Tie-In Agreements and First-Day Trading in Initial Public Offerings Hsuan-Chi Chen 1 Robin K. Chou 2 Grace C.H. Kuan 3 Abstract When stock returns in certain industrial sectors are rising, shares of initial public offerings (IPOs) in the same sectors become attractive to investors. And sentiment demand for these IPO shares increases. Sentiment demand prompts underwriters to seek IPOs in the sectors favored by investors. On the other hand, the loose listing standards in NASDAQ allow access for unprofitable firms to go public during the Internet bubble period. We estimate a probit model with the returns of matched industrial sectors and pretax incomes before the IPO offer dates. The empirical results indicate that unprofitable firms or firms in industrial sectors with higher returns are more likely to become laddering lawsuit targets. Moreover, we find that the estimated laddering likelihood has a positive relation with IPO underpricing, the number of trades, and order imbalance on the first trading day. The laddering likelihood offers another explanation for the phenomenal IPO underpricing during 1999-2000. However, the laddering likelihood is negatively related to the average volume per trade, indicating either higher retail investor sentiment or the possibility of stealth trading by institutional investors for laddering IPOs. The findings suggest that market sentiment, loose listing standards, and tie-in agreements have a significant impact on the aftermarket trading of IPOs during the internet bubble period. Keywords: Investor sentiment; IPO underpricing; Trading manipulation 1 Department of Finance, Yuan-Ze University, 135 Yuan-Tung Road, Chung-Li, Taiwan, Email: chenh@saturn.yzu.edu.tw 2 Department of Finance, National Central University, 300 Jungda Road, Chung-Li, Taiwan, Email: rchou@cc.ncu.edu.tw; 3 Department of Finance, Yuan-Ze University, 135 Yuan-Tung Road, Chung-Li, Taiwan, Email: gchkuan@gmail.com;
Tie-In Agreements and First-Day Trading in Initial Public Offerings Abstract When stock returns in certain industrial sectors are rising, shares of initial public offerings (IPOs) in the same sectors become attractive to investors. And sentiment demand for these IPO shares increases. Sentiment demand prompts underwriters to seek IPOs in the sectors favored by investors. On the other hand, the loose listing standards in NASDAQ allow access for unprofitable firms to go public during the Internet bubble period. We estimate a probit model with the returns of matched industrial sectors and pretax incomes before the IPO offer dates. The empirical results indicate that unprofitable firms or firms in industrial sectors with higher returns are more likely to become laddering lawsuit targets. Moreover, we find that the estimated laddering likelihood has a positive relation with IPO underpricing, the number of trades, and order imbalance on the first trading day. The laddering likelihood offers another explanation for the phenomenal IPO underpricing during 1999-2000. However, the laddering likelihood is negatively related to the average volume per trade, indicating either higher retail investor sentiment or the possibility of stealth trading by institutional investors for laddering IPOs. The findings suggest that market sentiment, loose listing standards, and tie-in agreements have a significant impact on the aftermarket trading of IPOs during the internet bubble period. Keywords: Investor sentiment; IPO underpricing; Trading manipulation 1
I. Introduction The first-day returns of initial public offerings (IPOs) have received much attention in the literature. Ritter and Welch (2002) summarize IPO activity for the period 1980-2001. They show that, during the 1980s, the first-day return was about 7%, doubling to 15% during the period 1990-1998. From 1999 to 2000, the NASDAQ composite index experienced record highs and Internet stock prices soared. During this time, the IPO first-day return jumped to more than 60%. However, after the stock market bubble collapsed, the first-day return in 2001 had returned to the level founded in the early 1990s. Loughran, Ritter and Rydqvist (1994) find that private firms generally prefer to go public when the market is optimistic. However, the explanations of a drastic increase for the first-day returns of IPOs in the bubble period may go beyond that explained by market timing. Ljungqvist and Wilhelm (2003) argue that the Internet bubble period witnessed marked changes in pre-ipo ownership, such as decreased CEO ownership and increasingly fragmented ownership, and a sharp decrease in the secondary sales of existing shares by pre-ipo shareholders at the time of the initial offer. Changes in pre-ipo ownership structure and insider selling behavior during this period would have reduced key decision makers incentives to monitor and bargain over the IPO offer price, and thus lead to greater underpricing. Alternatively, Loughran and Ritter (2004) attribute the higher underpricing during the bubble period to changes in the isuer sobjective function. They provide two reasons for why issuers were more tolerant of higher underpricing in the 1990s and during the Internet bubble period. First, analyst coverage becomes more important in choosing a lead underwriter and issuers pay the indirect costs of higher underpricing (see Cliff and Denis, 2004). Second, side payments through hot IPO share allocation to the personal brokerage accounts of CEOs and venture capitalists also plays an important role. Spinning practice creates an incentive for the decision makers to choose a lead underwriter who leaves money on the table in IPOs, in return for side payments to their own accounts. In December 2000, the Wall Street Journal reported the abuse of IPO share allocation by leading underwriters (see Pulliam and Smith, 2000a, 2000b). Underwriters may reward institutional investors who agree to bring in more business more initial IPO share allocation. To secure IPO share allocation, institutional clients may enter tie-in agreements with underwriters to purchase additional IPO shares in the aftermarket at prices that are likely higher than the IPO offer prices. This practice 2
is known as laddering. Hao (2006) shows that laddering has a larger effect on the market prices of IPOs with greater expected underpricing and greater expected momentum in the aftermarket. Furthermore, in her model, laddering simultaneously increases the IPO price and the aftermarket price, and thus does not necessarily increase the percentage of underpricing. As pointed out in Choi and Pritchard (2004), the laddering scheme benefits underwriters and their institutional clients. While individual investors lose out by buying IPO shares in the aftermarket with manipulated prices. For underwriters, tie-in agreements greatly reduce the risk borne by underwriters in dealing with firm commitment offerings. It also helps underwriters to gain reputation for successful IPO deals. This reputation may attract a larger number of financially vulnerable private firms. Moreover, institutional clients who participate in tie-in agreements could still easily earn greater profits with inflated market prices for their initially allocated IPO shares while kicking back part of the profits to the underwriters. The litigation settlement between the Securities Exchange Commission (SEC) and J. P. Morgan, Morgan Stanley, and Goldman Sachs reveals that, during Internet bubble period, underwriters aggressively approach their clients for tie-in agreements to help manipulate IPO trading. To receive hot IPO allocation, institutional clients promise higher aftermarket purchases for IPO shares and/or kick back some profits to the underwriters by paying unusually high commissions through stock trading (see Nimalendran, Ritter, and Zhang, 2006). In addition to Ljungqvist and Wilhelm (2003) and Loughran and Ritter (2004), Aggarwal, Purnanandam, and Wu (2006) argue that laddering manipulation can explain the extremely high level of IPO underpricing in that underwriters create artificial excess demand by requiring their customers to buy the IPO stocks in the aftermarket in exchange for the IPO share allocation. Similarly, our study provides an explanation for the underpricing witnessed in the Internet bubble period through the estimation of laddering likelihood. However, our econometric specification differs from that of Aggarwal, Purnanandam, and Wu (2006). We argue that when investors become more optimistic about specific industries, they are willing to place significantly more money in the IPO market, particularly in some industrial sectors. Underwriters may then take advantage of investors industrial sentiment and the discriminatory IPO share allocation allowed in the bookbuilding system to solicit additional buying orders in the aftermarket. This greater buying power would, in turn, 3
result in higher aftermarket prices and greater IPO initial returns. On the other hand, the loose requirements of NASDAQ listing standards open the door for unprofitable firms in the hot industrial sector to enter the IPO market. When engaging in firm commitment offers for financially weak IPOs, underwriters have an incentive to exert their discretionary power to boost up aftermarket prices of these IPOs. Therefore, we hypothesize that the likelihood for a firm to be a laddering target is related to investors sentiment, the NASDAQ looser listing standards, or both. Laddering likelihood is one of important factors to help explain the spectacular first-day returns and unusual trading patterns for IPOs during the Internet bubble period. Examining the first-day IPOs trading patterns enables us to investigate directly whether investor sentiment is an important factor in generating extremely high first-day returns, and the buying and selling strategies employed by institutional investors. Thus, unlike both Aggarwal, Purnanandam, and Wu (2006) and Hao (2006), our study makes contributions in analyzing the effect of laddering on the first-day trading of IPOs. Links among investor sentiment, IPO first-day returns, and long run performance have been recently studied by Dorn (2003), Ljungqvist, Nanda and Singh (2003), Derrien (2005), Chan and Meidan (2005), and Cornelli, Goldreich and Ljungqvist (2006). These studies argue that investor sentiment is a key factor in explaining greater underpricing and poor long run performance in a hot IPO market. The theoretical frameworks in these papers place emphasis on the underwriters decisions on offer prices and their impact on the long-run performance of IPOs. The empirical evidence in these studies collectively implies that the sentiment demand is able to explain the phenomenon observed during the Internet bubble period. Our study focuses on the link between investor sentiment and IPO trading manipulation, which is another controversial issue in IPO underwriting during the bubble period. When investor sentiment is high, we argue that underwriters play an aggressive role in both IPO pricing and trading, and they seek kick-backs from their institutional clients through tie-in agreements. In addition to investment sentiment, the loose NASDAQ listing standards based on market capitalization open the door for many unprofitable firms to go public. Both factors greatly contribute to laddering manipulation, which in turn help explain first-day returns and trading patterns observed in the IPO aftermarket during the bubble period. To assess the laddering likelihood, we estimate a probit model using a sample of 4
IPOs allegedly involved laddering. We use the average return of matched industrial sector before the IPO offer date as a proxy for market sentiment and pretax income as a measure of profitability. We find that the laddering likelihood has a positive relation to the proxy of market sentiment and a negative relation to the pretax income. The estimated laddering likelihood is a significant factor in explaining the large IPO first-day returns during the Internet bubble period. In addition, the laddering likelihood is positively associated with the number of trades and order imbalance on the first trading day. IPOs that are more likely to be laddering targets generate more trading interest and the net buying is also higher. However, the laddering likelihood has a negative relation to the average volume per trade, indicating that more retail traders are trading alleged IPOs, institutional investors are breaking trades for alleged IPOs, or both. Furthermore, we group trades based on their trade size and analyze the relation between the laddering likelihood and microstructure variables. For small and medium size trades, the laddering likelihood has a positive relation to the total number of trades, order imbalance, and the number of buyer-initiated trades. However, the laddering likelihood has a negative relation to the average volume per trade. These findings indicate that retail investors exhibit keen interests in buying IPO shares with a higher laddering likelihood. They are generally unaware of the poor profitability of these IPO firms and the tie-in agreements between underwriters and their clients. For large-size trades, the laddering likelihood also has a positive relation to the number of trades and the number of buyer-initiated trades. The negative relation to the average volume per trade also holds for large-size trades. However, the positive relation to order imbalance is less significant. It appears that institutional investors also actively participate in the trading of IPOs with a high laddering likelihood. However, they tend to trade these IPOs with a high laddering likelihood in smaller trade sizes. Since the relation to the number of buyer-initiated trades is significant, but the relation to order imbalance is not, it is possible that institutional investors buy the IPOs with a high laddering likelihood because of tie-in agreements with underwriters. The purpose of buying trades is to maintain the aftermarket prices at the opening levels. Institutional investors are also more likely to sell IPO shares with a high laddering likelihood to retail investors with frequent and smaller trades. The remainder of the paper is structured as follows. In section 2, we examine the difference in returns and trading characteristics between the IPOs alleged of 5
manipulation and those not. Section 3 discusses the possible causes of laddering manipulation, including investor sentiment and loose NASDAQ listing standards. In section 4, we estimate the laddering likelihood of an IPO firm based on its pre-listing profitability and investor sentiment. Given the estimated laddering likelihood, we investigate the impact of manipulation on first-day returns and the trading patterns during the first trading day. Section 5 concludes the study. II. IPO Returns and Trading Patterns of the First Trading Day A. Sample Description To investigate the impact of laddering on the underpricing and trading patterns for IPOs during the Internet bubble period, we ideally need to possess exact information on the tie-in agreements for IPOs involving in laddering manipulation. However, it is difficult for outside parties to access this information because it is only available from the managing underwriters. Due to the lack of information, we use IPOs that are subject to class-action lawsuits filed at the U.S. District Court (Southern District of New York) as substitutes. Our IPO data for the period of 1998-2000 are constructed from the US Domestic New Issues Database of Thomson Financial Securities Data. REITs, ADRs, close-end funds, and unit offerings are excluded from the sample. The IPO firms in the sample also need be covered by CRSP and COMPUSTAT. As the analysis requires the pre-listing accounting data and the microstructure data on the first trading day, the IPO sample firms also have to be covered by the SEC s EDGAR online database for S-1 filing and the Trade and Quote (TAQ) database compiled by the New York Stock Exchange (NYSE). Using the above criteria, we obtain a sample of 592 IPOs during the period 1998-2000. We use class-action lawsuits for IPO tie-in agreements filed at the Southern District of New York to identify the sample of IPOs that are likely to have involved manipulation. These class-action lawsuits allege that the underwriters violated Federal securities laws by manipulating IPO share prices, and the issuers violated the laws by failing to disclose the conduct to the public. These allegations are supported by the preliminary court findings. We find that 161 IPOs in our sample involve in the class-action lawsuits in the Southern District of New York. In this paper we refer these IPOs as alleged IPOs, with the remaining 431 firms termed as non-alleged IPOs. Table 1 presents the descriptive statistics for both the alleged IPOs and the 6
non-alleged IPOs. The Table shows that the alleged IPOs raise average proceeds of $95.5 million, while non-alleged IPOs raise an average of $68.5 million. The partial adjustment, which measures the adjustment of the offer price from the midpoint of the preliminary filing range, shows that the offer prices are significantly revised upwards for alleged IPOs. Table 1 also reports some accounting variables, such as the net tangible assets and pretax income of issuing firms at the time of listing. Net tangible assets are defined as total pro forma shareholders equity minus goodwill. The total pro forma shareholders equity is collected from the S-1 filing in the SEC s EDGAR database. The data on pretax income, goodwill, and net sales are collected from COMPUSTAT. The average pretax income for alleged IPOs is -$41 million and -$19 million for non-alleged IPOs. Alleged IPO firms appear to operate with more severe losses at the time of going public. However, the alleged IPOs have greater net tangible assets than the non-alleged IPOs. This suggests that most of the alleged IPOs listed based on the market capitalization standard. Matching the industries with the three-digit SIC codes, we also compute the average daily return of firms in the same industry as the sample IPOs for the 20 trading days before the offer date. The matching industrial daily return for alleged IPOs is twice as high as that for the non-alleged IPOs. We also calculate the first-day offer-to-close returns, offer-to-open returns, and open-to-close returns for the two sub-samples. For both alleged IPOs and non-alleged IPOs, most of offer-to-close returns come from price jumps from offer prices to opening prices. For instance, the average offer-to-open return for alleged IPOs is 136%, while the average open-to-close return is only 8%. Put differently, if investors purchase IPO shares in the aftermarket on the first trading day, there is little return to earn at the end of the day, compared with the offer-to-open return. Meanwhile, the magnitude of these returns implies that either institutional or regular investors who are allocated IPO shares initially at the offer prices can earn remarkably high returns. When comparing the two subsamples, the alleged IPOs earn on average higher offer-to-open returns and open-to-close returns than the non-alleged IPOs. B. The microstructure variables In this section we first investigate whether the alleged IPOs exhibit trading patterns that differ from the non-alleged IPOs. To receive IPO allocations, some underwriters s institutional clients would purchase alleged IPO shares and boost the aftermarket prices. Individual investors are attracted by the increasingly inflated 7
prices and actively purchase these shares due to higher sentiment. We examine the impact of laddering manipulation on six microstructure variables during the first trading day. These variables include the total number of trades, the average number of shares per trade, the average ask size, the average bid size, the order imbalance, and the average relative effective spread during the first trading day. The average number of shares per trade is calculated as the total trading volume divided by the total number of trades during the first trading day. The average ask (bid) size is the quoted size at the best ask (bid) price. The ask (bid) size is measured in hundreds of shares. We identify trades to be seller-initiated or buyer-initiated using the algorithm in Lee and Ready (1991) and calculate the aggregate order imbalance using a measure similar to that in Busse and Green (2002). The order imbalance is calculated as: IMBAL nbuys nsells it it it, nbuysit nsells it where nbuys it (nsells it ) is the number of buyer-initiated (seller-initiated) trades during time interval t for stock i. If this measure of order imbalance is close to one, it implies that most of the trades are buyer-initiated. Similarly, if the measure is close to -1, it implies that most of the trades are seller-initiated. Finally, the relative effective spread is calculated as: ES it Pit M it 2, M it where P it is the transaction price for stock i at time t and M it is the mid-point of the bid and ask quotes immediately before the transaction. Following Lee and Ready (1991), we require the quote be at least five seconds before the trade. Table 2 reports the summary statistics of the six microstructure variables for the entire sample, the alleged IPOs, and the non-alleged IPOs. Note that there is a significant difference in the total number of trades and the average volume per trade between the two sub-samples. The average total number of trades for the alleged IPOs is 20,644, while the average is only 8,252 for the non-alleged IPOs. Interestingly, the mean of the average number per trade for the alleged IPOs is less than half that of the non-alleged IPOs. It appears that alleged IPOs are traded with clusters of small trades. The mean values of the average ask size and bid size for the alleged IPOs are also smaller than the non-alleged IPOs. This shows that the aftermarket market depth 8
of the alleged IPOs is smaller partly because market makers avoid taking large positions on these IPOs. The average order imbalance shows that the alleged IPOs have more buyer-initiated trades and non-alleged IPOs have more seller-initiated trades. The relative effective spreads for the two sub-samples are at a similar level of 0.01. Preliminary investigation suggests that the first-day returns and the trading patterns of the alleged IPOs are strikingly different from those of the non-alleged IPOs. In the next section, we will discuss possible reasons that give rise to laddering manipulation in the IPO market. We argue that laddering manipulation is related to investor sentiment in some particular industrial sectors and the poor profitability of firms. In turn, laddering manipulation is one of the important factors used to explain the spectacular first-day returns and trading patterns observed for some IPOs during the Internet bubble period. III. Investor Sentiment, New Listing Standards, and IPO Manipulation A. Investor sentiment and IPO pricing Ljungqvist, Nanda, and Singh (2006) construct a model to explain the behavior of issuers in a hot IPO market with short sale constraints. They argue that institutional investors who receive initial IPO allocation hold IPO shares in inventory to maintain the price level and extract surplus from sentiment investors. Underpricing is the compensation for institutional investors for holding inventory as market sentiment may cease unexpectedly. Investor sentiment moves the first-day price and the offer price away from their fundamental values, which causes the underpricing and long run underperformance of IPOs. Derrien (2005) also develops an IPO pricing model based on investor sentiment. In his model, an underwriter determines the offer price based on the intrinsic value of an IPO firm and noise trader sentiment. The higher the sentiment is, the higher the offer price is. However, the offer price is only partially adjusted for sentiment as an underwriter is concerned about possible price support in the aftermarket. The underwriter is worried about that investors may turn pessimistic and therefore the underwriter chooses not to fully adjust the offer price. Derrien (2005) examines IPOs listed in the Paris stock exchange and finds support for his model. Chan and Meidan (2005) also study the relation between retail investor sentiment and IPO long run performance. They use order imbalance for small trades during the first trading day as the proxy for retail investor sentiment and examine the long run 9
performance of IPOs during the period 1994-2002. A negative relation between the sentiment proxy and the IPO long run performance is found during the Internet bubble period. The long run performance is unrelated to order imbalance of medium-sized and large-sized trades. In their paper, small trades have a trade size of less than 500 shares, medium trades have a trade size between 500 and 10,000 shares and large trades have a trade size of more than 10,000 shares. Although trades larger than 10,000 shares are normally initiated by institutional investors, they would be classified as block trades. It is unlikely that institutional investors who are involved in laddering would trade such large sizes so as to avoid attracting the attention of regulators. Therefore, with the trade size categories used in Chan and Meidan (2005), the impact of laddering manipulation on the trading volume, number of trades, and order imbalance cannot be easily identified. Dorn (2003) examines the trading of German IPOs in the pre- and post-ipo market. In the pre-ipo market, retail investors speculate on the share prices of IPOs that are shortly going public. He finds that retail investors tend to pay for new issues at prices higher than the fundamental values following the run-up of recent new offerings. The IPOs that are in great demand from retail investors exhibit high first-day returns and underperform in the long run. The empirical results show that investor sentiment can explain IPO first-day returns and long-run performance. Similar findings for the German IPO market are also documented in Oehler, Rummer and Smith (2005). Cornelli, Goldreich, and Ljungqvist (2006) examine the influence of overoptimism among retail investors on the first-day returns and long-run performance of IPOs. They use pre-ipo market prices as a proxy for retail investor sentiment and find an asymmetric relation between pre-ipo market prices and aftermarket prices. When pre-ipo market prices are high, aftermarket prices are positively correlated with pre-ipo market prices. This implies that retail investor sentiment is highly correlated with aftermarket prices. However, when pre-ipo market prices are low, the correlation is lower. To explain the asymmetric relation, Cornelli, Goldreich, and Ljungqvist (2006) argue that when retail investors are overoptimistic, institutional investors sell their inventory and push aftermarket prices to the level of retail investors reservation prices. When retail investors are pessimistic, institutional investors hold their allocated shares and aftermarket prices stay at the fundamental values. They further argue that the setting of offer prices depends on the bargaining power between issuing firms and underwriters. 10
Cornelli, Goldreich and Ljungqvist (2006) study IPOs in Europe and use pre-ipo market prices as a proxy for retail investor sentiment, unfortunately pre-ipo markets do not exist in the US to help reveal retail investor sentiment. In our paper, we use matched industrial returns prior to the offer dates as a proxy for investor sentiment. As retail investors are generally less sophisticated, industrial returns serve as a good indicator for investment sentiment. Meanwhile, in the IPO market, the valuation of IPO firms may refer to the valuation multiples of comparable firms that are likely to be in the same industry (see Kim and Ritter, 1999). Also, the loose NASDAQ listing standards based on market capitalization allows unprofitable firms to go public during the Internet bubble period. The vulnerable operating condition of these firms greatly reduces their bargaining power with underwriters who are eager to take advantage of the market boom. In the following section, we discuss the possible connections between investor sentiment, loose listing standards, and IPO trading manipulation. B. Investor sentiments, listing standards, and laddering manipulation Klein and Mohanram (2005) study the relationship between NASDAQ s National Market listing standards and the poor long-run performance of IPOs in the late 1990s. They find that the loose listing standards are partly responsible for the entry of many unprofitable firms. From August 1997 to June 2001, NASDAQ used three alternative sets of initial listing standards. The first alternative is based on the pre-listing profitability of firms. This requires a firm to have a minimum pretax income of $1 million for the year before listing and $6 million of net tangible assets at the time of listing. The second and the third alternatives are based on the market capitalization of firms before listing. The second alternative requires a firm to have a minimum of $18 million of net tangible asset and a two-year operating history. The third alternative requires a firm to have a market capitalization of $75 million or total assets and total revenues of $75 million. Of three alternatives, only the first requires firms to have positive prelisting earnings. Klein and Mohanram (2005) find that the majority of firms listing during the late 1990s are under the listing standards based on market capitalization. They find that these firms earn much higher first-day returns and continue to outperform those listed under the listing standard based on profitability in the first six months. However, these firms exhibit much worse performance in their stock returns and financial conditions in the long term. Many of them are delisted within three years. When firms are 11
classified into industries based on the three-digit SIC codes, 38% of new listings belong to the computer and data processing industry. This represents the highest proportion among all industries. About 86% of new listings in this industry went public under the listing standards based on market capitalization. Furthermore, 96% of Internet new listings enter the market with market capitalization standards. This shows that market capitalization based listing standards open a door for unprofitable firms in the technology industry to list on NASDAQ. Aggarwal, Purnanandam and Wu (2006) present a model that explains the return patterns based on laddering agreements between underwriters and their clients. The model assumes that, in the IPO market, there are ladders, momentum investors, and noise traders. Ladders are underwriters clients who agree to boost up the immediate aftermarket price in exchange for IPO share allocation. Momentum investors are uninformed. They tend to trade by observing prices and trading volume. The noise traders provide liquidity to the market. The model provides a link between the behavior of momentum investors and the likelihood of laddering. The difference between laddered IPOs and non-laddered IPOs is positively correlated with the number of momentum investors. If one treats momentum investors as sentiment investors, the likelihood of laddering increases when market sentiment is running high. The model also predicts that the pooling profile of the three types of investors is the key factor in determining the profitability of laddering. Aggarwal, Purnanandam and Wu (2006) find that, during the Internet bubble period, IPOs involved in laddering lawsuits exhibit the first-day returns seven times higher than those that were not. The laddered IPOs continue to perform better than non-laddered IPO in the first six months; however, the laddered IPOs experience major price reversal after six months. In the long run, laddered IPOs underperform. The patterns documented in Aggarwal, Purnanandam and Wu (2006) are similar to those found in Klein and Mohanram (2005). Given these similar return patterns, we raise questions on the connection between loose listing standards, investor sentiment, and IPO manipulation. To explain substantial IPO underpricing during the Internet bubble period, we argue that investor sentiment drives up the demand for IPO shares in certain industrial sectors. Market capitalization based listing standards provide a supply of unprofitable firms for IPO deals. The strong demand and supply conditions strengthen the discretionary power of underwriters in the IPO share allocation. Laddering is one of approaches used by 12
underwriters to make profits from deals of poorly performing firms in industrial sectors favored by sentiment investors. When stock returns in certain industrial sectors are increasing, IPO shares in these sectors tend to be more attractive to sentiment investors. As a result, underwriters seek firms in these sectors as targets for IPO deals. Alternatively, private firms in hot industrial sectors make use of the window of opportunity to go public. This could explain the industry concentration of IPOs during the bubble period. With strong sentiment demand by retail investors, underwriters have more control in setting IPO offer prices and aftermarket trading. On the other hand, the loose new listing standards of NASDAQ give underwriters access to many unprofitable firms in particular industries. In order to be qualified for market capitalization listing standards, poorly performing firms have little bargaining power in negotiating offer prices with underwriters even if they observe strong market sentiment. Besides, financially weak IPOs are more likely to become laddering targets. The aftermarket purchase through tie-in agreements greatly reduces the risk of underwriters for price stabilization. We therefore investigate two hypotheses on the relation of laddering likelihood to the firm s financial condition and to investor sentiment. The first hypothesis is that the more financially vulnerable the firm, the more likely the firm is involved in laddering. With market capitalization listing standards, firms with large market cap can go public, even if they are operating at a loss. Pre-listing profitability conditions of IPO firms are measured by pretax income and market capital conditions are measured by net tangible assets obtained from the SEC s EDGAR database and COMPUSTAT. The first hypothesis can be stated in the following form: Hypothesis 1: The likelihood of an IPO firm to be involved in laddering is negatively correlated with pretax income. The second hypothesis is that the higher the investor sentiment, the more likely the firm is involved in laddering. We use the industrial returns over the 20 trading days before the offer date of an IPO as a proxy for investor sentiment. The high stock returns in a particular industry may prompt buying from retail investors and demands for IPO shares in that particular industry. The second hypothesis is as follows. Hypothesis 2: The likelihood of an IPO firm to be involved in laddering is positively correlated with its industrial sector return. We now test whether the likelihood of laddering manipulation has an impact on IPO underpricing. We hypothesize that the keen interest of investors in certain sectors 13
and the loose listing standards of NASDAQ prompt underwriters to engage in laddering. The tie-in agreement to manipulate IPO shares causes the first-day returns of some IPOs to skyrocket. Our third hypothesis is: Hypothesis 3: The first day return of an IPO firm is positively correlated with the likelihood of laddering for the firm. We further investigate the impact of laddering likelihood on the trading patterns observed on the first trading day of the IPO during the bubble period. Retail traders may trade laddered IPOs more intensively because of higher sentiment, and institutional traders may trade smaller trade sizes, so as to avoid revealing information about the laddering agreement. We thus hypothesize that the large number of trades and the small trade sizes during the first trading day are related to laddering manipulation. Hypothesis 4: There is a positive relation between the likelihood of laddering for an IPO and its total number of trades on the first trading day. Hypothesis 5: There is a negative relation between the likelihood of laddering for an IPO and its trade size on the first trading day. As the underwriters clients promise to purchase IPO shares in exchange for initial IPO share allocation, we expect that the number of buyer-initiated trades will be higher than that of seller-initiated trades for laddered IPOs. Using the measure of order imbalance, this implies that IPO firms with a higher likelihood of laddering would involve larger positive order imbalance. The corresponding hypothesis is as follows: Hypothesis 6: IPOs with a higher likelihood of laddering will display a positive order imbalance. Ellis, Michaely and O Hara (2000) show that lead underwriters are always the dominant market maker for the IPO aftermarket trading. If market makers know about the laddering manipulation especially when they are lead underwriters or co-managers, then the market depth of laddered IPOs would be smaller as they try to avoid the risk of holding large inventory positions in these price-inflated IPOs by quoting a thinner depth and/or increasing the quoted bid-ask spreads. A market with a lower depth and a higher spread is characterized as being less liquid. We form two hypotheses on quote spread and market depth. Hypothesis 7: There is a positive relation between the likelihood of laddering and the average relative effective spread in the first trading day. 14
Hypothesis 8: There is a negative relation between the likelihood of laddering and its market depth on the first trading day. In the next section, we test Hypothesis 1 and Hypothesis 2 with a probit model using the sample of class action lawsuits involving laddering. Given the estimation of laddering likelihood estimated from the probit model, we test Hypothesis 3 to Hypothesis 8 to examine how laddering schemes affect first day returns and trading patterns. IV. Regression Results A. The likelihood of IPO manipulation Since the exact identification of laddering manipulation on IPO share allocation and aftermarket manipulation is not available, the IPOs in the class-action lawsuits in Southern District Court of New York are only a proxy for laddering practice. It is possible that an IPO that is not in the class-action lawsuit list has been a laddering target. Therefore, we test the relation among the likelihood of laddering, the financial conditions of IPO firms, and market sentiment with a probit model given the proxy sample of class-action lawsuits. In the probit model, we use pretax income to measure the pre-listing operating performance and net tangible assets to measure the market capitalization of IPO firms. Pretax income is the key accounting variable used to determine whether a firm could obtain listing based on the stricter profitability standard. If firms fail the profitability requirement, they need to have enough net tangible assets to pass the market capitalization requirement. For the investment sentiment, we use the matched industrial returns over the 20 trading days before the offer date of an IPO to measure investor sentiment in the same sector as the IPO. We hypothesize that the likelihood of laddering is higher for firms with poorer pretax income and industrial characteristics preferred by investors. Table 3 presents the estimation results of the probit model. For the first model, the dependent variable is regressed against net tangible assets, pretax income and matched industrial returns. Both pretax income and matched industrial returns are highly significant. The second model removes the matched industrial return. The removal of this variable lowers the likelihood ratio (LR) statistics for testing the goodness-of-fit. The third model removes the insignificant variable, net tangible assets, while maintaining the same level of LR statistic. 15
The results show that both pretax income and matched industrial returns play an important role in estimating the likelihood of laddering manipulation. The significance of the pretax income coefficient partially supports Hypothesis 1. Firms with poor pretax incomes are more likely to be the targets of laddering manipulation. The significance of the positive coefficient of matched industry returns also supports our Hypothesis 2. This shows that firms in an industrial sector that is currently performing well in the market are more likely to become the targets of laddering manipulation. B. Laddering and first-day returns We estimate the laddering likelihood from the pretax income and matched industrial return using probit model 3 in Table 3. The estimated laddering likelihood is used to gauge the impact of laddering practice on IPO underpricing. We regress first-day returns on a set of IPO variables in addition to the laddering likelihood. These variables include the logarithm of offer size, the partial adjustment dummy variable, gross spreads, the number of IPOs in the month prior to the offer day, the Carter and Manaster (1990) rank of the lead underwriter, and a venture-backed dummy. 4 We set the partial adjustment dummy to one for IPOs whose offer price is revised upward, and zero otherwise. The venture-back dummy is set to one for IPOs backed by venture capital, and zero otherwise. The regression results are presented in Table 4. Columns 2 to 4 in Table 4 show the results for the offer-to-close, offer-to-open, and open-to-close returns without the laddering likelihood variable. Columns 5 to 7 show the results after adding the estimated laddering likelihood to the regression. For the offer-to-close and offer-to-open returns, offer size, the partial adjustment dummy, and the gross spread are important factors in explaining the variation of the returns. The laddering likelihood is highly significant without reducing the significance of these IPO variables. The positive coefficient for the laddering likelihood indicates that, the higher the laddering likelihood, the higher the offer-to-close returns and offer-to-open returns. These results support Hypothesis 3 in that there is a positive relation between laddering practice and IPO first-day returns during the Internet bubble period. For 4 We use the underwriter reputation rank updated in Loughran and Ritter (2004), available at http://bear.cba.ufl.edu/ritter/rank.htm. 16
open-to-close returns, however, none of these variables is significant. C. How does laddering affect the first-day trading? We now investigate the impact of laddering manipulation on the trading patterns of IPOs during the first trading day. The microstructure variables used to examine the impact of laddering manipulation include the logarithm of the number of trades, the logarithm of the average volume per trade, the order imbalance, market depth, and the relative spread during the first trading day. We use the set of IPO variables used earlier to explain the trading patterns. We run three regression models for each of the microstructure variables to examine the impact of laddering likelihood on the first-day trading patterns. We first check whether these microstructure variables differ significantly for the alleged IPOs and non-alleged IPOs. The first regression model has the set of IPO variables and a laddering lawsuit dummy. The dummy variable is set to one for alleged IPOs and zero for non-alleged IPOs. The significance of the laddering lawsuit dummy variable in the first regression model reveals that alleged IPOs and non-alleged IPOs have significantly different trading patterns in the IPO aftermarket. In the second regression model, we examine whether the estimated laddering likelihood is able to explain the trading patterns for the first trading day. In the third regression model, we drop the laddering likelihood for the purpose comparison. Panel A in Table 5 presents the regression results with a laddering lawsuit dummy. The laddering dummy is significant for all of the microstructure variables except for the relative effective spread. These results show that the trading patterns of alleged IPOs and non-alleged IPOs are very different. The positive coefficients for the total number of trades and the order imbalance indicate that the alleged IPOs are traded more frequently and have more buyer-initiated trades. The negative coefficients for the average volume per trade indicate that alleged IPOs are traded with smaller size. The negative coefficient for the market depth indicates that alleged IPOs have a smaller market depth. Thus, the share prices of alleged IPOs are likely to be more volatile and are more easily affected by large trades. Panel B and Panel C in Table 5 present the regression results of the models with and without the estimated laddering likelihood, respectively. For the total number of trades, the laddering likelihood is positive and significant at the 5% level. Compared with the results of the regression without the laddering likelihood in Panel C, the 17
addition of the laddering likelihood does not mitigate the significance of the remaining variables. The estimated laddering likelihood is an additional important factor in explaining the number of trades observed in the first trading day. This result supports Hypothesis 4 in that there is a positive relation between the total number of trades and laddering manipulation. For the average volume per trade, the negative coefficient for the laddering likelihood indicates that an IPO with a higher laddering likelihood is traded with smaller trade size. The coefficient is significant at the 5% level, which is consistent with Hypothesis 5. One explanation for small size trading is that small trades are initiated by institutional investors who are aware of the laddering and wish to avoid the inventory risk of trading with large orders. It is also possible that retail investors who may not be aware of the laddering likelihood could cause more small trades. We examine these conjectures by sorting trades into various size groups in the next subsection. The estimated laddering likelihood is positive and significant at the 1% level for order imbalance. The statistics provide strong support for Hypothesis 6. Recall that the order imbalance is measured by subtracting the cumulative number of buyer-initiated trades from the cumulative number of seller-initiated trades and then dividing the result by the total number of trades. A positive coefficient indicates that IPOs with a higher laddering likelihood experience greater buying pressures. These buying pressures may come from retail investors who follow momentum in the IPO aftermarket, from institutional investors who trades for tie-in agreements, or both. For market depth and the relative effective spread, the laddering likelihood is not significant in explaining the patterns observed in both variables. This does not support Hypothesis 7 and Hypothesis 8. Based on our empirical results, we find that the laddering scheme has little impact on market depth and spread. To further investigate the relationship between the laddering likelihood and trades initiated by individual investors and institutional investors, we sort all IPO trades into three trade size groups. We then compute the microstructure variables for the trades in each group and examine how these variables are affected by the laddering likelihood. D. Trading patterns of different trade size groups 18
Barclay and Werner (1993) argue that informed traders tend to use trades of smaller sizes to hide information. These investors attempt to conceal their trades by spreading them over time. They refer to this as the stealth trading hypothesis. To examine how the tie-in agreement affects trades initiated by institutional investors and individual investors, we sort trades into large, medium and small sized groups. The size criteria are based on Griffin, Harris and Topaloglu (2005). They study which group of investors drives the price movements during the Internet bubble period. They classify investors into nine groups, including brokerage houses for individual investors, institutional investors, large investment banks, and hedge funds. Institutional trading is the most likely cause of the push in technology stock prices and the NASDAQ Composite Index during the Internet bubble period. They show that institutions tend to have an average trade size larger than 1000 shares. Individual investors tend to trade on an average trade size of 500 shares. In this paper, trades of more than 1000 shares are categorized as large. These trades are more likely to be initiated by institutions. Trades of less than 500 shares are categorized as small. Small size trades are more likely to be initiated by individual investors. Trades of size less than 1000 shares and more than 500 shares are designated as medium. Table 6 presents the sample averages of the microstructure variables for small-size, medium-size, and large-size trades. The means are computed for the overall sample, alleged IPOs, and non-alleged IPOs. From Panel A of Table 6, we find that the number of small trades for alleged IPOs is nearly three times that for non-alleged IPOs. For alleged IPOs, the small trades account for 84% of the total number of trades, but they account for only about 36% of the total trading volume. This again indicates that most of the trades for alleged IPOs are of small size. Note that the order imbalance for alleged IPOs is positive, which shows that more small-size trades are buyer-initiated. However, the order imbalance for non-alleged IPOs is negative. Because of the large underpricing and active trading patterns of alleged IPOs, these IPOs are more likely to grab the attention of individual investors who are unaware the laddering likelihood of these IPOs. In Panel C of Table 6, we find that alleged IPOs also have more large trades than non-alleged IPOs. The number of large trades accounts for only 6.5% of the total number of trades for alleged IPOs, but large trades account for 47% of the total volume. The order imbalance for alleged IPOs is -0.15 and -0.24 for non-alleged IPOs. Non-alleged IPOs experienced larger selling pressure from large trades than alleged 19