Risk Changes and External Financing Activities: Tests of the Dynamic Trade-off Theory of Capital Structure *

Transcription

1 Risk Changes and Exernal Financing Aciviies: Tess of he Dynamic Trade-off Theory of Capial Srucure * Marin J. Dierker Korea Advanced Insiue of Science and Technology (KAIST) dierkerm@business.kais.ac.kr Jun-Koo Kang Nanyang Technological Universiy of Singapore jkkang@nu.edu.sg Inmoo Lee Korea Advanced Insiue of Science and Technology (KAIST) inmool@business.kais.ac.kr Sung Won Seo Ajou Universiy seosw@ajou.ac.kr June 2017 * Phone numbers are (Dierker), (Kang), (Lee), and (Seo). We are graeful for valuable commens from Tim Loughran, Sheridan Timan, and seminar paricipans a KAIST, he 2012 Allied Korean Finance Associaion Meeings, and he 2013 Annual Conference on Asia-Pacific Financial Markes. We also hank Byoung Hyun Jeon for his excellen research assisance. All errors are our own.

2 Risk Changes and Exernal Financing Aciviies: Tess of he Dynamic Trade-off Theory of Capial Srucure Absrac We provide new insigh ino he relevance of he dynamic rade-off heory of capial srucure by examining firms exernal financing aciviies following risk changes. Consisen wih he predicion of he dynamic rade-off heory bu inconsisen wih he pecking order heory and he marke iming explanaion, we find ha firms issue equiy (deb) following risk increases (decreases). The resuls hold for subsamples of financially unconsrained firms and are robus o a variey of risk measures including sock reurn volailiy, defaul probabiliy, implied asse volailiy, and adjused Ohlson (1980) scores. Keywords: Capial srucure heory, Dynamic rade-off, Pecking-order, Marke iming, Risk change, Exernal financing aciviies JEL Classificaion: G32, G33, G35 3

3 I. Inroducion There has been much debae abou he relaive imporance and relevance of various capial srucure heories such as heories based on he rade-off beween he ax benefis of deb and he expeced coss of bankrupcy (Kraus and Lizenberger (1973), Miller (1977)), adverse selecion coss (Myers and Majluf (1984)), and marke iming (Baker and Wurgler (2002)). In paricular, in response o he debae on he relevance of he saic rade-off heory, 1 academics have urned o dynamic versions of he rade-off heory. Fischer, Heinkel, and Zechner (1989), for insance, show ha when capial srucure adjusmens are cosly, firms ake recapializaion acions only when he benefis of recapializaion ouweigh is coss. 2 As emphasized by several sudies, such adjusmen coss are nonrivial and can impose a serious challenge in esing he dynamic radeoff heory of capial srucure due o numerous assumpions required in he measuremen of adjusmen coss. 3 In his sudy, we evaluae he imporance of he dynamic rade-off heory relaive o wo oher capial srucure heories (pecking order and marke iming heories) by examining firms exernal financing decisions following changes in one of he key deerminans of he rade-off heory, risk. In spie of srong evidence on significan changes in firm risk over ime (e.g., Campbell e al. 1 For example, Myers (1993) and Graham (2000) poin ou ha a negaive correlaion beween profiabiliy and leverage raios is he mos criical evidence agains he saic rade-off heory, while Andrade and Kaplan (1998) argue ha, from an ex-ane perspecive, expeced financial disress coss are likely o be small in comparison o he ax benefis of deb. 2 In he absence of adjusmen coss, he rade-off heory suggess a posiive relaion beween profiabiliy and leverage raios, as profiable firms are more likely o uilize bigger deb ax shields. However, firms facing high adjusmen coss may find i opimal o remain inacive in he exernal financial marke. Hennessy and Whied (2005) and Srebulaev (2007) demonsrae ha adjusmen coss in a dynamic rade-off heory can explain he observed negaive relaion beween marke leverage raios and profiabiliy and oher empirical challenges. Welch (2004), however, argues ha sock reurns, no arge leverage, drive capial srucures. Leary and Robers (2005) find evidence on he imporance of adjusmen coss bu conclude ha furher work is needed o disinguish beween he predicions of he dynamic rade-off heory and hose of a pecking order heory modified for bankrupcy risk. 3 For example, while Srebulaev (2007) calibraes a muliude of parameers o generae meaningful cross-secional variaion, he sill assumes ha cerain parameers are equal for all firms. As an example, he assumes ha he presen values of ne payous and book asses are scaled o an idenical value for all firms a he iniial dae. 1

4 (2001), Ang e al. (2006), Adrian and Rosenberg (2008)) and he heoreical link beween risk changes and capial srucure, empirical evidence on he effecs of risk changes on a firm s capial srucure decision is scarce. 4 We fill his gap in he lieraure by using a firm s exernal financing aciviies as a seing for our sudy. Focusing on exernal financing aciviies has several advanages. Firs, as we discuss laer, i allows us o es he imporance and relevance of he hree capial srucure heories we consider in our sudy because hey yield quie differen predicions abou he opimal financing mehod when firms ener exernal capial markes following risk changes. Second, i allows us o examine he dynamic rade-off heory of capial srucure wihou measuring adjusmen coss since firms ha engage in exernal financing aciviies have already incurred adjusmen coss. This avoidance of adjusmen cos measuremen grealy reduces he challenge in esing he dynamic rade-off heory of capial srucure ha previous sudies have faced. Third, unlike many prior sudies ha examine leverage decisions over ime, our approach does no require he esimaion of a arge leverage raio, which is hard o define and difficul o measure. Insead of esimaing arge leverage raios, our approach uilizes he fac ha, according o he dynamic rade-off heory, an increase (decrease) in firm risk lowers (raises) is arge leverage raio, holding everyhing else consan. Thus, if he firm decides o ener exernal capial markes o raise capial afer experiencing risk changes, i is likely o choose a financing mehod ha moves i owards a lower (higher) arge 4 Leland (1994) heoreically shows ha increases in risk reduce deb capaciy, while Chen (2010) poins ou ha counercyclical variaion in risk premiums, defaul probabiliies, and defaul losses increases he presen value of expeced defaul losses, leading o lower opimal leverage raios. Gormley, Masa, and Milbourn (2012) empirically show ha leverage is relaed o he change in firm liigaion risk for a small se of firms. Numerous sudies also show ha leverage decreases wih asse or reurn volailiy (e.g., Harris and Raviv (1991), Ju, Parrino, Poeshman, and Weisbach (2005)). However, hese sudies do no explicily examine how firms respond o changes in risk and change heir capial srucures accordingly. 2

5 leverage raio. 5 In performing our ess, we address poenial problems arising from persisen unobservable facors such as indirec bankrupcy coss by including firm fixed effecs. Alhough previous sudies also examine a firm s capial srucure decision around is exernal capial raising period, 6 our approach differs from hese sudies in ha we focus on he choice of a firm s exernal financing mehod following a change in risk. Under he dynamic rade-off heory, a firm s capial srucure decision in response o a change in risk depends no only on is risk level, bu also on he (hard o observe) coss of raising exernal capial. When risk changes, he firm is likely o raise exernal capial o move oward is opimal leverage as far as he benefis from such a capial-raising aciviy is sufficienly greaer han is coss. 7 If an increase (decrease) in risk lowers (raises) he arge leverage raio and a firm decides o ener exernal capial markes o raise addiional exernal capial, he dynamic rade-off heory predics ha he firm is more likely o issue equiy (deb) or buy back deb (equiy) following an increase (decrease) in risk. Therefore, risk increases (decreases) are expeced o be associaed wih leverage-decreasing (increasing) aciviies. 8 5 Ineresingly, many previous sudies ha esimae arge raios do no explicily model risk as an explanaory variable. I is also imporan o noe ha, as shown by Chen and Zhao (2007) and Chang and Dasgupa (2009), he resuls in prior sudies ha firms pursue a arge leverage raio do no necessarily imply evidence in favor of he dynamic radeoff heory. Alhough he esimaion of arge leverage raios is no required in our analysis, we neverheless include esimaed arge leverage raios as a conrol in he regressions o make our resuls comparable o hose in prior sudies. 6 For example, Hovakimian, Hovakimian, and Tehranian (2004) focus on he period during which firms issue boh deb and equiy. Danis, Ral, and Whied (2014) pay close aenion o he case when firms simulaneously issue a large amoun of deb and pay ou a large amoun of inernal capial hrough cash dividends or share repurchases, whereas Koreweg and Srebulaev (2013) examine cases of refinancing in which firms ne deb (equiy) issuance is greaer han 5% of he book value of asses. 7 Cosly adjusmen of is capial srucure represens a real opion o a firm, and he value and opimal exercise iming of his opion also depend on he firm s risk level. Only if he benefis from raising (reducing) exernal capial are sufficienly large, he firm is expeced o raise (reire) a ype of exernal capial ha allows i o move closer o is opimal leverage. 8 Previous sudies focus mainly on adjusmen coss associaed wih raising exernal capial. However, since forgoing he opporuniy o inves in good projecs ha unexpecedly arrive afer he reducion of capial can be considered a par of he adjusmen coss, such adjusmen coss can also be significan. 3

6 In conras, he capial srucure explanaions by Myers and Majluf (1984) and Baker and Wurgler (2002) do no assign an equally imporan and explici role o firm risk. For example, under he pecking order heory, a firm s preference for deb over equiy does no depend on is risk level as far as he risk level does no affec he degree of he firm s informaion asymmery. An increase in risk, however, may resric he firm s abiliy o raise deb. However, a leas for a subsample of firms ha are financially unconsrained (and hus are unlikely o exhaus heir deb capaciy when risk increases), he pecking order heory predics ha hese firms will choose deb as he source of heir exernal capial whenever hey raise exernal capial regardless of he direcion of risk changes. The marke iming argumen of Baker and Wurgler (2002), which suggess ha firms are more likely o issue equiy a imes when heir valuaion (measured by he marke-o-book raio) is high, also does no explicily model firm risk and he complex inerplay beween dynamically changing risk levels and marke misvaluaion. However, heir argumen implicily suggess ha as a firm s risk increases, is equiy value ends o decrease, 9 for example, due o an increase in he cos of equiy capial, and herefore, i is less likely o issue equiy following an increase in is risk. 10 The above hree capial srucure heories also have differen predicions for firms exernal capial reducing aciviies (i.e., buyback decisions) following risk changes. While he dynamic rade-off heory predics ha, following risk increases (decreases), firms will buy back deb (equiy), he marke iming heory predics he opposie. Alhough Baker and Wurgler (2002) are 9 We empirically suppor his predicion laer in our analysis. 10 One may argue ha he change in risk affecs only a firm s fundamenal equiy value bu no is misvaluaion (i.e., difference beween observed and fundamenal values) and hus does no influence he firm s incenives o ake advanage of misvaluaion. However, given managers general view ha misvaluaion is correlaed wih pas sock reurn performance (Graham and Harvey (2001)), significan changes in prices following risk changes are likely o affec misvaluaion (or a leas managers percepion of misvaluaion of firms shares) and herefore, our predicion above is likely o hold. 4

7 no explici abou how firms buy back exernal capial in responses o risk changes, according o heir key argumen ha equiy misvaluaion drives firms financing decisions, firms are expeced o buy back securiies ha are undervalued (or relaively less overvalued). As we show laer, he marke-o-book raio ends o decrease (increase) following risk increases (decreases), all else being equal. If managers believe ha a firm s equiy is more likely o be undervalued (overvalued) following an increase (decrease) in he marke-o-book raio, he marke iming heory predics ha firms repurchase equiy following risk increases bu reire deb following risk decreases. On he oher hand, he pecking order heory predics ha firms reduce deb in response o boh increases and decreases in risk. Under he pecking order heory, as Shyam-Sunder and Myers (1999) argue, firms wih financing surpluses prefer deb reiremens over sock repurchases when hey use heir surpluses o buy back securiies, possibly o preserve heir deb capaciy or o avoid he paymen of high equiy prices when hey repurchase shares under asymmeric informaion. Thus, he pecking order heory suggess ha, irrespecive of he direcion of risk changes and wheher firms are financially consrained or no, firms prefer o reire deb when hey have o reduce exernal capial. 11 In sum, he differen predicions for he relaion beween risk changes and fuure exernal financing aciviies discussed above allow us o evaluae he imporance of he dynamic rade-off heory of capial srucure relaive o oher compeing capial srucure heories. Table 1 summarizes he preferred ype of securiy ha a firm issues (or buys back) as is risk changes under each of he hree compeing capial srucure heories. 12 To provide supporing evidence on he 11 One cavea is ha, due o asymmeric informaion, i is possible ha firms issue overvalued equiy or buy back undervalued equiy if he benefis obained from exploiing misvaluaion are greaer han he adverse selecion coss. This possibiliy predics ha firms repurchase equiy following risk increases bu reire deb following risk decreases, he predicions similar o hose of he marke iming heory. 12 Risk changes are likely o affec he widh of he arge range of leverage under he dynamic rade-off heory bu we do no explicily address his issue since our approach focuses on exernal financing aciviies ha occur afer 5

8 effec of risk changes on capial srucure under he dynamic rade-off heory, we perform a simulaion based on Srebulaev (2007). The resul in Figure 1 shows a negaive relaion beween risk changes and he opimal leverage. 13 Changes in oher deerminans of capial srucure could also be very useful o undersand a firm s opimal capial srucure decisions. We focus on risk changes since, as discussed above, risk changes play a differen role in a firm s choice of exernal financing mehods under he hree differen capial srucure heories. 14 In conras, hese predicions are unclear for oher deerminans of capial srucure such as asse angibiliy. 15 To perform our analysis, we use boh marke- and accouning-based risk measures. We use sock reurn volailiy, defaul probabiliy, and implied asse volailiy esimaes as marke-based measures of firm risk. In addiion, despie he limiaions of accouning-based measures documened by Hillegeis e al. (2004), we use an adjused Ohlson s (1980) O-score (Franzen, Rodgers, and Simin (2007)) as an alernaive measure of risk. For each of hese risk measures, we invesigae how changes in a firm s risk are associaed wih is fuure exernal financing aciviies and leverage changes. 16 We find ha all hese risk measures are highly persisen during our sample period. Since a firm would no necessarily need o reac o risk changes if hey were considering adjusmen coss and changes in he arge ranges of leverage. 13 Appendix A describes he deails on he procedures used o obain simulaion resuls in Figure Anoher reason for focusing on risk changes is ha, unlike risk levels, hey have been largely negleced in he lieraure. Moreover, unlike some oher deerminans ha are imporan in he rade-off heory, such as bankrupcy coss, marke-based measures of firm risk have he advanage of being observable a high frequency and displaying pronounced variaion over ime. 15 For example, increases in asse angibiliy predic increases in leverage under he rade-off and pecking order heories due o increased deb capaciy, while he marke iming heory does no provide a clear predicion excep ha he marke iming behavior is less likely o be observed since valuaion becomes easier. 16 To he exen ha hese risk measures are affeced by changes in leverage, conemporary changes in risk may be relaed o leverage changes. However, his is no a major concern in our paper since we examine exernal financing aciviies following risk changes insead of examining he conemporaneous relaion beween exernal financing aciviies and risk changes. 6

9 ransiory, his resul provides anoher raionale for why risk changes should be an imporan consideraion in he es of capial srucure heories. 17 Using a sample of firms lised on NYSE, Amex, or Nasdaq from 1972 o 2011, we find a significanly posiive relaion beween risk changes and leverage increasing exernal financing aciviies. Specifically, we find ha firms are more likely o issue equiy (deb) when hey raise exernal capial following risk increases (decreases). Similarly, firms are more likely o reduce exernal capial by buying back deb (equiy) following risk increases (decreases). These resuls are consisen wih he predicions of he dynamic rade-off heory bu inconsisen wih hose of he pecking order heory and he marke iming heory. In erms of economic significance, a onesandard-deviaion increase in annual changes in equiy volailiy (19%) leads o an increase in firms ne equiy issue (or a decrease in firms ne deb issue) of 0.55% of heir oal asses in he following year, which in urn leads o a decrease in marke leverage of 1.18%. Given ha he mean marke leverage for he full sample is 38.8%, his effec is economically large and significan. These resuls are more pronounced when we exend he observaion window from one o hree years afer he increase in risk and are also robus o using a variey of alernaive risk measures and conrolling for endogeneiy bias. We also find ha our resuls do no change when we limi our aenion o subsamples of firms facing fewer financial consrains, as measured by he Whied and Wu (2006) index of consrains and he size-age (SA) index proposed by Hadlock and Pierce (2010). The pecking order heory suggess ha hese firms paricularly prefer o issue deb over equiy because hey are likely o have easier access o deb markes due o heir large deb capaciy even afer risk increases. 17 Resuls are unrepored for breviy bu available upon reques. 7

10 Therefore, hese resuls furher suppor he dynamic rade-off heory bu dispue he pecking order heory. 18 Moreover, we find ha an increase in firm risk is associaed wih a fall in a firm s valuaion as measured by he marke-o-book raio. Thus, according o he marke iming heory, firms are less likely o issue equiy following risk increases since hey may perceive heir equiy no being overvalued. This resul, ogeher wih evidence ha firms are more likely o choose equiy financing following risk increases, furher suggess ha he dynamic rade-off heory explains a firm s capial srucure decision following risk changes beer han he marke iming heory. In sum, our resuls are mos consisen wih he implicaions of he dynamic rade-off heory bu inconsisen wih he resuls of recen sudies ha documen evidence agains he rade-off heory of capial srucure. 19 Our sudy conribues o he ongoing debae abou firms capial srucure decisions in several ways. Firs, we propose a simple, clear way o es he relevance of he dynamic rade-off heory relaive o he pecking order heory and he marke iming explanaion. Unlike previous sudies ha examine a firm s capial srucure decisions during is exernal capial raising period, we focus on he relaion beween risk changes and exernal financing decisions, which allows us o minimize he concern abou mismeasuring adjusmen coss and arge leverage raios. Second, our sudy emphasizes he imporance of changes in, no levels of, risk in capial srucure decisions. In spie of a srong heoreical link beween risk changes and capial srucure 18 Using Canadian daa, Dong e al. (2012) show ha firms ime heir equiy issuance when hey are no financially consrained. They furher show ha firms follow he pecking order only when heir shares are no overvalued. 19 For example, Hovakimian, Kayhan, and Timan (2012) find ha firms wih a higher likelihood of subsanial losses in bankrupcy end o choose capial srucures ha have greaer exposure o bankrupcy risk, which canno be easily reconciled wih he (saic) rade-off heory. 8

11 and empirical evidence on ime variaion in risk, previous sudies on capial srucure focus mainly on he relaion beween leverage and he level of risk. Third, our findings add o he lieraure on a firm s dynamic capial srucure choice in which he ax benefis of deb and expeced bankrupcy coss play an imporan role. While some recen sudies show ha he (saic) rade-off heory does no have significan power in explaining he cross-secional variaion of firms capial srucure choices (e.g., Hovakimian, Kayhan, and Timan (2012)), our sudy shows ha he ime-series variaion of observed firms financing choices is consisen wih he dynamic rade-off heory. This resul complemens he findings of previous sudies on he dynamic rade-off heory, which use oher approaches (e.g., srucural esimaion in Hennessy and Whied (2005), simulaion in Srebulaev (2007)) o explain persisen crosssecional paerns in firms capial srucures. The paper proceeds as follows. In Secion II, we describe our key risk measures and conrol variables and ouline he mehodology. Secion III presens our main empirical resuls. In Secion IV we furher examine he relevance of each capial srucure heory by performing addiional ess. Secion V summarizes and concludes. II. Daa and Mehodology A. Daa Our sample consiss of all NYSE, Amex, or Nasdaq firms available on boh CRSP and Compusa beween 1971 and As in Vassalou and Xing (2004), we sar our sample period in 1971 because here are insufficien deb-relaed financial daa prior o 1971 in Compusa. All he variables used in he paper are measured a fiscal year-ends. To focus on firms wih meaningful daa, we exclude firms wih a negaive book equiy value, a marke-o-book asse raio above 10, 9

12 or oal asses below US$ 10 million. We also exclude uiliy (SIC ) and financial (SIC ) firms since heir capial srucure decisions are subjec o regulaory consrains. In addiion, as in Kayhan and Timan (2007), we exclude firms wih book leverage raios above 100%. Finally, o miigae poenial problems caused by exreme ouliers, we winsorize all variables a he 1 s and 99 h perceniles in each year, as in Leary and Robers (2005) and Kale and Shahrur (2007). Since our analyses require he measuremen of changes in risk, we furher delee he firs year of our sample period. Our final sample consiss of 82,723 firm-year observaions over he period B. Risk Measures To es he imporance of firms risk changes in heir capial srucure decisions, we use various risk measures. Roll (1984) argues ha financial markes end o incorporae informaion abou firms in a imely and forward-looking manner, suggesing ha marke-based risk measures are good measures of firm risk and hus accuraely capure ime-series flucuaions in risk. Confirming his argumen, Hillegeis e al. (2004) show ha as predicors of financial disress, marke-based risk measures, such as hose obained by fiing he Meron (1974) model, significanly ouperform accouning-based risk measures. Therefore, we focus on he following hree marke-based risk measures as our key measures of firm risk: sock reurn volailiy, defaul risk, and implied asse volailiy. The laer wo are esimaed on he basis of he Meron (1974) model. 20 We also use a 20 In a dynamic capial srucure rade-off heory, he underlying source of risk comes ypically from he volailiy of a firm s asses, such as he volailiy of is unlevered asse value (Fischer, Heinkel, and Zechner (1989)), or he volailiy of cash flow generaed from is asses (Goldsein, Ju, Leland (2001), Sebulaev (2007)). Alhough equiy volailiy and he probabiliy of bankrupcy ypically play a less prominen role in developing dynamic capial srucure models, we sill use hem in our analyses since hey are easy o measure and, in realiy, can play an imporan role in a firm s capial srucure decision. For example, managers may pay a significan aenion o he probabiliy of bankrupcy when making borrowing decisions since a high level of he probabiliy of bankrupcy jeopardizes heir job securiy and relaed oher benefis. 10

13 risk measure based on financial saemens, namely, a version of Ohlson s (1980) adjused O-score (Franzen, Rodgers, and Simin (2007)), as an alernaive measure of firm risk. Firs, he volailiy of sock reurns reflecs uncerainy in he marke value of a firm s equiy. Alhough he volailiy of a firm s oal asses may provide a beer measure of is risk on heoreical grounds, we focus on equiy volailiy in measuring firm risk due o he illiquidiy of deb markes. To measure sock reurn volailiy, EquiyVol, we calculae he sandard deviaion of 52 weekly sock reurns in each fiscal year and muliply i by he square roo of 52 o annualize i. Due o he residual naure of equiy claims, he use of equiy volailiy may enail a poenial endogeneiy problem when sudying firms capial srucure using marke leverage since an increase in equiy risk refleced in he cos of equiy is likely o decrease he marke value of equiy more han he value of deb, hereby resuling in a conemporaneous increase in he leverage raio. However, his is no a major concern in our paper since we examine exernal financing aciviies following risk changes insead of he conemporaneous relaion beween exernal financing aciviies and risk changes. Furhermore, i should be noed ha his effec goes in he opposie direcion compared wih he effec prediced by he dynamic rade-off heory (i.e., firms reduce leverage when risk increases). Thus, all else being equal, his endogeneiy problem should make i harder for us o suppor he dynamic rade-off heory. Finally, we also presen resuls based on he book value of leverage, which is no affeced by his poenial endogeneiy problem. Second, defaul risk, which is effecively a measure of he probabiliy ha a firm will ener ino cosly financial disress, is measured based on he Meron (1974) model. We measure i, Meron, in a similar way as in Vassalou and Xing (2004). Third, implied asse volailiy capures he uncerainy in asse values, no equiy values, which ulimaely maer in avoiding financial disress. Anoher imporan reason o use asse volailiy as 11

14 one of our risk measures is he fundamenal role i plays in dynamic rade-off models. We compue implied asse volailiy, AsseVol, as he annualized sandard deviaions of daily changes in asse values calculaed in he process of esimaing Meron s defaul probabiliies in each year (i.e., esimaed σσ AA ). To annualize he sandard deviaion of daily changes in asse values, we muliply i by he square roo of 252, he approximae number of rading days per year. Finally, we use he adjused O-score (1980), O-Score, as our measure of accouning-based risk. This measure is esimaed following Franzen, Rodgers, and Simin (2007), who propose he adjusmen mehod for ne income, oal asses, and oal liabiliies o avoid misclassifying financially healhy R&D-inensive firms as financially disressed firms and o rea R&D in a more conservaive way. A deailed descripion on how Meron, AsseVol, and O-score are measured is provided in Appendix C. C. Dependen Variables To examine he effecs of risk changes on firms capial srucure decisions, we use hree measures as dependen variables: leverage-increasing exernal financing aciviies (LIEFA [+1]), book leverage raio, and marke leverage raio. As discussed above, he dynamic rade-off heory predics ha firms adjusing exernal capial following a risk increase (decrease) are likely o choose a financing mehod ha helps decrease (increase) heir leverage raio. To capure his exernal financial aciviy, we creae a variable, leverage-increasing exernal financing aciviies, LIEFA [+1], as one of our key dependen variables of ineres. LIEFA [+1] is compued as he scaled sum of exernal financing aciviies ha increase a leverage raio (i.e., issue of new deb and repurchase of equiy) minus hose ha reduce a leverage raio (i.e., reducion of deb and issue of 12

15 new equiy). 21 The reason for using LIEFA [+1] is as follows. If wo oherwise similar firms differ in heir adjusmen coss and invesmen opporuniies, i is possible ha one firm prefers o respond o a risk increase by issuing equiy, while he oher firm finds i more cos-effecive o buy back deb in order o reduce is leverage. LIEFA allows us o verify wheher firms respond o changes in risk in a manner consisen wih he dynamic rade-off heory, as boh of he aforemenioned firms will have negaive LIEFA. Specifically, LIEFA is measured as he raio of he difference beween ne long-erm deb issue and ne equiy issue in year +1 o lagged oal asses. The difference beween ne deb issue and ne equiy issue is calculaed as long-erm deb issuance (DLTIS) minus long-erm deb reducion (DLTR) minus sale of common and preferred socks (SSTK) plus purchase of common and preferred socks (PRSTKC). LIEFA [+2] and LIEFA [+3] are calculaed by summing LIEFAs over wo and hree years saring from year +1, respecively. 22 As alernaive dependen variables, we use he book (marke) leverage raio, measured as he raio of he book value of deb o he book (marke) value of oal asses. The marke value of oal asses is compued as oal asses (AT) minus he book value of equiy plus he marke value of equiy, and he book value of deb is compued as oal asses minus he book value of equiy. As in Kayhan and Timan (2007), he book value of equiy is esimaed as oal asses minus he sum 21 Wheher a posiive value of LIEFA indeed leads o an increase in leverage depends on boh he original level of leverage and oher facors such as reained earnings, deprecaion (book leverage), or sock reurns (marke leverage). For example, suppose ha a firm wih a deb-o-equiy raio of 10% raises 2% of exising equiy value (E) hrough equiy issuance and 1% of exising equiy value hrough deb issuance. Afer issuance, his firm s deb-o-equiy raio will increase from 10% o 10.78% (= (0.1 E E) / (E E) = 0.11/1.02), bu i will have a negaive LIEFA. This is one of several reasons why he relaion beween exernal financing decisions and leverage changes is known o be complex (Welch (2011)). Therefore, we also use changes in marke and book leverage raios as alernaive dependen variables in our analyses o check wheher he choice of exernal financing mehod is in line wih he direcion of changes in leverage raios. 22 As an alernaive way o measure leverage-increasing exernal financing aciviy over muliple years, we define LIEFA [+2] (LIEFA [+3]) as he raio of he difference beween ne long-erm deb issue and ne equiy issue during he year +1-year +2 (+3) period o oal asses in year. In unrepored resuls, we find ha he qualiaive resuls based on his alernaive measure are similar o hose repored in he paper. 13

16 of oal liabiliies (LT) and he liquidaion value of preferred sock (PSTKL) plus deferred axes, invesmen credi (TXDITC), and converible deb (DCVT). When PSTKL is no available, he redempion value (PSTKRV), or he carrying value (PSTK) if PSTKRV is no available, is used. 23 The marke value of equiy is measured a he fiscal year-end. D. Regression Specificaion and Conrol Variables To es wheher firms engage in leverage-increasing exernal financing aciviies following changes in risk, we run he following ordinary leas squares (OLS) panel regression model ha conrols for various facors ha affec a firm s capial srucure decision: LIEFA or Lev + s ( + s ) = β0 + β1 Risk + β2risk 1 + β3 MBi, + β4mbi, 1 + β5ri, + β LTA + β EBITD + β FD + β CRdef + β CRdummy, (1) 6 + β LdefB β YearD β FirmD ε 10 Our main independen variables of ineres in Eq. (1) are he change in risk, RRRRRRRR i, and lagged risk level, RRRRRRRR i 1. The oher variables are well known o explain firms capial srucure and exernal financing choices and are included as conrols. As shown in previous sudies (e.g., Loughran and Rier (1995)), a firm s financing decision may depend on is marke valuaion, which also affecs is capial srucure. To measure he marke valuaion of he firm, we esimae he marke-o-book oal asses raio, MB. Given ha changes in firm valuaion (or changes in invesmen opporuniies) can also affec firms exernal financing decisions and changes in leverage, we also conrol for changes in MB, MB Change (Baker and Wurgler (2002)) Annual Indusrial Compusa daa variable names are in parenheses. 24 In Eq. (1), for risk and MB, we use boh heir changes and lagged values. However, for LTA, FD, r and EBITD, we 14

17 Welch (2004) demonsraes ha pas sock reurns are a driver of marke leverage, which in urn may also affec financing decisions in subsequen periods. Thus, we include one-year sock reurn (r) during he fiscal year in our regression. Size can affec firms financing decisions in several ways. For example, large firms are more likely o be maure and diversified, and have beer access o capial markes. They are also less subjec o informaion asymmery, which, according o he pecking order heory, may help reduce he exen of price drops if he firm issues equiy. Finally, large firms informaion is easily available o ouside invesors and hey are subjec o fewer rading fricions, which is likely o reduce he likelihood of sock price misvaluaion in he sock marke, affecing he poenial for marke iming. Thus, o conrol for hese effecs, we include he naural logarihm of oal asses (AT), LTA, in he regressions. According o he dynamic rade-off heory, a firm s profiabiliy can be an imporan deerminan of is capial srucure since profiable firms can ake advanage of larger deb ax shields. Profiabiliy is also imporan under he pecking order heory, which suggess ha profiable firms are less likely o depend on exernal financing. Therefore, we conrol for profiabiliy, EBITD, defined as earnings before ineres, ax, and depreciaion (OIBDP) over oal asses a he beginning of he fiscal year in he regressions. How leverage changes over ime and wha ypes of capial firms choose also depend on he oal amoun of exernal financing raised. To measure he laer, similar o Frank and Goyal (2003), we define a firm s financial defici, FD, as he raio of he sum of ne equiy and long-erm deb use only heir values in year and do no include heir changes in he regression. We include changes in MB since hey are relaed o changes in boh firm valuaion (marke iming heory) and invesmen opporuniies (pecking order and dynamic rade-off heories) and hus are relevan o all hree capial srucure heories considered in he paper. However, in unrepored ess, we repea our analyses using boh levels of and changes in LTA, FD, r, EBITD, risk and MB in he regressions and find qualiaively similar resuls as hose repored in he paper. 15

18 issues o oal asses a he beginning of he year (i.e., [sale of common and preferred sock (SSTK) purchase of common and preferred sock (PRSTKC) + long-erm deb issuance (DLTIS) longerm deb reducion (DLTR)] divided by AT in year -1). Graham and Harvey (2001) and Hovakimian, Kayhan, and Timan (2009) show ha firms pay close aenion o heir arge credi raings. This finding suggess ha any gap beween arge and acual credi raings is likely o induce firms o adjus heir capial srucure in an effor o mainain heir credi raings a arge levels. As a credi raing is also closely relaed o firm risk, we need o conrol for i in our regression. We esimae a firm s arge credi raing, TRaing, by calculaing he fied value from an ordered probi regression esimaed in each year as in Hovakimian, Kayhan, and Timan (2009). The deails of his regression model and he resuls from he ordered probi regression used o esimae he arge credi raing in 2011 are presened in Appendix D. 25 We define credi raing defici (CRdef) as he difference beween TRaing and he acual credi raing and include i in he analyses. Since here are many firms ha do no have available credi raing informaion, we include a dummy variable, CRdummy, o indicae hose firms wih available credi raing informaion. Alhough our approach does no require us o measure arge leverage raios in esing he dynamic rade-off heory of capial srucure and, as discussed above, here is a debae on he exisence and measuremen of arge leverage raios, we conrol for hese raios in our analysis o faciliae comparison wih previous sudies (e.g., Hovakimian, Opler, and Timan (2001)). We esimae arge leverage raios (Tlev) using a similar mehod o ha in Kayhan and Timan (2007), which is described in Appendix E. We measure book leverage defici, LdefB, as he difference beween he arge book leverage raio and he acual book leverage raio, TlevB LevB. If a firm 25 Resuls for oher years are available upon reques. 16

19 pursues a arge leverage raio, we expec ha he firm s financing and capial srucure decisions depend on how far i is away from is arge (i.e., he leverage defici). 26 A non-zero leverage defici a ime can arise because he firm may have deviaed from is arge capial srucure in he pas or because he arge leverage raio has shifed over he las year. Thus, lagged leverage defici and changes in arge leverage ha are used in he previous sudies are subsumed in a single variable here for he sake of simpliciy. Despie he imporance of risk as a deerminan of he opimal arge leverage raio in he dynamic rade-off heory, previous sudies have ofen no explicily included firm risk in heir esimaion of arge leverage raios. The focus of our sudy is no o propose a marginal improvemen in esimaing arge leverage raios, bu o es if firms raionally respond o changes in risk levels in a way which is consisen wih he dynamic rade-off heory. Thus, we follow he mehod used in previous sudies and do no include risk in esimaing arge leverage raios o make he comparison wih hese sudies easier. I should be also noed ha some of our independen variables such as EBITD and MB are also used as inpus o calculae oher conrol variables. For insance, EBITD is included in he calculaion of FD and LdefB, and MB is included in LdefB. This inclusion is consisen wih he lieraure (e.g., Kayhan and Timan (2007)) and ensures ha our resuls are no driven by a simple correlaion beween risk changes and oher variables ha are known o be imporan deerminans of capial srucure decisions. Finally, we conrol for year and firm fixed effecs by using year and firm dummy variables, YearD and FirmD, respecively. Conrolling for firm fixed effecs miigaes he endogeneiy 26 We calculae boh marke and book leverage deficis and use hem in he regression analyses. However, since he resuls using hese wo leverage defici measures are qualiaively similar, we repor only he resuls based on book leverage deficis in he paper. 17

20 concern ha our resuls are driven by omied unobservable firm characerisics. However, as shown in Peersen (2009), including firm dummy variables is effecive only if firm fixed effecs are permanen. Therefore, as an addiional cauionary reamen, we use firm-clusered sandard errors in calculaing -saisics, as suggesed by Peersen (2009). All oher variables are defined in previous secions and are also summarized in Appendix B. III. Empirical Resuls A. Summary Saisics Table 2 shows he summary saisics for our sample firms. The average oal asses and marke capializaion are $2.07 billion and $2.04 billion (adjused o 2011 purchasing power using he Consumer Price Index), respecively. The average annual sock reurns and he profiabiliy (EBITD) are 18.2% and 14.5%, respecively. The mean marke (book) leverage is 38.8% (44.2%) while he mean annual change in marke (book) leverage is 0.8% (0.7%), indicaing ha during our sample period, on average, firms have slighly increased heir leverage raios. The mean book leverage defici (LdefB) is 1.4%, suggesing ha our sample firms book leverage raios are on average abou 1% lower han heir arge leverage raios. The average LIEFA[+1] is 0.54% while he median LIEFA[+1] is -0.22%. These resuls sugges ha on average, firms issued more deb han equiy during our sample period, albei he median suggess he opposie. The average (median) financial defici, FD, is 4.4% (0.0%), indicaing ha he average (median) annual oal amoun of exernal financing is around 4% (0%) of oal asses a he beginning of each year. Since he sum of FD and LIEFA[+1] represens wice 18

21 he amoun of ne long-erm deb issuance, on average, firms raised abou 2.5% ( (4.4% %)/2) of oal asses hrough ne long-erm deb issuance per year. 27 The average (median) annual equiy volailiy, implied annual asse volailiy, defaul risk, and adjused-ohlson s score are, respecively, 53% (46%), 50% (41%), 2.7% (0.0%), and -1.6 (-1.6). The average (median) annual changes in marke-based risk measures vary from -0.56% (-0.74%) o 0.19% (0.00%) and he annual average (median) change in Ohlson s score is 0.02 (-0.01). B. Correlaion and Univariae Analyses In Table 3, we repor he correlaions among changes in leverage, exernal financing aciviy, he level of risk, and he change in risk used in our analyses. We find ha risk change variables are significanly negaively correlaed wih changes in marke leverage and LIEFA in he following year, even hough he magniudes of heir correlaion coefficiens are no high. The correlaions beween risk changes and book leverage changes in he following year, and heir significance vary across risk measures. We also find ha our risk change variables are significanly posiively correlaed wih each oher, alhough he magniudes of he correlaion coefficiens vary across he pairs of risk change measures. There are also significan posiive correlaions among he levels of our risk measures. I is noeworhy ha he correlaions beween our risk measures are ypically lower han one, suggesing ha consisen wih our previous discussion, hey capure differen aspecs of firm risk. All hree marke-based risk measures are only weakly, albei significanly, correlaed wih he O- score, an accouning-based risk measure, perhaps due o he laer s lack of imely updaes. 27 This number is a rough esimaion because FD is measured as he sum of ne equiy and ne deb issuances in year whereas LIEFA[+1] is measured as he difference beween ne deb and ne equiy issuances in year

22 Financial defici, FD, is significanly posiively correlaed wih changes in marke and book leverage raios in he following year, suggesing ha afer raising exernal capial, leverage raios end o increase in he following year. However, i is negaively correlaed wih LIEFA in he following year. We do no find any consisenly srong correlaion beween FD and risk change (or risk level) variables. MB and sock reurns are significanly posiively correlaed wih changes in marke leverage in he following year, bu hey are significanly negaively correlaed wih changes in book leverage and LIEFA in he following year. Finally, MB and sock reurns are significanly negaively correlaed wih risk change variables, consisen wih he conjecure ha holding everyhing else consan, sock prices decrease as risk increases. In unabulaed ess, we check he correlaions beween risk change (level) measures and oher firm characerisics repored in Table 3 and find ha none of he correlaions is high enough o cause mulicollineariy problems in our subsequen empirical analyses. In Table 4, we repor univariae resuls for changes in MB, LIEFA, and leverage raios for each group formed on he basis of firms annual risk changes. In each year, firms in he op 20%, he middle 60%, and he boom 20% of each risk change measure are classified as High risk change, Middle risk change, and Low risk change firms, respecively. The number of observaions in each caegory is shown in he firs row, and he average change in risk is shown in he second row. As expeced, changes in he risk of High risk change firms are significanly greaer han hose of Low risk change firms, as shown in he las column. Consisen wih he resuls in Table 3, MB decreases significanly more for High risk change firms han for Low risk change firms, indicaing ha, on average, increases in risk lower firm valuaion. 20

23 The nex hree (las hree) rows show he average LIEFA and changes in leverage raios for each group during he one-year (hree-year) period following risk changes. As shown in he las four columns, where he differences of he averages beween High risk change and Low risk change groups are presened, we find ha relaive o firms experiencing a low change in equiy volailiy (our main risk measure), firms experiencing a high change in equiy volailiy are significanly less likely o engage in exernal financing aciviies ha increase heir leverage. These resuls hold boh for he subsequen year (+1) as well as over he following hree years (+3). As a resul, he average change in marke leverage for High risk change firms over he subsequen one-year (hree-year) period is 1.19 (2.46) percenage poins lower han he average change made by Low risk change firms. This difference is significan a he 1% level. The same paern is observed when we replace equiy volailiy wih he esimaed likelihood of defaul. Wih respec o he oher risk measures, alhough we find ha he effec has consisenly he same sign, i is no always saisically significan. Overall, he resuls in Table 4 sugges ha High risk change firms engage in less leverage-increasing exernal financing aciviies han Low risk change firms, consisen wih he predicion of he dynamic rade-off heory. C. Regression of LIEFA and Changes in Leverage on Risk Change Variables Table 5 presens he resuls from panel regressions of LIEFA and changes in marke (book) leverage in he following year on changes in risk. The dependen variables are as follows: LIEFA in columns (1) hrough (4), changes in book leverage raios in columns (5) hrough (8), and changes in marke leverage raios in columns (9) hrough (12). T-saisics based on clusered sandard errors a he firm level are repored in parenheses. In general, consisen wih he predicions of he dynamic rade-off heory, we observe significanly negaive associaions beween risk changes 21

24 and boh LIEFA and changes in leverage raios in he following year. The resuls in column (1) sugges ha a one sandard deviaion increase in annual changes in equiy volailiy (19%) leads o a decrease in LIEFA of 0.55% (= ) in he following year, which is close o he average absolue value of LIEFA for he full sample (0.54%). Such an increase in risk reduces book leverage by 0.21% (= , see column (5)) and marke leverage by 1.18% (= , see column (9)) on average, boh of which are significan a he 1% level. Managers may be concerned no only abou large increases in risk bu also abou high risk levels. Supporing his view, we find ha he risk level a he beginning of he year is significanly negaively relaed o LIEFA and changes in leverage in he following year. The resuls in columns (1), (5), and (9) indicae ha a one sandard deviaion increase in he level of equiy volailiy (29%) leads a firm o decrease LIEFA by 0.81% (= ) and marke [book] leverage by 1.77% (= ) [0.41% (= )] in he subsequen year. 28 Columns (2) o (4), (6) o (8), and (10) o (12) provide he resuls for he alernaive risk measures we consider. Overall, he resuls confirm our inuiion ha pronounced increases in risk as well as high risk levels lead firms o adop exernal financing choices ha serve o decrease leverage and ha hese choices indeed resul in lower leverage raios. Turning o he conrol variables, consisen wih previous sudies, we find ha issue aciviy (FD) and leverage changes are only weakly correlaed and ofen go in opposie direcions (Welch (2011)). In addiion, we find ha, as poined ou by Chen and Zhao (2007) and Chang and Dasgupa (2009), changes in he leverage raios of firms wih high or low leverage raios do no necessarily 28 Ineresingly, he coefficiens on lagged risk levels in columns (1), (2) and (9) are very similar o hose on risk changes. There is no heoreical reason for his o be he case, and indeed, unrepored resuls show ha his is no longer rue when we analyze LIEFA or leverage changes over wo- and hree-year windows following a risk change. 22

25 mach wih heir financing choices: for example, a firm wih 10% leverage needs o issue a leas nine imes more equiy han deb in order o reduce is leverage raio. I should be also noed ha he dependen variables used in our sudy (i.e., changes in fuure leverage) are differen from hose in mos previous sudies (i.e., levels of curren or fuure leverage raios), and his could lead o differences in resuls beween he sudies. For example, consisen wih Hovakiminan, Opler and Timan (2001), we find ha alhough profiabiliy is negaively associaed wih changes in book leverage, i is posiively associaed wih LIEFA. We also find ha he change in he marke-o-book raio is posiively associaed wih changes in marke leverage bu negaively associaed wih LIEFA. In addiion, he coefficiens on size are significanly negaive for changes in book leverage bu significanly posiive for boh LIEFA and changes in marke leverage. Alhough he coefficiens on he conrol variables do no always have consisen signs in explaining changes in book leverage, changes in marke leverage, and LIEFA, we find consisen resuls across differen dependen variables: The coefficiens on risk changes are significanly negaive for LIEFA and leverage changes, indicaing ha firms experiencing risk increases end o engage in fewer leverage-increasing financing aciviies in he following year. In Table 6, we examine wheher he relaion beween risk changes and leverage changes are consisen across posiive and negaive risk changes. We replace he risk change variables used in Table 5 wih he maximum (minimum) of risk change and zero for posiive (negaive) risk changes (i.e., posiive risk changes = max (risk change, 0) and negaive risk changes = min (risk change, 0)). We use his approach since firms wih posiive and negaive risk changes may face differen levels of difficuly in raising capial; for example, while firms ha experience a decrease in risk may find i relaively easy o increase leverage by buying back shares or issuing deb, firms ha experience an increase in risk may face greaer challenges in reducing leverage by issuing equiy 23