Essays on the Dynamics of Capital Structure

Unversty of New Orleans ScholarWorks@UNO Unversty of New Orleans Theses and Dssertatons Dssertatons and Theses 8-7-2003 Essays on the Dynamcs of Captal Structure Joseph Farhat Unversty of New Orleans Follow ths and addtonal works at: https://scholarworks.uno.edu/td Recommended Ctaton Farhat, Joseph, "Essays on the Dynamcs of Captal Structure" (2003). Unversty of New Orleans Theses and Dssertatons. 468. https://scholarworks.uno.edu/td/468 Ths Dssertaton s brought to you for free and open access by the Dssertatons and Theses at ScholarWorks@UNO. It has been accepted for ncluson n Unversty of New Orleans Theses and Dssertatons by an authorzed admnstrator of ScholarWorks@UNO. The author s solely responsble for ensurng complance wth copyrght. For more nformaton, please contact scholarworks@uno.edu.

ESSAYS ON THE DYNAMICS OF CAPITAL STRUCTURE A Dssertaton Submtted to the Graduate Faculty of the Unversty of New Orleans n partal fulfllment of the requrements for the degree of Doctor of Phlosophy n the Fnancal Economcs Program by Joseph Basheer Farhat B.Com. Economcs, Zagazg Unversty, 1991 M.S. Economcs, Unversty of Jordan, 1997 August 2003

DEDICATION To my famly, for ther overwhelmng love and support.

ACKNOWLEDGEMENTS I would lke to thank Dr. Ranjan D Mello, Dr. Sudha Krshnaswam, Dr. Tarun Mukherjee, Dr. Atsuyuk Naka, Dr. Gerald Whtney, and the 2002 FMA Doctoral Student Consortum partcpants n the panel chared by Dr. Sherdan Ttman for helpful comments. Many thanks go to John Graham for provdng the margnal tax data and to James Bartkus, Adel Sharkas, and Adel Bno for ther support. In addton, I would lke to thank specal and wonderful people Carmen, Ranjan, Sudha, and Tarun.I am thankful for so many thngs. My apprecaton for you wll never fade away.

TABLE OF CONTENTS DEDICATION... ACKNOWLEDGEMENTS... TABLE OF CONTENTS...v LIST OF TABLES...v LIST OF FIGURES...v CHAPTER I INTRODUCTION...1 2.1- THE TRADE-OFF THEORY... 4 2.1- TESTABLE HYPOTHESES...6 2.1- DATA AND SAMPLE SELECTION...7 2.1- TARGET LEVERAGE PROXIES...14 2.1- EMPIRICAL RESULTS...17 2.1- ROBUSTNESS CHECK...32 1.7- CONCLUSIONS...34 CHAPTER II INTRODUCTION...36 2.1- THE PECKING ORDER THEORY...40 2.2- THE PECKING ORDER MODEL...41 2.3- THE SYMMETRICAL BEHAVIOR ASSUMPTION...46 2.4- THE PROPORTIONS OF DEBT FINANCING...52 2.5- THE TRADE-OFF THEORY...65 2.6- THE PARTIAL ADJUSTMENT MODEL...65 2.7- THE SYMMETRICAL RATE OF ADJUSTMENT ASSUMPTION...66 2.8- FACTORS AFFECTING THE RATE OF ADJUSTMENT...66 2.9- THE FINANCING CHOICES AND THE SIZE OF ISSUE...76 2.10- CONCLUSIONS...86 BIBLIOGRAPHY...88 VITA...94 APPENDIX...95 v

LIST OF TABLES 1.1- Sample Dstrbuton across Industres-Years...10 1.2- Industry Classfcatons...12 1.3- Tests of the Stablty of the Intra-Industry Leverage...13 1.4- Regressons-Based Target Leverage...16 1.5- Tukey Parwse Comparsons Test-Mean Dfferences...18 1.6- Kruskal-Walls Parwse Comparsons Test-Locaton Parameters...20 1.7- Correlaton: Leverage Rato and Frm Value...23 1.8- Correlaton: Devaton from the Target Leverage Proxy and the Frm Value...25 1.9- Test of Proportons- Correlatons...27 1.10- Tukey Parwse Comparsons Test-Frm Value...28 1.11- Kruskal-Walls Parwse Comparsons Test-Frm Value...30 1.12- Test of Proportons-Frm Value Dfferences...33 2.1- Sample Dstrbuton across Years...45 2.2- Tests of Peckng Order Model-Symmetrcal Behavor Assumpton...50 2.3- Tests of Peckng Order Model-the Power of the Test...51 2.4- The Dstrbuton of the Proporton of Debt Fnancng...58 2.5- The Proporton of Debt Fnancng-the Frms Attrbutes...59 2.6- The Proporton of Debt Repurchases-the Frms Attrbutes...60 2.7- Correlaton Matrx...62 2.8- The Modfed Peckng Order Model...63 2.9- The Symmetrcal Rate of Adjustment Assumpton...69 2.10- The Short-Run Rate of Adjustment...72 2.11- The Short-Run Rate of Adjustment-the Frms Attrbutes...73 2.12- The Factors Affectng the Rate of Adjustment...75 2.13- The Fnancng Choces and the Sze of Issue...82 2.14- The Repurchases Choces and the Sze of Repurchases...85 v

LIST OF FIGURES 1.1- Leverage-Value Functon...8 1.2- Fnancng Defct-Surplus by Year...47 1.3- Fnancng Defct and Fnancng Surplus by Year...48 v

ABSTRACT Tests of the statc trade-off theory that posts that frms move towards the optmum captal structure necesstate a jont hypothess test - whether frms adjust toward target leverage, and whether the proxy used for target leverage s the true target leverage. Pror studes use the tmeseres mean leverage for each frm, the ndustry medan leverage, an estmated cross-sectonal leverage, and a tobt estmated leverage usng the factors suggested by the statc trade-off theory as proxes for the target leverage. In ths dssertaton, I examne whether these proxes are equvalent and test the consstency of the proxes wth the theorzed behavor of the true target leverage. My results ndcate that the four proxes we examne have sgnfcantly dfferent dstrbutons and ths holds across most ndustres. Further, the ndustry medan leverage s the proxy whch best exhbts behavor consstent wth the true target leverage. Frm value s hgher for frms closer to the ndustry medan and lower for frms away from the ndustry medan. A robustness check usng K-means cluster analyss confrms the superorty of the ndustry medan leverage over the other proxes of target leverage. Ths study complements the prevous studes on the peckng order theory and the trade-off theory. The man purpose of ths study s to nvestgate three ssues that are not consdered n the prevous studes. The adequacy of the specfcaton and the assumptons of the models used n testng the trade-off and the peckng order theory. The second ssue examned n ths study s the valdty to puttng the peckng order and the trade-off theores n a horse race. The fnal ssue examned n ths study s the factors drvng frms to ssue (repurchase) debt or equty or combnaton of both and smultaneously the factors affectng the sze of ssue (repurchase) v

1 CHAPTER I A COMPARATIVE ANALYSIS OF PROXIES FOR TARGET CAPITAL STRUCTURE Introducton Whle most of the academc lterature on the captal structure agrees on the mportance of the target captal structure role n many corporate fnancng models, no attenton has been gven to the accuracy of the dfferent proxes n measurng the optmal captal structure. Are the dfferent proxes equvalent? If not, whch proxy exhbts characterstcs that are most consstent wth the theorzed true optmal captal structure. 1 The tax beneft-bankruptcy cost trade-off models (Baxter (1967), DeAngelo and Masuls (1980), Kraus and Ltzenberger (1973), Robchek and Myers (1966), Scott (1976)) predct that frms wll seek to mantan an optmal captal structure by balancng the benefts and the costs of debt. The benefts nclude the tax sheld whereas the costs nclude expected fnancal dstress costs. Under the agency theoretcal models (Jensen and Mecklng (1976), Myers (1977), Jensen (1986), Stulz (1990), Hart and Moore (1995)) frms use the benefts of reducng potental free cash flow problems and other potental conflcts between managers and shareholders, to offset costs assocated wth undernvestment and asset substtuton problems. These theores predct that frms mantan an optmum captal structure where the margnal beneft of debt equal the margnal cost. The mplcaton of these theores, the target leverage hypothess, s that frms have target leverage and they adjust ther leverage toward the target over tme. Many emprcal studes (Marsh (1982), Jallvand and Harrs (1984), Ttman and Wessels (1988), Fsher, Henkel and Zechner (1989), Macke-Mason (1990), Rajan and Zngales (1995), Graham (1996a), Hull (1999), Hovakman, Opler and Ttman (2001)) fnd support for the target leverage hypothess. Recently, the target leverage hypothess has receved renewed attenton. Studes emerge n whch both the statc trade-off theory and the peckng order theory are jontly tested. Shyam- Sunder and Myers (1999) reject the statc theory and fnd strong confrmaton for peckng order behavor usng the mpact of the funds of flows defct on changes n the debt versus a target adjustment model. Hovakman, Opler, and Ttman (2001) examne the frms debt-equty choce and argue that the resduals of a cross-sectonal regresson on the debt rato are devatons from target. Fama and French (2002) ncorporate dvdend choce, whch s an mportant varable n the peckng order theory, and jontly test the trade off and peckng order models, they fnd evdence n favor and aganst each of the two models. Dssanake, Lambrecht, and Saragga (2001) fnd evdence that the trade-off theory can descrbe the debt polces of some frms. Hovakman (2003) examnes the role of the target leverage n securty ssues and repurchases, 1 There has been no test of whch proxy best serves as the target debt rato. Ths study s the frst to formally examne ths ssue.

2 and fnds that debt reductons are ntated to reduce the devaton from target captal structure whereas debt ssues, equty ssues and equty repurchase are not. The studes that test the target leverage hypothess use alternatve proxes for the target captal structure. For example, Jallvand and Harrs (1984) and Shyam-Sunder and Myers (1999) use the frms leverage mean durng the study perod as a proxy for the target leverage. Auerbach (1985), Hovakman, Opler, and Ttman (2001), Dssanake, Lambrecht, and Saragga (2001), Fama and French (2002) use regresson-based target, where the actual debt rato regressed over several frm- and ndustry-specfc factors suggested by the trade-off theory and prevous emprcal studes. 2 Hull (1999) and Hovakman (2003) use the ndustry medan leverage as a proxy for the target leverage. The need for a target leverage proxy stems from the fact that we cannot observe the true target leverage. Such a proxy needs to have characterstcs that are consstent wth the theorzed true optmal captal structure. The growng lterature that evaluates the trade-off versus the peckng order theory has produced evdence both n favor and aganst each of these theores. The studes that test the trade-off theory necesstate a jont hypothess test - whether the proxy for the target leverage used s the true target leverage, and whether frms adjust toward ths target. Rejectng the hypothess that frms adjust towards the leverage target may be due to the falure of the proxy used to represent the true target, and not because frms do not adjust toward ther target or vse versa. Such a jont hypothess has a measurable effect on the estmaton of the partal adjustment model and debt-equty choce models, and n general, on any hypotheses testng that requres the use of a proxy for optmal captal structure. Usng dfferent proxes for the optmal captal structure mples that these proxes are equvalent and have the same dstrbuton. My emprcal examnatons (parametrc and nonparametrc) of these proxes reveal that they have sgnfcantly dfferent dstrbutons. Ths mples that usng dfferent proxes of target leverage n testng any hypothess would lead to conflctng results. The asymmetry of those dfferent proxes dstrbutons suggest that those proxes are not substtutable, thus t s mportant the fnd out whch proxy has the most consstent characterstcs wth the theorzed target leverage. The trade-off theory provdes clear gudance of the behavor and characterstcs of the true optmal leverage.the measurable predcton of the trade-off models s that frms wll have a leverage level at whch the frms wll maxmze ther value. That s, when frms move closer to ther target they wll have hgher value, ceters parbus, than f they move away from the target leverage rato. Furthermore, for frms that operate below ther target leverage, a postve (negatve) relaton between the leverage rato (devaton from the target) and the frm value s predcted. For frms that operate above ther target leverage, a negatve (negatve) relaton between the leverage rato (devaton from the target) and the frm value s expected. Thus, among the dfferent proxes, the one that s the most consstent wth the true target leverage must exhbt such characterstcs n any partcular year as well as across tme and ndustres. To overcome the problem of usng a nosy proxy for the target leverage, I employ an emprcal approach that evaluates the consstency of dfferent proxes wth the theorzed behavor 2 Harrs and Ravv (1991) present a comprehensve survey of these factors.

3 of the true target leverage. After controllng for the non-homogenety n the nter-ndustry characterstcs, frms are classfed nto ndustres usng the Fama and French (1997) methodology. My emprcal results ndcate that the medan of the ndustry leverage s the most consstent proxy for the true optmal leverage among the dfferent proxes used n ths study. For about three-fourths of the ndustres n my sample, I fnd a negatve relaton between the devaton from the target leverage and frm value for frms that operate above ther ndustry leverage medan and a postve relaton for frms operate below ther ndustry leverage medan. On one hand, for about half of the ndustres such a relaton holds usng the Fama-MacBeth (1973) cross sectonal estmated leverage as proxy. 3 On the other hand, ths relaton holds for a lower percentage of the ndustres for the Tobt cross sectonal estmated leverage and the frm mean leverage as proxes for the target leverage. 4 The emprcal evdence ndcates that n two-thrds of the ndustres, frms ncrease ther value by movng toward the ndustry leverage medan and exhbt a value reducton by movng away from t. Usng the frm mean leverage, the Fama-MacBeth (1973) cross sectonal estmated leverage, and the Tobt cross sectonal estmated leverage as proxes, I fnd that ncreasng the frms value by movng toward the target leverage and the reducton n the frms value by movng away from t hold n about one-thrd, one-half, and one-fourth of the ndustres wth respect to the proxes mentoned above. Snce the leverage target hypothess mples that frms actual leverage wll fluctuate around stable long-run target leverage, ths yelds the mean-revertng property of actual leverage. The mean-revertng property suggests that frms wll cluster around a partcular leverage rato where the cluster wth the hghest market value should have the smallest devaton from the target. I examne ths hypothess usng K-means cluster analyss. I fnd that frms that cluster around a partcular leverage rato and have the hghest market value are those that are closer to the ndustry medan leverage, whch confrms the superorty of the ndustry medan leverage over the other proxes. Numerous emprcal studes support my fndng. For example, Schwarz and Aronson (1967), Scott and Martn (1975), Ferr and Jones (1979), Marsh (1982), Bradley, Jarrell, and Km (1984), Hovakman, Hovakman, and Tehranan (2002), and Welch (2002) observe sgnfcant ndustry effects n frms debt ratos and the frms debt equty choce. Moreover, Scott and Johnson (1982), Pnegar and Wlbrcht (1989), Kamath (1997), and Graham and Harvey (2001) surveys of chef fnancal offcers show that ndustry-wde ratos have an mportant nfluence on CFOs fnancng decson. Another fndng of my study s that frms that operate above the target leverage gan dfferent value relatve to the frms below the target by movng toward the target. Ths suggests that frms wll dffer n terms of how quckly they adjust toward the target dependng on ther 3 Ths reles on the use of year-by-year cross-sectonal regresson of the actual debt rato regressed over several frmand ndustry-specfc factors and then averages the coeffcents across years, where the tme-seres standard errors of the average coeffcents are used to draw nferences. 4 Snce that the leverage rato s bounded from below by zero ths necesstates the use of Tobt regresson. Ths estmaton reles on the use of year-by-year cross-sectonal Tobt regresson of the actual debt rato regressed over several frm- and ndustry-specfc factors and then averages the coeffcents across years, where the tme-seres standard errors of the average coeffcents are used to draw nferences. Whle the frm mean leverage s the hstorcal mean of debt rato for each frm.

4 poston relatve to the target leverage. Ths result s n lne of Byoun and Rhm (2002) who fnd evdence that frms above the target leverage have dfferent speed of adjustment toward the target leverage than those below ther target leverage. 1.1. THE TRADE-OFF THEORY Modglan and Mller (1958) are the frst to establsh the theoretcal foundaton of the modern research of fnancng decson-makng. Under ther famous rrelevance proposton, fnancng decsons do not matter, gven perfect captal markets. Snce then, huge developments n the captal structure theores have emerged. The trade-off models have domnated the captal structure lterature. The tax beneftbankruptcy cost trade-off model predcts that frms operatng at a leverage level beyond the optmum have hgher expected margnal costs of bankruptcy that exceed the margnal tax benefts of debt. Thus, movng toward the optmum wll ncrease the frms market value. Frms operatng at leverage levels below the optmum- where the margnal tax beneft of debt s hgher than the expected margnal cost of bankruptcy- can ncrease ther market value by ncreasng ther debt levels. Under such a model, a frm s proftablty wll be negatvely related to the expected fnancal dstress costs snce frms wth hgher and more stable profts wll have a lower probablty of bankruptcy. Large, well-dversfed frms have less proft volatlty and therefore, hgher debt ratos relatve to small-nondversfed frms. The tax-deductblty of nterest payments also encourages frms to ncur more debt. Because the margnal tax rate s drectly related to the frm s earnngs level, frms wth hgher earnngs are predcted to have more debt. Such a predcton s true as long as frms operate below ther optmal leverage level. When frms operate above ther optmal leverage rato, ncreasng debt wll ncrease the expected costs of fnancal dstress. Thus, f the expected margnal cost of bankruptcy s hgher than the margnal tax beneft of ncreasng debt, managers wll decrease the frm s debt as earnngs rse. Non-debt tax shelds also affect the debt-earnngs relatonshp under the trade-off model. Frms wth hgh non-debt tax shelds are less lkely to ncrease ther debt when ther earnngs ncrease, snce they already enjoy the tax benefts of the non-debt tax shelds. DeAngelo and Masuls (1980) model predcts that debt s less attractve to frms wth hgh non-debt tax shelds. Frms wth hgh non-debt tax shelds may operate below ther optmum leverage level compared to frms wth the same characterstcs but wth lower non-debt tax shelds. Fscher, Henkel, and Zechner (1989) ntroduce a dynamc captal structure model where, under the presence of recaptalzaton costs, a frm s leverage rato wll vary over tme. Thus, frms wll have lower and upper boundares of leverage ratos where they wll recaptalze. Frms below the lower bound wll recaptalze because they forego an ncreased amount of debt tax sheld f they do not. Frms that operate above the upper bound wll also recaptalze due to ncreasng bankruptcy costs and agency conflcts between shareholders and bondholders. Frms wthn the boundary wll not recaptalze persstently snce the beneft of recaptalzaton wll not exceed the recaptalzaton costs. Ther model predcts that frms wth the same characterstcs wll have the same recaptalzaton crtera. However, they could have dfferent leverage ratos

5 wthn common leverage boundares. Therefore, a frm s optmal leverage wll be a range, rather than a pont, wthn each frm tres to reman. The agency cost trade-off models (Jensen and Mecklng (1976), Easterbrook (1984), Jensen (1986), Stulz (1990), Harrs and Ravv (1990), Hart and Moore (1995)) consder the possble conflcts of nterests between the partes nvolved n the frm such as managers, shareholders, and bondholders. Debtholder and shareholder conflcts arse due to the rsk-shftng problem. If an nvestment yelds a hgh proft, shareholders capture most of the gan, whle f the nvestment fals; debtholders bear most of the loss because of the shareholders lmted lablty. Thus, shareholders have an ncentve to nvest n rsker projects after rasng captal n the bond market. Conflcts between shareholders and managers arse because managers do not wholly own the frm. Therefore, managers wll not capture the entre gan when they engage n proft ncreasng actvty, whereas they handle the entre cost of these actvtes. Such conflcts wll motvate managers to transfer frm resources to ther own purposes and to engage n value decreasng actvty. Increasng debt has the ablty to reduce these conflcts. Hgher debt levels wll reduce the free cash flow avalable to managers, lmtng ther capacty to engage n value decreasng actvty and ncrease ther fractonal ownershp. Under the agency cost models, frms dentfy ther optmal captal structure by balancng the costs and benefts of an addtonal dollar of debt. The benefts of debt nclude the reducton of free cash flow problems (Jensen (1986) and Stulz (1990)) and the potental reducton n agency conflcts between managers and shareholders. The costs of debt nclude: agency conflcts between shareholders and bondholders, costs of undernvestment (Stulz (1990) and Myers (1977)), and costs of asset substtuton (Jensen and Mecklng (1976)). Jensen and Mecklng (1976) and Myers (1977) models predct that frms wth low growth opportuntes (and therefore a low possblty of asset substtuton) wll be more levered. Frms wth hgher free cash flows and lmted growth opportuntes wll have hgher debt, whch mtgates the cost of the managershareholder conflcts (Jensen (1986) and Stulz (1990)). Harrs and Ravv s (1990) model predcts that frms wth hgher lqudaton value (tangble assets) are more lkely to have more debt n ther captal structure. In summary, the trade-off models predct that hgher debt wll be assocated wth hgher proftablty, lower non-debt tax shelds, low growth opportuntes, hgh asset tangblty, hgher free cash flows, and lower expected bankruptcy costs. Lower debt wll be assocated wth low proftablty, hgh non-debt tax shelds, hgh growth opportuntes, low asset tangblty, lower free cash flows, and hgher expected bankruptcy costs. In addton, t s predcted that frms wll ncrease ther value by movng toward ther optmal captal structure, whle a value reducton should be observed f they move farther from ther optmal captal structure.

6 1.2. TESTABLE HYPOTHESES 1.2.1. The Symmetry of the Dfferent Proxes Dstrbutons The mplct assumpton when usng any proxy of the commonly used proxes n the lterature to examne the leverage target hypothess s that those proxes are equvalent and substtutable. 5 On other words, ths assumpton means that those proxes have symmetrc dstrbutons. Such an assumpton can be examned usng both parametrc and non-parametrc tests. Rejectng the hypothess of the symmetrc dstrbutons of the dfferent proxes suggest that those proxes are not substtutable and the results of usng dfferent proxes are not comparable. Even f two proxes are close to each other, stll there s a need to nspect whch proxy has the closest characterstcs to the target leverage and has more capablty to mmc the theorzed behavor of the true optmal captal structure. 1.2.2. The Relaton between the Frm Value and the Devaton from the Target Captal Structure The trade-off models predct that the relaton between the frm value and the devaton from the true optmal captal structure s negatve for frms that operate below or above ther optmal captal structure. 6 Ths mples that the relaton between frm value and the actual leverage rato s postve for frms that operate below the optmal leverage and negatve for frms that operate above the optmal leverage. To study the frm value behavor relatve to the devaton from each proxy, frms are grouped by ndustry usng Fama and French (1997) ndustry classfcaton. In each ndustry-year, frms above and below the target proxy are dvded nto two sets (see fgure 1, leverage-value functon). The frst set contans frms wth a devaton from the target of less than the 50 th percentle of the devaton dstrbuton (Q 1 ). The second set contans those frms wth a devaton from the target of more than the 50 th percentle of the devaton dstrbuton (Q 2 ). My motvaton for classfyng frms nto ndustres s the well-documented evdence that ntrandustry frms are more homogenous n ther characterstcs, captal structure, and recaptalzaton crtera (Schwartz and Aronson (1967), Bowen, Daly and Huber (1982), Bradley, Jarrell, and Km (1984), Long and Maltz (1985), and Fscher, Henkel and Zechner (1989)) relatve to nter-ndustry frms. For example, DeAngelo and Masuls (1980) use ths evdence as one argument for the presence of an ndustry-related optmal captal structure. Under the dynamcs models of captal structure (e.g. Fscher, Henkel and Zechner (1989) and Leland (1994)), classfyng frms by ther devaton from the target leverage s motvated by 5 For example, Marsh (1982), Jallvand and Harrs (1984), Shyam-Sunder and Myers (1999), Nur and Archer (2001), and Byoun and Rhm (2002) use the frm mean as a proxy for the target leverage n ther studes. Hull (1999), Hovakman (2003), Hovakman, Hovakman, and Tehranan (2002) consder the ndustry medan leverage as a proxy for the target leverage. Auerbach (1985), Hovakman, Opler, and Ttman (2001), Dssanake, Lambrecht, and Saragga (2001), Le (2001) Fama and French (2002), Korajczyk and Levy (2003) use regresson-based proxes for the target leverage. 6 The frm value defnes as the market value (book value of assets plus the dfference between market value of equty and the book value of equty) standardzed by the total assets. The devaton from the target proxy defnes as the absolute value of the dfference between the actual debt rato and the target proxy.

7 the presence of the recaptalzaton costs, as well as by the substtutablty of debt tax sheld and non-debt tax sheld (DeAngelo and Masuls (1980)). For example, suppose that we have two frms, where the frst frm s somewhat closer to the target leverage than the second frm. These two frms could have the same value f the second frm has suffcent non-debt tax shelds to compensate for the debt tax sheld. Thus, t makes more sense to consder frms closer to the target leverage and to each other to have smlar value relatve to frms far way from the target, gven that the ndustry-year classfcaton controls for the movement on a gven leverage-value functon and the shft n the leverage-value functon. Under the true target leverage hypothess, I antcpate that the average frms value for those n Q 2 and below the target proxy to be less than that of frms that belong to Q 1 and are below the target proxy. For frms above the target, I expect to fnd that the average frms value for the frms n the Q 1 set s hgher than that for frms n the Q 2 set. To examne the second hypothess - the postve (negatve) relaton between frm value and the actual leverage rato for frms that operate below (above) the optmal leverage - frms are classfed nto two categores: the frst category contans frms above ther target leverage proxy whle the second contans frms below ther target leverage proxy. Then, for each category, the correlaton between frm value and leverage rato, and frm value and the devaton from the target s tested for each ndustry usng the parametrc (Pearson) and the nonparametrc (Spearman) correlatons. 1.3. DATA AND SAMPLE SELECTION The ntal sample conssts of all frms on the Compustat database for the perod 1981-2000. I classfy frms nto 48 group usng Fama-French (1997) ndustry classfcatons. As n prevous studes, fnancal frms (SIC 6000-6999) and non-classfable establshments (9900-9999) are excluded. 7 To enter the sample, fnancal data must be avalable to calculate the leverage rato (total debt/total assets) and market value (book value of assets plus the dfference between market value of equty and the book value of equty). Frms that have negatve debt or zero total assets n any gven year are excluded from the analyss n that year. Fnally, I requre that there are at least ffteen frms wthn the same ndustry n any gven year of the study perod. 8 Ths restrcton allows the statstcal tests to have the suffcent degrees of freedom to draw relable nferences. Table 1.1 shows the number of observatons n the sample for each ndustry ncluded n the analyss across all years. Applyng the above crtera, the fnal sample has 40 ndustres over the perod 1981-2000 and 130,939 frm-year observatons. Overall, the number of frms n my sample s ncreases over tme (from 5,180 n 1981 to 7,810 frms n 2000). The number of frms across ndustry vares dramatcally whereas wthn a gven ndustry group t shows more stablty over tme. A descrpton of Fama-French (1997) ndustry group s provded n table 1.2 and a detaled descrpton of the SIC n each ndustry group n appendx A1. 7 Non-classfable establshments are excluded to avod the non-homogenety n the frms characterstcs. 8 Only two ndustres n my sample have observatons close to ffteen n a gven year. The other ndustres have suffcent observatons to draw nferences from both the parametrc and nonparametrc tests.

8 Fgure 1.1 Proxy (Optmal Debt) Below Above Frm Value Devaton from optmalty Q 2 Q 1 Q 1 Q 2 Leverage-Value Functon

9 1.3.1. Leverage Rato Stablty over Tme The mean-revertng property of debt rato under the leverage target hypothess mples that debt rato should not vary randomly for a gven ndustry over tme. To examne the stablty of the leverage rato across years for ntra-ndustry frms, nonparametrc and parametrc methods are employed. The second column of Table 1.3 reports the P-Values of the Kruskal-Walls test for equalty of the locaton parameters of the debt rato dstrbutons across years for each ndustry. 9 The thrd column reports the P-Value of the ANOVA test for equalty of the means of the debt rato across years for each ndustry. The hypothess that an ndustry mean (medan) debt rato s the same across the twenty years perod of my study s rejected n two-thrd of the ndustres usng the Kruskal-Walls test and n one-thrd of the ndustres usng parametrc (ANOVA) test. Though t s possble that the locaton parameters of the debt rato dstrbutons across years are not the same but very few of them are sgnfcantly dfferent, to detect such dfferences, I use the Tukey parwse comparsons test whch allows to jontly perform all possble parwse comparsons of the means usng a sngle level of sgnfcance. For example, for each ndustry, the average debt rato n the year 1981 s compared wth the average debt rato n each other year, and so on for all the possble combnatons of years. Snce, I am comparng means of the same ndustry across 20 years, the possble number of parwse comparsons assocated wth the ANOVA s 190 pars for each ndustry ( H : µ =, = 1981,...,2000 j = 1982,..., 1999 j ). 10 0 µ j The fourth column of Table 1.3 reports the percentage of pars that shows a sgnfcant dfference out of all the possble pars. For example, the ANOVA test for the ndustry group number four (Beer & Lquor) rejects the hypothess that the average debt rato s the same across the twenty years of my study, but the percentage of pars that shows a sgnfcant dfference out of all the possble pars s 3.16%. In other words, there s a sgnfcant dfference n the average debt rato n only sx pars of all the possble pars. Thus, the ANOVA test rejects the hypothess that the average debt rato s the same across the twenty years, because of the sgnfcant dfference n the debt rato across few years. The Tukey parwse comparsons test reveals that the dfferences n the average debt rato are not persstent across all years. Whle the stablty of the leverage rato across all years s rejected n 37.5% of the ndustres, ths s due to dfferences n the leverage rato across some of the years, whch at most accounts for only 19% of the twenty years perod n my study. In general, the results ndcate that the ntra-ndustry leverage rato s stable over tme. These results are consstent wth Bowen, Daly, and Huber (1982) who fnd that ndustres tend to retan ther leverage rato rankng over tme. In addton, these results confrm the earler fndng of Schwartz and Aronson (1967) of the remarkable overall stablty n the fnancal structure of ndustres over tme. Taggart (1977), Marsh (1982), Auerbach (1985) fnd evdence on mean reverson n frms debt ratos, and those frms appear to adjust toward debt targets. The key mplcaton of leverage rato stablty s that under the trade-off model, the leverage rato should not vary 9 A P-Value less than 5% ndcates a rejecton of the null hypothess at 5% sgnfcance level. 10 The number of possble parwse comparsons of r s r! 2! ( r 2 )!.

10 Table 1.1- Sample Dstrbuton across Industres- Years The sample perod s 1981-2000. Fnancal frms are excluded. An ndustry s defned usng Fama-French (1997) ndustry classfcaton. To enter the sample, fnancal data must be avalable to calculate the leverage (total debt/total assets), Market Value (book value of assets plus the dfference between market value of equty and the book value of equty). Industres wth less than 15 frms n any gven year are excluded to allow for enough degrees of freedom for the statstcal tests. The fnancal data obtaned from Compustat database. Frms frequency by ndustry-year Industry 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 1 22 22 27 31 28 28 30 28 26 27 27 27 25 27 28 27 30 29 30 30 2 130 124 120 122 122 126 138 137 132 127 122 123 128 137 136 149 158 150 139 136 4 19 18 18 16 17 16 16 15 15 18 16 15 17 17 26 31 34 34 32 30 6 58 58 53 55 51 57 63 62 65 64 62 62 71 85 92 102 102 99 92 84 7 108 112 122 130 140 151 162 159 160 154 145 139 152 169 175 187 189 188 171 165 8 73 69 70 68 65 68 68 71 67 75 76 77 73 76 77 78 81 78 80 78 9 148 148 152 160 161 170 174 170 158 143 140 144 145 149 153 166 170 164 160 151 10 112 98 95 95 86 85 83 86 87 90 91 86 102 106 106 115 121 120 111 108 11 60 54 66 72 76 96 94 106 108 122 150 169 179 182 186 192 177 159 141 132 12 95 100 111 134 149 160 183 190 190 196 209 237 254 251 250 287 292 282 277 259 13 71 77 98 113 126 146 164 173 181 184 214 239 271 290 300 357 393 393 412 401 14 97 93 96 100 104 109 113 126 125 129 130 136 138 144 142 144 138 138 141 136 15 98 94 97 96 95 98 100 93 89 88 83 83 93 93 95 96 95 91 86 82 16 80 74 70 73 66 61 62 68 57 52 53 57 60 57 54 55 53 51 48 38 17 245 235 228 219 214 207 200 196 184 173 167 159 168 172 179 186 183 174 162 146 18 101 100 95 93 85 90 96 98 92 96 92 87 101 103 105 112 119 117 116 97 19 118 114 107 105 105 102 99 111 108 110 110 110 120 125 124 133 135 133 125 119 20 48 45 45 45 41 40 35 39 38 37 38 39 38 40 40 39 40 36 34 32 21 246 250 249 251 249 250 261 263 257 241 234 231 240 255 264 291 291 282 278 257 22 110 107 110 113 107 108 106 111 109 109 112 111 113 119 121 125 123 118 118 110

11 Table 1.1 contnued Frms frequency by ndustry-year Industry 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 23 108 106 108 106 106 104 105 107 106 104 101 105 113 126 128 130 130 125 120 107 24 41 39 36 37 33 32 32 32 31 33 32 31 29 28 27 31 33 31 28 27 25 17 17 18 16 15 15 15 16 16 17 16 18 17 19 20 20 20 19 19 15 27 31 29 38 40 49 58 69 73 67 76 75 73 70 64 70 74 65 55 49 45 28 36 33 33 32 29 29 31 40 41 40 37 37 37 37 33 33 37 37 37 37 30 398 477 491 481 464 427 402 382 370 368 366 346 353 371 364 372 372 346 322 312 31 256 255 258 256 255 251 245 238 238 235 230 227 228 229 229 224 224 217 209 197 32 163 161 179 182 198 202 217 221 228 230 235 245 268 278 299 351 361 358 420 410 33 45 47 49 52 55 60 65 65 63 66 68 68 64 71 85 104 101 99 92 91 34 334 339 394 452 471 549 570 567 544 523 528 570 632 697 804 1046 1145 1208 1454 1424 35 165 179 242 267 284 310 316 325 321 298 297 307 327 349 373 433 452 452 464 443 36 249 252 287 318 329 339 358 343 328 329 322 328 354 384 426 468 476 487 544 516 37 129 129 139 153 158 171 173 167 151 147 146 153 161 163 169 189 192 193 203 188 38 117 113 108 107 107 107 108 106 100 100 101 104 105 111 117 122 121 114 111 107 39 24 23 22 23 21 24 25 24 22 20 16 20 22 22 24 25 26 25 27 25 40 166 172 189 188 187 195 196 190 188 182 171 176 194 196 203 227 232 225 228 213 41 283 289 292 313 313 326 349 354 344 333 331 319 342 366 377 399 403 378 365 349 42 336 316 328 342 328 361 387 372 361 358 365 377 403 413 421 463 472 453 439 422 43 142 143 147 157 159 164 160 156 149 144 149 159 160 175 190 210 211 194 182 168 48 101 101 103 109 112 134 151 165 164 174 177 176 177 174 184 182 171 161 145 123 All 5180 5212 5490 5722 5760 6026 6221 6245 6080 6012 6034 6170 6544 6870 7196 7975 8168 8013 8211 7810

12 Table 1.2- Industry Classfcatons Industres are classfed n groups usng Fama and French (1997) ndustry classfcaton. A detal of the classfcaton method and the ndustry SIC s provded n appendx A1. Code Industry group Code Industry group 1 Agrculture 23 Automobles and Trucks 2 Food Products 24 Arcraft 4 Beer & Lquor 25 Shpbuldng, Ralroad Equpment 6 Recreaton 27 Precous Metals 7 Entertanment 28 Non-Metallc and Industral Metal Mnng 8 Prntng and Publshng 30 Petroleum and Natural Gas 9 Consumer Goods 31 Utltes 10 Apparel 32 Communcaton 11 Healthcare 33 Personal Servces 12 Medcal Equpment 34 Busness Servces 13 Pharmaceutcal Products 35 Computers 14 Chemcals 36 Electronc Equpment 15 Rubber and Plastc Products 37 Measurng and Control Equpment 16 Textles 38 Busness Supples 17 Constructon Materals 39 Shppng Contaners 18 Constructon 40 Transportaton 19 Steel Works Etc 41 Wholesale 20 Fabrcated Products 42 Retal 21 Machnery 43 Restaurants, Hotels, Motels 22 Electrcal Equpment 48 Mscellaneous

13 Table 1.3-Parametrc and Nonparametrc Tests of the Stablty of the Intra-Industry Leverage The Kruskal-Walls test s a nonparametrc test of equalty of the locaton parameters of the leverage rato dstrbutons for ntra-ndustry frms across years. The ANOVA test s a parametrc test of equalty of the means of the leverage rato for ntra-ndustry frms across years. The null hypothess n both tests s stated as the follows: H0:µ1981= µ1982= =. µ2000. A Tukey parwse comparsons test allows for smultaneously carryng out all 190- hypothess tests at a sngle, gven level of sgnfcance. The Tukey parwse comparsons test column report the percentage of years-pars n whch the dfferences n the means are sgnfcantly dfferent from zero at the 5% level of sgnfcance. Industry group Kruskal-Walls Test ANOVA Test Tukey Parwse Comparsons Test P-Value P-Value Percentage 1 0.36 0.14 6.84% 2 0.28 0.39 4.21% 4 0.10 0.00 3.16% 6 0.00 0.27 2.11% 7 0.11 0.26 0.00% 8 0.00 0.00 7.89% 9 0.00 0.00 11.58% 10 0.00 0.46 0.00% 11 0.02 0.18 0.00% 12 0.00 0.52 2.63% 13 0.00 0.53 0.00% 14 0.00 0.49 0.00% 15 0.00 0.00 2.11% 16 0.00 0.11 3.16% 17 0.09 0.51 0.00% 18 0.29 0.16 0.00% 19 0.11 0.61 0.00% 20 0.28 0.36 0.00% 21 0.00 0.06 2.63% 22 0.02 0.50 0.00% 23 0.00 0.00 7.89% 24 0.65 0.27 0.00% 25 0.56 0.77 0.00% 27 0.26 0.67 0.00% 28 0.12 0.32 4.21% 30 0.00 0.32 0.00% 31 0.00 0.00 2.11% 32 0.00 0.00 7.89% 33 0.10 0.10 0.00% 34 0.00 0.05 0.53% 35 0.00 0.02 10.53% 36 0.00 0.34 7.89% 37 0.00 0.16 1.58% 38 0.00 0.00 1.05% 39 0.00 0.00 2.11% 40 0.29 0.36 0.00% 41 0.00 0.06 4.74% 42 0.00 0.00 18.95% 43 0.00 0.45 0.53% 48 0.00 0.72 0.00%

14 randomly for a gven ndustry over tme. Such a result works n favor of the exstence of an optmal captal structure that frms try to preserve across tme. 1.4. TARGET LEVERAGE PROXIES In ths secton, I revew the major proxes for the optmal captal structure commonly used n the lterature and consdered n ths study. Marsh (1982) uses the average debt rato over the study perod for each frm (frm mean, hereafter) as a proxy for the target leverage n hs study of debt equty choce. Jallvand and Harrs (1984) use the frm mean n ther study of the target adjustment model. Shyam-Sunder and Myers (1999), Nur and Archer (2001), and Byoun and Rhm (2002) use the frm mean as a proxy for the target leverage n ther studes of testng the trade-off model vs. the peckng order model. Thus, the frm mean s ncluded n ths study as a one of the target leverage proxes. Hull (1999) uses the ndustry medan leverage as a proxy for the target leverage n hs study of the consstency of the market reacton to the debt equty swaps and equty ssue wth the sole purpose to reduce debt wth the trade-off models. Hovakman (2003), Hovakman, Hovakman, and Tehranan (2002) consder the ndustry medan leverage as proxy for the target leverage n studyng the role of target leverage n the frm decson to ssue and repurchase securty or to ssue combnaton of debt and equty. I nclude the ndustry leverage medan as one of the proxes n ths study (ndustry medan, hereafter). Auerbach (1985), Hovakman, Opler, and Ttman (2001), Dssanake, Lambrecht, and Saragga (2001), Le (2002), Fama and French (2002), Korajczyk and Levy (2003) use regresson-based proxes for the target leverage. They regress the actual debt rato over several frm- and ndustry-specfc factors suggested by the trade-off theory and prevous emprcal studes. The fact that leverage rato s bounded by zero necesstates the use of an econometrc technque such as a Tobt regresson to prevent the estmated leverage rato from beng negatve and to obtan consstent estmators. Hovakman, Opler, and Ttman (2001) are the frst to control for ths problem usng Tobt regresson model. I nclude two regresson-based proxes; the frst s estmated usng Fama-MacBeth (1973) procedures (cross-sectonal, hereafter) and the second are estmated usng a Tobt model estmaton to account for the fact that zero bound the debt rato from below (Tobt-cross-sectonal, hereafter). For the regresson-based proxes the actual debt rato s regressed over the followng ndependent varables: Growth optons: Rajan and Zngales (1995), Ttman and Wessels (1988) argue that frms wth hgh growth optons depend on equty fnancng more than on debt fnancng. Ther argument suggests a negatve relaton between leverage and growth optons. Lke Fama and French (2002) and Hovakman, Opler and Ttman (2001)), I use both the rato of market-tobook value and the rato of research and development to total assets as proxes for the frm s growth optons. Sze: Ttman and Wessels (1988) argue that larger frms are more dversfed, thus they face lower probablty of bankruptcy. Ths suggests that the larger the frm sze, the hgher the

15 frm s debt capacty. Thus, sze s expected to be postvely related to leverage. The logarthm of total assets used as proxy frm sze. Tangble Assets: In addton to sze, Ttman and Wessels (1988) argue that tangblty of assets, as a measure of collateral, s postvely related to leverage. Thus, frms wth a hgher percentage of tangble assets from ther total assets wll have a hgher capacty to rase debt. Non-debt tax shelds: accordng to Modglan and Mller (1958), the major ncentve for borrowng s to take advantage of nterest tax shelds. The presence of other non-debt tax shelds (deprecaton and amortzaton) mtgates such ncentve. The rato of deprecaton to total assets s used as a measure of the non-debt tax shelds. Proftablty: t has been argued that the probablty of bankruptcy rses as the volatlty of earnngs ncreases. Snce operatng ncome s ndependent of the effects of leverage and snce t represents the ncome avalable for nterest payments, hgher proftable frms expected to have hgher debt rato. Followng Fama and French (2002), earnngs before nterest and taxes scaled by total assets are used as a proxy for proftablty. Table 1.4 shows the results of the estmatons of both the cross-sectonal and the Tobtcross-sectonal proxes usng Fama and MacBeth (1973) and Tobt estmaton methods. The coeffcents of the ndependent varables represent the means across years. To account for the autocorrelaton n the annual coeffcents, I follow Fama and French (2002) procedure n approxmatng the nflaton factor of the standard errors of annual Coeffcents. 11 The frst autocorrelatons of the slopes are between 0.3 and 0.62 and for a longer lags t decay lke an AR1. Thus, I requre t-statstcs above 5.2 and 7.1 to nfer relablty at 0.05 and 0.01 levels respectvely. My results support the prevous emprcal results of Bradley, Jarrell and Km (1985), Long and Maltz (1985), Ttman and Wessels (1988), Rajan and Zngales (1995), Hovakman, Opler and Ttman (2001), and Fama and French (2002). All the ndependents varables coeffcents sgns are consstent wth ther predcted sgn, except for the market to book rato n the Fama- MacBeth regresson. Fama and French (2002) also fnd such a postve relaton usng the same estmaton method. 1.5. EMPIRICAL RESULTS To study whether the dfferent proxes have symmetrc dstrbutons, I report n Tables 1.5 and 1.6 the parametrc and nonparametrc tests of the dfferences n the locaton parameters of each proxy dstrbuton. Both the ANOVA and the Kruskal-Walls tests reject the hypothess that there are no dfferences n the locaton parameters among the four proxes across all the ndustres at 10% sgnfcance level. 11 As Fama and French pont out the procedure s a conservatve approach to account for the autocorrelatons n the annual coeffcents.

16 Table 1.4-Regressons-Based Target Leverage The dependent varable s defned as the total debt dvded by total assets. The Fama-MacBeth (1973) regressons are run for each year of 1981-2000 perod. The Tobt model estmaton regressons are run for each year of 1981-2000 perod. The coeffcents of the ndependent varables represent the means across years. To count for the autocorrelaton n the annual coeffcents, I follow Fama and French (2001) procedure by approxmatng the nflaton factor of the stranded errors of annual coeffcents. To count for the autocorrelaton n the annual slops. Thus, I requre t-statstcs above 5.2 and 7.1 to nfer relablty at 0.05 and 0.01 levels respectvely. ** and * ndcate that the coeffcent s statstcally dfferent from zero at 0.01 and 0.05 levels. Independent Varables Dependent varable debt rato Estmaton Method Fama-MacBeth Tobt model Constant 0.085** -1.708** Market To Book Rato 0.015** -0.036* R&D to Total Assets -0.196* -0.455* Instrument varable =1 f frm have no R&D 0.076* 0.136** Sze-Logarthm of Total Assets 0.011* 0.088** Tangble Asssts to Total Assets 0.158** 0.854** Deprecaton to Total Assets -0577* -0.463* Proftablty -0.213** -0.398** R-Square 0.221 0.183

17 However, gven the fact that I am nterested n knowng whch proxes are dfferent, I employ the Tukey parwse comparsons test (parametrc) and the Kruskal-Walls parwse comparsons test (nonparametrc) to dentfy whch pars of proxes are the source of the dfferences. Examnng the percentages of sgnfcant dfferences among the dfferent pars of proxes across all the ndustres obvously reveals that the proxes dstrbutons are not symmetrc. I fnd that there are no sgnfcant dfferences n the proxes dstrbutons n about 32.5% (30% usng nonparametrc test) of the ndustres between the frm mean leverage and the ndustry medan leverage. Whle, there are no sgnfcant dfferences n the proxes dstrbutons between the frm's mean leverage and the cross-sectonal leverage for about 55% (40% usng nonparametrc test) of the ndustres. For only 15% (0% usng nonparametrc test) of the ndustres there are no sgnfcant dfferences n the proxes dstrbutons between the frm mean leverage and the Tobt cross-sectonal leverage. Whle there are no sgnfcant dfferences n the proxes dstrbutons between the ndustry medan leverage and the cross-sectonal leverage proxes for about 45% (17.5% usng nonparametrc test) of the ndustres. The ndustry medan leverage and the Tobt cross-sectonal leverage proxes show no sgnfcant dfferences n ther dstrbutons n about 20% (2.5% usng nonparametrc test) of the ndustres. Fnally, there are no sgnfcant dfferences between the cross-sectonal leverage and the Tobt cross-sectonal leverage n about 15% (0% usng nonparametrc test) of the ndustres. The overall result shows, there s strong evdence that the dfferent proxes have sgnfcantly dfferent dstrbutons. Nevertheless, across all ndustres, the frm mean and crosssectonal proxes have the closest dstrbutons; ths s ndcated by both the parametrc and nonparametrc tests. The results above suggest that usng dfferent proxes of optmal captal structure n corporate fnance models could lead to dfferent results dependng on the proxy used. For example, the parametrc and nonparametrc tests ndcate that the dstrbuton of the frm mean proxy s to the left of the dstrbuton of the Tobt cross-sectonal proxy, suggestng that the frm mean as a target proxy yelds more over-leveraged frms relatve to the Tobt cross-sectonal proxy. Gven that the dfferent proxes are not equvalent, I turn now to nvestgate whch proxy exhbts characterstcs that are most consstent wth the theorzed true optmal captal structure. I frst examne the relaton between the leverage rato and the frm value for frms above and below ther target leverage. Table 1.7 presents the correlatons between leverage rato and frm value over tme for frms above and below ther target leverage, whle Table 1.8 presents the correlatons between the absolute value of the devaton from the target leverage and frm value. Table 1.9 shows, for each proxy, the proportons of ndustres that yeld consstent results wth the predcton of the trade-off model among all the ndustres. As t s obvous from table 1.9, among all the proxes, the ndustry medan s the most consstent wth the predcton of the tradeoff model.

18 Table 1.5-Tukey Parwse Comparsons Test -Mean Dfferences The dfferences n means among the deferent proxes are conducted usng T-test. Proxy1 represents the frm s leverage mean for the study perod, proxy2 represents the ndustry leverage medan (An ndustry s defned usng Fama-French (1997) ndustry classfcaton), proxy3 represents the estmated cross-sectonal leverage (estmated usng Fama-MacBeth (1973) procedures) and proxy4 represents the estmated cross-sectonal leverage usng Tobt model. The Tukey parwse comparsons test allows for smultaneously carryng a jont hypothess testng at a sngle gven level of sgnfcance. * ndcate that dfference n the means s statstcally dfferent from zero 0.05 levels. Industry ANOVA P-Value Proxy1-Proxy2 Mean dfferences Proxy1-Proxy 3 Mean dfferences Proxy1-Proxy4 Mean dfferences Proxy2-Proxy3 Mean dfferences Proxy2-Proxy4 Mean dfferences Proxy3-Proxy4 Mean dfferences 1 0.00 0.07* 0.07* 0.18* 0.01 0.11* 0.11* 2 0.00 0.01* 0.01 0.12* -0.01* 0.11* 0.12* 4 0.10-0.03-0.01 0.00 0.02 0.03 0.02 6 0.00-0.01-0.01 0.17* 0.00 0.18* 0.18* 7 0.00 0.04* 0.00 0.16* -0.03* 0.12* 0.15* 8 0.00 0.00-0.03 0.11* -0.02 0.11* 0.14* 9 0.00 0.01-0.02* 0.11* -0.03* 0.10* 0.13* 10 0.00 0.08* 0.04 0.18* -0.04 0.10* 0.14* 11 0.00 0.23* 0.03 0.03-0.21* -0.20* 0.01 12 0.00 0.13 0.02 0.21* -0.11 0.08 0.19* 13 0.00 0.11* 0.05* 0.19* -0.06* 0.08* 0.14* 14 0.00 0.03-0.02 0.07* -0.05* 0.04* 0.09* 15 0.00 0.04* 0.02* 0.16* -0.03* 0.12* 0.15* 16 0.00-0.04* -0.04* 0.00 0.00 0.04* 0.04* 17 0.00 0.01 0.00 0.12* 0.00 0.11* 0.12* 18 0.00 0.07* 0.02 0.18* -0.05* 0.12* 0.17* 19 0.00 0.01* 0.00 0.08* -0.01* 0.07* 0.08* 20 0.00 0.04* 0.00 0.12* -0.04* 0.08* 0.12* 21 0.00 0.05* 0.03* 0.16* -0.02* 0.10* 0.13* 22 0.10 0.17 0.07 0.17-0.09 0.01 0.10