Speed and consequences of venture captalst post-ipo ext Imants Paegls * and Paranen Veeren ** Ths verson: January, 2010 * John Molson School of Busness, Concorda Unversty, 1450 Guy St. Montreal, QC, H1H 1L8, Canada. e-mal: paegls@jmsb.concorda.ca. ** John Molson School of Busness, Concorda Unversty, 1450 Guy St. Montreal, QC, H1H 1L8, Canada. e-mal: p_veeren@jmsb.concorda.ca. For helpful dscussons, we thank Nlanjan Basu, Sandra Betton, and Rahul Rav. The authors gratefully acknowledge research fundng from the Insttut de Fnance Mathematque de Montreal (IFM 2 ) and the Socal Scences and Humantes Research Councl of Canada (SSHRC). We thank Aja Groza, Anna Kaolna, and Charles Martneau for excellent research assstance. We reman solely responsble for any remanng errors.
Speed and consequences of venture captalst post-ipo ext Abstract We examne the determnants of the speed of venture captalst post-ipo ext decson and ts nfluence on frm value. We hypothesze that the speed of VC ext wll be nfluenced by founder ownershp, whch both mpedes lqudty of a frm s stock and sgnfcantly nfluences post-ext frm value. Our results suggest that ths s ndeed the case. In partcular, we fnd a concave relatonshp between founder ownershp and the speed of venture captalst ext and a convex relatonshp between founder ownershp and changes n frm value around VC ext. The observed relatonshps are robust to alternatve econometrc specfcatons and to the presence of non-founder blockholders.
Speed and consequences of venture captalst post-ipo ext 1. Introducton Ext decson s an mportant part of a venture captalst s nvestment strategy. The exstng lterature has almost exclusvely focused on the choce between varous forms of ext (e.g., IPOs, mergers, and wrte-offs) and has treated IPO as an ext event. 1 It s, however, only after the IPO that venture captalst (VC) has to decde on the speed of hs ext. Ths decson, and especally ts consequences for the frm, have receved very lttle attenton n the lterature (for notable exceptons, see Baker and Gompers, 1998; Ln and Smth, 1998; and Luo, 2005). In ths paper, we fll ths gap n the lterature by examnng the relatonshp between the speed of venture captalst post-ipo ext (and changes n frm value ntated by t) and founder ownershp. 2 We hypothesze that founder ownershp wll nfluence the speed of VC ext for two reasons. Frst, founder ownershp s lkely to mpede lqudty of ther frm s stock, makng t harder for venture captalsts to ext (see Brockman et al., 2008). Ths negatve nfluence wll be most pronounced n frms wth hgh levels of founder ownershp. Snce n llqud markets the venture captalst wll be better off sellng hs block n peces (rather than at once), the lqudty consderatons mply a negatve relatonshp between founder ownershp and the speed of VC ext. In addton, the lqudty and ts changes due to the VC ext wll also nfluence frm value n the followng manner. The sale of venture captalst ownershp stake wll ncrease the lqudty of the frm s stock. Ths, n turn, wll lead to decreases n the lqudty premum and the stock s expected return and to an ncrease n the stock prce. Ths ncrease wll be most pronounced for llqud frms (.e., those wth hgh levels of founder ownershp). From now on we wll refer to 1 See, e.g., Cummng (2008). 2 Our focus on founder ownershp s motvated by the followng reasons. Frst, founders are by far the most mportant ndvdual blockholders n the newly publc frms. Second, emergng lterature on famly frms suggests that foundng famles are a unque group of blockholders that has a dstnct nfluence on frm value (see, e.g., Anderson and Reeb, 2003; and Vllalonga and Amt, 2006). We wll provde evdence on these two conjectures later n the paper and wll show that the relatonshps descrbed below are ndeed unque to founders. 1
ncreases n frm value stemmng from ncreased lqudty as lqudty effect. In summary, we expect a postve relatonshp between founder ownershp and the lqudty effect. Second, changes n a frm s ownershp structure (and, consequently, n the founder s ncentves) brought about by the departure of venture captalsts are also lkely to nfluence the speed of VC ext. 3 In partcular, upon the venture captalst ext, the founder s lkely to become the major shareholder n the frm. The nfluence of ths change n the ownershp structure on frm value (whch we wll refer to as ncentve effect) depends largely upon the level of founder ownershp. In partcular, frms wth ntermedate levels of founder ownershp wll have the largest (negatve) ncentve effect. 4 In ths range, the founder s ownershp s hgh enough to ensure hm a control of the frm (and thus lmt the effectveness of the market for corporate control) and yet too low to create an algnment of hs ncentves wth those of mnorty shareholders. At the same tme, the magntude of the ncentve effect wll be smaller for frms wth low and hgh levels of founder ownershp. In low founder ownershp frms the founder s ncentves wll be kept n lne wth those of mnorty shareholders by the market for corporate control. In hgh founder ownershp frms, the level of ownershp wll better algn nterests of founder and those of mnorty shareholders. In addton, VC ext from these frms wll leave ther ownershp structure relatvely unchanged (.e., the founder has been the major shareholder n the frm even n the presence of venture captalsts). In summary, we expect a convex relatonshp between founder ownershp and the ncentve effect of VC ext. The overall relatonshp between founder ownershp and the speed of venture captalst ext wll depend upon the relatve strength of the two effects dscussed above. In partcular, for 3 It should be noted that there s a dfference between the venture captalst certfcaton of founder qualty (see, e.g., Hellman and Pur, 2002) and the founder s ncentves after the departure of VCs. The certfcaton of qualty refers to the founder s abltes, background, and experence whle the ncentve effects dscussed n ths paper are largely related to the level of founder ownershp. 4 For the ease of exposton, from now on we wll refer to three dstnct ranges of famly ownershp. Hgh famly ownershp wll refer to ownershp close to 100%. Intermedate (low) famly ownershp wll refer to cases where the foundng famly holds a (less than a) controllng stake. A controllng stake s defned as the level of ownershp necessary to mantan control of the frm. Whle there has been a sgnfcant debate n the fnance lterature about the level of ownershp necessary to mantan the control of the frm, no consensus has been reached and the ssue has remaned an emprcal matter. 2
frms wth hgh levels of founder there wll be a large lqudty effect and a small ncentve effect, mplyng a slow speed of ext and large stock prce ncreases (smaller decreases) around the ext. As founder ownershp decreases, so wll the lqudty effect. At the same tme, the (negatve) ncentve effects wll be ncreasng. Therefore, the net mpact for frms wth ntermedate levels of founder ownershp s lkely to be smaller (relatve to frms wth hgh levels of founder ownershp). The potental prce gan, f any, from a partal ext wll make a complete ext a more attractve opton for the venture captalst. Fnally, for low founder ownershp frms both effects declne and the net mpact s an emprcal matter. If the lqudty effect domnates (s domnated by) ncentve effect, we expect a slower (faster) speed of VC ext and hgher (lower) stock prce reacton for these frms. Overall, we expect a negatve (concave) relatonshp between founder ownershp and the speed of VC ext f the ncentve (lqudty) effect domnates for frms wth low levels of founder ownershp. Smlarly, we expect a postve (convex) relatonshp between founder ownershp and changes n frm value around the venture captal ext. We test the above hypotheses usng a sample of all US venture-backed IPOs between 1993 and 2000. We track the venture captalst and founder ownershp n these frms untl the thrd proxy statement after the lockup expraton. In partcular, for each proxy statement we calculate the percentage change n the number of shares held by venture captalsts between the IPO and the proxy flng date, whch we use as our measure of the speed of VC ext. A hgher decrease n the VC holdngs ndcates a faster speed of ext. Our results can be summarzed as follows. Frst, we fnd a sgnfcant and concave relatonshp between founder ownershp and the speed of VC ext. Our results mply that VCs ext frms wth ntermedate levels of founder ownershp faster than they do those wth ether low or hgh levels of ownershp. Second, we fnd that VC ext nfluences frm value. In partcular, we fnd that frms wth ntermedate levels of ownershp experence the largest decrease n frm value around VC ext. Thrd, we fnd a concave and statstcally sgnfcant relatonshp between founder ownershp and lqudty of a frm s stock. Also, frms wth hgh levels of founder 3
ownershp experence larger ncreases n lqudty around the venture captalst ext. Fnally, we present addtonal two peces of evdence consstent wth entrenchment of founders wth ntermedate levels of ownershp. We fnd that frms wth ntermedate levels of ownershp are less lkely to attract nsttutonal blockholders. Ths leads to a lower degree of founder montorng n such frms. Further, we also fnd that frms wth ntermedate levels of ownershp ssue a lower fracton of shares n ther IPOs. We argue that ths evdence s consstent wth the founders of such frms beng protectve of ther ownershp postons (whch are crucal for the mantenance of ther control over the frm and control benefts thereof). Our results are robust to alternatve econometrc specfcatons. In addton, we show that the mpact of founder ownershp on the speed of VC ext and the value changed ntated by t s unque to founders and s not observed for ndvdual blockholders n general. The results regardng changes n frm value after VC ext descrbed above may ndcate a potental market rratonalty. In partcular, f the decrease n frm value documented above s predctable, as we have shown t to be, t s surprsng that such a decrease s not already ncorporated n the stock prce before the venture captalst ext. Part of the answer to ths puzzle may be found n the presence of retal nvestors, ther overoptmsm, and short sale constrants. In partcular, as reported above, frms wth ntermedate levels of ownershp have the lowest nsttutonal nterest and lowest float. Both of these peces of evdence mply that such frms are lkely to have the most bndng short sale constrants. In addton, shares of such frms are held prmarly by retal nvestors who are more lkely to be subject to fads and overoptmsm. Therefore, these frms are lkely to have the hghest potental overprcng n the post-ipo market. We, however, also fnd that frms wth ntermedate levels of founder ownershp have the hghest level of underprcng. Ths mples that such frms (and ther VCs) do not take full advantage of the potental overvaluaton n the post-ipo market. Our paper s related to several streams of lterature. The frst stream of lterature has examned venture captal exts after IPOs. Gompers and Lerner (1998) examnes the market 4
reacton to dstrbutons by venture captal funds. Luo (2005) examnes the relatonshp between the venture captalst post-ipo ext and abnormal accruals at the tme of IPO. There s also an extensve lterature that has examned the returns and tradng volume around lockup expraton (see, e.g., Cao et al., 2004; and Brav and Gompers, 2003). The rest of the paper s organzed as follows. Secton 2 develops the hypotheses that wll serve as the bass for our emprcal tests. Secton 3 revews the characterstcs of our sample. Secton 4 descrbes testng methodology and reports the results. Secton 5 concludes. 2. Theory Consder a venture captalst decdng on the speed of ext from the equty poston n hs portfolo company. We hypothesze that the speed of VC ext wll be an outcome of nteracton between two factors. Frst, the lqudty of a frm s stock wll nfluence the VCs ablty to sell hs shares. Lqudty tself, however, s lkely to be nfluenced by the founder ownershp. Second, the mpact of hs ext on a frm s ownershp structure (and consequently on founder s ncentves) s also lkely to nfluence the speed of hs ext. Thus, ultmately the speed of VC ext wll be related to the level of founder ownershp. We wll now dscuss these two factors (and ther relatonshps wth founder ownershp) n turn and then wll dscuss the nteracton between them. 2.1. Lqudty consderatons The lqudty of a frm s stock s lkely to have a sgnfcant nfluence on venture captalst s ablty to sell hs shares n the open market. To sell a large block of shares n an llqud market, the venture captalst has to face a lqudty dscount (.e. the lqudaton value of the block wll be lower than the market prce prevalent before hs trade). 5 In such a settng, the 5 For the ease of exposton, from now on we are assumng that the venture captalst hmself s sellng the block of shares n the open market. In practce, as reported by Gompers and Lerner (1998), n a majorty of cases the venture captalst dstrbutes hs shares to the lmted partners who then decde on the sale of these shares. As long as the lmted partners sell soon after the dstrbuton of shares by the VC, the nfluence of 5
optmal response for the venture captalst s to sell hs block n peces (.e., to stage hs ext and therefore to ext hs poston n the frm over a longer perod of tme). 6 Therefore, we expect to observe a slower speed of VC ext n frms wth less lqud stocks. Apart from the mnmzaton of the lqudty dscount dscussed above, the stagng of hs ext s lkely to have an addtonal advantage for the venture captalst. In partcular, VC sales are lkely to nfluence the stock s lqudty and therefore ts expected return and prce. As argued by Amhud and Mendelson (1986), llqud stocks have hgher expected returns due to the lqudty premum. Hgher expected returns, n turn, mply a lower prce for llqud stocks. As sales by venture captalsts ncrease lqudty of the frm s stock, ts lqudty premum and expected return wll decrease, leadng to an ncrease n the prce. In other words, VC ext, apart from any confoundng effects, wll be assocated wth an ncrease n the stock prce. Ths ncrease wll be more pronounced for llqud frms. To summarze, the above dscusson mples that VC ext from llqud frms wll be slower and assocated wth an ncrease n the stock prce. The lqudty of a frm s stock s lkely to be nfluenced by ts ownershp structure (.e., by the presence of blockholders). In a recent paper, Brockman et al. (2008) fnd a negatve relatonshp between the level of blockholder ownershp and lqudty of a frm s stock. 7 Ths suggests that frms wth hgh levels of founder ownershp wll have the most llqud stocks. The above dscusson on the relatonshp between lqudty and the speed of VC ext mples that frms wth hgh founder ownershp are lkely to have the slowest speed of venture captalst ext and to experence the largest stock prce ncreases followng t. Overall, we expect a negatve relatonshp between founder ownershp and the speed of venture captalst ext. 2.2. Changes n founder s ncentves such sales on lqudty and stock prce dscusses below should not be sgnfcantly dfferent from that observed n the case of VC sellng. 6 For the optmal ext strategy n llqud markets, see Subramanam and Yarrow (2001). 7 They fnd that ths relatonshp s drven prmarly by the lack of tradng rather than by the threat of nformed tradng. 6
Venture captalst ext s also lkely to nfluence the ncentves of a frm s founder. In partcular, the departure of venture captalst s lkely to leave the founder as the major shareholder n the frm. The wllngness and the ablty of the founder to use ths reganed status for hs own beneft wll largely depend on hs ownershp. The large lterature on the relatonshp between ownershp and frm value has contended that ncreasng ownershp can have two opposng effects on frm value. Frst, t can make the founder nsenstve to the market for corporate control (.e., to entrench hm) and thus allow hm ncreased consumpton of prvate benefts of control (see, e.g., Stulz, 1988). Second, t can also better algn founder s ncentve wth those of mnorty shareholders. Ths effect stems from the fact that an ncrease n founder s ownershp also ncreases the costs of prvate benefts of control that are borne by hm. 8 Consder three dstnct ranges of founder ownershp. Frms wth low levels of founder ownershp wll be subject to effectve montorng by the market for corporate control. In other words, the founder s ownershp stake wll be too low to ensure (an uncondtonal) control of the frm. For such frms, the venture captalst ext wll not have a sgnfcant detrmental nfluence on frm value. At the same tme, VC ext s lkely to have a negatve nfluence on the value of frms wth ntermedate levels of founder ownershp. In such frms, the ownershp stake of the founder wll be hgh enough to ensure hm control over the frm and yet not hgh enough to algn hs ncentves wth those of mnorty shareholders. To put t dfferently, after the venture captalst ext, the founder of these frms wll be solated from the market for corporate control and wll have a free regn over the frm. Fnally, n frms wth hgh founder ownershp, the level of ownershp wll ensure a greater algnment of founder s ncentves wth those of mnorty shareholders. In addton, n these frms the founder has been the major shareholder of the frm even n the presence of VCs and the latter s ext wll not change the founder s status. Therefore, 8 Despte a large emprcal lterature on the matter, there s no consensus about the domnance of one effect over the other over a partcular ownershp range. The emprcal estmates of the controllng stake vary sgnfcantly from study to study (see, e.g., Morck et al., 1988; McConnell and Servaes, 1990; Hermaln and Wesbach, 1991). 7
we expect ether no or relatvely small loss of value due to the venture captalst ext from these frms. 2.3. Hypotheses The above dscusson mples that VC ext from frms wth ntermedate levels of founder ownershp wll be assocated wth a decrease n stock prce. Due to a hgher level of lqudty of such frm s stock, venture captalsts wll be able and better off sellng ther holdngs at once rather than dong t n a pecewse fashon. Frst, the benefts from ncreased lqudty (and therefore lower expected return) due to hs sellng wll be smaller for these frms (relatve to those n frms wth hgh founder ownershp). Ths s due to lower mpedments to the lqudty by the founder ownershp. Second, there wll be a sgnfcant (negatve) ncentve effect whch wll reduce the attractveness of a staged ext. To put t dfferently, the postve lqudty effect wll be at least partally offset by the negatve ncentve effect. The above dscusson mples that VC wll ext frms wth ntermedate levels of founder ownershp faster (.e., he wll opt for a complete rather than a partal ext). In addton, these frms wll experence the lowest ncrease n the stock prce followng the VC ext (as compared to frms wth ether low or hgh levels of founder ownershp). For frms wth hgh levels of founder ownershp, the lqudty benefts of VC ext are lkely to exceed the (mnmal) losses due to changes n the founder s ncentves. Frst, as mentoned before, these frms are lkely to have the least lqud stocks (due to the hgh level of founder ownershp). Second, the ncentve effect s lkely to be relatvely small for these frms. Therefore we expect to observe the slowest speed of VC ext and the hghest ncrease n stock prce for frms wth hgh levels of founder ownershp. Fnally, for frms wth low levels of founder ownershp, there are two effects at work. Frst, as founder ownershp decreases, so wll the lqudty effect. Ths s due to the fact that these frms are lkely to have been the most lqud ones even n the presence of venture captalsts. 8
Second, the ncentve effect also decreases wth founder ownershp due to the effectve presence of the market for corporate control. The relatve mportance of these two effects wll determne the speed of VC ext and the prce mpact thereof for these frms. In partcular, f the net mpact s postve (.e., the lqudty effect domnates the ncentve effect), frms wth low levels of founder ownershp wll have a slower speed of VC ext and a less negatve stock prce response (as compared to frms wth ntermedate levels of founder ownershp). If, on the other hand, the ncentve effect domnates the lqudty effect, low founder ownershp frms wll have a faster speed of ext and a more negatve stock prce reacton. The domnance of one effect over the other s an emprcal matter. The above dscusson leads to the followng testable hypotheses. If the lqudty effect domnates the ncentve effect, we expect the followng hypotheses to hold: Hypothess 1a: There wll be a concave relatonshp between founder ownershp and the speed of venture captalst ext. Hypothess 2a: There wll be a convex relatonshp between founder ownershp and changes n frm value followng VC ext. If, on the other hand, the ncentve effect domnates the lqudty effect, we expect the followng: Hypothess 1b: There wll be a negatve relatonshp between founder ownershp and the speed of venture captalst ext. Hypothess 2b: There wll be a postve relatonshp between founder ownershp and changes n frm value followng VC ext. 3. Data and sample selecton We start wth all US IPOs of common equty between 1993 and 2000, obtaned from the SDC/Platnum New Issue database. We elmnate REITs, closed-end funds, unt offerngs, equty carve-outs, fnancal (all frms wth SIC codes between 6000 and 6999), utltes, foregn frms, leveraged buyouts, and roll-ups. We also elmnate frms, whch are ether not found n the Center 9
for Research n Securty Prces (CRSP) and COMPUSTAT databases or for whch the CRSP and the SDC databases show dfferent frst dates of tradng. Ths leaves us wth a total of 2,613 frms. Snce the focus of ths paper s on the speed and consequences of venture captalst ext, we are nterested only n the venture-backed frms. Venture captalsts are dentfed from varous ssues of Pratt s Gude to Venture Captal Sources. Of the 2,613 frms, 1,465 had venture captalsts present at the tme of IPO. These frms then represent our ntal sample. Panel A of Table 1 shows how we arrved at ths sample. We then track these frms up to the thrd proxy statement after the lockup expraton. Panel B of Table 1 reports the dstrbuton of our sample by post-ipo year. 1,099 of the 1,465 frms survve untl the thrd proxy statement. Our total sample s comprsed of 5,253 frm-year observatons wth avalable ownershp data. Fnally, Panel C of Table 1 shows how varous data requrements affect our sample. For 5 of these 5,253 frm-year observatons COMPUSTAT data was not avalable, whle n the addtonal 58 cases no CRSP data was avalable for the requred date range. Ths leaves us wth a fnal sample of 5,190 frm-year observatons. We defne a founder frm as a frm n whch the founder or hs/her descendants ether hold at least 5% of the frm s outstandng shares or are actvely nvolved n the management (or governance) of the frm. 9 We dentfy founders of our sample frms usng the nformaton n the management sectons of IPO prospectuses. The data on the founder s (and hs famly s) ownershp s collected from the IPO prospectus and the subsequent proxy statements. Data on varous accountng varables s collected from COMPUSTAT, whle the stock return data comes from the CRSP. Appendx lsts varables used n ths study, whle Table 2 descrbes the evoluton of founder and venture captalst ownershp from the tme of IPO up to the thrd proxy statement after the lockup expraton. At the tme of IPO, founder and hs famly own 13.9 % (8.3%) of the 9 Securtes and Exchange Commsson (SEC) regulatons requre the dsclosure of ownershp postons of (1) all offcers and drectors and (2) all shareholders holdng more than 5% of any class of shares. 10
shares outstandng n an average (medan) sample frm. The correspondng number for the founder frm subsample s 19% (13%). Founders at the tme of IPO are present n 1,077 (73.5%) of sample frms. At the tme of IPO, venture captalsts own 41.2% (39.2%) of the shares outstandng n an average (medan) sample frm. These numbers declne to 8.6% (0.5%) by the thrd proxy statement after the lockup expraton. By the tme of the frst proxy statement after the lockup expraton VCs sell, on average (medan), 34% (23%) of shares held at the tme of IPO. These numbers ncrease to 67% (98%) by the tme of the thrd proxy statement. The percentage of frms wth a complete VC ext also ncreases over tme and reaches 47.9% by the thrd proxy statement. Fnally, Table 3 provdes relevant summary statstcs of dependent and ndependent varables used n our emprcal tests. 4. Emprcal tests and results Ths secton descrbes the testng methodology and reports results of our emprcal tests. We start by conductng unvarate tests of hypotheses descrbed above. The results of these tests are reported n Secton 4.1. We then turn to the examnaton of the relatonshp between the speed of venture captalst ext and founder ownershp n a regresson framework (Secton 4.2). In Secton 4.3 we examne the change n frm value around the venture captalst ext. In Secton 4.4 we provde evdence on changes n lqudty around the VC ext, whle n Secton 4.5 we provde addtonal supportng evdence. 4.1. Unvarate tests and results In ths secton we perform unvarate tests of hypotheses descrbed n Secton 2. We do so by splttng our sample nto two subsamples: non-founder frms and founder frms. We then splt the founder frms nto the followng three subgroups based on the level of founder ownershp. Low founder ownershp frms are defned as those n whch founder and hs famly control less than a thrd of cashflow rghts. Intermedate founder ownershp frms are those n whch the 11
founder owns between 33.3% and 66.7% of shares outstandng and hgh founder ownershp frms are those wth founder ownershp greater than 66.7%. Our measure of the speed of venture captalst ext, VCEXIT, s the percentage change n the number of shares held by venture captalsts between the IPO and proxy flng date multpled by negatve one. We makes ths adjustment (multply by negatve one) to make our measure consstent wth the term speed of ext. Thus, a hgher value of VCEXIT mples a faster speed of VC ext. Our measure s ndependent of the percentage of shares outstandng held by the venture captalsts and allows us to examne sellng by the venture captalsts rather than a jont declne n VC ownershp due to the sellng and the dluton of IPO (as well as follow-on SEOs, f any, and other dlutng events). Note that use of an alternatve measure, the change n the venture captalst holdngs as a percentage of shares outstandng, would be susceptble to the above crtcsm. Wth ths alternatve measure we would observe a declne n VC holdngs even n the absence of any sellng by the venture captalsts purely due to the dluton caused by the IPO and other dlutng events. The results of our unvarate tests regardng the relatonshp between founder ownershp and the speed of venture captalst ext are reported n Panel A of Table 4. We fnd that for an average (medan) non-founder frm the declne n the shares held by VCs s 35% (17%) of the shares they held at the tme of the IPO. For an average (medan) low founder ownershp frm ths declne s 39% (25%). The mean and medan dfferences between these two subsamples are statstcally sgnfcant at the 1% level. The speed of venture captalst ext for ntermedate founder ownershp frms s n-between that of the two subsamples mentoned above and s not statstcally dfferent from ether one. Fnally, we fnd that frms wth hgh levels of founder ownershp have a speed of VC ext smlar to that of the other three subsamples. The results of our unvarate tests regardng the relatonshp between founder ownershp and changes n frm value around the venture captalst ext are reported n Panel B of Table 4. Our measure of the frm value, VALUE, s the sum of market value of equty and book value of 12
long term debt dvded by the sum of book value of equty and long term debt. Consstent wth our conjecture that frms wth hgh levels of founder ownershp have the lowest ncentve effect, we fnd no sgnfcant change n the frm value for these frms. The largest declne n value for the latter frms s consstent wth our conjecture that such frms wll have the greatest value loss due to changes n the founder s ncentves trggered by the VC ext. 4.2. The speed of venture captalst ext We now turn to the examnaton of the relatonshp between founder ownershp and the speed of venture captalst ext n a multvarate settng. We do so by estmatng the followng regresson equaton: VCEXIT = α + α FOWN + α FOWNSQ + α FWEDGE + α LNBVA + α LFAGE 6 0 1 7 2 8 + α TECH + α RRANK + α RESID + α RET _ V + α VCSYND + 3 9 4 10 5 n k= 1 α YD k k + u (1) FOWN s the percentage of cashflow rghts controlled by the founder and hs famly. FOWNSQ s FOWN squared. We control for the nfluence of dual class share structures by usng FWEDGE, defned as the dfference between the percentage of votng rghts controlled by the founder and hs famly, and FOWN. All other varables are as defned n the Appendx. We use year dummes (YD) to control for the varaton n the speed of VC ext over tme. We expect the coeffcent estmate of FOWNSQ to be negatve. The expected sgn of the coeffcent estmate of FOWN depends on whch of the two effects (ncentve or lqudty) domnates for frms wth low levels of founder ownershp. In partcular, f Hypothess 1a holds, we expect a postve coeffcent estmate, whle f Hypothess 1b holds, we expect a negatve coeffcent estmate. The results of the emprcal tests of Eq. (1) are reported n column 1 of Table 5. We fnd that the coeffcent estmate for α 1 s postve, whle the coeffcent estmate for α 2 s negatve. Both coeffcent estmates are statstcally sgnfcant at the 1% level. Our results are consstent 13
wth Hypothess 1a and mply a concave relatonshp between founder ownershp and the speed of venture captalst ext. In partcular, frms wth ntermedate levels of founder ownershp have the fastest speed of VC ext. The dfferences n the speed of VC ext are also economcally meanngful. In partcular, at ts hghest pont, the dfference n the speed of venture captalst ext between founder and non-founder frms reaches negatve 11%. 10 Ths represents around 29% (47%) of average (medan) speed of VC ext n our sample. Ln and Smth (1998) fnd that a vast majorty of cases where lead venture captalsts sell ther shares at the IPO nvolve sales by managers and offcers as well. To control for ths nteracton between founder and VC sellng, we use FSEC_IPO, a dummy varable that takes on a value of one f a frm s founder sells hs shares at the IPO, and zero otherwse, as an addtonal control varable. The results are reported n column 2 of Table 5. We fnd that, whle the founder sellng at the IPO has a postve and sgnfcant nfluence on the speed of VC ext, our results reman qualtatvely unchanged n ths alternatve specfcaton. 11 So far we have not controlled for the level of venture captalst ownershp at the tme of IPO. It s, however, possble that stocks of frms wth a hgh venture captalst ownershp may not be lqud enough to absorb sgnfcant blocks of shares, thus mpedng the VCs ablty to ext fast. The level of VC ownershp at the tme of the IPO s determned jontly wth the level of founder ownershp. Therefore, to examne the nfluence of the sze of VC holdngs at the tme of IPO on the speed of hs ext, we use the two-stage least squares approach. In partcular, we ntroduce VCOWN as an addtonal varable n Eq. (1) and use the followng nstruments: the rato of book value of long term debt to total assets (LEVERAGE), the rato of R&D expendture to net sales (RND), the rato of property, plant, and equpment to total assets (PPE), the natural logarthm of the total amount rased by venture captal funds n the year of the frst venture captal fundng 10 Ths happens at the nflecton pont, whch, based on the results reported n column 1 of Table 5, s at 40.5% founder ownershp. 11 We have also used the VC nvestment duraton as an addtonal control varable and fnd that our results reman qualtatvely unchanged n ths alternatve specfcaton. 14
round, and the percentage of shares held by non-vc nsttutonal blockholders at the tme of IPO. 12 13 All nstruments, except for the fundrasng amount, are measured at the tme of IPO. The results of ths estmaton are reported n column 3 of Table 5. 14 Frst, the relatonshp between founder ownershp and the speed of venture captalst ext remans qualtatvely unchanged n ths alternatve specfcaton. Second, as expected, the coeffcent estmate of VCOWN_IPO s negatve and statstcally sgnfcant, mplyng a slower speed of ext n frms wth hgh levels of VC ownershp. The speed of venture captalst ext s also lkely to be nfluenced by the lqudty of a frm s stock. Lqudty of a frm s stock, however, as dscussed n Secton 2, s also lkely to be related to founder ownershp. Therefore, we examne the nfluence of the lqudty on the speed of venture captalst ext usng the two-stage least squares approach. In partcular, followng Brockman et al. (2008) we use two measures of lqudty: the medan rato of daly tradng volume dvded by the number of shares outstandng, calculated over 100 tradng days endng on the day precedng the proxy statement flng date (TURNOVER) and the natural logarthm of the medan number of trades (as reported by CRSP), calculated over 100 tradng days endng on the day precedng the proxy statement flng date (NTRADES). 15 Followng Fang et al. (2009), we use ndustry medan TURNOVER (IMEDTURN) and ndustry medan NTRADES (IMEDNTRADES) as nstruments. Industry medans are calculated for all CRSP frms n the same year and Fama-French 48 ndustry group. The results of these estmatons are reported n columns 4 and 5 of Table 5. The relatonshp between founder ownershp and the speed of 12 These varables have been used by Baker and Gompers (1999) and Chahne et al. (2007) to explan varaton n IPO-tme ownershp. 13 If the date of the frst venture captal fundng round s not avalable n VentureXpert, we use the natural logarthm of the total amount rased by venture captal funds n the 5 th year before IPO or the year of a frm s foundng, whchever s greater. 14 The decrease n the number of observatons between columns 1 and 3 s due to the presence of frms wth zero sales n our sample whch make t mpossble to calculate our measure of research and development (R&D) ntensty. We have replcated our results usng the R&D expendtures scaled by total assets as an alternatve to our measure. Our results are qualtatvely unchanged n ths alternatve specfcaton. 15 For the tme of IPO, both of these measures are calculated over 100 tradng days startng on the 10 th day after the IPO. 15
venture captalst ext remans statstcally sgnfcant even after controllng for the lqudty of a frm s stock. Snce our focus s on the predctablty of the speed of venture captalst ext, we have used the founder ownershp at the tme of IPO as our man explanatory varable. Founder ownershp, however, s also lkely to change over tme. 16 To examne the robustness of our results to the use of actual post-ipo founder ownershp, we replcate our prevous results usng ths alternatve measure of ownershp. The results of these estmatons are reported n columns 6 to 9 of Table 5. The results reman qualtatvely unchanged wth ths alternatve measure. 4.3. Value changes around the venture captalst ext We now turn to the examnaton of the relatonshp between founder ownershp and changes n frm value around venture captalst ext. We do so by estmatng the followng regresson equaton: VALUE = γ + γ FOWN + γ FOWNSQ + γ FWEDGE + γ FOWN _ VC + γ FOWNSQ _ VC 0 + γ ROA + γ LEVERAGE + γ IMEDVALUE + + γ D _ VCEXIT + γ VCOWN + γ LNBVA + γ LFAGE + γ RESID + γ RND 6 12 1 13 2 7 14 8 3 9 4 n k =1 γ YD k 10 k + ε 5 11 (2) D_VCEXIT s a dummy varable that takes on a value of one for complete exts, and zero otherwse. FOWN_VC s a product of FOWN and D_VCEXIT. FOWNSQ_VC s a product of FOWNSQ and D_VCEXIT. All other varables are as defned n the Appendx. As before, we use year dummes (YD) to control for the varaton n valuaton over tme. Snce the ncentve (lqudty) effect s expected to declne (ncrease) wth the level of founder ownershp, we expect the coeffcent estmate of FOWNSQ_VC to be postve. The expected sgn of the coeffcent estmate of FOWN_VC depends on the effect domnatng at low levels of founder ownershp. In 16 Ths can be due to dluton of founder s ownershp (as a result of the IPO and subsequent, f any, SEOs) as well as the founder s ext from the frm. 16
partcular, f Hypothess 2a (2b) s true, we expect a negatve (postve) coeffcent estmate of FOWN_VC. The results of the emprcal tests of Eq. (2) are reported n column 1 of Table 6. We fnd that the coeffcent estmate for γ 4 s negatve, whle the coeffcent estmate for γ 5 s postve. Both coeffcent estmates are statstcally sgnfcant at the 1% level. Our results mply a convex relatonshp between founder ownershp and changes n frm value around venture captalst ext. 17 The results support our hypothess that frms wth ntermedate levels of founder ownershp have the largest ncentve effect and therefore the largest declne n frm value. Ths declne s economcally meanngful. In partcular, at ts lowest pont, the dfference n the loss of frm value between founder and non-founder frms reaches negatve 1.2. 18 Ths represents around 32.5% (42.5%) of average (medan) sample frm s market-to-book rato (VALUE). In a recent paper, Fang et al. (2009) show that frms wth lqud stocks have hgher value. Therefore, t s mportant to control for lqudty n Eq. (2). Snce lqudty s lkely to be related to founder ownershp, we use the two-stage least squares approach. As before, we use TURNOVER and NTRADES as measures of lqudty and ndustry medans (IMEDTURN and IMEDNTRADES) as nstruments. The results of these estmatons are reported n columns 2 and 3 of Table 6. We fnd that the relatonshp between founder ownershp and changes n frm value remans qualtatvely unchanged even after controllng for lqudty. Also, consstent wth Fang et al. (2009), we fnd that both measures of lqudty have a sgnfcant and postve nfluence on frm value. So far we have used the founder ownershp at the tme of IPO to examne ts nfluence on frm value changes around venture captalst ext. Snce founder ownershp s lkely to change over tme, we examne the robustness of our results to the use of actual post-ipo founder 17 We have also replcated our results usng an alternatve specfcaton of D_VCEXIT based on 75% (nstead of a complete) ext. The results are unchanged wth ths alternatve specfcaton of VC ext. 18 Ths happens at the nflecton pont, whch, based on the results reported n column 1 of Table 6, s at 36% founder ownershp. 17
ownershp. The results of these estmatons are reported n columns 4 to 6 of Table 6. The results are qualtatvely unchanged wth ths alternatve measure of founder ownershp. Snce, as shown above, the choce of the speed of venture captalst ext s a functon of founder ownershp (and therefore endogenous), the coeffcent estmates for D_VCEXIT (and nteractve dummes based on t) n Eq. (2) may be nconsstent. To correct for ths potental selfselecton bas, we use the swtchng regresson model wth endogenous swtchng. 19 In partcular, we estmate the followng model: D _ VCEXIT = Z ' γ + ε * (3.1) ' D _ VCEXIT= 0 _ = Xβ 1 + u (3.2) VALUE 1 ' D _ VCEXIT= 1_ = Xβ 2 + u (3.3) VALUE 2 Equaton (3.1) s the selecton equaton as specfed by Eq. (1). The sample s then splt nto two subsamples based on VC ext (D_VCEXIT). The second-stage equatons, Eqs. (3.2) and (3.3), estmated separately for pre- and post-vc-ext subsamples, respectvely, are specfed as follows: VALUE = γ + γ FOWN + γ FOWNSQ + γ FWEDGE + γ VCOWN + γ LNBVA 0 n k=1 γ YD k + γ LFAGE + γ RESID + γ RND + γ ROA + γ LEVERAGE 6 + γ IMEDVALUE + 11 1 7 2 8 k 3 + ε 9 10 4 5 (4) The model s estmated usng full nformaton maxmum lkelhood. The results of the estmaton of Eqs. (3.2) and (3.3) are reported n Table 7. We fnd that, even after controllng for the possble self-selecton bas, there are stll sgnfcant changes n frm value around the venture captalst ext. In partcular, we test for the equalty of the coeffcent estmates for γ 1 and γ 2 across the two equatons usng a Wald-test (n column 3) and fnd that they are statstcally sgnfcantly dfferent. Our fndngs suggest that the OLS results reported above are not drven by the self-selecton bas. As before, we replcate our results usng actual post-ipo founder ownershp. The results, reported n columns 4 to 6, show that our fndngs are qualtatvely unchanged wth ths alternatve measure of founder ownershp. 19 For a detaled dscusson of ths model, see Maddala (1983). Ths model has been used by Dunbar (1995) and Fang (2005). 18
4.4. Lqudty changes around the venture captalst ext In ths secton we examne the relatonshp between founder ownershp and lqudty as well as ts changes around the venture captalst ext. We do so by estmatng the followng regresson equaton: TURNOVER = γ 0 + γ1fown + γ 2FOWNSQ + γ 3FWEDGE + γ 4FOWN _ VC + γ 5FOWNSQ _ VC + γ 6D _ VCEXIT + γ 7LNBVA + γ8lfage + n + γ10vcsynd + γ11imedturn + γ kydk + ε k =1 γ 9RESID (5) As before, we use year dummes (YD) to control for the varaton n lqudty over tme. We also estmate Eq. (5) usng NTRADES as the dependent varable and replacng IMEDTURN wth IMEDNTRADES. If founder ownershp mpedes lqudty, we expect the coeffcent estmates of FOWN and FOWNSQ to be negatve and sgnfcant. We expect the coeffcent estmate of D_VCEXIT to be postve and sgnfcant, ndcatng an ncrease n lqudty followng the venture captalst ext. Fnally, we expect the coeffcent estmates of FOWN_VC and FOWNSQ_VC to be postve and sgnfcant, ndcatng a larger ncrease n lqudty for frms wth hgh founder ownershp. The results of the estmaton of Eq. (5) are reported n Table 8. Frst, we fnd that there s a concave and statstcally sgnfcant relatonshp between founder ownershp and lqudty of a frm s stock before the venture captalst ext. The results are consstent across the two measures of lqudty. Our results are also economcally meanngful. In partcular, at ts hghest pont, the dfference n TURNOVER between founder and non-founder frms reaches 0.0004. 20 Ths represents around 6.9% (10.4%) of average (medan) sample frm s turnover (TURNOVER). Ths fndng provdes an addtonal support to our conjecture that the postve relatonshp between 20 Ths happens at the nflecton pont, whch, based on the results reported n column 1 of Table 8, s at 16.7% founder ownershp. 19
founder ownershp and the speed of VC ext observed at low levels of ownershp s drven by the domnance of the lqudty effect. The coeffcent estmate of D_VCEXIT s, as expected, postve and statstcally sgnfcant. Ths fndng ndcates an ncrease n lqudty followng the venture captalst ext. Our results ndcate that VC ext ncreases turnover by about 50% (75%) of an average (medan) frm s TURNOVER. Also, as expected, the ncrease n lqudty s hgher for frms wth hgh founder ownershp. In partcular, the coeffcent estmates of FOWN_VC and FOWNSQ_VC are negatve and postve, respectvely. The hgher ncrease n lqudty for frms wth hgh founder ownershp are consstent wth the lack of a sgnfcant negatve change n frm value around the venture captalst ext reported above. 4.5. Addtonal tests and results In ths secton we perform several tests that provde addtonal evdence consstent wth and n support of our conjectures above. In partcular, n Secton 4.5.1 we provde addtonal evdence consstent wth entrenchment of founders wth ntermedate levels of ownershp. Then (n Secton 4.5.2) we dscuss the mplcatons of our fndngs for market ratonalty and venture captalst reputaton. Fnally, n Secton 4.5.3 we provde evdence that our results are unque to founders and do not hold for ndvdual blockholders n general. 4.5.1. Are founders wth ntermedate levels of ownershp entrenched? We have argued that the evdence presented above s consstent wth entrenchment of founders wth ntermedate levels of ownershp. We now present two addtonal peces of evdence consstent wth ths conjecture. We start by examnng the arrval of post-ipo nsttutonal blockholders. Insttutonal shareholders are known to be better at pckng best performng stocks (see, e.g., Feld and Lowry, 2009) and to be actvely nvolved n montorng of ther portfolo companes (see, e.g., Del 20
Guerco and Hawkns, 1999; and Parrno et al., 2003). In addton, nsttutonal blockholders are lkely to take over the montorng of a frm s founder from the departng venture captalsts. If founders wth ntermedate levels of ownershp are ndeed entrenched, as we have argued above, we expect to observe a convex relatonshp between the founder ownershp and the holdngs of nsttutonal blockholders. We examne ths hypothess by estmatng the followng regresson equaton: INSTOWN = γ + γ FOWN + γ FOWNSQ + γ FWEDGE + γ VCOWN + γ LNBVA + γ LFAGE + γ RRANK + γ RESID + 6 0 1 7 2 8 3 n k =1 k γ YD k 4 + ε 5 (6) INSTOWN s the percentage of shares outstandng held by the nsttutonal blockholders (as reported n proxy statements). Snce we do not observe the arrval of nsttutonal blockholders at the tme of IPO, we estmate Eq. (6) usng only post-ipo frm-year observatons. The results of the estmaton of Eq. (6) are reported n column 1 of Table 9. We fnd that there, ndeed, s a convex and statstcally sgnfcant relatonshp between founder ownershp and INSTOWN. In addton, the negatve and sgnfcant coeffcent estmate of VCOWN mples that departng venture captalsts are replaced by nsttutonal blockholders. Next, we explctly allow for the change n the relatonshp between founder ownershp and INSTOWN after the venture captalst ext by ntroducng D_VCEXIT and the nteractons thereof wth FOWN and FOWNSQ. As can be seen from the results reported n column 2 of Table 9, the rate of nsttutonal nvestor arrval s sgnfcantly lower for frms wth ntermedate founder ownershp as evdenced by negatve (postve) coeffcent estmate of FOWN_VC (FOWNSQ_VC). Our fndngs are consstent wth nsttutonal blockholders sceptcsm of frms wth ntermedate levels of founder ownershp and mply a lower degree of montorng, whch allows the founders of such frms a (largely) free and unmpeded control over the frm. We now turn to the examnaton of the number of shares offered as a percentage of the shares outstandng after the IPO (FLOAT). We hypothesze that entrenched founders wll be 21
protectve of ther ownershp postons and therefore wll try to lmt the extent of dluton resultng from the IPO. Snce the benefts from lower levels of dluton are of lesser mportance for founders wth low and hgh levels of ownershp, we expect to observe a convex relatonshp between founder ownershp and FLOAT. 21 We examne ths hypothess by estmatng the followng regresson equaton: FLOAT = α + α FOWN + α FOWNSQ + α FWEDGE + α LNBVA + α LFAGE 0 + α TECH + α RRANK + 6 1 7 2 n k =1 α YD k 3 k + u 4 5 (7) The results of the estmaton of Eq. (7) are reported n column 3 of Table 9. As expected, we fnd a convex relatonshp between founder ownershp and FLOAT. Ths mples that frms wth ntermedate founder ownershp do ndeed ssue a smaller fracton of shares at IPO and thus lmt the dluton faced by ther founders. 4.5.2. Are venture captalsts sellng overvalued equty? It can be argued that the results reported above represent a challenge to the market ratonalty. In partcular, f the decrease n frm value documented above s predctable, as we have shown t to be, t s surprsng that such a decrease s not already ncorporated n the stock prce before the venture captalst ext. In other words, f the market foresees the actons of founders after the VC ext, they would have ncorporated ths nformaton nto the stock prce. We argue that an answer to ths puzzle may be found n the presence of retal nvestors, ther overoptmsm, and short sale constrants. In partcular, as reported above, frms wth ntermedate levels of ownershp have the lowest nsttutonal nterest. Ths fndng has two mplcatons. Frst, t mples that shares of such frms are held prmarly by retal nvestors who are more lkely to be subject to fads and overoptmsm. Second, several recent papers (see, e.g., Nagel, 2005) fnd that 21 Dluton s of lesser mportance for founders wth hgh levels of ownershp because any reasonable level of dluton s unlkely to affect ther control over the frm. Dluton s also of lesser mportance for founders wth low levels of ownershp because ther ownershp s already too low to ensure control over the frm. 22
the man determnant of short sales constrants s the percentage of shares held by nsttutonal nvestors, who are the man lenders n short sale transactons. Thus, lower shareholdngs by nsttutonal nvestors, coupled wth the lowest float for frms wth ntermedate levels of ownershp, mply that such frms are lkely to have the most bndng short sale constrants and 22 23 thus the hghest degree of overvaluaton of n the mmedate post-ipo market. On a related note, changes n frm value descrbed above may also be seen as a challenge to the mportance of VC reputaton n the IPO market. In partcular, VCs are repeat players n the IPO market and therefore are lkely to suffer serous reputatonal consequences for sellng overvalued equty. Yet, decreases n the value of frms wth ntermedate levels of founder ownershp around venture captalst ext perssts. Are VCs ndeed sellng overprced equty? If venture captalsts are takng advantage of the overvaluaton of frms wth ntermedate levels of founder ownershp, they would be settng a hgher IPO offer prce for these frms (to reflect the potental overvaluaton n the post-ipo market). Ths, n turn, would mply no sgnfcant dfferences n IPO underprcng between frms wth ntermedate founder ownershp and the others. To examne ths conjecture, we estmate the followng regresson equaton: UNDERP = α + α FOWN + α FOWNSQ + α FWEDGE + α LNBVA + α LFAGE + α TECH + α RRANK + 6 0 1 7 2 n k=1 k α YD k 3 + u 4 5 (8) The results of the estmaton of Eq. (8) are reported n column 4 of Table 9. We fnd a concave and sgnfcant relatonshp between founder ownershp and IPO underprcng. Our fndngs ndcate that frms wth ntermedate levels of founder ownershp have the hghest underprcng 22 As argued by Duffe et al. (2002) and Ljungqvst et al. (2006), n the presence of short sale constrants, the prce of an asset can be bd above the most optmstc nvestors valuaton. 23 Venture captalst ext can be seen as an event that ncreases both the avalable float and the holdngs of nsttutonal nvestors and therefore lowers the short sale constrants. As such, t s possble that our results regardng changes n frm value are drven by the relaxaton of short sale constrants and not by changes n founder s ncentves. To address ths potental concern, we replcated our results usng float and nsttutonal shareholdngs as addtonal control varables. We fnd that our results (not reported, but avalable from the authors upon request) are qualtatvely unchanged n ths alternatve specfcaton. 23