Why Don t Issuers Get Upset About Leaving Money on the Table in IPOs?

Transcription

1 Why Don t Issuers Get Upset About Leavng Money on the Table n IPOs? Tm Loughran Unversty of Notre Dame P.O. Box 399 Notre Dame IN voce Loughran.9@nd.edu and Jay R. Rtter Unversty of Florda P.O. Box 768 Ganesvlle FL voce rtter@dale.cba.ufl.edu October 8, 00 Abstract One of the puzzles regardng ntal publc offerngs (IPOs) s that ssuers rarely get upset about leavng substantal amounts of money on the table, defned as the number of shares sold tmes the dfference between the frst-day closng prce and the offer prce. The average IPO leaves $9. mllon on the table. Ths number s approxmately twce as large as the fees pad to nvestment bankers, and represents a substantal ndrect cost to the ssung frm. We present a prospect theory model that focuses on the covarance of the money left on the table and wealth changes. Our reasonng also provdes an explanaton for a second puzzlng pattern: much more money s left on the table followng recent rses than after falls. Ths results n an explanaton of hot ssue s. We also offer a new explanaton for why IPOs are underprced. Keywords: Intal publc offerngs; Prospect theory; Behavoral fnance; Hot ssue s We wsh to thank Hsuan-Ch Chen and Donghang Zhang for useful research assstance. Hsuan- Ch Chen, Kent Danel, Harry DeAngelo, Crag Dunbar, Bruce Foerster, Jason Karcesk, Robert Korajczyk, Alexander Ljungqvst, M. Nmalendran, Mchael Ryngaert, Hersh Shefrn, Wllam Wlhelm, Kent Womack, L-Anne Woo, and an anonymous referee have provded useful comments, as have semnar partcpants at the Australasan consortum of unverstes vdeoconference, DePaul, Cornell, Georgetown, and Oho State Unverstes, the Unverstes of Florda, Georga, and Kansas, the ABN Amro Conference on IPOs n Amsterdam, the 8 th Pacfc Basn Accountng, Economcs, and Fnance Conference n Bangkok, the NYSE Equty Markets n Transton Conference n Scottsdale, the Socety of Fnancal Studes Conference at Northwestern, and the September 00 Chcago Quanttatve Allance Conference.

2 Why Don t Issuers Get Upset About Leavng Money on the Table n IPOs?. Introducton Durng , companes gong publc n the U.S. left over $27 bllon of money on the table, where the money left on the table s defned as the frst day prce gan multpled by the number of shares sold. If the shares had been sold at the closng prce rather than the offer prce, the proceeds of the offerng would have been hgher by an amount equal to the money left on the table. Alternatvely, the same proceeds could have been rased by sellng fewer shares, resultng n less dluton of the pre-ssue shareholders. The nvestors profts come out of the pocket of the ssung company and ts pre-ssue shareholders. The $27 bllon left on the table s twce as large as the $3 bllon n nvestment banker fees pad by the ssung companes that we study. These same companes generated profts of approxmately $8 bllon n the year before gong publc, so the amount of money left on the table represents more than three years of aggregate profts. In some cases, the numbers are extreme: In Netscape s August 995 ntal publc offerng (IPO) wth Morgan Stanley as the lead underwrter, 5 mllon shares were sold to nvestors at $28.00 per share. Wth a closng prce of $58.25, $5 mllon was left on the table. Yet, n spte of ths huge wealth transfer from Netscape s pre-ssue shareholders to those lucky enough to have been allocated shares at the offer prce, Netscape s major shareholders were satsfed wth the prcng of the offerng. (Most companes gong publc n the U.S. are relatvely young frms wth large blocks of equty owned by the managers. Thus, throughout most of ths paper we wll use the terms ssung company and pre-ssue shareholders nterchangeably.) Netscape retaned Morgan Stanley as the lead underwrter for the November 996 follow-on offerng. And ths reacton s not unusual. Krgman, Shaw, and Womack (0) report that ssung frms do not vew large amounts of money left on the table as an mportant consderaton n choosng underwrters for a follow-on offerng. They report that for 5 IPOs wth frst-day returns n excess of 60% that subsequently conducted follow-on offerngs, all 5 retaned the lead underwrter from the IPO. The fact that ssuers rarely complan about leavng large amounts of money on the table has long puzzled fnancal economsts. As Brealey and Myers state on p. 389 of the ffth edton

3 2 of ther textbook, after dscussng an IPO that trpled n value on ts frst day of tradng, Contentment at sellng an artcle for one-thrd of ts subsequent value s a rare qualty. Why don t ssuers get upset about leavng money on the table? We propose a prospect theory answer to ths queston. Prospect theory assumes that ssuers care about the change n ther wealth, rather than the level of wealth. Prospect theory predcts that, n most stuatons occurrng n the IPO, ssuers wll sum the wealth loss from leavng money on the table wth the larger wealth gan from a prce jump, producng a net ncrease n wealth for pre-ssue shareholders. Emprcally, we show that most of the money left on the table comes from a mnorty of IPOs. Indeed, although the average amount left on the table s $9. mllon, the medan s only $2.3 mllon. The IPOs leavng a lot of money on the table are those where the offer prce s revsed upwards from what had been antcpated at the tme of dstrbutng the prelmnary prospectus. The offer prce s ncreased n response to ndcatons of strong demand, but t could have been ncreased even further. Thus, at the same tme that underprcng s dlutng the pressue shareholders of these frms, these shareholders are recevng the good news that ther wealth s much hgher than they had antcpated. Our model s bult upon ths covarance of the amount of money left on the table and unantcpated wealth changes. An example wll llustrate the argument. In Netscapes IPO, James Clark, a company cofounder, held 9.34 mllon shares. Approxmately one month before gong publc, Netscape fled a prelmnary prospectus wth the Securtes and Exchange Commsson (S.E.C.). Ths prospectus contaned a projected number of shares to be ssued and an antcpated prce range for the offerng. Based upon the mdpont of the fle prce range of $2-4, the expected value of hs Netscape holdngs equaled $2 mllon at the tme that the prelmnary prospectus was fled. At the closng prce on the frst day of tradng, hs shares were worth $544 mllon, a 350% ncrease n ths component of hs pre-tax wealth n the course of a few weeks. So at the same tme that he dscovered that he had been dluted more than necessary due to the large amount of money left on the table, he dscovered that hs wealth had ncreased by hundreds of mllons of dollars. Would many people be upset f they found themselves n ths stuaton? Most of the tme that there s a large stock prce runup, the offer prce has been ncreased above the fle prce range. The emprcal pattern that the frst day return s related to the revson

4 3 n the offer prce was frst documented by Hanley (993). IPOs where the offer prce s revsed upwards see much hgher frst-day prce jumps, on average, than those where the offer prce s revsed downwards. The magntude of the dfference s large: ssues where the fnal offer prce s below the mnmum of the fle prce range have average frst-day returns of 4%, whereas those that are prced above the maxmum of the fle prce range have average frst-day returns of 32%. The receved wsdom among academcs for why ths partal adjustment exsts s based upon the Benvenste and Spndt (989) model of IPO underprcng. Ths dynamc nformaton acquston model argues that regular nvestors, n order to truthfully reveal ther demand to an underwrter durng the book-buldng phase of an IPO s ng, must be rewarded wth more underprcng on deals for whch there s strong demand. Thus, deals n whch the offer prce s revsed upwards wll have greater underprcng. The Benvenste and Spndt model predcts that there should be partal adjustment of the offer prce wth respect to prvate nformaton. It does not predct that there should be partal adjustment to publc nformaton, such as recent movements that are readly observable to all partes. Yet, as we show, the frst-day returns on IPOs are predctable based upon movements n the three weeks pror to ssue. The quanttatve effect s large: each % ncrease n the durng the three weeks before ssue results n a frst-day return that s.3% hgher (.3% rather than 0.0%, for example). Because there s partal adjustment to publc nformaton, frst-day returns are predctable. Followng a rse, IPOs that were n the pre-sellng perod wll have hgher than average expected frst-day returns. Smlarly, when there s a fall, IPOs that come to n the next few weeks wll have low expected frst-day returns. Our prospect theory model does not dstngush between publc and prvate nformaton, and thus predcts that average frst-day returns wll be predctable based upon publc nformaton. The prospect theory explanaton of condtonal underprcng thus leads to a theory of hot ssue s. As many authors have documented, IPOs are underprced, on average. We argue that ths underprcng s a form of ndrect compensaton to underwrters. At frst glance, leavng money on the table comes at a cost to underwrters. Snce the percentage gross spread s typcally negotated before the fnal offer prce s establshed, rasng the offer prce ncreases the revenues of underwrters. Then why do underwrters choose a lower offer prce, and n so dong leave more money on the table? Investment bankers beneft n two ways. Frst, t makes t easer to

5 4 fnd buyers for IPOs, reducng ther ng costs (Baron (982)). Second, nvestors wll engage n rent-seekng behavor to mprove ther prorty for beng allocated shares n hot IPOs. Among other thngs, they do ths by tradng wth the brokerage arm of the underwrters and overpayng for commssons. Ths rent-seekng behavor on the part of potental IPO nvestors ncreases the revenues of the underwrter beyond that measured when focusng exclusvely on the gross spread. Although underwrters want to leave money on the table, t s unlkely that they are able to gan $ n pre-tax profts from qud pro quos for every $ left on the table. Thus, there s some leakage, n that underwrters beneft more from a $ ncrease n drect fees than from leavng $ more on the table, even though the effect on the net proceeds receved by the ssuers s the same. Consequently, we are tryng to explan why underwrters prefer to use a relatvely neffcent mechansm to collect revenue. We argue that underwrters are able to acheve hgher revenue f part of ther compensaton s less transparent than f t all was n the form of gross spreads. Ths s because ssuers treat the opportunty cost of leavng money on the table as less mportant than the drect fees (see Thaler (980)). Our prospect theory explanaton can be recast n terms of a barganng model n whch underwrters want a lower offer prce and ssung frms desre a hgher offer prce. 2 We are argung that when unexpectedly strong demand becomes apparent durng the pre-sellng perod, ssung frms acquesce n leavng more money on the table. When demand s unexpectedly weak, ssung frms negotate more aggressvely, leavng lttle money on the table. The structure of the rest of ths artcle s as follows. Secton 2 descrbes the data used. Secton 3 descrbes the partal adjustment phenomenon. Secton 4 provdes a prospect theory explanaton for why ssuers aren t upset. Secton 5 presents an explanaton of why underwrters ntentonally underprce IPOs, on average. Secton 6 addresses the queston of why offer prces In a Wall Street Journal artcle on an S.E.C. probe of mutual funds overpayng on commssons, Lucchett (999) states Other fund executves pont out that hgher commssons can be justfed by the access they can provde to ntal publc stock offerngs. 2 The objectve functon of the nvestment banker can be vewed as beneftng from underprcng, provded t s not carred to an extreme (see Beatty and Rtter (986) and Dunbar (00)). If certan condtons are satsfed, models wth ntertemporal maxmzaton problems predct that some ssuers may want to ntentonally underprce. These models, however, do not account for dynamc prce adjustment between the fle prce range and the fnal offer prce.

6 5 do not make a full adjustment to movements n between the settng of the fle prce range and the fnal offer prce. Secton 7 develops the mplcatons of our model for the predctablty of frst-day returns, and provdes a theory of hot ssue s. Secton 8 provdes a summary and conclusons. 2. Data Our sample s composed of 3,025 IPOs lsted by Securtes Data Co. from 990 to 998 that meet several crtera. To reduce the nfluence of mcrocap stocks, we exclude all IPOs that had a mdpont of the fle prce range of below $8.00 per share. Unt offerngs are excluded, as are closed-end funds, Real Estate Investment Trusts (REITs), partnershps, and Amercan Depostory Recepts (ADRs). The frst closng prce s taken from the Center for Research n Securtes Prces (CRSP), as s the post-ssue number of shares outstandng. Table reports the number of IPOs n our sample for each year from 990 to 998, along wth the average frst-day return. The average frst-day return, measured from the offer prce to the closng prce, s 4% for our sample frms. Throughout the paper, we do not adjust for movements n reportng frst-day returns. Ths s because movements are small n comparson (an average of 0.05% per day), and thus have lttle mpact on the conclusons. The average frst-day return of 4% s n lne wth the underprcng reported n other studes. The last three columns of Table report the proporton of IPOs prced below, wthn, and above the ntal fle prce range. Durng our sample perod, approxmately one-quarter are prced below, one-half wthn, and one-quarter above the fle prce range. Snce our sample ncludes only completed deals, wthdrawn offerngs, most of whch would have been prced below the mnmum, are not reflected n the proportons. Fgure presents a hstogram of the frst-day returns. Durng , less than.0% of IPOs doubled n prce on the frst day (the proporton has been much hgher after 998). Sxteen percent of IPOs close the frst day at the offer prce, a feature wdely attrbuted to stablzaton actvtes on the part of underwrters [see Hanley, Kumar, and Segun (993), Jenknson and Ljungqvst (996, p. 76), and Aggarwal (00)].

7 6 3. The partal adjustment phenomenon Table 2 llustrates the partal adjustment phenomenon, frst documented by Hanley (993). In ths table, we categorze IPOs on the bass of the fnal offer prce relatve to the orgnal fle prce range. Table 2 reports that the average IPO left $9. mllon on the table, a number that works out to an aggregate of over $27 bllon n For those IPOs prced below the fle range, the mean amount s $.5 mllon, and the medan amount s only $0.2 mllon. For those prced wthn the fle range, the mean s $6.4 mllon, and the medan s $2. mllon. For those prced above the fle range, the mean s $23.0 mllon, and the medan s $2.7 mllon. Over 60% of the amount of money left on the table s from the 24% of IPOs prced above the fle range. Ths skewness n the dstrbuton of the amount of money left on the table suggests that ths should be a bg concern for only some ssuers. 4 The last column of Table 2 reports the revaluatons of the pre-ssue equtyholders stakes durng the IPO process. Ths represents the change n ther wealth durng the nterval between when the fle prce range s set and the close of tradng on the day of ssue. We compute the revaluaton as the number of shares retaned by pre-ssue shareholders multpled by the change n the value per share from the mdpont of the flng prce range to the closng prce on the frst day of tradng, plus the shares sold by sellng shareholders n the IPO multpled by the change n the value per share from the mdpont of the fle prce range to the offer prce. 5 On 3 In all of our calculatons, we do not take nto account the exercse of overallotment optons. The reason for ths excluson s the unrelablty of ths nformaton n the SDC new ssues database. Because a 5% overallotment opton s present n almost all IPOs, and the overallotment opton s much more lkely to be exercsed n full when there s strong demand, ncludng nformaton on the exercse of overallotment optons would strengthen all of our results. 4 Frequently the number of shares s revsed n the same drecton as the prce. Logue, Rogalsk, Seward, and Foster-Johnson (00, Table ) report the average change n the number of shares offered, usng the same three classfcatons for revsons n the offer prce that we use. For IPOs where the offer prce s below the mnmum of the fle prce range, the average revson n the number of shares offered s -4%. For IPOs prced wthn the fle range, the number of shares s revsed upwards by 5%, and for deals prced above the maxmum, the number of shares s revsed upwards by an average of 9%. 5 As an example, Csco Systems, a February 990 IPO, had 9.5 mllon shares retaned by pre-ssue shareholders, 2.43 mllon newly ssued shares, and 0.37 mllon secondary shares n ts offerng at $8.00 per share. The frst closng prce was $22.25, and the mdpont of the fle prce range was $4.50. So for the 9.5 mllon shares retaned, the revaluaton of $7.75 per share resulted n a wealth gan of $73.6 mllon. For the 0.37 mllon shares sold by exstng shareholders, the revaluaton of $3.50 per share resulted n a wealth gan of $.3 mllon. Thus, the total revaluaton was $74.9 mllon for the pre-ssue shareholders, as contrasted wth the $.9 mllon left on the table.

8 7 average, ths revaluaton s postve, reflectng the underprcng phenomenon. But mportantly, ths revaluaton covares wth the amount of money left on the table. For upward revsons, where on average $23.0 mllon s left on the table, ths revaluaton averages $3.4 mllon. For downward revsons, where on average only $.5 mllon s left on the table, the revaluaton s a negatve $26.4 mllon. In Fgure 2, we present hstograms of the frst-day return dstrbutons, condtonal on the revson n offer prce relatve to the orgnal fle prce range. Inspecton of these hstograms shows that the dstrbutons are very dfferent. Condtonal upon a downward revson n the offer prce, the chance of a large frst-day return s qute low. Condtonal upon an upward revson, the probablty of a hgh frst-day return s qute hgh. 4. Prospect theory and ssuer complacency Let s return to the example of Netscape s IPO, where co-founder James Clark held 9.34 mllon shares. Based upon the mdpont of the fle prce range of $2-4, the expected value of hs Netscape holdngs was $2 mllon at the tme that the prelmnary prospectus was fled. At the closng prce on the frst day of tradng, hs shares were worth $544 mllon, a 350% ncrease n hs pre-tax wealth n the course of a few weeks. So at the same tme that he dscovered that he had been dluted more than necessary due to the large amount of money left on the table, he dscovered that hs wealth had ncreased by hundreds of mllons of dollars. Snce he owned 28.2% of the company before gong publc, $43 mllon of the $5 mllon wealth transfer from pre-ssue shareholders to new nvestors came out of hs pocket. After the offerng, he owned 24.5% of Netscape, but f the same proceeds had been rased by sellng 2.4 mllon shares at $58.25 nstead of 5.0 mllon shares at $28.00, he would have owned 26.3%. Suppose nstead that the offer prce had been revsed downwards to $6.00 per share, and then jumped to $2.50. Ths 08% frst-day return s the same percentage ncrease as the actual jump from $28.00 to $ But at a $2.50 prce, hs 9.34 mllon shares would be worth $7 mllon, about the same wealth as he had been expectng a few weeks earler. In ths scenaro, $32.5 mllon would have been left on the table, of whch $9 mllon would have been hs. Now he should be mad: he has been dluted, and there s no offsettng good news. In fact, hs $7 mllon ex post holdng of Netscape s below the $2 mllon he had expected a few

9 8 weeks earler. We conjecture that he would be much more upset at the nvestment bankers for leavng $32.5 mllon on the table n ths scenaro than he was when $5 mllon was actually left on the table, but accompaned by the good news that hs wealth had ncreased by 350% n a matter of weeks. Ths ntutvely plausble feelng can be formalzed. Prospect theory [Kahneman and Tversky (979), Shefrn and Statman (984)] s a descrptve theory of choce under uncertanty, and s not based upon normatve postulates about the way people should behave. Each ndvdual has a value functon, whch s smlar to a utlty functon, but the value functon s defned n terms of gans and losses, rather than levels. The value functon s llustrated n Fgure 3. The value functon s concave n gans and convex n losses, wth the functon beng steeper for small losses than for small gans (loss averson). In the context of gong publc, gans and losses are computed relatve to the prce that the executves of the ssung frm have anchored on. We argue that ths reference pont s the mdpont of the fle prce range, rather than the hstorcal cost (snce n many cases executves have receved ther shares as sweat equty, n leu of cash compensaton). 6 Prospect theory argues that when an ndvdual s faced wth two related outcomes, the ndvdual can ether treat them separately or as one. Fgure 4 llustrates the regons for whch segregaton and ntegraton of the outcomes wll occur. Due to the concavty of the value functon for gans, two gans wll be segregated. Ths s because a person feels better about two gans than about one gan of twce the sze. Because of the convexty of the value functon for losses, two related losses wll be ntegrated. But for a gan and a loss, whether the ndvdual feels better by ntegratng or segregatng the events depends upon ther magntudes. The lnes demarcatng the ntegrate and segregate regons of the second and fourth quadrants n Fgure 4 are below the 45 degree lne because of the loss averson feature of prospect theory. By 6 The June 29, 999 IPO of e-loan s descrbed n the cover story of the Sept. 6, 999 ssue of Busness Week [Hof (999)]. In the artcle, company presdent Janna Pawlowsk boycotts the pre-ipo dnner sponsored by the lead underwrter, Goldman Sachs, because she and the other members of the management team are upset wth Goldman for choosng a $4 offer prce. In the face of strong demand (the offer was oversubscrbed 26 tmes) followng the roadshow, the ssuers had argued for a $6 offer prce at the prcng meetng. The orgnal fle prce range was $ to $3, so ths seems to be nconsstent wth our prospect theory explanaton. But the key s that management anchored not on the $2 mdpont of the fle prce range, but nstead on $6, the prce that Softbank had been wllng to pay n a proposed prvate equty nfuson shortly before the IPO. The closng prce on the day of ssue was $37 per share, leavng $80 mllon on the table.

10 9 ntegratng the bad news (a small amount of dluton) wth the good news (a hgh net worth ncrease), the stockholders can feel good about the net gan. In Fgure 4, ssues that are underprced followng an upward revson n the offer prce wll be n the lower rght-hand quadrant. In general, the wealth gan for pre-ssue shareholder from the revaluaton s greater than hs or her share of the money left on the table when the followng condton s met: [shares retaned + secondary shares sold ][OP mdpont] + shares retaned [P - OP] > [P - OP][secondary shares sold + prmary shares sold(shares retaned /shares retaned)] where P s the prce, OP s the offer prce, prmary shares sold are beng sold by the frm, secondary shares sold are exstng shares beng sold by shareholder, and the shares retaned wthout a subscrpt are for all shareholders combned (see Barry (989) for related work). If ths condton s met, a pre-ssue shareholder wll fnd oneself n the ntegraton regon of the lower rght-hand quadrant of Fgure 4. For ssuers wth an upward revson of the offer prce and underprcng, ths condton wll be met unless shareholder s sellng a large number of shares n the offerng or the offerng s huge relatve to the pre-ssue number of shares. In practce, young companes gong publc rarely have any secondary shares n the IPO. We have provded a cogntve psychology argument for why some ssuers wll not be greatly upset wth leavng money on the table n IPOs. The key element s the covarance of money left on the table and wealth gans accrung to the ssuer. Ths s an example of the mportance of framng. If ssuers vewed the opportunty cost of underprcng by tself, ssuers would be more resstent to severe underprcng. But because t comes as part of a package that ncludes the good news of an ncrease n wealth, there s much less resstance. We do not clam that ths condtonal underprcng s an optmal contract among the class of all possble contracts. Indeed, our suspcon s that book-buldng s favored by underwrters partly because t allows them to take advantage of rsk-averse ssung frms. The road show perod mmedately before an IPO s a hgh-stress perod for ssung frms. The terms of the offerng are subject to substantal revsons, and there s a non-trval chance that the offerng may be completely cancelled due to forces outsde of management s control, such as a sharp drop. Thus,

11 there s a sense of relef wth a completed offerng, especally f the proceeds are hgher than had been expected. And the meda assocates a large prce jump wth a successful IPO Money on the table as ndrect underwrter compensaton Many reasons have been gven for why underwrters underprce IPOs, on average (see Jenknson and Ljungqvst (996) and Ibbotson, Sndelar, and Rtter (994) for surveys). In general, these reasons are not mutually exclusve. Underprcng comes at a cost to the underwrters, n that hgher gross proceeds from a hgher offer prce would result n hgher fees, gven that the percentage gross spread typcally does not change whether the proceeds are revsed upward or downward (Chen and Rtter (00)). We are focusng on why, condtonal upon strong demand becomng apparent durng the sellng perod, underwrters ntentonally underprce some IPOs severely. Underprcng does make t easer to an IPO, but the beneft of reduced ng efforts would seem to be mnmal once an offerng s oversubscrbed by a factor of ten or more. By usng underprcng as well as gross spreads, underwrters are able to get ssuers to pay much hgher average total costs than f all of the costs were mpounded nto drect fees for two reasons. The frst reason s that ssuers treat opportunty costs (the money left on the table) as less of a cost than the drect cost of the gross spread. 7 The second reason s that the amount of money left on the table s state-contngent. Even though underwrters may be able to capture only a fracton of the money left on the table n the form of qud pro quos, they are able to get hgher total compensaton than f all of ther compensaton was n drect fees. 8 The necessary condton for ths to be true s that the percentage qud pro quo per dollar left on the table exceeds the percentage gross spread that would be receved on an extra dollar of gross proceeds. 7 Anecdotal evdence regardng the relatve mportance of drect costs versus opportunty costs s contaned n Uttal s (986) descrpton of the prcng meetng for Mcrosoft s 986 IPO. At the prcng meetng, the lead underwrter, Goldman Sachs, nformed Mcrosoft that the IPO would probably start tradng at $25 or more, well above the $6-9 fle prce range. Uttal descrbes the negotaton of the offer prce n ncrements of a dollar per share, and of the gross spread n ncrements of a penny per share: Mcrosoft and Goldman Sachs had no trouble agreeng on a fnal prce of $2 Havng agreed farly easly over dollars, the two sdes bogged down over pennes. Mcrosoft and ts sellng shareholders pad $3.66 mllon n drect fees and left $8.87 mllon on the table, wth the stock prce closng at $27.75 on the frst day of tradng. 8 Habb and Ljungqvst (00), however, present emprcal evdence that the opportunty cost of underprcng has a one-to-one margnal tradeoff wth drect expenses.

12 We are not argung that underprcng s necessary because the drect fees are nsuffcent to compensate underwrters for ther efforts. Instead, we are argung that wth the ndrect benefts of underprcng, underwrtng IPOs s exceptonally proftable for nvestment bankers. Why, then, doesn t entry from competng nvestment bankers erode these profts? We beleve that there are sgnfcant barrers to entry, largely due to the perceved mportance of coverage by nfluental analysts. The mportance of analyst coverage allows the hgh-prestge nvestment bankers to attract ssuers n spte of leavng large amounts of money on the table. Although opportunty costs are not vewed as equvalent to drect costs by ssuers, opportunty costs do matter. In general, the smaller s the offerng, the less s the dluton assocated wth a gven frst-day return. Thus, our analyss predcts that IPOs sellng a larger percentage of the frm, and wth more secondary shares, should have less underprcng. Habb and Ljungqvst (00) make a smlar predcton, and present supportng emprcal evdence. Lastly, t should be noted that nether the drect cost of the gross spread nor the ndrect cost of money on the table shows up on an ssuer s ncome statement. The net proceeds appears on the balance sheet, but nether the drect nor ndrect costs drectly affect the ncome statement. Whle most academcs may fnd ths accountng treatment rrelevant, our suspcon s that f these costs appeared on the ncome statement, managers would pay more attenton to mnmzng these costs. 6. Why s there partal adjustment to -wde nformaton? 6.a. Emprcal patterns In Table 3, we categorze IPOs by the movement n the three weeks (5 tradng days) pror to ssue. 9 We report that followng declnes, the average frst-day return s 0.0%, whereas after perods when the has rsen by at least 2.0%, the average frst-day 9 Durng January 997-August 999, the length of tme between flng a regstraton statement (form S- for most IPOs) wth the S.E.C. and gong publc has averaged 78 days, for IPOs where ths length of tme took less than 80 calendar days (takng more than sx months ndcates that the orgnal plans were postponed at some pont). Typcally, 4-5 weeks before the expected offer date, the ssuer fles an amended regstraton statement S-/A gvng a fle prce range. Ths prelmnary prospectus (also known as a red herrng, due to the red letters on the cover page dentfyng t as prelmnary) s then dstrbuted to potental nvestors. The frm may fle further amendments as well. The S.E.C. mandates that the fnal offer prce must be wthn a range of % below the mnmum to % above the maxmum of the latest fle prce range. Our use of 5 tradng days (three weeks) s a mnmal measure of the length of tme between the settng of the orgnal fle prce range and the offer date.

13 return s 8.5%. The ablty of recent movements to predct frst-day returns has been documented for decades [Logue (973, Table ), Hanley (993, Table 3), Benvenste, Wlhelm, and Yu (00, Table 3)]. 0 Thus, underwrters do not fully adjust the offer prce wth respect to publc nformaton. Indeed, the evdence suggests that there s very lttle adjustment of the offer prce wth respect to movements. The dfference n average value-weghted returns between our top and bottom categores n Table 3 s 5.52%. The dfference n average frst-day returns s 8.47%. Followng rses, ssuers leave more than twce as much money on the table as followng declnes ($2.7 mllon versus $5.6 mllon). In Fgure 5, we sort IPOs by both the revson n the offer prce from the fle prce range, and by pror movements. Inspecton of Fgure 5 dscloses that, condtonal on revsons, lagged returns predct frst-day returns. Whle we have shown n Table 2 that frst-day returns are predctable usng the revson n the offer prce from the fle prce range, these revsons are n themselves somewhat predctable, as llustrated n Fgure 6. Here, for each month between March 99 and August 998, the percentage of IPOs that are prced above ther fle prce range s graphed on the top, and the percentage of IPOs that are prced below ther fle prce range s graphed on the bottom. (The sample starts n March 99 because many of the months from January 990 to February 99 had few IPOs, as was the case n September and October 998.) For a gven month, these two percentages wll generally add up to less than 00% because of other offerngs that are prced wthn the fle prce range. Inspecton of the fgure dscloses that there s persstence n the proporton of upward and downward revsons from month to month. Indeed, we can compute the average percentage change from the mdpont of the fle prce range to the fnal offer prce each month. The frst-order autocorrelaton of ths seres of monthly observatons s 0.6 (p = 0.000), and the autocorrelaton at two lags s 0.28 (p = ). Informaton about prevous frst-day returns, whch of course s publc nformaton, can be used to predct the average frst-day return n a month. In Fgure 7, we show the average frstday return by month for the same perod as n Fgure 6. For the March 99 to August Maksmovc and Unal (993), however, do not fnd predctablty when examnng the frst-day returns on the IPOs of mutual fnancal nsttutons convertng to stock ownershp. The offer prce for these IPOs must be approved by regulators.

14 perod, the frst-order autocorrelaton of monthly average frst-day returns s 0.50 (p-value = 0.000), and the second-order autocorrelaton s 0.8 (p-value = 0.089). Ths persstence has been prevously documented n the lterature (Ibbotson and Jaffe (975), Ibbotson, Sndelar, and Rtter (994)). Not surprsngly, there s contemporaneous correlaton of monthly average frstday returns and the average revson n the offer prce durng a month. The contemporaneous correlaton coeffcent s 0.77, whch s sgnfcantly dfferent from zero wth a p-value of b. The dynamc nformaton acquston and prospect theory predctons There are a number of alternatve explanatons for why underwrters do not fully adjust the offer prce to nformaton about the state of demand. These theores are not mutually exclusve. The Benvenste-Spndt (989) dynamc nformaton acquston hypothess predcts that regular nvestors should be rewarded for revealng ther prvate nformaton. In the dynamc nformaton acquston framework, condtonal underprcng s a result of an ncentvecompatblty constrant. There should, however, be full adjustment to publc nformaton, such as whether the went up or down durng the weeks between when the fle prce range was set and the fnal offer prce was set. Our prospect theory explanaton of the partal adjustment phenomenon does not dstngush between prvate and publc nformaton, so the partal adjustment of the offer prce wth respect to pror movements s as predcted. Investment bankers can selectvely underprce some IPOs by combnng the bad news that there has been excessve dluton wth the good news that they are wealther than had been expected. 6.c. Optmal rsk sharng Any theory (ncludng our prospect theory explanaton) that has partal adjustment to publc nformaton mples that the ssuer s beng partly nsured aganst movements. It should be noted that publcly traded nstruments, such as put optons on the and varous ndustry ndces, exst, and they would appear to be the most effcent manner for provdng ths nsurance. Thus, any complete theory must nclude an explanaton for why these nstruments aren t used to nsure the ssuer aganst movements. A smple extenson of Mandelker and Ravv (977) would show that wth a rsk-averse prncpal (the ssuer) and a rsk-neutral agent (the underwrter), and wth no prvate nformaton, 3

15 the optmal contract s for the ssuer to be gven a fxed net proceeds. The underwrter would bear the rsk of changes n valuaton durng the presellng perod. Wth prvate nformaton about frm value that may leak out durng the presellng perod, the analyss would have to be modfed. Wth the ssung frm possessng prvate nformaton, the net proceeds to the ssung frm should not change based upon publc nformaton about movements, but there should be partal adjustment wth respect to frm-specfc valuaton changes (Baron (982)). A more realstc model must take nto account that there are three partes, not two, nvolved. Changes n valuaton must be apportoned among ssuer, underwrter, and nvestors. How changes n valuaton are borne between underwrters and nvestors determnes the adjustments n gross spreads versus adjustments n the amount of money left on the table. Both our prospect theory explanaton and the optmal rsk-bearng explanaton predct that there wll be state-contngent wealth transfers from ssuers to the two other partes nvolved, underwrters and nvestors. The optmal rsk-sharng explanaton cannot explan why publcly traded nstruments, such as put optons on a ndex, are not used to nsure ssuers aganst adverse movements. Our prospect theory explanaton, on the other hand, argues that ssuers let down ther guard when there s good news, and do not bargan aggressvely for a hgher offer prce, whether ths s due to publc nformaton or not. In addton to makng predctons about the level of wealth transfers n varous states of the world, the optmal rsk-sharng and prospect theory explanatons also make predctons about the form of payments (spreads versus money on the table). The optmal rsk-sharng explanaton for leavng money on the table must explan why nvestors, rather than underwrters, perform the rsk-bearng functon. If underwrters bore the rsk, then percentage spreads would be hgher n the good state of the world than n the bad state of the world. Wthout makng addtonal auxlary assumptons, we do not have an obvous explanaton for why ths does not occur f optmal rsk-sharng s the salent factor for explanng why the amount of money left on the table vares n a predctable manner. Our prospect theory explanaton states that ssuers make a dstncton between drect costs (spreads) and opportunty costs (money on the table). Thus, f underwrters substantally 4 Sternberg (989) contans a barganng model to ratonalze the partal adjustment phenomena, but he assumes a Rubensten splt-the-dfference equlbrum.

16 ncreased ther percentage spreads n the good state of the world, ssuers would bargan harder than f the wealth transfer was n the form of an opportunty cost. Thus, underwrters fnd that leavng money on the table may be more advantageous, n that rent-seekng behavor by nvestors wll allow underwrters to recapture ths wealth transfer from ssuers. 6.d. Alternatve explanatons for the partal adjustment phenomenon Prevous agency theores of nvestment bankers takng advantage of ssuers (.e., Baron (982)) have nvestment bankers usng ther superor knowledge about prvate nformaton to underprce IPOs. As wth Benvenste and Spndt, Baron s model doesn t make the predcton that there wll be partal adjustment to publc nformaton. An explanaton offered by nvestment bankers as to why the offer prce adjusts only partally s as follows. Potental nvestors anchor on the mdpont of the fle prce range, just as ssung frms do. If the offer prce s rased too far above ths, some potental nvestors wll defect (see Hof (999) for an example of ths argument), even f the ncrease s n response to publcly avalable nformaton. Thus, because of the anchorng, underwrters are loath to adjust the offer prce too much. There are two varants of ths argument, whch we call the unsophstcated and the sophstcated varant. The unsophstcated verson s just statng that a hgher offer prce deters some buyers. In other words, there s a negatvely sloped demand curve. Rasng the offer prce reduces the excess demand. Although the underwrter can exercse more dscreton n allocatng underprced shares f there s excess demand, t s not clear why ths s of great advantage to the ssuer. The sophstcated verson of the argument that rasng the offer prce n the face of strong demand nduces some buyers to defect s based upon Welch s (992) cascades argument. The logc s that f nvestors are payng attenton to what other nvestors are dong, ncreasng the offer prce s rsky because f some nvestors decde not to buy, many others may suddenly wthdraw ther purchase orders as well. We are skeptcal of ths argument for the followng reason. Insttutonal nvestors are well aware of the partal-adjustment phenomenon, so that an ncrease n the offer prce generally results n ncreased demand because t s sgnallng to nvestors that other nvestors want to buy the IPO. Because of ths, once a fle prce range has been set, there tends to be a postvely sloped demand curve. Two axoms among IPO nvestors 5

17 are Cut the deal, cancel my order and Increase the deal, double my order (Ftzgbbon (998)). One reason that s sometmes gven by academcs for why IPOs are underprced, on average, s that hgh-qualty frms are tryng to sgnal ther type by ntentonally leavng money on the table. Whle ths may partly explan why some IPOs are underprced, t does not explan why there s partal adjustment to publc, as opposed to prvate, nformaton. An alternatve explanaton for partal adjustment s sometmes voced by underwrters. Ths s the leanng aganst the wnd theory. Investors n the IPO tend to overreact. Thus, when the s too receptve, the prce s bd up too hgh, above ts long-run value. Underwrters prce the ssues wth long-run value n mnd. Ths leanng aganst the wnd theory has the testable mplcaton that there should be a negatve correlaton of frst-day returns and subsequent long-run returns. Ths mplcaton s tested n Table 4, where we report three-year buy-and-hold returns for the IPOs n our sample after categorzng them on the bass of revsons n the offer prce, and pror movements. We also report the compounded value-weghted returns over the same holdng perods, and the wealth relatves, computed as the average gross return on the IPOs dvded by the average gross return. Inspecton of the table dscloses no obvous patterns, wth most of the wealth relatves near one, whether the offer prce had been adjusted up or down. 2 The lack of a pattern s also present when, n unreported results, we use style benchmarks (sze and book-to-). The lack of any relable patterns s nconsstent wth the leanng aganst the wnd hypothess If we categorze IPOs by a fner parttonng of the revson of the offer prce from the fle range, the lack of patterns n long-run returns contnues to hold. 3 Evdence supportng the leanng aganst the wnd hypothess s contaned n Rtter (99) and Krgman, Shaw, and Womack (999). Rtter reports that IPOs wth the hghest frst-day runups have the lowest long-run szeadjusted returns. Part of the reason s that Rtter s results, usng IPOs from , are domnated by mcrocap IPOs. Rtter ncludes all IPOs wth an offer prce of $.00 or above, whereas we exclude IPOs wth a mdpont of the fle prce range below $8.00. Krgman, Shaw, and Womack report that IPOs wth frst-day returns n excess of 60% subsequently underperform, whereas offerngs wth moderate frst-day returns outperform a sze-matched benchmark durng ther frst year. Ther subsample of IPOs wth bg frst-day returns from January 988-May 995 contans only 33 IPOs. The late 990s saw a number of frms wth hgh frst-day returns (such as Yahoo!) produce extremely hgh returns durng the next few years. Consstent wth our evdence, Logue, Rogalsk, Seward, and Foster-Johnson (00) also report no long-run predctve ablty when IPOs are categorzed by revsons n the offer prce.

18 Other explanatons that probably have some mert nclude the followng: If an ssuer questons an underwrter about severe underprcng, the nvestment banker wll always be wllng to argue that the prce jump was due to a successful job of ng the ssue by the nvestment banker. And ssuers are well aware that the prce of stocks that have jumped a lot n prce can come down by the tme that a lockup provson expres. There s also an element of nvestment banker psychology that comes nto play. When there s weak demand for an offerng, nvestment bankers do not lke havng to delver bad news to an ssuer. To mnmze the confrontaton, they do not lower the offer prce (and the proceeds) as much as would be needed to mantan a target 0-5% underprcng. If the offer prce s lowered too much, there s also the danger that the ssung frm wll decde to cancel the offerng and seek alternatve fnancng A theory of hot ssue s One of the puzzles regardng IPOs s the exstence of hot ssue s. Followng Ibbotson and Jaffe (975), a hot ssue s defned as a month n whch the average frst-day return s above the medan month s average frst-day return. 4 There s strong postve seral correlaton n the monthly average frst-day returns. Fgure 7 shows the monthly average frstday returns on IPOs. The frst-order autocorrelaton coeffcent s 0.50, and the second-order autocorrelaton coeffcent s 0.8. Currently, the lterature offers no explanaton that s consstent wth ratonal behavor on the part of nvestors that can generate ths postve autocorrelaton. When there s a rse n the, our prospect theory explanaton predcts an ncrease n the expected underprcng of all IPOs that are n the sellng perod, whether they wll be gong publc tomorrow or fve weeks later. Thus, we can explan the autocorrelaton of average frst-day returns. It s an equlbrum explanaton, n that no one s actng rratonally. Prvate nformaton s serally uncorrelated, so there should be no autocorrelaton n frstday returns accordng to the dynamc nformaton acquston model. Prospect theory, on the other hand, does not dstngush between prvate and publc nformaton as a reason for wealth changes. Because sellng perods overlap, there wll be autocorrelaton n frst-day returns. 4 Other authors mplctly or explctly use some measure of volume. Typcally, the volume of IPOs lags average frst-day returns by several months (Ibbotson, Sndelar, and Rtter (994)).

19 Frst-day returns wll be hgher followng rses, and ths effect wll be present n all IPOs where the sellng perod ncludes the perod of the rse. Thus, rses wll be followed by hgh average frst-day returns for one to two months, snce the tme between choosng a prce range on whch ssuers anchor and actually gong publc s typcally n ths range. These deas can be formalzed n equatons () and (2). Equaton () expresses the offer prce as equal to the mdpont of the fle prce range plus several addtonal terms reflectng new nformaton that arrves durng the book-buldng perod. Ths book-buldng perod starts at tme 0 and ends at tme T, the date of settng the fnal offer prce. These addtonal terms reflect the return, m r 0,T, addtonal publc nformaton such as ndustry-specfc movements, publc r 0,T, and frm-specfc dosyncratc nformaton, prvate r 0,T. By defnton, ths last component s uncorrelated across frms, no matter when they go publc. The return and other publcnformaton return for date t, however, affect all IPOs whose book-buldng perod ncludes tme t. 8 OP = mdpont + α + α + α publc prvate [ r0,t 2r0,T 3r0,T ] + τ () P, = mdpont β r γ r γ r ] + ε publc prvate.4 [ + 0, T + 2 0, T + 3 0, T (2) Equaton (2) expresses the prce as a functon of the orgnal valuaton, multpled by addtonal terms reflectng new nformaton that arrves durng the book-buldng perod. In equaton (2), the scalar of.4 s present because the mdpont of the fle prce range s a based estmator of the frst closng prce; our sample s underprced by 4%, on average. The predctons of the Benvenste-Spndt dynamc nformaton acquston model (hereafter, B-S) and our prospect theory model (hereafter, P-T) for the coeffcents of equatons () and (2) are as follows: n equaton (), B-S predcts that α should equal the Sharpe-Lntner beta, α 2 should equal the relevant factor senstvty, and α 3 should be between 0 and (partal adjustment to prvate nformaton). P-T predcts that α should be a fracton of beta, α 2 should be a fracton of the factor senstvty, and α 3 should be between 0 and, snce there should be

20 partal adjustment to both prvate and publc nformaton. We can dstngush between the two theores based upon the dfference n the coeffcents n equatons (2) versus (). If B-S s a full descrpton, then the coeffcent on α n equaton () should equal that on β n equaton (2). If P-T s descrptve of the process, then the coeffcent on α n equaton () should be closer to zero than β n equaton (2). Before testng these predctons, both equatons are dvded by the mdpont of the fle prce range, so that the dependent varables are expressed as revsons from the mdpont. (The transformatons are gven as equatons ( ) and (2 ) n the appendx.) As a result of ths dvson, the explanatory varables are expressed as returns, rather than returns tmes the mdpont. Snce the only observable varable that we use s the return, we run unvarate regressons. The resdual term n these regressons ncludes not only the error term, but also the terms representng other publc nformaton and prvate nformaton. In other words, we run sngle-varable regressons wth the return as the explanatory varable. These are reported n Table 5, after we subtract from each sde and multply by 00 to convert nto percentage changes. When usng the return on the IPO from the mdpont of the fle prce range to the frst closng prce, the coeffcent on the return durng the 5 tradng days pror to the offer date mples a beta of 2.37 (see Row 3 of Panel A). Whle hgh, ths s consstent wth the fndngs of Clarkson and Thompson (990) and Chan and Lakonshok (992) that the beta of IPOs shortly after gong publc s n excess of 2. Usng the percentage revson n the offer prce as the dependent varable, the slope coeffcent s only 0.76 (Row 2 of Panel A). The rato of coeffcents, 0.76/2.37, s only 0.32, far below the.00 that would be mpled by the B-S model. 5 Ths mples that only one-thrd of the publc nformaton about returns durng the book-buldng perod s ncorporated nto the offer prce. Ths s consstent wth our prospect theory explanaton beng economcally mportant relatve to the dynamc nformaton acquston model. 9 5 Snce some of the IPOs where downward revsons occur are wthdrawn, some of the populaton of IPOs wth negatve error terms are deleted from the sample. The resultng OLS slope coeffcents wll be based towards zero. It s not obvous whether the 0.32 rato s based upwards or downwards, however.

21 In Row 2 of Table 5 s Panel A, where the dependent varable s the percentage revson of the offer prce from the mdpont of the fle prce range, the ntercept can be nterpreted as the average percentage change n the offer prce relatve to the mdpont n a flat. The ntercept of -.47 mples that, n a flat, the offer prce s only.47% below the mdpont. Thus, the mdpont s qute close to beng an unbased estmate of the fnal offer prce, condtonal upon the IPO beng completed, snce wthdrawn deals are not n our sample. In Panels B and C, where we categorze IPOs on the bass of whether the had gone down or up pror to the offer, the ntercepts n the offer prce revson regressons are even closer to zero. Furthermore, n all three panels, the ntercepts n the regressons usng the closng prce n the dependent varable are close to the 4% uncondtonal average return for our sample. What autocorrelaton of monthly average frst-day returns does ths mply? The answer depends upon the relatve magntude of the varance of the error terms, the varance of publc nformaton, and the varance of prvate nformaton. To take extreme cases, f all nformaton s prvate, the autocorrelaton wll be zero. If all nformaton s publc nformaton about returns and the partal adjustment coeffcent s zero, we have a stuaton analogous to autocorrelaton of an ndex ntroduced by nonsynchronous tradng (Atchson, Butler, and Smonds (987)). In the appendx, we derve the predcted autocorrelaton of the monthly average frst-day returns. Obvously, the exact numbers depend upon parameter values. Assumng that the lagged return s the only publcly avalable nformaton, we derve a frst-order autocorrelaton coeffcent n the vcnty of 0. to 0.2. Although postve, these values are notceably below the emprcal coeffcent of In our analyss, we do not nclude publcly avalable nformaton about ndustry returns. The extreme underprcng of many nternet IPOs n 999, at the same tme that most non-nternet IPOs are not severely underprced, ndcates that ndustry factors are economcally mportant. Incluson of lagged ndustry returns would undoubtedly ncrease the predcted autocorrelaton. Lowry and Schwert (00) document that there s much more of an adjustment downwards followng declnes than upwards followng ncreases. In Panels B and C of Table 5, we confrm ths. Panel B reports regresson results for IPOs followng declnes. The coeffcent of.5 (t-statstc of 4.58) on the return n row 2 shows that