BEHAVIOR OF LIQUIDITY AND RETURNS AROUND CANADIAN SEASONED EQUITY OFFERINGS. Current Version: December 2008

Transcription

1 BEHAVIOR OF LIQUIDITY AND RETURNS AROUND CANADIAN SEASONED EQUITY OFFERINGS Lawrence Kryzanowski, 1 Skander Lazrak 2 and Ian Rakia Curren Version: December Concordia Universiy Research Chair in Finance. Finance Deparmen, John Molson School of Business, Concordia Universiy, 1455 de Maisonneuve Blvd. Wes, Monreal, Quebec, Canada, HG 1M8. Telephone: (514) , X2782; lawrence.kryzanowski@concordia.ca 2 Assisan Professor of Finance. Deparmen of Finance, Operaions and Informaion Sysems, Faculy of Business, Brock Universiy, 500 Glenridge Avenue, S. Caherines, Onario, Canada, L2S A1. Telephone: (905) , X511; slazrak@brocku.ca Associae Professor of Finance. Finance Deparmen, John Molson School of Business, Concordia Universiy, 1455 de Maisonneuve Blvd. Wes, Monreal, Quebec, Canada, HG 1M8. Telephone: (514) , X2786; rakia@jmsb.concordia.ca Financial suppor from he Auorié des Marchés Financiers (AMF), Concordia Universiy Research Chair in Finance and SSHRC (Social Sciences and Humaniies Research Council of Canada) are graefully acknowledged. The usual disclaimer applies in ha he views expressed herein are solely our own and do no necessarily reflec he official posiions or policies of he providers of financial suppor of his paper or he views of heir saff members. Please do no quoe wihou he auhors permission. Commens are welcomed.

2 BEHAVIOR OF LIQUIDITY AND RETURNS AROUND CANADIAN SEASONED EQUITY OFFERINGS ABSTRACT Liquidiy improves for announcemen and closing windows for Canadian SEOs. The adverse selecion (emporary) spread cos follows an approximae V-shaped paern achieving is lowes level a he announcemen (closing) window. The adverse selecion cos of privaely-placed Canadian SEOs decreases afer Mulilaeral Insrumen reduced he lock-up period o four monhs in Consisen wih he reducions in informaion asymmery, negaive (no robus) abnormal reurns occur in announcemen and closing windows for undiffereniaed SEOs. Abnormal reurns are significanly differen for public (significanly negaive) versus privae (insignificanly posiive) SEOs. Condiional residual volailiies decrease pos-announcemen, consisen wih a diminished emporary spread cos. Keywords: Seasoned equiy offerings, liquidiy, lock-up period, reurns, condiional volailiy. JEL Classificaion: G10, G12, G14, G15

3 BEHAVIOR OF LIQUIDITY AND RETURNS AROUND CANADIAN SEASONED EQUITY OFFERINGS 1. INTRODUCTION Unlike an iniial public offering (IPO) wherein a firm issues equiy o he public for he firs ime, a seasoned equiy offering (SEO) occurs when a firm is already publicly raded and is simply selling addiional common sock. Alhough i is much less exensive han he IPO lieraure, he SEO lieraure examines sraegic behavior around he ime of he SEO announcemen (e.g., Gerard and Nanda, 199; Bayless and Chaplinsky, 1996; Safieddine and Wilhelm, 1996; and Teoh e al., 1998), 1 he long-run marke and operaing underperformance of firms following SEO issuance (Loughran and Rier, 1997; Herzel e al., 2002; and Desrosiers e al., 2004), he informaion conen of SEOs (e.g., Slovin e al., 1991; McLaughlin e al., 1998), and he marke behavior of reurns (e.g., Asquih and Mullins, 1986; Masulis and Korwar, 1986; and Slovin e al., 1994) and o a much lesser exen liquidiy and asymmeric informaion (e.g., Tripahy and Rao, 1992; and Brooks and Pael, 2000) for SEO announcemens. We address a number of shorcomings in he exising lieraure in his paper. The firs shorcoming is ha he exising lieraure does no examine he behavior of various liquidiy proxies such as dollar volumes, spreads (oal and componens) and dephs hrough all of he various idenifiable sages (henceforh, windows) of he SEO issuance cycle. As a resul, effecs for he pos-announcemen window are confounded wih he effecs of he SEO closing window. By examining four insead of wo windows over he SEO issuance cycle, we are able o ensure ha he pos SEO window does no include SEO closing effecs. Our esable expecaion is ha share liquidiy improves over he SEO issuance cycle and ha some of he iniial improvemen in share liquidiy dissipaes afer he SEO compleion window. 1 Conrary o he predicion of he model of Baker and Sein (2004) ha implies ha managers should ime SEO issues when liquidiy is high (windows of opporuniy), DeAngelo e al. (2007) find ha firms issue SEOs due o cash shorages.

4 The second shorcoming is ha he exising lieraure does no examine he impac of wo imporan regulaory changes on liquidiy behavior hrough he SEO issuance cycle. 2 To he bes of our knowledge, we are he firs o es he efficacy of a raionale given for he reducion in he required minimum holding period for privae placemens for qualifying issuers. Specifically, he SEC s (2007, p. 1) saed raionale for shorening he holding period requiremen under Rule 144 o six monhs for resriced securiies of issuers on February 15, 2008 was [w]e believe ha he amendmens will increase he liquidiy of privaely sold securiies and decrease he cos of capial for all issuers wihou compromising invesor proecion. In his paper, we examine he impac on liquidiy hrough he SEO issuance cycle of he adopion of Mulilaeral Insrumen on November 0, 2001 in Canada ha reduced he required minimum holding period for privae placemens o four monhs for qualifying issuers. This leads o he esable expecaion ha he asymmeric informaion associaed wih privaely-placed Canadian SEOs decreased afer he adopion of Mulilaeral Insrumen To he bes of our knowledge, we also are he firs o examine he impac on liquidiy (including spread componens) hrough he SEO issuance cycle of he reducion of he minimum price incremen and inroducion of decimal pricing on he TSX in April 1996 (henceforh TSX decimalizaion). Various papers find ha he move o decimalizaion reduced quoed and effecive spreads and quoed dephs on he TSX (e.g., Chung e al., 1996; Ahn e al., 1998; Bacidore, 1997; and Porer and Weaver, 1997). Li and Parker (2005) find ha he adverse selecion cos is significanly smaller and inversely relaed o firm size and rading volume afer decimalizaion in U.S. markes. We examine wheher hese findings for he overall Canadian and U.S. markes can be generalized o Canadian SEOs by esing our expecaion ha oal spreads and is adverse selecion cos componen for SEOs decreased afer TSX decimalizaion. 2 Mioo (2006) documens he impac of he implemenaion of he Canada and U.S. mulijurisdicional disclosure sysem (MJDS) in 1991 on Canadian for cross-border SEO issues. The U.S. Securiies and Exchange Commission (SEC) has ofen reviewed and revised Rule 144 under he U.S. Securiies Ac of 19, which creaes a safe harbor for he sale of resriced (and conrol) securiies. In 1997, Rule 144 was amended o shoren he iniial holding period from wo o one year and he ulimae holding period for nonaffiliaes from hree o wo years. Securiies and Exchange Commission, Release No (Feb. 20, 1997) [62 FR 9242] ( he 1997 Adoping Release ). 2

5 The hird shorcoming is ha no sudy o he bes of our knowledge conducs a comprehensive comparison of he various ypes of SEO issuing mechanisms, some of which (e.g., bough deals) are no used in he U.S. o raise new equiy financing. Thus, we examine he behavior of liquidiy and reurns over he SEO issuance cycle for four issuing mechanisms individually and for wo of heir pairings (bough versus no bough, and privae versus public). Bough deals are a widely employed issuing mechanism in Canada and he U.K (Slovin e al., 2000). In a bough deal, a single invesmen dealer conacs an issuing firm wih a ready-made deal for a firm commimen by he dealer o purchase a fixed number of shares a a fixed price generally before finalizing a group of poenial (generally insiuional) buyers. An abbreviaed regisraion saemen (or shor-form prospecus) is ofen used o reduce he normal issuance ime frame and associaed coss (including lower issuer fees). Since underwriers purchase securiies in bough deals and resell such securiies o heir large cliens, bough deals should signal confirmaions (or cerificaions) of issue values o all marke paricipans ha should heoreically reduce informaion asymmery. This leads o he esable expecaion ha bough deals are associaed wih less informaion asymmery (and hus lower adverse selecion coss) han no bough deals. Unlike heir publicly placed counerpars, invesors in privaely placed SEOs may benefi from reduced issuing coss, more relaxed disclosure requiremens, speed of issue closing and collecion of issue proceeds, and ime-limied rading resricions (Brau e al., 2005; Brav and Gompers, 200) ha may signal an enhanced alignmen of he ineress of insiders wih new share purchasers. The few sudies ha examine privae placemen SEOs find significan posiive (low) mean share price effecs for he issuing firms over he shor (long) run and do no direcly examine liquidiy effecs (e.g., Wruck, 1989; Herzel and Smih, 199; Herzel e al., 2002). This leads o he esable expecaion ha privae placemens as commimen devices reduce informaion asymmery as measured by he adverse selecion componen of he bid-ask spread (and hus, oal spreads). The fourh shorcoming is ha he lieraure does no examine he behavior of liquidiy and valuaions over he SEO issuance cycle for a resource or commodiy based economy such as Canada. The naural resource secor (paricularly, mining, oil and gas) due o he imporance of new discoveries o replace

6 depleing reserves and he echnical complexiy of heir operaions provides relaively greaer opporuniies for maerial informaion asymmeries beween informed and uninformed invesors. This leads o he esable expecaion ha SEOs for resource issuers are associaed wih greaer informaion asymmery (and hus higher spreads) han SEOs for non-resource issuers. The fifh and final shorcoming is ha he ess repored in he lieraure for valuaion effecs around SEO iniial and closing announcemens rely on uncondiional reurn models. As a resul, hey provide no explici conrols for he possible effec of ime-varying reurn volailiies and liquidiies on valuaion changes for boh ypes of SEO announcemens. We provide a formal es of he possible changes in volailiies pos-seo and in valuaions around boh SEO announcemen and closing daes. Our esable expecaions wih regard o valuaion effecs are ha abnormal reurn differences will favor privae over public SEOs, bough over no bough SEOs and resource over no resource SEOs due o expeced differences in heir adverse selecion coss. Before proceeding, a legiimae quesion o pose is: Why examine he behavior of Canadian SEOs? We argue ha an examinaion of Canadian SEOs no only allows us o minimize daa snooping biases ha arise from examining he same (U.S.) daase repeaedly bu i also allows us o provide a benchmark for issues ha have no already been examined in he lieraure, such as he liquidiy behavior for bough deals hrough he SEO issuance cycle. Similarly, our ou-of-sample examinaions of some issues previously examined in oher markes (primarily, U.S.) ha are quie similar in many respecs o he Canadian marke provide us wih a useful benchmark for assessing he generalizabiliy of inferences based on research conduced in hose markes. The significan differences repored in he lieraure beween Canadian and U.S. IPOs sugges ha maerial differences are likely o exis for Canadian versus non-canadian SEOs. This IPO lieraure finds marked Canadian versus U.S. differences in he role of overallomen opions in erms of underwriing fees and price sabilizaion (Chung e al., 2000), in underwrier fee clusering wih is absence for Canadian IPOs (Kryzanowski and Rakia, 1999), in abnormal reurns around unlock days wih significance abnormal reurns primarily in he U.S. 4

7 (Kryzanowski and Liang, 2008) and in firs-day mean reurns wih Canadian IPOs among he lowes worldwide (e.g., Kryzanowski e al., 2005; Jay Rier s websie a hp://bear.cba.ufl.edu/rier/in.pdf). Our major empirical findings are as follows. Firs, we find ha spreads and heir componens (dollar volumes and dephs) follow an approximae (invered) V-shaped paern hrough he SEO issuance cycle. SEOs resul in posiive liquidiy benefis for shareholders ha are maximized during announcemens or closings and exend beyond SEO closings. This increased liquidiy resuls from a reducion of boh adverse selecion and emporary coss as SEOs are informaive evens ha resul in an increase in he rading aciviy of invesors (including hose ha are liquidiy seeking). While boh spread componens follow approximae V-shaped paerns, he adverse selecion cos falls upon announcemen and sars revering a closing while he emporary cos coninues falling and his i lowes level around he closing dae. Thus, mos informaion asymmery is reduced around he SEO announcemen when he firm issues a saemen declaring is inenion o seek new financing. SEOs also increase he poenial or acual floa and increase invesor recogniion of heir firms. Second, we find ha wo imporan regulaory changes have a significan impac on liquidiy hrough he SEO issuance cycle. The adverse selecion spread componen (oal spread) is (no) significanly affeced upwards by TSX decimalizaion in 1996 and is affeced significanly downwards by he reducion in he required holding period for privae placemens in While our findings do no suppor our esable expecaion ha oal spreads and adverse selecion coss decreased afer TSX decimalizaion, hey do suppor our esable expecaion ha asymmeric informaion of privaely-placed Canadian SEOs decreased afer he adopion of Mulilaeral Insrumen Third, we find ha quoed and effecive spreads and he wo spread componens are significanly lower for bough versus no bough deals and significanly higher for resource versus no resource SEOs, and quoed spreads and he adverse selecion coss are significanly higher for privae placemens afer conrolling for various firm-specific deerminans of spreads. This suppors he hypohesis ha bough deals are associaed wih less informaion asymmery (and hus lower oal spreads) han no bough deals and ha SEOs for resource issuers are associaed wih greaer informaion asymmery (and hus higher 5

8 spreads) han SEOs for no resource issuers. However, our findings do no suppor our esable expecaion ha privae placemens as commimen devices help o alleviae moral hazard problems by reducing informaion asymmery as measured by he adverse selecion componen of he bid-ask spread (and hus, oal spreads). Fourh and finally, we find negaive (bu no robus) abnormal reurns in announcemen and closing windows ha are relaed o he reducion in informaion asymmery caused by he SEO announcemens. While our liquidiy proxy is no a significan deerminan of reurns around hese evens, condiional residual reurn volailiies decrease pos-announcemen indicaing reduced uncerainy due o he parial resoluion of informaion asymmery. Furhermore, only he differences in abnormal reurns beween public and privae SEOs during he announcemen window are significan wih he former being significanly negaive and he laer being insignificanly posiive. Thus, our findings only suppor our esable expecaion ha abnormal reurn differences will favor privae over public SEOs due o expeced differences in heir adverse selecion coss. The remainder of his paper is organized as follows. In he nex secion, he SEO sample and daa are described. The resuls of our invesigaion ino shor-run liquidiy and rade aciviy, as measured by dollar volume, he proporional effecive spread and he quoed deph based on univariae and mulivariae approaches, are repored and analyzed in secions hree and four, respecively. In secion five, he componens of he bid-ask spread for six ypes of SEOs are esimaed and discussed. A range of reurn models incorporaing a liquidiy proxy and allowing for condiional volailiies are esimaed and discussed in secion six. Some concluding commens are offered in secion seven. 2. SAMPLE AND DESCRIPTION OF THE DATA Canadian SEOs during are idenified using he Financial Pos s Record of New Issues. Deb, iniial public offerings, uni offerings and preferred shares and issues wih offer prices below $2 are eliminaed. Each of he reained 756 equiy issues is classified as a privae offering, public offering, bough deal, no bough deal, resource issue and no resource issue. 6

9 Inraday rade and quoe daa are obained from he TSX. Firms ha did no rade for five successive days in eiher of he pre-announcemen or pos-closing windows are eliminaed for he liquidiy-relaed ess o ensure ha liquidiy can be reliably measured. 4 All quoes and rades ouside normal rading hours (9:0 am o 4:00 pm ET) are eliminaed. Quoes are eliminaed if negaive or heir corresponding spread is less han zero or more han 0 percen of he mid-spread. Trades are eliminaed if hey are for a negaive price or number of shares or have special selemen condiions or are for delayed delivery or cancellaion or resul in rade-by-rade reurns exceeding 50 percen. Table 1 repors an annual accoun of he number of SEOs and mean issue sizes (no) differeniaed by disribuion mehod and issuer ype for our final sample of 667 SEOs. Paricularly weak SEO years occur in 1995 for all SEOs, public offerings, bough deals and no resource issues, in for privae placemens, in 1998 for no bough deals, and in 2000 for resource issues. A paricularly srong SEO year occurs in 1997 for all SEOs and for he six differeniaed SEO samples. [Please place able 1 abou here]. LIQUIDITY BEHAVIOR AROUND CANADIAN SEOs The lieraure examining liquidiy behavior around SEOs is no exensive. Tripahy and Rao (1992) examine changes in percenage quoed spreads around announcemen and offer daes for U.S. over-hecouner SEO issues. Using issue size as a proxy for he exen of informaion asymmery, hey find ha spreads reach "normal" levels before he firs public disclosure for larger issues and only on he offer dae for smaller issues. They conclude ha he former finding is consisen wih he dealer reducing adverse informaion risk hrough informaion gahering during he underwriing process and ha he laer finding is consisen wih a reduced informaion asymmery effec and wih afer-marke liquidiy suppor by he dealer. Brooks and Pael (2000) repor ha larger changes in informaion asymmery (as proxied by he 4 The five-day screen reduced he sample size from 756 o 667 SEOs. All of he SEOs in he final sample have rades for a leas wo of he hree days in heir announcemen and closing windows. 7

10 adverse selecion componen of bid-ask spreads) a announcemens are correlaed wih larger reducions in wealh for a sample of 48 NYSE and AMEX SEOs issued in Before proceeding o an examinaion of liquidiy behavior around our sample of SEOs, we need o decide on he windows o be examined. Pael and Brooks (2000) compare he asymmeric componen for each day in he announcemen window (i.e., he 7 days cenered on he announcemen day) o a 1-day pre-announcemen window consising of days -40 o -10, and for his pre-announcemen window o a pos-announcemen window consising of days +10 o We do no use his delineaion since our SEO announcemen and closing daes are separaed by abou 15 rading days (21 calendar days) on average wih a maximum of 20 days and a sandard deviaion of 17 rading days. Thus, o avoid a comingling of closing announcemens wih pos SEO announcemen windows and o provide for a beer examinaion of changes over he SEO issuance cycle, our subsequen analysis of liquidiy is conduced over four ime windows: window 1 (W1) goes from 80 hrough 20 rading days before he announcemen dae; windows 2 and (W2 and W) cover he hree rading days cenered on he SEO announcemen and closing daes, respecively; and window 4 (W4) goes from 20 hrough 80 rading days afer he SEO closing dae. Table 2 conains he cross-secion means and medians of hree measures of liquidiy for he four SEO windows for he full ime period and for he pre- and pos-tsx-decimalizaion periods. 6 The firs measure is he average of each day s oal raded dollar volume (henceforh, dollar volume). Generally, his measure is highes during he wo windows wih significanly elevaed informaion flow (i.e., he hreeday announcemen and closing windows). The mean and median dollar volumes for he announcemen window (W2) are approximaely double (and significanly differen a 1% from) heir counerpars preannouncemen for he ime periods (no) differeniaed by TSX decimalizaion and o wheher or no he 57 SEOs wih he same announcemen and effecive daes are included. 5 Frijns e al. (2006) repor a mean (median) inerval beween SEO announcemens and SEO closings of 41 (1) rading days for heir final sample of 1245 U.S. SEOs issued beween 1984 and Since he resuls for quoed spreads are similar o hose for effecive spreads, hey are no abulaed o save valuable journal space. 8

11 [Please place able 2 abou here] Mean and median dollar volumes decline subsanially in he closing window (W) bu are sill abou 50% higher han hose of he pre-announcemen window. The mean for W is significanly differen a < 1% from hose of W1 and W2. The mean bu no he median dollar volume in he hree calendar monhs pos-closing window (W4) declines furher bu is sill consisenly higher han during he preannouncemen window (W1). Similarly, dollar volumes for periods differeniaed by TSX decimalizaion increase dramaically around he SEO announcemen and hen decline during he SEO closing window when compared o he announcemen level bu remain higher han heir pre-announcemen levels. Furhermore, mean and median pos-decimalizaion dollar volumes are approximaely double heir respecive pre-decimalizaion levels in each of he hree windows and are significanly differen. This resul is parially expeced as dollar rading volumes generally increase wih ime. The second measure is proporional effecive spreads, or wice he absolue difference [ 2 Z ] beween he ransacion price P and he prevailing mid-quoe [ M =(A +B )/2] scaled by he prevailing mid-quoe based on a one-second lag as in Henker and Wang (2006). 7 To obain he saisics repored in able 2 for his measure, we firs compue he iner-daily average for each window using he average proporional effecive spread for each SEO firm for every day using heir inra-daily spreads, and hen compue he cross-secion mean and median spreads over he SEOs included in our sample for each window. The mean (median) proporional effecive spread falls from a pre-seo announcemen value of 1.89% (1.66%) o 1.66% (1.42%) upon announcemen. Conrary o dollar volume, he spread coninues o fall around he SEO closing dae o a significan mean (median) of 1.48% (1.2%) before increasing pos closing o 1.70% (1.52%). Alhough he effecive spread increases afer he closing dae, i remains significanly lower han is pre-announcemen levels based on univariae paired ess. No only is he behavior of effecive spreads over he SEO issuance cycle robus o decimalizaion differeniaion bu he mean effecive spreads are no significanly differen pre- and pos-decimalizaion a 1.92% for W1 and a 7 The findings repored in able 2 do no change maerially if a zero- or five-second lag is used insead, as in Bessembinder (200) and Lee and Ready (1991), respecively. 9

12 respecively 1.72% and 1.76% for W4. 8 We invesigae he impac of decimalizaion furher in he nex secion using a mulivariae approach o accoun for oher mixing facors. The hird and final measure is quoed deph or [( bid * bid size + ask * ask size)/ 2] for each valid quoe where size refers o he number of shares bid or offered. To obain he saisics repored in able 2, we firs compue daily averages for each SEO for each window, hen average successively over he days of each window, and finally over he cross-secion of he 667 SEOs. The ime-series behavior of quoed dephs is similar o ha for dollar volumes. The means and medians of he dephs increase significanly for he announcemen window (W2), decline during he closing when compared o heir announcemen levels, and decline furher during he pos-closing window bu remain near or above hose observed during he pre-announcemen window. A window-o-window comparison of mean and median dephs pos- compared o pre-decimalizaion indicaes ha dephs are significanly lower (generally a < 1%) pos-decimalizaion. Resuls of similar ess for he six SEO ypes are repored in able. All mean dollar volumes are significanly differen a < 1% and more han 50 percen higher in he announcemen (W2) versus preannouncemen window (W1). Excep for he no bough deal mean, dollar volume ends o decline in he closing (W) versus announcemen window (W2). Excep for privae placemens and bough deals, dollar volume ends o decline in he pos-closing (W4) versus closing window (W). Mean (and especially median) dollar volumes are much smaller for privae SEO placemens, which may be explained by he paricularly small mean sizes of privae placemens compared o heir public counerpars ($26.67 million versus $8.92 million) along wih relaively low public floa ha is no alered following he closing of he issues due o share lockup requiremens. Similarly, mean issue sizes are larger for bough deals versus all oher SEO ypes ($74.85 million versus $62.8 million) and smaller for resource versus no resource firms ($41.22 million versus $88.67 million). 8 The mean values for hese wo sub-samples are higher han ha for he full sample because he full sample includes he SEOs where he sar of decimalizaion (April 15, 1996) is included in one of heir four windows (i.e., W1 o W4). The mean proporional effecive spread for W1, for example, for his unrepored sub-sample is much lower a 1.0%. 10

13 [Please place able abou here] Wih respec o proporional effecive spreads and quoed dephs, privae placemens are consisenly he highes and lowes, respecively, among he six SEO ypes. Privae placemen spreads a 2.45% in he pre-announcemen window are relaively large and decline hrough he closing window before rebounding o 2.1% in he pos-closing window. Bough deals and no resource issues are very similar in heir mean and median spreads over he four windows alhough bough deals are larger in erms of issue size (mean of $62.8 million versus $41.22 million). Size appears o be a possible explanaion for any differences in spread changes over he issuance cycle for hese six ypes of SEOs. For example, bough deals end o be more liquid han no bough deals probably due a leas in par o he primarily insiuional buyers of he SEOs of firms ha issue bough deals. In he nex secion, we conrol for facors such as size ha may explain he cross-secion of bid-ask spreads across SEO ype o deermine he incremenal effec of each sage in he SEO issuance cycle. The main conclusion drawn from he findings repored in his secion of he paper is ha liquidiy dramaically increases a SEO announcemen. Liquidiy improves furher upon issue closing if proxied by effecive spreads and deerioraes slighly using quoed dephs or rading volumes bu remains higher han is pre-announcemen level. Depending upon he proxy used, liquidiy deerioraes pos-closing and remains a or above is pre-announcemen level. 4. MULTIVARIATE EXAMINATION OF THE DETERMINANTS OF SEO SPREADS We now use a mulivariae approach o es for he effec of specific cross-secion feaures on SEO quoed and effecive spreads including SEO ype, TSX decimalizaion, and he reducion in he lockup period for privae placemens. Since i is used as a conrol for spreads, he cross-secional feaures of dollar raded volume are no examined. We explain he cross-secion variaion in quoed spreads using he size, rading volume and volailiy of reurns of he firms issuing he SEOs, as used in a differen conex by Bessembinder (1999) and Soll (2000). A panel daa approach is used o solve for he missing variables problem, quanify spread changes as he SEO issuance cycle unfolds, and disinguish beween 11

14 he spread effecs of paricular cross-secional feaures such as SEO ype. The four windows for each SEO issuance cycle represen he ime dimension of our panel. The specific panel model esimaed is: Spreadi, = β1 Ln( Volume) i, + β2volailiyi + βln( Size) i+ β4decimalizaioni + β5privaei, (1) + β Pos Privae + β Bough + β Resources + α + γ W + γ W + γ W + ε 6 i 7 i 8 i i i, where Spreadi, is he proporional quoed or effecive spread for SEO i during window. Ln( Volume ) i, is he naural logarihm of he average daily raded dollar volume for SEO i during window. Volailiyi is he sandard deviaion of daily reurns exraced from he CFMRC daabase for SEO i over he four windows. Ln( Size) i is he naural logarihm of he oal asses for SEO i, which was exraced for mos (608) of he SEOs from he las balance shee disclosed before each SEO announcemen from various daabases including Compusa, Sockguide, EDGAR and SEDAR (he Canadian equivalen o EDGAR). zero oherwise. Decimalizaion i is a dummy variable ha is equal o one if SEO i is pos-decimalizaion and Privae i, Pos Privae i, Boughi and Resources i are dummy variables ha are equal o one if SEO i was respecively privaely placed, privaely placed afer he November 2001 lockup period reducion, issued as a bough deal and issued by a resource company and zero oherwise. αi is he (random) cross-secional effec for SEO i. W 2, W and W4 are dummy variables ha are equal o one, respecively, for windows wo (during announcemen), hree (during closing) and four (pos-closing) and zeros oherwise. The γ 1, γ 2 and γ parameers capure he change in he bid-ask spreads as he SEO cycle unfolds and can be hough of as a ime dimension random effec. For insance, γ 1 and γ 2 measure he incremenal effec on each spread measure of announcing and compleing he SEO compared o he pre-announcemen level. As some of he regressors are ime-consan explanaory variables, a random raher han a fixedeffec model is used o esimae he parameers of ineres. Summary resuls of panel regressions for boh spread measures wih(ou) he PosPrivae dummy variable are repored in able 4. Whie sandard errors, which are robus o cross-secion correlaions, are 12

15 used in ess of saisical significance. Coefficien esimaes for rading volume are negaive and significan (< 1%) for boh spread measures. This conforms o expecaions, since higher dollar volumes should resul in lower fixed coss per dollar as order processing coss are fixed per ransacion. Coefficien esimaes for reurn volailiy are posiive and significan (< 5%) for boh spread measures. This conforms o he underlying raionale behind he invenory bid-ask spread models where liquidiy providers require higher compensaion wih higher reurn volailiy. Coefficien esimaes for oal asses are negaive as expeced (Bessembinder, 1999) bu no significan a convenional levels for boh spread measures. Some of he effec of firm size may be subsumed by dollar volumes and reurn volailiy ha end o increase and decrease wih firm size. [Please place able 4 abou here.] Only he coefficien esimae for he effec of TSX decimalizaion on effecive spreads of 12 bps for he panel regression wihou he PosPrivae dummy variable is significan (10% level). This resul is consisen wih findings repored earlier in able where, for example, average effecive spreads are equal a 192 bps boh pre- and pos-decimalizaion during he pre-announcemen window. The esimaed coefficien for he privae placemen dummy variable for quoed spreads is a significan (< 1%) 22 and 25 bps wih and wihou he inclusion of he PosPrivae dummy variable. The esimaed coefficien for is counerpar for effecive spreads of 11 bps is only significan (<10%) wih he inclusion of he PosPrivae dummy variable. Thus, he 77 bps univariae difference in he mean proporional effecive spreads beween privaely and publicly placed SEOs during he pre-announcemen window in able is o a large exen explained by he oher facors accouned for in our regression. The coefficien esimaes for he PosPrivae dummy variable are negaive bu insignifican. The esimaed coefficiens of 10 and 7 bps for he bough deal dummy variable for quoed and effecive spreads, respecively, are significan. As underwriers are expeced o be beer informed han he general public, heir willingness o secure each deal sends an indicaion o he marke of heir faih in hese SEOs and reduces he adverse selecion componen of rading coss. The esimaed coefficiens of 27 (26) and 18 bps for he naural resource dummy variable for quoed and effecive spreads, respecively, 1

16 wih(ou) he PosPrivae dummy variable are significan. This is consisen wih he noion ha he ype of informaion released by naural resource firms is no only complex bu ofen requires echnical experise o correcly inerpre heir (un)favorableness (e.g., a drill-core assay resul). The remaining parameer esimaes repored in able 4 relae o he change in spreads as he SEO cycle unfolds. The coefficien esimaes for W2 or γ 1 in able 4 indicae ha quoed spreads are reduced by a significan 12 bps and effecive spreads do no change significanly upon SEO announcemen. The coefficien esimaes for W or γ 2 in able 4 indicae ha quoed and effecive spreads are significanly lower by 40 and 1 bps, respecively, upon closing compared o heir pre-announcemen levels. Based on a Wald es for a Chi-square disribuion wih one degree of freedom, he differences in he esimaed coefficiens (i.e., γ1 γ 2in able 4) around SEO closings versus SEO announcemens for quoed and effecive spreads correspond o Chi-square saisics of and 72.75, respecively. Is single linear resricionγ 1 γ 2 = 0 is rejeced for boh regressions a convenional confidence levels. Thus, quoed and effecive spreads are significanly lower by 28 and 0 bps around SEO closings versus SEO announcemens. The coefficien esimaes for W4 or γ in able 4 indicae ha he quoed and effecive spreads are significanly lower by 11 and 7 bps, respecively, relaive o heir pre-announcemen levels. The significan esimaed coefficiens of he differences for boh spreads given by γ 2 γ in able 4 indicae ha quoed and effecive spreads are higher by 29 and 24 bps respecively for SEO pos-closing versus SEO closing. Thus, we find ha boh spread measures follow an approximae V-shaped paern hrough he SEO issuance cycle. Spreads decrease marginally upon announcemen, fall more srongly as he SEO is closed, and increase pos-closing o a level ha is sill significanly lower han pre-announcemen. Therefore, we conclude ha SEOs resul in posiive spread cos benefis for shareholders ha exend beyond SEO closing. In he nex secion, we examine which spread componen(s) are he source of he enhanced liquidiy as measured by oal spreads around SEO issuance. 14

17 5. DECOMPOSITION OF SEO BID-ASK SPREADS The bid-ask spread conains wo main componens; namely, a emporary cos (composed of order processing and invenory coss) and a permanen or adverse selecion componen ha persisenly impacs he value of he raded asse o compensae he uninformed for rading agains privaely informed raders. As discussed previously, Brooks and Pael (2000) repor a significan decline in adverse selecion coss for equiy SEOs for individual days in he even window cenered on he announcemen dae compared o he pre-even window. Unlike Brooks and Pael (2000), we invesigae changes in spread componens for he enire SEO issuance cycle ha includes he closing daes for (non-)differeniaed samples o address hree quesions. Firs, how does each spread componen vary over he four windows for each SEO? Second, are hese wo spread componens influenced by he srucural breaks caused by he TSX s move o decimalizaion on April 15, 1996 and by he required holding period reducion for privae placemens on November 0, 2001? Third, do he spread componens for he six SEO ypes vary individually across he even windows and wih respec o heir paired ypes (i.e., privae versus public, bough versus no bough and resource versus no resource) wihin each even window? 5.1 Spread Decomposiion Mehodology Alhough we use he Lin e al. (LSB) (1995), Glosen and Harris (GH) (1988) and Neal and Whealey (1998) models because hey are based on quie differen assumpions, we repor resuls for he LSB model due o he similariy of he resuls. Of he hree models relaed o he realized spread ha were considered, we chose he LSB model over he unpublished Masson (1995) model ha is used by Pael and Brooks (2000) because he LSB model also considers rades ha occur wihin he bid-ask spread. From he numerous indicaor models, we chose he GH model because i has a role for volume, has less parameers o esimae han he Madhavan e al. (1997) model, and due o he findings of Brooks and Masson (1995) ha models such as Soll (1989) and he adaped version by George e al. (1991) suffer from a generaed regressor problem. The NW model used by us is a modificaion of he George e al model. 15

18 Wih all erms as previously defined, he proporion of he spread due o adverse selecion or λ is esimaed for our sample of 667 TSX lised SEOs using sacked inra-day daa for each SEO window in he following regression for he LSB model: ( M M ) = LN( P M ) e ΔM = LN 1 λ +. (2) Alhough we expec boh spread componens o follow an approximae V-shaped paern over he SEO issuance cycle, we expec he imings of heir minimum values o differ. Since emporary coss are affeced by various facors, such as porfolio rebalancing and underwrier marke sabilizaion, we expec lower emporary coss due o increased rader aciviies o occur in he SEO announcemen windows and o reach heir minimums in he SEO closing windows when he marke weighs of SEO issuers change. In conras, we expec minimum adverse selecion coss o occur in he announcemen windows since SEO announcemens are generally more unexpeced informaion evens han are SEO compleion announcemens Empirical Resuls The spread componens for and heir differences beween even windows for he (non-)differeniaed samples are repored in able 5. A cursory examinaion of able 5 suggess ha mean/median emporary coss are always beween wo and hree imes he corresponding values for adverse selecion coss for same-window comparisons wihin each SEO ype. The relaively more liquid ypes of SEOs (such as public offerings, bough deals due o heir larger issue sizes and no resource issues) have consisenly lower emporary spread coss compared o same-window levels for heir counerpars (i.e, privae placemens, no bough deals and resource issues, respecively). [Please place able 5 abou here] The wihin-seo caegory of window o window changes in mean/median emporary and adverse selecion coss exhibi several similariies. Boh cos componens generally decline significanly in he announcemen (W2) versus pre-announcemen window (W1). Temporary coss due o he increase in 9 Frijns e al. (2006) repor ha abou eigh and fifeen percen of heir iniial and final samples of U.S. SEOs issued beween 1984 and 2000 were wihdrawn. 16

19 rading aciviy usually decline significanly during he closing hree-day window (W) versus eiher W1 or W2, and rebound significanly during he pos-closing window (W4) bu remain significanly below heir pre-announcemen levels. Temporary cos differences condiioned on TSX decimalizaion are always significan in he preannouncemen window (W1). Significan differences predominanly a < 1% wihin each of he hree remaining windows are observed for all paired caegories. Thus, he emporary cos in each of hese hree windows is larger in he pre- versus pos-decimalizaion period for resource versus no resource issues, for bough versus no bough deals and for privae versus public offerings. 10 Significan differences exis in he same-window adverse selecion coss for all four windows and for he hree pairs of SEO ypes (a < 1%) bu no for he pre- versus pos-decimalizaion periods. Neiher average adverse selecion cos changes significanly beween SEO announcemen and closing daes. Similar o emporary coss, adverse selecion coss rebound pos-closing bu remain marginally lower han heir values pre-announcemen. 11 Differences in he levels for like window comparisons beween he subsamples include public offerings, which have an average pre-announcemen adverse selecion cos of 51 bps while privaely placed SEOs have a corresponding cos of 91 bps. Thus, firms ha proceed o issue via privae placemens face higher adverse selecion coss when compared o heir public placemen counerpars. Similarly, firms ha use bough deals versus hose ha did no have on average a 52 insead of a 71 bps adverse selecion cos pre-announcemen. Wihou furher esing using a mulivariae framework, his suggess ha eiher bough deals are associaed wih lower adverse selecion as we argued earlier due o srong underwrier signals or ha firms ha use bough deals already face lower adverse selecion due o heir larger size. We now use a mulivariae approach similar o ha described in secion four o accoun for commonaliy. The approach is formally saed in he following wo separae models: 10 Same window ess beween pre- and pos-decimalizaion periods and beween similar SEO ypes are discussed herein bu are no abulaed. 11 For privae placemens, he mean permanen cos of 91 bps does no fall significanly upon announcemen based on he paired -ess while he median declines significanly based on he Wilcoxon es. 17

20 SpTemp = c + β Ln( Volume) + β Volailiy + β Decimalizaion + β Privae + β PosPrivae i, 1 i, 2 i 4 i 5 i 6 i + β Bough + β Resources + α + γ W + γ W + γ W + μ 7 i 8 i i i, () SpAdverse = c + β Ln( Size) + β Decimalizaion + βprivae + βposprivae i, i 4 i 5 i 6 i + β Bough + β Resources + α + γ W + γ W + γ W + ξ 7 i 8 i i i, (4) where SpTempi, and SpAdverse i, are respecively he emporary and adverse selecion spread coss for SEO i during window as esimaed using he LSB mehodology; and he remaining variables are as described earlier in secion four. Volume and volailiy are used as regressors for he emporary componen as hey relae o he order processing and invenory spread componens. Size is used in he adverse selecion cos componen regression due o is relaionship wih informaion asymmery. As expeced and based on columns -4 and 7-8 of able 6, emporary cos is srongly and negaively relaed o volume. Since mos of he order processing cos is fixed per rade, is value per raded dollar falls as rading volume increases. Similarly, emporary cos is significanly and posiively relaed o volailiy as expeced based on he invenory cos componen heory. Since heir risk of ruin is higher wih higher volailiy, marke makers emporarily increase heir spreads and uninformed invesors are more likely o place limi over marke orders. Surprisingly, he emporary cos is no significanly lower following TSX decimalizaion. Unlike he findings for he simple univariae ess conduced earlier, he emporary coss for privaely versus publicly placed SEOs and for he pos- versus pre-window reducion in he holding period in 2001 for privae placemens are no longer significanly differen. Significanly lower emporary coss are observed for bough versus no bough deals and for firms in no resource versus resource secors. As for changes hrough he SEO issuance cycle, we find ha he emporary cos falls by 6 bps upon SEO announcemen and coninues o fall by an addiional and significan 28 bps (a <1%) around SEO closing. Subsequenly, he emporary cos increases by a significan 25 bps bu neverheless remains nine bps significanly lower when compared o he pre-announcemen window. Columns 5-6 and 9-10 of able 6 conain he panel regression resuls for he adverse selecion cos componen. As expeced, adverse selecion cos is significanly lower for larger firms which are generally 18

21 followed by more financial analyss and invesors. The adverse selecion cos componen is lower by a significan seven bps (six wih he inclusion of PosPrivae) for bough versus no bough deals, and is higher by a significan 14 bps (15 wih he inclusion of PosPrivae) for resource versus no resource issuers. Unlike he case for he oal spread in secion four, he coefficien esimae for he Posprivae dummy for he adverse selecion cos is a significan -1 bps. Thus, reducing he lock-up period conribues o a reducion in informaion asymmery. As for he changes hrough he SEO issuance cycle, he adverse selecion cos falls by 11 bps upon announcemen ( γ 1 in able 6) and increases by hree bps for SEO closing versus SEO announcemen ( γ 1 γ 2 in able 6), and hence remains eigh bps significanly lower han is pre-announcemen value. The adverse selecion cos increases by a furher hree bps posclosing versus closing. As for he oal spread, boh spread componens follow approximae V-shaped paerns wih differen minimums (see figure 1). The lowes poin for he emporary cos is around closing when liquidiy raders are likely o be he mos acive. In conras, he lowes poin for he adverse selecion cos occurs upon SEO announcemen where mos of he informaion asymmery is reduced hrough he public disseminaion of informaion ha a SEO is imminen. [Please place figure 1 abou here.] 6. SEO RETURNS In his secion, we es he valuaion impacs of SEO announcemens and closings, including any associaed changes in liquidiy, and we es for volailiy changes upon SEO closing and he impac of any such volailiy changes on increased liquidiy (as was documened earlier). Hess and Fros (1982) conclude ha he negaive abnormal reurns (ARs) associaed wih SEO announcemens of NYSE-lised firms do no exceed ransacion coss. Asquih and Mullins (1986) find ha over 80% of he indusrial firms in heir SEO sample exhibi negaive ARs. Masulis and Korwar (1986) find ha he negaive SEO price effecs upon announcemen are larger for indusrial firms han for 19

22 uiliies. Slovin e al. (1994) find an average AR of -2.9% for he firs SEO afer a NASDAQ IPO, and an inverse relaionship beween ARs and he proporion of shares sold by insiders. 12 Mioo (1997) finds significanly negaive abnormal reurns for SEO announcemen daes for a sample of 108 SEOs issued beween 1982 and 199 by Canadian firms lised in he TSX-00 index. Ignoring he condiional naure of reurn volailiy can lead o misleading inferences in even sudies. For example, Kryzanowski and Zhang (199) repor ha modelling condiional residual variances as ARCH processes maerially affecs esimaes of ARs and sysemaic risk for sock-spli ex daes. Similarly, Corhay and Rad (1996) find subsanially differen AR esimaes for a sample of divesiures using a marke model incorporaing GARCH effecs. Tradiionally, SEO ARs are examined for he announcemen window under he resricive assumpion of consan sysemaic risks, residual reurn volailiies and liquidiy reurn premiums hrough he SEO issuance cycle. To alleviae his possible shorcoming, we now use six marke models (see Appendix 1) ha allow for ime-varying residual volailiies hrough a GARCH (1,1) specificaion, for shifs in sysemaic risk (denoed by β * ) in hree of he six models, and for he possibiliy ha liquidiy is no only a priced facor bu also may change significanly around boh SEO even daes in wo models ( and 6). 1 Allowing for ime-varying volailiy also provides us wih he opporuniy o es wheher SEOs direcly impac volailiy. For ess repored in his secion of he paper, our iniial sample of 756 SEOs is reduced o a final sample of 65 SEOs afer we delee SEOs ha do no have a leas 200 daily reurns in he CFMRC daabase from one calendar year before SEO announcemen o one calendar monh afer SEO closing. The marke porfolio is proxied by he value-weighed global CFMRC index ha includes common shares 12 Oher non-canadian sudies include Parsons and Raviv (1985), Hull and Kerchner (1996) and Slovin e al. (2000); and Canadian sudies include Mioo (1996, 2006) and Kryzanowski and Rubalcava (2004). 1 The resuls from models using condiional volailiy specificaions [such as he asymmeric EGARCH and GARCH of Glosen e al. (199)] are no abulaed o save valuable journal space since hey generae similar resuls. Building on he seminal work of Amihud and Mendelson (1986), various sudies such as Acharya and Pedersen (2005) and He and Kryzanowski (2006) find ha liquidiy is priced in U.S. and Canadian markes, respecively. 20

23 wih prices over $2 (consisen wih our SEO selecion filer). Risk-free ineres raes, as proxied by yields on one-monh, consan-mauriy Canadian Treasury Bills, are exraced from Daasream. An examinaion of able 7 reveals ha he mean and median measures of sysemaic risk ( β ) for he pre-announcemen periods are less han one and consisenly significan for all models a < 1% level. The * change in sysemaic risk ( β ) beginning wih he announcemen window is no significanly differen from zero (see models 1-). Consisen wih he lieraure, average ARs for he announcemen window ( κ 1 ) are negaive for all models bu only significanly differen from zero for he medians (a < 1%). Similarly, he average ARs during he closing window ( κ 2 ) from models 1 and 4 are negaive and only he medians are marginally significan (a < 10%). These ARs appear o be consisen wih he magniudes of he reducions in informaion asymmeries caused by hese SEO evens, as repored in previous secions of his paper. [Please place able 7 abou here] All GARCH model average parameer esimaes are significan (a < 5%). Thus, prior-period squared residuals (based on he φ esimaes) and condiional volailiies (based on he ψ esimaes) have significan and posiive impacs on he esimaions of he conemporaneous condiional volailiies (h). The poin esimae for condiional volailiy (ω ) is significanly posiive. Condiional volailiy decreases, as capured by a significanly negaive κ, wih he parial resoluion of uncerainy pos-announcemen. This is consisen wih our earlier finding ha he emporary spread cos is lower afer SEO closing versus pre-announcemen for all caegories. Since volailiy is he main driver of he invenory spread componen, liquidiy improves ceeris paribus as he non-sysemaic componen of volailiy falls. The esimae of he quoed spreadθ, which is he proxy for liquidiy (S) in models and 6, is negaive bu no significan. Along wih he similariy of resuls across models, his suggess ha we need only discuss he resuls for model 1 (he mos general of he remaining four specificaions) for he samples differeniaed by TSX decimalizaion and SEO ype. Based on able 8, he esimaed beas are 21

24 less han one and significanly differen from zero, and are significanly differen beween pre- and posdecimalizaion windows (smaller) a < 1% level. Significan differences are idenified for he esimaed beas beween public offerings (smaller) and privae placemens (significan a < 5%) and beween resource (larger) and no resource SEOs (significan a < 1%). Marginally significan differences (only for means a < 10%) are found for he esimaed beas beween bough deals (smaller) and no bough deals. [Please place able 8 abou here] Some significan differences exis in he ARs during he announcemen window ( κ 1 ) by SEO ype. Mean (median) ARs are negaive and significan a < 5% (< 1%) pre- and pos-decimalizaion. Only he median AR for public SEOs is significan a he 5% level (and negaive), and neiher average (posiive) AR for privae SEOs is significan a he 5% level. Only he median AR for bough deal SEOs is negaive and significan a he 5% level, while boh average ARs are negaive and significan a he 5% level for no bough SEOs. The median (no mean) ARs for boh resource and no resource SEOs are negaive and significan a he 5% level. The differences in heir respecive averages for W2 are significanly differen a he 5% level for public versus privae SEOs bu no for pre- versus pos-decimalizaion, bough versus no bough and for resource versus no resource. Similarly, only he difference in he mean ARs beween bough and no bough SEOs is significan a he 5% level for he closing window. 14 Finally, esimaed GARCH parameer values are generally significan a he 5% level individually. Excep for esimaes ofφ (he coefficien of he prior-period squared residuals), which are significanly differen in ess of pos- versus pre-decimalizaion and resource versus no resource issues, heir differences are no significan in he oher paired comparisons. 7. CONCLUDING REMARKS 14 The numerical oupu for saisical ess on he sum of he ARs in he announcemen and closing windows (i.e., κ1+ κ2) are no presened since heir inference did no differ from ha for he announcemen window when significance was presen. 22