Downward Sloping Demand Curves for Stock and Leverage

Transcription

1 Downward Sloping Demand Curves for Sock and Leverage Liem Pei Fun * Faculy of Economics Lecurer, Pera Chrisian Universiy Surabaya pfun@peer.pera.ac.id Absrac This research aemps o invesigae he effec of downward sloping demand curves for sock on firms financing decisions. For he same size of equiy issuance, firms wih seeper slope of demand curves for heir socks experience a larger price drop in heir share price compare o heir counerpars. As a consequence, firms wih a seeper slope of demand curves are less likely o issue equiy and hence hey have higher leverage raios. This research finds ha he seeper he slope of demand curve for firm s sock, he higher he acual leverage of he firm. Furhermore, firms wih a seeper slope of demand curves have higher arge leverage raios, signifying ha hese firms prefer deb o equiy financing in order o avoid he adverse price impac of equiy issuance on heir share price. Keywords: slope of demand curves for socks, leverage, financing decisions. * The auhor is indebed o Dr. Xin Chang (The Universiy of Melbourne) for his valuable inpu, insighful commens, and helpful suggesions.

2 INTRODUCTION In an economic sense, he price of all asses is deermined by supply and demand. However, in finance, he share price which reflecs fundamenal value of a firm is normally assumed no o be deermined by supply and demand, bu is obained from discouning he firm s expeced fuure cash flows by is cos of capial. This is because classical finance heories assume a horizonal demand curve for firms equiy and hence he share price is independen of supply. The erm horizonal demand curve, perfecly elasic demand curve, infinie elasic demand curve is idenical and hey will be used inerchangeably in his paper. Conrary o he assumpion of horizonal demand curves, many researchers (among ohers, Shleifer (1986), Loderer, Cooney, and Drunen (1991), Kaul, Mehrora, and Morck (2000)) have found ha he demand curve for firms equiy is acually downward sloping. Conrolling for informaion effecs, Shleifer (1986) ess he hypohesis of downward sloping demand curves by examining socks price movemen afer hey are included ino S&P 500 Index. One would no expec index inclusions o resul in a price effec if demand curve is horizonal. In conras, he finds a share price increase a he announcemen of he inclusion suggesing ha demand curves for sock do slope down. This downward sloping demand curves for sock imply ha shares need no be priced exacly a heir fundamenal values. A downward sloping demand curve for socks suggess ha new equiy issues resul in sock price decreases, herefore firms need o ake ino accoun his price effec when making financing decisions. Since he magniude of his price effec depends on he slope of each individual firm s demand curve, he following ineresing quesion can be raised: How does downward sloping demand curve for socks affec individual firm s financing decisions? Inuiively, firms wih a seeper slope of demand curves are more concerned abou he price impac of equiy issuance because he same amoun of addiional equiy supply (issuance) causes a larger price drop in heir share prices compared o firms wih a flaer slope of demand curves. Consequenly, ceeris paribus, firms wih a seeper slope of demand curves for heir socks are less likely o issue equiy and hence we should observe higher acual leverage raios for hese firms. Furhermore, o he exen ha each firm has an opimal leverage raio and cos of equiy issuance is posiively associaed wih he slope of demand curve, firms wih a seeper slope of demand curves would have higher arge leverage raios. These predicions are he main hypoheses invesigaed in his paper. Hypoheses esing are performed using cross-secional leverage regressions and he arge adjusmen model. Following Peajiso (2004), Haggard and Pereira (2005), Hir and Pandher (2005), Baker, Coval, and Sein (2006), his research paper employs idiosyncraic risk, esimaed using various mehods, o proxy for he slope of demand curve. Leverage raio is calculaed in boh marke and book erms. Main findings in his research are consisen wih preceding predicions. Firs, firms wih a seeper slope of demand curve for sock have higher acual leverage raios. The slope of demand curves for socks is posiively and significanly relaed o acual leverage raios, even afer conrolling for ohers facors shown in prior sudies o influence firms leverage. Second, he slope of demand curves for sock is a posiive and significan facor deermining firms arge leverage raios suggesing ha firms wih 2

3 seeper slope of demand curves are less likely o issue equiy and prefer deb insead as a mean of financing. This research can poenially conribue o he lieraure in several ways. Firs, i sheds ligh on he relaionship beween he slope of he demand curve for socks and firms capial srucure. The resuls show ha firm s leverage is posiively and significanly affeced by he slope of demand curve for is sock. This finding also imply ha sudies in capial srucure should consider he slope of demand curves for sock as one of he conrol variable when hey aemp o examine he impac of paricular facor on he firms leverage. Second, his research parly fills he gap beween capial srucure heories and observed behaviour of firms financing decisions. Main findings in his research indicae ha firms do and should concern abou he slope of demand curves for heir shares when hey make financing decisions. In addiion o he facors menioned in prior lieraures, he resuls sugges ha he slope of demand curves for sock provides addiional explanaion for he cross-secional differences in firms leverage. Specifically, he downward sloping demand curves for sock is anoher imporan reason why firms end o use deb financing insead of equiy financing. The remainder of his research paper is organized as follows. Secion 2 provides a brief lieraure review. Secion 3 presens he hypoheses developmen. Secion 4 describes he empirical mehodology adoped in his research. Secion 5 presens he daa and esimaion procedures. Secion 6 repors he resuls and provides discussions of he resuls. The final secion concludes. LITERATURE REVIEW In his par, i will be discussed he downward sloping demand curves for socks, followed by an ouline of major capial srucure heories and he implicaion of downward sloping demand curves for socks on each capial srucure heory. Downward Sloping Demand Curves for Socks One of he mos imporan assumpions underlying several prominen finance heories is he invesors abiliy o buy and sell any amoun of a firm s equiy wihou any price impacs, which suggess he demand curve for a firm s equiy is horizonal. For example, he home leverage argumen behind Modigliani-Miller capial srucure heorem relies on he exisence of perfec capial marke, where he horizonal demand curve is one of he key condiions. Shleifer (1986) poins ou ha in CAPM and APT models, sock price is unbiased predicor of fundamenal value, mainained hrough he workings of arbirage. Provided close subsiues exis for a sock, is fundamenal value, which equals o he expeced cash flows discouned by he cos of capial, is independen of he supply of equiy. Therefore, he demand curve for a sock should be almos perfecly horizonal and we should observe virually no price impac (Peajiso, 2004). In oher words, firm can sell whaever quaniy of sock i desires wihou concerning abou he falling in is share price, because a horizonal demand curve suggess ha he marke can always absorb he exra supply a he fundamenal value. Conrary o he heoreical assumpion, here are number of sudies suggesing ha he demand curves for socks are downward sloping. Scholes (1972), Holhausen, Lefwich and Mayers (1990), and Mikkelson and Parch (1985) examine sock price reacions o buyer and seller iniiaed large block rades and hey documen negaive 3

4 price reacions o large block sales and posiive price reacions o large block purchases. However, in a world of informaion asymmery, a block rade o buy (sell) may signal good (bad) news abou he sock and hus enailing a price increase (decrease). Since hese prior ess fail o disinguish wheher i is he signalling effec or he downward sloping demand curve causes he price impac, hey can no provide conclusive evidence on he hypohesis of he downward sloping demand curves for socks. Forunaely, subsequen sudies address his problem. To conrol for he impac of informaion effecs o minimal, Shleifer (1986) ess he hypohesis of downward sloping demand curves by examining socks price movemen afer hey are included ino S&P 500 Index. If demand curve is horizonal, one would no expec index inclusions o resul in any price effec. However, if demand curves are downward sloping, we should observe a share price increase on he announcemen of he inclusion. His conclusion is in favour of downward sloping demand curves for sock. Similarly, in sudy of posiive price response o share repurchase announcemens, Haggard and Pereira (2004) find ha price response is greaer for socks wih seeper demand curves. Kaul, Mehrora, and Morck (2000) use change in supply even o deec downward sloping demand curves. If horizonal demand curves for socks exis, wihou any new informaion, sock prices should no be affeced by a shif in supply. They use an even when Torono Sock Exchange implemened a previously announced redefiniion of he public floa. Public floa is percenage of he firm s equiy ha mus be freely raded on he exchange in order o mainain a public firm saus. When an exchange redefines public floa, his acion does no conain any informaion on firms fundamenals, i merely change number of freely raded shares available in he marke. Therefore, his redefiniion can be perceived as pure supply even. Since he revision conveyed no informaion, one would no expec o observe any price effec if he demand curve for sock is fla. Conrary o he expecaion, he affeced socks experienced saisically significan excess reurn during he even week and no price reversal occurred as rading volume reurned o normal levels. These findings suppor he hypohesis ha demand curves for socks slope down. Loderer, Cooney, and Drunen (1991) sudy he price elasiciy of demand for he common sock of an individual corporaion by invesigaing he price change afer announcemen of sock offering. Their finding suggess ha he price decline is caused by finie price elasiciies of demand, herefore hey conclude ha issuing firms canno rea he demand for heir sock as if i were perfecly elasic. Prior sudies no only deec he presence of downward sloping demand curves bu also aemp o provide probable reasons for heir exisence. The plausible driving forces behind downward sloping demand curves can be classified ino he following four caegories: limied arbirage, liquidiy induced compensaion, shareholder heerogeneiy, and divergence opinion among analyss. Wurgler and Zhuravskaya (2002) claim ha he absence of perfec subsiue makes arbirage aciviies risky, hence risk-averse arbirageurs are relucan o engage in unlimied arbirage, which resuls in downward sloping demand curve for sock. They examined he exen of socks price jumps afer inclusion ino he S&P 500 Index and found ha arbirage forces are weakes, and oher pricing anomalies are severes, among socks wihou close subsiues. In line wih he argumen above, Shleifer and Summers (1989) argue ha arbirage is insufficien o couner invesor senimen and Mayshar 4

5 (1978) poins ou ha any firm may provide invesors wih a ype of hedging hey canno duplicae wih he shares of oher firms. Even on organized exchanges, he marke is no uniformly broad, deep, and resilien. Since invesors value liquidiy, raders are willing o ake on large posiions in illiquid asses only if appropriaely compensaed wih a lower price (Amihud & Mendelson (1986), Kamara (1989), and Warga (1990)). If he required discoun is increases wih he size of he posiion, he aggregae demand schedule for a financial asse could be downward sloping (Loderer e al., 1991). Incomplee informaion cause risk averse invesors araced o differen ses of risky asses (Meron, 1987). Alernaively, due o differen informaion available (Parsons and Raviv, 1985) or differen inerpreaions of he same informaion, invesors may have differen reservaion prices for he same securiy. Since i akes a lower price o induce a larger number of invesors wih differen reservaion prices o buy, he aggregae demand schedules could be downward sloping (Loderer e al., 1991). Disagreemen among analyss gives a good indicaion of he riskiness of a securiy which in urn resuls in higher required reurn from invesor. Therefore, when selling equiy, firm has o offer discoun from he fundamenal value of is sock. This suggess ha demand curve of firm s sock is no longer horizonal (Miller, 1977). Taken ogeher, he downward sloping demand curve implies ha asses need no be priced exacly a heir fundamenal values (Peajiso, 2004). A new sock issue will cause a permanen sock price decrease (Barclay & Lizenberger, 1988), even he issuance iself does no conain any negaive signalling effec. Therefore, he implicaion for issuer is when an equiy issue is conemplaed, he share price decrease resuling from downward sloping demand curve is a major cos ha has o be raded off agains he possible benefis from equiy issue. Capial Srucure Theories Ever since Modigliani and Miller (1958) irrelevance proposiion, a large number of sudies claim ha financing decisions does maer. To dae, here are four main heories explaining firms financing decisions. They are rade-off heory, pecking order heory, agency heory, and he marke iming heory. In rade off heory, he balance beween coss and benefi of deb is he major deerminan of opimal capial srucure. Benefis of deb include ax deducibiliy of ineres while coss of deb include poenial bankrupcy coss. The rade-off heory predics ha he firm will achieve he opimal capial srucure by borrowing up o he poin where he marginal value of ax shields on addiional deb is jus offse by he increase in he presen value of possible coss of financial disress. We believe ha downward sloping demand curve for socks herefore has no impac on firm financing decisions under his rade-off heory as he shape of demand curves affecs neiher axsavings nor bankrupcy coss. In pecking order heory proposed by Myers (1984), firms finance new invesmens wih he following order of preference: reained earnings, safe deb, risky deb, and ouside equiy. This heory predics ha capial srucure of he firms is he resul of he pecking order financing and variaion in a firm s leverage is no driven by he coss and benefis of deb bu by he firm s financing defici. Pecking order heory says ha he firm will borrow, raher han issuing equiy, when inernal cash flow is no sufficien o fund capial expendiures, herefore observed deb raios will reflec he cumulaive 5

6 requiremen for exernal financing a requiremen cumulaed over an exended period (Myers, 1984). Driven by informaion asymmery problem, pecking order heory suggess ha when exernal financing is necessary, firms prefer o issue deb while reaing equiy as he las resor of financing. When informaion asymmery exiss, invesors associae equiy issuance wih managemen s belief ha he firm s equiy is currenly overvalued. As a resul, equiy issuance is usually accompanied by share price decrease. To avoid he falling share price, equiy is he leas preferred mehod of exernal financing. The downward sloping demand curve for socks amplifies he negaive price impac when firms issue equiy. Equiy issuance can be seen as increase in supply which leads o decline in share price if he demand curve for socks is downward sloping. Therefore, combining pecking order heory and he downward sloping demand curve for socks, we inerpre ha firm s equiy issuance will be more unlikely. Under he agency heory, agency coss can arise from conflics beween bondholders and sockholders and conflics beween sockholders and managers. Conflics beween deb-holders and equiy-holders inroduce incenive disorion problems namely deb overhang (under-invesmens), risk shifing (asse subsiuion), managerial myopia (shor sighed), and relucance o liquidae (hang on o losers). These incenive disorion problems creae agency coss of subopimal invesmens and operaing decisions. Thus, many growh firms end o rely on equiy o avoid losing he financing flexibiliy in he fuure and he agency coss beween equiy-holders and debholders. On he conrary, agency conflics beween equiy holders and managers can be resolved hrough high deb raio. The need o consanly service deb paymen forces managers o generae and pay ou cash. As a resul, he free cash flow available and hence manager s opporuniies for value desroying behaviour are grealy reduced. Since his heory essenially describes he cos and benefis of using deb from he perspecive of agency problems, he downward sloping demand curve for sock has no implicaion for corporae financing decisions under his heory. Marke iming heory suggess ha firms end o issue equiy when marke value is high, relaive o book value and pas marke values, and o repurchase equiy when marke value is low. Hence, Baker and Wurgler (2002) poin ou ha curren capial srucure is he cumulaive oucome of pas aemps o ime he equiy marke. Graham and Harvey (2001) repor ha he exen o which companies shares are perceived o be overvalued or undervalued is one of managers mos imporan consideraions, when hey choose he iming for equiy issuance. This resul is clearly supporing marke iming behaviour by firms. The downward sloping demand curves for socks sugges ha firms face price pressure when hey issue equiy. The exen of price pressure varies beween firms due o he difference in he slope of heir sock s demand curves. Compared o heir counerpars wih flaer demand curve, firms wih seeper demand curve face higher drop in heir share price. Hence, we predic ha hey are more relucan o issue equiy when he condiion is unfavourable (when share price is undervalued), while issuing more equiy when iming becomes favourable. In shor, we argue ha firms wih seeper demand curve for heir socks will be less likely o issue equiy bu more likely o engage in marke iming behaviour as hey have more incenive o do so. 6

7 Capial Srucure Decisions: A Final Trade-Off Each capial srucure heory oulined above has is own srengh and drawback. There is no single heory o dae ha can fully explain he mix of deb and equiy for each firm. Each heory works beer han ohers under paricular circumsances. As poined ou by Myers (2001) There is no universal heory of he deb-equiy choice, and no reason o expec one. There are several useful condiional heories, however. To conclude, he final rade-off faced by firms when hey deermine heir capial srucure is synhesized in he following able. Advanages of Deb Disadvanages of Deb 1. Tax Benefis: he higher he ax raes, he higher he ax benefi. 2. Added Discipline: he greaer shareholder-manager agency problems, he greaer he benefi of deb. 3. Avoid Sending Adverse Signals: he greaer he informaion asymmery, he greaer he benefi of issuing deb han equiy because deb issuance has less pronounce announcemen effecs. 1. Direc Bankrupcy Coss: he higher he business risk, he higher he cos of bankrupcy. Hence lower he abiliy o ake financial risk. 2. Incenive Disorion: he greaer he separaion beween sockholders and lenders, he higher he cos because he incenive disorion problems will be more severe. 3. Loss of Fuure Financing Flexibiliy: he greaer he uncerainy abou fuure financing needs, he higher he cos. 4. Forgo he Gain of Timing he Equiy Marke This final rade-off implies ha each firm has arge leverage raio which is he balance beween he advanages and disadvanages of deb usage. To he exen ha informaion asymmery and sock marke condiions affec managers financing decisions, he arge leverage raio should reflec he effec of downward sloping demand curves for socks. HYPOTHESIS DEVELOPMENT In erms of predicing firm s capial srucure, alhough downward sloping demand curve for socks has no impac on eiher classical rade-off heory or he agency heory, i implies ha equiy issuance is less likely under pecking order heory and marke iming heory. A each poin in ime, differen firms will likely o have demand curve wih differen slope. Therefore, he exen of price pressure faced by equiy-issuing firms varies. Firms wih a seeper slope of demand curves will suffer more if hey issue equiy because he drop in heir share price will be larger han firms wih a flaer slope of demand curves. As a resul, firms wih a seeper slope of demand curves for heir sock end o prefer deb han equiy hence we should observe higher deb raio for hese firms. This leads o he firs hypohesis, oher hings being equal, i is expeced ha firms wih a seeper slope of demand curves have higher leverage raios. 7

8 Hypohesis 1: Oher hings being equal, firms wih a seeper slope of demand curves end o have higher (acual) leverage raios. The firs hypohesis scruinizes he relaionship beween he slope of demand curve for sock and deb raio a a poin in ime. However, he final synhesized rade-off beween he benefi and cos of deb suggess ha firms inend o mainain a arge deb raio, hence when a firm s deb level is above he arge raio, i should issue equiy; when below he arge, i should issue deb. This leads us o he second hypohesis as follows. Hypohesis 2: Oher hings being equal, firms wih a seeper slope of demand curves end o have high (arge) leverage raio. EMPIRICAL METHODOLOGY In his secion, proxies for he slope of demand curves will be discussed firs before presening he empirical models o es hypoheses regarding he relaionship beween he slope of demand curve for socks and firms financing decisions. Proxies for he Slope of Demand Curve This research employs idiosyncraic risk as proxy for he slope of demand curve. This proxy is commonly used by many researchers (Baker, Coval, and Sein (2006), Haggard and Pereira (2005), Hir and Pandher (2005), Peajiso (2004), ec). Wurgler and Zhuravskaya (2002) reveal ha socks wihou close subsiues have higher arbirage risk and experience higher price jumps. Idiosyncraic risk make closer subsiues more unlikely and hence resuling in downward slope of demand curve. Likewise, Haggard and Pereira (2005) poin ou ha idiosyncraic risk prevens riskless arbirage hrough eliminaion of perfec subsiues for socks, which resuls in he downward-slope demand curve. Greaer magniude of he proxy represens seeper demand curve (less elasic demand). Idiosyncraic risk is calculaed using CAPM (marke model) and Fama and French s (1993) exended four-facor model wih Carhar s (1997) momenum facor as he fourh facor. Time series regression is performed o esimae idiosyncraic volailiy of individual sock. Fama and French (1993) hree-facor asse pricing model is considered o do a beer job han CAPM in capuring he cross-secional average reurn on US socks. However, as saed in Fama and French s (1996) paper, Fama and French (1993) hreefacor model canno explain profiabiliy of momenum sraegies or he coninuaion of shor erm reurns documened by Jegadeesh and Timan (1993). Therefore, in addiion o Fama and French s (1993) hree-facor model, momenum facor is included in his research as he fourh facor following Carhar s four-facor model (1997). In he marke model, individual sock s daily excess reurn is regressed on marke risk premium: r (1) r i imkt m, Where: = reurn of sock i in day in excess of he risk-free rae r i, r m, = i w i ri,, wih w i, = weigh of firm i a dae. I is value-weighed excess marke reurn on all NYSE, AMEX, and NASDAQ socks (from CRSP) minus he onemonh Treasury bill rae (from Ibboson Associaes). 8

9 Using he esimaed alpha and bea from he regression, he daily excess sock reurn can be prediced by subsiuing he marke premium ino he equaion. The difference beween his prediced value and acual excess sock reurn is he daily residual risk for each sock. ˆ ˆ r (2) r i imkt m, Idiosyncraic risk of he sock can be esimaed in wo ways. Firs, following he mehod used by Haggard and Pereira (2005), idiosyncraic volailiy is calculaed as sandard deviaion of daily residual risk obained from equaion (2). sd 1 N N 1 (3) sd = sandard deviaion of he residual risk = residual risk for sock i a day 1 N = N 1 N 2 = mean or average of he residual risk for sock i = number of rading days Under he second approach which has been used by many researchers (e.g. Peajiso (2004), Hir and Pandher (2005)), idiosyncraic risk volailiy is esimaed using roo mean square error. As a kind of generalized sandard deviaion, Roo Mean Squared Error (RMSE) is calculaed as following: RMSE ˆ r i, 1 N N N N yi yˆ i ri rˆ,,, N N 1 (4) = prediced reurn of sock i in day in excess of he risk-free rae = residual risk for sock i a day N = number of rading days Generally, he smaller he RMSE, he beer he performance of he model will be. Esimaion of he idiosyncraic risk using Fama and French s exended four-facor model is done similarly. Firs, we regress individual sock s daily excess reurn on Fama and French s (1993) exended four-facor o esimae heir facors loadings: r r SMB HML UMD (5) i imkt m, ismb ihml iumd Where: = reurn of sock i in day in excess of he risk-free rae r i, r m, = i w i, ri, wih w i, = weigh of firm i a dae. I is value-weighed excess marke reurn on all NYSE, AMEX, and NASDAQ socks (from CRSP) minus he onemonh Treasury bill rae (from Ibboson Associaes). SMB (Small Minus Big) = he difference beween he reurns on small and large capializaion porfolios (he average reurn on he hree small porfolios minus he average reurn on he hree big porfolios). 2 9

10 HML (High Minus Low) = he difference beween he reurns on high and low book o marke porfolios (he average reurn on he wo value porfolios minus he average reurn on he wo growh porfolios). UMD (Up Minus Down) = he difference beween he reurns on winner and loser porfolios (he average reurn on he wo high prior reurn porfolios minus he average reurn on he wo low prior reurn porfolios). These facors, definiion of facors, and daily porfolio reurn are obained from Kenneh French websie (hp://mba.uck.darmouh.edu/pages/faculy/ken.french/daa_library.hml). Afer we ge he esimaion for alpha, bea, and he facor loadings for SMB, HML, and UMD, we pu he acual sock s daily excess reurn on he esimaion (he line of bes fi) o find he residual risk for each day. Subsequenly, idiosyncraic risk of sock reurn is esimaed similarly using sandard deviaion of daily residual risk (equaion (3)) and roo mean squared error of a regression of daily excess reurns on he Fama and French s exended four-facor (equaion (4)). I should be noed ha hin rading or liquidiy problem can lead o a bias in he regression resuls. To couner his problem, idiosyncraic risk measures are se o missing if for a sock he number of days when rading ake place are less han 50 days. Empirical Model Slope of Demand Curve and Leverage Raio To es he firs hypohesis ha firms wih seeper demand curve end o have higher leverage, he following regression is performed: LEV 1 C 1DCSlope (6) TDM 1 C 1DCSlope (6a) TDB 1 C 1DCSlope (6b) LEV, he leverage raios have wo variaions; marke leverage (TDM) and book leverage (TDB). We run he regression for boh as here is no agreemen among researchers over which measuremen is he bes one. DCSlope is he slope of demand curve for sock. C denoes pre-deermined conrol variables (will be discussed shorly), lagged one period. Afer he necessary facors correlaed wih cross-secional differences in leverage are conrolled, he impac of downward sloping demand curve for socks on firm s financing decisions can be observed by running he regression model above. 1, he coefficien of DC Slope, is prediced o be posiive. To invesigae he relaionship beween he slope of demand curve for sock and leverage, we need o conrol for oher variables which affec he firm s leverage. Frank and Goyal (2004) show seven core facors which are reliably imporan for predicing leverage decisions of publicly raded US firms from 1950 o Their sudy provides useful inpu for pre-deermined conrol variable in his research paper. Conrol variables suggesed by Frank and Goyal (2004) are mainly used because in he process of obaining hose facors, hey have ake ino accoun he parsimony consideraion and conrol for mulicollineariy using Bayesian Informaion Crierion (BIC) mehod. This research also includes few addiional facors used by Chang e al. (2006). Alhough here is conformiy among researchers abou which facors influence capial srucure, here is less agreemen in he inerpreaions of he effecs. This paper provides 10

11 commonly acceped inerpreaions of differen conrol variables. Alernaive inerpreaions are possible. Facors which serve as conrol variables in his research are summarised below. Firm Size. Trade-off heory predics ha larger firms should have higher leverage because generally hey have lower defaul risk. In line wih ha, Harris and Raviv (1991) documened ha leverage is posiively relaed o firm size. Therefore, log of he book value of asses as a proxy for firm size is included as one of conrol variables. Median Indusry Leverage. Frank and Goyal (2004) show ha indusry leverage is major deerminans of corporae leverage. Firms in high leverage indusry, measured using median indusry leverage, end o have high leverage. In accordance wih ha, Hovakimian, Opler and Timan (2001) find ha firms adjus heir deb raios owards indusry median deb raios. Hence, median indusry leverage is included in he conrol variables. Tangibiliy. From rade-off and agency heory perspecive, firms ha have more angible asses end o have more leverage because angible asses could serve as collaeral. Indeed, empirical sudies suppor his predicion. Research and developmen expense o sales raio. Research and developmen expense scaled by sales can proxy for a variey of firm characerisics for insance growh poenial or uniqueness of he produc (Timan, 1984). Hence, i is included as one of he conrol variables. This paper also includes research and developmen dummy variable which equals o one if research and developmen expense is missing and zero oherwise. Firm age or mauriy of he firm. Maure firms ypically have more repuaion in deb markes and hence face lower agency coss of deb. For his reason, rade-off heory predics posiive relaion beween firm s age and leverage. To accoun for his, he age of he firm is included as a conrol variable. Marke o book raio. Marke o book raio is included o conrol for growh opporuniies. Empirical sudies documen ha high growh firms end o have less leverage. According o agency heory, his is possibly because growh firms end o avoid incenives disorion problems. Profiabiliy. Frank and Goyal (2004) documened ha firms wih more profis end o have less leverage. This can be explained using pecking order heory of capial srucure which saes ha profiable firms borrow less because hey have more inernal source of funds. Dividends. Frank and Goyal (2004) find ha firms which pay dividends end o have less leverage han non-payers. Their finding can be explained in a dynamic rade-off framework bu canno be explained by pecking order heory because pecking order heory predics he conrary. Frank and Goyal (2004) propose wo inerpreaions o accoun for negaive relaion beween leverage and dividends. Firs, dividend-paying firms have lower agency coss of equiy and hence allow firms o issue more equiy. Consequenly, hey have less leverage. Second, dividend-paying firms migh be hose ha generae more cash from operaions relaive o invesmen opporuniies so hey pay ou he difference. Such firms would be unlikely o raise more deb since ha would incur he unnecessary ransacions coss herefore dividend paying firms would have less leverage. Share urnover. Share urnover is included in he conrol variables o proxy for liquidiy. Illiquidiy of firms sock prevens hem o raise equiy. They will prefer deb financing insead hence hey will end o have higher leverage. 11

12 Sock reurn. Consisen wih marke iming heory and adverse selecion argumens, firms are more likely o issue equiy when heir sock performance has been good as firms are less likely o be undervalued during such periods. Therefore, o conrol for pas sock performance, cumulaive sock reurns is included in his research. Earning volailiy and Alman s unleveraged z-score. According o rade-off heory, firms reac o risk by reducing leverage. Hence, earning volailiy and Alman s unleveraged z-score are included o conrol for he risk faced by he firm (Chang e al. (2006)). Expeced inflaion. Firms end o have high leverage when expeced inflaion is high. According o Taggar (1985) his may reflec feaures in he ax code ha favour deb when inflaion is expeced. As noed by Frank and Goyal (2004), i migh also reflec effors by managers o ime he marke. Slope of Demand Curve and Targe Adjusmen Model Targe adjusmen model is performed o es wheher firms wih seeper demand curve have higher arge deb raio (he second hypohesis). In perfec capial marke, firms always mainain heir arge leverage raio. However, adjusmen coss may preven immediae adjusmen o a firm s arge. The sandard arge adjusmen model allows parial adjusmen of he firm s iniial leverage raio oward is arge wihin each ime period (Flannery and Rangan, 2004). The sandard arge adjusmen model is esimaed as follow: * LEV LEVi, 1 1 LEVi, LEVi, 1 (7) * TDM TDMi, 1 1 TDM TDMi, 1 (7a) where * TDB TDBi, 1 i TDB TDBi, 1 1, (7b) LEV i, is leverage raio a ime for firm i and 1 LEV is leverage raio lagged one period. Final synhesized rade-off suggess ha firms have a arge deb raio which hey wan o achieve hence firms ha are above a arge deb raio should issue equiy and firms ha are below a arge deb raio should issue deb. Therefore, we include deviaion of lagged one-period deb raio from he esimaed arge deb raio in he regression. LEV denoes he arge leverage raio for firm i a ime, which can be expressed as a * funcion of demand curve slope and a se of predeermined (lagged one period) conrol variables (C): * LEV C DCSlope (8) TDM TDB C DCSlope (8a) * C DCSlope (8b) * By subsiuing equaion (8) ino (7), we reduce he arge adjusmen model o: LEV 3 ( 1 ) LEV 1 C 1 DCSlope 1 (9) TDM 3 ( 1 ) TDM 1 C 1 DCSlope 1 (9a) TDB 3 ( 1 ) TDB 1 C 1 DCSlope 1 (9b) where and. is speed of adjusmen of he firm o he arge deb raio. Using he arge adjusmen regression model in equaion (9), i is prediced ha is posiive. 12

13 Regression Specificaions This research uses Ordinary Leas Square (OLS) regression which assumes ha errors have zero mean, consan variance (homoscedasiciy), are uncorrelaed wih each oher and normally disribued. To address he concern ha he error erms are likely o violae OLS assumpions of homoscedasiciy and cross-secional independence, we also esimae parameers and sandard errors based on he Fama and MacBeh (1973) approach. In oher words, we esimae regression equaions each year, and conduc saisical ess on he ime-series means of he esimaed coefficiens. The resuls from Fama-MacBeh regression are similar o he ones from OLS regression herefore hey are no repored for he purpose of conciseness. Sudy by Flannery and Rangan (2004) confirm his since hey also find ha OLS yields similar coefficien esimaes for he same specificaions. Therefore, we can infer ha regressions resuls in his research are reliable and do no suffer from a serious bias. DATA AND ESTIMATION PROCEDURES Sample The sample is consruced from all firms ha exis in Compusa any ime beween 1971 and Following he common pracice, non-indusrial firms are excluded from he sample, specifically financial (SIC ) and regulaed (SIC ) firms. The reason of his exclusion is because he capial decisions of financial and regulaed firms may reflec special facors (Flannery and Rangan, 2004). Firms wih missing book value of asses and firms which have less han wo consecuive years of daa are also excluded because lagged one period daa is needed for mos of he variables in he regression. Exclusion of non-indusrial firms and firms which have less han wo years consecuive daa resul in complee informaion for 99,397 firm-year observaions (7,785 firms). When firms wih missing book value of asses are also excluded (o run book leverage regressions), i leads o 98,715 firm-year observaions. Variey of Compusa and CRSP variables are employed. Financial saemen daa (deb, oal asses, EBITDA, research and developmen expenses, ec.) are obained from Compusa. Sock price and sock reurn are obained from he Cener for Research in Securiy Prices (CRSP). Adjusmens are made, when he daa is merged, o ensure ha hey mached each oher correcly. For example, Compusa repors he number of shares in million while i is repored in housand in CRSP. Hence, CRSP number of shares mus be divided by 1,000 o arrive a he same measuremen wih Compusa. CRSP repors monhly daa while Compusa repors annually herefore CRSP daa need o be adjused on yearly basis when hey are merged. Ouliers or exreme observaions can lead o misleading conclusions hence hey canno be ignored. To deal wih his problem, we use winsorizaion under which he mos exreme ails of he disribuion are replaced by he mos exreme value ha has no been removed. Table 1 repors he summary saisics of variables which we use and he definiion of each variable. All of hese variables are winsorized a 0.5% of boh sides of he disribuion o miigae he impac of exreme observaions and mis-recorded daa. 13

14 Variables and Variables Definiion The dependen variable is leverage, measured using marke leverage (TDM) and book leverage (TDB). Marke leverage is raio of oal deb divided by marke value of asses while book leverage is raio of oal deb divided by book value of asses. The independen variables are slope of demand curve for sock and conrol variables. All explanaory variables are lagged one period. Proxy for he slope of demand curve for sock is idiosyncraic risk of sock reurns which is measured using RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF. RMSE-CAPM is roo mean square error of residuals from a regression of daily excess reurns on he marke model. RMSE-FF is roo mean square error of residuals from a regression of daily excess reurns on he Fama and French (1993) exended four-facor. RSD-CAPM is sandard deviaion of he residuals from a regression of daily excess reurns on he marke model. RSD-FF is sandard deviaion of he residuals from a regression of daily excess reurns on he Fama and French (1993) exended four-facor. Conrol variables are size, age, angibiliy, indusry median leverage, profiabiliy, urnover, sock reurn, marke o book raio, research and developmen o sales raio, research and developmen dummy, z-score, dividend o asses raio, and earning s variabiliy. Firm size (LNBVAD1) is log of book value asses deflaed by Consumer Price Index o accoun for inflaion. Firm age (LNAGE1) is log of he firm s age or number of years since he Iniial Public Offering (IPO) year. Tangibiliy (TANG1) is he ne propery, plan, and equipmens o asse raio. Indusry median leverage (MDLEV1) is he median of he raio of oal deb o he marke value of asses by indusry and by year. Reurn on Asses (ROA1), a profiabiliy measuremen, is he operaing income before depreciaion and amorizaion divided by oal asses. Share urnover (MDTNOVER1), proxy for firm s sock liquidiy, is median value of monhly shares raded (volume) divided by shares ousanding over a welve monh period. Sock Reurn (STKRTN1) is compounded annual sock reurn over a welve-monh period, obained from CRSP daase. Marke o Book Raio (MBA1), well-known as proxy for growh opporuniies, is marke value of asses divided by book value of asses. Research and Developmen o Sales (RNDSLS1) is research and developmen expenses divided by ne sales. Research and Developmen Dummy (RNDD1) is dummy variable which ake value of one if firm does no have research and developmen expenses, zero oherwise. Alman Un-leveraged Z-score (ZSCORE1) is equals o [(3.3*preax income + sales + 1.4*reained earnings + 1.2*(curren asses curren liabiliies))/oal asses]. Dividend o Asses Raio (DIVAST1) is common socks dividends divided by oal asses. Earning s Variabiliy (EBITSD) is hisorical sandard deviaion of he raio of EBITDA o oal asses. Summary Saisics Table 1 displays descripive saisics of employed variables in overall sample. All of hese variables are winsorized a 0.5% of boh side of he disribuion o avoid he influence of exreme observaions and mis-recorded daa. The mean, median, and sandard deviaion of each variable are repored in he able. On average, marke leverage raio in overall sample is 20% while book leverage raio is 24%. The median of marke leverage raio and book leverage raio are 16% and 22% respecively. Idiosyncraic volailiy (proxy for he slope of demand curve for sock) as measured using RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF has mean, median, and sandard deviaion of 4%, 3%, and 3% respecively. When four decimal poins are used 14

15 insead of wo decimal poins, we can see he sligh difference of idiosyncraic volailiy esimaion using hose various mehods i.e. RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF. However, he difference is rivial. Average age of firms lised in Compusa beween 1971 and 2004 is years. The minimum number of years per firm is 2, he maximum is 34, and he median is 14. Table 1: Summary Saisics Variable Mean Median Sandard Deviaion Marke Leverage (TDM) Book Leverage (TDB) Firm Size (BVAD) Age (years) Tangibiliy (TANG) Indusry Median Leverage (MDLEV) Reurn On Asses (ROA) Share Turnover (MDTNOVER) Sock Reurn (STKRTN) Marke o Book Raio (MBA) R&D o Sales Raio (RNDSLS) R&D Dummy (RNDD) Un-leveraged Z-score (ZSCORE) Dividend o Asses (DIVAST) Earning Variabiliy (EBITSD) RMSE-CAPM RMSE-FF RSD-CAPM RSD-FF Table 2 repors he correlaion coefficiens among several key variables of ineres. From pairwise correlaion coefficiens among various proxies for he slope of demand curves for sock, i can be inferred ha various measuremen (RMSE-CAPM, RMSE- FF, RSD-CAPM, and RSD-FF) o esimae idiosyncraic volailiy do no differ much and heir correlaion coefficiens are close o one. This suggess ha hose various measuremens are consisen wih each oher in esimaing idiosyncraic risk of individual sock. 15

16 Table 2: Correlaion Coefficien beween Key Variables Correlaion TDM TDB BVAD TANG MDLEV ROA MDTNOVER MBA RMSE-CAPM RMSE-FF RSD-CAPM RSD-FF TDM TDB *** BVAD *** *** TANG *** *** *** MDLEV *** *** *** *** ROA ** *** *** *** *** MDTNOVER *** *** *** *** *** *** MBA *** *** *** *** *** *** *** RMSE-CAPM *** *** *** *** *** *** *** *** RMSE-FF *** *** *** *** *** *** *** *** *** RSD-CAPM *** *** *** *** *** *** *** *** *** *** RSD-FF *** *** *** *** *** *** *** *** *** *** ***

17 Table 2 also displays he correlaion coefficiens beween slope of demand curves and leverage. This univariae analysis shows ha he slope of demand curve for sock (as shown by RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF) is posiively relaed o he leverage raio. Greaer magniude of RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF represen seeper demand curve for a firm s sock. This preliminary evidence suppors he idea ha firms wih seeper demand curve end o have higher leverage raio. Pairwise correlaion coefficiens beween conrol variables and leverage are repored in able 2 as well. Mos correlaion coefficiens have expeced signs. For insance, size and leverage are posiively correlaed. Tangibiliy and leverage are posiively correlaed suggesing ha firms which have more angible asses end o have more leverage. The low correlaions among independen variables indicae ha one can include hese variables in he same regressions wihou concerning abou mulicollineariy. RESULTS AND ANALYSIS Slope of Demand Curve for Sock and Acual Leverage Raio Acual Marke Leverage Table 3 presens he regression resuls of acual marke leverage on he slope of demand curve for sock. Resuls show ha here is posiive and saisically significan relaionship beween he slope of demand curve for sock and acual marke leverage and are consisen across various measuremen mehods of he slope of demand curve for sock (RMSE- CAPM, RMSE-FF, RSD-CAPM, and RSD-FF). Even afer conrolling for oher facors shown in prior sudies o influence leverage, he coefficien of he slope of demand curve for firm s sock on firm s acual marke leverage is posiive and saisically significan a 1% level demonsraing ha firms wih seeper slope of demand curves end o have higher acual marke leverage. This resul also implies ha he slope of demand curves for sock is an addiional deerminan facor in predicing firm s marke leverage. Mos conrol variables are significan and have he expeced signs. For insance, firm size, angibiliy, and indusry median leverage are posiively relaed o leverage. This is consisen wih prior lieraures (e.g. Harris and Raviv (1991), Frank and Goyal (2004)). Harris and Raviv (1991) documen ha larger firms have higher leverage. I is well known ha firms wih more angible asses end o have higher leverage as hey can pledge heir asses in suppor of deb. Frank and Goyal (2004) find ha indusry leverage is prominen deerminan of firm s leverage, which is exacly wha is found in his research. They provide explanaion ha firms in high leverage indusry have higher leverage since firms in he same indusry mus face many common facors. 17

18 Table 3: Acual Marke Leverage and he Slope of Demand Curve (1) (2) (3) (4) LEVERAGE (TDM) RMSE-CAPM RMSE-FF RSD-CAPM RSD-FF DCSlope 1.227*** 1.227*** 1.230*** 1.240*** (16.9) (16.8) (16.9) (16.8) LNBVAD *** 0.022*** 0.021*** 0.022*** (23.2) (23.2) (23.2) (23.2) LNAGE *** *** *** *** (-2.9) (-2.9) (-2.9) (-2.9) TANG *** 0.081*** 0.081*** 0.081*** (11.9) (11.9) (11.9) (11.9) MDLEV *** 0.511*** 0.511*** 0.511*** (45.2) (45.2) (45.2) (45.2) ROA (0.7) (0.7) (0.7) (0.7) MDTNOVER *** *** *** *** (-17.2) (-17.1) (-17.2) (-17.1) STKRTN *** *** *** *** (-15.8) (-15.7) (-15.8) (-15.7) MBA *** *** *** *** (-22.6) (-22.5) (-22.6) (-22.5) RNDSLS *** *** *** *** (-10.7) (-10.7) (-10.7) (-10.7) RNDD *** 0.016*** 0.016*** 0.016*** (6.3) (6.2) (6.3) (6.2) ZSCORE *** *** *** *** (-7.5) (-7.5) (-7.5) (-7.5) DIVAST *** *** *** *** (-21.8) (-21.8) (-21.8) (-21.8) EBITSD *** *** *** *** (-12.6) (-12.6) (-12.6) (-12.6) Consan (1.4) (1.4) (1.4) (1.3) Observaions R-Squared Ohers significan facors, namely mauriy of firms, sock s liquidiy, sock reurn, marke o book raio, R&D o sales raio, Alman un-leveraged z-score, dividend o asses, and earning variabiliy are negaively relaed o firm s acual leverage raio. Consisen wih prior sudies and agency heory, marke o book raio (as proxy for growh opporuniies) is found negaively relaed o firm s leverage possibly because growh firms wan o avoid deb overhang problems. In line wih ha, R&D o sales raio is negaively relaed o firm s leverage raio. This research also found ha sock reurn, o conrol for pas sock performance, is negaively relaed o leverage. This resul is consisen wih he marke iming heory which saes ha firms are more likely o issue equiy hence having lower deb proporion when heir sock performance has been good. In accordance wih Frank and Goyal (2004), his paper finds ha dividend paying firms end o have lower leverage raio. Conrary o he view ha maure firms ypically have higher leverage because hey have more repuaion in deb markes hence can borrow a lower cos of deb, his research find ha maure firms end o borrow less. This probably because maure firms have more cash 18

19 flow in he firs place hus no only need less ouside financing, bu also end o pay-off deb insead of borrows more. Firms wih more liquid socks end o have less leverage. The liquidiy of firm s socks suggess ha firms can raise equiy more easily and cheaply in he marke because markes can absorb firm s socks in he even of equiy issuance. The more liquid firms socks, he more likely firms will issue equiy hence he lower heir leverage will be. Finally, as suggesed by rade-off heory, firms reac o risk by reducing leverage. In line wih his view, his research paper finds ha earnings variabiliy (proxy for risk) is negaively relaed wih leverage. Acual Book Leverage The regression of acual book leverage on various proxies of he slope of demand curve for sock is repored in able 4. When book leverage is used insead of marke leverage, he same resuls are found. The slope of demand curves for sock is a posiive and saisically significan facor in predicing book leverage of he firm, even hough oher variables shown o affec leverage had been conrolled. Across various proxies of he slope of demand curve for sock (RMSE-CAPM, RMSE-FF, RSD-CAPM, and RSD-FF), he coefficien of he slope of demand curve for sock o he acual book leverage is posiive and significan a 1% level suggesing ha firms wih seeper slope of demand curves have higher acual book leverage. Oher facors i.e. he conrol variables have same signs and magniude as in marke leverage regression. The only remarkable difference is ha reurn on asses (ROA) as one of he profiabiliy measures is a posiive and saisically significan facor o explain book leverage while i is insignifican for marke leverage. This is possibly because ROA as an accouning raio, is based on book value and more backward looking which are feaures also shared by book leverage. 19

20 Table 4: Acual Book Leverage and he Slope of Demand Curve (1) (2) (3) (4) LEVERAGE (TDB) RMSE-CAPM RMSE-FF RSD-CAPM RSD-FF DCSlope 0.990*** 0.994*** 0.992*** 1.004*** (14.3) (14.2) (14.3) (14.2) LNBVAD *** 0.022*** 0.022*** 0.022*** (21.8) (21.9) (21.8) (21.9) LNAGE *** *** *** *** (-4.8) (-4.8) (-4.8) (-4.8) TANG *** 0.102*** 0.102*** 0.102*** (13.1) (13.1) (13.1) (13.1) MDLEV *** 0.411*** 0.411*** 0.411*** (33.9) (33.9) (33.9) (33.9) ROA *** 0.047*** 0.047*** 0.047*** (5.2) (5.2) (5.2) (5.2) MDTNOVER *** *** *** *** (-11.3) (-11.2) (-11.3) (-11.2) STKRTN *** *** *** *** (-4.0) (-3.9) (-4.0) (-3.9) MBA *** *** *** *** (-10.7) (-10.6) (-10.7) (-10.6) RNDSLS *** *** *** *** (-8.9) (-8.9) (-8.9) (-8.9) RNDD *** 0.024*** 0.024*** 0.024*** (8.5) (8.5) (8.5) (8.5) ZSCORE *** *** *** *** (-12.2) (-12.2) (-12.2) (-12.2) DIVAST *** *** *** *** (-19.7) (-19.7) (-19.7) (-19.7) EBITSD *** *** *** *** (-8.2) (-8.2) (-8.2) (-8.2) Consan 0.060*** 0.060*** 0.060*** 0.060*** (7.9) (7.9) (7.9) (7.8) Observaions R-Squared To sum up, he resuls from leverage regression show ha slope of demand curve for sock has posiive and significan effec on he firm s acual leverage raio (measured using eiher marke or book leverage), even afer conrolling for oher facors which have been shown in pas lieraures o influence firm s leverage. The resuls confirm he firs hypohesis ha firms wih seeper demand curve for heir sock suffer more from he share price drop if hey issue equiy hence are less likely o issue equiy. Consequenly, we will observe higher acual leverage raio for hese firms, which is exacly wha is found from he empirical sudies in his research. Slope of Demand Curve for Sock and Targe Leverage Raio Targe Marke Leverage Table 5 repors arge adjusmen regression resuls when marke leverage is used. The resuls show ha he slope of demand curves for socks is a posiive and saisically 20