Lmts of arbtrage and corporate fnancal polcy Massmo Massa INSEAD * Urs Peyer INSEAD * Zhenxu Tong INSEAD * Frst draft: March 2004 Ths draft: September 2004 Abstract We focus on an exogenous event that changes the cost of captal of a company: the addton of ts stock to the S&P500 ndex and nvestgate how the companes react to t by modfyng ther corporate fnancal, and nvestment polces. Ths allows us to test captal structure theores n an deal controlled experment, where the effect of the ndex addton on the stock prce s exogenous from a manager s pont of vew. Consstent wth the trade-off and market tmng theores, but nconsstent wth the peckng order theory, we fnd more equty ssues, ncreases n nvestment, but reductons n debt n response to a hgher abnormal announcement return. Moreover, n the 24 months after the ndex addton, frms that ssue equty and ncrease nvestment dsplay negatve abnormal returns and they perform worse than frms that ssue but do not ncrease nvestment. Ths fndng s consstent only wth the market tmng theory of Sten (1996) and supports a lmts of arbtrage story n whch the stocks dsplay a downward slopng demand curve and companes themselves act as arbtrageurs takng advantage of the wndow of opportunty. We would lke to thank Jeremy Sten and semnar partcpants at INSEAD for ther comments. * Department of Fnance, INSEAD, Boulevard de Constance, 77305 Fontanebleau, France. Emal for Massmo Massa: massmo.massa@nsead.edu. Emal for Urs Peyer: urs.peyer@nsead.edu. Emal for Zhenxu Tong: zhenxu.tong@nsead.edu.
1 Introducton The lterature on lmts of arbtrage argues that, f nsttutonal nvestors are constraned by the lack of long-term resources, they cannot arbtrage away dfferences n prce between otherwse dentcal - n terms of cash flow and rsk profle - stocks. A classc example of ths phenomenon s the abnormal announcement return observed when a stock s added to the S&P500 ndex (e.g., Shlefer, 1986). Gven that some nvestors are nsttutonally requred to nvest n stocks of S&P500 frms (e.g., ndex funds), the absence of suffcent arbtrage captal mples that the addton to the ndex ncreases prces n the announcement perod due to a demand effect. Ths prce dscrepancy s expected to persst n the absence of suffcent arbtrage captal. 1 The lterature has always looked for arbtrageurs focusng on nvestors. However, there exsts one player who could arbtrage away the dfference that has been largely gnored: the company tself. That s, f a msprcng exsts, then we would expect companes to take advantage of ths stuaton. For example, f the ncrease n equty value due to the announcement return of the S&P500 ndex addton leads to overvaluaton n the eyes of the managers, these should ssue more equty. More generally, we expect that the ncrease n equty value observed around the addton to the S&P500 ndex wll lead managers to adjust the frm's captal structure and potentally also the nvestment and cash holdng polces. Snce the ndex addton s exogenous from the pont of vew of the managers, the event provdes an deal settng to test captal structure theores where the frm s cost of equty changes exogenously. Usng a sample of 222 companes added to the S&P500 ndex n the years 1981-1997 we fnd an average abnormal announcement return of 3.2%, that translates n a reducton n the cost of equty on average, by about 1.2 percentage ponts. The fact that the average frm reacts to the drop n the cost of equty by also ncreasng nvestment s consstent wth both ratonal theores (trade-off theory and peckng order theory) and Sten s (1996) market tmng theory. Indeed, all theores predct that managers take advantage of the lower cost of captal by ssung equty. However, the fact that the average frm that ssues equty also reduces debt s consstent wth the trade-off and market tmng theory but nconsstent wth the peckng order theory. 1 We use the term lmts of arbtrage nterchangeably wth lmted arbtrage to nclude the case where all market partcpants are ratonal but arbtrageurs resources are lmted (Shlefer and Vshny, 1997). 1
It s the analyss of long-run stock returns that allows us to dfferentate between ratonal and behavoral theores. Ratonal models are based on the assumpton that the drop n the cost of captal s not a temporary msprcng but a permanent change. Therefore, ssung equty and ncreasng nvestment wll lead to a hgher frm value n the long run relatve to ssung but not nvestng. The reason s that a lower cost of captal decreases the hurdle rate for new projects and more nvestment should be realzed on average. Sten s (1996) theory, on the other hand, posts that managers take advantage of temporary msprcng to ssue equty. The way they use the proceeds depends on the horzon of the managers. Short-horzon managers wll use the temporarly lower cost of equty as the hurdle rate and therefore ncrease nvestment, whle long-horzon managers wll use the proceeds of the equty ssue to reduce debt or ncrease cash, but not to take projects that are value destroyng gven the true cost of equty. Ths mples that ssung and nvestng, relatve to ssung and not nvestng, wll result n a lower frm value n the long run f the change n the cost of equty s due to msprcng as the market tmng theory posts. Thus, the study of the long-run effects of equty ssuance and nvestment ensung the ndex addton allows us to test the two alternatve hypotheses. Ratonal theores are supported f frms that ssue equty and ncrease nvestment outperform those that ssue but do not nvest. Behavoral theores, nstead, requre that frms that ssue equty and nvest underperform those that ssue but do not nvest. The latter ndcates that short-horzon managers have correctly exploted a temporary decrease n the cost of captal by ssung equty and nvestng. Consstent wth the market tmng theory and n contrast to the predcton of the tradeoff theory, we do fnd that the subsample of frms that ssue equty and ncrease nvestment dsplay sgnfcantly negatve long-run abnormal returns. In lne wth the predctons of the market tmng theory, we also fnd that long-horzon managers who ssue equty and ncrease cash perform better, whle frms that ssue equty and reduce debt do even better. One potental ssue wth ths nterpretaton, s that we may just be capturng the effects of fnancal constrants. To do address ths ssue we follow Baker, Sten and Wurgler, (2003), who show that managers of fnancally constraned frms act as f they had a short horzon. Thus, we expect the long-run returns of fnancally constraned frms to be negatve (Lamont, Polk and Saa-Requejo, 2001). Issung and nvestng should not affect ths negatve 2
predcton. However, unconstraned frms presumably the ones wth long-horzon managers should explot the wndow of opportunty by ssung equty and not usng the proceeds to ncrease nvestment. Thus, unconstraned frms that ssue and nvest reveal themselves to be managed by short-horzon managers and should dsplay negatve long-run returns. We fnd that the long-run abnormal returns for fnancally constraned frms proxed by the modfed KZ-ndex (Baker, Sten and Wurgler, 2003) are sgnfcantly negatve regardless of whether they ncrease nvestment. On the other hand, the long-run abnormal returns of fnancally unconstraned frms, whle n general postve, turn negatve f frms ssue equty and ncrease nvestment. Thus, frms revealng themselves to be operated by short-horzon managers suffer sgnfcantly n the long run, consstent wth Sten s (1996) market tmng theory. Our fndngs confrm all the predctons of the market tmng theory, but are nconsstent wth the ratonal theores. Our paper makes two mportant contrbutons. Frst, we are able to test predctons of captal structure theores usng the S&P500 ndex addton as an event an event that s exogenous to the frm s decsons. Whle there are many studes that analyze corporate fnancal polcy ssues, few can escape the crtcsm of endogenety. Investment decsons are affected and affect fundng decsons. Payout polcy s nterlnked wth the fundng and nvestment decson. The corporate fnancal structure (leverage) s often the result of a seres of separate decsons taken over a long perod of tme (Baker and Wurgler, 2002) and not necessarly based on an optmzng decson (Welch, 2004). Ths mples that there are very few stuatons where t s possble to fnd a proper dentfyng restrcton that allows the researcher to break the smultanety nherent n any corporate fnancal decson (Lamont, 1997). We beleve the addton of a company s stock to the S&P 500 ndex, whch s an event that s largely outsde management s control and thus exogenously changes the cost of equty, s one such stuaton. Second, our results suggest that equty ssues offset the ntal reason for the ncrease n stock prces by ncreasng the supply of shares. Thus, companes de facto act as arbtrageurs. Ths offers a potental reconclaton between the fndngs of Shlefer (1986) and Garry and Goetzmann (1986) who show that the addton to the ndex has a permanent effect on the share prce, and Harrs and Gurel (1986), Benesh and Whaley (1996) and Lynch and Mendenhall (1997), who fnd a reversal. Companes reactons to the event tself 3
seem to affect the long-run performance, thus leavng room for dfferent fndngs at dfferent tme horzons. It s also nterestng to note that these results queston the degree of market effcency. Indeed, f companes are able to take advantage of stock market reactons, ether the market does not ratonally foresee companes reactons, or s prevented by some constrant to optmally adjust to the long run value. Ths provdes further ndrect evdence on the lmts of arbtrage and suggests the exstence of constrants analogous to the ones found n the case of lack of arbtrage captal n mergers and acqustons (Baker and Savasoglu, 2002). The paper s structured as follows. In the next secton, we provde a lterature revew and compare state our contrbuton. In Secton 3, we hghlght the predctons of the tradeoff, peckng order and market tmng theores as they pertan to the change n the cost of equty. In secton 4, we ntroduce the sample and analyze the change n the cost of equty around the ndex addton. Secton 5 reports the results of studyng the changes n the corporate fnancal polces around the ndex addton usng a system of equatons framework. Secton 6, analyzes the long-run stock returns of the event frms. In secton 7, we provde a new test of the market tmng theory. We dscuss the mplcatons of our fndngs for market effcency n secton 8. A bref concluson follows. 2 Lterature revew The emprcal lterature has found compellng evdence that lmts of arbtrage affect ndvdual stock prces through downward slopng demand curves. Indrect evdence s provded by the studes on the effects of compostonal changes n broad ndexes, block tradng and nternatonal flows of funds. For example, studes analyzng addtons and deletons to the S&P500 ndex fnd that addtons to the ndex ncrease share prces, whle delstngs decrease share prces (e.g., Garry and Goetzmann, 1986, Harrs and Gurel, 1986, Shlefer, 1986, Dhllon and Johnson, 1991, Benesh and Whaley, 1996, and Lynch and Mendenhall, 1997). Smlar effects have been documented n studes of ndex addtons n other countres (e.g., Masse et al., 2000). More recently, Wurgler and Zhuraskaya (2000) assess the effect of lmts of arbtrage n terms of cross-stock substtutablty, whle Barbers, Shlefer and Wurgler (2003) show that the addton to the ndex causes the stock to co-move more wth other S&P500 frms and ncreases the beta wth respect to the S&P 500 ndex. 4
Besdes the growng lterature focusng on ndex addtons, the lterature on block tradng fnds temporary prce pressure on ndvdual securtes condtonal upon unusual demand or supply (Lakonshok, Shlefer and Vshny, 1991 and 1992; Chan and Lakonshok 1993, 1995). Asquth and Mullns (1986) fnd a sgnfcantly negatve announcement effect for prmary stock offerngs - the larger the ssue, the more negatve the effect. Smlar results are reported by Loderer, Cooney, and Van Drunen (1991) for prmary offerngs by regulated frms. Unless the negatve announcement returns can be explaned by changes n the stock s lqudty or new nformaton, those fndngs are consstent wth a downward slopng demand curve for stocks. In an nternatonal context, Froot, O'Connell and Seasholes (2001) show that local stock prces are senstve to flows of funds of nternatonal nvestors, and that transtory nflows have a postve mpact on future returns. More drect evdence on the mpact of flows on stocks s provded by Warther (1995), Zheng (1999), Goetzmann and Massa (2002) and Teo and Woo (2003) who show that demand pressure from mutual funds drectly mpacts stock prces and ncreases stocks cross-correlatons. However, to our knowledge, there are no studes that focus on the mpact of lmts of arbtrage on corporate fnancal polcy. In the present paper, we brdge ths gap. Our goal s to study how the managers choce of the captal structure, nvestment and cash holdng polces s affected by the stock s abnormal return caused by the ndex addton. In other words, f lmts to arbtrage prevent nvestors to arbtrage away dfferences n prces, how are managers reactng to ths stuaton? We beleve that the addton to the S&P500 ndex s an deal event to test predctons of the market tmng theory snce t s exogenous from the perspectve of management. Moreover, t s drven by a change n the cost of equty not by changes n expected future cash flows. 3 Hypothess development In the followng secton we revew the predctons of the captal structure theores as they pertan to managers reacton to the short-term abnormal announcement return. We start wth the trade-off theory n whch the assumpton s that the change n the cost of captal s a permanent change drven by ratonal sources. Second, we dscuss the peckng order theory that makes the assumpton that prces can devate from the true value known by the nsders but wll revert to t f managers act on the prce dscrepancy. Thrd, we 5
descrbe the predctons of the market tmng theory, whch s based on the assumpton that the market prce can be wrong and companes can take advantage of the msprcng. 3.1 The trade-off theory The trade-off theory predcts that leverage s set to a level at whch the costs and benefts of debt are balanced. The benefts of debt are ts tax sheld (Modglan and Mller, 1963) and a potental reducton n the free cash flow problem (Jensen, 1986). These benefts are traded off aganst the cost of fnancal dstress and the potental agency cost of debt (Jensen and Mecklng, 1976; Myers, 1977; Morellec, 2004). The addton to the S&P500 ndex affects at least three of the above costs and benefts, thus shftng the optmal captal structure to nclude less debt. 2 Gven that n our sample a hgher announcement return mples a bgger reducton n the cost of equty, the trade-off theory predcts a reducton n the optmal leverage. Ths clear mplcaton s unque to our experment because the change n the equty value s due to a change n the dscount rate, not a change n the expected cash flows. If the change n equty value were prmarly due to changes n expected cash flows, then the optmal leverage could ncrease because tax shelds become more valuable. Ths sets ths experment apart from pror results (Baker and Wurgler, 2002, and Welch, 2004). The trade-off theory provdes two other mplcatons. Frst, a frm wth a hgher announcement return should ncrease ts cash-leverage more. The reason s the followng. Gven that the relevant level of debt n the trade-off theory s gross debt mnus cash, one way of reducng debt would be to ncrease cash. However, f frms also choose an optmal cashleverage (e.g., Opler, Pnkowtz, Stulz and Wllamson, 1999) then our tests need to consder the mpact of the change n the cost of equty on debt and cash separately. Accordng to Opler et al. (1999), a drop n the cost of captal ncreases nvestment opportuntes and thus rases the value of the precautonary motve of cash. Also, a reducton n the opportunty cost of holdng cash due to the lower cost of captal ncreases the optmal level of cash holdngs. Therefore, we expect frms wth hgher announcement returns to ncrease ther cash-leverage more. 2 In partcular, a reducton n the cost of captal ncreases the number and sze of the postve NPV projects. Ths lowers the free cash flow at hand and thus the agency cost of equty. At the same tme, more valuable nvestment opportuntes ncrease fnancal dstress and agency cost of debt. These three effects lower agency cost of equty; hgher fnancal dstress; and agency cost of debt concur to reduce the amount of debt n the optmal captal structure. 6
The other predcton s related to nvestment. The drop n the cost of equty should also postvely affect nvestment because the hurdle rate for projects decreases. Therefore, the trade-off theory predcts a postve correlaton between the abnormal return and nvestment, equty ssues and cash-leverage and a negatve correlaton wth debt ssues. The proceeds from an equty ssue may be employed ether towards reducng debt, ncreasng cash, 3 or ncreasng nvestment. 3.2 The peckng order theory The peckng order theory (Myers, 1984; Myers and Majluf, 1984) s based on the assumpton that managers have superor nformaton and act n the nterest of current shareholders. The exstence of an nformaton asymmetry between the managers and the market mples that equty ssues, by sgnalng overvaluaton, nduce the prce to drop. Therefore, managers would use nternal funds frst before rasng captal through bond and equty ssues. A strct nterpretaton of the peckng order theory posts that frms wll never ssue equty (Myers, 1984; Mayer and Sussman, 2004) or wll ssue equty only f they have found a way out of the asymmetrc nformaton problem (Fama and French, 2002). Thus, once we control for changes n the level of nformaton asymmetry, the strct verson of the peckng order theory predcts that managers do not react to the change n the prce of equty by ssung equty, because of the hgh cost of adverse selecton. A more flexble nterpretaton of the peckng order theory (Myers, 1984, and Myers and Majluf, 1984) allows for equty ssues f nformaton asymmetres exst only for a short perod and the cost of undernvestng s suffcently large. That s, f the change n the cost of captal ncreases the value of the nvestment opportuntes, frms should frst use cash and ssue debt and only as a last resort ssue equty to make nvestments. Therefore, n our test settng, we expect frms to ssue equty n response to a reducton n the cost of equty only f the frm s debt capacty s exhausted and the costs of undernvestng are hgh. Ths mples that the proceeds from the equty ssue should be used to ncrease nvestments. Moreover, even f the proceeds were not used for nvestment purposes, a better use of the proceeds 3 The exact proporton of the equty ssues proceeds used to ncrease cash or reduce debt depends on whether there s also an optmal level of cash holdng. Indeed, the postve correlaton between the announcement return and the change n cash holdngs s based on the assumpton that frms move towards the new optmal cash level. 7
would be to ncrease cash, because cash s nsenstve to asymmetrc nformaton problems. Unlke the trade-off theory, the peckng order theory does not predct that the proceeds of the equty ssue would be used to repurchase debt, unless the cost of fnancal dstress s suffcently large (Myers, 1984). However, usng our test settng of frms that get added to the S&P500 let s us vrtually exclude ths restructurng argument. Ths mples that the frst man predcton dstngushng the trade-off theory from the peckng order theory s that we expect the average frm to reduce debt f ssung equty f the trade-off theory s true, but do not expect to observe ths f frms follow the peckng order theory. The second man dfference s based on the effects of a shft n the level of nformaton asymmetry. The peckng order theory predcts that a shft n the degree of nformaton asymmetry can be exploted to ssue rsky securtes (Myers and Majluf, 1984 and Fama and French, 2002). In partcular, frms experencng a reducton n the degree of nformaton asymmetry should, controllng for the change n the cost of equty, ncrease the use of equty. Gven that lower nformaton asymmetry mples less undernvestment, the proceeds of an equty ssue should be nvested, not held n cash. 4 3.3 Market tmng theory The market tmng theory (Sten, 1996) assumes that equty can be msprced due to nvestor rratonalty and that managers can explot the temporary msprcng, by ssung equty. 5 Ths stands n stark contrast wth the peckng order theory, whch lnks msprcng to nformaton asymmetry problems. Issung equty s a sgnal of overvaluaton that leads to a negatve stock prce reacton. The ensung drop n prce should make t mpossble for companes to explot the temporary msprcng by ssung overvalued equty. The market tmng theory also dffers from the trade-off theory n that the decson to nvest depends on the managers nvestment horzon 6. Short-horzon managers use the mpled lower cost of equty to dscount project cash flows. Thus, f the average frm s run by short-horzon managers, the market tmng theory predcts a postve correlaton between 4 In fact, cash holdngs are expected to decrease f the reducton n nformaton asymmetry perssts because the precautonary motve for stackng cash loses mportance. 5 The market tmng theory only predcts equty ssues f the company s equty s overvalued. Undervalued frms should not ssue equty. However, n our sample, hgher abnormal announcement returns ncrease the probablty of beng overvalued. 6 Sten (1996) defnes managers as havng a short horzon f they can realze gans from ther actons before the truth s revealed. Essentally, the gans are based upon wealth transfers between new and old nvestors. 8
the abnormal return and nvestment. In contrast, long-horzon managers realze that makng nvestment decsons based upon the temporarly low(er) cost of captal would be detrmental to long-term shareholder value. Therefore, whle they would explot the wndow of opportunty and ssue equty, they would use the proceeds to repurchase debt and/or ncrease cash nstead of ncreasng nvestment. In sum, both the market tmng and trade-off theory predct a postve correlaton between the abnormal announcement return and equty ssues. However, the trade-off theory predcts a postve correlaton between abnormal return and nvestment whereas the market tmng theory predcts such a relaton only f the company s managers have a short horzon. Managers wth a long horzon would not ncrease nvestment n response to a too low a cost of equty. We summarze the predctons relatng the changes n the corporate polces to the abnormal announcement return n table A.1. n the Appendx. 4 Index addton and change n the cost of captal We proceed n three steps. Frst, we study whether the average postve short-term abnormal return around the ndex addton translates nto a decrease n the cost of equty. Second, we consder whether the corporate fnancal polces change around the perod of the addton to the S&P500 ndex. We use a seres of t-tests to assess whether and how leverage, equty ssues, debt ssues, nvestment and cash holdngs have changed. Thrd, we nvestgate the determnants of ths change. In partcular, we relate the abnormal return to the changes n the corporate fnancal polces, nvestment and cash holdngs and other control varables that proxy for the change n the cost of debt, agency problems, asymmetrc nformaton and nvestment opportuntes. Snce all the corporate polces are nterdependent, we use a smultaneous equaton framework to account for these nterdependences. Ths allows us to nvestgate whether managers react to the change n the cost of equty accordng to ether the trade-off, the peckng order or market tmng theores as descrbed n the prevous secton. 4.1 Sample selecton and data descrpton We study frms that were added to the S&P 500 ndex over the perod from 1981 to 1997. We start wth a sample of 397 frms but exclude events where frms are added to the S&P 500 ndex due to a merger, spn-off, or some other form of restructurng (85 events). 9
We also exclude fnancal frms defned as frms operatng n the SIC between 6,000 and 6,999, because ther choce of leverage dffers from those of non-fnancal frms (59 events). We requre that our sample frms have data avalable n Compustat and CRSP for stock prces and fnancal data, IBES for analyst coverage, and TFN/Insttutonal for nsttutonal ownershp. We also requre data on nsder ownershp n the year pror to the ndex addton. Ths data s collected from Valuelne and proxy flngs. These requrements elmnate another 26 events such that we end up wth 222 frms that have all the data avalable to be ncluded n our sample of S&P 500 ndex addtons. 4.2 Index addton and the cost of captal We show n table 1, panel A, that the sample frms dsplay a sgnfcant postve average abnormal announcement return of 3.21%. Ths s n lne wth pror research on ndex addton (e.g., Shlefer, 1986). The abnormal return s measured n the wndow of 5 to +5 days around the ndex addton announcement. We estmate the parameters of the market model over 254 days endng 46 days pror to the S&P 500 ndex addton announcement, usng the equally-weghted CRSP ndex as the market return. We also fnd that the nsttutonal ownershp fracton ncreases by an average 1.4 percentage ponts from the quarter before, to the end of the quarter of the ndex addton. The ncrease s statstcally sgnfcant at the 1% level. The ncreased nsttutonal ownershp s n lne wth the nterpretaton that ndex funds need to buy the newly added stock, thus causng an ncrease n demand. If the demand curve s not flat, ths phenomenon can lead to a prce ncrease. 7 These fndngs are thus n lne wth Shlefer s nterpretaton that the announcement return s caused by a downward slopng demand curve. Shlefer (1986) argues that the postve announcement return s a reflecton of a drop n the dscount rate 8 rather than a change n expected cash flows because S&P does not change ther stock ratng around the ndex addton. We drectly test ths. We follow Foerster and Karoly (1996) and estmate the change n the cost of equty by usng the Fama-French three-factor model. We consder alternatve specfcatons n whch the factor loadngs are 7 In the regresson framework we wll nvestgate the correlaton between AR and the change n nsttutonal ownershp further. 8 Ths can be seen most easly by thnkng of the present value of a perpetuty wth cash flow CF and dscount rate r 1. The event ncreases the present value from PV 1 to PV 2 wthout changng CF. Thus the new dscount rate r 2 would need to be lower than r 1. 10
computed n the 24 (12) months, endng one month before the event and n the 24 (12) months after the event, usng daly returns. For each frm, we compute the dfference n the factor loadngs (Beta Market, Beta SMB and Beta HML ) around the event and multply t by the average value of the factors (Market, SMB and HML) calculated over the 60 months pror to the event (or alternatvely over the perod from 1977 to 2000). The results are reported n table 1, panel B. They show that the cost of equty decreases usng any of the four combnatons used to calculate the change n the cost of equty. For example, when we use 24 months to calculate the betas before and after, and use the average rsk premum for the factors n the 60 months pror to the event, we observe a sgnfcant drop n the cost of equty of 1.26 percentage ponts for the average frm 9. The economc magntude s smlar to the change n the cost of captal found n the ADR lterature (e.g., Foerster and Karoly, 1999). 10 Whle our fndng of a drop n the cost of equty s consstent wth Shlefer s argument of a downward slopng demand curve for stocks, we cannot exclude the possblty that other factors, such as lqudty, mght affect the cost of equty. Indeed, f the stocks added to the ndex become more lqud, some nvestors e.g., nsttutonal nvestors wth short-term nvestment horzon and/or hgh portfolo turnover may prefer to nvest n them. The prce they would pay for ths addtonal lqudty should be related to the abnormal return. The lterature has reached mxed results about the exstence of a lqudty mprovement after the ndex addton. Benesh and Whaley (1996) fnd a permanent ncrease n the tradng volume but only a temporary decrease n the quoted spread. They therefore reject the hypothess that an ncrease n lqudty s the man drver of the abnormal return experenced by companes added to the ndex. In contrast, Hegde and McDermott (2003) fnd evdence of a long-term sustaned ncrease n the lqudty of the added stocks, mostly due to 9 In what follows we use ths verson of computng the change n the cost of equty. However, smlar results obtan usng the other versons (not shown). 10 The most sgnfcant change s the drop n Beta SMB. Also sgnfcant s the drop n Beta HML. In contrast to Barbers, Shlefer and Wurgler (2003), we do not fnd a sgnfcant change n Beta Market although the ndex addton s movng the market beta closer to 1. Ths can be explaned by our use of a three-factor model where the market return s the value-weghed CRSP ndex as opposed to Barbers et al. (2003) who use a onefactor model wth the S&P500 ndex as the market. Indeed, f we re-estmate the change n the cost of equty usng a sngle factor model, the results are consstent wth the ones of Barbers et al. (2003) (not tabulated). 11
a reducton n the drect cost of transactng. Ths takes the form of an ncrease n the quoted dollar depth, tradng volume and tradng frequency. Nevertheless, n order to control for ths lqudty ssue, we construct a varable called lqudty change. Ths varable s the dfference between the average monthly volume n the nterval [+1,+12] and [-12,-1], where volume s the number of shares traded dvded by the number of shares outstandng. In table 1, panel A, we show that tradng volume ncreases sgnfcantly, thus leavng room for the possblty that the abnormal announcement return reflects expected mprovements n lqudty of the stock. One reason for the ncreased lqudty could be that ndex funds are more wllng to lend the stock, thus reducng the cost of transactons for short sellng. 4.3 The effect on corporate polces We now defne our proxes for the changes n the corporate polces. We frst compute raw measures, whch are based on changes n the event frm only. The varables of nterest are leverage, stock and debt ssues, nvestment and cash holdngs. We use data provded by Compustat and follow Baker and Wurgler (2002) n defnng leverage as the book value of debt dvded by total assets (#6) 11, where the book value of debt s total assets mnus book value of equty. The book value of equty s defned as total assets mnus total labltes (#181) and preferred stock (#10), plus deferred taxes (#35) and convertble debt (#79). We defne cash holdngs as cash and short term nvestment (#1) dvded by total assets. Investment s captal expendtures (#128) plus acqustons (#129), dvded by total assets. We defne stock ssues as the common and preferred stock ssued mnus common and preferred stock repurchased (#108-#115), dvded by total assets. 12 Smlarly, debt ssue s long-term debt ssued mnus long-term debt retred (#111-#114) plus changes n short-term debt (#301), dvded by total assets. For the leverage and cash holdngs, we use the change from the fscal year end before to the fscal year end one year after the event,.e., the level at year t+1 mnus the level at year t-1. For nvestment, stock ssue and debt ssue, we use the 11 The numbers n parentheses ndcate the reference to the Compustat data tem number. 12 Qualtatvely smlar results obtan f we use the change n the book value of equty usng the balance sheet nformaton nstead of the cash flow statement. Equty ssues not-for-cash are also ncluded n the balance sheet measure. 12
average of the varables n year 0 and year 1. We choose these defntons because frms may have a target leverage and cash holdng, whch requres studyng the change from the prevous levels, whle for nvestment, stock ssue and debt ssue, no such target level exsts. Flow measures are thus approprate to study the cross-sectonal mplcatons of the captal structure theores. 13 The use of raw measures has two man lmtatons. Frst, over the two-year perod over whch we gauge these changes other common factors could affect a company s choce of leverage, nvestment or cash holdng. Second, durng the long perod on whch our analyss focuses (1981-1997), many changes may have taken place n the way both the market and managers react to the ndex addton announcement. Ths could affect our analyss. In order to control for these two potentally confoundng effects, we compute a net measure. The net measure s based on the dfference between the event frm and a sample of control frms. To mplement ths, we construct a portfolo of three comparable frms for each event frm. These comparable frms are not n the S&P500 ndex and are not added to t durng the event year. To fnd the comparable frms, we match the event frm wth all the other non-s&p 500 frms n the same ndustry, measured at the one-dgt SIC level. The three matchng frms are selected accordng to fve crtera: sze, market-to-book, leverage, number of analysts, and nsttutonal ownershp. In partcular, we frst compute the absolute value of the percentage dfference between the event and non-event frms along all the fve crtera. Then, we rank the non-event frms accordng to the overall dfference, whch we defne as the sum of the percentage dfference of each of these fve crtera. Fnally, we choose the three non-event frms wth the smallest overall dfference to form an equally weghted portfolo, whch we call the matchng sample frms. Based upon the dfferences between the event and matchng sample frms, we construct net measures as our prmary proxes to assess the changes n captal structure, nvestment and cash holdngs around the ndex addton. In table 2, panel A, we report unvarate statstcs of the corporate fnancal polces, nvestment and cash holdng varables. Consstent wth the argument that the announcement return has reduced the cost of equty, we fnd an ncrease, albet nsgnfcant, n the average net stock ssues of 0.56 percentage ponts (medan 0.48 percentage ponts and sgnfcant at 13 For robustness tests we have also used SDC data on equty ssue and repurchase announcements to only select equty transactons n the 24 months after the ndex addton. SDC data also excludes ncreases n equty due to opton exercses. Results are smlar to the ones reported below and are omtted for brevty. 13
the 5% level) 14. There s no sgnfcant ncrease n the net debt ssues. Overall, ths leads to a sgnfcant reducton n net leverage of 1.67 percentage ponts (medan 2.77 percentage ponts and sgnfcant at the 1% level). These unvarate statstcs seem to lend some support to the noton that managers are reactng to the change n the cost of equty mpled by the ndex addton by ssung more equty and reducng leverage. We do not fnd a consstent result for the change n net nvestment. The mean ncreases sgnfcantly by 1.29 percentage ponts, the medan, however, decreases by 0.05%. We also do not fnd any sgnfcant change n net cash holdngs around the event. Therefore, t s dffcult to argue that the event had a sgnfcant effect on the asset sde. However, a word of cauton s requred n nterpretng these unvarate results. Based on our hypotheses developed n secton 3, t s obvous that the corporate polces are heavly nterlnked. Before drawng any conclusons about managers reactons to the event, we wll have to study the changes n a smultaneous equaton framework usng addtonal control varables. 5 Determnants of changes n corporate fnancal polces 5.1 Methodology We now proceed to test how companes change ther corporate polces as a functon of the change n the cost of captal due to the S&P500 ndex addton. The predctons of the models outlned n secton 3 above can only be tested f we study all the corporate polces polces smultaneously. For example, equty ssues may smultaneously affect cash holdngs, nvestment and debt ssues. By the same token, an equty ssue can also be affected by the need for cash, nvestment opportuntes, or adjustments to leverage. To control for these factors, we use a smultaneous equaton framework. The system contans four equatons determnng the corporate polces, and one determnng the abnormal return. To estmate the system, we use a three-stage least squares (3SLS) estmator. Ths procedure s consstent and asymptotcally effcent for normally dstrbuted dsturbances (e.g., Greene, 1997). Moreover, t has the advantage of estmatng the full covarance matrx, 14 Whle t s nterestng to note that equty ssues have margnally ncreased relatve to the matchng sample frms after the ndex addton, the average ncrease s economcally relatvely small. One possble explanaton s that the average change n the stock prce due to the ndex addton s small compared to the transacton costs of ssung equty. However, the transacton cost argument should not affect the cross-sectonal predctons of the captal structure theores,.e., hgher abnormal announcement returns should be postvely related to hgher net equty ssues. 14
thus properly accountng for the correlatons n error terms across regressons. Ths s especally mportant n our test settng as all the four corporate polcy equatons refer to the same company and tme. 15 We estmate the followng system: AR = a 1 + a NetDebtIssue = c + c 2 NetStockIssue NetCashChange = d + d NetInvestment InstownerChange + a = b + b 1 1 1 1 + a 2 2 = e + e 2 6 AR + b AR + c 2 3 3 AR + d AR + e 3 3 3 NetStockIssue + a NetInvestment + µ C 1 1 + ε 1 NetDebtIssue + b NetStockIssue + c NetStockIssue + d NetStockIssue + e 4 4 4 4 4 NetDebtIssue + a NetCashChange + b NetCashChange + c NetDebtIssue + d NetDebtIssue + e 5 5 5 5 5 NetCashChange NetInvestment + µ C NetInvestment + µ C NetCashChange + µ C 3 5 2 4 3 NetInvestment + µ C 5 2 + ε 4 + ε + ε 3 + ε 5 2 4 (1) where NetStockIssue, NetDebtIssue, NetCashChange and NetInvestment are the corporate polces of company. AR s the abnormal return n the nterval 5 to +5 days around the ndex addton announcement, and C x s a vector contanng control varables. Snce the set of control varables are specfc to each regresson n order to dentfy the system, we next defne and descrbe n whch regresson they are ncluded. The unvarate statstcs of these varables are reported n table 2, panel B and the exact defntons are gven n the Appendx. 5.2 Regresson specfcaton We nclude the change n nsttutonal ownershp n the AR regresson, as the ndex addton creates demand for the stock from ndex funds. Wth a downward slopng demand curve for stocks, we expect a postve coeffcent on ths varable n the AR regresson. We also nclude nsttutonal ownershp n the other regressons to control for the possblty that nsttutonal owners affect corporate polcy drectly, for example, through ther shareholder actvsm. In partcular, one could expect that ncreased nsttutonal ownershp could lead to better access to external captal. Not controllng for the change n nsttutonal ownershp could thus lead to a spurous correlaton between the event abnormal return, and equty or debt ssues. In the AR regresson, we also nclude our proxy for the changes n lqudty of the stock. Ths varable s also added to the net stock ssue regresson to control drectly for the effect of changes n lqudty around the event on net stock ssues. In addton, we control for 15 Usng 2SLS nstead, we fnd the coeffcents to be smlar but less sgnfcant as expected (e.g., Greene, 1997). 15
the lagged level of lqudty n the net stock ssue regresson to control for the possblty that the level of lqudty affects the amount of stock ssues. Insder ownershp, an nverse proxy for agency problems between managers and shareholders, could affect the abnormal return as well as the reacton to t on the corporate sde. We therefore nclude ths varable n all the regressons. The change n the number of analysts s used as a proxy for varatons n nformaton asymmetry around the event. The peckng order theory predcts a postve correlaton between the change n the number of analysts and nvestment, equty and debt ssues, but a negatve correlaton wth changes n cash holdngs. The varable s thus ncluded n all regressons. The net change n bond ratngs s used to control for the potental change n the cost of debt and s ncluded n all regressons. We also control for pror stock market condtons, both market-wde as well as stockspecfc. The former nclude the pror market return and the pror market volatlty. Potentally, market wde fluctuatons can affect a company s decson to ssue equty and determne the ntensty of the market s reacton to the ndex addton. We thus nclude these two varables n the AR and net stock ssue regressons. The stock-specfc condtons are proxed by the stock s net run-up. Net run-up controls for pror frm-specfc excess stock return that has been shown to affect equty ssue decsons (Lucas and McDonald, 1990) as well as for market expectatons. For example, f the market antcpates the company s lkelhood of gettng added to the ndex, then arbtrageurs mght ncrease ther nvestment pror to the announcement. We also nclude net run-up n the debt, nvestment and cash holdng regressons, snce a run-up can proxy for recent changes n valuaton that affects a frm s cost of equty. We nclude the term premum as a proxy for the slope of the term structure n the debt ssue, net cash change and AR regresson. Ths allows us to control for the fact that the nterest rate envronment n the early 1980s was very dfferent from that n the 1990s, dfferentally affectng the choce of debt versus cash polcy. Accordng to the trade-off theory, the benefts of debt are smaller f the cash flow stream s rsker. We use earnngs volatlty, measured as the standard devaton of earnngs over the fve years pror to the ndex addton, as a proxy for the tax benefts and bankruptcy 16
cost of debt. Gven that the trade-off theory s concerned wth debt mnus cash, we also nclude earnngs volatlty n the cash regresson. In all corporate polcy regressons, we control for the effects of sze and market-tobook and level of cash holdngs n year -1. If there s an optmal cash leverage, then t s possble that the changes n cash holdngs reflect mean reverson. Fnally, we nclude dentfyng nstruments for each equaton separately. In the AR regresson we add the pror stock return volatlty measured over the 12 months before the ndex addton. Accordng to Pastor and Verones (2003), hgher return volatlty mples hgher nformatonal uncertanty and accordng to Baker and Savasoglu (2002), hgher arbtrage rsk. These effects weaken arbtrage, preventng an mmedate adjustment of the stock prce, reducng the abnormal return around the event. Therefore, we expect a negatve correlaton between abnormal return and stock volatlty. It s worth notng that volatlty effectvely proxes for arbtrage behavor after the announcement. Before the announcement, any potental leakage of nformaton has been already controlled for by the ncluson of the runup varable, proxng for possble propostonng of arbtrageurs. We argue that the pror stock return volatlty does not affect the net corporate polces because the rsk factor s controlled for by the matchng procedure. In the net stock ssues equaton we add the lagged dvdend to asset rato. John and Wllams (1985) predct a postve correlaton argung that dvdends serve to reveal the true value of the stock and makng equty ssues less costly. On the contrary, Loderer and Mauer (1992) fnd evdence rejectng ths hypothess. Thus, we expect the dfference between the sample and matchng frm s stock ssue to be negatvely affected by the level of the dvdend. The dentfyng nstrument n the net debt regresson s the lagged level of the ratngs. We expect frms wth lower ratngs to be more lkely to dffer from ther matchng sample frms because debt ssues are more dffcult/expensve for such frms. Snce frms that do not have a debt ratng cannot be ftted nto ths ordnal ratng varable, we add separately a dummy varable equal to one f the frm s rated. The dentfyng nstrument n the net cash change regresson s the sales volatlty, measured as the standard devaton of the logarthm of sales over the fve years pror to the ndex addton. A hgher sales volatlty s expected to negatvely affect the change n net cash because frms wth hgher sales volatlty hold a hgher cash-to-asset rato and thus any change s expected to be relatvely smaller. In the net nvestment regresson, we nclude the rato of nvestment 17
to total assets of the event frm n year-1. Ths allows us to control for the possblty that nvestment - whch ncludes acqustons - s lumpy. We expect t to be postvely related to net nvestment. 16 We do not expect the lagged nvestment to affect current net fnancng decsons. In the frst-stage regresson, we use as nstrumental varables, all the exogenous varables, plus year dummes. 17 The results of the frst stage, the R-squares of the frst stage (not reported) regressons are between 12% (net debt ssue regresson) and 41% (net equty ssue regresson). 5.3 Results The results of the estmate of the system of equatons are reported n table 3. Frst, the change n nsttutonal ownershp s sgnfcantly postvely correlated wth the abnormal announcement return. Ths fndng s n lne wth the earler studes (e.g., Shlefer, 1986), argung that at least a part of the market s reacton s due to a downward slopng demand curve for stocks. The nsgnfcant coeffcent on the proxy for the change n lqudty suggests that ether our proxy s not good enough or that lqudty s n fact not a major determnant of the announcement return. Ths s n lne wth the concluson drawn by Benesh and Whaley (1996) 18. We also fnd a negatve but nsgnfcant coeffcent on the net run-up. The negatve sgn s consstent wth the nterpretaton that some nsttutonal nvestors already buy stocks n antcpaton of the stock beng added to the ndex thus reducng the market mpact at the announcement. Second, and more mportantly, there s evdence that ncluson n the ndex nduces managers to ncrease the share of equty n the captal structure. The coeffcent n the net stock ssue regresson s 0.812, ndcatng that frms wth a hgher abnormal announcement return ssue more equty than the matchng sample frms do n the same perod. Besdes rejectng the strct verson of the peckng order theory, ths correlaton does not allow us to tell the captal structure theores apart as t s consstent wth both the peckng order and the 16 Imposng ths structure on the system of equatons makes all equatons overdentfed. Also the rank condtons are satsfed snce we excluded at least four exogenous varables from each equaton. See secton 5.5 for a dscusson of the robustness tests of the 3SLS estmaton. 17 As shown n Baltag, 1998, (p. 278) all exogenous varables need to be ncluded n each frst-stage regresson. 18 Notce that our fndng supports ther concluson that lqudty s not a major determnant of AR by relatng lqudty drectly to AR whle they come to ther concluson based on the lack of a permanent change n the bd-ask spread, despte the fact that the average tradng volume ncreased. 18
trade-off theory and s also n lne wth the market tmng theory. Nevertheless, the fndng s mportant as t suggests a potental role for the companes themselves as arbtrageurs. Thrd, regardng the asset sde, the abnormal return s sgnfcantly postvely correlated wth net nvestment, and sgnfcantly negatvely correlated wth changes n net cash holdngs and net debt ssues. The former correlaton s n lne wth our earler fndng that the cost of equty decreases after the event and suggests that nvestment s ncreased n response to the equty prce change. The ncrease n nvestment s consstent wth the tradeoff theory, the peckng order theory as well as wth the market tmng theory under the assumpton that the average frm s run by managers wth a short horzon. The negatve correlaton between the abnormal return and the change n cash holdngs s nstead the opposte of what the trade-off and peckng order theory would have predcted. One potental explanaton for the negatve correlaton s that frms where the abnormal return s hgher use more of the cash for nvestments or reductons n debt. As we mentoned, one dstngushng feature that sets the peckng order theory apart, s the predcton that proceeds from equty ssues are not used to reduce debt as allowed by the other theores. We fnd that net stock ssues are negatvely correlated wth net debt ssues, as predcted by the trade-off and market tmng theores. For example, the coeffcent on net stock ssues n the net debt regresson s -0.277 and sgnfcant at the 2% level. Economcally, ths suggests that for every one-dollar ncrease n net equty proceeds, the frm reduces ts net debt ssues by 27 cents. It s nterestng to observe that the negatve correlaton s drven by frms that ssue new equty, and not by frms that ssue debt and repurchase equty, snce the coeffcent on net debt ssue n the net equty regresson s nsgnfcant. 19 Furthermore, the abnormal return s sgnfcantly negatvely related to net debt ssues wth a coeffcent of 0.431 (p-value of 0.04) also supportng the trade-off theory argument that the change n the cost of equty reduces the amount of debt rased. Let us now focus on the change n the degree of nformaton asymmetry. Accordng to the peckng order theory, a reducton n nformaton asymmetry should nduce frms to 19 Snce one could argue that a negatve correlaton between debt and equty ssues could also be consstent wth the peckng order argument where frms ssue debt f they have debt capacty and equty but no debt f they don t, we have re-estmated the system wth only the frms that have ssued equty (189 frms) and also fnd a sgnfcantly negatve correlaton between equty ssues and debt ssues ndcatng that the correlaton s drven by frms that ssue equty to reduce debt (results not tabulated for brevty). 19
ssue more equty and have lower cash holdngs 20. The fndngs support ths predcton. Indeed, there s a postve and sgnfcant correlaton between our proxy for the reducton n the level of nformaton asymmetry - net change n the number of analysts - and net stock ssues. Moreover, there s a negatve, albet nsgnfcant correlaton between the net change n the number of analysts and the net change n cash. However, the peckng order theory also posts that a reducton n nformaton asymmetry should nduce frms to under-nvest less as they are now, on average, more valued n lne wth managers' vew. In other words, frms should rase external captal prmarly to cover nvestment expenses. Ths predcton s not supported by the data gven the lack of correlaton between the change n nformaton asymmetry and nvestment as well as equty ssues and nvestment. Therefore, the analyss of the changes n the corporate polces condtonal on changes n nformaton asymmetry s only partally consstent wth the predctons of the peckng order theory. Fnally, the lack of correlaton between equty ssues and nvestment s also contrary to the predctons of the agency models based on empre buldng managers. However, the negatve correlaton between net equty ssue and net debt ssue s n lne wth the predctons of agency models (Jensen and Mecklng, 1976, Jensen, 1986, Morellec, 2004). Indeed, ths would be consstent wth a story n whch managers react to an unexpected ncrease n stock value by ssung more equty, and usng the proceeds to reduce debt gven ther preference for low(er) leverage (Morellec, 2004). To test the agency model more drectly, we nclude the level of nsder ownershp at the fscal year end pror to the ndex addton. We fnd that frms wth hgher nsder ownershp rase sgnfcantly more equty, reduce net cash holdngs, and do not dsplay a dfference n ether nvestment or debt ssues. Thus, assumng that hgher nsder ownershp proxes for fewer agency problems, the data does seem to be nconsstent wth some of the predctons of the agency model, namely that hgher agency problems lead to hgher nvestment (more overnvestment), hgher use of equty and lower use of debt. A word of caveat apples here. It s possble that the sample we have selected, namely S&P500 frms, at 20 However, t s worth notng that a smlar predcton s made by the trade-off theory as hghlghted n Opler, Pnkowtz, Stulz and Wllamson (1999). If the wedge between the cost of nternal and external captal decreases, the precautonary motve for holdng cash looses mportance, therefore predctng a negatve correlaton between changes n the net number of analysts and cash holdngs. 20