Strategic Underwriting in Initial Public Offers

Strategic Underwriting in Initial Public Offers Gerard Hoberg∗(mailto: gerard.hoberg@yale.edu) October 28, 2003 Yale University School of Management 135 Prospect Street, Box 208200 New Haven, CT 06520-8200 Abstract This study presents a model of strategic underwriting in which underwriters compete for and price IPOs. The model explains three IPO pricing anomalies: (1) IPO underpricing, (2) the partial adjustment phenomenon, and (3) the tendency of some underwriters to persistently underprice more than others (holding size and industry constant). New empirical evidence documenting underwriter specific persistence in residual initial returns supports the model. I also present a new measure of underwriter reputation based on past initial returns that, unlike existing measures based on underwriter size or tombstone placements, is among the most significant predictors of future initial returns. Large underwriters with high rankings later experience superior market share growth. Gerard Hoberg is from the Yale School of Management. I wish to thank Ivo Welch, Matt Spiegel, Florencio de Silanes-Lopez and Lily Qiu for excellent comments and advice. All errors are the author’s alone. ∗ 1 In a sample of US initial public offerings from 1984 to 2000, (1) the typical IPO experienced initial returns of 22.7%; (2) IPOs priced below, within and above the initial filing range experienced initial returns of 3.4%, 12.1% and 54.0% respectively; And (3) underwriters who were in the highest quartile of past underpricing brought to market future IPOs with size and industry adjusted initial returns that were 14.7% larger than underwriters in the lowest quartile. Though roughly half as large if the late 1990s are excluded, all three phenomena are statistically and economically significant throughout the 1980s and the 1990s. The former two results, first documented in Ibbotson (1975) and Hanley (1993), respectively, are referred to as “IPO underpricing” and the “partial adjustment phenomenon”. The latter result, the “underwriter persistence phenomenon”, is first documented in this study. The underwriter persistence phenomenon is unique because the choice of lead underwriter is known early in the IPO process. Thus, in the absence of barriers to IPO investing, this result would suggest that IPO investors can earn 14.7% higher one day returns simply by purchasing shares from an underwriter with high past initial returns. In contrast, agents cannot condition their participation on the partial adjustment phenomenon because the final IPO price is not known until the IPO date. Large profits from underwriter persistence are thus a mystery. Consistent with Fulghieri and Spiegel (1992) and Loughran and Ritter (2002), the existence of these large profits suggests they may not be free, and they may only be available to investors who return a large portion of them to their underwriter via quid pro quo arrangements. I present a rational model of strategic underwriting, which assumes that underwriters maximize profits extracted via quid pro quo arrangements. The model can explain all three IPO pricing phenomena and suggests that (1) IPO underpricing arises from imperfect underwriter competition;1 (2) the partial adjustment phenomenon arises because underwriters, who face an uninformed government regulator, adjust the IPO price less when they can extract more of the issuing firm’s value; and (3) the underwriter persistence phenomenon arises from adverse selection, where informed underwriters are better at identifying, and profiting from, undervalued issuers. Economically, the partial adjustment phenomenon is perhaps the largest IPO pricing anomaly, yet few existing studies explain how it might arise. Book building models based on Benveniste and Spindt (1989) suggest that partial adjustment compensates investors for revealing private information. Though popular among academics, this theory faces some criticism. Maksimovik and Pichler (2002) show that informed investors do not require compensation when the underwriter has no share allocation constraints, which is Though the US IPO market consists of over 50 active underwriters, Arkebauer (1991) explains that only 2-4 typically compete for any one IPO and thus the IPO market may indeed be concentrated. 1 2 likely given that IPOs are oversubscribed.2 Though Sherman and Titman (2002) show that this problem can be addressed by assuming that information is costly, Ritter and Welch (2002) suggest that IPO underpricing is too large to be explained by theories of bookbuilding. I present a rational agency explanation in which informed underwriters partially adjust when they can extract more of the issuing firm’s value. In contrast to my study and bookbuilding models, which explain partial adjustment to private information unknown by the issuer, 2 studies explain the partial adjustment to public information such as market returns. In an agency model based on prospect theory, Loughran and Ritter (2002) suggest that issuers bargain less aggressively for a higher IPO price when faced with good news. Edelen and Kadlec (2003) suggest that issuers rationally trade issue proceeds for greater IPO success probability when faced with good news. My study can be viewed as a complement to Loughran and Ritter (2002), and the combined results can explain the partial adjustment to both public and private information using related agency theory.3 Because it is first documented here, existing studies do not explain how the underwriter persistence phenomenon may arise. To fill this void, I test 5 alternative hypotheses in addition to my strategic underwriting model. (1) By controlling for the Carter/Manaster rank (see Carter and Manaster (1990)) and underwriter market share (see Megginson and Weiss (1993)), I find that existing measures of underwriter quality cannot explain underwriter persistence. (2) By adding controls for industry fixed effects, I find that industry-specialization cannot explain underwriter persistence. (3) By adding controls for market returns and the initial returns of recent IPOs, I find that short-term “hot market” effects cannot explain underwriter persistence. (4) Habitual (behavioral) underpricing by some underwriters may be the most viable alternative. However, I find that large underwriters with high past initial returns experience growing (not declining) market share, which is not consistent with their setting the least efficient prices. (5) Because firms underwritten by underwriters with high past initial returns visit the reissue market less (not more) frequently, underwriter persistence likely does not compensate underwriters for providing better service. In contrast to these 5 alternatives, my strategic underwriting model is consistent with these results. Moreover, the superior market share growth of large underwriters with high past initial returns supports the model’s prediction that these underwriters are better informed and more profitable. Numerous existing studies explain how IPO underpricing may arise. Signaling models following Welch (1989), Allen and Faulhaber (1989), and others propose that underpricSee Amihud, Hauser, and Kirsh (2001), who document IPO oversubscription, for example. Both studies make the common assumption that underwriters can profit by extracting IPO returns from investors using quid pro quo arrangements. This poses an agency problem because underwriters can only extract a fraction of these returns, making them an ineffcient form of compensation. 3 2 3 ing is a costly signal of issuer quality. However, empirical support for signaling models is modest at best. Book building models following Benveniste and Spindt (1989) propose that underpricing compensates informed investors for revealing information. However, as noted, investors require little or no compensation and bookbuilding models cannot explain the larger price anomalies of the late 1990s. Rock’s (1986) winner’s curse receives some support from Michaely and Shaw (1994) and others, who compare underpricing across markets. However, Rock’s model also cannot explain results from the late 1990s. Other hypotheses such as protection against negative cascades as in Welch (1992) are difficult to test. Because viable explanations for IPO underpricing are numerous, a thorough empirical test of each is outside the scope of my study. Instead, I show that imperfect underwriter competition can also explain why IPOs are underpriced. The model is perhaps most similar to Baron (1982), who includes an entrepreneur and a single underwriter. However, my model is unique because it (1) explains 2 additional IPO pricing anomalies; (2) includes N underwriters and shows that underwriter concentration may determine the level of expected underpricing; and (3) shows that higher firm-level price uncertainty can explain why pricing anomalies were larger in the late 1990s. In the empirical section of this study, I create a new underwriter reputation measure, “UWpastIR”, which is based on each underwriter’s past initial returns. I summarize the empirical results as follows: • UWpastIR is the single most important predictor of future initial returns known on the initial filing date. This result is robust to adding controls for existing reputation measures, issuer size, underwriter size and industry fixed effects. • Among the larger set of variables known on the IPO date, UWpastIR (1) still predicts initial returns and (2) is partially subsumed by variables related to the partial adjustment phenomenon. • Large underwriters with higher UWpastIR experience growing market share. Small underwriters with higher UWpastIR experience declining market share. • Firms underwritten by high UWpastIR underwriters visit the reissue market less frequently (ie: they issue fewer post-IPO debt and equity issues). • Underwriter reputation is two dimensional: (1) the Carter-Manaster rank and underwriter market share both measure the underwriter’s ability to attract large issuers. (2) UwpastIR measures his ability to attract issuers who are initially undervalued and require upward price adjustment. • 2 underwriter concentration proxies predict higher initial returns in cross section. 4 This study makes 3 main contributions. First, I present a theoretical model of strategic underwriting that can explain 3 IPO pricing phenomena. Second, I present a new empirical measure of underwriter reputation based on past IPO prices that is distinct from existing measures. Third, it is the first study to document that some underwriters experience initial returns that are higher than others. The differences are economically large, persist over long periods of time, and are robust to numerous controls. The remainder of this paper is organized as follows. Section I presents a strategic underwriting model where two underwriters compete as a duopoly. Section II generalizes the model and assumes N underwriters compete as an oligopoly. Section III presents comparative statics and numerical examples. Section IV describes the variables and the data used in the empirical section. Section V presents empirical evidence relating to the predictability of initial returns. Section VI examines whether reputable underwriters experience superior market share growth. Section VII studies the relationship between underwriter reputation and the frequency of visits to the reissue market (post-IPO debt and equity issues). Section VIII studies the relationship between initial returns and underwriter concentration and section IX concludes. I A Theory of Duopoly Underwriting In a two period game, I model a duopoly of 2 underwriters competing for the business of a single entrepreneur issuing an initial public offering. The 2 underwriters are type H (partially informed) and L (uninformed) regarding the value of the entrepreneur’s firm. The game models the firm commitment IPO mechanism, in which the entrepreneur and the underwriters initially agree to issue the IPO without knowing the final IPO price.4 A key innovation of this paper is to recognize that the IPO price can be written as a sum of two sequential parts: the initial filing midpoint and the final IPO date price adjustment.5 I assume that each is an endogenous result of two sequential pricing games. In period 1, there is a “courting game” where underwriters compete for the entrepreneur’s business. Period 1 ends on the IPO’s initial filing date. The IPO’s initial filing midpoint, size, and allocation of shares among underwriters are determined In a firm commitment IPO, the entrepreneur and the underwriters agree upon terms in a “letter of intent” (see Arkebauer (1991)). This document is drafted when the lead underwriter is chosen, which occurs several months before the IPO date (when the IPO price is determined). 5 2 to 3 months prior to the IPO date, issuers file a preliminary prospectus with the SEC that includes an initial price estimate. On the IPO date, underwriters adjust the price after collecting information from investors. The actual IPO price is equal to the intial estimate plus the price adjustment. 4 5 endogenously.6 In period 2, there is a “price adjustment game” in which underwriters are regulated by an uninformed government. Period 2 ends on the IPO date. The IPO’s final price adjustment and government approval are determined endogenously. A The Entrepreneur The risk neutral entrepreneur (E) has access to convex production technology that produces shares of his new firm. Where QE is the total number of shares produced, his production technology is described by the following convex cost function: C[QE ] ≡ RQE + 1 2 Q . 2S E (1) The parameter R is the entrepreneur’s reservation value, the minimum share price needed to profitably produce a positive number of shares.7 Because a larger S allows the entrepreneur to produce more shares at a lower marginal cost, S is the effective size of the entrepreneur’s firm. The entrepreneur does not have access to IPO investors and can only sell his newly issued shares through an underwriter. Thus, if the entrepreneur issues an IPO with QE shares at an IPO price Pipo , his profits are πE = QE Pipo − RQE − 1 2 Q . 2S E (2) The value of the entrepreneur’s firm will evolve as three prices: Pmid , Pipo , and Pmkt . The initial filing midpoint (Pmid ) is an endogenous result in period 1 and is an uninformed expectation of the firm’s value. The IPO price (Pipo ) is an endogenous result in period 2. It is the outcome of a strategic process involving the underwriters and the uninformed government. The after-market trading price (Pmkt ) is a random variable that is realized at the end of period 2 and is the true value of the firm. Thus, period 1 and period 2 end on the IPO’s filing date and IPO date respectively. If ∆P is the period 2 “price adjustment”, Pipo can be written: Pipo = Pmid + ∆P . 6 (3) Issuers may, in practice, adjust the size of the IPO after the initial filing date. The main results of the model still obtain if this late size adjustment is permitted. The assumption that size is finalized in period 1 is thus a simplifying one. 7 This can be seen by observing that the initial marginal cost of production is equal to R: C’[0]=R. 6 Consistent with the firm commitment IPO mechanism, the entreprenuer must commit to the IPO in period 1 without knowing the final sale price Pipo .8 Instead, the risk neutral entrepreneur will base his production decision on the expected IPO price given the period 1 price estimate: E[Pipo | Pmid ]. Equations (2) and (3) can be used to state the entrepreneur’s objective as: MAX QE (Pmid + E[∆P | Pmid ]) − (RQE + QE 1 2 Q ). 2S E (4) The unique solution to the quadratic maximization gives the entrepreneur’s supply curve: QE [Pmid ] = (Pmid + E[∆P | Pmid ] − R)S. (5) The supply curve confirms that the parameter R can be interpreted as the entrepreneur’s reservation price, the expected sale price at which he would issue zero shares. Similarly, it confirms that S is a scaling factor for the size of his firm. Pmkt , the firm’s true value, is Pmkt = V + ˜ + ˜ , n p (6) where V is known by all agents in period 1 and is the expected value of the firm based on public information (all information contained in the preliminary prospectus). The random variables ˜ and ˜ are independent and take the values {−σ,0,σ} with equal p n probability. σ is a parameter indicating the magnitude of firm-level price uncertainty. ˜ p is learned by all agents at the start of period 2. It summarizes public information learned between the filing date (period 1) and the IPO date (period 2). For example, ˜ includes p market returns, industry returns, and the firm’s continued reporting as permitted by regulations.9 ˜ summarizes all private information that is learned by the underwriters n (see section B) at the start of period 2, but is unknown by the entrepreneur.10 ˜ may, n for example, include the firm’s true growth rate, the firm’s potential to expand into new markets, or the firm’s likelihood of being acquired. In the US IPO market, the entrepreneur does not actually make a legal commitment on the filing date. However, they do “effectively commit” because the IPO process is costly to abandon or restart. 9 The SEC, for example, does not permit the issuing firm to make new announcements concerning the company’s future that are not stated in the prospectus during the waiting period. However, the issuer may discuss factual information concerning the firm’s present activities and the firm may continue to report quarterly earnings. 10 See Benveniste and Spindt (1989) for example, who suggest that underwriters become highly informed just prior to the IPO date by gathering information from informed IPO investors. Underwriters are thus likely to be more informed on the IPO date than both the entrepreneur and the government. 8 7 B The Underwriters Two risk neutral underwriters, type H (partially informed) and type L (uninformed), compete for shares of the new firm. At the start of period 1, H receives the partially ˜ ˜ informative signal ηH regarding the value of ˜ . Like ˜ , ηH can take on the values n n ˜ ˜ {−σ,0,σ}, and with probability λ, where 0 λ 1, ηH is equal to ˜ . Otherwise, ηH n 1 1 1 11 is independently drawn from {−σ,0,σ} with probability { 3 , 3 , 3 }. Values of λ close to ˜ one (zero) thus imply that ηH is more (less) informative as follows: ˜ ˜ E[ ˜ | ηH ] = ληH . n I make the following assumption about underwriting profit: A1 Underwriting profit is equal to an extraction rate “γ,” times IPO underpricing: Pmkt − Pipo . When underwriting QH and QL shares, H and L thus realize profits of πH = γQH (Pmkt − Pipo ) and πL = γQH (Pmkt − Pipo ) respectively.12 A1 reflects a growing belief among researchers that underwriters profit by extracting investment returns from investors through quid pro quo arrangements.13 A1 also explains why I do not formally include IPO investors in this model: their participation is guaranteed by the remaining fraction (1-γ) of the profits from expected IPO underpricing that the underwriters are unable to extract. The presence of positive investor profits also informally predicts, consistent with observation,14 that IPOs will be oversubscribed. Equations (3) and (6) can be used to restate underwriter profits as: πH = γQH (V + ˜ + ˜ − Pmid − ∆P ), n p πL = γQL (V + ˜ + ˜ − Pmid − ∆P ). (8) n p (7) Figure 1 displays the timeline of the strategic underwriting game. In period 1, H and L fully learn the entrepreneur’s supply curve (see equation 5) and engage in Cournot competition, where they simultaneously select QH and QL , the number of shares each will underwrite respectively.15 The underwriters make up a duopoly with a total demand of It does not matter whether H knows if a given realization of η˜ is actually equal to ˜ or not. H n An earlier version of the model added a fixed commission rate times gross proceeds to the profit function. This feature was omitted because the model generated similar predictions as this one with considerable added complexity. 13 Fulghieri and Spiegel (1993) show that underwriters can use share allocations to induce quid pro quo transfers from their investors. Loughran and Ritter (2002) also make the assumption that underwriters can profit from extracted share underpricing. Aggarwal, Prabhala and Puri (2002) show that underwriters indeed favor their institutional clientele when shares are more deeply underpriced reflecting quid pro quo arrangements. 14 See Amihud, Hauser, and Kirsh (2001) for example. 15 Though underwriters do not actually compete on the basis of quantity, I assume Cournot competition because it (1) permits underwriters to earn positive profits and (2) is consistent with the fact that underwriters do not compete on the basis of price alone (see Arkebauer (1991)). 12 11 8 QH + QL shares.16 At the end of period 1, total demand is equated to the entrepreneur’s supply curve from equation (5) to determine the period one price estimate Pmid :17 QH + QL = QE [Pmid ] ⇒ Pmid = R + QH + QL ˜ − E[∆P | ηH ]. S (9) Like the entrepreneur, H and L must commit to the IPO without knowing the final IPO price Pipo . Instead, the risk neutral underwriters will maximize expected profits ˜ E[πH | ηH ] and E[πL ] respectively. Equations (8) and (9) can be used to write H and L’s period 1 objectives as:18 ˜ ˜ MAX γQH (V + E[ ˜ | ηH ] + E[ ˜ | ηH ] − R − n p QH QH +QL ), S MAX γQL (V + E[ ˜ ] + E[ ˜ ] − R − n p QL QL +E[QH ] ). S (10) Equation (7), and the properties of the random variables ( ˜ , ˜ , ηH ), can be used to n p ˜ restate H and L’s objectives: MAX γQH (V − R − QH QH +QL S ˜ + ληH ), (11) MAX γQL (V − R − QL QL +E[QH ] ). S In period 1, underwriters balance the profit gains from underwriting more shares against the resulting profit losses associated with a higher expected IPO price. Because: (1) Pmid is determined after H and L select QH and QL ; and (2) H and L act simultaneously, L ˜ cannot infer the value of H’s signal ηH before choosing QL . Thus, QH is stochastic from L’s perspective and L’s strategy is a function of its expected value E[QH ]. At the start of period 2, all agents learn the value of ˜p , but only the underwriters (H and L) learn the value of ˜n (see Figure 1).19 This reflects the fact that underwriters engage in book building and are likely to be more informed than both the government and the entrepreneur on the IPO date. As a result, H only has an advantage over L in period 1 and both underwriters are equally and fully informed in period 2. The duopoly can also be viewed as an underwriting syndicate, where underwriters pool their resources to sell an issue. In the US, the majority of medium and large IPOs are syndicated. 17 I substitute E[∆P | Pmid ] with E[∆P | η˜ ] in the derivation of equation (9) because Pmid fully H reveals η˜ to the entrepreneur in equilibrium. H 18 Note that I assume, see Figure 1, that the underwriters do not yet know Pmid at the time QH and QL are chosen. 19 In the general case, when λ < 1, even H must learn the value of ˜n because H’s signal η˜ is only H partially informative on average. 16 9 After observing ˜ and ˜ , H and L choose a price adjustment ∆P . The chosen ∆P n p must then comply with binding government regulations. Because QH , QL , Pmid , and Pmkt are known to H and L in period 2, they will select ∆P to maximize the known profit functions πH and πL in equation (8). It follows that the period 2 underwriters will optimally choose the most negative ∆P allowed by regulation (see section C).20 C The Government In period 2, an uninformed government regulates the underwriter’s price adjustment ∆P . Consistent with a primary objective of the Securities Exchange Commission (SEC), I assume that the government requires ∆P to be “transparent”. Specifically, ∆P must only reflect new information made available to the public as follows. R1 Because ˜ is public information, the price adjustment ∆P = ˜ is immediately p p transparent. R2 Because ˜n is non-public information, a price adjustment of of the form ∆P = ˜ +α ˜ , where 0 < α 1, is initially non-transparent. However, the underwriters p n can make such a ∆P transparent by distributing a revised prospectus that reveals the fraction α of ˜ to the public. n R3 Any price adjustment not satisfying R1 or R2 is initially non-transparent. Such a ∆P also cannot be made transparent because the underwriters, who only know ˜ and ˜ , cannot document its validity by revising the prospectus. Any ∆P not p n satisfying R1 or R2 will thus result in a failed IPO. Non-transparent price adjustments do not comply with regulations. By R1 and R2, the set of regulation compliant choices for ∆P thus includes all prices in the closed interval [ ˜ , ˜ + ˜ ] if ˜ 0, or the closed interval [ ˜ + ˜ , ˜ ] if ˜ < 0. Because the p p n n p n p n underwriters prefer to set ∆P as negative as possible (see section B), the underwriters’ optimal price adjustment is ∆P ∗ = ˜ + min[0, ˜ ]. p n (12) The government’s regulation prevents the underwriters from setting ∆P arbitrarily low. However, the government’s uninformed status allows the underwriters to act strategically. For example, when ˜ = 1, the underwriters will conceal their information to n avoid an upward price adjustment. Similarly, when ˜ = −1, the underwriters will fully n reveal ˜ in order to adjust the price down as much as possible. In equilibrium, this n behavior will explain how the partial adjustment phenomenon arises in the model. In equilibrium, H and L will choose the same ∆P because they have the same period 2 objective, which is to select the most negative ∆P allowed by regulation. Thus, all shares are sold at the same IPO price and I will not place a subscript on ∆P . 20 10 Observed IPO pricing patterns confirm that government regulation, not market forces, may indeed place binding restrictions on ∆P . For example, a theory based on market forces would predict that underwriters face declining investor demand when setting ∆P arbitrarily high. Yet empirical results show that IPOs are typically welloversubscribed,21 so this constraint is generally not binding. Similarly, the underwriters face the issuer’s right to withdraw if they set ∆P arbitrarily low. Yet few issuers withdraw on the IPO date and issuers are generally satisfied with their underwriter (even when the shares are underpriced)22 , so this constraint is also not binding. In contrast, Hanley (1993, pg 239) shows that the likelihood of IPO delay due to prospectus revision (a regulatory matter), is negatively related to the price adjustment ∆P .23 This evidence suggests that government regulation over the process of price revelation and price adjustment, not market forces, may indeed constrain ∆P . D Equilibrium Equilibrium in the two period game is described by the set of optimal underwriting decisions {QH ∗ , QL ∗ , ∆P ∗ } and the entrepreneur’s optimal production QE ∗ . I solve for equilibrium using backward induction: in period 1, agents assume that the underwriters will select the optimal ∆P in period 2. I prove the following theorem in the appendix: Theorem 1 Consider the duopoly underwriting game defined by the exogenous parameters {V, λ, σ, γ, R, S}. If (1) all parameters are non-negative, (2) V > R and (3) 0 λ 1, then the game has the following unique Nash equilibrium: ∆P ∗ = ˜p + min[0, ˜n ] QL ∗ = S(V − R) , 3 QE ∗ = QH ∗ + QL ∗ S(V − R) Sλ˜ H η + 3 2 2S(V − R)) Sλ˜ H η = + 3 2 QH ∗ = (13) The total size of the IPO, QE ∗ , increases linearly with S, the size parameter. The IPO is also larger when the reservation value R is smaller. This reflects scenarios in which the IPO adds more value and the firm is worth more publicly traded than privately held. Because L is uninformed in period one, the equilibrium strategy QL ∗ is deterministic. ˜ In contrast, QH ∗ depends on the signal ηH . When λ is zero, H does not have an advantage 21 22 See Amihud, Hauser, and Kirsh (2001) for example. See Krigman, Shaw and Womack (2001) for example. 23 My model predicts this negative relationship between price adjustment and prospectus revision. 11 ˜ ˜ and QH ∗ will equal QL ∗ . Otherwise, QH ∗ is larger when ηH is positive. H finds a large ηH to be favorable because the uninformed government cannot prevent him from extracting more of the firm’s larger than expected value through underpricing. II A Theory of Oligopoly Underwriting In this section, I extend the duopoly model and assume that N underwriters compete for the entrepreneur’s shares as a Cournot oligopoly. The extended model will demonstrate that the level of underwriter concentration determines the level of underpricing in the economy and the size of the information rents earned by H underwriters. A The Oligopoly Model The timeline of the oligopoly model is identical to that of the duopoly model and is presented in figure 1. In the extended model, N underwriters compete for the single entrepreneur’s business where N = NH + NL . NH of the N underwriters are type H (partially informed) and NL are type L (uninformed). ˜ At the start of period 1, all NH of the H underwriters receive identical signals ηH regarding the value of ˜n . Each H and L underwriter then selects QH and QL , the number of shares each will underwrite respectively. I only consider symmetric equilibria, where (1) all NH of the H underwriters select identical quantities QH and (2) all NL of the L underwriters select identical quantities QL . The underwriters thus form an oligopoly with a total demand of NH QH + NL QL shares. As in the duopoly model, total demand is equated to the entrepreneur’s supply curve in equation (5) to determine the period 1 filing date midpoint Pmid : QH + QL = QE [Pmid ] ⇒ Pmid = R + NH QH + NL QL ˜ − E[∆P | ηH ]. S (14) Equations (8), (7), and (14) can be used to write H and L’ period 1 objectives as MAX γQH · (V − R − QH QH +QH− +NL QL S ˜ + ληH ), (15) MAX γQL · (V − R − QL QL +QL− +NH E[QH ] ). S Each objective describes the problem faced by a single H or L underwriter. The quantity QH− is the total quantity underwritten by the given H underwriter’s (NH − 1) type H- 12 rivals.24 QL− is similarly defined. As in the duopoly model, QH , QL , Pmid , and Pmkt are known to all underwriters in period 2 so they will select ∆P to maximize the (known) profit functions πH and πL in equation (8). The period 2 underwriting objective is thus to select the most negative ∆P allowed by regulation. The following theorem is proven in the appendix: Theorem 2 Consider the oligopoly underwriting game defined by the exogenous parameters {V, NH , NL , λ, σ, γ, R, S}. If (1) all parameters are non-negative, (2) V > R and (3) 0 λ 1, (4) NH + NL 1, then the game has the following unique symmetric Nash equilibrium: ∆P ∗ = ˜p + min[0, ˜n ] QL ∗ = S(V − R) , 1 + NH + NL QH ∗ = S(V − R) Sλ˜ H η + 1 + NH + NL 1 + NH (16) QE ∗ = NH QH ∗ + NL QL ∗ = η S(NH + NL )(V − R) Sλ˜ H NH + 1 + NH + NL 1 + NH Consistent with the duopoly model, the total size of the IPO QE ∗ is increasing in S and decreasing in R. The total size also increases when more underwriters compete (larger NH or NL ). This result obtains because members of a larger oligopoly will price more aggressively and the entrepreneur will be induced to issue more shares. In contrast, QH ∗ and QL ∗ both decrease when more underwriters compete because a larger oligopoly implies greater competition from rivals. III A Comparative Statics and Numerical Examples Comparative Statics This section presents comparative statics for the strategic underwriting model. Throughout this section, I assume that the 4 regularity conditions listed in Theorem 2 hold. In the appendix, I provide an overview of how the expectations in this section are computed and I prove implication 3 as an example. Implication 1: If V > R, then the model predicts that IPOs will be underpriced on average. V −R (17) E[Pmkt − Pipo ] = 1 + NH + NI 24 H (L) underwriters hold QH− (QL− ) fixed when solving for their optimal QH (QL ), see appendix. 13 Because the expectation is positive when V > R and inversely related to the number of competing underwriters (NH + NL ), the model can explain IPO underpricing and further suggests that it arises from oligopolistic/imperfect competition. Underpricing is also proportional to (V − R), the IPO’s “value added”, which is the firm’s expected value as a publicly traded firm less its reservation value as a private firm. Implication 2: If σ > 0, the model predicts the partial adjustment phenomenon. +(1−λ E[Pmkt − Pipo | ∆P > 0] = E[Pmkt − Pipo ] + σ 1+2λ6(1+NH ) )NH 2 2 E[Pmkt − Pipo | ∆P = 0] = E[Pmkt − Pipo ] E[Pmkt − Pipo | ∆P > 0] = E[Pmkt − Pipo ] − σ 1+2λ +(1−λ ))NH 12(1+NH 2 2 (18) The first term in each expectation is the average underpricing in the economy. Because expected underpricing following upward (∆P > 0), zero (∆P = 0), and downward (∆P < 0) price adjustment have magnitudes that are in descending order when σ > 0, the implication is verified. The model suggests that the partial adjustment to positive information arises because underwriters will conceal their information ˜ when it is n positive. The government is uninformed, so its price transparency regulations do not correct this behavior. The full adjustment to negative information arises because the underwriters will fully reveal their information ˜ when it is negative. These actions n are optimal for the undewriters because they maximize extracted underpricing. The magnitude of the partial adjustment phenomenon is also proportional to firm-level price uncertainty σ. Implication 3: If λ > 0 and σ > 0, then IPOs with H as lead underwriter (QH > QL ) experience larger underpricing than IPOs with L as lead underwriter (QL > QH ). E[Pmkt − Pipo | QH > QL ] = E[Pmkt − Pipo ] + E[Pmkt − Pipo | QH = QL ] = E[Pmkt − Pipo ] E[Pmkt − Pipo | QH < QL ] = E[Pmkt − Pipo ] − λσ 1+NH λσ 1+NH (19) The first term in each expectation is the average underpricing in the economy. Because expected underpricing when H is lead (QH > QL ), when H and L are co-lead (QH = QL ), and when L is lead underwriter (QL > QH ) have magnitudes that are in descending order when λ > 0 and σ > 0, the result is verified. I will refer to this result as the “underwriter persistence phenomenon” because, when repeated, the game predicts that underwriters with high past underpricing experience larger future underpricing. λ Because 1+NH is proportional to λ, this result is explained by adverse selection favoring 14 H underwriters. H underwriters use their superior information to attract issuers who are initially undervalued by their type L rivals. H underwriters then profit by only partially adjusting the price to reflect this undervaluation. The magnitude of the underwriter persistence phenomenon is also proportional to firm-level price uncertainty σ. Implication 4: If (1) γ > 0, (2) S > 0, and (3) V > R, then H and L both earn positive profits. If, in addition, λ > 0 and σ > 0, then H earns larger profits than L in the form of information rents. E[πL ] = γS(V − R)2 , (1 + NH + NL )2 E[πH ] = E[πL ] + 2γλ2 Sσ2 3(1 + NH )2 (20) The expressions, which are positive under the given assumptions, confirm the implication. L’s profits E[πL ] decrease in the number of underwriters NH + NL , reflecting the increased competition associated with a larger Cournot oligopoly. Similarly, the information rents (E[πH ] − E[πL ]) decrease with the number of H underwriters, reflecting the increased competition within a larger H-oligopoly. The size of the information rents also increase in H’s advantage λ and firm-level price uncertainty σ. The model also makes the informal prediction that IPOs will be oversubscribed. This result obtains because the fraction (1 − γ) of IPO underpricing is captured by IPO investors. The other fraction γ is extracted by underwriters by assumption A1. The availability of these “free profits” explains why investors will seek large allocations and thus why IPOs are oversubscribed.25 B Numerical Examples Table 1 displays numerical examples of the strategic underwriting game. Closed form expressions for each quantity are presented in section A and in the appendix. The “Calibrated Scenario” in Table 1 generates results that are roughly consistent with observed IPO pricing patterns from 1984 to 1997. In the appendix, I discuss the methodology for selecting the calibrated parameter values. On average, IPOs in the “calibrated scenario” are 10% underpriced, compared to the 12.1% actually observed from 1984 to 1997. It also predicts the partial adjustment phenomenon: downward, zero and upward price revisions are followed by initial returns of 5.8%, 10%, and 18.3% respectively. This 12.5% spread is roughly half the observed 27.7% from 1984 to 1997. Because my model only explains the partial adjustment to private information, not public information such as market returns, its ability to explain only half of the partial This also explains why I do not formally model IPO investors in the model: the free profits trivially guarantee their participation. 25 15 adjustment phenomenon is consistent and expected. When combined, my model and Loughran and Ritter (2002) (who explain partial adjustment to public information), likely can fully explain the observed magnitude. The “Calibrated scenario” also shows that H underprices 6.7% more than L. This 6.7% spread approximates the 7.2% observed from 1984 to 1997.26 I conclude that a small informational advantage (λ = .20) can generate large differences in underwriter-specific levels of underpricing. The “High σ Scenario” assumes that firm-level price uncertainty σ is twice as large as in the “Calibrated Scenario”, and may explain why IPO pricing phenomena roughly double in magnitude when the late 1990s are included in the sample. The “Calibrated Scenario” and the “High σ Scenario” respectively predict that the standard deviation of underpricing σunder will be 23.3% and 46.5% respectively. These predictions roughly match the observed 20% in 1984 to 1997 and 45% in 1984 to 2000 respectively. Following upward price adjustment, the “High σ Scenario” underprices 25% more than it does following downward price adjustment. This 25% spread is twice as large as the “Calibrated Scenario’s” 12.5%. The observed 27.7% spread in 1984 to 1997 also doubles and is 50.6% in 1984 to 2000. The “High σ Scenario” also shows that H underwriters underprice 13.3% more than L underwriters. This 13.3% spread is twice as large as the “Calibrated Scenario’s” 6.7% spread. The observed spread of 7.2% in 1984 to 1997 also doubles to 14.7% in 1984 to 2000. These observed two fold increases closely match the predictions made by the strategic underwriting model. However, the “High σ Scenario” and the “Calibrated Scenario” generate identical levels of unconditional IPO underpricing equal to 10%. In contrast, the observed levels increase from 12.7% in 1984 to 1997 to 22.7% 1984 to 2000 respectively. This can be interpreted as a limitation of the model. However, the observed rise in unconditional underpricing may be explained by the high market returns of 1998 to 2000, particularly within the technology industry. When market returns are high, the partial adjustment to public information (my model only explains partial adjustment to private information) can predict the observed increase.27 When combined, my model and Loughran and Ritter’s (2002) model may thus explain all observed changes in the late 1990s, and attributes them to (1) increased price uncertainty and (2) high market returns. The “Monopolistic Scenario”, which has just 2 underwriters, experiences larger IPO underpricing of 16.7% because underwriters compete less aggressively when the Cournot oligopoly is smaller. In contrast, the “Competitive Scenario”, which has 8 underwriters, experiences lower underpricing of 5.6%. The “Monopolistic Scenario” also generates a From 1984 to 1997, underwriters who were in the highest quartile of past underpricing brought to market future IPOs with size and industry adjusted initial returns that were 7.2% larger than underwriters in the lowest quartile. 27 See Loughran and Ritter (2002) for details relating to the partial adjustment to public information. 26 16 larger underwriter persistence phenomenon of 12.0%, compared to the 4.8% spread of the “Competitive Scenario”. This result obtains because a larger NH implies that more H underwriters will compete for the information rents, making them smaller. The “High λ” scenario magnifies H’s advantage and increases the underwriter persistence phenomenon to a larger spread of 16%. The “Zero λ” scenario, in which H and L are identically informed, demonstrates that (1) the underwriter persistence phenomenon is a direct result of underwriter informational heterogeneity and (2) the partial adjustment phenomenon and IPO underpricing depend little on this heterogeneity. IV A Empirical Setup Empirical Variables Two following two return variables are common in the existing literature. ∆P = Pipo − Pmid , Pmid IR = Pmkt − Pipo . Pipo (21) Pmid , Pipo , and Pmkt are the filing date midpoint, the IPO price and the after-market trading price respectively. ∆P is underwriter’s actual price adjustment from the filing date midpoint to the IPO price. IR (initial return) is the return on an investment purchased at Pipo and then sold at Pmkt . Investors who purchase shares at the IPO price Pipo can realize returns equal to IR by selling their shares at the closing price on the first day of public trading. Implication 3 of the strategic underwriting model suggests that an underwriter’s past initial returns should predict his future initial returns. Consider an underwriter J who has issued T IPOs over the past 5 years and who is about to file for his (T+1)-th IPO. {IRJ,1 , ..., IRJ,T } denote the initial returns of his T past IPOs. {IRmktJ,1 , ..., IRmktJ,T } denote the market-wide average initial returns for all IPOs issued in the same calendar month as each past IPO respectively. I define UWpastIR as the following 5 year average T t=1 (IRJ,t UWpastIR = − IRmktJ,t ) . T (22) The numerator is a sum of “abnormal initial returns”: actual initial returns less marketwide average initial returns during the same period. I account for underwriter mergers as follows: (1) An IPO issued by a post-merger underwriter is assigned a value of UWpastIR based on IPOs completed by either pre-merger firm or the post-merger firm itself in the 17 past 5 years. (2) A pre-merger underwriter’s UWpastIR is based only on the IPOs completed by the pre-merger firm itself. Underwriter mergers are identified using the list presented in Ljungqvist, Marston, and Wilhelm (2003). By construction, UWpastIR only includes transactions completed before a given IPO’s filing date. Thus, UWpastIR can be used to predict an IPO’s initial returns early in the IPO process, when the lead underwriter is chosen. Throughout this study, I control for the following variables. ∆P+: Positive price adjustment max[∆P, 0]. ∆P–: Negative price adjustment min[∆P, 0]. UWshare: Equity market share of the lead underwriter in the previous calendar year. CMrank: Carter Manaster Rank from 1 to 9 from Carter, Dark, and Singh (1998). Overhang: Shares retained by the entrepreneur (for all classes) divided by shares filed (including primary and secondary shares). InvPrice: A proxy for issuer risk equal to the reciprocal of the filing midpoint Pmid . LogSize: Natural logarithm of the original filing amount. VC: Dummy variable equal to unity if the firm is VC-backed, zero otherwise. Mkt15: NASDAQ return for the 15 trading days preceeding the issue date. Hot30: Average initial return of IPOs issued in the 30 days before the issue date. B Data and Descriptive Statistics The data are from the Securities Data Company (SDC) U.S. New Issues Database. The sample initially consists of all U.S. IPOs from January 1, 1980 to December 31, 2000. I eliminate ADRs, unit issues, REITs, Financial firms, and firms with offer prices less than 5 dollars. Because UWpastIR is based on underwriter history, observations satisfying either of the following 2 conditions are only used to compute stable starting values for UWpastIR (and are otherwise excluded): (1) IPOs issued in first four years of the sample 1980 to 1983 and (2) IPOs underwritten by a lead underwriter who has completed fewer than 10 past IPOs.28 Because the number of past IPOs is public information, no bias is introduced by this exclusion. 3,950 observations remain for 1984 to 2000. I also consider the sub-samples 1984 to 1989, 1990 to 1997, and 1984 to 1997. The 1984 to 1997 sample tests for robustness to excuding the late 1990s technology boom. Table 2 displays summary statistics and confirms that IPO prices are uncertain. For example, the standard deviation of the underwriter’s price adjustment ∆P is 24.3% from 28 Similar results obtain when this cutoff is placed at 15 or 20 IPOs. 18 1984 to 2000. The standard deviation of initial returns IR is 48.8%. This confirms that underwriters constructively revise IPO prices, but do so imperfectly. Like the Carter/Manaster rank and underwriter market share, UWpastIR is a property of the lead underwriter and is a measure of underwriter reputation. UWpastIR’s standard deviation of 4.8% indicates that underwriters experience meaningful differences in their past initial returns. Its mean near zero results from its construction based on abnormal initial returns. Average underwriter market share (UWshare) is 4.8% and the average Carter/Manaster rank (CMrank) is 7.9. The relatively high CMrank results because (1) reputable underwriters transact more IPOs (reported averages are transaction weighted) and (2) as noted, IPOs issued by inexperienced underwriters are excluded. C Underwriter Reputation Table 3 presents Pearson correlation coefficients and suggests that underwriter reputation may be two-dimensional: (1) CMrank and UWshare measure the underwriter’s ability to attract larger issuers (56% and 45% correlated with LogSize respectively) with lower risk (-57% and -28% correlated with InvPrice respectively); (2) UWpastIR measures the underwriter’s ability to attract somewhat riskier issuers (23% correlated with InvPrice) that are initially undervalued and require upward price adjustment (18% correlated with ∆P). This correlation with ∆P is consistent with the strategic underwriting model’s prediction that high UWpastIR underwriters can identify issuers who are initially undervalued. UWpastIR’s near zero correlation with Nasdaq market returns (Mkt15) and recent market wide initial returns (Hot30) confirms that UWpastIR, by construction, is not related to short-term market effects. The table also confirms that UWpastIR is a distinct measure of underwriter reputation because it correlates just -16% and 2% with CMrank and UWshare respectively. In contrast, CMrank and UWshare correlate 48%. Table 4 summarizes IPO characteristics versus underwriter reputation. Panel A shows that initial returns increase monotonically with increasing UWpastIR from 15.8% to 32.9% for the sample period 1984 to 2000. This 17.1% spread is economically large. The “Residual IR” column displays each group’s average residuals from the following OLS regression equation: IR = β1 LogSize + β2 LogSize2 + [Fama-French 48 industry fixed effects] + ˜. (23) The residuals, which control for size and industry effects, still increase monotonically from –5.3% to 9.4% versus increasing UWpastIR. This 14.7% spread is still economically large. Panel B shows that underwriter market share (UWshare) and the Carter-Manaster 19 Rank (CMrank) also sort raw initial returns. For example, they increase monotonically from 16.6% to 31.8% versus increasing UWshare. However, (1) neither CMrank nor UWshare sorts “Residual IR” in a stable fashion and (2) the ability of CMrank and UWshare to sort initial returns is mainly attributed to observations from the technology bubble (1998 to 2000).29 In contrast, section V will show that UWpastIR’s ability to predict initial returns is both stable and economically large across sub samples. Table 4 also shows that high CMrank and high UWshare underwriters strongly favor larger issuers. In contrast, high UWpastIR underwriters have only a small tendency to favor large issuers. Though all 3 reputation measures show some ability to sort underwriter spread (transaction cost), the residual spread column shows that spreads are constant versus underwriting reputation when size and industry controls are added. Figure 2 displays scatter plots of abnormal initial returns (IR) versus abnormal price adjustment (∆P ) for Morgan Stanley (172 IPOs) and Lehman Brothers (135 IPOs) from 1984 to 1997. An IPO’s abnormal IR and ∆P are equal to its raw IR and ∆P less market wide averages in the month the IPO was completed. This adjustment eliminates the public information component from both variables. Both underwriters have the highest Carter/Manaster rank (CMrank) of 9, confirming that CMrank cannot explain the figure’s results. In contrast, Morgan Stanley has a high UWpastIR of +6.7%, compared to Lehman’s -6.2%. The figure confirms the strategic underwriting model’s prediction that UWpastIR may be driven by adverse selection because (1) points to the right of the vertical bar experience higher abnormal initial returns due to partial adjustment; and (2) only 28% of Lehman’s IPOs are to the right compared to Morgan Stanley’s 61%. If underpricing is indeed profitable by assumption A1 (see section I), Lehman suffers from an adverse selection favoring less profitable overvalued issuers. In contrast, Morgan Stanley attracts issuers who they can profit from by partial adjustment. This result is not unique as a plot of Goldman Sachs (UWpastIR=+4.3%) versus Salomon Brothers (UWpastIR=-5.5%), who both have a CMrank of 9, is virtually identical. V Predictability of Initial Returns Table 5 displays the results of Fama-MacBeth (1973) style regressions that predict initial returns (IR). I first compute separate cross-sectional regression coefficients for IPOs issued each calendar year. Consistent with the Fama-MacBeth method, the table’s reported coefficients and T-statistics are based on the average yearly cross-sectional coefficients. Pooled OLS regressions (not reported) generate results that are more significant 29 Based on rank tables constructed for the subsample 1984 to 1997 (not reported). 20 than the reported Fama-MacBeth results.30 I present Fama-MacBeth results to show that UWpastIR’s ability to predict initial returns is a feature of IPO pricing that is both stable and persistent over time. Panel A of Table 5 restricts the explanatory variables to those known on the initial filing date. Rows (1) to (4) show that UWpastIR, with a T-statistic that ranges from 2.5 in the 1980s to 7.8 in the 1990s, is a significant predictor of initial returns in all sub-samples. The overall coefficient of .62 (.96 in pooled OLS, not reported) suggests that 62% (96%) of an underwriter’s average initial returns over the past 5 years are expected to continue in the following year. I will refer to this result as the “underwriter persistence phenomenon”. I consider the following hypothesis to explain this result. H1 The underwriter persistence phenomenon is explained by the strategic underwriting model (see implication 3 in section III). I also test the following 5 alternative hypotheses. AH1 UWpastIR is a proxy for existing reputation measures such as CMrank and UWshare, which in turn explain the underwriter persistence phenomenon. AH2 The underwriter persistence phenomenon is explained by industry specialization, where some underwriters serve industries with exogenously higher initial returns. AH3 The underwriter persistence phenomenon is explained by short-term “hot market” effects. This hypothesis predicts a relationship between UWpastIR and market returns or the initial returns of recently issued IPOs. AH4 The underwriter persistence phenomenon is explained by habitual (behavioral) underpricing by a subset of underwriters. AH5 The underwriter persistence phenomenon compensates high UWpastIR underwriters for providing better service. I test hypotheses AH1, AH2, and AH3 in this section. Hypotheses AH4 and AH5 are indirectly tested in section VI and section VII respectively. A Discussion Two results from Table 5 show that hypothesis AH1 cannot explain underwriter persistence. (1) Panel A shows that UWpastIR predicts initial returns even when controls Many existing studies, such as Bradley and Jordan (2002) use pooled OLS regressions when testing whether a given set of variables can predict initial IPO returns. 30 21 for the Carter-Manaster Rank (CMrank) and underwriter market share (UWshare) are included.31 (2) The CMrank coefficient changes sign from the 1980s (-.30) to the 1990s (+1.18), which reproduces a result previously reported in Beatty and Welch (1996).32 In contrast, the UWpastIR coefficient is positive in rows 1 to 4 and ranges from .48 in the 1980s to .70 in the 1990s. I conclude that AH1 cannot explain underwriter persistence. Panel B tests hypotheses AH2 and AH3.33 I test AH2 by adding industry fixed effects based on the Fama-French 48 industries to the regressions in Panel A.34 The table shows that the UWpastIR coefficient declines by only 8%, from .62 to .57, when industry effects are included.35 I conclude that the majority of the underwriter persistence phenomenon, at least 90%, cannot be explained by AH2. Panel B tests AH3 by adding controls for Recent Nasdaq market returns (Mkt15) and the initial returns of recently issued past IPOs (Hot30). The table shows that Mkt15 significantly predicts initial returns. Consistent with Lowry and Schwert (2001), Hot30 does not predict initial returns.36 However, the UWpastIR coefficient is not significantly affected by the addition of either Mkt15 or Hot30. I conclude that AH3 cannot explain underwriter persistence. Panel C adds positive and negative price adjustment, ∆P+ and ∆P–, to the regressions in Panel B. As documented in Hanley (1993) positive price adjustment ∆P+ is the most significant predictor of initial returns. This result is known as the partial adjustment phenomenon because it suggests that underwriters do not fully adjust the IPO price to reflect positive information learned during the waiting period. The table shows that the UWpastIR coefficient is 56% subsumed by ∆P+ and ∆P–. The ability of price adjustment to subsume UWpastIR is predicted by the strategic underwriting model (hypothesis H1), which explains that high UWpastIR underwriters are better informed and attract issuers who are initially undervalued. High UWpastIR underwriters then profit by only partially adjusting the IPO price to correct this undervaluation. This result is further supported by regressions (not reported) that remove CMrank and UWshare from the model presented in Table 5. These regressions show that UWpastIR coefficient does not change materially when CMrank and UWshare are omitted or included. 32 Though UWshare and CMrank are somewhat correlated, the results of these regressions do not change significantly when either is omitted from the regression. 33 Results are identical if AH2 and AH3 are tested separately. 34 See Ken French’s website for the 48 Fama-French industry definitions. 35 When the fixed effects are based on 2-digit or 3-digit SIC codes (not reported), the UWpastIR coefficient declines by roughly 10% instead of 8%. I present results using the 48 industries because 3 or 4 digit SIC codes consume a large fraction of the available degrees of freedom in the yearly cross sectional regressions. 36 In pooled OLS regressions (not reported), Hot30 does significantly priedict initial returns. However, controlling for Hot30 still does not impact the UWpastIR coefficient. 31 22 Panel D predicts initial returns using residual ∆P+ and ∆P–, instead of their raw values. The residuals are from OLS regressions of the following form: ∆P+ = α + β1 UWpastIR + ˜ (24) This specification is relevant because UWpastIR becomes public information months before ∆P+ and ∆P–.37 All other explanatory variables in Panel D remain in their raw form. When the residual ∆P+ and ∆P– are substituted, the UWpastIR coefficient rises to 1.04 with a T-statistic of 6.0. This coefficient near unity suggests that an underwriter’s past initial returns are fully expected to continue. Panel D summarizes the main results of this section, which are consistent with hypothesis H1: (1) UWpastIR is among the most significant predictors of future initial returns and cannot be explained by AH1, AH2, or AH3; (2) underwriter persistence is positively related to the partial adjustment phenomenon; and (3) the role of UWpastIR is especially important because it becomes known before ∆P+ and ∆P–. This latter fact suggests that investors can condition their participation on UWpastIR, and can add 14.7% to their first-day returns (see Table 4) simply by purchasing shares from a high UWpastIR underwriter. The existence of these large one-day returns is consistent with assumption A1 (see section I), which suggests that these profits are not free, and that underwriters extract a fraction of them via quid pro quo arrangements. VI Underwriter Market Share Table 6 displays the results of Fama-MacBeth (1973) regressions that predict future changes in underwriter market share. One observation is an underwriter in a calendar year. The dependent variable is the percentage change in market share from the past 5 calendar years (t-5 to t-1) to the given calendar year (t+1). The independent variables are the positive and negative components of UWpastIR measured from year t-5 to t-1 UWpastIR+ = max[UWpastIR, 0], UWpastIR− = min[UWpastIR, 0]. (25) Table 6 shows that large underwriters, who underwrite the majority of IPOs and IPOdollars, have UWpastIR+ and UWpastIR- coefficients that are positive and significant. For example, large underwriters in 1984 to 1997 experience market share growth of 5.4% for each percentage point that their past initial returns exceed market-wide averages. UWpastIR is known at the time the issuer selects a lead underwriter, which typically occurs before both the filing date and the IPO date. In contrast, ∆P+ and ∆P– only become known on the IPO date when the final IPO price is announced. 37 23 Similarly, they experience a 3.5% market share decline for each percentage point that their past initial returns are below market-wide averages. Both support hypothesis H1 because implications 3 and 4 in section III suggest that higher UWpastIR underwriters are more profitable due to their ability to extract information rents. Because more profitable underwriters should experience market share growth, the positive sign for both the UWpastIR+ and UWpastIR- coefficients support H1. Hypothesis AH4 is not supported because it suggests that underwriters with high (low) UWpastIR price IPOs less (more) efficiently. Panel A contradicts this preciction because it would suggest that less efficient underwriters are rewarded with growing market share. The results in Panel A are important because large underwriters issue 78% of all IPOs and 92% of all IPO dollars. Panel B shows that, like large underwriters, small underwriters have a positive UWpastIR- coefficient in all subsamples. In contrast to large underwriters, however, they have a negative UWpastIR+ coefficient in all subsamples. For example, small underwriters with high UWpastIR in 1984 to 1997 experience a 2.14% market share decline for each percentage point that their initial returns exceed market wide averages. This result is inconsistent with H1. Similarly, the results are also inconsistent with AH4 because the UWpastIR- coefficient is also positive, which would suggest that efficient underwriters lose market share. In contrast to H1 and AH4, Panel B supports Beatty and Ritter (1986), who predict that underwriters with either positive or negative UWpastIR will experience declining market share. The authors explain that only underwriters with initial returns near market-wide averages can successfully balance the opposing objectives of both investors and issuers. Though interesting, the importance of the results in Panel B are minimal because small underwriters issue just 22% of new issues and 8% of IPO-dollars. The overall regressions in Panel C also support Beatty and Ritter (1986). This is primarily because, consistent with Beatty and Ritter (1986), the regression methodology is equal weighted across all underwriters. An underwriter who issues 80 IPOs is thus weighted the same as an underwriter who issues just 5 IPOs. Overall, Panel A shows that the majority of IPOs and IPO dollars are consistent with H1. Moreover, the results of all 3 panels are inconsistent with AH4. VII The Reissue Market The reissue market, which I define to include both debt and equity transactions, is where publicly traded firms raise capital in the years following their IPO. Where one observation is one IPO, Table 7 displays Fama-MacBeth regressions that test whether 24 the lead IPO underwriter’s reputation can predict the frequency of future visits to the reissue market. I define “LogReissue” as the natural logarithm of one plus the number of times a firm visits the reissue market in the five years following its IPO. Panel A shows that firms underwritten by high UWpastIR underwriters visit the reissue market less frequently. This result is significant at the 5% level except 1990 to 1997, where it is significant at the 10% level. Panel B shows that this result is robust to adding controls for LogSize, InvPrice, venture capital financing and existing reputation measures. Because firms who need to visit the reissue market more frequently likely choose underwriters who offer better service, I conclude that underwriter persistence likely does not compensate high UWpastIR underwriters for offering better service. This suggests that AH5 likely cannot explain underwriter persistence. Row (7) in Panel B adds a control for past visits to the reissue market made by firms previously underwritten by the lead underwriter. The “number of past visits” is the equal weighted average of LogReissue over all firms underwritten by the lead underwriter in the past 5 calendar years. Adding this variable to the regression partially subsumes the UWpastIR coefficient. However, UWpastIR remains negative and significant at the 10% level. Despite the reduced significance, AH5 still likely cannot explain underwriter persistence because it predicts that the UWpastIR coefficient will be positive and significant, a result that is not supported in any specification. In contrast to UWpastIR, Panel A shows that firms underwritten by high UWshare and high CMrank underwriters do visit the re-issue market more frequently. Thus, high UWshare and high CMrank underwriters may offer better service. However, Panel B shows that this result is explained by issuer size and the number of past visits. Panel B thus confirms the link between CMrank, UWshare, and Size described in section C. VIII Underwriter Concentration The strategic underwriting model, hypothesis H1, suggests that entrepreneurs who face a more concentrated underwriting market will experience higher underpricing (see implication 1, section III). In this section, I test whether two measures of concentration can explain initial returns in cross section. The first is “Industry Underwriting HHI”, the prevailing industry-by-industry Herfindahl index, which is the sum of squared underwriter market shares (based on number of transactions) for the most recent 15 IPOs completed in a given industry.38 The second is “Solo Lead Underwriter”, which is equal Industry definitions are based on the Fama-French 48 industry groupings, which are available on the Ken French website. Results are robust to using the 10 or 20 most recent IPOs instead of 15 and also to using 2 or 3 digit SIC codes. 38 25 to 1 if an IPO is managed by a single lead underwriter and is zero otherwise. This measure proxies for concentration because IPOs with more (fewer) lead underwriters are likely underwritten in a more competitive (concentrated) environment. Where one observation is one IPO, Table 8 displays the results of Fama-MacBeth regressions that predict initial returns using both concentration measures. The table confirms that issuers who face more concentrated underwriting markets experience higher initial returns. For example, Panel A shows that, with the exception of the 1984 to 1989, each percentage point increase in the Industry Herfindahl index generates a .29% to .85% increase in expected initial returns. Similarly, an IPO underwritten by a solo lead underwriter experiences initial returns that are 2.8% to 4.5% higher. Panel B shows that both results are also robust to adding controls for LogSize, reputation measures, the inverse of the IPO price, and venture capital financing. The results broadly support hypothesis H1. IX Conclusion This study presents a model of strategic underwriting that can explain three empirical IPO pricing phenomena. (1) IPO underpricing arises from imperfect underwriter competition. (2) The partial adjustment phenomenon arises because underwriters, who face an uninformed government regulator, adjust the IPO price less when they can extract more of the issuing firm’s value. (3) The underwriter persistence phenomenon, the tendency of some underwriters to underprice more than others, arises from adverse selection, which allows informed underwriters to better identify, and profit from, undervalued issuers. The model also makes predictions about underwriter profitability. The empirical section of this study considers three measures of underwriter reputation. The results suggest that (1) The Carter/Manaster rank and underwriter market share summarize an underwriter’s ability to attract large issuers. Neither predicts stable variation in initial returns. (2) A new reputation measure (UWpastIR), which is based on each underwriter’s past initial returns, summarizes the underwriter’s ability to attract issuers who are initially undervalued and require upward price adjustment. UWpastIR is among the most stable and significant predictors of initial returns. This “underwriter persistence phenomenon” is robust over subsamples and to numerous controls. Because the choice of lead underwriter is known early in the IPO process, this suggests that investors can condition their participation on underwriter persistence. The size of the resulting 14.7% improvement one-day profits supports the notion that these profits are not free and that underwriters extract a fraction of them via quid pro quo arrangements. 26 I also find that (1) large underwriters with high UWpastIR experience growing market share; (2) firms underwritten by high UWpastIR underwriters visit the reissue market less frequently; and (3) issuers who face a more concentrated underwriting market experience higher initial returns. These empirical findings support the strategic underwriting model and limit the admissibility of alternative explanations. X A Appendix Calibration Table 1 displays numerical examples of the strategic underwriting game. Closed form expressions for each quantity are presented in section III. In this section, I discuss how values of the model’s exogenous parameters {S, γ, V, R, NH , NL , λ, σ} were chosen for the “Calibrated Scenario” displayed in Table 1. The goal is to choose parameter values that roughly match observed values during the sample period 1984 to 1997. I exclude 1998 to 2000 from this calibration in order to show that features of the technology boom can be explained by higher price uncertainty σ. NH and NL : Consistent with Arkebauer (1991), I assume that N=4 underwriters compete in the oligopoly. For simplicity, I assume that NH = 2 are informed and NL = 2 are uninformed. Arkebauer (1991) suggests that only 2 to 5 underwriters actually compete for any given IPO. He suggests that competition is limited in this way because underwriters (1) have heterogeneous distribution capabilities, (2) may specialize in certain industries, and (3) are reluctant to work with an entrepreneur who “shops” more than 4 or 5 underwriters. σ: From 1984 to 1997, the standard deviation of initial IPO returns was roughly 20%. In the strategic underwriting model, the expression for this standard deviation is: 2((1 + NH )2 − λ2 σ2 (NH 2 + 2NH − 2)) σunder = (26) (1 + NH )2 The chosen value σ = .5 generates a σunder of 23%, which roughly matches the observed value. λ: This parameter measures the informational heterogeneity among underwriters. Because this quantity is difficult to measure in practice, I choose a value of .20 specifically to calibrate the model to reproduce the underwriter persistence phenomenon observed from 1984 to 1997. During this period, underwriters who were in the highest quartile of past underpricing brought to market future IPOs with size and industry adjusted initial returns that were 7.2% larger than underwriters in the lowest quartile. Because λ can take on values between zero and one, the required value of .20 confirms that just a small amount of informational heterogeneity is needed to generate high levels of underwriter persistence. 27 V and R: V is the firm’s initial expected value as a publicly traded firm. I choose V=1 for convenience so that price quantities will be displayed as percentages. I choose the reservation value R=.5 to be half of V. This is consistent with the notion that privately held firms are less valuable than their publicly traded counterparts. S and γ: S and γ are scaling factors and do not impact IPO prices (see comparative statics in III). I choose S=100 and γ = .5 for convenience. B Proof of Theorems 1 and 2 Theorem 1 is a special case of Theorem 2, where NH = NL = 1, so it is sufficient to prove Theorem 2. I begin with the period 1 underwriting objectives from equation (15): MAX γQH · (V − R − QH QH +QH− +NL QL S ˜ + ληH ) (27) MAX γQL · (V − R − QL QL +QL− +NH E[QH ] ). S The objectives describe the problem faced by a single H or L underwriter. QH− is the total quantity underwritten by an H underwriter’s (NH − 1) type H rivals. The H underwriter holds QH− fixed when maximizing QH . QL− is similarly defined. These quadratic objectives have the following solutions assuming a symmetric equilibrium: QH ∗ = S(λ˜ H + V − R) − NL QL ∗ η , 1 + NH QL ∗ = S(V − R) − NH E[QH ∗ ] . 1 + NL (28) ˜ Because ηH has a mean of zero, the expectation E[QH ∗ ] is: E[QH ∗ ] = S(V − R) − NL QL ∗ . 1 + NH (29) After substituting (29) into (28), the solution to the two simultaneous equations is: QL ∗ = S(V − R) , 1 + NH + NL QH ∗ = S(V − R) Sλ˜ H η + . 1 + NH + NL 1 + NH (30) The equilibrium price adjustment ∆P ∗ results from selecting the lowest ∆P allowed by regulation and is stated in equation (12). The proof of uniqueness is trivial because the solution to the quadratic period 1 objectives is unique and the solution to the constrained linear period 2 objective is also unique. The solution has an interpretation when the 4 regularity conditions stated in theorem 2 hold. 28 C Comparative Statics and Proof of Implication 3 Table 9 lists the 27 states that are possible in the economy and their respective probabilities. Each state is labeled from “s1 ” to “s27 ”. The unconditional probability measure in this economy is: 27 E[x] = i=1 Pr[si ] · E[x | si ] (31) The conditional probability measure is: E[x | s ∈ S] = s∈S Pr[si | s ∈ S] · E[x | si ], where S ⊆ {s1 , ..., s27 } (32) Each of the comparative statics is proven using either the unconditional or the conditional probability measure. I prove implication 3 as an example. To simplify the notation, I define U as underpricing U ≡ Pmkt − Pipo . The first goal of the proof is to write the equilibrium level of underpricing (U∗ ) in terms of the model primitives: the parameters {V, NH , NL , γ, R, S, λ} and the random variables { ˜ , ˜ , ηH }. To do this, I substitute expressions for Pmkt , Pipo , n n ˜ and Pmid from equations (6), (3), and (14) as follows: U ≡ Pmkt − Pipo = V + ˜ + ˜ − R − p n NH QH + NL QL ˜ + E[∆P | ηH ] − ∆P S (33) The equilibrium solution ∆P ∗ in equation (16), and the properties of the random variables ˜ and ηH , can be used to solve the expectation: ˜ n ˜ E[∆P | ηH ] = − 1−λ ˜ + λ min[0, ηH ] 3 (34) By substituting the equilibrium solutions {QH ∗ , QL ∗ , ∆P ∗ } from equation (16) and equation (34) into (33), the following result obtains: ˜ U∗ = max[0, ˜ ]+λ min[0, ηH ]+ n ˜ λ(1 + NH − 3NH ηH ) (3V − 3R − 1 − NH − NL ) + (35) 3(1 + NH ) 3(1 + NH + NL ) The equilibrium quantities QH ∗ and QL ∗ in equation (16) show that H is lead under˜ ˜ writer (QH > QL ) when ηH = 1. Similarly, L is lead underwriter when ηH = −1. H and ˜ ˜ L are co-lead underwriters when ηH = 0. Table 9 verifies that ηH is equal to {1, 0,-1} 29 each for 9 of the 27 states. The conditional probability measure in equation (32) can be used to write the desired expectations as: E[Pmkt − Pipo | QH > QL ] = s∈S Pr[s | s ∈ S] U∗ [s], Pr[s | s ∈ S] U∗ [s], s∈S where S = {s1 , ..., s9 } where S = {s10 , ..., s18 } where S = {s19 , ..., s27 }. (36) E[Pmkt − Pipo | QH = QL ] = E[Pmkt − Pipo | QH < QL ] = s∈S Pr[s | s ∈ S] U∗ [s], For each of the 9 states “s” corresponding to each sum, Table 9 describes how the random variables { ˜ , ˜ , ηH } should be substituted into U∗ [s] when evaluating equation n n ˜ (36). The conditional probabilities Pr[s | s ∈ S] are evaluated by applying Bayes rule to the state probabilities listed in Table 9. After collecting and simplifying terms, the following results obtain, completing the proof: E[Pmkt − Pipo | QH > QL ] = E[Pmkt − Pipo | QH = QL ] = E[Pmkt − Pipo | QH < QL ] = V−R 1+NH +NL V−R 1+NH +NL V−R 1+NH +NL + λ 1+NH (37) − λ 1+NH 30 XI References Aggarwal, Reena, N. Prabhala, and Manju Puri, 2002, “Institutional allocation in initial public offerings: empirical evidence,” Journal of Finance 57(3), 1421-42. Allen, Franklin, and Gerald Faulhaber, 1989, “Signaling by underpricing in the IPO market,” Journal of Financial Economics 23, 303-324. Amihud, Yakov, Shmuel Hauser, and Amir Kirsh, 2001, “Allocations, adverse Selection, and cascades in IPOs: Evidence from Israel,” working paper, New York University. Arkebauer, James, and Sam Schultz, 1991, “Cashing Out: The Entrepreneur’s Guide to Going Public” (Harper Business, New York, NY). Baron, David, 1982, “A model of the demand for investment banking advising and the distribution services for new issues,” Journal of Finance 37(4), 955-976. Barry, C., 1989, “Initial Public Offering Underpricing: The Issuer’s View, A Comment,” Journal of Finance 54, 1099-1103. Beatty, Randolf and Jay Ritter, 1986, “Investment banking, reputation and the underpricing of initial public offerings,” Journal of Financial Economics 15(1-2), 213-32. Beatty, Randolf and Ivo Welch, 1996, “Issuer expenses and legal liability in initial public offerings,” Journal of Law and Economics 39(2), 545-602. Benveniste, Lawrence and Paul Spindt, 1989, “How investment bankers determine the offer price and allocation of new issues,” Journal of Financial Economics 24, 343-362. Bradley, Daniel and Bradford Jordan, 2002, “Partial adjustment to public information and IPO underpricing,” Journal of Financial and Quantitative Analysis 37(4), 595-616. Carter, Richard and Steven Manaster, 1990, “Initial public offerings and underwriter reputation,” Journal of Finance 45(4), 1045-67. Carter, Richard; Dark, Frederick, and Ajai Singh, 1998, “Underwriter reputation, initial returns, and the long run performance of IPO stocks,” Journal of Finance 53(1), 285-311. Edelin, Roger, and Gregory Kadlec, 2003, “Issuer surplus and the partial adjustment of IPO prices to public information,” Wharton School and Pamplin College Working Paper. Fama, Eugene, and J. MacBeth, 1973, ”Risk, Return and Equilibrium: Empirical Tests,” Journal of Political Economy, 71, 607-636. Fulghieri, Paulo and Matthew Spiegel, 1993, “A theory of the distribution of underpriced initial public offers by investment banks,” Journal of Economics and Management Strategy Winter 1993 2(4), 509-30. Hanley, Kathleen Weiss, 1993, “The underpricing of initial public offerings and the partial adjustment phenomenon,” Journal of Financial Economics 34, 231-250. 31 Hansen, R, 2001, “Do investment banks compete for IPOs? The advent of the ‘7% plus contract’,” Journal of Financial Economics 59, 313-346. Jenkinson, Tim, and Alexander Ljungqvist, 2001, “Going Public: The Theory and Evidence on How Companies Raise Equity Finance,” 2nd Edition (Oxford University Press, Oxford, UK). Johnson, James and Robert Miller, 1988, “Investment banker prestige and the underpricing of initial public offerings,” Financial Management 17, 19-29. Krigman, I., W. Shaw, and K. Womack, 2001, ”Why do firms switch underwriters?,” Journal of Financial Economics, 60, 245-284. Ibbotson, Roger, 1975, “Price performance of common stock new issues,” Journal of Financial Economics 2, 235-272. Ljungqvist, Alexander, Felicia Marston, and William Wilhelm, 2003, “Competing for Securities Underwriting Mandates: Banking Relationship and Analyst Recommendations,” New York University Working Paper. Logue, Dennis, Richard Rogalski, James Seward, and Lynn Foster-Johnson, 2002, “What’s special about the role of underwriter reputation and market activities in IPOs,” Journal of Business 75, 213-243. Loughran, Tim, and Jay Ritter, 2002, “Why dont issuers get upset about leaving money on the table in IPOs,” Review of Financial Studies 15, 413-433. Lowry, Michelle and William Schwert, 2002, “IPO Market Cycles: Bubbles or Sequential Learning,” Journal of Finance, 57(3), 1171-1200. Maksimovik, Vojislav, and Pegaret Pichler, 2002, “Structuring the initial offering: who to sell to and how to do it,” University of Maryland and Boston College Working Paper. Megginson, William and Kathleen Weiss, 1991, “Venture capitalist certification in initial public offerings,” Journal of Finance 46(3), 879-903. Michaely, Roni, and Wayne Shaw, 1994, “The pricing of initial public offerings: tests of adverse selection and signaling theories,” Review of Financial Studies 7, 279-319. Sherman, Ann, and Sheridan Titman, (2002), “Building the IPO order book: underpricing and participation limits with costly information”, Journal of Financial Economics 65(1), 3-29. Ritter, Jay and Ivo Welch, 2002, “A review of IPO activity, pricing, and allocations,” Journal of Finance 57(4), 1795-1828. Rock, Kevin, 1986, “Why new issues are underpriced,” Journal of Financial Economics 15, 187-212. Welch, Ivo, 1989, “Seasoned offerings, imitation costs, and the underpricing of initial public offerings,” Journal of Finance 44, 421-450. Welch, Ivo, 1992, “Sequential Sales, Learning and Cascades,” Journal of Finance 47, 695-732. 32 Table 1. Numerical examples of the strategic underwriting game. Explanation: The table displays the predicted values of various statistics related to unconditional underpricing, conditional underpricing, profitability and the likelihood of delay (delay results when the underwriter chooses to reveal all or part of ˜ , see section C). The model parameters are {N, NH , NL , σ, λ, V, R, S, γ}. N is n the total number of underwriters, NH of which are partially informed (type H) and NL of which are uninformed (type L). σ is a factor identifying the level of firm-level price uncertainty. λ identifies the accuracy of H underwriters’ signal and thus determines the magnitude of underwriter heterogeneity. V is the firm’s initial valuation in period 1 based on all public information. R is the issuer’s reservation value (the price at which he would choose to issue zero shares). S is a scaling factor indicating the size of the entrepreneur’s firm. γ is the assumed fraction of underpricing that underwriters are able to extract from their investors via quid pro quo arrangements. Closed form expressions for each expectation are presented in the comparative statics section and the methodology for evaluating the expectations is presented in the appendix. The “Calibrated Scenario” generates predictions that roughly match empirically observed values (see appendix for discussion). The “High σ Scenario” assumes a higher level of price uncertainty than the Calibrated Scenario. The “Monopolistic” scenario assumes fewer underwriters than the Calibrated Scenario. The “Competitive” scenario assumes more underwriters than the Calibrated Scenario. The “High λ” scenario assumes greater underwriter heterogeneity and a higher λ than the Calibrated Scenario. The “Zero λ” scenario assumes the underwriters are homogeneous and thus sets λ to zero. Variable Assumed Parameter Values Expression Calibrated Scenario High σ Scenario Monopolistic Scenario Competitive Scenario High λ Scenario Zero λ Scenario 33 Average Underpricing E[Pmkt − Pipo ] σunder 10.0% 23.3% 10.0% 46.5% 16.7% 23.5% The Partial Adjustment Phenomenon E[Pmkt − Pipo | ∆P > 0] E[Pmkt − Pipo | ∆P = 0] E[Pmkt − Pipo | ∆P < 0] 18.3% 10.0% 5.8% 26.7% 10.0% 1.7% 25.2% 16.7% 12.4% E[Pmkt − Pipo | QH > QL ] E[Pmkt − Pipo | QL > QH ] 13.3% 6.7% 16.7% 3.3% 21.7% 11.7% Underwriter Profits E[πH ] E[πL ] 0.54 0.50 0.65 0.50 1.47 1.39 Total Number of Underwriters Number of H Underwriters Number of L Underwriters Price Uncertainty Factor H’s Information Advantage Public Value Reservation Value Size Factor Extraction Rate NH NH NL σ λ V R S γ 4 2 2 0.50 0.20 1.00 0.50 100 0.50 4 2 2 1.00 0.20 1.00 0.50 100 0.50 2 1 1 0.50 0.20 1.00 0.50 100 0.50 8 4 4 0.50 0.20 1.00 0.50 100 0.50 4 2 2 0.50 0.40 1.00 0.50 100 0.50 4 2 2 0.50 0.00 1.00 0.50 100 0.50 Average Underpricing Standard Deviation of Underpricing 5.6% 23.2% 10.0% 22.3% 10.0 % 23.6% Underpricing After Up Revision Underpricing After Zero Revision Underpricing After Down Revision 13.8% 5.6% 1.5% 18.3% 10.0% 5.8% 18.3% 10.0% 5.8% Underpricing for Informed (H) and Uninformed (L) Underwriters 7.6% 3.6% 16.7% 3.3% 10.0% 10.0% Underpricing by H Underpricing by L Underwriter H’s Profits Underwriter L’s Profits 0.17 0.15 0.65 0.50 0.50 0.50 Table 2. Summary Statistics Explanation: Summary statistics are reported for IPOs issued in the US from 1984 to 1997 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). IR is the implied return from the IPO price to the after market trading price. ∆P is the implied return from the filing date midpoint to the IPO price and thus reflects the underwriter’s actual price adjustment. For a given IPO, UWpastIR is equal to the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding the filing date of the given IPO. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s equity market share computed over the previous calendar year. Overhang is equal to the pre-IPO shares retained by the issuer divided by the shares filed, both primary and secondary. InvPrice is the reciprocal of the filing date midpoint. LogSize is the natural logarithm of the original filing amount. VC is a dummy that is equal to one if the firm is VC-backed. ∆P+ and ∆P– are the positive and negative truncated components of ∆P (explained above). Mkt15 is the cumulative NASDAQ return over the 15 trading days prior to the issue date. Hot30 is the average underpricing for all IPOs issued in the 30 day window preceeding the issue date. “Industry HHI” is the prevailing industry-by-industry underwriting Herfindahl index, which is computed using the most recent 15 IPOs completed in a given industry using the 48 Fama-French industries before a given IPO’s filing date. Industry HHI is the sum of squared market shares (based on number of transactions) of the lead underwriters who completed the 15 IPOs. “Solo Lead” is equal to 1 if only one lead underwriter manages the given IPO and is zero otherwise. Standard Mean Price Variables 0.227 0.020 Explanatory Variables 0.001 7.902 0.048 3.086 0.084 19.645 0.463 0.094 –0.075 0.011 0.236 1.386 0.060 2.126 0.027 0.941 0.499 0.180 0.114 0.048 0.285 Concentration Measures 0.102 0.957 0.024 0.215 0.067 0.000 0.102 1.000 0.227 1.000 0.048 –0.304 1.000 0.000 0.000 0.011 16.792 0.000 0.000 –0.677 –0.219 –0.028 –0.005 8.000 0.026 2.558 0.080 19.508 0.000 0.000 0.000 0.013 0.143 0.198 9.000 0.435 31.692 0.250 24.521 1.000 2.200 0.000 0.250 4.735 0.243 –0.677 0.488 –0.328 0.078 0.000 6.975 2.200 Deviation Minimum Median Maximum Number of Observations Variable Description IR Initial returns 3,950 3,950 ∆P % Price adjustment from Pmid to Pipo 35 UWpastIR Lead underwriter’s past abnormal IR 3,950 3,950 3,950 3,950 3,950 3,950 3,950 3,950 3,950 3,950 3,950 CMrank Lead underwriter’s Carter-Manaster rank UWshare Lead underwriter’s equity market share Overhang Shares retained / shares filed InvPrice One divided by the offer price LogSize Natural Logarithm of filing amount VC Dummy = 1 if venture capital backed ∆P+ Positive truncated component of ∆P ∆P– Negative truncated component of ∆P Mkt15 NASDAQ returns 15 days before IPO date Hot30 Average IR 30 days before IPO date UW HHI Underwriting HHI by industry 3,582 3,582 Solo Lead Dummy = 1 if single lead underwriter Table 3. Correlation Table Explanation: Pearson Correlation Coefficients are reported for IPOs issued in the US from 1984 to 1997 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). IR is the implied return from the IPO price to the after market trading price. ∆P is the implied return from the filing date midpoint to the IPO price and thus reflects the underwriter’s actual price adjustment. For a given IPO, UWpastIR is equal to the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding filing date of the given IPO. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s equity market share computed over the previous calendar year. Overhang is equal to the pre-IPO shares retained by the issuer divided by the shares filed, both primary and secondary. InvPrice is the reciprocal of the filing date midpoint. LogSize is the natural logarithm of the original filing amount. VC is a dummy that is equal to one if the firm is VC-backed. ∆P+ and ∆P– are the positive and negative truncated components of ∆P (explained above). Mkt15 is the cumulative NASDAQ return over the 15 trading days prior to the issue date. Hot30 is the average underpricing for all IPOs issued in the 30 day window preceeding the issue date. Variable ∆P Description IR UWpastIR CMrank UWshare Price Change Variables 1.000 0.537 Explanatory Variables 0.208 –0.002 0.076 0.165 0.127 –0.035 0.112 0.563 0.344 0.204 0.140 0.176 0.063 0.123 0.125 0.072 0.040 0.061 0.832 0.829 0.137 0.237 1.000 –0.163 0.019 0.049 0.230 –0.143 0.092 0.136 0.153 –0.004 –0.026 –0.163 1.000 0.476 0.177 –0.570 0.562 0.099 0.103 0.006 –0.005 0.034 0.019 0.476 1.000 0.228 –0.277 0.447 0.032 0.139 0.071 0.011 0.008 1.000 0.537 0.208 0.176 –0.002 0.063 0.076 0.123 IR Initial returns ∆P % Price adjustment from Pmid to Pipo 36 UWpastIR Lead underwriter’s past abnormal IR CMrank Lead underwriter’s Carter-Manaster rank UWshare Lead underwriter’s equity market share Overhang Shares retained / shares filed InvPrice One divided by the offer price LogSize Natural Logarithm of filing amount VC Dummy = 1 if venture capital backed ∆P+ Positive truncated component of ∆P ∆P– Negative truncated component of ∆P Mkt15 NASDAQ returns 15 days before IPO date Hot30 Average IR 30 days before IPO date Table 4. IPO Characteristics versus UW Reputation Quartiles Explanation: Mean characteristics are reported for IPOs issued in the US from 1984 to 1997 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). Within each calendar year, IPOs are grouped into quartiles based on the reputation of the lead underwriter. Reputation is measured using UWpastIR, CMrank, and UWshare for Panels A, B, and C respectively. The reported characteristics are equal weighted averages over all observations grouped into each quartile. IR is the implied return from the IPO price to the after market trading price. ∆P is the implied return from the filing date midpoint to the IPO price and thus reflects the underwriter’s actual price adjustment. ∆P is the percentage implied return from the filing date midpoint to the IPO price and thus reflects the underwriter’s actual price adjustment. For a given IPO, UWpastIR is equal to the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding filing date of the given IPO. UWpastIR is expressed as a percentage. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s percentage market share computed over the previous calendar year. Size is the original filing amount in millions. Spread is the underwriter’s percentage commission rate. Residual IR, Residual ∆P, and Residual Spread control each given variable for size and industry effects. Each is computed as the residuals from the following OLS regression equation, replacing “variable” with IR, ∆P, and Spread respectively: Variable = β1 LogSize + β2 LogSize2 + [Fama-French 48 industry fixed effects] + ˜ . IPO Pricing Variables Transaction Costs Underwriter Quality Measures 37 IR ∆P Panel A: UWpastIR Quartiles –5.3 –3.7 –0.4 9.4 –3.5 –1.5 4.4 8.4 –4.5 –2.8 1.4 5.9 6.8 6.9 6.9 7.1 –0.1 0.0 0.0 0.1 Panel B: CMrank Quartiles 0.8 –0.5 0.4 –0.7 –2.6 2.4 3.6 4.4 –1.4 0.2 1.1 0.1 7.4 7.0 6.8 6.6 0.1 –0.0 0.0 –0.1 Panel C: UWshare Quartiles 1.2 –4.9 –0.4 4.1 –2.3 –0.1 3.1 7.1 –1.2 –2.0 0.3 2.8 7.3 6.9 6.9 6.6 0.1 –0.1 –0.0 –0.0 6.5 7.9 8.4 8.7 6.2 8.1 8.4 8.9 8.0 7.8 8.0 7.8 Quartile Residual IR Residual ∆P Underwriter Spread Residual Underwriter Spread CMrank UWshare UWpastIR Size Observations Lowest UWpastIR Quartile 2 Quartile 3 Highest UWpastIR 15.8 17.5 24.4 32.9 3.3 4.3 5.7 5.8 –5.2 –2.0 1.8 5.7 283.7 304.2 331.5 392.0 982 987 1,001 980 Lowest CMrank Quartile 2 Quartile 3 Highest CMrank 16.0 22.9 24.7 27.0 0.9 3.0 5.9 9.3 –0.7 0.9 –0.1 0.2 114.9 206.1 374.0 615.9 982 988 996 984 Lowest UWshare Quartile 2 Quartile 3 Highest UWshare 16.6 18.1 24.2 31.8 0.3 1.8 4.4 12.7 –1.0 –0.3 0.5 1.2 136.1 223.8 349.4 603.2 988 967 1,022 973 Table 5. Fama-MacBeth Style Regressions Predicting Intial Returns Explanation: Fama-MacBeth style regression coefficients and T-statistics (in parentheses) are reported for various subsamples of the set IPOs issued in the US from 1984 to 2000 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). The main result that UWpastIR predicts future initial returns is also robust to using the Fama-MacBeth (1973) regression methodology instead of OLS. The dependent variable, initial return (IR), is the implied return from the IPO price to the after market trading price. For a given IPO, UWpastIR is equal to the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding filing date of the given IPO. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s equity market share computed over the previous calendar year. Overhang is equal to the pre-IPO shares retained by the issuer divided by the shares filed, both primary and secondary. InvPrice is the reciprocal of the filing date midpoint. LogSize is the natural logarithm of the original filing amount. VC is a dummy that is equal to one if the firm is VC-backed. ∆P+ and ∆P– are the positive and negative truncated components of ∆P, which is the implied return from the filing date midpoint to the IPO price (the underwriter’s price adjustment). Mkt15 is the cumulative NASDAQ return over the 15 trading days prior to the issue date. Hot30 is the average underpricing for all IPOs issued in the 30 day window preceeding the issue date. If industry effects are set to “Yes”, then industry fixed effects based on 48 Fama-French industry codes (see Ken French website) are included in the regression equation. The Fama-MacBeth method first computes annual cross-sectional regression coefficients for the IPOs issued in each calendar year. The displayed coefficients and T-statistics are computed from the time series average of the cross sectional coefficients. The R-Square is the average R-Square of the cross sectional regressions. Separate Fama-MacBeth regressions are computed for the observations in each group. In Panel D, the variables ∆P+ and ∆P– are residuals from the following OLS regressions, where “variable” is replaced by ∆P+ and ∆P– respectively: Variable = α + β1 UWpastIR + ˜ . Underwriter Quality Measures UWpastIR CMrank UWshare Panel A: IPO Characteristics known on the filing date 0.604 (6.15) 0.477 (2.45) 0.699 (7.84) 0.621 (3.59) 1.068 (1.53) 0.472 (3.01) 0.031 (3.71) –0.041 (–2.73) 0.748 (2.61) 1.180 (3.80) 0.380 (3.19) 0.030 (4.61) –0.034 (–3.34) 0.464 (3.84) –0.295 (–0.44) 0.038 (0.74) 0.002 (0.88) 0.004 (0.66) 0.331 (4.43) 0.015 (1.75) 0.013 (1.18) 0.034 (1.68) 0.548 (1.45) 0.233 (2.79) 0.018 (3.39) –0.018 (–2.17) 0.407 (5.37) 0.014 (1.98) Overhang LogSize InvPrice VC Mkt15 Hot30 Filing Date Control Variables Market and Price Variables ∆P+ ∆P– Industry REffects? Squared Observations Row Sample Period 38 0.572 (3.18) 0.971 (1.37) 0.582 (2.75) 0.030 (3.79) –0.042 (–2.45) 0.341 (2.11) 0.024 (1.34) 1.058 (5.18) 0.251 (1.79) 0.050 (0.08) 0.425 (2.75) 0.020 (3.64) –0.044 (–2.72) –0.459 (–1.27) 0.010 (0.82) 0.737 (4.22) Panel D: Use residual ∆P+ and ∆P– 1.037 (6.01) 0.050 (0.08) 0.425 (2.75) 0.020 (3.64) –0.044 (–2.72) –0.459 (–1.27) 0.010 (0.82) 0.737 (4.22) (1) 1984 to 1997 0.130 3,073 (2) 1984 to 1989 0.120 774 (3) 1990 to 1997 0.138 2,299 (4) All (1984-2000) 0.141 3,937 Panel B: Add Industry Affects and Market Variables 0.357 (4.88) Yes 0.359 3,937 (5) All (1984-2000) Panel C: Add Price Adjustment Variables ∆P+ and ∆P– 0.025 (0.30) 0.844 (8.06) 0.189 (4.79) Yes 0.548 3,937 (6) All (1984-2000) (7) All (1984-2000) 0.025 (0.30) 0.844 (8.06) 0.189 (4.79) Yes 0.548 3,937 Table 6. Fama-MacBeth Style Regressions Predicting Changes in Underwriter Market Share Using Past Abnormal Initial Returns (UWpastIR) Explanation: One observation is one underwriter’s issuing activity in one calendar year. The dependent variable is the underwriter’s percent change in market share (market share is based on the number of IPOs, consistent with Beatty and Ritter (1986)). For each observation, change in market share is equal to the underwriter’s market share in the given calendar year less his market share in the five year interval preceding the given calendar year. UWpastIR is the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. For a given underwriter in a given calendar year, this average includes all IPOs issued by the given underwriter in the 5 years preceding the given calendar year. The independent variables, “UWpastIR+” and “UWpastIR-” are the positive and negative components of UWpastIR respectively. The Fama-MacBeth regression methodology first computes cross sectional regression coefficients for the observations in each year. Reported coefficients are equal weighted averages of the yearly coefficients and the T-statistics are computed from the standard error of each average. R-Square is the average R-Square of the cross sectional regressions. The “number of observations” is equal to the number of underwriter-year observations included in the market share regression. The “Number of IPOs underwritten” is equal to the total number of IPOs underwritten by the underwriters included in the given size grouping (expressed as percentage in parentheses). The “Total Size of IPOs Underwritten” is the total gross proceeds of all IPOs underwritten by these underwriters (also expressed as a percentage in parentheses). An observation is grouped into “Large Underwriters” if the underwriter’s gross proceeds over the past 5 year period exceeds the median over the same period. Otherwise it is grouped into “Small Underwriters”. Separate Fama-MacBeth regressions are computed for the observations in each group. Regression Results UWpastIR+ Panel A: Large Underwriters Only 5.411 6.303 4.742 8.212 Panel B: Small Underwriters Only –2.138 –3.659 –0.997 –2.838 (–2.27) (–4.09) (–0.70) (–2.87) 5.022 7.045 3.504 5.469 (2.32) (1.76) (1.45) (2.85) 0.119 0.145 0.100 0.163 332 145 187 393 650 212 438 681 (21.7%) (26.2%) (20.0%) (19.1%) 49,882 11,570 38,311 53,248 (7.8%) (10.3%) (7.3%) (5.1%) (2.06) (1.23) (1.66) (2.32) 3.470 3.847 3.187 3.632 (4.09) (2.95) (2.69) (4.78) 0.096 0.087 0.102 0.115 338 148 190 401 2,347 (78.3%) 597 (73.8%) 1,750 (80.0%) 2,887 (80.9%) 587,385 100,728 486,658 987,131 (92.2%) (89.7%) (92.7%) (94.9%) UWpastIR− R-Square Number of Observations Number of IPOs Underwritten Total Dollars Underwritten Row Sample Period 39 Panel C: All Underwriters –1.259 –1.977 –0.721 –0.515 (–2.06) (–2.90) (–0.77) (–0.67) 3.669 5.192 2.526 4.122 (3.08) (2.40) (1.95) (3.78) 0.053 0.062 0.047 0.057 670 293 377 794 (1) (2) (3) (4) 1984 to 1997 1984 to 1989 1990 to 1997 All (1984-2000) (5) (6) (7) (8) 1984 to 1997 1984 to 1989 1990 to 1997 All (1984-2000) (9) (10) (11) (12) 1984 to 1997 1984 to 1989 1990 to 1997 All (1984-2000) 2,997 (100.0%) 809 (100.0%) 2,188 (100.0%) 3,568 (100.0%) 637,267 (100.0%) 112,298 (100.0%) 524,969 (100.0%) 1,040,379 (100.0%) Table 7. Fama-MacBeth Style Regressions Predicting the Frequency of Future Visits to the Reissue Market Explanation: Fama-MacBeth style regression coefficients and T-statistics (in parentheses) are reported for various subsamples of the set IPOs issued in the US from 1984 to 2000 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). The dependent variable is the natural logarithm of one plus the number of times a given firm visits the reissue market in the 5 year period after it’s IPO. I define a visit to the reissue market to include all transactions where a given firm issues new debt or equity in the 5 years after it’s IPO. For a given IPO, the independent variable UWpastIR is equal to the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding filing date of the given IPO. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s equity market share computed over the previous calendar year. InvPrice is the reciprocal of the filing date midpoint. LogSize is the natural logarithm of the original filing amount. VC is a dummy that is equal to one if the firm is VC-backed. “Number of Past Visits” is the average number of visits to the reissue market made by firms previously underwritten by the given lead underwriter in the past 5 calendar years. The Fama-MacBeth method first computes annual cross-sectional regression coefficients for the IPOs issued in each calendar year. The displayed coefficients and T-statistics are computed from the time series average of the cross sectional coefficients. The R-Square is the average R-Square of the cross sectional regressions. Separate Fama-MacBeth regressions are computed for the observations in each group. Underwriter Quality Measures UWpastIR Panel A –0.824 (–3.08) –1.051 (–2.59) –0.653 (–1.78) –0.732 (–3.23) 0.024 (2.68) 0.834 (2.87) Panel B 0.020 (1.84) 0.614 (2.48) 0.032 (1.64) 1.130 (1.48) 0.025 (2.49) 0.835 (2.42) CMrank UWshare LogSize InvPrice VC Issuer Characteristics Number of Past Visits R-Square Number of Observations Row Sample Period 40 –0.548 (–2.52) –0.591 (–2.86) –0.458 (–1.67) –0.008 (–0.73) –0.009 (–0.77) –0.003 (–0.29) 0.444 (1.39) 0.405 (1.25) 0.342 (0.99) 0.099 (5.45) 0.102 (5.31) 0.105 (5.25) –0.036 (–0.09) 0.071 (0.20) (1) 1984 to 1997 0.033 2,959 (2) 1984 to 1989 0.045 743 (3) 1990 to 1997 0.024 2,216 (4) All (1984-2000) 0.035 3,768 (5) All (1984-2000) 0.067 0.079 0.103 (0.88) 0.091 3,768 3,768 3,768 (6) All (1984-2000) (7) All (1984-2000) 0.012 (0.50) 0.013 (0.55) Table 8. Fama-MacBeth Style Regressions Predicting Initial Returns using Measures of Concentration Explanation: Fama-MacBeth style regression coefficients and T-statistics (in parentheses) are reported for various subsamples of the set IPOs issued in the US from 1984 to 2000 excluding: firms with an issue price less than five dollars, ADRs, financial firms, REITs, and IPOs underwritten by an underwriter that has issued fewer than 10 IPOs since 1980 (observations from 1980 to 1983 are used only to compute stable starting values for UWpastIR). The dependent variable, initial return (IR), is the implied return from the IPO price to the after market trading price. For a given IPO, the independent variable “Industry Underwriting HHI” is the prevailing industry-by-industry Herfindahl index, which is computed using the most recent 15 IPOs completed in a given industry using the 48 Fama-French industries before a given IPO’s filing date. Industry Underwriting HHI is the sum of squared market shares (based on number of transactions) of the lead underwriters who completed the 15 IPOs. “Solo Lead Underwriter” is equal to 1 if one lead underwriter manages the given IPO and is zero otherwise. This measure proxies for concentration because IPOs with fewer lead underwriters were likely underwritten in a more concentrated underwriting environment. UWpastIR is the lead underwriter’s average past abnormal initial returns. Abnormal initial returns are actual initial returns less market-wide average initial returns in the same month. This average includes all IPOs issued by the given underwriter in the 5 years preceding filing date of the given IPO. CMrank is the Carter/Manaster rank as listed in Carter, Dark and Singh (1998). UWshare is the lead underwriter’s equity market share computed over the previous calendar year. InvPrice is the reciprocal of the filing date midpoint. LogSize is the natural logarithm of the original filing amount. VC is a dummy that is equal to one if the firm is VC-backed. The Fama-MacBeth method first computes annual cross-sectional regression coefficients for the IPOs issued in each calendar year. The displayed coefficients and T-statistics are computed from the time series average of the cross sectional coefficients. The R-Square is the average R-Square of the cross sectional regressions. Separate Fama-MacBeth regressions are computed for the observations in each group. Concentration Measures Underwriting HHI UWpastIR Panel A CMrank UWshare LogSize Solo Lead Underwriter Underwriter Quality Measures Issuer Characteristics InvPrice VC R-Square Number of Observations Row Sample Period 41 0.286 (1.99) –0.061 (–0.44) 0.546 (2.94) 0.853 (2.09) Panel B 0.028 (1.98) 0.006 (0.33) 0.045 (2.30) 0.045 (2.05) 0.046 (2.06) 0.048 (2.25) 0.971 (2.01) 0.869 (2.07) 0.870 (2.09) 0.045 (2.12) 0.802 (5.65) 0.657 (4.46) 0.654 (4.40) 0.654 (4.01) 0.855 (1.03) 1.441 (1.88) 1.468 (1.90) 1.385 (1.89) 0.593 (3.19) 0.537 (3.06) 0.543 (3.10) 0.504 (3.17) (1) 1984 to 1997 0.010 0.011 0.009 0.012 2,783 636 2,147 3,582 (2) 1984 to 1989 (3) 1990 to 1997 (4) All (1984-2000) (5) All (1984-2000) 0.088 0.116 0.112 0.107 3,582 3,582 3,582 3,582 (6) All (1984-2000) (7) All (1984-2000) (8) All (1984-2000) –0.020 (–2.03) 0.008 (1.31) 0.005 (0.95) 0.009 (1.53) 1.771 (3.57) 1.830 (3.50) 1.824 (3.45) 0.048 (1.86) 0.048 (1.89) 0.050 (1.89) Table 9. States and Corresponding Probabilities State Description {˜ H , ˜n , ˜p } η State s1 {1,1,1}, State s4 {1,-1,-1}, State s7 {1,0,-1}, State s10 {0,0,1}, State s13 {0,-1,-1}, State s16 {0,1,-1}, State s19 {-1,-1,1}, State s22 {-1,0,0}, State s25 {-1,1,0}, State s2 {1,1,0}, State s5 {1,0,0}, State s8 {1,-1,1}, State s11 {0,0,0}, State s14 {0,1,1}, State s17 {0,-1,0}, State s20 {-1,-1,0}, State s23 {-1,1,1}, State s26 {-1,0,-1}, State s3 {1,1,-1} State s6 {1,-1,0}, State s9 {1,0,1} State s12 {0,0,-1} State s15 {0,-1,1}, State s18 {0,1,0}, State s21 {-1,-1,-1} State s24 {-1,0,1}, State s27 {-1,1,-1} λ 9 λ 9 State Probability λ 9 + 1−λ 27 1−λ 27 + 1−λ 27 1−λ 27 + 1−λ 27 1−λ 27 42 Figure 1. Timeline of the Strategic Underwriting Game Period 1 begins 1a All agents observe V and the underwriter types H and L. H observes the signal η˜ . H 1b The entrepreneur determines his supply curve QE [Pmid ]. QE [Pmid ] becomes public information. 1c H and L simultaneously select QH and QL . Filing Date (period 1 ends) Period 2 begins 1d Pmid is determined by equating supply QE and demand QH + QL . 2a All agents observe ˜ . Only the p underwriters (H and L) observe ˜ . n 2b H and L select a price adjustment ∆P . 2c The government tests ∆P for compliance. IPO date (period 2 ends) 2d Profits are realized. e¡ 34 Figure 2 The figure depicts scatter plots of abnormal initial returns versus abnormal price adjustment for Morgan Stanley (172 IPOs) and Lehman Brothers (135 IPOs) from 1984 to 1997. Both underwriters have the highest Carter/Manaster rank of 9. An IPO's abnormal initial returns and abnormal price adjustment are equal to their raw values less market wide averages in the month the IPO was completed. This adjustment eliminates components of each that arise from public information. Each underwriter's UWpastIR is the equal weighted average of their past abnormal initial returns over the entire sample. Points denoted by black diamonds (gray squares) have negative (positive) abnormal initial returns. If underwriters can extract profits from their investors via quid pro quo arrangements, then the gray squares represent more profitable IPOs. 28% (61%) of Morgan Stanley's IPOs are to the left (right) of the vertical bar placed at zero abnormal price adjustment. 63% (28%) of Lehman's IPOs are to the left (right) of the vertical bar. IPOs to the left (right) are ex-ante overvalued (undervalued). Morgan Stanley (UWpastIR= +6.7%) Abnormal Initial Returns (IR) 1 0.75 0.5 0.25 0 -0.25 -0.5 -0.5 28% revised down 61% revised up -0.25 0 0.25 0.5 0.75 172 IPOs Abnormal Price Adjustment (∆P) Lehman Brothers (UWpastIR= -6.2%) Abnormal Initial Returns (IR) 1 0.75 0.5 0.25 0 -0.25 -0.5 -0.5 -0.25 0 0.25 0.5 0.75 135 IPOs 63% revised down 28% revised up Abnormal Price Adjustment (∆P)