THE IMPACT OF FINANCIAL CONSTRAINTS ON THE RELATION BETWEEN INVESTOR-LEVEL TAXES AND CAPITAL STRUCTURE DECISIONS. Stephen John Lusch

Transcription

1 THE IMPACT OF FINANCIAL CONSTRAINTS ON THE RELATION BETWEEN INVESTOR-LEVEL TAXES AND CAPITAL STRUCTURE DECISIONS by Stephen John Lusch Copyright Stephen John Lusch 2014 A Dissertation Submitted to the Faculty of the DEPARTMENT OF MANAGEMENT In Partial Fulfillment of the Requirements For the Degree of DOCTOR OF PHILOSOPHY WITH A MAJOR IN ACCOUNTING In the Graduate College THE UNIVERSITY OF ARIZONA 2014

2 2 THE UNIVERSITY OF ARIZONA GRADUATE COLLEGE As members of the Dissertation Committee, we certify that we have read the dissertation prepared by Stephen John Lusch, titled The Impact of Financial Constraints on the Relation between Investor-Level Taxes and Capital Structure Decisions and recommend that it be accepted as fulfilling the dissertation requirement for the Degree of Doctor of Philosophy. Date: (3/28/2014) Dan S. Dhaliwal Date: (3/28/2014) Leslie G. Eldenburg Date: (3/28/2014) Kirsten A. Cook Final approval and acceptance of this dissertation is contingent upon the candidate s submission of the final copies of the dissertation to the Graduate College. I hereby certify that I have read this dissertation prepared under my direction and recommend that it be accepted as fulfilling the dissertation requirement. Date: (3/28/2014) Dissertation Director: Dan S. Dhaliwal

3 3 STATEMENT BY AUTHOR This dissertation has been submitted in partial fulfillment of the requirements for an advanced degree at the University of Arizona and is deposited in the University Library to be made available to borrowers under rules of the Library. Brief quotations from this dissertation are allowable without special permission, provided that an accurate acknowledgement of the source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the copyright holder. SIGNED: STEPHEN JOHN LUSCH

4 4 ACKNOWLEDGEMENTS I will be forever thankful for all the support and encouragement I have received from my family, friends, fellow doctoral students, faculty, co-authors, and dissertation committee members over the past five years at the University of Arizona. I am particularly grateful to my dissertation committee: Dan Dhaliwal (chair), Kirsten Cook, and Leslie Eldenburg for their support and guidance on my dissertation. My dissertation has also benefited from the helpful comments and suggestions of Dane Christensen, Katharine Drake, Andrew Finley, Nathan Goldman, Curtis Hall, Timothy Hinkle, Dave Kenchington, Phil Lamoreaux, Jim Seida, Shyam V. Sunder, brownbag participants at the University of Arizona, and workshop participants at the University of Arizona. I wish to thank all of the University of Arizona accounting faculty and doctoral students for everything that I have learned from you all. In particular, I owe additional recognition to the accounting (Phil Lamoreaux) and finance (Filippo Curti, DJ Fairhurst, and Hayden Kane) PhD students who entered the PhD program in fall Pushing through the coursework in the first year was made easier and more enjoyable by having such a good group of colleagues to work with. Many thanks are also due to the doctoral students in the classes above me, in particular James Chyz, Logan Steele, John Campbell, Dane Christensen, Dave Kenchington, Landon Mauler, Curtis Hall, Timothy Hinkle, and Ben Hoffman, for listening to random research ideas and answering questions about research design and SAS. I hope that I have been able to provide the students who entered the program after me the same sort of support and guidance that you all gave to me. As I will be entering the postulant program with the Congregation of Holy Cross upon graduation, I give many thanks to those who helped me discern my call to religious life. This includes Fr. Drew Gawrych, C.S.C., Fr. Jim Gallagher, C.S.C., Fr. Bill Remmel, S.D.S, Fr. Bart Hutcherson, O.P., Br. Jim Siwicki, S.J., Fr. Dave Farnum, C.S.P., Lupita Parra, Stella Samorano, Jim Seida, Ken Milani, Pat Murphy, and Gene Laczniak. I am particularly grateful for the guidance of Fr. Dan Parrish, C.S.C., who served as my vocations mentor and as a fellow PhD student was able to provide me guidance in both matters related to faith and academia. Finally, words cannot express the gratitude I owe my family for all the love and support they have provided to me during not only the PhD program, but throughout life. In particular the love of my parents, my brother, and my sister are unparalleled and has undoubtedly been an important factor in all that I have achieved.

5 5 DEDICATION To my father, Robert, for being such a strong role model as a father, teacher, and researcher To my mother, Virginia, for your compassion and absolute willingness to selflessly serve others To my twin brother, Mark, for being my closest friend and confidant, we have roamed this earth for 27 years together and I look forward to many more To my sister, Heather, for being one of my biggest fans and supporters, you never doubted that I could achieve whatever I set my mind to Finally, in the words of my friend Monte Shaffer, I dedicate this to all those who, whether by faith or science, seek truth

6 6 TABLE OF CONTENTS LIST OF FIGURES AND TABLES...8 ABSTRACT...9 I. INTRODUCTION...10 II. LITERATURE REVIEW...16 II.A Taxes and Capital Structure...16 II.B Jobs and Growth Tax Relief Reconciliation Act of III. HYPOTHESES DEVELOPMENT...26 IV. RESEARCH DESIGN...35 IV.A Regression Models...35 IV.B Variable Definitions...35 IV.C Sample...39 V. RESULTS...40 V.A Descriptive Statistics...40 V.B Multivariate Results...41 VI. SUPPLEMENTAL ANALYSES...45 VI.A Increased Distributions for Unconstrained Firms...45 VI.B Alternative Hypothesis: Increased Distributions and Financing Deficits...46 VI.C Combined Financial Constraint Measure...50 VII. CONCLUSION...53 APPENDIX A: Derivation of equations (7), (8), and (9)...55 APPENDIX B: Reconciliation of Table 1 to Total Derivatives...58 APPENDIX C: Variable Definitions...59 APPENDIX D: Figures...60 APPENDIX E: Tables...63

7 REFERENCES

8 8 LIST OF FIGURES AND TABLES TABLE 1: Gains from Leverage Pre- and Post Act...31 FIGURE 1: Optimal Financial Policy...61 FIGURE 2: Changes in Debt in the Falsification and 2003 Act Periods...62 TABLE 2: Sample Selection...64 TABLE 3: Descriptive Statistics...65 TABLE 4: Correlations...68 TABLE 5: Change in Debt from 2002 to 2004 and Financial Constraints for Tax Sensitive and Non-Tax Sensitive Firms...69 TABLE 6: Difference-in-Differences for Change in Debt from and Falsification Periods...71 TABLE 7: Change in Equity Distributions for High Individual Ownership Firms...73 TABLE 8: Change in Debt from 2002 to 2004 and Financial Constraints for Non- Distribution Increasing Firms Only...74 TABLE 9: Change in Debt from 2002 to 2004 using a Combined Financial Constraint Measure...76

9 9 ABSTRACT This study addresses the question of whether the relation between investor-level taxes and a firm s capital structure decisions varies predictably with financial constraints. Using the setting of the 2003 reduction in individual tax rates for ordinary income, dividends, and capital gains, this study documents that constrained firms decrease their debt use in response to the 2003 tax cuts, while unconstrained firms increase their debt use over the same period. I find these effects are only evident among firms with relatively high individual ownership, which is the group of firms that theory suggests will react to the tax cuts. This paper contributes to the literature on how investor-level taxes influence firms financing decisions as well as the literature pertaining to the 2003 Tax Act.

10 10 I. INTRODUCTION This study examines whether financial constraints influence the relation between investor-level taxes and firms capital structure decisions. 1 In particular, whether the investor-level tax cuts of the Jobs and Growth Tax Relief Reconciliation Act of 2003 (the 2003 Act) elicit different capital structure shifts for constrained and unconstrained firms, controlling for other determinants of capital structure. Answering this question increases our understanding of the factors that influence firms capital structure decisions and provides additional empirical evidence regarding the effect of financial constraints on managerial decision-making. The firm s choice to use debt involves a cost-benefit trade-off. The deductibility of interest payments for corporate income tax purposes encourages firms to finance projects with debt rather than external equity; however, adding additional debt to the balance sheet comes at a cost to the company, such as increasing the expected costs of bankruptcy (e.g., Kraus and Litzenberger 1973). Noting that the incremental expected costs of bankruptcy associated with incremental debt use are rather small, Miller (1977) argues that the primary cost associated with adding additional debt to the capital structure of the firm is the personal tax penalty arising from a tax system with a higher individual tax rate on interest payments than on equity distributions. In particular, when investors face a higher tax rate on the interest income from debt securities compared to income from equity securities, investors will demand a higher pre-tax rate of return on the debt relative to when the tax rates that investors face on debt and equity income are equal. 1 Consistent with the literature, the term investor-level tax rates pertains to the tax rates faced by individual investors. Thus this paper will use the terms investor-level taxes and individual taxes synonymously.

11 11 This asymmetry in tax rates on debt income and equity income causes the cost of debt to increase relative to the cost of equity. However, when a firm is financially unconstrained it has the ability to fund new projects with internal capital, leading to differences in capital structure decision-making for financially constrained versus unconstrained firms. Hennessy and Whited (2005) determine that increased debt use by high liquidity firms (i.e., unconstrained) is less beneficial than for low liquidity firms (i.e., constrained), as the use of additional debt serves to increase shareholder distributions rather than replace costly external equity. Accordingly, Kemsley (2013) incorporates a budget constraint, based on whether the firm has the ability to finance new projects completely with internal funds, into the Miller (1977) trade-off model. When the firm is constrained, the budget constraint is binding, and the trade-off model reduces to the original Miller (1977) model. However, for unconstrained firms, the budget constraint is non-binding, resulting in a trade-off model that contains an additional investor-level tax penalty equal to the tax rate on equity distributions. Intuitively, when an unconstrained firm adds additional debt to its capital structure, the additional dollar of debt is replacing internal cash as a means to fund new projects. This differs from the Miller (1977) model in which the additional dollar of debt is replacing external equity as a means to fund new projects. Therefore, the model suggests that when an unconstrained firm adds an additional dollar of debt to its capital structure, it results in an additional dollar of excess cash that will be distributed by the firm, and thus triggers investor-level equity distribution taxes. 2 2 The underlying assumption of the model is that the excess cash must be distributed at some point in time. However, the model does not make any assumption as to the timing of the distributions (i.e., whether the distribution occurs in the current year or in the future).

12 12 When the individual tax rates on equity distributions are decreased (as in the 2003 Act), ceteris paribus, the asymmetry between the individual tax rates on debt distributions and equity distributions increases, which increases the personal tax disadvantage of debt. Therefore, relative to debt, equity becomes a more attractive means of financing new projects. Dhaliwal, Erickson, and Krull (2007) and Lin and Flannery (2013) both find evidence of tax sensitive firms decreasing their debt use in response to the individual tax rate cuts of the 2003 Act. However, in the case of unconstrained firms, the additional investor-level tax penalty associated with using debt decreases when individual tax rates on equity distributions are decreased. This leads to the preference for equity financing versus debt financing in the post-2003 Act period to be more pronounced for constrained firms relative to unconstrained firms. Consistent with this conjecture, Dai, Shackelford, Zhang, and Chen (2013) find that constrained firms realize larger decreases in their cost of equity capital than unconstrained firms around the 2003 Act. However, the 2003 Act not only reduced the individual tax rate on equity distributions, it also reduced the individual tax rate on interest income, which also influences the Miller (1977) gains from leverage equation. When the individual tax rate on interest income is decreased, ceteris paribus, the asymmetry between the individual tax rate on equity distributions and on interest income decreases, leading to an increase in gains from leverage. In summary, the reduction in the individual tax rates on equity distributions decreases gains from leverage, but to a greater extent for constrained firms than for unconstrained firms and the reduction in the individual tax rate on interest income increases the gains from leverage uniformly for constrained and unconstrained

13 13 firms. Therefore, the magnitudes of the two countervailing effects must be considered to determine whether firms will increase or decrease their debt use in the post-act period. As demonstrated in this paper, for constrained firms the decrease in gains from leverage from the reduction in individual equity distribution taxes is greater than the increase in gains from leverage from the reduction in individual interest income taxes; therefore, constrained firms are expected to decrease debt use in the post-2003 Act period. However, for unconstrained firms the decrease in gains from leverage from the reduction in individual equity distribution taxes is smaller than the increase in gains from leverage from the reduction in individual interest income taxes resulting in a positive change in gains from leverage and an expected increase in debt use in the post-2003 Act period. I use the 2003 Act as a quasi-experimental setting to test whether time-series variation in individual taxes affects capital structure decisions differently for constrained and unconstrained firms. 3 The 2003 Act is a strong setting to test whether individual taxes influence capital structure as the individual tax rate on dividends decreased from a maximum rate of 38.6 percent to 15 percent, the individual tax rate on long-term capital gains decreased from 20 percent to 15 percent, and the individual tax rate on interest income decreased from a maximum rate of 38.6 percent to 35 percent. 4 In addition, the 3 The literature on financial constraints is broad, and contains a number of different conceptual definitions of financial constraint. Hennessy and Whited (2007) say a firm can be judged to be more or less financially constrained according to two logically distinct metrics (p.1707). The first metric is a firm s need for external funds; the second is the cost of external funds. For the purposes of this paper, financial constraint is based on the first metric, whether a firm has sufficient internal capital to finance new projects. In the case that the firm does not have these internal funds, they are required to access the external capital market if they wish to invest in new projects, and are thus labeled financially constrained. As a result, financial constraint in this study deals purely with whether the firm must access the external capital market, not the ease of access to the external capital market. 4 For individuals, interest income is taxed at the ordinary income tax rate. The maximum statutory individual income tax rate bracket pre-2003 Act was 38.6 percent, and the 2003 Act reduced the top bracket to 35 percent. In addition, the second highest bracket was reduced from 35 percent to 33 percent, the third highest bracket from 30 percent to 28 percent, and the fourth highest bracket from 27 percent to 25 percent.

14 14 legislative timeline was rather short and the uncertainty of whether the bill would pass was rather high, giving firms little time or incentive to alter their capital structure before the rate changes were effective. Finally, these tax cuts were the main provisions of the 2003 Act, as opposed to many tax law changes that are minor provisions in larger pieces of legislation. Using two multivariate research designs, I find results consistent with the prediction that constrained firms decrease their debt use in the post-2003 Act period and unconstrained firms increase their debt use in the post-2003 Act period, and that this effect is concentrated among firms with high levels of individual ownership (tax sensitive). As a validation that the findings capture the effect of the 2003 Act and not inherent differences among tax sensitive constrained and unconstrained firms, I estimate the difference-in-difference model in the two year periods immediately preceding and proceeding the 2003 Act ( and ) and find that the results of the 2003 Act period do not hold in these falsification periods. Overall, this study contributes to the literature on the effect of investor-level taxes on a firm s capital structure by identifying a more nuanced relation between investorlevel taxes and capital structure decisions than identified in prior literature. In particular, building upon the analytical work of Hennessy and Whited (2005) and Kemsley (2013), this paper provides empirical evidence that financial constraints influence the relation between investor level taxes and capital structure. In addition, this study contributes to the literature that uses the 2003 Act as a setting to test how investor-level taxes influence firm decision-making. Specifically, the results of this paper suggest that not all tax sensitive firms responded to the 2003 Act by reducing debt use (Dhaliwal, Erickson, and As a result, all individuals who had taxable income above $28,400 ($56,800 married filing jointly) realized a decrease in the tax rate on interest income as a result of the 2003 Act.

15 15 Krull 2007; Lin and Flannery 2013), as unconstrained firms have a positive shift in their gains from leverage around the 2003 Act, which leads unconstrained firms to increase their debt use in the post-act period. Finally, the study is useful to policymakers as it highlights an outcome of investor-level tax rate changes. In particular, the results of this study suggest that the financial constraints a firm faces influences how it responds to investor-level tax rate changes. To the extent that a firm s choice to alter its capital structure around an investor-level tax rate change interacts with the legislatures intended objectives (e.g., increased capital investment) of the rate change, the influence of financial constraints on these capital structure decisions could either strengthen or diminish the intended outcomes of the tax rate change. Thus it is of importance for policymakers to be cognizant of the relation between financial constraints, investor-level tax rate changes, and capital structure decisions. The remainder of the paper is organized as follows. Section II reviews the literature on the relation between taxes and capital structure as well as the literature on the 2003 Act. Section III develops the hypotheses. Section IV provides an overview of the methodology used to test the hypotheses. Section V reviews the results of the tests of the hypotheses. Section VI provides additional analyses. Section VII concludes.

16 16 II. LITERATURE REVIEW A. Taxes and Capital Structure The influence of taxes and capital structure on firm value has been examined in the accounting, economics, and finance literatures for over fifty years. In a stylized model, including the absence of taxes, Modigliani and Miller (1958) posit that the value of the levered firm is equal to the value of the unlevered firm, and thus a firm s capital structure does not influence the value of the firm. 5 In their 1963 paper, Modigliani and Miller responded to the mounting criticism of the absence of taxes in the 1958 paper by conceding that corporations benefit from the deductibility of interest for corporate income tax purposes. When taxes are added to the model, the value of the levered firm is equal to the value of the unlevered firm plus the corporate debt tax shield, and therefore, capital structure does influence the value of the firm. In addition, the value of the corporate debt tax shield increases in both the corporate tax rate and the amount of debt used by the firm. However, researchers observed that many companies appear to use less debt than required to maximize the benefits of the corporate debt tax shield. 6 As a result, unless companies make inefficient decisions (i.e., leave money on the table ), the benefits from adding more debt (the corporate debt tax shield) must be offset by costs of adding additional debt. In his 1977 presidential address to the American Finance Association, Miller concluded that the corporate tax advantage of debt is primarily offset by the personal tax disadvantage of debt. The crux of his argument is that when debt is used in 5 The Modigliani and Miller (1958) model assumes the absence of taxes, bankruptcy costs, agency costs, and information asymmetry. 6 e.g., Miller (1963) finds that the mean debt/asset ratio of a nonfinancial corporation in the 1950s was similar to that of the 1920s, even though the corporate tax rate had increased four-fold.

17 17 place of equity, there are changes in the firm s payouts in the form of dividends and capital gains versus payouts in the form of interest payments. As debt use increases relative to equity, the percentage of interest payouts increases relative to dividend or capital gains payouts. Since individual taxes are higher on interest payouts than on equity payouts, investors will demand a higher pre-tax return from debt, raising the cost of debt relative to the cost of equity. Therefore, the personal tax disadvantage of debt will reduce, or even eliminate, the corporate tax advantage of debt. DeAngelo and Masulis (1980) identify another factor driving firms to appear under-levered. In their model, firms have varying amounts of non-debt tax shields, such as depreciation expense and investment tax credits, that will reduce corporate income taxes. Given that the benefit to the corporation of using debt is contingent on having taxable income, when the company has already reduced taxable income using non-debt tax shields, the benefits of using additional debt decrease. Consistent with this prediction, they find that debt use is lowest in industries with the highest amounts of non-debt tax shields. Mackie-Mason (1990) extends the tax shield literature by clarifying the relationship between non-debt tax shields and the incentives to use debt. In particular, he determines that non-debt tax shields should only decrease the incentives to use debt if the non-debt tax shields lower the firm s marginal tax rate. By focusing on a non-debt tax shield that is expected to influence the marginal tax rate (tax loss carry-forwards) and a non-debt tax shield that is not expected to influence the marginal tax rate (investment tax credits), he finds that higher tax loss carry-forwards decrease the probability of new debt

18 18 issuance while higher investment tax credits do not. 7 This study is one of the first to identify a positive relation between incremental debt use and taxes, a finding that had eluded other researchers (Myers 1984; Bradley, Jarrell, and Kim 1984; Titman and Wessels 1988; Smith and Watts 1992; Gaver and Gaver 1993). In addition, Trezevant (1992) corroborates the findings of DeAngelo and Masulis (1980) (substitution effect) and Mackie-Mason (1990) (tax exhaustion hypothesis) around the Economic Recovery Tax Act of 1981 which increased investment tax shields. Since incremental debt policy should be associated with marginal tax rates, as suggested by Mackie-Mason (1990), researchers have estimated firm specific marginal tax rates to develop cleaner tests of how taxes influence firms investment, financing, and compensation decisions. Shevlin (1990) recognizes that a firm s marginal tax rate in the current year is function of the firm s taxable income in prior, present, and future years. As a result, to effectively estimate current year marginal tax rates, researchers must estimate future taxable income for the firm. Making use of a firm s current net operating loss positions and estimates of the firm s future taxable income, Shevlin estimates the present value of a change in taxes payable from earning an extra dollar of taxable income. 8 Graham (1996) improves Shevlin s (1990) estimation process by incorporating the effects of investment tax credits and the alternative minimum tax into the model. Graham (1996) also greatly expands the sample used in Shevlin (1990) by estimating marginal tax rates for over 10,000 COMPUSTAT firms from In addition to simulating the marginal tax rates, Graham (1996) also identifies a positive relation 7 Tax credits and deductions only influence a firm s marginal tax rate to the extent that they move firms into a lower tax bracket, which for corporations, only occurs very near or below zero taxable income. (Mackie-Mason 1990). 8 Shevlin (1990) uses a random walk model with a drift term to estimate the values of future taxable income.

19 19 between the firm s marginal tax rate and incremental debt policy. Using these marginal tax rates, Graham (2000) estimates that 44 percent of firms could increase their debt twofold and still receive the full tax benefit from interest deductions. Blouin, Core, and Guay (2010) provide the latest improvement to the marginal tax rate estimation process. They argue that the random walk forecasting approach employed in Shevlin (1990) and Graham (1996) produces estimates of future income that are too high for high income firms and too low for low income firms, as well as severely underestimating the volatility of future income. As a result, Blouin, Core, and Guay (2010) employ a non-parametric simulation technique to forecast future income and find that only 11 percent of firms could increase their debt by 200 percent and still be able to capture tax benefits at the top statutory tax rate. The trade-off model in Miller (1977) focuses on a firm s trade-off between the corporate benefits of debt and the investor-level penalty of debt to identify whether firms can increase firm value through incremental debt use. 9 In the case where the firm cannot increase firm value by increasing debt, it will opt to finance new investments through external equity issuance. However, Hennessy and Whited (2005) argue that not all firms make a choice between debt and equity to finance new projects. In particular, firms with high liquidity have the option to finance new projects with internally generated funds. Accordingly, Kemsley (2013) incorporates an internal budget constraint into the Miller (1977) model that is binding when the firm cannot finance new investments internally (i.e., constrained firms) and non-binding when firms can finance new projects completely 9 Miller s (1977) trade-off model differs from the earlier trade-off theory work that models the trade-off between the tax benefits of debt and the costs of bankruptcy (e.g. Kraus and Litzenberger 1973). Miller (1977) argues that costs of bankruptcy, in expectation, are too small to offset a sizable portion of the tax benefits of debt.

20 20 with internal funds (i.e., unconstrained firms). This is consistent with the findings of Korajczyk and Levy (2003) who recognize that firms facing financial constraints do not choose capital structure in the same manner as unconstrained firms. In particular, they find that highly profitable firms are more likely to use internally generated funds rather than debt to finance new investments, and in the case that they eventually do access the external capital markets, these highly profitable firms are more likely to issue new debt than new equity. 10 In summary, starting with Modigliani and Miller (1958), taxes have been a focal point of the research surrounding the determinants of firms capital structure decisions. While the deductibility of interest payments for corporate income tax purposes provides a benefit to the firm, the use of debt also has costs. These costs include such things as agency costs (e.g., Jensen and Meckling 1976) and the expected costs of bankruptcy (e.g., Kraus and Litzenberger 1973). However, Miller (1977) argues that the primary cost that offsets the corporate tax benefit of debt is the disadvantageous treatment of interest payments to individual investors compared to the tax treatment of equity distributions. Accordingly, if personal tax rates do influence the magnitude of the personal tax disadvantage of debt, then changes in these rates should elicit changes in firm capital structure. B. Jobs and Growth Tax Relief Reconciliation Act of 2003 The Jobs Growth Tax Relief Reconciliation Act of 2003 (the 2003 Act) represents one of the largest shifts in individual income tax policy in the past twenty-five years. In addition, the relatively short legislative timeline and close votes in both the House and 10 The modeling of the capital structure decisions of unconstrained firms in Kemsley (2013) as well as the results in Korajczyk and Levy (2003) aligns with the pecking order theory (Myers and Majluf 1984).

21 21 the Senate provided very little certainty as to whether the rates would eventually be cut, which provides researchers with a quasi-experimental setting. 11 As a result, the 2003 Act spurred research investigating the impact of taxes on firm decision-making and on tax capitalization. Theory suggests that investor-level taxes should influence the cost of equity capital (e.g., Modigliani and Miller 1958; Modigliani and Miller 1963; Brennan 1970; Miller 1977). Therefore, when the tax rates on dividends and capital gains are decreased, as in the 2003 Act, the cost of equity capital should decrease because investors can now earn the same after-tax return on a lower pre-tax return. Consistent with theory, Dhaliwal, Krull, and Li (2007) find that the 2003 Act induced a 1.02 percent average decrease in the cost of equity capital, and that this decrease was concentrated among firms with lower levels of institutional ownership (tax sensitive). Dai, Shackelford, Zhang, and Chen (2013) extend Dhaliwal et al. (2007b) by testing whether the elasticity of demand for external equity capital influences the magnitude of the reduction in the cost of equity capital when investor-level taxes are decreased. In particular, Dai et al. (2013) argue that financially constrained firms will have an inelastic demand for external equity capital relative to unconstrained firms and will therefore realize a larger reduction in the cost of equity capital when investor-level taxes are cut. Using the Hadlock and Pierce (2010) definition of financial constraints, Dai et al. (2013) estimate that the average firm s cost of equity capital will be twenty basis points lower after the 2003 Act, 11 The 2003 Act passed both the House ( ) and the Senate (50-50 with Vice President Cheney breaking the tie with a yea vote) on May 23, 2003 and was signed into law by President George W. Bush on May 28, The tax rate cuts on dividends were effective retroactively to January 1, 2003, and the rate cuts on capital gains were effective on sales of stock after May 6, 2003.

22 22 and that financially constrained firms realize an additional four basis point decrease in the cost of equity capital. One outcome of a lower cost of equity capital after the 2003 Act is that the cost of equity capital relative to the cost of debt decreases, making financing through equity more attractive and financing through debt less attractive than in the pre-2003 Act period. Dhaliwal, Erickson, and Krull (2007) test this conjecture by studying new debt and equity issuance in the pre- and post-2003 Act periods. They find that the 2003 Act decreased the probability of a firm issuing new debt relative to new equity, and this effect is mitigated when the firm has higher institutional ownership. These findings are corroborated by Lin and Flannery (2013) who find that firms with high individual ownership decrease leverage ratios by five percentage points, on average, from the pre- to post-2003 Act period. In addition, they find that firms with poor access to external capital markets are not able to readily respond to the tax cuts and therefore do not reduce their leverage ratios as much. My study directly extends this line of research by identifying a subset of firms (unconstrained firms) that do not decrease their debt use around the 2003 Act, but instead increase their debt use. Given that decreasing the investor-level tax rate on dividends increases the aftertax cash flow investors receive from dividends, many researchers have studied whether dividend behavior changed around the 2003 Act. Chetty and Saez (2005) find that dividend payments increased twenty percent after the 2003 Act for their sample of nonfinancial, nonutility, public firms. They conclude that the effect is driven by principal-agent issues as firms with large holdings by executives, particularly firms whose executives hold a large number of shares but a low number of options, were most

23 23 likely to increase dividends. Chetty and Saez (2006) follow their 2005 study by examining whether the increase in dividends continued into 2005, and were not just a one-time increase associated with the tax cut. They find the higher levels of dividend payments observed around the 2003 Act are part of a longer-term shift in dividend policy. In addition, they observe an increase in special dividends and a spike in dividend initiations after the 2003 Act. Taken together, the Chetty and Saez (2005, 2006) findings align with survey evidence in Brav, Graham, Harvey, and Michaely (2008) who find that while dividend taxes are not a first order determinant of dividend policy, they are often a factor considered by firms in forming their payout policy. 12 In addition, Hanlon and Hoopes (2013) study the anticipated January 1, 2011, expiration of the 2003 tax cuts. Consistent with the studies investigating payout policy around the 2003 Act, they identify a surge in special dividends at the end of 2010 as well as the shifting of regular dividends from January 2011 to December 2010 in anticipation of higher dividend tax rates. Also, consistent with prior studies, they find this effect is particularly strong among firms with high executive ownership. Similarly, Hribar, Savoy, and Wilson (2013) document a pattern of dividend shifting and special dividends around the anticipated January 1, 2013, expiration of the 2003 tax cuts, and find that investors react positively to these events. While Chetty and Saez (2005, 2006) focused only on dividend policy, Brown, Liang, and Weisbrenner (2007) incorporate stock repurchases into their research design to study the effect of the 2003 Act on the total payout policy of the firm. Consistent with Chetty and Saez (2005, 2006) they find an increase in dividend payments in the post Act period, and this increase is particularly prevalent among firms with high 12 The Brav et al. (2008) results are based on a survey of 384 financial executives and additional in-depth interviews with 23 of those financial executives percent of the executives surveyed said that dividend taxes are important or very important to their company s dividend decisions.

24 24 executive ownership. However, they also find that about one-third of the firms that initiated new dividends after the 2003 Act also reduced share repurchases by an amount sufficient to reduce total payouts to shareholders. Blouin, Raedy, and Shackelford (2011) also study how managers alter their dividend-repurchase mix in response to the 2003 Act. They find that managers of firms with disproportionately large individual investor bases (tax sensitive) modify their payout policies by increasing the dividend portion of their total payouts. In addition, since the preference for dividends, ceteris paribus, by individual investors increased due to the 2003 Act, wealth-maximizing individual shareholders will rebalance their portfolios to capture higher dividend yields. Consistent with this conjecture, Blouin, Raedy, and Shackelford (2011) observe that insiders simultaneously increase their holdings in a stock when the firm shifts its payout policy more toward dividends; but the authors do not find this effect for individual investors in general. However, using Survey of Consumer Finances data, Kawano (2014) finds that households in the highest individual income tax bracket increase their portfolio dividend yields by 23 percent between 2001 and 2004, which is 13 percent more than the households in the second highest individual income tax bracket. By forming hypothetical portfolios of firms, she determines that this effect is not only driven by the increase in dividend payments by firms (e.g., Chetty and Saez 2005, 2006), but also active portfolio rebalancing toward stocks with higher dividend yields. Finally, one of the stated goals of the 2003 Act was to increase corporate capital expenditures. Campbell, Chyz, Dhaliwal, and Schwartz (2013) make two tax-motivated predictions associated with increases in investment post-2003 Act. First, firms with more

25 25 tax sensitive investor bases are most impacted by the 2003 Act; therefore, increases in capital expenditures should be larger for firms with largely tax sensitive investor bases. Second, they predict that the 2003 Act will spur a larger increase in investment for firms that finance investments with new equity issuance rather than internal funds as these firms receive a greater benefit from the cost of equity capital reductions associated with the 2003 Act (Dhaliwal et al. 2007b). Consistent with their predictions, they find that investment increases between 8.5 and 10.2 percent. However, when the percentage of the investor base that is tax sensitive increases by one standard deviation the increase in post Act investment rises to between 17.7 and 19.7 percent. In addition, when the firm is likely to finance its new investments by issuing external equity versus using internal funding, the increase in post-2003 Act investment rises to between 18.2 and 19.1 percent. In summary, the 2003 Act has spurred a significant stream of literature on the influence of investor-level tax rates on corporate decision-making. In particular, the large individual tax rate shifts of the 2003 Act provided a rich setting to corroborate the findings of earlier studies on whether investor-level taxes impact capital structure. For example, Dhaliwal et al. (2007a), Dai et al. (2013), and Lin and Flannery (2013) all provide evidence consistent with the hypothesis that investor-level taxes are a determinant of firm capital structure decisions. In addition to capital structure, researchers have also determined the investor-level tax changes of the 2003 Act influenced firm payout policy (e.g., Chetty and Saez 2005; Blouin et al. 2011), investor portfolio composition (e.g., Blouin et al. 2011; Kawano 2014), corporate capital expenditures (e.g., Campbell et al. 2013), and reduced the cost of equity capital for firms (e.g., Dhaliwal et al. 2007b).

26 26 III. HYPOTHESES DEVELOPMENT Miller (1977) explores the role of both corporate taxes and investor-level taxes on the use of debt within the firm. In particular, he argues that the corporate tax advantage of debt, which arises through the deductibility of interest payments for corporate tax purposes, may be offset by the personal tax disadvantage of receiving debt interest payments instead of equity distributions. 13 The takeaway from his model is that gains from leverage are positive as long as the net debt tax shield (corporate tax benefit from debt less the personal tax disadvantage from debt) is greater than zero. Miller s gain from leverage is specified as follows: G = r (1 t ) (1 t ) 1 t, (1) where: r is the pretax return, t i is the individual tax rate on interest income, t c is the corporate income tax rate, and t p is the individual tax rate on equity distributions. The equation can be rearranged as: G = r t t (1 t )t. (2) Here t c is the gross corporate tax advantage of debt and t (1 t )t is the personal tax disadvantage of debt. The gain from leverage can be expressed in terms of the capitalized present value of a perpetuity. Miller (1977) uses the discount rate of r(1-t i ). Thus the capitalized value is: 13 It is worth noting that the trade-off models assume that a firm issues a single class of debt, and are thus focused on the optimal amount of debt, not the structure of debt (i.e., bank debt versus market debt). Hackbarth, Hennessy, and Leland (2007) examine the optimality of bank debt versus market debt and determine that firms have a preference for bank debt as banks can engage in a private negotiation process with the firm in the event of financial distress, while public debt requires a costly bankruptcy process to renegotiate terms. In the case of firms that have high bargaining power in private negotiation, the optimal level of debt cannot be secured through only bank debt, so they will also enter the public debt market to reach their target level of total debt. In addition, for firms to maximize the amount of bank debt, it is placed senior to market debt. Consistent with this, Rajan and Zingales (1995) find that bank debt is a lower portion of total debt in countries where absolute priority violations during bankruptcy are common.

27 27 G : = ( ). (3) ( ) The gain from leverage in (3) specifies the personal tax disadvantage of debt when comparing debt versus equity issuance. However, corporations that have ample internal cash have a third option for financing new projects. Instead of choosing between debt and equity, firms with large cash reserves can finance new projects through debt, equity, or internal financing. Kemsley (2013) adapts the Miller (1977) model for cash unconstrained firms to include the choice to finance new projects internally. In particular, for firms that are not cash constrained, the financing decision is not between debt and equity, but is instead between debt and internal financing. This is consistent with the Pecking Order Theory in which firms prefer internal over external financing, and when accessing external capital markets, firms prefer debt over equity (Myers 1984; Myers and Maluf 1984), and with the empirical findings in Korajczyk and Levy (2003) who find that that highly profitable firms are more likely to use internally generated funds rather than debt to finance new investments, and in the case that they eventually do access the public capital markets, these highly profitable firms are more likely to issue debt than new equity. In addition, if external financing allows the firm to increase distributions around the 2003 Act, as prior research suggests, the unconstrained firms will have a preference for debt as issuing new external equity is in essence a negative distribution to current shareholders because it dilutes current share ownership and would therefore offset the

28 28 increased distributions that firms seek to deliver to shareholders. 14 Kemsley adapts formula (1) as follows: G = r 1 t (1 t ) (1 t ) 1 t. (4) Comparing constrained firms (1) and unconstrained firms (4), it can be seen that the gain from leverage for the unconstrained firm is smaller than for the constrained firm by a factor of (1-t p ), Kemsley (2013) defines this new term as a personal tax toll investors must pay to use debt financing in lieu of internal equity financing (p. 18). Intuitively he explains it as a cash unconstrained firm having the choice to either use internal cash to finance new projects or to pay distributions to shareholders through dividends or share repurchases. Therefore, if a cash unconstrained firm decides to use debt to finance a new project; it will have excess cash to pay dividends or repurchase, which leads to an equity distribution tax for the investor. This is consistent with Hennessy and Whited (2005) who note, that each dollar of debt issued by this high liquidity firm would serve to increase the distribution to shareholders, rather than replacing external equity (p. 1130). It is important to note that the increase in distributions will not necessarily occur in the current period, but that the increased levels of internal cash will lead to higher distributions either in the current period or in the future. Rearranging (4) is equal to: G = r t t (1 t )t (1 t )t. (5) 14 Accordingly, Fama and French (2005) state that If asymmetric information means stock must be issued at below equilibrium value, issuing stock to pay dividends decreases the wealth of current shareholders (p.560). In addition, Asquith and Mullins (1986) find that on average abnormal returns are positive for dividend initiations and subsequent increases and negative for equity issue announcements.

29 29 Comparing constrained firms (2) to unconstrained firms (5) it can be seen that the personal tax disadvantage of debt is larger in (5) than in (2), as a result, the larger the term (1 t )t becomes, the more likely unconstrained firms are to finance internally instead of using debt. Therefore, as t p decreases (as in the 2003 Act), ceteris paribus, the attractiveness of debt compared to internal financing increases. In addition, as t i decreases (as in the 2003 Act), ceteris paribus, the attractiveness of debt compared to internal financing decreases. As in Miller (1977), Kemsley (2013) expresses (5) as a capitalized present value of a perpetuity, this results in the following formula: G : = ( ) ( ) t. (6) As seen in equation (6), unconstrained firms should use debt until the net marginal debt tax shield is equal to the personal tax rate on equity distributions (t p ), which differs from equation (3) in which the constrained firms should use debt until the net marginal debt tax shield is equal to zero. Since the 2003 Act changed both the personal tax rate on equity distributions and the personal tax rate on interest income, the partial derivatives of the gains from leverage formula with respect to both t p and t i need to be considered. The effect of a change in the personal tax rate on equity distributions on the gain from leverage is obtained by taking the partial derivate of the gain from leverage equation with respect to t p. 15 Therefore, the effect of a change in t p on the gain from leverage for constrained firms (the partial derivative of (3) with respect to t p ) is: : = ( ) ( ), (7) and the effect of a change in t p on the gain from leverage for unconstrained firms (the partial derivative of (6) with respect to t p ) is: 15 Step-by-step partial derivation is provided in Appendix A

30 30 : = ( ) 1. (8) ( ) Similarly, the effect of a change in the personal tax rate on interest income on the gain from leverage is obtained by taking the partial derivative of the gain from leverage equation with respect to t i. The effect of a change in t i on the gain from leverage is uniform across constrained and unconstrained firms. The partial derivative of (3) and (6) with respect to t i is: : = : = ( ) ( ). (9) The total derivative of the gains from leverage function can be used to make predictions about the impact of the 2003 Act on the gains from leverage, and how that effect varies across financially constrained and unconstrained firms. The total derivative is of the following form: dg =, dt +, dt +, dt. (10) There is no change in t c as part of the 2003 Act; therefore, the last term of the total derivative is zero and can be disregarded for the purposes of making predictions about the impact of changes in t p and t i on the gains from leverage. Substituting equations (7), (8), and (9) into the total derivative (10) results in: dg : = ( ) ( ) t : t : + ( ) ( ) (t : t : ), (11) and dg : = ( ) ( ) 1 t : t : + ( ) ( ) (t : t : ). (12) The 2003 Act reduced the rate on dividend income from a maximum rate of 38.6 percent to 15 percent, the rate on long-term capital gains from 20 percent to 15 percent,

31 31 and the tax rate on interest income from a maximum rate of 38.6 percent to 35 percent. The following table examines the effects of the 2003 Act on a firms decision to finance using debt, equity, or internal financing. For the analyses, assume a corporate tax rate of 35 percent and a pre-2003 Act tax rate on equity distributions of 29.3 percent, which represents the average of the dividend and the long-term capital gains tax rate in the pre Act period. 16 Table 1: Gains from Leverage: Pre- and Post-2003 Act 17 Constrained Unconstrained G L Pre 2003 Act G L Post 2003 Act Change. (. (. ). ) (. ). (. (. ). ) (. ) = =. (. (. ). ) (. ). (. (. ). ) (. ) = = Consistent with Hennessy and Whited s (2005) conjecture that the marginal increase in debt is more attractive when it serves as a replacement for external equity and is less attractive when it finances an increase in distributions to shareholders (p. 1130), the gains from leverage are smaller for the unconstrained firms in both the pre- and post Act periods. However, the tax rate decreases of the 2003 Act result in different post-2003 Act changes in the gains from leverage for constrained versus unconstrained firms. In particular, for constrained firms, with a tax rate of 35 percent, the tax cuts decrease gains from leverage by , indicating that the preference for debt over 16 The 29.3 percent rate assumes that 50 percent of earnings are eventually returned to shareholders as dividends and 50 percent are eventually returned to shareholders through capital gains. It should be noted that the change in gains from leverage from the pre-2003 Act to the post-2003 Act period is still negative for constrained firms and still positive for unconstrained firms at the endpoints of the dividend payout spectrum where 100 percent of earnings are eventually paid out as dividends (which would result in a pre Act rate on equity distributions of 38.6 percent) and where zero percent of earnings are eventually paid out as dividends (which would result in a pre-2003 Act rate on equity distributions of 20 percent). Therefore, this assumption does not influence the hypotheses, but only serves to aid in the quantification of the shifts in the gains from leverage around the 2003 Act in Table Appendix B ties the changes documented in Table 1 using the gains from leverage formula to the total derivatives in (11) and (12).

32 32 equity decreases for these firms in the post-2003 Act period. Dhaliwal, Erickson, and Krull (2007) and Lin and Flannery (2013) document decreases in new debt issuance and leverage ratios respectively in the post-2003 Act period, consistent with the decrease in gains from leverage post-2003 Act. However, when the rates on dividends, long-term capital gains, and interest income are decreased, the gains from leverage increase by Therefore, in the post-2003 Act period, the unconstrained firms are incentivized to add debt to their capital structure. As Table 1 suggests, the 2003 Act will result in a decrease in gains from leverage for constrained firms, and an increase in gains from leverage for unconstrained firms. However, within the constrained and unconstrained groups, the effect of the tax cut on the gains from leverage will be larger for some firms than for others. Tax clientele theory predicts that the impact of investor-level tax rate changes on a given firm will vary based on the tax rate faced by the firm s investors (e.g., Miller and Modigliani 1961; Scholes et al. 2009). Equations (3) and (6), and therefore the calculations in Table 1, assume that all investors are taxed at personal tax rates. However, this assumption does not generalize well to the current state of the economy where, in addition to individual investors, there are institutional investors that are tax-exempt (such as endowment funds or pensions) and institutional investors that are taxed at corporate rates (such as banks and insurance companies). Consistent with tax clientele theory, prior research finds that firms that are largely owned by institutional investors, and are therefore tax-exempt or tax-favored, are less affected by investor-level tax rate changes (e.g., Ayers, Cloyd, and Robinson 2002; Ayers, Lefanowicz, and Robinson 2004, 2007; Chetty and Saez 2005; Dai, Shackelford, Zhang, and Chen 2013; Dhaliwal, Li, and Trezevant 2003; Dhaliwal, Krull, Li, and

33 33 Moser 2005; Dhaliwal, Krull, and Li 2007; Dhaliwal, Erickson, and Krull 2007; Lin and Flannery 2013; Moser and Puckett 2009). The assumption here is that firms which are largely owned by individual investors will be more sensitive to changes in individual tax rates. For the purposes of this study I will refer to these firms as tax sensitive. Therefore, if the 2003 Act incentivizes constrained firms to decrease their debt use and unconstrained firms to increase their debt use, then this effect should be particularly salient among firms with a more tax sensitive investor base. These observations lead to the following hypotheses (stated in the alternative): H1: Tax sensitive constrained firms decrease debt use around the 2003 Tax Act. H2: Tax sensitive unconstrained firms increase debt use around the 2003 Tax Act. Figure 1 provides a stylized visual representation of the hypotheses. MC is the marginal cost of debt function, which increases in the total amount of debt used by the firm. The four horizontal lines represent the marginal benefit of debt under different scenarios. First, the marginal benefit of debt differs based on whether the additional dollar of debt is replacing an additional dollar of equity (constrained firms) or leading to an increase in distributions to shareholders (unconstrained firms). 1+γ represents the marginal benefit of debt for constrained firms, where the additional dollar of debt replaces additional equity;; therefore, γ is the amount by which the cost of equity exceeds the cost of debt. The constrained firm will then issue debt until the point that the marginal cost of debt equals the marginal benefit, leading to debt use of amount C. On the other hand, 1-t p represents the marginal benefit of debt for unconstrained firms, where the additional dollar of debt leads to an increase in distributions to shareholders. Accordingly, t p is the additional tax shareholders will bear from an increase in

34 34 distributions. Again, the unconstrained firm will use debt until the marginal cost equals the marginal benefit, leading to debt use of amount U. The tax cuts of the 2003 Act influence the marginal benefit of debt for both constrained and unconstrained firms. As shown in Dhaliwal et al. (2007b) and Dai et al. (2013) the tax cuts of the 2003 Act reduced the cost of equity capital for firms. As a result, the spread between the cost of debt and the cost of equity is smaller in the post Act period than in the pre-2003 Act period (i.e., γ < γ). This decreases the marginal benefit of debt for constrained firms, shifting debt use for constrained firms from C to C. In Figure 1, C is less than C, consistent with H1 which posits that constrained firms will decrease their debt use in the post-2003 Act period. For unconstrained firms, the level of the marginal benefit of debt is dependent on t p, which is the personal tax rate on distributions. Since the 2003 Act reduced the tax rate on dividends and long-term capital gains (i.e., t p < t p ), the marginal benefit of debt for unconstrained firms will increase. This leads to debt use for unconstrained firms shifting from U to U. According to the figure, U is greater than U, leading to H2 that unconstrained firms will increase their debt use in the post-2003 Act period.

35 35 A. Regression Models IV. RESEARCH DESIGN To test hypotheses one and two, the following multivariate models are estimated. The first model is estimated on tax sensitive and non-tax sensitive sub-samples separately. Model two employees a difference-in-differences design to test the full sample of both tax sensitive and non-tax sensitive firms in a single model. 18 Debt = β + β Unconstrained + β Constrained + β MTR + β CF + β Momentum + β MB + β Ln(Assets) + β Ln(Assets) + β Col + β MB + β Intang + β CF + β Ddiv + δ IndustryFE + ε (13) Debt = β + β Unconstrained + β Constrained + β HighIndOwn + β Unconstrained x HighIndOwn + β Constrained x HighIndOwn + β MTR + β CF + β Momentum + β MB + β Ln(Assets) + β Ln(Assets) + β Col + β MB + β Intang + β CF + β Ddiv + δ IndustryFE + ε (14) B. Variable Definitions Debt is defined as the 2002 to 2004 change in the book value of long-term debt deflated by the lagged market value of the firm, which is defined as the book value of debt plus the market value of equity. This change measure is suggested by Graham (1996) to investigate changes in debt policy, and the two-year change from 2002 to 2004 is consistent with Lin and Flannery s (2013) investigation of changes in leverage ratios around the 2003 Act. For robustness, only observations with Debt it >.02 are included in the analysis (Graham 1996), as it is more likely to capture firms that are making intentional changes in debt policy. 18 The primary difference between the two models is that (13) is estimated on tax sensitive and non-tax sensitive firms separately, while (14) is estimated with the full sample of firms. The implicit assumption then in (14) is that the influence of the control variables on incremental debt use is uniform across tax sensitive and non-tax sensitive firms. It is not clear whether this is the case or not, thus consistent results across the two models provide stronger support for the hypotheses.

36 36 To form the constrained and unconstrained sub-samples, financial constraint is measured using the KZIndex, described in Lamont, Polk, and Saa-Requejo (2001) based on the work of Kaplan and Zingales (1997). I use the KZIndex instead of the measures for financial constraint in Hadlock and Pierce (2010) or Lin and Flannery (2013) because the KZIndex more closely aligns with the budget constraint modeled in Kemsley (2013). In particular, the KZIndex is a relative measure of reliance on external financing (Hennessy and Whited 2007) while many of the other constraint measures capture a dimension of financial constraint that is more focused on the ease of access to the capital markets. Following is the five-factor model used to calculate the KZIndex: KZIndex = ( ) ( ) + ( ) ( ) ( ) (15) The variables in equation (11) are defined in COMPUSTAT as follows: IB is income before extraordinary items, DP is depreciation and amortization, PPENT is net property, plant, and equipment, AT is total assets, CEQ is common equity, TXDB is deferred taxes reported on the balance sheet, DLTT is long-term debt, DLC is debt in current liabilities, SEQ is stockholder s equity, DVC is common dividends, DVP is preferred dividends, CHE is cash and short-term investments, and MktEq is common shares outstanding (CSHO) times share price at calendar year end (PRCC_C). The dichotomous variables Unconstrained and Constrained are coded one for firms identified as unconstrained and constrained respectively based on the KZIndex. The KZIndex is measured in This 19 The cut-offs for unconstrained and constrained are based on using the upper quartile (KZIndex 1.29 = constrained) and lower quartile (KZIndex = unconstrained) of all domestic firms besides utilities and financial firms in the 2002 COMPUSTAT universe with the necessary data to calculate the KZIndex. The results are robust to using the top three deciles of this paper s 1,964 observation sample (KZ Index.975 = constrained) and lower three deciles of this paper s 1,964 observation sample (KZIndex =

37 37 results in a sample with three constraint groups, financially constrained firms, financially unconstrained firms, and a control group of firms that are neither financially constrained nor unconstrained. To differentiate between tax sensitive and non-tax sensitive firms, HighIndOwn is coded one if the firm has individual ownership above the mean for the sample, and zero otherwise based on quarterly institutional ownership data from the Thomson-Reuters Institutional Holdings (13F) Database. 20 HighIndOwn is then interacted with the dichotomous variables of Unconstrained and Constrained to isolate the effect that financial constraints have on changes in debt use for the tax sensitive group. 21 The results are robust to an above/below median variable definition of HighIndOwn. The control variables are consistent with Lin and Flannery (2013), and are known determinants of changes in debt and the cost function of debt (van Binsbergen, Graham, and Yang 2010). All levels control variables are measured in 2002, and the change control variables are changes from 2002 to 2004, consistent with the measurement period of the dependent variable. CF is cash flow defined as operating income before depreciation (OIBDP) scaled by total assets. Momentum is the firm s buy-and-hold return in the 12 months before the end of the fiscal year. MB is market-to-book defined as market value of equity over book value of equity. Ln(Assets) is the natural log of total assets. COL is collateralized assets defined as the sum of inventories and property, plant, unconstrained), which is the methodology used to divide samples into constrained/unconstrained groups in Almeida et al. (2004). 20 The SAS program written by Denys Glushkov, Rabih Moussawi, and Luis Palacios from WRDS is used to extract ownership data from 13F. The program is available at 21 Blouin, Bushee, and Sikes (2011) identify a set of institutional investors that are also tax sensitive. In addition, Chyz and Li (2012) find that tax sensitive and non-tax sensitive institutional investors respond differently to the capital gains tax rate reduction of the Taxpayer Relief Act of The inclusion of any tax sensitive institutions in my LowIndOwn group would introduce bias against finding a difference-indifferences response to the 2003 Act.

38 38 and equipment scaled by total assets. Intang is intangible assets scaled by total assets. Ddiv is equal to one if the firm paid a dividend, zero otherwise. Firms are grouped into industries using the Fama and French 17 industry portfolios and industry fixed effects are included in the models. 22 Model two is a difference-in-differences (DID) model in which the intercept term measures the change in debt from 2002 to 2004 for the group of firms that are identified as neither constrained nor unconstrained. The coefficients on the variables Constrained and Unconstrained represent changes in debt for the group of constrained firms and unconstrained firms relative to the intercept. Finally, the coefficients of interest are the interaction terms Constrained x HighIndOwn and Unconstrained x HighIndOwn. These represent the change in debt for the group of constrained firms with high individual ownership and the change in the debt for the group of unconstrained firms with high individual ownership relative to those constrained and unconstrained firms without high individual ownership, respectively. Consistent with Blouin, Raedy, and Shackelford (2011) and Lin and Flannery (2013), the DID model is also estimated over time periods in which there were no investor level tax rate changes (i.e., falsification periods). Therefore, model two is estimated over the period as well as the period. The expectation is that the interaction terms Constrained x HighIndOwn and Unconstrained x HighIndOwn will be insignificant in these falsification periods. C. Sample The primary results use data from 2002 to 2004, specifically the changes in dependent and independent variables between 2002 and As a result, there is only 22 The industry portfolio classification schemes are available on Ken French s data website.

39 39 one observation per firm in the sample. The sample selection procedure is documented in Table 2. An initial sample of 6,124 observations is obtained by merging the 2002 COMPUSTAT universe with the Blouin, Core, and Guay (2010) marginal tax rate database available through WRDS. Next, to calculate the variables for the two regression models, data from 2001 as well as 2004 is required; this reduces the sample to 5,192 observations. Consistent with prior studies on the impact of taxes on capital structure, financial firms and regulated utilities are excluded from the sample, resulting in 4,798 observations. Because firms must be classified as constrained or unconstrained, variables required for the KZIndex (equation 15) cannot be missing. This restriction reduces the sample to 2,759 observations. Excluding firms with missing values of the dependent and independent variables in the two regression models reduces the sample to 2,463 observations. To minimize the influence of extreme observations, the variables Debt, CF, Momentum, MB, MB, and KZIndex are truncated at the 1 st and 99 th percentiles, reducing the sample to 2,223 observations. 23 Finally, as institutional ownership data is required to identify those firms that are expected to be affected by the 2003 Act, matching to the Thomson-Reuters Institutional Holdings (13F) Database is necessary, this restriction results in the final sample size of 1, The results are robust to winsorizing at the 1 st and 99 th percentiles instead of truncating 24 For comparison, Dhaliwal et al. (2007a) have 1,831 observations in their 2003 Act tests and Lin and Flannery (2013) have 3,034 observations in their main 2003 Act test. However, Lin and Flannery (2013) do not require the data necessary to calculate the KZIndex, which is the largest data restriction in my sample. In addition, the total market capitalization of the sample at the end of 2002 was $4.292 billion, given that the end of 2002 market capitalization of the S&P 500 was $7.891 billion (before excluding financials and utilities, which are excluded from the sample in this study) the sample represents an economically meaningful portion of the economy.

40 40 V. RESULTS A. Descriptive Statistics Descriptive statistics for the sample are reported in Table 3. As expected, the mean 2002 to 2004 change in debt differs between the constrained and unconstrained samples ( and respectively). This difference is statistically significant at p< The difference is larger in the intentional debt change sample, with mean changes in debt of and for the constrained and unconstrained sample respectively. By design, the KZIndex is statistically different in the unconstrained compared to the constrained subsamples. Furthermore, in both the full sample and the intentional debt change sample, constrained firms are more likely (p< 0.01) to have high individual ownership (i.e., above the full sample mean) than unconstrained firms and are thus more likely to be tax sensitive. In addition, unconstrained firms have higher marginal tax rates (p< 0.01), higher cash flows (p< 0.01), higher market-to-book (p< 0.01), larger changes in assets (p< 0.01), larger changes in collateralized assets (p< 0.01), larger changes in intangible assets (p< 0.01), and are more likely to pay dividends (p< 0.01) than the constrained firms. Many of these variables also influence a firm s choice to use debt financing; therefore, it is necessary to conduct multivariate analyses to confirm the univariate differences in change in debt for the constrained and unconstrained firms are due to the rate changes of the 2003 Act. Table 4 presents the Pearson and Spearman correlations for the variables in the regression models. The Pearson correlation for Unconstrained and Debt is and the correlation for Constrained and Debt is Both of these correlations are statistically significant, which provides initial evidence consistent with unconstrained

41 41 firms increasing debt and constrained firms decreasing debt in response to the 2003 Act. In addition, consistent with theory, MTR is positively correlated ( Pearson and Spearman) with Debt. In addition, HighIndOwn is positively correlated with Constrained (0.0763), and negatively correlated with MTR ( ), CF ( ), and Ln(Assets) ( ); therefore, firms with high individual ownership are more likely to be constrained, have lower marginal tax rates, cash flows, and are smaller than unconstrained firms. B. Multivariate Results Table 5 presents the results for the first multivariate model (equation 13), where the multivariate analyses is performed on a partitioned sample based on individual ownership. A positive and significant coefficient on Unconstrained and a negative and significant coefficient on Constrained in Panels (ii) and (iv) provide evidence consistent with H1 and H2. Since Panels (i) and (iii) are for the non-tax sensitive sub-sample (where HighIndOwn = 0), the coefficients on Unconstrained and Constrained are expected to be insignificant. Panels (i) and (ii) present the results for the full sample of 1,946 firms. Consistent with H1 and H2, unconstrained firms with high individual ownership increase their debt use (p< 0.01) and constrained firms with high individual ownership decrease their debt use (p< 0.01) after the 2003 Act. However, in the non-tax sensitive sample, the coefficients on Unconstrained and Constrained are both insignificant. These results indicate that only those firms with relatively higher levels of individual ownership, and thus more sensitive to changes in individual-level taxes, shifted their capital structure in response to the 2003 Act. In terms of economic significance, unconstrained firms increase their debt by an average of 8.21 percentage points of the 2001 market value of

42 42 the firm, and constrained firms decrease their debt by an average of percentage points of the 2001 market value of the firm. Panels (iii) and (iv) of Table 5 present the results for the intentional debt change (those with Debt >.02) sample of 1,284 firms. Consistent with the results of the full sample, the coefficient on Unconstrained is positive and significant (p< 0.01), and the coefficient on Constrained is negative and significant (p< 0.05). Since the intentional debt change sample eliminates all firms with small changes in their capital structure, the economic significance of the coefficients in the intentional debt change sample are larger than the full sample (19.05 percentage points and percentage points of the 2001 market value of the firm for unconstrained and constrained firms respectively). Across all panels, two of the primary drivers of debt use are changes in assets, which are positively related to changes in debt (p< 0.01 in all panels), and changes in cash flows, which are negatively related to changes in debt (p< 0.05 or better in all panels). 25 To determine whether the capital structure changes of the constrained and unconstrained firms are significantly different, the linear hypothesis that the coefficient on Unconstrained (β 1 ) minus the coefficient on Constrained (β 2 ) equals zero is tested. For non-tax sensitive firms, both the full sample and the intentional debt change sample, the test is insignificant meaning that I fail to reject the null hypothesis that the unconstrained and constrained firms have the same reaction to the 2003 Act. However, this test is significant (p< 0.01) 25 The change in assets and change in cash flows are also consistently determinants of the change in capital structure in Lin and Flannery (2013). I do note that Lin and Flannery (2013), in general, have more control variables loading significantly than in my study. This could be due to two reasons. First, their dependent variable is change in leverage ratio which is influenced by changes in debt and changes in assets versus the measure of change in debt that I use, as suggested in Graham (1996), which uses a constant lagged scalar so the variable is only influenced by changes in the numerator (i.e., debt). Second, as noted in Table 3, many of these control variables are significantly different between the unconstrained and constrained subsamples. To the extent that the relation between the controls and change in debt interacts with the relation between the dichotomous financial constraint variables and change in debt, the significance of the controls as a determinant of change in debt may be diminished.

43 43 for the tax-sensitive firms, which provides evidence consistent with the hypothesis that tax sensitive unconstrained and constrained firms had significantly different changes in debt as a result of the 2003 Act. Table 6 presents the results for the second multivariate model (equation 14), where a difference-in-differences design is used to allow for testing the differential response to the 2003 Act for the tax sensitive unconstrained and constrained firms without partitioning the sample. Panel (ii) presents the results for the change, the period in which the 2003 Act became effective. As predicted by H1 the coefficient on the interaction Constrained x HighIndOwn is negative and significant (p< 0.05), providing evidence that tax sensitive constrained firms decreased their debt use as a result of the 2003 Act, which is consistent with the full sample results of prior literature. In comparison, the coefficient on the interaction between Unconstrained and HighIndOwn is positive and significant (p< 0.01), consistent with H2. Similar to Table 5, change in assets is positively associated with changes in debt (p< 0.01) and change in cash flows is negatively associated with changes in debt (p< 0.05) in the DID model. Panels (i) and (iii) of Table 6 provide the results for the DID model for the and change in debt respectively. These are the two-year periods immediately before and after the 2003 Act. Since investor-level tax rates did not change in these periods, it is expected that the interaction coefficients (Unconstrained x HighIndOwn and Constrained x HighIndOwn) will be insignificant. If these coefficients are not insignificant, then the results found in panel (ii) are likely not driven by the 2003 Act. For the period, as predicted, the coefficients on the interaction terms are statistically insignificant. In addition, for the period, the coefficient on

44 44 Unconstrained x HighIndOwn is insignificant, however, the coefficient on Constrained x HighIndOwn is negative and marginally significant (p< 0.10). This could be due to constrained firms continuing to alter their debt policy for a longer horizon in response to the 2003 cuts. This explanation is consistent with financial constraints influencing the speed at which firms are able to adjust their capital structure (e.g., Byoun 2008; Faulkender, Flannery, Hankins, and Smith 2012; and Öztekin and Flannery 2012). 26 Alternatively, there could be another event, which is currently unexplained by the model, which is driving constrained firms to continue to alter their debt policy in the two-year period immediately following the 2003 Act. In both falsification periods, change in assets remains positively associated with change in debt, and change in cash flows remains negatively associated with changes in debt. As a whole, the results of the period and the results of the period provide evidence that the results observed in the period are not driven by inherent differences between tax sensitive unconstrained and constrained firms that persist through all time periods. 26 Byoun (2008) finds that financial constraints influence a firm s ability to adjust capital structure in response to negative cash flow shocks. Faulkender et al. (2012) find that financial constraints (measured as firms without rated debt and firms that did not pay dividends in t-1) are negatively associated with adjustment speeds. Öztekin and Flannery (2012) find, in a cross-country setting, that firms in countries with imposed financial constraints (requiring a minimum percentage of net income to be distributed as dividends or requiring firms to liquidate if they cannot maintain a minimum equity level) converge more slowly to its optimal leverage.

45 45 VI. SUPPLEMENTAL ANALYSES A. Increased Distributions for Unconstrained Firms Given that the 2003 Act decreased the tax rates on dividends and long-term capital gains, individual taxable investors now have a stronger preference for equity distributions in the post-2003 Act period as their after-tax cash flow on every dollar of equity distributions has now increased. As discussed in Section II, prior research found a shift in payout policy around the 2003 Act (e.g., Chetty and Saez 2005, 2006; Brown, Liang, and Weisbrenner 2007; Blouin, Raedy, and Shackelford 2011). Therefore, if one motivation for managers of unconstrained firms to increase debt is an increased demand for distributions by shareholders, then I expect to find evidence of increased distributions by the unconstrained firms in my sample with high individual ownership. To test this conjecture I use the following model adapted from Blouin and Krull (2009): Payout = β + β Unconstrained + β Constrained + β Size + β CapEx + β ROA + β ROA + β Debt + β DivYield + β Cash + β Payout + ε. (16) Where Payout is equal to the change in total dividends plus net share repurchases divided by total assets from 2002 to 2004, where net share repurchases are measured from COMPUSTAT. 27 Unconstrained and Constrained is measured in 2002, and the firms identified as Unconstrained and Constrained in this test are the same as in the primary analyses. Size is measured in 2001 and is the natural log of the market value of equity. CapEx is measured as the 2001 to 2002 change in the ratio of capital 27 Banyi, Dyl, and Kahle (2008) compare several available measures of share repurchases and determine that COMPUSTAT purchases of common stock and preferred stock (PRSTKC) is the best estimate of actual repurchases when adjusted for overstatements. Overstatements are due to the inclusion of amounts spent on the retirement/redemption of preferred stock as well as the conversion of preferred stock into common stock (Stephens and Weisbach 1998). To correct for this overstatement Grullon and Michaely (2002) and Kahle (2002) recommend subtracting decreases in the value of preferred stock from the COMPUSTAT annual purchase measure.

46 46 expenditures to total assets. ROA is the 2002 to 2004 change in ROA. Debt is 2001 total long-term debt scaled by total assets. DivYield is the 2001 dividend yield. Cash is the 2001 to 2002 change in the ratio of cash and cash-equivalents to total assets. The model is estimated on the 1,023 high individual ownership firms from the 1,964 firm full sample. I find results consistent with the expectation that unconstrained firms with high individual ownership will increase distributions post-2003 Act. In particular, the results in Table 7 suggest that the average tax sensitive unconstrained firm increases total distributions by 5.73 percent of total assets ( ), which is an additional 2.15 percent increase over the firms in the control group (the high individual ownership firms that are neither constrained nor unconstrained). B. Alternative Hypothesis: Increased Distributions and Financing Deficits While prior literature has documented an increase in equity use in the post-2003 Act period (Dhaliwal et al. 2007a; Lin and Flannery 2013), literature pertaining to the capital structure decisions of firms offers a contradictory prediction for a subset of firms, in particular those that do not have debt capacity concerns. Capital structure theory fits into two broad categories, trade-off theory and pecking order theory. In trade-off models (e.g., Kraus and Litzenberger 1973; Miller 1977), firms identify a target leverage ratio by finding the point in which the benefit of debt is equal to the cost (Fama and French 2002). The pecking order theory (Myers 1984; Myers and Majluf 1984) at its foundation is based on the notion that the cost of issuing new securities (namely information asymmetry) greatly outweighs the forces that determine optimal leverage in the trade-off model. As a result, firms will finance investments first with internal funds, then with

47 47 debt, and finally with external equity to minimize the problems associated with information asymmetry. Many studies have tested which of these two theories of capital structure best approximate what is observed in the economy. Shyam-Sunder and Myers (1999), using a small sample of firms, conclude that firms largely fund financing deficits with debt, which supports the pecking order theory. However, while Fama and French (2002) find that short-term variation in the financing deficit is primarily absorbed by debt, they also find, along with Frank and Goyal (2003), that small, high-growth firms are the primary issuers of equity. These studies conclude that this is contrary to the pecking order theory as these small/high-growth firms should have higher information asymmetry costs relative to large/mature firms and thus the small/high-growth firms should be more likely to follow the pecking order theory. However, most studies that test the trade-off theory and pecking order theory against each other do not consider the influence of debt capacity in their research design. Debt capacity needs to be considered in studies that test the pecking order theory as measuring debt capacity identifies those firms that are able to add additional debt to their capital structure. Analytically, Bolton and Friexas (2000) model a financial market with information asymmetry where bonds, bank debt, and external equity exist in equilibrium. In their model, bank debt is more costly than public debt because while bank debt allows the borrower to minimize the costs of financial distress, the intermediation costs of the bank are passed on to the borrow. Therefore, the most financially healthy firms (with high debt capacity) will access the public debt market because their probability of financial distress is low enough that it is not worth paying the higher costs of bank debt to

48 48 minimize the costs of financial distress. The middle group (with lower debt capacity) has a higher probability of financial distress and will therefore accept the higher costs associated with bank debt because it is more easily renegotiated in the event of financial distress. Finally, the riskiest group of firms (those with very little or no debt capacity) are shut out of the debt market and must use external equity to raise funds. Similarly, in a model of the choice between public and private debt, Holmstrom and Tirole (1997) conclude that firm s with higher debt capacity use public debt, while those with lower debt capacity use private debt. Lemmon and Zender (2010) empirically test whether debt capacity influences the choice to absorb financing deficits with debt or equity. To identify firms with high debt capacity they use a predictive model of whether a firm has a bond rating in a given year, as the aforementioned analytical work suggests that firms that access the public debt market have the highest debt capacity. 28 They partition the sample on the probability that the firm has rated debt and regress net debt issued in the year on the financing deficit. They find that the size of the coefficient on financing deficit increases monotonically as the probability that the firm has publicly rated debt (their proxy for debt capacity) increases. They interpret their findings to mean that as a firm s debt capacity increases the firm absorbs more of its financing deficit with debt. In addition, they find that the small/high-growth firms identified in prior research (e.g., Fama and French 2002; Frank and Goyal 2003) as examples of firms that violate the pecking order theory by issuing 28 Lemmon and Zender (2010) use the predictive model rather than the observed presence of rated debt because there are likely some firms that have decided to rely on external equity for reasons outside of the pecking order theory and therefore could issue public debt but would not have an observed debt rating. In addition, there may also be firms that have negative financing deficits (either large amounts of internal funds or low investment opportunities) that could issue public debt but do not need to as they do not need external financing at all.

49 49 high levels of equity also have very low debt capacity. This suggests that these firms did not necessarily violate the pecking order theory as they likely did not have the debt capacity available to take on additional debt, and thus had to issue equity even though they faced high levels of information asymmetry. Overall, prior literature has documented an increase in equity use after the 2003 Act; however, corporate finance research suggests that firms with high debt capacity will generally respond to increases in its financing deficit by issuing debt. Given that the 2003 Act incentivized firms to increase payouts, which increases financing deficits, it is plausible that high debt capacity firms actually increased debt in the post-2003 Act period even though the trade-off model in Miller (1977) suggests that these firms would decrease debt. To determine whether the primary results of this study are driven by firms with high debt capacity increasing distributions and thus increasing their debt in the post Act period to absorb their financing deficit, Table 5 is re-estimated on the subsample of firms that do not increase equity distributions around the 2003 Act. If the main results are driven by this alternative hypothesis, then the expectation is that the unconstrained firms will only increase debt use in the post-2003 Act period if they also increase distributions. The results for this test are presented in Table 8. Panels (i) and (ii) of Table 8 provide the test results for the sub-sample of firms that do not increase distributions in the post-act period. For tax sensitive firms, the coefficient on unconstrained is ; however, it is not statistically significant. The coefficient on constrained is negative and significant (p< 0.01) as in the main tests. While panel (ii) does not provide evidence consistent with tax sensitive unconstrained firms increasing debt use around the 2003 Act if they do not increase distributions, the test of

50 50 whether unconstrained and constrained firms behave differently around the 2003 Act (beta 1 beta 2 = 0) is positive and significant (p< 0.01) suggesting that unconstrained firms increased their debt use relative to constrained firms post-act. As in Table 5, the intentional debt change sample (where Debt it >.02) is tested separately as suggested in Graham (1996). Panel (iv) focuses on the subset of firms that are tax sensitive, do not increase distributions, and intentionally alter their capital structure. If the main results are completely driven by firms that increase distributions then the unconstrained firms in this sample should not increase debt use in the post-2003 Act period. However, the tests indicate that these unconstrained firms do increase debt use, which is consistent with the initial hypotheses of this study that the relation between investor-level taxes and capital structure decisions (Miller 1977) is influenced by the financial constraints of the firm. Overall, while a portion of the main results are due to firms increasing distributions and absorbing the additional financing deficit with debt, this alternative hypothesis does not completely subsume the primary results of this study. C. Combined Financial Constraint Measure Lin and Flannery (2013) investigate changes in leverage ratios around the 2003 Act and find that tax sensitive firms as a whole decrease their leverage ratios after the 2003 Act, consistent with the cost of equity capital decreasing relative to the cost of debt. In addition, they find that this effect is incrementally higher for financially unconstrained firms. They measure financial constraints using a measure of financial market access versus the need for external capital as in this study. In particular, they identify a firm as financially unconstrained (hereafter the LF method) if 66th percentile in 2002,

51 51 or 66th percentile in 2002, or 66th percentile in 2002, or 66th percentile in Since their findings are dissimilar to those of this study, I combine the measure of financial constraint based on the KZIndex (a constraint measure of need for external financing) with the LF method (a constraint measure of the cost of external financing) to determine which facet of financial constraint prevails. 29 Accordingly, I label a firm as unconstrained if the firm is determined to be unconstrained under both the KZIndex and the LF method (KZIndex constrained LF method constrained). One difference between my method and the LF method is that I have three groups of firms (unconstrained, constrained, and a control group that are neither constrained nor unconstrained) while Lin and Flannery (2013) label any firm that is not identified as unconstrained using the LF method as a constrained firm, resulting in only two groups. To maintain three groups in this analysis, I label a firm as constrained for this test if it is identified as constrained by the KZIndex and is in the bottom tercile of all four of the LF method financial market access measures. As a result, the sample for this test includes a group of firms that are constrained under both proxies, a group of firms that are unconstrained under both proxies, and a third group of control firms I would not expect the results of this study and the results of Lin and Flannery (2013) to be consistent as we are using different measures of financial constraint that capture different dimensions of financial constraint. In particular, the KZIndex proxies for the need for external funds while the LF method proxies for the cost of external funds. Therefore, the purpose of this supplemental analysis is to examine the firms that are in the intersection of the two measures (those which are identified as unconstrained based on both the need for external funds and the cost of external funds). 30 This group of control firms include firms that are constrained/unconstrained only under one of the proxies or firms that are not constrained or unconstrained under both proxies.

52 52 Table 9 presents the results for the combined constraint measure analysis. In panel 1, the multivariate model is estimated on the high individual ownership and non-high individual ownership sub-samples separately. Consistent with the results using only the KZIndex (Table 5), the combined constraint measure variables Constrained and Unconstrained do not explain changes in debt for the non-high individual ownership sample. However, for the high individual ownership sample (where HighIndOwn=1), the combined variable Unconstrained is significantly associated (p< 0.01) with positive changes in debt around the 2003 Act. Therefore, while Lin and Flannery (2013) find that their unconstrained firms with high individual ownership incrementally reduce leverage ratios around the 2003 Act and I find that high individual ownership firms identified as unconstrained using the KZIndex increase debt use around the 2003 Act, the results of the combined constraint analysis suggest that the intersection of high individual ownership firms that are unconstrained under both the KZIndex and the LF method also increase debt use around the 2003 Act. Panel 2 provides the difference-in-differences results using the combined constraint measure. Consistent with the primary analysis (Table 6 Panel ii), the unconstrained firms with high individual ownership incrementally increase debt use around the 2003 Act as evidenced by the positive and significant (p< 0.01) coefficient on the interaction Unconstrained*HighIndOwn.

53 53 VII. CONCLUSION This study provides evidence that investor-level taxes influence a firm s capital structure decisions, and this effect varies predictably based on the firm s financial constraint. Using the Jobs and Growth Tax Relief Reconciliation Act of 2003 as a quasiexperimental setting, the results indicate that tax sensitive constrained firms decrease their debt use post-2003 Act and tax sensitive unconstrained firms increase their debt use post-2003 Act. The decrease in debt use for the constrained firms is consistent with the prediction that the investor-level tax rate cuts in the 2003 Act lowered the cost of equity capital relative to the cost of debt; therefore, making equity a more attractive financing vehicle in the post-2003 Act period. In contrast, unconstrained firms are able to finance new investment entirely with internal capital. As a result, the unconstrained firms are not forced to enter the external capital market to invest in new projects. This creates a situation in which unconstrained firms are choosing between internal financing and external debt financing. Therefore, when the investor-level taxes are reduced on equity distributions (making the pre-tax cost of equity distributions cheaper for the firm), firm value is maximized by using internal funds to pay distributions to shareholders and using external debt to finance new investments. These effects are not found in the two years immediately preceding and proceeding the 2003 Act, providing evidence that the effects identified from are driven primarily by the rate cuts of the 2003 Act. However, while I perform analyses to try and rule out alternative explanations, the results of this study are subject to at least two limitations. First, the hypothesis development relies on the underlying assumptions of the Miller (1977) model and its subsequent adaptations; if these analytical assumptions are not valid in the archival

54 54 setting then the results documented in this study may be attributable to other phenomena. Second, the 2003 Act is unique in that the magnitude of the individual tax rate changes lead to opposite predictions for change in debt use for tax sensitive constrained and unconstrained firms. I have yet to identify another tax act in which these predictions hold. As a result, the results are not currently replicable using a different tax act event window. Therefore, I cannot completely rule out the possibility that there is another event, in addition to individual tax rate changes, that occurred in the 2003 Act period that lead tax sensitive unconstrained and constrained firms to differentially alter their capital structures around the 2003 Act. This paper contributes to the growing literature on firm s capital structure decisions, particularly the studies that examine how investor-level taxes influence capital structure, by providing evidence that financial constraints alter the relation between timeseries changes in investor-level taxes and debt use. Therefore, I document a more nuanced relation between investor-level taxes and capital structure decisions than presented in prior literature. Future research can investigate the subset of tax sensitive firms that are predicted to alter their capital structure around investor-level tax rate changes, but do not. The characteristics of these non-conforming firms as well as whether the market punishes these firms for not changing their capital structure would both be fruitful questions to pursue.

55 55 Appendix A: Derivation of equations (7), (8), and (9) (7): Partial derivative of (3) with respect to t p : = ( ) ( ) = ( ) ( ) = ( ) ( ) = = = ( ( ) ) ( ) ( ) ( )( ) ( ) ( ) ( ) = ( ) ( ) = = = (7) (8): Partial derivative of (6) with respect to t p : = ( ) ( ) t = ( ) ( ) t

56 56 = ( ) ( ) t = ( ( ) ) (t ) = = = ( ( ) ) + 1( 1) ( ) ( ) ( )( ) 1 ( ) ( ) ( ) 1 = ( ) ( ) 1 = 1 = 1 = (8) (9): Partial derivative of (3) and (6) with respect to t i. Note: The only difference in the gains from leverage formula for constrained (3) and unconstrained (6) firms is the additional t p term subtracted on the end of the unconstrained gains from leverage formula. Since (9) is the partial derivative of gains from leverage with respect to t i the additional t p term in (6) becomes zero and therefore both constrained and unconstrained firms have the same partial derivative of gains from leverage with respect to t i. = ( ) ( ) = ( ) ( ) = ( ) ( )

57 57 = ( ) ( ) ( ) ( ) ( ( ) ) = ( ) ( ( ) ) ( ) = ( ) ( ) = (9)

58 58 Appendix B: Reconciliation of Table 1 to Total Derivatives Total derivative for constrained firms dg : = ( ) ( ) t : t : + ( ) ( ) (t : t : ) = (. ) (. ). ( ) + (. ). ( ) (. )(. ) = Total derivative for unconstrained firms dg : = ( ) ( ) 1 t : t : + ( ) ( ) (t : t : ). = (. ) (. ). 1 ( ) + (. ). ( ) (. )(. ) =

59 59 Appendix C: Variable Definitions Debt = The 2002 to 2004 change in the book value of long-term debt scaled by the lagged market value of the firm. Where the market value of the firm is defined as the book value of debt plus the market value of equity KZIndex = *[(IB+DP)/PPENT] *[(AT+MVE-CEQ- TXDB)/AT] *[(DLTT+DLC)/(DLTT+DLC+SEQ)] *[(DVC+DVP)/PPENT] *[CHE/PPENT] Where IB is income before extraordinary items, DP is depreciation and amortization, PPENT is net property, plant, and equipment, AT is total assets, CEQ is common equity, TXDB is deferred taxes reported on the balance sheet, DLTT is long-term debt, DLC is debt in current liabilities, SEQ is stockholder s equity, DVC is common dividends, DVP is preferred dividends, CHE is cash and short-term investments, and MktEq is common shares outstanding (CSHO) times share price at calendar year end (PRCC_C) Unconstrained = Dichotomous variable coded 1 if the firm is identified as unconstrained by the KZIndex, 0 otherwise Constrained = Dichotomous variable coded 1 if the firm is identified as constrained by the KZIndex, 0 otherwise MTR = Marginal tax rate as defined by Blouin, Core, Guay (2010) CF = Operating income before depreciation scaled by total assets Momentum = 12 month buy-and-hold return MB = The ratio of the market value of equity to the book value of equity Assets = Total assets Col = Collateralized assets defined as the sum of inventories and property, plant and equipment scaled by total assets Intang = Intangible assets scaled by total assets DDiv = Dichotomous variable coded 1 if the firm paid a dividend in the year, 0 otherwise HighIndOwn = Dichotomous variable coded 1 if the firm is above the sample mean for individual ownership, 0 otherwise IndustryFE = A set of industry fixed effect dichotomous variables coded as 1 if the firm is in the industry as defined by the Fama and French 17 industry classification scheme, 0 otherwise Payout = The 2002 to 2004 change in total dividends plus net share repurchases scaled by total assets. Where net share repurchases are measured as COMPUSTAT purchases in common stock and preferred stock (PRSTKC) minus any decrease in the value of preferred stock in the year CapEx = The ratio of capital expenditures to total assets ROA = The ratio of income before extraordinary items to total assets Size = The natural log of market value of equity Debt = Total long-term debt scaled by total assets DivYield = Dividends per share divided by stock price Cash = Cash and cash-equivalents scaled by total asserts

60 60 Appendix D: Figures Figure 1 Figure 2 Optimal Financial Policy...61 Changes in Debt in the Falsification and 2003 Act Periods..62

61 61 Figure 1 Optimal Financial Policy Figure 1 presents a graphical representation of the hypotheses, where MC is the marginal cost of debt and the horizontal lines represent (from top to bottom) the marginal benefits of debt for constrained tax sensitive firms pre-2003 Act, constrained tax sensitive firms post-2003 Act, unconstrained tax sensitive firms post-2003 Act, and unconstrained tax sensitive firms pre-2003 Act respectively. The expected increase in debt use as a result of the 2003 Act for tax sensitive unconstrained firms is depicted by the increase in debt (the x axis) from U to U. The expected decrease in debt use as a result of the 2003 Act for tax sensitive constrained firms is depicted by the decrease in debt (the x axis) from C to C.