IPO underpricing and long-term underperformance

Transcription

1 Francisco Santos Stanford GSB <Job Market Paper> November 30, 2010 Abstract This paper lls in a gap in the IPO literature by documenting a close connection between IPO underpricing and the long-term underperformance of IPOs. Firms going public in periods of low underpricing do not underperform in the long run, while rms going public in high underpricing periods do. Furthermore, IPOs in later stages of high underpricing periods underperform even relative to their o er prices, which suggests that many of the most "underpriced" IPOs are in fact priced above fundamental value. This result cannot be explained by di erences in risk, and it is unlikely to be driven by a peso problem. I also nd that rms going public in later stages of high underpricing periods display worse operating performance and pro tability, lower asset growth, lower investment rates and higher cash holdings. Finally, I provide evidence that investor sentiment is stronger in high-underpricing periods. These results are consistent with a story in which low quality rms, in periods in which the average underpricing in the market is high, try to exploit investors sentiment by going public. I am extremelly grateful to my dissertation committee: Dirk Jenter (chair), Francisco Perez-Gonzalez, and Ilan Kremer for many insightful comments and guidance. I am also grateful to Anat Admati, John Beshears, Jules van Binsbergen, Darrel Du e, Liran Einav, Arthur Korteweg, Charles Lee, Andrey Malenko, Nadya Malenko, Paul P eiderer, Ilya Strebulaev, Yanruo Wang and Je rey Zwiebel for helpful comments and discussions. I am responsible for all remaining errors. Address for correspondence: Graduate School of Business, Stanford Business, 518 Memorial Way, Stanford, CA, fjsantos@stanford.edu 1

2 1 Introduction The initial public o ering (IPO) is an important step in a rm s life and a central issue in corporate nance. As such, the topic has received widespread attention in the literature. Enhancing the attention given to IPOs is the fact that IPOs are associated with some interesting empirical patterns. Two of the most important ones, also called anomalies of the IPO market, are positive rst day returns (so-called underpricing) and long-run underperformance. The underpricing of IPOs is one of the most studied anomalies in nancial economics (McDonald and Fisher, 1991; Logue, 1973; Ibbotson, 1975; Ritter, 1984; Ibbotson, Sindelar and Ritter, 1988; Ritter, 1991; Ritter and Welch, 2002; Ritter and Loughran, 2002; among others), and this attention is driven in part by the magnitude of the phenomenom - for the last 40 years, the average rst-day return is around 20%. The long-run underperformance describes the observation that IPO rms, over the 1 to 5 year period, subsequent to the IPO, tend to underperform relative to the market and relative to public rms with similar characteristics (Ritter, 1991; Loughran and Ritter, 1995; among others). Although extensive research has been done trying to explain either IPO underpricing or the long-term underperformance of IPOs, the two issues are usually not analyzed jointly as most papers address one issue without taking into account the existence of the other 1. The objective of this paper is to examine the connection between IPO underpricing and long-term underperformance. More speci cally, I analyze whether rm-speci c underpricing or average underpricing in the market, de ned as the average underpricing for all recent o erings, determines the degree of subsequent long-term underperformance. My central nding is the existence of a close relation between periods of high/low rst day returns in the market and subsequent long-term performance. Five-year wealth relatives comparing rms going public in low-underpricing or in the beginning of high-underpricing periods with public rms matched on book value, market value, industry and operating performance are not signi cantly di erent from one. Wealth relatives for rms going public in the late stages of high-underpricing periods are well below one. This implies that the long-run underperformance of IPO rms documented by Ritter (1991) and Loughran and Ritter (1995), among 1 Notable exceptions are Ritter(1991), Krigman, Shaw and Womack (1999), and Ljungqvist, Nanda and Singh (2006). 2

3 others, is caused by rms going public in the late stages of high-underpricing periods. Moreover, assuming that the long-run price proxies for fundamental value, I nd that rms going public in the late stages of high-underpricing periods underperform even relative to their o er price. This result is di cult to reconcile with the view that IPO underpricing is a discount to fundamental value. Instead, it suggests that IPO prices, in these late stages of high-underpricing periods, include a premium over fundamental value. Having established a strong link between average IPO underpricing in the market and long-term performance, I subsequently examine the underlying forces driving such correlation. First, I show that di erences in systematic risk cannot explain the relation between IPO underpricing and long-term underperformance. For risk to explain long-term underperformance, rms going public in high-underpricing periods would have to be safer. I test this explanation by looking at earnings volatility, delisting rates, total return volatility, and the betas of rms going public. There is no evidence that IPOs in periods of high underpricing are safer than IPOs in low-underpricing periods. I also do not nd that rm risk decreases within high-underpricing periods, which would be needed to explain why the long-term underperformance is caused by rms going public in the end of high-underpricing periods. Furthermore, and in contrast to the risk-based story, I nd that large underpricing is correlated with signi cantly weaker (not stronger) rms. I show that rms going public in high-underpricing environments, subsequent to the IPO, exhibit striking di erences in terms of operating performance. Speci cally, these rms exhibit worse ROA and pro tability, invest less, display lower asset growth, and hold more cash than rms that undertake their IPOs in low-underpricing periods. I also provide suggestive evidence that the relation between IPO underpricing and longterm underperformance is not caused by a peso problem. For growth opportunities to explain the relation between underpricing and long-run performance, we would need that rms in a high underpricing environment are, on average, worse, but also more likely to become the next Microsoft, i.e. more likely to yield extremely high returns in the long-run. I test this explanation by looking at the fraction of rms going public in low-, neutral-, and highunderpricing periods that earn extremely high returns, de ned as more than 500 or 1000%, in the ve-year period after the IPO. I do not nd that IPOs undertaken in high-underpricing periods are more likely to deliver extremely high returns. Moreover, I nd that the probability 3

4 of yielding extremely high returns is the lowest in the late stages of high-underpricing periods, when long-term underperformance is more severe. This is inconsistent with the view that investors are paying high prices in high-underpricing periods hoping to discover the next Microsoft. Next, I provide evidence that investor sentiment is stronger in high-underpricing periods than in low-underpricing periods. First, I nd that rms going public in high-underpricing periods experience a higher number of stock trades and higher turnover in their rst day of trading. This is consistent with the idea that over-optimistic retail investors are driving the stock price above fundamental value on the rst day of trading. Second, I nd that three widely-used measures of sentiment - the dividend premium, the percentage of equity issues in total issues, and the University of Michigan Consumer Con dence Index - indicate that investor sentiment is stronger in high-underpricing periods. Third, I show that investors react more positively to news in periods of high underpricing. Estimating investors response to earnings announcements, I nd that cumulative abnormal returns on the day of the announcement and the next day are on average fty basis points higher in periods of high underpricing in the IPO market. This e ect is more pronounced for extreme positive news. These result are consistent with a story where some rms go public to take advantage of overvalued equity due to the presence of overly-optimistic investors. In periods of high underpricing in the market, when IPO prices seem to include a premium over fundamental value, rms with and without investment opportunities have an incentive to go public. In low-underpricing periods, when IPO prices do not include a premium over fundamental value, only rms that need nancing for positive NPV projects have an incentive to go public. Hence, post-ipo performance is di erent for rms going public in high- and low-underpricing periods. Finally, looking at the magnitude of earnings surprises and the number of earnings estimates by analysts, I provide suggestive evidence that periods of high underpricing are correlated with periods of high information asymmetry. As I will discuss in Section 5, this result may explain why rms, although raising IPO prices above fundamental value, are willing to leave money on the table in high-underpricing periods. This paper contributes to the IPO literature by establishing a close relation between average underpricing in the market at the time of the IPO and subsequent long-term underperformance. My results identify a need for explanations that jointly address both IPO 4

5 underpricing and long-term underperformance. Moreover, my results are inconsistent with the standard view that underpricing represents a discount to fundamental value (Rock, 1986; Welch, 1992; Benveniste and Spindt, 1989; Benveniste and Wilhelm, 1990; and Spat and Sristava, 1991) or a costly instrument that rms use to signal quality (Welch, 1989; Allan and Faulhaber, 1989; Chemmanur, 1993). Instead, periods of high underpricing appear to be periods in which rms are able to raise their IPO prices above fundamental value. Note that this nding o ers a simple explanation for the puzzle why so many rms appear keen on going public in periods of high underpricing. Firms leave money on the table but, at the same time, they are getting more than what they are worth 2. Finally, I nd that there are, post-ipo, substantial di erences in observable operating performance for rms that go public as a function of the average underpricing in the market. To my knowledge, this is a novel result. Theories trying to explain IPO underpricing and long-term underperformance should also be able to explain di erences in operating performance. The outline of the paper is as follows: Section 2 brie y summarizes the prior literature. Section 3 describes the data, the return benchmarks, and the procedures used to identify high- and low-underpricing periods. Section 4 presents the empirical results. I analyze the relation between IPO underpricing and long-term underperformance; rm risk; the likelihood of extremely high returns as a function of initial underpricing; the quality of IPO rms; and the relation between periods of high average underpricing in the market and investor sentiment. Section 5 discusses potential explanations for why rms may be willing to leave money on the table even if they are extracting a premium from sentiment investors. Section 6 concludes the paper. 2 Literature Review Empirical evidence on the relation between underpricing and long-run performance This paper examines the relation between the average level of underpricing in the market and the long-run underperformance of IPO rms. Two prior studies have examined the 2 I will discuss several explanations for why rms may be willing to leave money on the table in section 5. 5

6 connection between rm-speci c underpricing and long-term underperformance. Ritter (1991) documents that rm-speci c underpricing and long-run performance are negatively correlated. Using a sample of IPOs for the period , Ritter compares aftermarket returns for quintiles of industry adjusted initial returns. He shows that rms with high underpricing have the worst aftermarket performance. Krigman, Shaw and Womack (1999) also look at the relation between rm-speci c underpricing and long-term underperformance. They partition a sample of IPOs for the period by rst-day returns and show that only the most extreme rst day returns predict future performance. They nd that IPOs with underpricing above 70% and IPOs with negative rst day return underperform in the long run. I examine both rm-speci c and average underpricing in the market as a predictor of long-run underperformance. I nd that the average underpricing in the market determines the degree of subsequent long-term underperformance. Firm-speci c underpricing cannot explain long-run performance once the market e ect is taken into consideration. IPO underpricing A sizeable literature provides evidence that IPO underpricing is a persistent phenomenon in the IPO market - see Reilly (1973); Ibbotson (1975), Ritter and Welch (2002), Loughran and Ritter (2002), among others. In response to the empirical evidence, the literature has developed a variety of theories to explain underpricing. Most theories of underpricing rely on some type of asymmetric information between issuers, underwriters and investors. If the issuer is more informed than investors, the money left on the table through underpricing can work as a signal of higher quality (Welch, 1989; Allan and Faulhaber, 1989; Chemmanur, 1993). Alternatively, in a model where some investors are more informed than rms, for example about market conditions, rms provide a discount either to guarantee that uninformed investors break even (Rock, 1986) or to induce informed investors to share information (Benveniste and Spindt, 1989; Benveniste and Wilhelm, 1990; Spat and Srivastava, 1991; Welch,1992). Another set of theories tries to look at IPO allocations as the main driver for underpricing - Benveniste and Spindt (1989), Booth and Chua (1996), Brennan and Franks (1997), Mello 6

7 and Parsons (1998), Stoughton and Zechner (1998) Sherman (2000), Sherman and Titman (2002), and Loughran and Ritter (2002) are examples. Underpriced o erings create excess demand which allows discretion by rms on whom to allocate shares. Implicit, is the argument that some shareholders are more desirable than others. Long-run underperformance of IPO rms Another large literature provides evidence that IPO rms tend to underperform in a 1-5 year period subsequent to the IPO. For example, Ritter (1991), and Loughran and Ritter (1995) report a 16 percent buy-and-hold return for IPO rms while their comparable sizematched rms earn 66 percent over the ve year period. Loughran and Ritter interpret their evidence as investors being too optimistic about the prospects of rms issuing equity for the rst time. Moreover, Loughran, Ritter and Rydqvist (1994) claim that rms time their IPOs to coincide with periods of excessive optimism, consistent with ndings in Lee, Shleifer, and Thaler (1991) that more companies go public when investor sentiment is high. Another contribution of my paper is that I look at buy-and-hold returns including and excluding rst day returns. Including the rst day return means that the long term performance is measured from the IPO price, while excluding the rst-day return means that performance is measured from the rst day closing price. Previous papers examine long term underperformance starting from the closing price at the rst day of trading. Observing longterm underperformance excluding the rst-day return is not enough to claim that rms take advantage of excessive optimism. However, long-term underperformance including the rst day return implies that IPO prices are above fundamental value which would be consistent with the claim that IPO prices include a sentiment premium. It is still unclear how abnormally poor post-ipo performance is. Long-run returns, even if extremely low, are su ciently noisy to make statistical inference di cult. As Welch and Ritter (2002) point out in Brav (2000) it can require an abnormal return of -40 percent to reject the hypothesis of underperformance. However, the evidence strongly suggests that IPOs and rms with similar characteristics exhibited poor performance in times in which the overall market performs exceptionally well. Miller (1977) tries to explain IPO long-term underperformance with short-sale constraints on IPOs and heterogeneous beliefs about fundamental values. Immediately after the IPO, the 7

8 marginal investor is the most optimistic among all investors but as time passes, di erences in valuations decrease and the marginal investor s valuation converges to the mean valuation. Hence, in the long run IPO rms underperform. Most common explanations for the long run underperformance rely on some type of overcon dence - Teoh, Welch, and Wong (1998), Heaton (2001), Bernardo and Welch (2001), Daniel, Hirshleifer, and Subramanyam (1998). The main idea is that investors in the shortrun overshoot fundamental value and in the long run prices revert to the correct level. To summarize, the prior literature in underpricing and long-term underperformance suggests that rms go public at a discount to fundamental value which leads to positive rst day returns; investors overshoot fundamental value on the rst day of trading, and in the long run, prices revert to their fundamental value. Connecting underpricing and long-term underperformance The empirical results presented in this paper are in line with the theoretical work of Ljungqvist, Nanda and Singh (2006). Di erent from most of the IPO literature, which usually addresses each issue separately, they develop a model of IPO pricing that connects underpricing and long-term underperformance. A major assumption of their model is that, on occasion, some investors are "irrationally exuberant" about IPOs. In a setting where shorting IPO shares is di cult or too costly, the presence of sentiment investors implies that some periods should display long-term underperformance excluding the rst day return. Moreover, in times where these sentiment investors are likely to show up, rms raise IPO prices in order to exploit them. This is consistent with my result that in certain periods we observe long-term underperformance even relative to the IPO price. Ljungqvist, Nanda and Singh (2006) model also tries to explain why, although rms raise IPO prices above fundamental value, they do not fully exploit sentiment investors. In their model, high underpricing is compensation given by rms to intermediaries for bearing the risk of carrying overvalued IPO shares in inventory. 8

9 3 Data, IPO cycle and Benchmark 3.1 Data The initial sample includes all US IPOs completed between January 1, 1973 and December 31, 2008 as reported by Thomson Financial s Securities Data Company (SDC) database. Unit o ers, spin-o s, closed-end funds, IPOs with an o er price below $1.00, Real Estate Investments Trusts (REITs), American Deposit Receipts (ADRs) and nancial rms as described by SDC are excluded from the sample. From SDC I obtain the date of issue, the dollar value of proceeds raised and the rst day return. I obtain trading price histories from the Center for Research in Security Prices (CRSP) database. I exclude from the sample observations for which no match is found on CRSP based on the Cusip reported by SDC. I also exclude observations for which mismatches between CRSP and SDC data make it di cult to identify the correct IPO date. In particular, I exclude cases for which the trading price history on CRSP starts on a date prior or four days after the reported IPO date by SDC. This yields a sample size of 6,256 IPOs for which there is information on rst-day underpricing. Table 1 provides information on IPO activity in the U.S. from 1973 to I report the number of IPOs, average rst day returns, average and aggregate proceeds, and average and aggregate money left on the table per year. Proceeds are de ned as number of shares issued times IPO price; money left on the table is de ned as proceeds times rst day return. As documented by the prior literature, we see that the number of IPOs per year is highly volatile. In 1975, only 5 rms undertook an IPO, while 599 rms decided to go public in In terms of average rst day returns, not a single year from 1973 to 2008 exhibits a non-positive average rst day return. As with the number of IPOs, the yearly time series of average rst day return displays high volatility, with values ranging from 1.9% in 1974 to 78.6% in Figure 1 displays the number of IPOs per month and the monthly average underpricing; underpricing and volume volatility is clearly visible as well. The amount of money raised per IPO is increasing overtime. For the entire sample period, the average IPO raises 72.5 million dollars, but, this amount doubles for the 2000 to 2008 sub period. Overall, this sample of IPOs was able to raise 496 billion dollars, but left 122 billion dollars on the 9

10 table through underpricing. In 1999 and 2000 alone, the money left on the table amounted to 67 billion dollars as a result of high underpricing and high volume. The magnitude of money left on the table is puzzling for at least two reasons. First, if underpricing is a discount on fundamental values, why do we observe higher IPO volume when this discount increases? Second, according to asymmetric information models, high levels of underpricing are caused by severe information asymmetries between rms, underwriters, and investors. Hence, any strategy that helps decrease information asymmetries would lower underpricing. For example, bundling IPOs would decrease the level of uncertainty about the average rm. However, we do not observe such strategies being implemented even when the money left on the table amounts to 67 billion dollars over two years. 3.2 High- and low-underpricing periods The primary goal of this paper is to study the relation between the level of IPO underpricing in the market, measured as monthly average underpricing, and subsequent long-run performance. Motivating this choice is the fact information asymmetries between rms, underwriters, and investors can explain underpricing but not long-term underperformance. It might then be the case that the long-term underperformance is caused by sentiment investors driving up prices in the rst-trading day. If rms do not fully exploit these sentiment investors, we would observe high rst day returns in the IPO market. Hence, it seems reasonable that average underpricing in the market correlates with the presence of optimistic investors more powerfully than rm-speci c underpricing. The rst step is to de ne if, at a given moment, the IPO market is displaying low or high levels of underpricing. In line with Helwege and Liang (2002), I divide the IPO cycle into three phases: low, neutral and high underpricing. I calculate a time series of the average monthly underpricing from 1973 to For each month I compute the distribution of the previous 120 monthly average underpricing measures. A month is considered high underpricing if its underpricing is higher than the 75th percentile; low if its underpricing is below the 25th percentile; all remaining months are classi ed as neutral. It might be that one single month of high underpricing is driven by outliers and does not re ect the state of the IPO market. Hence, using low, neutral, and high months I 10

11 de ne low, high, and neutral periods. A high-underpricing period consists of three or more consecutive high-underpricing months. Low-underpricing periods are de ned analogously, while all other periods still unde ned are considered neutral-underpricing periods. In order to capture potential dynamics within high-underpricing periods, I subdivide high-underpricing periods into beginning, middle, and end. I do this in a simple manner by dividing each period into three parts (and giving the middle part the biggest share when the number of months is not divisible by three). The prior literature subdivides the IPO market into hot and cold periods based on IPO volume, not underpricing 3. To compare results I perform a similar classi cation. For each month I compute the distribution of the number of IPOs per month during the last 120 months. A month is considered high volume if the number of IPOs in that month is higher than the 75th percentile; low if the number of IPOs is below the 25th percentile; all remaining months are de ned as neutral. High-, low-, and neutral-volume periods are de ned analogously to what is done under the underpricing de nition. Figure 2 shows that "hot" periods in terms of underpricing and volume do not coincide. For example, in 1978 we see high underpricing but low volume, while in 1986 and 1987 we observe high volume but low rst day returns. In the analysis section below, we will nd that the two classi cations of the IPO market have very di erent implications for the subsequent performance of IPO rms. Table 2 provides information on the number of IPOs and the average rst day return by the state of the IPO market. Panel 1 uses the classi cation into months while Panel 2 uses the division into longer periods. 688 rms go public in low-underpricing periods with an average rst day return of 6.3%. In high-underpricing periods, the number of rms going public more than doubles and rst day returns jumps to 41.6%. Looking at the volume de nition, there are no signi cant di erences in average underpricing between high- and neutral-volume periods. Within high-underpricing periods we do not observe signi cant variance or any clear trend in the level of rst-day returns. However, the number of IPOs almost doubles from the beginning to the end of high-underpricing periods. This increase in volume within high-underpricing periods is puzzling if higher underpricing represents a large discount to fundamental value, as suggested by prior literature. However, the increase in volume in high- 3 Notable exceptions are Ritter(1984), and Helwege and Liang (2002), among others. 11

12 underpricing periods makes sense if high underpricing coincides with IPO prices that are set above fundamental value. 3.3 Benchmarking returns A proper benchmark is fundamental to analyze IPO rm performance. However, such benchmarking is di cult for rms that just went public due to the lack of information about rm characteristics prior do the date of the IPO. In this paper, I follow the literature and match IPO rms with public rms with similar characteristics at the time of the IPO. The criteria used to match IPO rms with public rms is based on Lie (2001) and uses book value of assets, market size, return on assets (ROA), and industry. For each IPO rm, I retrieve from Compustat the rm s book value, market size, ROA, and SIC code at the end of the IPO year. Then, for each IPO rm, I try to obtain control rms from the universe of rms on Compustat. I want a matching procedure that is precise but also does not result in a considerable loss of observations. To satisfy these two goals, the matching process implements sequential sets of criteria. Speci cally, I de ne the following three sets: I - same three-digit SIC code, and market size, ROA, and book value of assets between 70% and 130% of the IPO rm values; II - same two-digit SIC code, and market size, ROA, and book value of assets between 70% and 130% of the IPO rm values; III - market size, ROA, and book value of assets between 70% and 130% of the IPO rm values. I do not allow a control rm to be chosen if its IPO was less than ve years prior to the IPO of the event rm as I do not want IPO rms controlling for IPO rms. I apply criteria I and choose the ve rms closest to the IPO rm in terms of market size. If criteria I does not yield ve control rms, then I apply criteria II. If I and II are still not enough I apply criteria III. In the end, for each IPO rm I obtain a portfolio of ve control rms as benchmark. Choosing only the closest rm as the benchmark, or using book-to-market instead of market size as the factor to decide closeness to the IPO rm, does not change the results presented below. 12

13 4 Empirical Analysis 4.1 Underpricing and long term underperformance Following Ritter (1991), Loughran and Ritter (1995) among other, the strategy implemented to test for long run underperformance is to compare buy-and-hold returns for IPO rms and control rms matched on book value, market value, industry, and operating performance. I compute ve-year buy-and-hold returns using daily and monthly returns from CRSP. I demand that IPO rms have at least one year of return history after the IPO but if, after one year, information is missing for an IPO stock I do not drop that rm. For the benchmark rms, I only include returns for the period for which data is non-missing for the IPO rm. For example, if an IPO rm has only four years of returns I use those four years for the IPO rm and benchmark rms. This procedure mitigates sample selection bias by not dropping rms with limited return histories. Finally, for each rm I compute 5y CAR 1d and 5y CAR 0d which represent, respectively, ve year cumulative return excluding and including rst day return. Most studies of IPO long term underperformance exclude the rst day return, reasoning that the price at the end of the rst trading day is a better proxy for rm value than the IPO price. I follow a more general approach by computing buy-and-hold returns both relative to the IPO price and relative to the price at the end of the rst trading day. Di erences between these two measures should help identify the correct interpretation of IPO underpricing. If underpricing is to be interpreted as a discount to fundamental value, then one should not observe long-term underperformance relative to the IPO price. If, however, rms underperform even relative to the IPO price, then underpricing cannot be interpreted as a discount to fundamental value as it seems that IPO prices are set above fundamental value. The next step is to analyze the relation between long term underperformance, measured by 5y-CARs, and underpricing at the time of the IPO. For that purpose, I run several regressions of 5y CAR 1d on di erent de nitions of underpricing, testing for a relation at the rm and market level. Table 3 presents the rst important result of this paper. There is a strong negative relation between long-term performance of IPO rms and the average IPO underpricing at 13

14 the market level. Moreover, this long-run underperformance is more severe for rms going public after several months of high underpricing in the market. Column (1) shows a negative relation between rm speci c underpricing and subsequent long-term performance. A one percentage point increase in initial underpricing leads to a statistical signi cant decline of 43 basis points (bps) in the 5y CAR 1d. However, regression (1) is unable to show if subsequent performance of a rm going public is a consequence of its own underpricing or overall IPO market underpricing. Regression (2) tackles this issue by including the average market underpricing in the month before the IPO. The estimates show that the market e ect dominates as the rm-speci c underpricing no longer predicts future long-term underperformance. Also, the magnitude of the long-term underperformance driven by market underpricing is larger. A one percentage point increase in market underpricing one month before the IPO implies a decline of 107bps in the 5y CAR 1d. Results in regression (2) shows that the long-term underperformance of IPO rms is more severe for IPOs in periods of high average market underpricing. Column (3) provides further evidence. I replace the value of the average market underpricing by two dummies: Neutral Month and High Month which are equal to one if the month preceding the IPO classi es as neutral- or high-underpricing, respectively. Firms going public after a high-underpricing month perform substantial worse than rms going public after a low-underpricing month. Regressions (4) and (5) show that the long-term underperformance in high-underpricing periods is more pronounced for rms going public after several high-underpricing months. Including the number of consecutive low-, neutral- and high-underpricing months prior to the IPO plus the interaction with rm-speci c underpricing, I nd a negative relation between the number of consecutive high-underpricing months before the IPO and 5y CAR 1d. One additional month in the number of consecutive high-underpricing months before the IPO entails a decrease on 5y CAR 1d of 202 bps if I use the value of average market underpricing, or, 454bps if I use dummies for average market underpricing. Firm-speci c underpricing and number of consecutive low-underpricing months do not seem to explain long-term underperformance. Table 4 presents evidence that rejects a relation between rms going public in periods of low/high volume and long-run performance. Regressing 5y CAR 1d on rm speci c underpricing and number of IPOs in the month preceding the IPO, I nd that high volume 14

15 can predict subsequent poor long-run performance. However, adding the average underpricing in the market in the month prior to the IPO shuts down both the rm speci c underpricing and the IPO volume predictive power. Replacing number of IPOs by dummies representing how "hot" the market is in terms of number of IPOs in the month prior to the IPO does not change the results. Under the underpricing de nition, long-term underperformance is more severe for IPOs in later stages of high-underpricing periods. We do not observe the same e ect using periods of high volume. Regressions (4) and (5) show that rms going public after several months of high volume do not experience worse long-run performance. Tables (3) and (4) provide evidence that, between rm-speci c underpricing, IPO volume, and average market underpricing, the latter e ect is the dominant one in terms of predicting long-run underperformance. Another way to address the relation between IPO underpricing and long-run underperformance is to compare wealth ratios between low, neutral and high underpricing/volume periods. I compare how much money an investor gets by investing in an IPO rm and holding it for ve years to the same investment in the portfolio of ve rms matched on book value, market capitalization, industry, and operating performance. A wealth ratio below one means that the investor would have been better o with an investment in the control rms. Table 5 presents wealth ratios between IPO and benchmark rms for the di erent phases of the IPO cycle based on the underpricing and volume de nitions. Panel 1 excludes rst day returns, which is the correct measure for an investor buying at the end of the rst-trading day. Panel 2 includes the rst day return, thus, captures the return obtained by an investor who is allocated shares at the IPO price. For the period, investors who are allocated shares at the IPO price do not lose money, as the wealth ratio of 0.97 is not statistically di erent from one. However, an investor that buys at the end of the rst trading date would be better o investing in the control rms as she only achieves a wealth ratio of 0.84 by investing in IPO rms. Dividing the sample into periods of low-, neutral-, and high-underpricing periods, we observe that long-term underperformance is more severe in periods of high-underpricing. Excluding the rst day return, IPOs in high-underpricing periods display a ve year wealth ratio of only 0.66, while, for IPOs in low-underpricing periods, the wealth ratio is Moreover, within high-underpricing periods, we see that long-run underperformance is particularly se- 15

16 vere for IPOs in the later stages. In the end of high-underpricing periods, IPO rms earn a ve-year buy-and-hold return of -7.8%, while benchmark rms earn 61.2%. Measuring longrun performance relative to the IPO price reveals an interesting result. Firms going public in low and neutral periods do not under or over-perform benchmark rms but rms going public in high-underpricing periods do underperform. Subdividing these high-underpricing periods, we see that in the beginning there is no underperformance but in later stages this underperformance is large as evidenced by a wealth ratio of This result is di cult to reconcile with the view that IPO underpricing is a discount to fundamental value. Assuming that the long-run price proxies the fundamental value, it seems that IPO prices are set equal to fundamental value in low-, neutral-, and beginning of highunderpricing periods; and above fundamental value in the late stages of high-underpricing periods. Under a classi cation of periods based on volume we do not see major di erences in longterm underperformance. Excluding the rst day return, there are no di erences between neutral- and high-volume periods or within high-volume periods. Including the rst day return, all wealth ratios are not signi cantly di erent from one. In this section I presented the two main ndings of this paper. The rst nding is that rms going public in low-underpricing periods do not underperform in the long-run but rms that go public in high-underpricing do. This long-run underperformance is due to rms going public in the later stages of high-underpricing periods. The second nding is that rms that issue in the late stages of high-underpricing periods exhibit long-term underperformance even relative to the o er price. Established this relation between average market underpricing and long-term underperformance, in the next section, I test if risk can explain the relation. 4.2 Long-term underperformance and risk One way to rationalize the results in long-term underperformance presented in the previous section is through risk. In order for risk to explain the worse long-run performance of rms going public in high-underpricing periods, we should see safer rms going public in these periods. Furthermore, rm risk should decrease within high-underpricing periods given that we observe worse long-run performance for rms going public in the late stages of high- 16

17 underpricing periods. I test this hypothesis by looking at several measures of risk: earnings volatility, delisting rates, volatility of returns, and rm s beta over the ve-year period after the IPO. Earnings volatility is computed as the standard deviation of reported quarterly earnings; volatility of returns is the standard deviation of monthly returns; and the rm s beta is estimated from the CAPM. Quarterly earnings are retrieved from Compustat, and delisting codes and daily returns come from CRSP. For delisting rates, the dependent variable is one if the rm is delisted or zero if it stays on CRSP for the ve-year period. If the patterns in returns are caused by risk we should observe lower earnings volatility, delisting rates, return volatility and betas for rms going public in high-underpricing periods and even lower for IPOs in the late stages of high-underpricing periods. Tables 6 to 10 reject this claim. Table 6 shows there is no relation between earnings volatility and IPO underpricing. Firm-speci c underpricing and average market underpricing at the time of the IPO is not related to subsequent earnings volatility. In Table 7, I present logit estimates for delisting rates. Speci cations (3) and (5) show that IPOs following a high-underpricing month are more likely do be delisted from CRSP than IPOs following a low-underpricing month. This e ect is not ampli ed with number of high-underpricing months preceding the IPO. Results in volatility of returns on Table 8 point into the same direction as delisting rates. We see that rm-speci c underpricing is positively correlated with return volatility. Including average market underpricing we see that higher market underpricing is associated with higher subsequent volatility. Using dummies for level of underpricing in the market does not change the result that rms issuing after a high-underpricing month exhibit higher return volatility. Adding the number of high-underpricing months preceding the IPO magni es the e ect. One concern about this measure is that it does not separate between idiosyncratic and systematic risk. Table 9 addresses the issue by presenting results for CAPM betas. Results on betas are consistent with the ones on delisting rates and total volatility. Firms going public in high-underpricing periods display higher betas. Speci cations (4) and (5) also show that betas are higher for rms going public after several consecutive high-underpricing months. These results suggest that, if anything, rms going public in high-underpricing periods are more risky, and de nitely not safer, than rms going public in low-underpricing periods. 17

18 In table 10, I present estimates for the following regressions: Y i = Neutral i + 2 High i + " i Y i = Low i + 2 Neutral i + 3 Middle of High i + 4 End of High i + " i where the dependent variables are earnings volatility, delisting rates, return volatility and CAPM betas. Neutral, High, Middle of High and End of High are dummy variables equal to one if the IPO is undertaken in a neutral-, high-, middle of high-, or end of high-underpricing period, respectively. In panel 1, we see that, in the ve years following the IPO, rms going public in highunderpricing periods display higher total return volatility and higher betas, and are more likely to be delisted. There are no di erences in terms of earnings volatility. Panel 2 shows that rms going public in the beginning of high-underpricing periods are more likely to get delisted, exhibit higher earnings volatility, return volatility and higher betas than rms going public in low and neutral-underpricing periods. We do not observe signi cant di erences within high-underpricing periods. Results presented in this section are not consistent with a risk-based explanation for the relation between average market underpricing and subsequent long-run underperformance. We do not observe safer rms going public in high-underpricing periods or in the late stages of high-underpricing periods. Results suggest the opposite as it seems that riskier rms decide to go public in high-underpricing periods. 4.3 Peso problem In the previous section, I showed that the documented relation between average underpricing in the market and subsequent long term-underperformance is not driven by risk. An alternative explanation is that such relation is caused by a peso problem. Under this explanation, di erent returns in the long run are explained by di erent growth opportunities at the time of the IPO. For such explanation to hold, we would need that the average rm going public in high-underpricing is worse than the average rm going public in low-underpricing periods, and also that the probability of providing huge returns is higher for high-underpricing pe- 18

19 riods. The long-term underperformance comes from investors paying more for worse rms, hoping to discover the next Microsoft. I test this explanation by analyzing if rms going public in high-underpricing periods are more likely to provide extremely high returns in the long run. Table 10 provides suggestive evidence that the relation between IPO underpricing and long-term underperformance is not driven by a peso problem. Table 10 shows that IPOs in high-underpricing periods are not more likely to provide extremely high returns, de ned as ve-year buy-and-hold returns above 500 or 1000%. We actually observe that is more likely to nd the next Microsoft in rms going public in low-underpricing periods than in high-underpricing periods. Moreover, the probability of yielding extremely high returns is the lowest in the late stages of high-underpricing periods, when long-term underperformance is more severe. Table 11 reinforces results in Table 10. Performing logit regressions where the dependent variable is one if the rm provides extremely high returns, I do not nd that IPOs in highunderpricing environments are more likely to provide these huge returns in the long run. Furthermore, looking at speci cation (5), we see that rms going public after several highunderpricing months are less likely to yield extremely high returns after the IPO. This result is inconsistent with the proposed explanation. If in fact, the relation between IPO underpricing and long-term underperformance is caused by a peso problem, we should see exactly the opposite. 4.4 IPO timing In previous sections, I showed that rms going public in the late stages of high-underpricing periods display the worst subsequent long-run performance. These rms even underperform relative to the o er price. This result suggests that, during high-underpricing periods, IPO prices are set above fundamental value. The idea that rms are sold at a premium although uncommon is not an original one. Using intrinsic value based on industry-matched price/sales and price/ebitda from comparable publicly traded rm, Purnanandam and Swaminathan (2001) nd that o er prices are priced 50% above comparables. This premium over fundamental value can be the result of rms trying to exploit excessively optimistic investors. If 19

20 that is the case, then rms have an incentive to time their IPOs in order to coincide with periods of overvaluations by investors. Pagano et al. (1998) nds support that rms try to time their IPOs so they can take advantage of industry-wide overvaluations, rather than to nance their growth. Based on a sample of Italian rms, and using both ex ante and ex post information on characteristics and performance, they nd that rms are more likely to go public when the average marketto-book ratio of public rms is high. Moreover, they nd that higher market-to-book ratios do not re ect investment opportunities as companies tend do go public after and not before periods of high investment. We see rms going public in low-underpricing periods, when it seems there is no premium to fundamental to be exploited. It might be that these rms decide to go public because they need capital to nance investment opportunities. The optimal scenario is to go public when IPO prices include a premium but if rms cannot wait because, for example, investment opportunities be exploited by others, then rms are willing to undertake an IPO even in a low-underpricing period. In high-underpricing periods, given the premium to fundamental value, both rms with and without investment opportunities have an incentive to go public. Hence, post-ipo, we should observe di erences in rms characteristics between rms going public in low- and high-underpricing periods. I test this claim by looking at operational performance measured as return on assets (ROA), total pro tability, investment rates, asset growth, and cash holdings. ROA is de ned as operating income before depreciation over assets; total pro tability is income before extraordinary items over assets; investment rate is capital expenditures over assets; and cash holding is cash over assets. All data is retrieved from Compustat. Everything else the same, after the IPO, rms going public in low-underpricing periods should display higher ROA, higher pro tability, should invest more, hold less cash, and exhibit higher asset growth. I run the same set of regressions as in previous sections, where the dependent variables are the di erences between the IPO rm and control rms, in each speci c measure, and one year after the IPO. Di erences in each measure are winsorized at the ve percent level. Independent variables are rm-speci c underpricing, average market underpricing in the month prior to the IPO, dummies for neutral and high month if the month prior to the IPO classi es as neutral- or high-underpricing month, and number of consecutive low-, neutral-, or 20

21 high-underpricing months prior to the IPO. Tables 13 to 18 present evidence that rms going public in high-underpricing periods seem to be of lower quality as they exhibit, after the IPO, worse ROA and pro tability, invest less, grow more slowly, and hold more cash. Moreover, these patterns are substantially more pronounced for rms going public in the late stages of high-underpricing periods. Table 13 and Table 14 present the results for ROA and overall pro tability. Regressions (1) and (2) show that rm-speci c and average market underpricing have a negative e ect on ROA and pro tability. These e ects are economically signi cant as an increase of one percentage point in average underpricing in the month prior to the IPO predicts a decrease of around 17 basis points in ROA and pro tability. Replacing average market underpricing by dummies representing level of underpricing in the market still yields negative coe cients for high underpricing but breaks the statistical signi cance of the e ect. I also nd that the number of consecutive high underpricing months has a negative impact in future ROA and pro tability. The fact that rms going public in the late stages of high-underpricing periods register the worst performance in terms of subsequent ROA is consistent with a story in which low quality rms rush into the market trying to exploit a premium to fundamental value. Results in asset growth and investment rates are presented in Tables 15 and 16. We see that average market underpricing in the month prior to the IPO is negatively related with asset growth and investment rates on the rst year after the IPO. Moreover, speci cation (5) suggest, once again, that rms going public after several high-underpricing months are the ones experiencing the lowest asset growth and lowest investment rates. Finally, results concerning cash holdings are presented in Table 17. In this case, rmspeci c underpricing, rather than market underpricing, has a positive e ect on cash holdings. However, regressions (5) and (6) show that rms going public after several high-underpricing months hold more cash. An alternative way to test di erences in post-ipo performance is to run the following regressions: Y i = Neutral i + 2 High i + " i Y i = Low i + 2 Neutral i + 3 Middle of High i + 4 End of High i + " i 21

22 where dependent variables are the performance measures presented in this subsection. Neutral, High, Middle of High and End of High are dummy variables equal to one if the IPO is undertaken in neutral-, high-, middle of high-, or end of high-underpricing period, respectively. Panel 1 of Table 18 con rms that rms going public in high-underpricing periods exhibit, post-ipo, lower ROA, pro tability, asset growth, and higher cash holdings. Economic magnitude of the e ects is large, ranging from six to seven percentage points. Panel 2 shows that the di erences observed in Panel 1 are caused by rms going public in the middle- and end of high-underpricing periods. Firms going in the late stages of high-underpricing periods display substantial lower ROA, pro tability, asset growth, investment rates and higher cash holdings. There are no signi cant di erences between post-ipo performance of IPOs in the beginning of high-underpricing periods and neutral- or low-underpricing periods. Results shown in this section are consistent with a story in which rms try to time their IPOs to coincide with periods of high underpricing, when IPO prices include a premium to fundamental value. This explains why the number of IPOs goes up when underpricing is high (recall Table 2). Firms with and without investment opportunities rush into the market to take advantage of overvaluations. It is still puzzling why rms do not fully exploit overvaluations as they still leave money on the table, but is not a puzzle that they prefer to go public when prices are above fundamental. 4.5 Underpricing periods and investor sentiment In this section, I test if investor sentiment is stronger in high-underpricing periods. Results presented in previous sections are consistent with a story in which rms in high-underpricing periods try to extract rents from overly optimistic investors in the aftermarket. For this to hold, we should observe stronger investor sentiment in these periods. Clearly, this is not an easy task since measuring sentiment is di cult and no consensual measure exists. I propose to address the issue by looking at the behavior on the rst-trading day, correlation of sentiment proxies with underpricing, and investors response to earnings announcements. First, I look at the trading behavior on the rst-trading day. If high-underpricing periods are driven by sentiment investors we should observe more "excitement" on the rst-trading 22

23 day of high-underpricing periods. I test this claim by regressing the number of stock trades and turnover percentage, measured as volume over number of shares outstanding, on the rsttrading day on dummies representing neutral-, and high-underpricing periods. I also analyze sub periods of high-underpricing periods. In summary, I perform the following regressions: Y i = Neutral i + 2 High i + u i Y i = Low + 2 Neutral i + 3 Middle i + 4 End i + i where dependent variables are number of stock trades and turnover. Estimates on Table 19 show that in high-underpricing periods, number of stock trades, and turnover on the rst-trading day of rms going public in high-underpricing periods is higher than in low-underpricing periods. Panel 1 shows that, in low-underpricing periods, 16% of all shares outstanding of an IPO rm are transacted on the rst-trading day. This is done with roughly 2,300 trades. Turnover for rms going public in high-underpricing periods is 27%. Furthermore, this eleven percentage point di erential in turnover is accompanied by triple the number of trades. Panel 2 provides further insights on the question. Within highunderpricing periods, I do not nd signi cant di erences in number of trades, and turnover on the rst-trading day. These results suggest that "excitement" on the rst-trading day is higher in high-underpricing periods and remains relatively constant throughout these periods. As a second approach to test if high underpricing is correlated with investor sentiment, I analyze the relation between IPO underpricing and widely used proxies of investor sentiment. Namely, I use the same variables as Baker in Wurgler (2006): dividend premium, number of IPOs, NYSE turnover, closed-end fund discount, percentage of equity in total issues, and I add the University of Michigan Consumer Con dence Index (CCI). The dividend premium is de ned as the di erence between the average market-to-bookvalue ratios of dividend payers and nonpayers (Baker and Wurgler, 2004). Dividend-paying stocks have a more predictable stream of incomes which is a characteristic of safety. Baker and Wurgler (2004) claim that rms when deciding whether to pay dividends appear to accommodate sentiment for or against "safety". Higher dividend premium is interpreted as a signal that investors are pessimistic and looking for safer investments. IPO volume is commonly associated with investor sentiment. Occasionally, there are 23

24 "windows of opportunity" that seem to ease the process of going public. If these "windows of opportunity" are triggered by optimistic investors then we should see a positive relation between number of IPOS and investor sentiment for a given period. The NYSE turnover is the ratio of trading volume to the number of shares listed on the New York Stock Exchange. Volume can be viewed as an investor sentiment index. If shortselling is more costly than opening and closing long positions, then sentiment investors are more likely to trade when they are optimistic (Baker and Stein, 2004). Hence, high NYSE turnover suggests high investor sentiment. The closed-end fund discount is another used proxy for investor sentiment. Closed-end funds are funds with a xed number of shares that trade on stock exchanges. The closed-end fund discount (sometimes premium) is the di erence between the net asset value of the fund s security holdings and the fund s market price. If closed-end funds are largely held by retail investors, the observed discount can proxy sentiment (Zweig, 1973; Lee, Schleifer, and Thaler, 1991; Neal and Wheatley, 1998). When the average discount on closed-end funds goes up it suggests that retail investors are pessimistic. The percentage of equity in total issues is also argued by Baker and Wurgler (2000) to be related to investor sentiment. They nd that the equity share can foretell low stock market returns and that the pattern can be the result of rms shifting between equity and debt, successfully reducing cost of capital. A high equity share suggests high optimism by sentiment investors. I also look at the University of Michigan Consumer Con dence Index (CCI). Although this survey does not ask directly consumers for their views on securities prices, changes in the CCI correlate highly with changes with the UBS/Gallup index that targets investors. Moreover, Qiu and Welch(2006), and Lemmon and Portniaguina (2006) show that CCI changes correlate especially strongly with small stocks and returns of rms held mainly by retail investors. Hence, a positive relation between CCI and investor sentiment is expected. These investor sentiment proxies also re ect economical fundamentals to some extent. For example, IPO volume is partially determined by investment opportunities. To iron out information about fundamentals I regress each proxy on a set of macroeconomic variables: growth in industrial production, real growth in durable, nondurable, and services consumption, growth in employment, and an NBER recession indicator. I use the residuals from these 24

25 regressions as sentiment proxies. Sentiment proxies are then dependent variables of the following regression: Sentiment i = Neutral i + 2 High i + u i Table 20 shows that c 2 has the predicted sign for the six sentiment proxies, although the closed-end fund discount is not statistically signi cant. The relation is particularly strong for the dividend premium, percentage of equity in total issues and the CCI. For number of IPOs and NYSE turnover, although we observe a signi cant di erence between high- and low-underpricing periods, we cannot reject that sentiment in high- and neutral-underpricing periods is di erent. Finally, I test if in high-underpricing periods investors are being more optimistic than usual. It might be the case that investors are not only optimistic about IPO rms and that they can also drive up other prices. One way to test if optimism is indeed higher in high-underpricing periods is to look at price reactions to news. For example, price reactions surrounding earnings announcements. The motivation is that earnings surprises generate immediate reactions in prices. If in certain periods investors are more optimistic, then we should observe weaker negative reactions to negative surprises and stronger positive reactions to positive surprises. In order to test if price reactions to earnings announcements are more favorable in periods of high underpricing in the IPO market, I retrieve data from I/B/E/S, Compustat and CRSP from January 1, 1984 to December 31, To determine the surprise in earnings announcements I collect actual and forecasted earnings. I retrieve all quarterly earnings announcements from I/B/E/S for which there is at least one analyst forecast 90 days before the announcement. During the 90 days prior to the earnings announcements we may observe several updates on analyst forecasts. I only keep the forecast closest to earnings announcement yielding a sample of 307,442 observations. It is important that we have the correct earnings announcement dates. Hence, I also retrieve earnings announcement dates from Compustat. The sample is restricted to observations for which there are announcements dates on I/B/E/S and Compustat and di erence between the two dates is less than ve days. In order to impute a unique date to each earnings announcement I follow the optimal imputation rule used in 25

26 Dellavigna and Pollet (2009): when I/B/E/S and Compustat dates disagree I impute the earnings announcement date to be the earliest; for the period before January 1, 1990 if the two dates agree I impute the announcement date to be the previous trading day to the I/B/E/S and Compustat date; for the period after January 1, 1990 if the two dates agree I impute the announcement date to be the I/B/E/S and Compustat date; This yields a sample size of 179,590 observations. I de ne earnings surprise as the di erence between expected and actual earnings per share normalized by share price as in Kotari (2000). For expected earnings I use I/B/E/S consensus analyst forecast de ned as the median forecast among all analysts. To eliminate any noise in prices I use the price seven business days before the announcement. All observations for which this information is not available and all penny stocks (price below $1) are dropped. Sample is then reduced to 171,222 observations. Then I compute s t;k ; the earnings surprise for rm k in quarter t as: s t;k = e t;k p t;k be t;k where e t;k is the actual earnings per share of rm k in quarter t, be t;k the forecasted earnings per share and p t;k the corresponding price seven business days before the announcement date. I drop 124 observations for which the actual earnings are larger in absolute value than the price of the share. Given that market reaction to earnings surprises may be non-linear I sort them into quantiles as in DellaVigna and Pollet (2009). I create a total of 11 quantiles. Quantiles 1 to 5 concern negative surprises, quantile 6 has all announcements with no surprises while quantiles 7 to 11 have all positive earnings surprises. The breakpoints for the quantiles are computed year by year. Given that we have more positive surprises than negative surprises, we do not have the same number of observations per bin. The goal is to obtain abnormal returns for the day of the announcement and the next one. Thus, I match the sample with CRSP to obtain daily returns and I estimate betas for 26

27 each rm-quarter observation from the regression: R u ; k = t;k + t;k R u;m where R u ; k and R u;m are respectively the daily return for rm k and market at day u and u 2 [ 300; 45] with beeing the announcement date. Given b ; we can compute buy and hold abnormal return for the period [ + h; + H] for rm k, quarter t as R h;h t;k = +H j=+h (1 + R j;k ) 1 b h i t;k +H j=+h (1 + R j;m) 1 Finally, I obtain from Compustat market capitalization and book to market for each rm-quarter observation. The nal sample has 163,216 observations. I divide the sample in two subgroups: announcements during high-underpricing periods and announcements during low- or neutral- underpricing periods. Figure 3 depicts preliminary evidence that investors react more favorably to news during high-underpricing periods. For each quantile of earnings surprises, I compute average CAR 0;1 for the high- and non-high underpricing subsamples. As expected, for all quantiles, cumulative abnormal returns are higher in the high-underpricing sample. This is especially true for positive news. CAR 0;1 for surprises in bin eleven, announced in low-, and neutral-underpricing periods, is 2.15%. For the same level of surprises, announcements in high-underpricing periods generate a CAR 0;1 of 3.16% - one percentage point di erence in a two day window. To rigorously analyze the question I run the following regressions: (1) CAR it = + 1 High it + 1 Earn_Surp it + 2 High it Earn_Surp it + " it (2) CAR it = + 1 High it + 1 Earn_Surp it + 2 High it Earn_Surp it Controls it + 2 Controls it Earn_Surp it + " it (3) CAR it = + 1 High it + 1 BotQ it + 2 T opq it + 3 High it BotQ it + 4 High it T opq it Controls it + 1 Controls it BotQ it + 2 Controls it T opq it + " it (4) CAR it = + 1 High it + 1 Bot2Q it + 2 T op2q it + 3 High it Bot2Q it + 4 High it T op2q it Controls it + 1 Controls it Bot2Q it + 2 Controls it T op2q it + " it 27

28 where High it is a dummy variable equal to one when earnings are announced in a highunderpricing period in the IPO market; Earn_Surp it represent earnings surprises; Controls it include decile of market capitalization, decile of book to market, year and month of the announcement; BotQ=Bot2Q is a dummy variable equal to one if the surprise is in quantile 1/1&2 of earnings surprises; T opq=t op2q is a dummy variable equal to one if the earnings surprise is in quantile 11/10&11. Results in Table 21 con rm that investors react more favorably to earnings surprises in high-underpricing periods. In the four speci cations, we see that c 1 is positive and signi cant. Controlling for decile of market capitalization, decile of book to market, year and month of the announcement does not eliminate this e ect. Moreover, speci cations (3) and (4) show that this reaction is stronger for surprises in the two top bins. CARs are fty basis points higher for announcements in high-underpricing periods and an extra fty basis points for the most positive surprises. This is a big e ect and strong evidence that high-underpricing periods are correlated with high optimism. Results presented in this section suggest investor sentiment is particular strong in periods where the IPO market exhibits high underpricing. This is consistent with a story in which rms time their IPOs to periods when they can take advantage of these "exuberantly" optimistic investors. 5 Discussion about money left on the table Evidence presented in this paper suggests rms try to exploit sentiment investors in highunderpricing periods. IPO prices in the late stages of high-underpricing periods are set above fundamental value. However, we still observe high rst day returns for these rms going public. It is still a puzzle why rms do not fully exploit sentiment investors. In this section, I brie y discuss four potential explanations for this puzzle. The Ljungqvist, Nanda and Singh (2006) model has rms selling their shares to intermediaries, who then try to unload them in the aftermarket. Sentiment investors might show up in the market, driving prices up. When these investors are expected, rms extract rents from them by demanding higher o er prices from underwriters. However, the fact that rms exploit overly optimistic investors does not imply zero underpricing. High underpricing can 28

29 still occur as compensation given by rms to intermediaries for bearing the risk of carrying overvalued IPO shares in inventory. Although consistent with the results presented in this paper, the story has two drawbacks: within high-underpricing periods, the likelihood that sentiment investors show up in the aftermarket increases, which should lead to a decrease in the level of underpricing within these periods. The data does show this decline in underpricing. Moreover, the empirical magnitude of underpricing should provide strong incentives for rms to choose a di erent mechanism to go public such as auctions - we also do not observe it in the data. A second explanation is based on information asymmetry. Unlike the traditional literature on IPOs, the information asymmetry is not on fundamental value but on the maximum value that sentiment investors are willing to pay. In this sense, the high rst day return is a discount not on fundamental value, but on sentiment investors valuations. I examine analysts earnings forecasts to see if periods of high underpricing are associated with a high degree of information asymmetry. Table 22 shows that errors in earnings forecasts tend to be higher in high-underpricing periods than in low underpricing. Also, the number of analysts forecasts per rm tends to be smaller in high-underpricing periods. These two results are consistent with a higher degree of information asymmetry in high-underpricing periods. A third possible explanation for why rms do not fully exploit sentiment investors relies on a very particular and strong form of sentiment sentiment investors form their valuations by anchoring on IPO prices. Imagine again that rms sell shares to underwriters who then re-sell them in the aftermarket. Assume that when sentiment investors show up in aftermarket they always value shares at a premium over the IPO price. Under these assumptions, rms and underwriters cannot fully exploit sentiment investors. When sentiment investors are likely to show up, rms demand IPO prices at a premium to fundamental value, given the high probability that underwriters can re-sell at a higher price to sentiment investors. However, given the anchoring on IPO prices, there is no way for rms to fully extract rents from sentiment investors. Two problems of this explanation are the strong assumption on the nature of investors sentiment, and the strong incentive to perform the IPO through an auction. A fourth and perhaps the most reasonable explanation is that underwriters and rms share the rents from exploiting sentiment investors. Underwriters might be able to convince 29

30 rms to not fully exploit sentiment investors in return for better deals in the future. Firms are willing to go public as they are sold above fundamental value. Underwriters can then unload these shares, priced above fundamental value but below what sentiment investors are willing to pay, to favored clients in return for future participation in less attractive IPOs or quid pro quos 4. Determining the exact reason for why rms are willing to leave money on the table is beyond the scope of this paper. However, understanding that IPO prices include a premium helps to understand why rms decide to go public in high-underpricing periods. The level of underpricing remains a puzzle, but the puzzle of increased IPO volume can be explained as a reaction of rms to the presence of sentiment investors. 6 Conclusion This paper documents a close connection between the average underpricing in the market and the long-term underperformance of IPOs. I show that the average underpricing in the market determines the degree of subsequent long-run performance. The two central ndings of the paper are that the well documented long-term underperformance of IPOs is driven by rms going public in periods of high underpricing; and that underpricing is best understood not as a discount to fundamental value, but as a discount to in ated rst day prices. I provide evidence that di erences in long term underperformance cannot be explained by di erences in risk, and it is unlikely to be driven by a peso problem. I also nd that rms going public in later stages of high-underpricing periods display worse operating performance and pro tability, lower asset growth, lower investment rates and higher cash holdings. These results are consistent with a story in which some rms go public to take advantage of overvalued equity. 4 CSFB allocated shares of IPOs to more than 100 customers who, in return, funneled between 33 and 65 percent of their IPO pro ts to CSFB. These customers typically ipped the stock on the day of the IPO, often gaining tremendous pro ts. They then transferred a share of their ipping pro ts to CSFB by way of excessively high brokerage commissions... The customers paid these commissions on uneconomic, limited-risk trades in highly liquid, exchange-traded shares unrelated to the IPO shares trades that they e ected for the sole purpose of paying IPO ipping pro ts back to CSFB. (SEC News Release ). 30

31 Finally, I nd that investor sentiment is stronger in high-underpricing periods. The presence of exuberantly optimistic investors in the market drives prices up. Firms and underwriters exploit these optimistic investors, but the fact that underpricing still remains high implies that they do not fully adjust IPO prices. Why rms still leave money on the table is still an unresolved puzzle. However, results presented shed light on why the number of IPOs goes up when underpricing goes up. The increase in volume in high-underpricing periods is puzzling if higher underpricing represents a large discount to fundamental value, as suggested by the prior literature. However, the increase in number of IPOs in high-underpricing periods makes sense if high underpricing coincides with IPO prices that are set above fundamental value. References [1] Aggarwal, Reena, and Patrick Conroy, 2000, Price discovery in initial public o erings and the role of the lead underwriter, Journal of Finance 55, [2] Allen, Franklin, and Gerald R. Faulhaber, 1989, Signaling by underpricing in the IPO market, Journal of Financial Economics 23, [3] Alti, Aydogan, 2005, IPO market timing, Review of Financial Studies 18, [4] Baron, David, 1982, A model of the demand for investment banking advising and distribution services for new issues, Journal of Finance 37, [5] Beatty, Randolph, and Jay Ritter, Investment banking, reputation, and the underpricing of initial public o erings, Journal of Financial Economics 15, [6] Benveniste, Lawrence M., and Paul A. Spindt, 1989, How investment bankers determine the o er price and allocation of new issues, Journal of Financial Economics 24, [7] Benveniste, Lawrence M. and William J. Wilhelm, 1990, A comparative analysis of IPO proceeds under alternative regulatory environments, Journal of Financial Economics 28, [8] Bernardo, Antonio E., and Ivo Welch, 2001, On the evolution of overcon dence and entrepreneurs, Journal of Economics and Management Strategy 10,

32 [9] Booth, James R., and Lena Chua, 1996, Ownership dispersion, costly information, and IPO underpricing, Journal of Financial Economics 41, [10] Booth, James R. and Richrad L. Smith, 1986, Capital raising, underwriting, and the certi cation hypothesis, Journal of Financial Economics 15, [11] Brav, Alon, and Paul A. Gompers, 1997, Myth or reality? The long-run underperformance of initial public o erings: Evidence from venture and nonventure capital-backed companies, Journal of Finance 52, [12] Brennan, Michael J., and J. Franks, 1997, Underpricing, ownership and control in initial public o erings of equity securities in the UK, Journal of Financial Economics 45, [13] Chemmanur, Thomas J., 1993, The pricing of initial public o ers: A dynamic model with information production, Journal of Finance 48, [14] Choe, H., R. Masulis, and V. Nanda, 1993, Common Stock O erings across the Business Cycle, Journal of Empirical Finance, 1, [15] Daniel, Kent, David Hirshleifer, and A. Subrahmanyam, 1998, Investor psychology and security market under- and over-reactions, Journal of Finance 53, [16] Dellavigna, Stefano and Joshua M. Pollet, 2009, Investor Inattention and Friday Earnings Announcements, Journal of Finance 64, [17] Ellis, Katrina, Roni Michaely, and Maureen O Hara, 2000, When the underwriter is the market maker: An examination of trading in the IPO aftermarket, Journal of Finance 55, [18] Field, Laura C., and Dennis P. Sheehan, 2001, Underpricing in IPOs: Control, monitoring, or liquidity? Working paper, Pennsylvania State University. [19] Hanley, Kathleen Weiss, 1993, The underpricing of initial public o erings and the partial adjustment phenomenon, Journal of Financial Economics 34, [20] Heaton, J. B., 2002, Managerial optimism and corporate nance, Financial Management. [21] Ho man-burchardi, 2001, Clustering of Initial Public O erings, Information Revelation and Underpricing, European Economic Review, 45,

33 [22] Hughes, Patricia J., and Anjan V. Thakor, 1992, Litigation risk, intermediation, and the underpricing of initial public o erings, Review of Financial Studies 5, [23] Ibbotson, R. G., 1975, Price performance of common stock new issues, Journal of Financial Economics 2, [24] Ibbotson, Roger G., Jody L. Sindelar, and Jay R. Ritter, 1988, Initial public o erings, Journal of Applied Corporate Finance 1, [25] Ibbotson, R. J. Sindelar, and J. Ritter, 1994, The Market s Problem with the Pricing of Initial Public O erings, Journal of Applied Corporate Finance, 7, [26] Jain, B. and O. Kini, 1994, The Post-Issue Operating Performance of IPO Firms, Journal of Finance, 49, [27] Jegadeesh, Narasimhan, Mark Weinstein, and Ivo Welch, 1993, An empirical investigation of IPO returns and subsequent equity o erings, Journal of Financial Economics 34, [28] Korajczyk, R., D. Lucas, and R. McDonald, 1991, The E ect of Information Releases on the Pricing and Timing of Security Issues, Review of Financial Studies, 4, [29] Lie, Erik, 2001, Detecting Abnormal Operating Performance: Revisited, Financial Management, [30] Ljungqvist, A., V. Nanda, and R. Singh, 2006, Hot Markets, Investor Sentiment, and IPO Pricing, Journal of Business 79, [31] Loughran, Tim, and Jay R. Ritter, 2002, Why don t issuers get upset about leaving money on the table in IPOs? Review of Financial Studies 15, [32] Lowry, Michelle, 2003, Why does IPO volume uctuate so much? Journal of Financial Economics 67, [33] Lowry, Michelle, and G. William Schwert, 2002, IPO market cycles: Bubbles or sequential learning? Journal of Finance 57, 3, [34] Mello, Antonio S., and John E. Parsons, 1998, Going public and the ownership structure of the rm. Journal of Financial Economics 49,

34 [35] Michaely, R. and W. Shaw, 1994, The Pricing of Initial Public O erings: Tests of Adverse- Selection and Signaling Theories, Review of Financial Studies, 7, [36] Mikkelson and Shah, 1997, Performance of Companies Around Initial Public O erings, Journal of Financial Economics, 44, [37] Miller, Edward M., 1977, Risk, uncertainty, and divergence of opinion, Journal of Finance 32, [38] Muscarella, Chris J. and Michael R. Vetsuypens, 1989, A simple test of Baron s model of IPO underpricing, Journal of Financial Economics 24, [39] Pagano, Marco, Fabio Panetta, and Luigi Zingales, 1998, Why do companies go public? An empirical analysis, Journal of Finance 53, [40] Purnanandam, Amiyatosh K., and B. Swaminathan, 2001, Are IPOs underpriced? Working paper, Cornell University. [41] Rajan, R., and H. Servaes, 1997, Analyst Following of Initial Public O erings, Journal of Finance, 52, [42] Ritter, Jay R., 1984, The "hot issue" market of 1980, Journal of Business 57, [43] Ritter, Jay R., 1991, The long run performance of initial public o erings, Journal of Finance 46, [44] Rock, Kevin, 1986, Why new issues are underpriced, Journal of Financial Economics 15, [45] Sherman, Ann, 2000, IPOs and long-term relationships: An advantage of bookbuilding, Review of Financial Studies 13, [46] Sherman, Ann, and Sheridan Titman, 2002, Building the IPO order book: Underpricing and participation limits with costly information, Journal of Financial Economics. [47] Smith, Cli ord W., 1986, Investment banking and the capital acquisition process, Journal of Financial Economics 15, [48] Spatt, Chester S., and Sanjay Srivastava, 1991, Preplay communication, participation restrictions, and e ciency in initial public o erings, Review of Financial Studies 4,

35 [49] Stoughton, Neal M., and Josef Zechner, 1998, IPO-mechanisms, monitoring and ownership structure, Journal of Financial Economics 49, [50] Stoughton, Neal M., Kit Pong Wong, and Josef Zechner, 2001, IPOs and product quality, Journal of Business 74, [51] Taylor, Shelley E., and Jonathan D. Brown, 1988, Illusion and well-being: A social psychological perspective on mental health, Psychological Bulletin 103, [52] Tinic, Seha M., 1988, Anatomy of initial public o erings of common stock, Journal of Finance 43, [53] Teoh, S., I. Welch, and T. Wong, 1998, Earnings Management and the Long-Run Market Performance of Initial Public O erings, Journal of Finance, 53, [54] van Bommel, Jos, and Theo Vermaelen, 2001, Post-IPO capital expenditures and market feedback, Journal of Banking and Finance. [55] Welch, Ivo, 1992, Sequential sales, learning, and cascades, Journal of Finance 47, [56] Welch, Ivo, 1989, Seasoned O erings, Imitation Costs, and the Underpricing of Initial Public O erings, Journal of Finance, 44, [57] Welch, Ivo and Ritter, Jay, 2002, A Review of IPO Activity, Pricing and Allocations, Journal of Finance 57,

36 Figure 1. Number of IPOs and monthly average rst day return of U.S. IPOs from 1973 to The sample is composed of U.S. IPOs completed between January 1, 1973 and December 31, 2008 as reported by Securities Data Company (SDC). Unit o ers, spin-o s, closed-end funds, IPOs with an o er price below $1.00, Real Estate Investments Trusts (REITS) and nancial rms as described by SDC are excluded from our sample. 36

37 Figure 2. IPO level of underpricing and volume from 1973 to The sample is composed of U.S. IPOs completed between January 1, 1973 and December 31, 2008 as reported by Securities Data Company (SDC). Unit o ers, spin-o s, closed-end funds, IPOs with an o er price below $1.00, Real Estate Investments Trusts (REITS) and nancial rms as described by SDC are excluded from our sample. High underpricing/volume periods consist of three or more consecutive high underpricing/volume months.low underpricing/volume periods are de ned analogously and all other periods still not de ned are considered neutral underpricing/volume periods. 37