 To further test for clientele effects within an industry, we examine Internet IPOs and find similar patterns to those in Panel A. 16

19 bought through the co-manager. 18 Panel B shows that only five co-managers have clients that buy more than this amount despite the fact that they receive small allocations to begin with. To examine the marketing hypothesis related to underwriting activity more thoroughly, Panel A of Table 3 reports the mean and median client net buying for four underwriter groups based on activity. Consistent with both laddering (H1.4) and marketing hypotheses, highly active lead underwriters have higher net client buying. Bookrunners underwriting more than 45 IPOs have clients that buy an additional percent of the total shares issued on the first trading day, whereas clients of brokerage houses that underwrite fewer than five IPOs buy only an additional 3.96 percent. Tests reported in Panel B indicate significant differences across lead underwriter activity categories. Active bookrunners are associated with greater net client buying, consistent with larger brokers generating more demand through marketing or more prevalent laddering agreements where access to future hot IPOs is lucrative. If marketing is a major factor in net client demand, we expect to see clients of large brokerage houses acting as co-managers engaging in more net client buying. However, in contrast to marketing hypothesis H3.2, Table 3 shows little relation between co-manager activity and client buying. Even the most active co-managers face net client selling, suggesting that marketing efforts do not create the large net buying through the active bookrunners. A. 3. Underwriter reputation Net client buying might also be related to reputation effects where bookrunner clients bet on a successful offer due to past underwriter reputation. However, if reputation is visible to all investors, this hypothesis predicts no differences between lead underwriter clients and clients of other brokerage houses, holding reputation constant. Panel C of Figure 2 presents the trading 18 The average bookrunner buying as a percentage of their allocation is percent (=8.79/79.4). If the co-manager clients bought in the same proportion to their estimated allocation, it would amount to 1.77 (=11.07*( )*0.7775) percent of shares issued. 17

20 behavior of the clients both when their broker is the lead and when their broker is a co-manager in an IPO with a high-reputation lead underwriter. 19 Although this combination limits our analysis to 24 underwriters, clients of highly reputable firms are almost always net buyers (23 of 24 have positive bookrunner client net buying) when the broker is the lead, but are generally net sellers when the broker is a co-manager. Overall, the underwriter s role as the bookrunner, rather than reputation, seems to drive the observed patterns in client trading. A. 4 Information hypotheses Table 3 also provides evidence against the informational advantage hypothesis (H4). If bookrunner clients have superior information, one might expect large purchases only in those IPOs where they have material information. However, Table 3 shows that median levels of IPO client buying are similar to mean levels, suggesting no skewness in the distributions. More importantly, clients of the most active underwriters are net buyers in percent of IPOs, providing evidence that bookrunner clients almost always buy. Given historically low long-run IPO returns and median firm s poor performance at different horizons in our sample, 20 it is clearly not the case that these IPOs are bought because they are underpriced in the aftermarket. Building on the intuition of Benveniste and Spindt (1989), information regarding the true value of an issue may be held by investors, but only fully revealed in aftermarket trading perhaps through the lead underwriter. If this is the case, we expect to see initial buying through the bookrunner clustering at the market open when the information will first be revealed. In unreported results, we examine buying on the first trading day at five-minute intervals and find that although volume clusters at the open and close, the proportion of net buying through the bookrunner does 19 High-reputation underwriters are those scoring 8 or 9 for the period according to data on Jay Ritter s website. 20 Abnormal returns are computed by subtracting the return for the corresponding size-b/m portfolio as in Brav & Gompers (1997). The percentage of IPOs that outperform the benchmark is 55.40, 48.54, 39.91, 31.36, 21.91, and at the one-month, three-month, six-month, one-year, two-year, and three-year horizons, respectively. 18

21 not cluster and is consistently positive throughout the day. We re-examine the link to information by looking at short and long-run returns in Section 5. B. Trade size Both laddering and strategic allocation predict that buy imbalances primarily involve institutional clients. Generally, only institutions or extremely wealthy traders can execute block trades of 10,000 shares or more. 21 We break down buy, sell, and buy-sell imbalances for client trades of each dealer type by trade size in Table 4, revealing several interesting facts. First, bookrunner client trading is dominated by large trades. For example, over 20 times more volume is executed through block purchases compared to small purchases through the bookrunner. Co-managers have three times more volume through block purchases than small purchases, but similar amounts of block purchases and block sales. Conversely, most non-syndicate member client trading is in small trades. Second, all of the net buying through the bookrunner is due to trades of 1,000 shares or more. Trades of fewer than 1,000 shares through the bookrunner are slightly more likely to be sales. Third, clients of co-managers and non-syndicate members sell more than they buy in the two largest trade-size groups. Clients of non-syndicate brokerages are net buyers in trades for less than 1,000 shares. Overall, large trades drive the bookrunner client buy imbalance and these patterns are distinctly different from those of other syndicate or non-syndicate brokerages. C. Execution quality The bookrunner may offer differential execution costs that induce net client purchases in the aftermarket. To examine this hypothesis (H6), we benchmark all first-day transactions to the posted (inside) bid-ask quotes and present the percentage of buying activity by clients of the bookrunner that occurs a) below the best posted bid; b) at or below the bid-ask midpoint; and c) at or below the 21 With the average buy trade price of in our sample, a typical block trade requires more than $250,000 of capital. 19

22 best posted ask price. Since it is advantageous for clients to buy at lower prices, higher fractions of these numbers across brokerage houses represent superior execution. Since quotes are only valid up to a posted depth, execution costs may vary substantially across trade size and we present results by trade-size categories. Panel A of Figure 3 compares the execution of client buy trades through bookrunners to those of non-syndicate market makers at various prices relative to the offer price. While the fraction of clients who buy below the bid is small near the offer price, the fraction increases at prices above the offer price (likely because of rapidly moving prices) and increases with trade size (likely reflecting trades exceeding the quoted size). 22 Although we present all trade size categories, we focus on the larger trades that are primarily responsible for the persistent bookrunner client buy imbalances. In the less than 1,000-share trade-size category results are mixed. However, in the trade-size group of 1,000 to 5,000 shares, the proportion of trades at or below midpoint is approximately the same between clients of the bookrunner and non-syndicate members, but non-syndicate member clients are able to buy at or below the ask a greater fraction of the time. In the 5,000 to 10,000 and greater than 10,000 share trade-size groups, clients of non-syndicate members buy at prices at or below the ask and at or below the midpoint a greater fraction of the time. 23 Since all of the buy imbalances through the bookrunner are due to the large trades and in these large trade-size groups the bookrunner often offers worse buy executions, these results lend no support to the hypothesis that the excessive buying activity through the bookrunner is due to cheaper prices. The net buy imbalance could also be driven by a lack of selling through the bookrunner if the bookrunner offers poor selling executions to discourage flipping. Panel B of Figure 3 reports the 22 Timing mismatches between trades and quotes may also result in trades outside the quotes. Although market makers are only required to report trades within 90 seconds, we lag each quote two seconds before mapping to trades based on conversations with Nasdaq officials. Further, Nasdaq commissions are embedded in transaction prices during our sample period, which may generate trade reports that appear outside of quoted spreads. 23 These results contrast with Ellis (2003) who finds that bookrunners generally execute client block buy trades at better prices than other market makers over an earlier October 1996 to June 1997 Nasdaq period. One difference in our analyses is that we examine executed prices relative to the offer price rather than unconditionally. 20

23 fraction of trades where sellers receive prices a) above the best ask price; b) at or above the midpoint; and c) at or above the bid. For the smallest three trade-size groups, Panel B shows that investors generally receive better selling prices when executing trades through a non-syndicate market maker than through the bookrunner. However, for trades at or above 10,000 shares (those that generate the large buy imbalances) there is little difference between executions through the bookrunner and non-syndicate market makers. In sum, for block trades that produce the largest buy imbalances, non-syndicate dealers offer slightly better buy prices and similar sell prices. Overall, there is little evidence that underwriter client buy imbalances can be explained by superior execution quality. D. Long-term share holding patterns Strong net client buying through large trades may stem from the bookrunner reallocating secondary shares to long-term shareholders. Since our trading data does not include the identity of the investors, we utilize quarterly 13F Spectrum data where all institutional investors are required to report. 24 We identify the institutional investors who hold shares at the time of the first 13F quarterly reporting date following the IPO and track their holdings through the following four quarters. 25 If bookrunner client buying is due to long-term shareholders aggregating secondary market shares, then we expect to see those brokerage houses with large initial buy imbalances to have more original shareholders who continue to hold shares across quarters (H7). Of course, we cannot capture ownership changes between the first day of aftermarket trading and the first reporting date. Consequently, if transitory investors purchase shares on the first day but sell before the first reporting date, our estimates will understate the amount sold and bias us in favor of H7. 24 The data has many known reporting issues that we thoroughly account for before applying to our sample. 25 Aggarwal, Prabhala, and Puri (2002) and Ljungqvist and Wilhelm (2002) show that 72.8 and 80.4 percent of IPO shares are allocated to institutional investors in their respective U.S. and international samples. 21

24 We regress the change in holdings of the institutional shareholders at the time of the first reporting date on the magnitude of first-day client buying while controlling for the first-day excess IPO return, the log of one plus the number of days from the IPO date to the first Spectrum reporting date, lockup dummies, and fixed effects for each quarter and report the results in Table 5. Panel A contains raw changes in initial institutional holdings as the dependent variable, while Panel B uses abnormal changes in initial holdings for each firm by computing the difference between the change in institutional holdings for the IPO and the change for matched firms from the same size and book-to-market portfolio. 26 Both panels show that at horizons up to the following four reporting quarters, the change in ownership for the shareholders as of the first reporting date is significantly negatively related to bookrunner client net buying. These results are inconsistent with the hypothesis that strategic allocation leads to stable long-term shareholders, but consistent with the laddering alternative where aftermarket client buying likely represents transient investors. E. Buying across price levels The laddering hypothesis predicts that clients unconditionally buy. Client buying may be particularly beneficial to the underwriter (and costly to the client) at prices at or near the offer price and in cold IPOs. Alternatively, if aftermarket buying is driven by familiarity or marketing through road show activities then one might expect this over-optimistic behavior to be greatest in the hottest IPOs--those that receive the most attention and enjoy high returns on the first trading day. To examine if bookrunner clients buy even at low prices we examine a) net client buying for the first 21 days for IPOs with varying degrees of initial price movement; b) the breakdown of the first-day activity by trade-size and IPO hotness; and c) the first-day intradaily dynamics relative to the offering price for these categories. 26 We follow Brav and Gompers (1997) and require book-to-market equity (BE/ME) to be available within one-year of the offering date. This reduces the size of the sample in the benchmark analysis from 1,118 with 1 st quarter Spectrum holdings to 958 IPOs with size and BE/ME. 22