Pyramd IPOs on the Chnese Growth Enterprse Market Martn Holmén and Peng Wang July, 2014 Abstract Ths paper nvestgates Intal Publc Offerngs (IPOs) of hgh-tech frms on the Chnese Growth Enterprse Market (GEM). Almost half of the GEM IPOs are set up n pyramd structures. The lkelhood of a pyramd structure ncreases wth the sze of the IPO frm and state control. Our results do not suggest that pyramds are set up to overcome fnancal constrants. However, we document that pyramd IPOs are dscounted before the IPO. The subscrpton prce to book rato s sgnfcantly lower for pyramd IPOs and ths translates nto hgher underprcng. We conclude that IPO nvestors demand a hgher rsk-premum when nvestng n pyramd IPOs. Key words: pyramd ownershp structure, Intal Publc Offerng, Growth Enterprse Market, Underprcng, Oversubscrpton JEL: G32 Martn Holmén (martn.holmen@cff.gu.se), correspondng author, s professor of fnance at Centre for Fnance and Department of Economcs, Unversty of Gothenburg, Box 640, 405 30 Gothenburg, Sweden and a vstng professor at the Department of Fnance and Statstcs of Hanken School of Economcs, Arkadagatan 22, 00101, Helsnk, Fnland. Peng Wang (peng.wang@hanken.f) s a Ph.D. canddate at the Department of Fnance and Statstcs of Hanken School of Economcs, Arkadagatan 22, 00101, Helsnk, Fnland. The authors wsh to thank an anonymous referee, the edtor Al Kutan, and conference partcpants at the Asan Fnance Assocaton annual meetng 2013 and workshop partcpants at Copenhagen Busness School and Hanken School of Economcs for ther helpful suggestons. Fnancal support by VINNOVA (grant 2010-02449 Holmen) and Wallenberg Center for Fnancal Research s gratefully acknowledged. 1
Ths paper explores the causes and consequences of the decson to create a pyramd structure and separate cash flow rghts from votng rghts at Intal Publc Offerngs (IPOs). We document that pyramd frms are dscounted already before the IPO,.e. before majorty shareholders could have exproprated any mnorty shareholders. Thus, the dscount on pyramd frms documented by e.g. Claessens et al. (2002) appears to be drven, at least partly, by an ex ante rsk-premum assocated wth nvestng n pyramds and not only by majorty shareholders actually expropratng mnorty shareholders. Mechansms that separate votng rghts from cash flow rghts, e.g. pyramd structures, dual class shares and cross-shareholdngs, are wdespread around the world and ths s well documented (see e.g. La Porta et al., 1999). However, Adams and Ferrera (2008) pont out that the emprcal lterature on the determnants and the effects of the actual decson to separate cash flow rghts from votng rghts s surprsngly small. Almeda and Wolfenzon (2006) provde a ratonale for pyramdal ownershp n famly busness groups. In ther theoretcal model a pyramd allows a famly to use all retaned earnngs of a frm t controls to set up new frms. Furthermore, Masuls et al. (2011) argue that pyramds are used to allevate fnancng constrants and lower the cost of rasng external equty. The emprcal lterature on pyramd structures has, however, gnored the analyss of how pyramd structures ntally were set up and nstead focused on the performance and behavor of mature pyramd frms. 1 Furthermore, the IPO lterature has overlooked the effect of ndrect ownershp and pyramd structures on the IPO process. 2 By lookng at the actual set-up of pyramd frms,.e. the IPO, we shed lght on the type of frms that become part of a pyramd. The orgnal owners can choose between holdng the controllng stake drectly or through a pyramd structure. Based on Almeda and Wolfenzon's (2006) and Masuls et al. (2011) we derve hypotheses about captal constrants, frm sze, and rsk among pyramd IPO frms. We also derve hypotheses about the effect of 2
pyramd structures on the IPO process based on the assumpton that pyramd structures are assocated wth nvestor expectatons about tunnelng of corporate resources to the ultmate owner. If expected expropraton s dscounted by external nvestors, they wll only nvest n pyramd structures f they receve an addtonal dscount on pyramd IPOs. We use the Chnese Growth Enterprse Market (GEM) n Shenzhen as our laboratory. The GEM was launched n 2009 n order to facltate for young growng frms to go publc and rase external equty. Surprsngly, 46 percent of the hgh tech IPOs on the GEM durng 2010 were set up n pyramd structures. In Europe, the typcal pyramdal frm s a large, old frm n a mature ndustry (see e.g. Holmén and Högfeldt, 2009). Our sample conssts of 173 IPOs on the GEM durng 2009 and 2010. We only fnd lmted support for pyramd frms beng captal constraned. The result ndcatng that pyramd frms have hgher leverage s not robust n our sample. Furthermore, there are no ndcatons that pyramd frms have hgher captal expendture. However, n lne wth Almeda and Wolfenzon s (2006) argument, pyramdal IPOs are sgnfcantly larger than the stand-alone IPOs. We fnd no ndcatons that pyramdal IPOs are rsker than the stand alone IPOs. Second, we nvestgate how pyramdal structures affect the IPO process. Usng treatment effect models we document that the pre IPO (subscrpton) prce to book rato s sgnfcantly lower for pyramd IPOs. Furthermore, the underprcng,.e. the prce ncrease the frst day of tradng, s sgnfcantly hgher for pyramd IPOs. These results are consstent wth Almeda and Wolfenzon (2006) and Masuls et al. (2011). Mnorty shareholders demand compensaton for expected expropraton whch results n a larger dscount on pyramdal IPOs. Usng detaled nformaton about the IPO process we fnd evdence that the oversubscrpton rate among nsttutonal nvestors s hgher for the more underprced pyramd IPOs. The oversubscrpton rate among ndvdual nvestors s not sgnfcantly dfferent for 3
pyramdal IPOs. The results appear to be consstent wth Rock s (1986) model where nformed nvestors only subscrbe to underprced IPOs. Even f the better nformed nsttutonal nvestors subscrbe to all IPOs on GEM, ther demand s hgher for the more underprced IPOs. The rest of the paper s organzed as follows. The next secton outlnes our hypotheses. Secton three presents the data and the IPO process on GEM whle the emprcal results are reported n secton four. Secton fve puts our results n perspectve and concludes. Hypotheses We explore the causes and consequences of pyramd structures at the IPO n lght of Almeda and Wolfenzon (2006) and Masuls et al. (2011). Almeda and Wolfenzon (2006) argue that pyramd structures wll be more prevalent among larger frms. The larger the frm s, the more dffcult t s for a controllng shareholder wth lmted wealth to keep control of the frm wthout separatng ownershp and control. Furthermore, snce dual class shares are not allowed n Chna, a pyramd structure s the natural choce for a controllng shareholder who wants to keep control of a large frm. Hypothess 1: The pyramd IPO frms are larger than the stand alone IPO frms. Masuls et al. (2011) argue that pyramds are used to allevate fnancng constrants and lower the cost of rasng external equty. Compared to drect fundng, the pyramdal structure levers the nternal captal under control whch helps rase external fundng for projects wth large captal requrements and low ntal cash flows. Ths argument s the bass for our second hypothess. Hypothess 2: The pyramdal IPO frms are more fnancally constraned than the stand-alone IPO frms. Masuls et al. (2011) also argue that famles have ncentves to place rsky frms n a pyramd. By usng a pyramd structure the famles can leverage ther nvested captal and 4
control many frms but at the same tme reduce the rsk and ths argument s partcularly vald for rsky frms. Smlar arguments can be found n Almeda and Wolfenzon (2006) and Attg et al. (2004). Hypothess 3: The pyramdal IPO frms are more rsky than the stand-alone IPO frms. Almeda and Wolfenzon's (2006) pyramdng model combnes the dea that the levered control over the frm's cash flow creates a fnancal advantage wth the key assumpton that tunnelng of corporate resources s sgnfcant when protecton of mnorty shareholders s weak. 3 The controllng owner use the captal of an already exstng frm together wth new equty captal from external nvestors to fnance a new frm that they also control at the pyramd level below. If the new external nvestors ratonally antcpate the cost of expropraton they wll only co-nvest alongsde the controllng shareholders n pyramd structures f they receve an addtonal dscount when nvestng n pyramdal IPO frms. 4 These arguments are the bass for our fourth hypothess. Hypothess 4: The subscrpton prce to book rato s lower for pyramd frms compared to stand alone frms. If the subscrpton prce on pyramd IPOs s set low delberately n order to attract nvestors t wll lead to an expected larger ntal returns,.e. larger underprcng. Beatty and Rtter (1986) also argue that the expected ntal return s an ncreasng functon of the uncertanty nvestors feel about the market prce of the IPO frm after lstng. Thus, f a pyramd structure ncreases the uncertanty nvestors feel about the IPO frm, t wll translate nto larger underprcng. Hypothess 5: The underprcng of pyramd IPO frms s hgher than for standalone IPO frms. Data Sample and data sources 5
Our sample conssts of 173 frms that were lsted on the Shenzhen Growth Enterprse Market (GEM) n 2009 and 2010. These frms are typcally young and n hgh-tech sectors. No fnancal or utlty frms exst n the sample. Publcly traded frms n Chna have to report the detaled ownershp nformaton n ther annual reports. The nformaton ncludes a fgure of the ownershp structure of the frm, the name and status of the ultmate controllng owner as well as the other nne bggest shareholders for both tradable and non-tradable shares. Ths regulaton makes t possble to dentfy the ultmate owner and the pyramdal chan f t s applcable. Due to regulaton on GEM, the largest shareholders of the frms are not allowed to sell ther shares or buyback shares for at least three years after IPO. Thus, at the tme of the data-collecton for ths study, there were no changes n controllng shareholders ownershp n the sample. In addton to the manually collected ownershp structure, we collected the fnancal data from Thomson Reuter and IPO data from RESSET (www.resset.cn). The IPO process on Growth Enterprse Market The contract between the IPO frm and the underwrter on GEM s a standby commtment, wth whch the ssung frm and ts underwrter frst ssue a prelmnary prospectus and solct nterests from potental nvestors. 5 After approval from the Chna Securtes Regulatory Commsson (CSRC), the ssung frm and ts underwrter hold a prcng meetng where the offer prce and the number of shares to be sold are agreed upon, and fnally they reach the stage of the share ssuance. Unlke other markets, the share ssuance n GEM nvolves both on-lne ssuance and off-lne ssuance. The on-lne ssuance s manly for ndvdual nvestors. The procedure s, once an ndvdual nvestor apples for a certan amount of shares, he or she receves a lottery number. After the ssuance of on-lne shares, the frm announces the numbers that wn the lottery, and only nvestors wth those numbers can actually get a certan amount of shares. The number of shares s usually much smaller than the amount that 6
the nvestor appled for beforehand. And for those who do not wn the lottery, the money s to be returned. The off-lne ssuance s manly amed for nsttutons, and t bascally follows the same rule as the on-lne ssuance. Snce the number of nsttutons s much smaller than the number of ndvdual nvestors, usually all of the elgble nsttutons can get a certan amount of shares of the IPO. The actual number of shares that each nsttuton s able to acqure depends on the number the nsttutons that appled durng the off-lne ssuance. Under a standby commtment contract, the underwrter s responsble for buyng any remanng shares that are not subscrbed n the two- to four- week standby perod. In Chna, non-votng shares or other devces for dscrmnatng aganst dfferent shareholders are not permtted. B-shares or H-shares carry the same votng rghts as A- shares. 6 Frms n GEM are too small to ssue B-shares or H-shares. We also fnd that two ratos are fxed. The porton of a frm sold at the IPO s fxed at 25% and the tradable shares sold to ndvdual nvestors (.e., onlne ssuance) are fxed as four tmes that sold to nsttutons (.e., offlne ssuance). These fxed ratos reflect the strct regulaton features of GEM. Varables We categorze the IPOs on GEM durng 2009 and 2010 nto three groups wth respect to the characterstcs of the ultmate owners,.e. famly-owned, state-owned and wdely held. In our sample 156 frms are famly controlled, 7 11 frms are state controlled, and only 6 frms are wdely held. Snce the wdely held frms do not have a controllng shareholder they can by defnton not be set up n a pyramd structure and these 6 frms are not ncluded n our tests of the probablty of a pyramd structure. 8 Thus, our tests of the probablty of a pyramd structure use 167 frms. State owned frms only make up 7 percent of our sample. In comparson, Fan et al. (2005) fnd that n 2001, the government was the 7
ultmate owner of about 76% of Chnese publcly lsted frms. Ths dfference reflects that the Chnese government has encouraged more prvate corporatons to lst ther shares publcly durng the latest ten years. The man varable of nterest and dependent varables n the probt regressons s Pyramd. It s a dummy varable that takes the value of one f the frm s set up n a pyramdal structure and zero otherwse. 73 IPO frms (43.7%) are set up n a pyramd structure. In the treatment effect regressons we use four dfferent dependent varables. The frst s the (subscrpton) prce to book rato pre IPO. PBpreIPO s calculated as the offerng prce dvded by net asset value per share (NAP) before IPO. The second s the underprcng of the IPO. IPO underprcng rate s calculated as (P c -P o )/ P o where P c s frst day closng prce and P o s offerng prce. The thrd and fourth dependent varables are the onlne and offlne, respectvely, oversubscrpton rates. The oversubscrpton onlne (offlne) s defned as the number of shares ndvdual (nsttutonal) nvestors have appled for, dvded by the number of shares avalable for sale to ndvdual (nsttutonal) nvestors. Our control varables nclude: Sales denotes the total sales of the frm n 2010 n logarthmc form and s used as a proxy for frm sze. As an alternatve sze measure we use the number of employees n the frm (EMPL). Leverage s calculated as the frm s total debt dvded by total asset n the year of 2010. CAPEX stands for the captal expendtures related to property plant and equpment and s used to proxy for the frm s nvestment. Leverage and CAPEX are used as proxes form captal constrants. We use yearly stock market volatlty (Volatlty) as proxy for rsk. 9 It s calculated as the standard devaton of daly stock returns from February 2011 to September 2011 of each frm and multpled by the square root of 252 n order to have an estmate of yearly volatlty. As a proxy for the frm s growth potental we construct a Salesgrowth varable. It s estmated as the total sales n 2010 dvded by total 8
sales n 2009 mnus one. In order to control for the effect of state ownershp, we construct a dummy named State whch takes the value of one f the ultmate controller s the state, and zero otherwse. Smlar control varables are used n earler studes on pyramds, e.g. Masuls et al. s (2011) nternatonal study, Fan et al. s (2005) Chnese study and Attg et al. s (2004) Canadan study. Emprcal results Unvarate comparson of pyramd, non-pyramd and wdely-held frm characterstcs Table 1 reports the medan dfference tests for the frm characterstcs of pyramd frms, nonpyramd frms and wdely-held frms. The sgnfcance level of the dfference n medan s based on Wlcoxon ranksum test. 10 We frst compare the pyramd frms to the closely held non-pyramdal frms. The Wlcoxon ranksum tests suggest some sgnfcant dfferences. Pyramdal frms ) are larger (Sales and number of employees (EMPL)), ) nvest more (CAPEX), ) have hgher Leverage, and v) have hgher Salesgrowth than non-pyramdal frms. We fnd no sgnfcant dfference n terms of the other varables. Second we compare the wdely held frms to the closely held pyramd and non-pyramdal frms. Wdely-held frms have sgnfcantly hgher captal expendtures (CAPEX) than closely held frms. These are no sgnfcant dfferences between wdely-held frms and closely held frms n other frm characterstcs. [Insert Table 1 here] Causes of pyramds We frst estmate a probt regresson to analyze what factors that nfluence the ultmate owners' endogenous decson to set up the IPO frm n the pyramd structure or not. The dependent varable s the pyramd dummy, whch equals one f the IPO frm s part of a pyramd structure, and zero otherwse. 9
Our frst hypothess s that larger frms are more lkely to be set up n a pyramd structure. We use Sales or EMPL as proxes for frm sze. Our second hypothess s that pyramdal frms are captal constraned. We therefore nclude captal expendture over property plant and equpment (CAPEX) and the rato of total debt over total asset (Leverage) as ndependent varables n the regressons. Fnally, our thrd hypothess states that pyramdal frms are more rsky and we therefore nclude our proxy for frm rsk, stock market volatlty after lstng (Volatlty), as an ndependent varable. To control for growth opportuntes we nclude Salesgrowth. In Chna, stateowned companes (SOEs) stll domnate the economy, so we nclude a State dummy to capture the potental governmental effect. In some estmated models we also nclude 16 ndustry dummes. Thus, we estmate the followng probt regresson. Pyramd Sze CAPEX State 6 1 16 j 1 2 Industry j j Leverage 3 Volatlty 4 Salesgrowth 5 The results are reported n table 2. In lne wth our frst hypothess, Sales and EMPL are postve and sgnfcant n most of the estmated models ndcatng that larger sze ncreases the lkelhood that a frm wll be set up n the pyramdal structure. The result that pyramd frms are larger n sze s supported by the theoretcal arguments n Almeda and Wolfenzon (2006) and can also be found by Attg et al. (2004) and Masuls et al. (2011). We do not fnd any support for our second hypothess statng that pyramdal frms are captal constraned. CAPEX and Leverage are nsgnfcant n all estmated models. Thus, whle CAPEX has been found to sgnfcantly ncrease the lkelhood of pyramd structures on the Chnese man market (Fan et al., 2005), n Canada (Attg et al., 2004), and n multple countres (Masuls et al., 2011), t does not appear to be sgnfcant on GEM. One possble explanaton s that almost all frms on the GEM are rapdly growng hgh-tech frms. 10
They are all nvestng ntensvely and t mght be dffcult to document a dfference between pyramdal frms and stand-alone frms even f dfferences exst. Earler results concernng the relaton between leverage and the lkelhood of pyramd structures are mxed. Masuls et al. (2011) fnd that leverage s negatvely related to the lkelhood of pyramds. Attg et al. (2004), on the other hand, fnd no sgnfcant relaton between leverage and the lkelhood of pyramd structures. Smlarly, we fnd no support for our thrd hypothess that pyramdal frms are rsker. Earler results concernng the relaton between frm rsk and the lkelhood of pyramd structures are mxed. Masuls et al. (2011) fnd that rsk s postvely related to the pyramd layer. Attg et al. (2004), on the other hand, fnd no sgnfcant relatons between rsk and the lkelhood of pyramd structures. State s postvely sgnfcant and ths fndng s supported by the decentralzaton theory by Fan et al. (2005). They argue that the government cannot relnqush control by sellng off ts stake n the frm freely due to strct regulaton prohbtng the dluton of state ownershp. Therefore, a pyramd can serve as a devce for the government to decentralze control rghts to the frm managers. Salesgrowth s manly nsgnfcant n table 3. It s postve and sgnfcant at the 10 percent level n one of the estmated models. [Insert Table 2 here] Treatment effect models Endogenety concern s an mportant ssue n studes of corporate and nvestor behavor, respectvely, and ownershp structure (Demsetz and Lehn, 1985). In order to control for the fact that orgnal owners endogenous decde whether to set up the frm n a pyramd structure we estmate treatment effect models. Treatment effect models consder the effect of an endogenously chosen bnary treatment--the pyramd dummy--on another endogenous contnuous varable, condtonal on two sets of ndependent varables. Related papers, for 11
example, Vllalonga and Amt (2006) and Masuls et al. (2011) also apply treatment effect models. The treatment effect model conssts of two steps. The frst step s a probt model wth the bnary treatment varable (.e., the Pyramd dummy) as the dependent varable, and nstrument(s) as ndependent varable(s). The second step s a regresson of the contnuous dependent varable, y, on the Pyramd dummy controllng for Salesgrowth, frm rsk (Volatlty), State ownershp, frm sze (Sales), and ndustry effects. Thus, we estmate y Pyramd Salesgrowth Volatlty State Sales 1 2 3 4 5 16 j 1 Industry where y represents ) the pre IPO Prce to book rato (PBPreIPO), ) Underprcng, ) Oversubscrpton onlne, and Oversubscrpton offlne, respectvely. In the frst step we use EMPL as our nstrument varable snce ) Almeda and Wolfenzon s (2006) theoretcal model suggests that pyramdal frms are large, ) n our sample, pyramdal frms are ndeed larger n sze and have more employees than stand-alone frms, ) EMPL s orthogonal (unrelated) to all the dependent varables n the second step. 11 Our other sze measure, Sales turned out to be hghly correlated wth oversubscrpton rates. 12 In the frst step, the treatment Pyramd s modeled as the outcome of an unobserved latent varable, Pyramd *. j Pyramd 1, 0, f Pyramd * 0 otherwse All models are estmated wth Maxmum Lkelhood and reported wth heteroskedastcty corrected standard errors n parenthess (Whte, 1980). 13 Subscrpton prce to book rato before IPO We start by testng our fourth hypothess statng that the subscrpton prce to book rato s lower for pyramd frms. PBpreIPO s used as the dependent varable and t s calculated as the subscrpton (offerng) prce dvded by net asset value per share (NAP) before IPO. Thus, t s an ex ante measure whch should capture the underwrters' nformaton about the 12
nvestors' nterest n the IPO. A lower PBpreIPO, ceters parbus, s consstent wth nvestors usng a hgher dscount rate when valung the frm. Holmen and Högfeldt (2004) run smlar tests on Swedsh IPOs. The results are reported n table 3 model 1. In lne wth hypothess 4 we fnd that the coeffcent on the pyramd dummy s negatve and sgnfcant. The result suggests that underwrters antcpate that the nvestors wll dscount the prce of pyramdal frms due to concerns of expropraton or asset appropraton. Wang (2013) fnds that, other thngs beng equal, pyramdal structures on GEM have a detrmental effect on frm performance, whle Gao and Klng (2008) document that a frm belongng to a pyramdal busness group wll experence exacerbated asset appropraton on the Chnese man market. Among other varables, only Salesgrowth s sgnfcantly assocated wth a hgher subscrpton prce to book rato pre IPO. [Insert Table 3 here] IPO underprcng We next test hypothess 5 statng that underprcng of pyramd IPO frms s hgher than for stand-alone IPO frms. The motvaton for nvestors to take the potental addtonal rsk assocated wth co-nvestng along wth the controllng shareholder n a pyramd structured IPO mght be that they are compensated by larger returns durng the frst days of stock lstng,.e. hgher underprcng. The lower prce to book rato of pyramd IPOs suggests that ths mght be the case. The dependent varable s calculated as (P c -P o )/P o where P c s frst day closng prce and P o s offerng (subscrpton) prce. The results are reported n table 3 model 2. In lne wth hypothess 5, the coeffcent on the Pyramd dummy s ndeed postve and sgnfcant, suggestng that underprcng s more substantal for pyramd frms. We nterpret ths result from both the nvestors and the underwrters' ponts of vews. Frst, from the nvestors sde, ths result s 13
consstent wth Beatty and Rtter (1986) who argue that the expected ntal return s an ncreasng functon of the uncertanty nvestors feel about the market clearng prce of an IPO. Investors antcpate or seek more compensaton when uncertanty s larger. For example, the uncertanty to nvest n a relatvely more complcated frm structure (pyramd frms) s larger than to nvest n a relatvely smpler structured frm (stand-alone frm). Second, from the underwrters pont of vew, a pyramd structure should ncrease the lkelhood that nvestors mght not be nterested n the IPO and not subscrbe to shares. Thus, n order to reduce the rsk of not beng able to sell all shares, the underwrters ntentonally underprce pyramd IPOs more. In lne wth earler research (e.g. Habb and Ljunqvst, 2001) larger IPOs n terms of total sales are assocated wth sgnfcantly lower underprcng. And n lne wth e.g. Gompers et al. (1996), more rsky frms experence sgnfcantly hgher underprcng. Furthermore, the coeffcent over the State dummy s postve and sgnfcant. One nterpretaton of ths result s that state-owned frms tend to underprce more n order to beneft nvestors, consstent wth the helpng hand hypothess of central government as controllng shareholder. 14 Cheung et al. (2010) fnd that consstent wth the helpng hand hypothess, central government connectons beneft the mnorty shareholders of lsted frms. In contrast, consstent wth the grabbng hand hypothess, the mnorty shareholders n frms controlled by the local government are exproprated as they lose up to half of the value of related party transactons. Onlne/Offlne oversubscrpton rate We have shown above that pyramdal frms are dscounted already before the IPO. Underwrters of frms n pyramds appear to be aware of the potental rsk of not beng able to sell all the shares and therefore ntentonally underprce pyramd IPOs more. So do nvestors actually stay away from pyramd IPOs due to concerns of expropraton? Or do nvestors 14
show hgher nterest n pyramd IPOs n order to obtan hgher frst day ntal returns? We next examne f and how the pyramd structures affect the IPO subscrpton onlne and offlne. 15 We report the results n table 6 models 1 (onlne) and 2 (offlne). The coeffcent of the Pyramd dummy s nsgnfcant n model 1 but sgnfcantly postve n model 2. Onlne subscrpton s manly for ndvdual nvestors whle offlne subscrpton s manly for nsttutonal nvestors, and a hgher oversubscrpton rate mples a larger nterest among nvestors. The results suggest that ndvdual nvestors nterest n IPOs appear not be affected by pyramd structures. Insttutons, on the other hand, are keener on pyramd frms than standalone frms. Insttutonal nvestors larger demand for the more underprced pyramd IPOs appear to be n lne wth the general percepton that nsttutonal nvestors, compared to ndvdual nvestors, have more experence, expertse, and better nformaton (See e.g. Chakravarty, 2001; Sas and Starks, 1997; and Walther, 1997). On the one hand and n lne wth Rock s (1986) theoretcal model of underprcng of IPOs, nformed nvestors (nsttutons) has an opportunty to proft from superor nformaton by bddng for msprced securtes (pyramdal IPOs). They are compensated for ther costly nformaton processng about e.g. complcated pyramd structure, by larger frst day returns. On the other hand, ndvdual nvestors are usually regarded as less sophstcaton and they mght have lttle knowledge about the pyramd structure. For example, Feng and Seasholes (2003) examne a sample of 90,478 actvely nvestng ndvduals n Chna, and document that Chnese ndvdual nvestors are much younger and have less nvestng/tradng experences than typcal ndvdual nvestors n U.S. In addton, Volatlty s postve and sgnfcant suggestng that nvestors regard IPO nvestments as lotteres wth postvely skewed returns. The lkelhood that a partcular 15
IPO wll be the next Apple ncreases wth frm rsk and therefore rsk mght ncrease nvestor nterest. [Insert Table 4 here] Addtonal tests We have performed a battery of unreported robustness test. Frst, we have deleted the 11 state owned frms from all tests. Our man results reman unchanged. Second, we have ncluded CAPEX and Leverage n the second step of treatment effect models. It does not alter our man results. Thrd, we nclude Return on Assets (ROA) n the models but t does not change our man results. Furthermore, Masuls et al (2011) argue that pyramd structures mght be used to overcome fnancal constrants. In the reported tests we have used CAPEX and Leverage to proxy for fnancal constrants. Another measure of fnancal constrants s of course dvdend payments. Fnancal constraned frms should be less lkely to pay dvdends. However, we note that most frms n our sample are rapdly growng frms and most of them do not pay dvdends, or pay very small dvdends, rrespectve of beng wthn a pyramd structure or not. Almeda and Wolfenzon (2006) argue that pyramds are set n order to derve prvate benefts of control and exproprate mnorty shareholders. Proxes for the value of the prvate benefts of control are dffcult to construct usng publc nformaton snce they are by defnton prvate (Zngales, 1998). The most commonly used measures, the votng premum on hgh votng stock and the premum at block transfer transactons, are not possble to construct for our sample snce all frms are one share - one vote and no major ownershp changes have taken place n the sample yet. Due to data lmtaton, we cannot test the drect related party transacton that may beneft the ultmate controller n the pyramdal group ether (see, Hu et al. 2012). However, we try to construct a measure by collectng the salares pad 16
by the IPO frms to the ultmate owners. Payng herself a very large salary could be one prvate benefts of control. We do not fnd any sgnfcant dfference between pyramds frms and stand-alone IPOs n terms of the salares pad to the ultmate owner. The ultmate owner may of course place frends and relatves on well-pad executve postons n the (pyramdal) frms as well and ths would be another type of prvate benefts of control. However, as far as we understand, ths s very uncommon n Chna snce t would generate an outcry among nvestors. We have also consdered varous nteracton effects. For example, nspred by Khanna and Palepu (1997) and Attg et al. (2004), we have ncluded the nteracton effects of Pyramd and Salesgrowth, and Pyramd and Volatlty n OLS regressons. The nteracton terms are nsgnfcant and the other results are not changed n a sgnfcant way. Fnally, we have replaced the pyramd dummy wth the actual separaton of control rghts from cash flow rghts created by the pyramd structure. Ths varable s not sgnfcant n smlar tests as the ones reported above. Thus, t appears as f the pyramd structure has an ndcator effect at the IPO,.e. there s no lnear relaton between the actual degree of separaton of control rghts from cash flow rghts and prce to book, underprcng, and oversubscrpton. Concluson and Dscusson In ths paper we have nvestgated Intal Publc Offerngs (IPOs) of hgh-tech frms on the Chnese Growth Enterprse Market (GEM). As far as we know, ths s the frst paper that nvestgates the IPO process n combnaton wth pyramdal structures on the new GEM n Chna. We frst show that almost half of the hgh-tech IPOs on the GEM are set up n pyramd structures. Based on Almeda and Wolfenzon (2006) and Masuls et al (2011) we formulate fve hypotheses and we fnd support for three of them. The lkelhood of a pyramd structure 17
ncreases wth the sze of the IPO frm (Hypothess 1). The subscrpton prce to book rato s sgnfcantly lower for pyramd IPOs compared to stand-alone IPOs,.e. pyramd IPOs are dscounted before the IPO (Hypothess 4). Furthermore, the underprcng s hgher for pyramd IPOs (Hypothess 5). We conclude that nvestors demand a hgher rsk-premum when nvestng n pyramd IPOs and ths translates nto a hgher underprcng. Our results do not suggest pyramd IPOs are captal constraned (Hypothess 2) or rsker (Hypothess 3). We thnk our results contrbute to the exstng lterature along at least two dmensons. Frst, most pror studes use nformaton about mature pyramdal frms and document that at least under some condtons, pyramd structures are assocated wth lower frm value and worse frm performance (see e.g. Claessens et al., 2002, Lemmon and Lns, 2003 and Fan et al. 2005). These results are often nterpreted n terms of expropraton of mnorty shareholders by the controllng shareholder. Pan et al. (2008) document expropraton wthn Chnese pyramds on the man board. Our IPO results supplement these results and suggest that pyramd afflated frms are dscounted already at the IPO. Second, we thnk our approach and the use of detaled nformaton about the IPOs on GEM let us document a more complete pcture than s typcally done. The analyss of ) the prce to book rato at the subscrpton prce, ) the underprcng, and ) oversubscrpton ratos also provdes a consstent story. A hgher underprcng could be due to a large nterest n the IPO whch creates overvaluaton once the frm shares begn to trade. Or, a hgher underprcng could be due to underwrters settng a lower subscrpton prce n order to attract nvestors. Our results for pyramd IPOs on the GEM suggest that the second explanaton s more plausble. 18
Notes 1. For a survey of the lterature on pyramds, see Morck et al. (2005). 2. Some studes have analyzed IPOs by frms wth dual class shares. See e.g. Taylor and Whttred (1998), Amoako-Adu and Smth (2001), Smart and Zutter (2003), and Holmén and Högfeldt (2004). 3. Several recent papers fnd evdence of tunnelng n Chnese lsted frms from dfferent perspectves. See e.g. Gao and Klng (2008) Hu et al. (2012) and Cheung et al. (2009). Chna s also perceved as a country wth poor nvestor protecton, for example, accordng to a corporate governance ratng research conducted by Governance Metrcs Internatonal (GMI) n 2009, Chna only got 3.01 (where 10 s the hghest score) and belonged to the lowest ranked countres. Ireland ranked hghest (7.44) followed by the UK (7.36), Canada (7.35), Australa (7.32) and the U.S. (7.18). 4. Due to prvate benefts of control, the new frm has postve present value for the controllng owner even f the subscrpton prce s dscounted at the IPO. 5. All frms n our sample have a Chnese nvestment bank as lead underwrter. 6. A and B shares are the most commonly traded shares n Chna, and H shares (also known as Hang Seng Chna Enterprses Index) are manly traded n the Hong Kong Stock Exchange. A-shares are traded and denomnated by Chnese Yuan, but B-shares are traded and denomnated by US dollars n the Shangha Stock Exchange and Hong Kong dollars n the Shenzhen Stock Exchange. H-shares are denomnated and traded n Hong Kong dollars. 7. In lne wth Masuls et al. (2011), a frm where a famly (ths ncludes sblngs, relatves, etc) or an ndvdual s the largest ultmate shareholder s defned as famly controlled. 8. Snce we use treatment effect models, where the probablty of a pyramd structure s estmated n the frst step, to estmate the effect of pyramd structures, the wdely held frms drop out of these estmatons as well. 9. Ex-post IPO stock market volatlty has been used as a proxy for IPO rsk by e.g. Gompers (1996) and Zhu (2009). 10. T-tests on mean dfferences generate smlar results. 11. A vald nstrument n the treatment effect model should be sgnfcant related to the bnary varable (pyramd dummy) n the frst step, and be orthogonal (unrelated) to the dependent varables (PBpreIPO, Underprcng, Oversubscrpton rate) n the second step. 12. Hmmelberg et al. (1999), Vllalonga and Amt (2006) and Masuls et al. (2011) employ dosyncratc rsk as ther nstrument for ownershp structure, e.g. pyramd structures n the frst step. We tred the dosyncratc rsk from the sngle factor model and t turned out to be hghly correlated wth Underprcng. Furthermore, the other sgnfcant varable n the probt models reported n table 2, the State dummy, s also sgnfcantly related to Underprcng and therefore not a vald nstrument. 13. Estmatng the Treatment Effect models wth the two-step Heckman procedure generate smlar but somewhat statstcally weaker results. 14. Nne out of eleven state-owned enterprses n our sample are controlled by the central government. 15. Unlke the allocaton rule proposed by London Stock Exchange where dscrmnaton can only appear on the bass of the sze of the applcatons (Brennan and Franks, 1997), CSRC allows for no dscrmnaton aganst any nvestors (e.g., the sze of applcaton and dentty) n the allocaton process of the IPOs n Chnese frms. Frms and ther nvestment bankers should notarze ther lottery processes, so the allocaton s on a far bass, and ths allevates our concern about any potental selecton bas of the successful subscrbers. 19
References Adams, B. R., and Ferrera, D. 2008. "One-share, One-vote: the Emprcal Evdence." Revew of Fnance 12, no. 1: 51-91. Almeda, H. and Wolfenzon, D. 2006. "A theory of pyramdal ownershp and famly busness groups." Journal of Fnance 61, no. 6: 2637-2680. Amoako-Adu, B. and Smth, B. 2001. "Dual class frms: Captalzaton, ownershp structure and recaptalzaton back nto sngle class." Journal of Bankng and Fnance 25, no.6: 1083-1111. Attg, N., Fscher, K. and Gadhoum, Y. 2004. "On the determnants of pyramdal ownershp: evdence on dluton of mnorty nterests." Workng paper. http://papers.ssrn.com/sol3/papers.cfm?abstract_d=434201 Beatty, R. and Rtter, J. 1986. "Investment bankng, reputaton, and the underprcng of ntal publc offerngs." Journal of Fnancal Economcs 15, no. 1-2: 213-232. Brennan, M.J. and Franks, J. 1997. "Underprcng, ownershp and control n ntal publc offerngs of equty securtes n the UK." Journal of Fnancal Economcs 45, no. 3: 391-413. Chakravarty, S. 2001. "Stealth-tradng: Whch traders trades move stock prces?" Journal of Fnancal Economcs 61, no.1: 289-307. Cheung, Y.L. Rau, P. R. and Stourats, A. 2010. "Helpng hand or grabbng hand? Central vs Local government shareholders n Chnese lsted frms." Revew of Fnance 14, no.4: 669-694. Claessens, S., Djankov, S., Fan, J.P.H. and Larry, L.H.P. 2002. "Dsentanglng the ncentve effect from the entrenchment effect of large shareholdngs." Journal of Fnance 57, no.6: 2741-2771. Claessens, S., Djankov, S., Fan, J.P.H. and Larry, L.H.P. 2002. "Expropraton of mnorty shareholders n East Asa." Journal of Fnancal Economcs 58, no.7: 81-112. Demsetz, H. and Lehn, K., 1985. "The structure of corporate ownershp: causes and consequences." Journal of Poltcal Economy 93, no.6: 1155-1177. Fan, J.P.H., Wong, T.J. and Zhang, T.Y. 2005. "The emergence of corporate pyramds n Chna." CEI workng paper seres, No. 2005 16. Feng, L and Seasholes, M.S. 2003. "A profle of ndvdual nvestors n an emergng stock market." Workng paper. Gao, L. and Klng, G. 2008. "Corporate governance and tunnelng: Emprcal evdence from Chna." Pacfc-Basn Fnance Journal 16, no.5: 591-605. Gompers, P. 1996. "Grandstandng n the Venture Captal Industry." Journal of Fnancal Economcs 42, no. 1: 133-156. Habb, M. and Ljungqvst A. 2001. "Underprcng and Entrepreneural Wealth Losses n IPOs: Theory and Evdence." Revew of Fnancal Studes 14, no.2: 433-458. Hmmelberg, C. P., Hubbard, G.R. and Pala, D. 1999. "Understandng the determnants of manageral ownershp and the lnk between ownershp and performance." Journal of Fnancal Economcs 53, no.3: 353-384. Holmén, M. and Högfeldt, P. 2004. "A law and fnance analyss of ntal publc offerngs." Journal of Fnancal Intermedaton 13, no.3: 324-358. Holmén, M. and Högfeldt, P. 2009. "Pyramdal dscounts: tunnelng or overnvestment?" Internatonal Revew of Fnance 9, no.1-2: 133-175. Hu, S. H., L, G. Xu, Y. H. And Fan, X. A. 2012. Effects of nternal governance factors on cross-border-related party transactons of Chnese companes. Emergng Markets Fnance and Trade 48, no. supplement 1: 58-73. 20
Khanna, T. and Palepu, K. 1997. "Why focused strateges may be wrong for emergng markets." Harvard Busness Revew 75, no. 4: 41-51. La Porta, R., Lopez-De-Slanes, F. and Shlefer, A. 1999. "Corporate ownershp around the world." Journal of Fnance 54, no. 2: 471-517. Loughran, T., Rtter, J. R. and Rydqvst, K. 1994. "Intal publc offerngs: Internatonal nsghts." Pacfc-Basn Fnance Journal 2, no. 2-3: 165-199. Lemmon, M.L. and Lns, K.V. 2003. "Ownershp structure, corporate governance, and frm value: Evdence from the East Asan fnancal crss." Journal of Fnance 58, no.4: 1445-1468. Masuls, R., Pham, P., and Zen, J. 2011. "Famly Busness Groups around the World: Fnancng Advantages, Control Motvatons, and Organzatonal Choces." Revew of Fnancal Studes 24, no. 11: 3556-3600. Morck, R. Wolfenzon, D. and Yeung, B. 2005. "Corporate governance, economc entrenchment, and growth." Journal of Economc Lterature 43, no. 3: 655-720. Pan, H. B., Xa, X. P. and Yu, M. G. 2008. "Expropraton: Evdence from rght ssues n Chna." Emergng Markets Fnance and Trade 44, no. 1: 5-20. Rock, K. 1986. "Why new ssues are underprced." Journal of Fnancal Economcs 15, no. 1-2: 187-212. Sas, R.W. and Starks, L.T. 1997. "Return autocorrelaton and nsttutonal nvestors." Journal of Fnancal Economcs 46, no. 1: 103-131. Smart, S. and Zutter, C. 2003. "Control as a motvaton for underprcng: A comparson of dual and sngle-class IPOs." Journal of Fnancal Economcs 69, no. 1: 85-110. Taylor, S. and Whttred, G. 1998. "Securty desgn and the allocaton of votng rghts: Evdence from the Australan IPO market." Journal of Corporate Fnance 4, no. 2: 107-131. Vllalonga, B. and Amt, R. 2006. "How do famly ownershp, control and management affect frm value?" Journal of Fnancal Economcs 80, no. 2: 385-417. Walther, B.R. 1997. "Investor sophstcaton and market earnngs expectatons." Journal of Accountng Research 35, no. 2: 157-179. Wang, P. 2013. Ownershp structure of the Chnese Growth Enterprse Market. Workng paper http://papers.ssrn.com/sol3/papers.cfm?abstract_d=2223989 Whte, H. 1980. A heteroskedastcty-consstent covarance matrx estmator and a drect test for heteroskedastcty. Econometrca 48, no. 4: 817-838. Zhu, Y. 2009. The relaton between IPO underprcng and ltgaton rsk revsted: Changes between 1990 and 2002. Fnancal Management 38, no. 2: 323-355. Zngales, L. 1998. Why t s worth beng n control, n G. Bckerstaffe, ed.: The Complete Fnance Companon, FT Ptman Publshng, London. 21
Table 1 Dfferences between pyramd frms, non-pyramd frms and wdely-held frms, respectvely In ths table we report dfference n medan test between pyramd frms, non-pyramd frms and wdely held frms. The sample conssts of 173 frms that lsted on the growth enterprse market (GEM) n Chna n 2010. Accountng data and stock market data are collected from Thomson One and IPO data are from RESSET (www.resset.cn), ndustry classfcaton s from Chna Securtes Regulatory Commsson (CSRC) Industry Classfcaton, and ownershp nformaton s manually collected by the authors. PBpreIPO refers to (subscrpton) prce to book rato pre IPO, and s calculated as the offerng prce dvded by net asset value per share (NAP) before IPO. IPO underprcng s equal to (P c -P o )/P o, where P c s frst day closng prce and P o s offerng prce. Oversubscrpton onlne (offlne) s equal to the number of shares ndvdual (nsttutonal) nvestors have appled for, dvded by the number of shares avalable for sale to ndvdual (nsttutonal) nvestors. Sales denotes the total sales n Chnese Yuan of the frm n 2010 n logarthmc form. EMPL stands for employee number, whch s the logarthmc of total employee number of the frm. Volatlty s equal to the yearly stock market volatlty and t s estmated as the standard devaton of daly stock returns from February 2011 to September 2011 of each frm multpled by the square root of 252. Salesgrowth s the growth of total sales n 2010. Leverage s equal to the frm s total debt dvded by total asset n the year of 2010, and t s n a percentage scale. CAPEX s equal to the captal expendture over property plant and equpment. Medan dfferences tested by means of Wlcoxon ranksum test. ***denotes sgnfcant at 1% level. ** denotes sgnfcant at 5% level. *denotes sgnfcant at 10% level. (1)Wth Pyramd (2)No Pyramd (3)Wdely Held Wlcoxon ranksum test varable n=73 n=94 n=6 (1) and (2) (1) and (3) (2) and (3) medan medan medan PBpreIPO 11.390 11.500 13.946-0.110-2.556-2.446 IPO underprcng 0.346 0.397 0.261-0.051 0.085 0.136 Oversubscrpton 136 153 136-17 0 17 onlne Oversubscrpton 50.3 58.9 54.9-8.7-4.6 4.0 offlne Sales (n log) 4.065 3.566 3.970 0.499*** 0.095-0.404 EMPL (n log) 6.522 6.226 6.359 0.296*** 0.163-0.133 Volatlty (%) 48.8 49.8 49.6-1.0-0.8 0.2 Salesgrowth 0.336 0.278 0.270 0.058* 0.066 0.008 Leverage 14.537 10.312 11.760 4.225** 2.777-1.448 CAPEX 8.070 4.800 11.050 3.270*** -2.980* -6.250*** 22
Table 2 Probt regresson, the dependent varable s the pyramd ndcator varable In ths table we use 167 closely held frms that lsted on the growth enterprse market (GEM) n Chna n2010. Accountng data and stock market data are collected from Thomson One, and ownershp nformaton s manually collected by the authors. The dependent varable, Pyramd, s a bnary varable that takes the value of one f the frm s n a pyramd structure and zero otherwse. EMPL stands for employee number, whch s the logarthmc of total employee number of the frm. Salesgrowth s the growth of total sales n 2010. Volatlty s equal to the yearly stock market volatlty and t s estmated as the standard devaton of daly stock returns from February 2011 to September 2011 of each frm multpled by the square root of 252. State s a dummy varable equals to one f the ultmate owner s the state, and zero otherwse. Leverage s equal to the frm s total debt dvded by total asset n the year of 2010, and t s n a percentage scale. CAPEX s equal to the captal expendture over property plant and equpment. Robust standard errors are reported n the parentheses. ***denotes sgnfcant at 1% level. ** denotes sgnfcant at 5% level. *denotes sgnfcant at 10% level. M1 M2 M3 M4 M5 M6 Sales 0.603*** 0.552*** - - 0.756*** - (0.216) (0.194) (0.166) EMPL - - 0.151 (0.152) 0.268* (0.145) - 0.377*** (0.138) Salesgrowth 0.356 0.348 0.588 0.605* - - (0.353) (0.322) (0.392) (0.354) Volatlty -1.969-1.331-2.230* -1.404 - - (1.244) (1.08) (1.229) (1.160) State 0.910* 1.087** 1.190*** 1.307** - - (0.472) (0.466) (0.454) (0.468) Leverage 0.006 0.006 0.010 0.010 - - (0.009) (0.008) (0.011) (0.0078) CAPEX 0.007 0.009 0.017 0.014 - - (0.012) (0.012) (0.011) (0.001) Constant -1.407 (1.1) -2.030** (0.885) -0.164 (1.209) -1.779 (1.094) -3.100*** (0.651) -2.559*** (0.889) Industry Yes No Yes No No No Dummes Pseudo-R 2 0.2348 0.1393 0.2057 0.1204 0.1009 0.0383 N 167 167 167 167 167 167 23
Table 3 Treatment effect models wth the (subscrpton) prce to book rato pre IPO (PBpreIPO) and IPO underprcng, respectvely, as dependent varables. In ths table we use 167 closely held frms that lsted on the growth enterprse market (GEM) n Chna n 2010. Accountng data and stock market data are collected from Thomson One and IPO data are from RESSET (www.resset.cn), ndustry classfcaton s from Chna Securtes Regulatory Commsson (CSRC) Industry Classfcaton, and ownershp nformaton s manually collected by the authors. In M1, the dependent varable, PBpreIPO refers to (subscrpton) prce to book rato pre IPO, and s calculated as the offerng prce dvded by net asset value per share (NAP) before IPO. In M2, the dependent varable, IPO underprcng, s equal to (P c -P o )/P o, where P c s frst day closng prce and P o s offerng prce. Sales denotes the total sales n Chnese Yuan of the frm n 2010 n logarthmc form. Salesgrowth s the growth of total sales n 2010. Volatlty s equal to the yearly stock market volatlty and t s estmated as the standard devaton of daly stock returns from February 2011 to September 2011 of each frm multpled by the square root of 252. State s a dummy varable equals to one f the ultmate owner s the state, and zero otherwse. The endogenous bnary varable, Pyramd, takes the value of one f the frm s n a pyramd structure and zero otherwse. The models are estmated by Maxmum Lkelhood, and robust standard errors are reported n parentheses. ***denotes sgnfcant at 1% level. ** denotes sgnfcant at 5% level. *denotes sgnfcant at 10% level. We use EMPL (employee number) as the nstrument n the frst step. In order to save space, only the results from the second step of treatment effect model are reported. Model 1 Second step PBpreIPO Pyramd -8.265*** (2.263) Salesgrowth 2.986** (1.262) Volatlty -0.384 (3.215) State 1.814 (1.677) Sales 0.419 (0.663) Constant 15.872*** (3.371) Industry dummes Yes Yes N 167 167 Model 2 Second step IPO underprcng 0.446** (0.178) -0.054 (0.291) 1.504*** (0.291) 0.371*** (0.141) -0.081* (0.046) -0.093 (0.195) 24
Table 4 Treatment effect models wth oversubscrpton rate onlne/offlne as dependent varables In ths table we use 167 closely held frms that lsted on the growth enterprse market (GEM) n Chna n 2010. Accountng data and stock market data are collected from Thomson One and IPO data are from RESSET (www.resset.cn), ndustry classfcaton s from Chna Securtes Regulatory Commsson (CSRC) Industry Classfcaton, and ownershp nformaton s manually collected by the authors. In M1 (M2), the dependent varable Oversubscrpton onlne (Oversubscrpton offlne) s equal to the number of shares ndvdual (nsttutonal) nvestors have appled for, dvded by the number of shares avalable for sale to ndvdual (nsttutonal) nvestors, note that both measures are n logarthmc form. Salesgrowth s the growth of total sales n 2010. Volatlty s equal to the yearly stock market volatlty and t s estmated as the standard devaton of daly stock returns from February 2011 to September 2011 of each frm multpled by the square root of 252. State s a dummy varable equals to one f the ultmate owner s the state, and zero otherwse. The endogenous bnary varable, Pyramd, takes the value of one f the frm s n a pyramd structure and zero otherwse. The models are estmated by Maxmum Lkelhood, and robust-standard errors are reported n parentheses. ***denotes sgnfcant at 1% level. ** denotes sgnfcant at 5% level. *denotes sgnfcant at 10% level. We use EMPL (employee number) as the nstrument n the frst step. In order to save space, only the results from the second step of treatment effect model are reported. Pyramd 0.085 (0.259) Sales growth -0.196 (0.148) Volatlty 1.722*** (0.439) State 0.116 (0.147) Sales -0.103 (0.072) Constant 4.192*** (0.375) Industry Yes Dummes N 167 Model 1 Second step Oversubscrpton onlne Model 2 Second step Oversubscrpton offlne 1.033*** (0.161) -0.038 (0.095) 3.058*** (0.360) 0.595 (0.183) -0.171* (0.095) 2.588*** (0.503) Yes 167 25