Personal Income Tax and Corporate Investment

Transcription

1 Personal Income Tax and Corporate Investment Murray Z. Frank, Rajdeep Singh and Tracy Yue Wang This version: October 13, 2009 ABSTRACT Existing studies report that the 2003 dividend tax cut had no effect on corporate investment. In this paper we show that dividend taxation, in theory should have, and empirically does have, important effects on corporate investment among publicly traded U.S. firms from 1977 to Cash-poor firms respond to a dividend tax cut by increasing investment. Cash-rich firms respond to a dividend tax cut by reducing investment and increasing dividends. The opposite reactions from these two types of firms significantly reduce the observed aggregate or average response of firms investment to dividend tax changes. JEL classification: G31, G32, H24. Keywords: corporate investment, dividend taxation, personal income tax, tax overhang, dividend capitalization. The authors are at the Carlson School of Management, University of Minnesota. Contact: Murray Frank: murra280@umn.edu; Raj Singh: rajsingh@umn.edu; Tracy Yue Wang: wangx684@umn.edu. We would like to thank seminar participants at Simon Fraser, University of Minnesota, the 4th FIRS Conference on Banking, Corporate Finance and Intermediation, and the 2008 European Summer Symposium in Financial Markets in Gerzensee Switzerland. We also thank Richard Arnott, Alan Auerbach, Reint Gropp, Thomas Hellmann, Josh Lerner, Vojislav Maksimovic, Jack Mintz, and Clemens Sialm for helpful comments. As usual, all mistakes are ours.

2 The Jobs and Growth Tax Relief Reconciliation Act of 2003 significantly cut the tax rate on dividend income for individuals. Since the tax cut increased the investment returns from an investor s point of view, the Act was expected to stimulate corporate investment. However, in a study of this issue Desai and Goolsbee (2004) found that the tax reform did not generate the expected investment. The evidence... suggests that the dividend tax reductions enacted in 2003 had little or no effect on investment. Their study calls into question the empirical importance of dividend tax effects on corporate investment. If tax cuts had no effect on corporate investment, then perhaps tax increases scheduled for 2010 (when the Bush tax cuts expire), will also not have much effect on corporate investment. Furthermore, if dividend taxes do not affect corporate investment, then perhaps some reconsideration of the corporate finance theory about the cost of capital is necessary. This paper shows that in fact, dividend taxation has empirically important effects on corporate investment. To understand the tax effect, it is crucial to recognize crosssectional differences in firms responses to personal tax changes. The cross-sectional difference under study is due to the cash position that each firm inherits from past retained profits. For our purposes the past profits could equally be due to luck or good past decisions. Either way a firm that has a profitable history is likely to be cash-rich. A firm without a profitable history will be cash-poor. The cash-rich firm being on the distribution margin can finance investment using retained earnings, while the cash-poor firm being on the financing margin must raise funds if it is to invest. As a result, even if facing the exact same real investment opportunity, these firms may make different decisions. To motivate the empirical analysis, we develop a simple illustrative model to show how personal taxes affect the firm s incentive to invest. For a firm that needs to raise new funds in order to invest ( cash-poor ), there is the conventional impact of double taxation, which raises the cost of investing. However, for a firm that has enough retained earnings ( cash-rich ) there are other effects. By reinvesting profits within the firm, the firm s investors are able to defer personal taxes in much the same way as in a tax deferred savings account. We call this the tax-overhang effect. It is related to the tax-free wealth accumulation in life insurance analyzed by Miller and Scholes (1978). The reinvestment 1

3 may also be associated with more beneficial tax treatments through an effect that we call the dividend-capitalization effect. By investing retained earnings the firm may convert investors dividend tax liabilities to capital gains liabilities. 1 These two tax effects for the cash-rich firm work in the opposite direction from the double-taxation effect. Thus in comparison to a personal-tax-free first-best investment rule, the cash-poor firm tends to underinvest, while the cash-rich firm tends to overinvest. A dividend tax cut reduces both distortions, moving both types of firms closer to the first best. The cash-rich firm tends to respond to a dividend tax cut by increasing payout and reducing investment. The cash-poor firm tends to respond by increasing investment financed by newly raised funds. In the aggregate the two opposite reactions can significantly reduce the observed response of firms to tax cuts and can even make it insignificant. The empirical work focuses on the impact of dividend tax cuts on corporate investment rather than dividend payout. 2 Using the dividend tax rates and the capital expenditures of U.S. publicly traded companies from 1977 to 2006, the implications of the model are tested. Cash-richness is empirically important. When this is taken into account there is strong evidence that dividend taxation has a first order effect on corporate investment. As predicted by the model, a cash-rich firm and a cash-poor firm react in opposite ways to a dividend tax change. There is also evidence that potential startup firms, an important example of cash-poor firms, behave in a way consistent with our model implication. New business formation surged in the U.S. following the 2003 dividend tax cut. 1 The idea that the financing margin matters is not new to our paper. The influential study of Hennessy and Whited (2005) examines the impact of this distinction on corporate capital structure. 2 Investment seems to be a key concern for policy makers. For example, the impetus for the stimulus bill of 2009 came from a desire to spur investment in the economy. Academics have also similar concerns. Auerbach and Hassett (2009) argue that there is a need for further study of the impact of the 2003 dividend tax change on corporate investment. Moreover, the empirical effect of dividend taxes on dividend payments that are observed is not too surprising. Similar results have been documented by Chetty and Saez (2005) and others for the 2003 dividend tax cut. Blouin et al. (2004) provide a number of caveats, including the important observation that Fisher Black s (1976) dividend puzzle remains true both before and after the tax change in Dharmapala (2009) provides a valuable up-to-date review of that literature. Going beyond 2003 we show that the effect of dividend taxes on dividend payments is actually fairly general. 2

4 A rich set of tests are performed to ensure the effectiveness of the empirical identification strategy. We show that changes in dividend taxes are not proxying for systemic changes in personal income taxes or other confounding events. Firms that have a larger fraction of shares held by tax-exempt investors exhibit, as expected, weaker dividend tax effects on investment. Cash-rich dividend payers significantly increase dividend payout following dividend tax cuts, while cash-poor dividend payers do not. Further, the theory implies that the differing dividend tax effects on cash-rich firms and cash-poor firms should weaken among firms with high debt capacity. This is supported by the data. The results are also quite robust. Using alternative definitions of cash-richness and examining longer-horizon changes in investment leave the key results unchanged. When the entire time-series is studied the effects found for the sub-sample of large notable tax changes continue to hold. Not surprisingly, the effects are larger when the tax changes themselves are larger. We also extend the analysis to incorporate competition between firms with similar or different cash positions. Industries with both cash-poor and cash-rich firms exhibit significantly larger investment responses to dividend tax cuts than industries with mostly cash-poor (or mostly cash-rich) firms. Given that the effects are significant, and appear to be reliable, it becomes of interest to ask whether the effects are large enough to be important. Consider a dividend tax cut similar in magnitude as the one in That is a 14.3 percentage point reduction in the marginal dividend tax rate. According to the base-line estimates, a cash-poor firm with one billion dollars of assets will on average increase its capital expenditures by $13 million, while a same-size cash-rich firm will on average decrease its investment by $15 million. In our sample the average annual capital expenditures of firms with about one billion dollars of assets is $73 million. Therefore, the estimated changes in both types of firms capital expenditures due to the tax cut are economically significant. However, the estimated change in the aggregate investment in 2003 due to the dividend tax cut is only 0.8% of the aggregate corporate assets, consistent with the flat investment response documented in other studies. The rest of the paper is structured as follows. Section I builds a simple model to illustrate how personal taxes affect the firm s investment incentive. Section II develops 3

5 the hypotheses and the empirical design based on the model implications. Section III presents the empirical analysis. Section IV discusses the related literature. Section V concludes. I. The Model The purpose of this section is to clarify the basic forces that will be tested empirically. We start with a particularly simplified situation in which all corporate distributions are taxed at the personal level at the same rate. 3 This provides a sharp illustration of the taxoverhang effect. Then the possibility of investors getting a more beneficial tax treatment is modelled. To do this a more realistic version of the tax code is studied which accounts for differing dividend tax and capital gains tax rates. This permits the issue of dividend capitalization to arise. 4 Consider the investment problem of a firm that maximizes the investor s welfare. There is no agency issue. Capital is perfectly divisible and does not depreciate. There are three dates that represent the past t = 0, the present t = 1, and the future t = 2. At t = 0, the firm raised B dollars from the investor and invested in a long term (two-period) project. This is called the original project. At t = 1, the original project either had a successful first period (net profit of π 1 > 0) or an unsuccessful first period (net profit of 0). If the firm has a positive cash balance from a successful first-period operation, it is called a cash-rich firm. If 3 Our framework is similar in spirit to the model in section I of Hennessy and Whited (2005) in that both they and we are concerned about whether money is better left in the firm, or in the hands of the investor. Hennessy and Whited (2005) study the debt versus equity choice. In contrast, we assume no flotation costs, we study investment decisions rather than capital structure decisions, and we allow for capital gains taxation. As a result, while there is some similarity, the key results are different. 4 Our analysis abstracts from a whole host of potentially interesting issues such as optimal payout policy, optimal capital structure, optimal cash retention, managerial agency, tax or liquidity motivated trading, optimal tax policy etc. Many of these could provide valuable deeper insights and qualifications to what we have found in the current study. In an interesting recent working paper Chetty and Saez (2007) study the impact of agency conflicts to explain corporate reactions to dividend tax changes. They argue that dividend taxation can exacerbate an under-monitoring problem. In contrast, our analysis sticks to the traditional setting in which managers operate in the best interests of the investors. 4

6 the firm has zero cash balance as a result of an unsuccessful first-period operation, then it is called a cash-poor firm. At t = 2 the original project has a net profit of π 1 > 0 for sure. After this period s profits are realized, the firm stops operations and returns the capital to the investor. The key decision for the firm takes place at date t = 1. Irrespective of the cash balance, at date t = 1 the firm has a decreasing returns to scale investment opportunity with a net profit of π (I), where I is the amount invested. The Inada conditions are assumed to be satisfied. Specifically, π ( ) > 0, π ( ) < 0, π (0) = and π ( ) = 0. The investor s alternative investment is simply depositing money in a bank. At t = 1 the investor has capital K invested in the bank. The bank pays a scale-independent riskless rate of return denoted r. The cash-rich firm is at the distribution margin. It can either pay out the first-period operating profits, or reinvest the profits in the new project. In contrast, the cash-poor firm is at the equity-financing margin. It can either stand pat with the existing capital, or raise new equity capital from the investor to invest in the new project. 5 The tax code treats various corporate cash flows in different ways. Let τ c be the corporate level tax on profits, τ d be the personal tax on dividends, τ g be the personal tax on capital gains, and τ i be the personal tax on interest income from the bank. A. Personal Tax Overhang Effect We start by assuming that given a firm s payout regime, all cash paid out by the firm will be taxed at the same personal tax rate for equity, τ e = t (τ d, τ g ), where t τ d > 0 and > 0. At t = 2 the firm liquidates and the investor gets to deduct her basis before the t τ g personal tax is levied. The distinction between dividend taxes and capital gains taxes is considered in the next section. 5 We discuss the impact of debt financing in Section III.C.3. Also, given that the firms will cease to exist after one period, there is no role of conserving cash for the future. Due to double taxation at the corporate level, investing in the bank by the firm is dominated by investing by the investor on her personal account. 5

7 The Cash-Poor Firm: The cash-poor firm has no internal cash from past operations. Suppose that at t = 1 the cash-poor firm raises I P dollars from the investor and invests in the new project. At t = 2, the firm pays out its after-corporate-tax earnings (1 τ c ) [π 1 + π (I P )], liquidates its capital, and returns B + I P to the investor. The tax basis of the investor is also B + I P. The investor, at t = 1 also has (K I P ) dollars invested in the bank. Hence, the investor s wealth at t = 2 is given by V P = (K I P ) }{{} Bank Deposit [1 + r (1 τ i )] + [(1 τ c ) [π 1 + π (I P )] + B + I P ] }{{} Firm s t=2 Payout Payout Tax Basis {}}{{}}{ τ e (1 τ c ) [π 1 + π (I P )] + B + I P (B + I P ) } {{ } Personal tax at t=2 The firm maximizes the investor wealth V P by choosing I P. The first order condition is given by V P I P = (1 + r (1 τ i )) + (1 τ e ) (1 τ c ) [π (I p )] + 1 = 0. (1) The firm invests to the point at which the after-tax return of investment is equal to the after-tax cost of capital. Solving the first order condition gives, 6 π (I p ) = r (1 τ i ) (1 τ e ) (1 τ c ) π P. (2) The interpretation of π P is quite natural. In (2) the term π P is the marginal pre-tax required rate of return. The cash-poor firm invests until the point at which its pre-tax return from investment is equal to π P. The taxes on interest income (τ i ), equity income (τ e ), and corporate profits (τ c ) all affect the optimal scale on which the cash-poor firm invests. 6 The Inada conditions insure that there exists an interior solution to the FOC. The second order condition is clearly satisfied as 2 V P = (1 τ IP 2 e ) (1 τ c ) π (I p ) < 0. Hence, the solution to the FOC is the maxima. 6

8 The Cash-Rich Firm: The problem for the cash-rich firm is similar to that of the cash-poor firm. The main difference is that the cash-rich firm s cash balance at t = 1 is assumed to be high enough so that the firm does not need to raise fresh capital to invest. Instead the firm must decide how much money, if any, to return to the investor. The cash-rich firm has earnings at t = 1, and it pays corporate tax of τ c π 1. The firm invests I R dollars and returns rest of the earnings (1 τ c ) π 1 I R to the investor. At time t = 2 the firm first pays corporate taxes and pays out (1 τ c ) [π 1 + π (I R )] to the investor. The firm also liquidates its capital (B + I R ) and returns this to the investor. The investor has a basis of B. Accordingly the investor s wealth at t = 2 is given by Bank Deposit {}}{ V R = (1 τ e ) [(1 τ c ) π 1 I R ] + K [1 + r (1 τ }{{} i )] + [(1 τ c ) [π 1 + π (I R )] + B + I R ] }{{} After tax t=1 dividend Payout {}}{ τ e (1 τ c ) [π 1 + π (I R )] + B + I R Tax Basis {}}{ B. } {{ } Personal tax at t=2 Firm s t=2 Payout The first order condition for this problem is given by V R I R = (1 τ e ) (1 τ c ) π (I R ) + (1 τ e ) (1 + r (1 τ i )) (1 τ e ) = 0 (3) Concavity of π ( ) is again sufficient to satisfy the second order condition. The firm invests until its after-tax return of investment is equal to the after-tax cost of capital. Solving the first order condition gives, π (I R ) = r (1 τ i) (1 τ c ) π R. (4) Here π R is the cash-rich firm s marginal pre-tax required rate of return. The basic difference between the cash-rich firm s and the cash-poor firm s marginal conditions is clear. The personal tax rate on equity does not affect the marginal condition 7

9 of the cash-rich firm. In the next section it will be shown that this simplification does not carry over, under reasonable conditions, to the case in which there is a distinction between dividend taxes and capital gains taxes. The Impact of Tax Changes on Firm Investment A key purpose for the analysis is to clarify the differing impact of personal tax changes on the investment incentives of the cash-poor and the cash-rich. The firm s optimal level of investment is determined by its investment opportunity set and the required rate of return derived above. Let the firm s optimal investment for the cash-poor and the cash-rich firms respectively be denoted by IP and I R. These are determined by solving, π (I P ) = π P ; π (I R) = π R (5) The different investment incentives of the cash-poor firm and the cash-rich firm arise from the difference in their marginal required rate of return (π P π R Δ π ). Δ π = r (1 τ i) τ e (1 τ e ) (1 τ c ) > 0 (6) The intuition can be described fairly simply. If the cash-poor firm invests $1 at t = 1, the opportunity cost from the investor s point of view is the after-tax bank interest forgone given by r (1 τ i ). For the cash-rich firm, by investing $1 the firm s investor avoid paying dividend taxes of τ e at t = 1. Hence, the direct opportunity cost for the after-tax interest forgone is given by r (1 τ i ) (1 τ e ). This difference is what we refer to as the personaltax-overhang effect. It comes from the cash-rich firm s ability to defer the investor s personal tax liabilities. The personal-tax overhang effect is conceptually very similar to the advantage obtained by popular tax-deferred investments such as 401-K accounts. The model helps guide the main empirical tests to follow. Empirically the point is to compare the investment responses of cash-rich firms and cash-poor firms to changes in the dividend tax rate. The model s prediction for these investment responses depends 8

10 on how the pre-tax required rate of return changes with respect to the change in the tax rate. The model implies that π P τ d π R τ d = π ( P τ e = τ e τ d r (1 τ i ) (1 τ e ) 2 (1 τ c ) = π R τ e τ e τ d = (0) τ e τ d = 0. ) τe τ d > 0 A cash-poor firm s pre-tax required rate of return is increasing in the dividend tax rate. This is due to double taxation on corporate profits. For every $1 of dividend paid at t = 2 the investor only gets (1 τ e ). Hence, the cash-poor firm requires a higher pre-tax return from its investment as the personal tax rate increases. The cash-rich firm is also subject to double taxation. However, there is an offsetting force in the case of the cash-rich firm that is absent in the case of the cash-poor firm. The current earnings has already accrued a tax liability. By investing the retained earnings the firm helps the investor defer the tax liability. This accrued tax-liability is the basis of the tax overhang. Thus, an increase in the personal tax rate reduces the after-tax return from investment but also increases the benefit of tax deferment in the same proportion. The two effects exactly cancel out, resulting in no net change in the required rate of return. 7 The above discussion is summarized in the following proposition. Proposition 1 The cash-rich firm s optimal level of investment is greater than that of the cash-poor firm (IR > I P ). The cash-poor firm s optimal investment is decreasing in the dividend tax rate ( I P τ d < 0). The cash-rich firm s optimal investment is independent of the dividend tax rate ( I R τ d = 0). Proof: I R > I P results from π R < π P and the concavity of π(i). I P τ d Similarly, I R τ d = I P π P π P τ d. Equation (5) implies that I P π P = 1 < 0. Since π π P (I) τ d = I R π R π R τ d = 0 because π R τ d = 0. This completes the proof. > 0, I P τ d < 0. 7 The comparative statics of π P and π R with respect to τ c and τ i are straightforward and intuitive. r(1 τ i) > 0; π (1 τ e)(1 τ c) 2 R τ c = r(1 τi) > 0; π (1 τ c) 2 P r τ i = (1 τ < 0; and π R e)(1 τ c) τ i = r (1 τ < 0. c) π P τ c = 9

11 The results so far imply that following a dividend tax cut, there should be no effect on the investment by cash-rich firms but an increase in the investment by cash-poor firms. B. Dividend Capitalization Effect In practice the distinction between dividend taxation and capital gains taxation can be important. How much of an effect will this have on the analysis? It depends on how firms behave. Unfortunately there is no fully accepted theory of either corporate dividend policy, or of corporate financing policies. 8 Thus we make simple assumptions that are consistent with the intent of the tax code. According to Spilker et al. (2009), The tax law defines a dividend as any distribution of property made by a corporation to its shareholders out of its earnings and profits (E&P) account. 9 Thus, we assume that the payout that arises from periodic earnings is subject to the dividend taxation (τ d ). The payout that arises from liquidated capital is subject to the capital gains taxation (τ g ). The Cash-Poor Firm The analysis is very similar to what came before. The cash-poor firm invests I P time t = 1. At t = 2, the after-tax earnings from the original project and the new project are paid out as dividends. Also at t = 2, the capital invested in both projects, B + I P, is recovered and returned to the investor. As in the US tax code we assume that the 8 For recent a survey of corporate dividends and share repurchases, see Kalay and Lemmon (2008). For a recent survey of capital structure, see Frank and Goyal (2008). These are both large literatures. To explore either would be well beyond the scope of our current study. 9 A detailed discussion of the tax issues is provided by Spilker et al. (2009, chapter 18). Precise definitions of distributions that are considered dividends can be obtained from the U. S. Code, Title , entitled Dividend Defined, which reads in part: For purposes of this subtitle, the term dividend means any distribution of property made by a corporation to its shareholders (1) out of its earnings and profits accumulated after February 28, 1913, or (2) out of its earnings and profits of the taxable year (computed as of the close of the taxable year without diminution by reason of any distributions made during the taxable year), without regard to the amount of the earnings and profits at the time the distribution was made. at 10

12 distribution of liquidated capital is treated as a capital gain. The tax basis of the investor is also B + I P. Hence, the investor s wealth at t = 2 is given by V P = (1 τ d ) (1 τ c ) [π 1 + π (I P )] }{{} After-tax dividend + B + I }{{ P} + (K I P ) (1 + r (1 τ }{{} i )). Return of Capital Bank Deposit The first order condition is given by V P I P = (1 τ d ) (1 τ c ) [π (I p )] + 1 (1 + r (1 τ i )) = 0. (7) The first order condition is exactly the same as in equation (1). Solving it gives, π (I p ) = r (1 τ i ) (1 τ d ) (1 τ c ) π P. (8) Note that the cash-poor firm s marginal required rate of return is not affected by allowing the dividend tax and capital gains tax rates to differ. This is not surprising. The Cash-Rich Firm For the cash-rich firm things are not quite as simple. This happens because there is an extra effect that was not previously present. We call this effect dividend capitalization. At t = 1, the cash-rich firm pays corporate tax of τ c π 1. It invests I R dollars and returns the rest of the cash balance of (1 τ c ) π 1 I R to the investor. This distribution is taxed as a dividend because it comes from periodic earnings. At t = 2 the firm pays out the earnings from the original and the new projects (1 τ c ) [π 1 + π (I R )] as a dividend. The firm then liquidates its capital (B + I R ) and returns it to the investor as a capital gain. The investor s tax basis is B. Thus the portion of the distribution subject to the capital gains taxation is ((B + I R ) B) = I R. 11

13 The investor s wealth at t = 2 is given by Bank Deposit {}}{ V R = (1 τ d ) [(1 τ c ) π 1 I R ] + K [1 + r (1 τ }{{} i )] + (1 τ d ) (1 τ c ) [π 1 + π (I R )] }{{} After tax t=1 dividend After-tax t=2 dividend + B + I R τ g I }{{ R }. After-tax Cap-gains payout The way the problem is set up ignores the possibility of the firm paying out cash and simultaneously raising fresh cash at t = 1. This would only be optimal if the capital gains tax rate were significantly higher than the dividend tax rate. To rule out this unusual possibility we assume that τ g < (1 + r (1 τ i )) τ d. 10 This assumption is empirically natural given the current and historical personal tax rates in the United States. The first order condition is given by V R I R = (1 τ d ) (1 τ c ) π (I R ) + (1 τ g ) (1 + r (1 τ i )) (1 τ d ) = 0 (9) Solving this condition provides the marginal pre-tax required rate of return given by, π (I R ) = r (1 τ i) (1 τ c ) (τ d τ g ) (1 τ d ) (1 τ c ) π R. (10) As before the key issue is the different investment incentives of the cash-poor firm and the cash-rich firm. This difference is created by the difference in their marginal required rate of return (π P π R Δ π ). Direct substitution gives, Δ π = r(1 τ i )τ d τ d τ g +. (11) (1 τ d )(1 τ c ) (1 τ }{{} d )(1 τ c ) }{{} Dividend tax-overhang Dividend capitalization 10 Consider a firm that on the margin pays $1 of dividend and raise $1 of fresh equity at t = 1. The investor would be forced to realize a divided tax of τ d, and the wealth at t = 2 will reduce by τ d (1 + r (1 τ i )). If instead the firm retains $1 and invests it at t = 1 and returned it as a capital gain at t = 2, the investor will pay τ g as taxes at t = 2. Thus, when τ g > (1 + r (1 τ i )) τ d, raising fresh equity dominates investing retained earnings. 12

14 The required rate of return difference is positive as long as τ g < τ d [1 + r(1 τ i )]. Not coincidentally this is the same condition needed for reinvesting retained earnings to be optimal for the cash-rich firm. Historically, the capital gains tax rate has been lower than the dividend tax rate in the United States. Thus the lower required rate of return enjoyed by the cash-rich firm exists under realistic tax parameters. How should we think about (11)? The assumed structure of the tax code allows the cash-rich firm to achieve two objectives when investing retaining earnings. First, the firm defers the investor s dividend tax liability to a later period. This tax-overhang effect is the first item on the right hand side of (11). Second, the firm converts the t = 1 dividend payout to a capital gains payout, which may be subject to a lower tax rate. This is the second item on the right hand side of (11). In other words, by investing retained earnings the firm is able to capitalize the dividend payment. That is why we call this effect the dividend-capitalization effect. With this richer specification of the tax code, there are two benefits for the cashrich firm s investor relative to the cash-poor s investor. The investor of the cash-rich firm benefits from both the dividend-tax-overhang effect and the dividend-capitalization effect. Accordingly, the required rate of return is lower for the cash-rich firm. The next proposition compares the investment incentives of the cash-poor firm and the cash-rich firm in a tax regime that allows for dividend capitalization. Proposition 2 If it is optimal for the cash-rich firm to invest retained earnings (τ g < τ d [1 + r(1 τ i )]), then the cash-rich firm s optimal level of investment is greater than that of the cash-poor firm (IR > I P ). The cash-poor firm s optimal investment is decreasing in the dividend tax rate ( I P τ d < 0). The cash-rich firm s optimal investment is increasing in the dividend tax rate ( I R τ d > 0). Proof: From (11) we have π R < π P when τ g < τ d [1 + r(1 τ i )]. I R > I P π R < π P and the concavity of π(i). results from 13

15 The comparative statics of π P and π R with respect to τ d are π P r(1 τ i ) = τ d (1 τ d ) 2 (1 τ c ) > 0; π R 1 τ g = τ d (1 τ d ) 2 (1 τ c ) < 0. Combining these with equation (5) gives the result that I P τ d completes the proof. < 0 and I R τ d > 0. This The major difference between Proposition 2 and Proposition 1 is that once we allow for dividend capitalization the cash-rich firm s investment becomes increasing in the dividend tax rate. This is because an increase in the dividend tax rate increases both the benefit from deferring the dividend taxes (a larger dividend-tax-overhang effect) and the benefit from converting dividends to capital gains (a larger dividend-capitalization effect). The increased benefit more than offsets the increase in double taxation. 11 The question of whether investors benefit from dividend capitalization is largely an institutional and empirical question. Our analysis shows that if this effect is relevant, then the cash-rich firm s investment is increasing in the dividend tax rate ( I R τ d > 0). If it is not relevant, then the cash-rich firm s investment is independent of the dividend tax rate ( I R τ d = 0). These are the predictions to be tested. Finally, it is easy to see that in the absence of dividend taxation (τ d = 0), the required rate of return for the cash-poor firm is π P τd =0 = r(1 τ i) (1 τ c) < π P τd >0. This implies that IP τ d >0 < IP τ d =0. The cash-poor firm underinvests in the presence of dividend taxation. On the other hand, π R τd =0 = r(1 τ i)+τ g (1 τ c) > π R τd >0, implying that IR τ d >0 > IR τ d =0. The cash-rich firm overinvests in the presence of dividend taxation. The model implies that a dividend tax cut will decrease both the under-investment by the cash-poor and the 11 The result that the cash-rich firm s required rate of return decreases with the dividend tax rate can arise in other settings. For example Hennessy and Whited (2005) consider a dynamic model with endogenous investment and capital structure decisions. They also show that the marginal cost of capital for a cash-rich firm is decreasing in the dividend tax rate. 14

16 over-investment by the cash-rich. Empirical work that does not control for this type of firm heterogeneity is likely to generate misleading results. 12 II. Model Implications and Hypothesis Development The implications to be empirically examined are comparative statics stated in the propositions how changes in dividend tax rates affect changes in corporate investment. The model implies that a reduction in the dividend tax rate will lead to either no change or a decrease in investment by cash-rich firms, and an increase in investment by the cashpoor firms. To test the model implications it is natural to consider changes in the tax code. Such quasi-natural experiments are attractive since changes in the tax code are reasonably exogenous to the investment decisions of an individual firm. Let I i,t be firm-i s investment in year t, τ d,t be the dividend tax rate in year t, and Δ be the first-difference operator. The model focuses on the implication of the firm s cash position for the dividend taxation effect. Thus we use CashP oor i,t 1 to indicate whether the firm is cash-poor or not at the beginning of year t. Let X i,t denote a vector of conventional firm-specific factors that may affect investment but are not in the model. Finally, let ε i,t be a regression error term. Consider the following basic linear empirical model, ΔI i,t = αδτ d,t + β(cashp oor i,t 1 Δτ d,t ) + γ 1 CashP oor i,t 1 + γ 2 ΔX i,t + ε i,t. (12) In this specification investment and the tax rates are both in the first-difference form, but CashP oor is in the level form. This reflects the structure of the propositions being tested the effect of changes in dividend tax rates on changes in investment. According to the analysis, this impact depends on whether the firm is cash-rich or cash-poor. Thus the cash position is supposed to enter in the level form. Since the main variables of interest are in the first-difference form, the other control variables are also in the first-difference 12 In earlier drafts of this paper we analyzed a number of extensions including the use of debt financing, and the impact of imperfect competition. These extensions generate fairly natural results that are described in the empirical section. To save space the derivations are not presented. 15

17 form. The first-difference model has the added benefit of accounting for unobservable factors that have a constant effect on a firm s investment such as firm fixed effects or industry fixed effects. 13 The parameter α represents the effect of changes in dividend taxes on changes in the cash-rich firm s investment. The model (assuming dividend capitalization) predicts that α > 0. That is, a dividend tax cut will lead to a decrease in the firm s investment. If there is no dividend capitalization, then in the model we can have α = 0. The public finance literature has developed both an old view (e.g., Poterba and Summers (1984)) and a new view (e.g., Auerbach and Hassett (2003)) of the impact of dividend taxation on corporate investment. Under the old view double taxation is key, so a dividend tax cut reduces the tax burden and stimulates investment. According to the old view α < 0. Under the new view investment is financed out of retained earnings, and a dividend tax cut has no effect on corporate investment. According to the new view α = 0. Therefore, the empirical framework allows us to test between the different views about the dividend tax effect on a cash-rich firm s investment incentives. The previous literature is discussed in section IV. For the cash-poor firm, the effect of dividend taxes on investment is given by (α + β). The model predicts that the effect is negative due to double taxation. That is, a dividend tax cut will lead to an increase in the firm s investment. Since α 0, the prediction is that β < 0 and β > α. The new view could be interpreted to mean that all firms are cash-rich. If so, it contains no prediction for cash-poor firms. It could also be interpreted to mean that β = 0. Similarly there is some ambiguity in how best to interpret the implication of the old view for β. It is possible to interpret the old view to mean either β < 0 or β = 0. In some sense one could think of the old view as assuming that all firms are like our cash-poor firms. 13 In Wooldridge (2002) chapters 10.6 and 10.7 discuss in detail the advantages of using a first-difference model to address the fixed effect problem. It is worth stressing that the dividend tax change affects both cash-poor firms and cash-rich firms. Cash-poor firms are not a control group for cash-rich firms. The theory has specific predictions about the reaction of each type of firms, not just the difference between the two. 16

18 The parameter γ 1 gives the impact of being cash-poor on the change in investment. This effect is not directly predicted by the model. But since the theory makes a prediction on the interaction effect between cash-poorness and changes in the tax rates, it is important to include the direct effect of cash-poorness to avoid a biased estimation of the interaction effect. According to the model cash-poor firms will have a lower level of investment than cash-rich firms. But the empirical model does not regress the level of investment on the level of cash-poorness. It explains the change in investment in terms of the level of cashpoorness. Suppose that empirically the change in being cash-poor is correlated with the level of being cash-poor. Then we would expect to find that γ 1 < 0. But it should be stressed that this prediction, while seemingly reasonable, is not as tightly connected to the theory as the other predictions. III. Empirical Analysis This section presents the empirical analysis. Section A describes the data and empirical specification. Section B presents the main empirical results. Section C examines the effectiveness of the identification strategy. Section D discusses robustness issues. A. Data and Empirical Specification A.1. Dividend Tax Rates There are a number of alternative measures that have been used to proxy for the tax rates. Graham (1996) provides a useful study. We use two popular measures. The main measure for Δτ d is the yearly change in the U.S. federal average marginal dividend tax rate from 1977 to This is the dollar weighted average marginal individual dividend income tax as calculated by the NBER TAXISM model from micro data for a sample of U.S. taxpayers. 14 As a robustness check, we also use the time-series of the statutory 14 See detailed information at taxsim/marginal-tax-rates/federal.html. 17

19 dividend tax rate from 1981 to 2006 from the OECD tax database. 15 This tax rate is computed as the net top statutory dividend income tax rate to be paid at the shareholder level in the U.S., taking account of all types of reliefs and gross-up provisions at the shareholder level. Figure 1 plots the time series of the marginal dividend tax rate and that of the top statutory dividend tax rate. The two time series are very similar to each other. It is apparent from Figure 1 that in the following three periods the dividend tax rate underwent a significant change: (i) in years 1982 and 1983 the marginal dividend tax rate decreased by 7.6 percentage points, and the tax rate decreased by 5.5 percentage points in 1982 alone; (ii) from 1986 to 1988 there was another 7.4 percentage-point reduction; and (iii) in 2003 there was a large 14.3 percentage-point reduction in the dividend tax rate in a single year. The focus on these three periods in the main analysis has two basic motivations. First, small changes in the marginal dividend tax rates in the 1990s correspond to changes in tax clienteles rather than real dividend tax changes. Second, if the theory has predictive power, then the identification of the dividend tax effect must come from these relatively large changes in the dividend tax rate. But we do check the robustness of the main results in the entire period from 1977 to Table 1 Panel A shows that the marginal dividend tax rate decreased from 34.4% to 12.3% as a result of the dividend tax cuts in the sample period. The median annual change in the dividend tax rate is -2.9 percentage points. A.2. Corporate Investment and Cash-Poorness The firm level financial data is obtained from the COMPUSTAT industry annual database. The empirical proxy for ΔI is the yearly change in a firm s capital expenditure ratio (COMPUSTAT item #128 divided by lagged item #6). Since the focus is on capital expenditures, we exclude firms in the finance, insurance, and real estate sectors (2-digit SIC code between 60 and 69). Also excluded are firms with the 2-digit SIC code equal to See detailed information at 18

20 or above because those firms are either non-private companies or shell holding companies that have no real activities. Financial variables are truncated at the top and bottom 1% of the distribution to reduce the influence of outliers. As shown in Table 1 Panel A, the median investment ratio in the sample is 4.9 percentage points, and the median annual change in the investment ratio is -0.3 percentage point. For the variable CashP oor, the theory implies that it should be a state variable reflecting a firm s financing margin. This variable should depend on not only a firm s cumulative cash balance, but also the firm s optimal investment, which economists usually do not perfectly observe. To capture the cash-holding effect, our primary proxy for CashP oor is one minus a firm s cash-to-asset ratio (1 #1/#6). This measure is between zero and one, and thus can serve as a probability measure. The higher the CashP oor measure, the more likely that the firm is cash-poor and is on the equity-financing margin. For robustness, we also compute three alternative measures of CashP oor. CashP oor2 is a dummy variable that equals one if a firm s cash holding (item #1) is less than the sum of its average level of capital expenditures (item #128) and the average level of nonnegative non-cash working capital (max[0, item#4 item#1 item#5]) in the past three years, and equals zero otherwise. 16 The idea here is to use the firm s average investment spending in the recent past as a proxy for its optimal investment. This is reasonable because corporate investment generally exhibits strong auto-correlation (see, e.g., Eberly et al. (2008)). CashP oor3 is a dummy variable that equals one if a firm s cash holding is less than the sum of its average level of long-term and short-term investment (as in CashP oor2) and the average level of cash dividend payout (item #127) in the past three years, and equals zero otherwise. CashP oor3 tells us whether a firm s cash holding is sufficient to cover its investment needs and regular dividend payout. Bates et al. (2008) provide an empirical model for a firm s need to hold cash (see model (1) of Table 3 in their paper). In their model a firm s need for holding cash is positively related to its growth potential (measured by the ratio of market value of 16 Positive non-cash working capital is like short-term investment that needs to be financed. 19

21 assets to book value of assets), investment (capital expenditures, R&D expenditures, and acquisition cash flow), dividend practice (a dummy variable for common-dividend payers), and industry cash flow risk (the average cash flow volatility in a two-digit SIC code industry in the past ten years). 17 Accordingly we compute the abnormal cash ratio as the residual term from their regression model. CashP oor4 is defined as one minus the abnormal cash ratio. CashP oor3 and CashP oor4 take into consideration not only a firm s investment need for cash, but also a payout need for cash and a precautionary motive for holding cash. Although our theory does not incorporate these other needs for cash, they may be empirically important for the distinction between cash-rich firms and cash-poor firms. Table 1 Panel A shows that the median cash-poorness is 0.94, which corresponds to a median cash ratio of 6%. A little over three quarters of the firms are cash-poor according to the definitions of CashP oor2 and CashP oor3. The median CashP oor4 is 1.004, which corresponds to a median abnormal cash ratio of -0.4% based on the Bates et al. model. Panel B shows that all four cash-poorness proxies are highly correlated with each other. CashP oor has a correlation of 0.69 with CashP oor2, 0.70 with CashP oor3, and 0.79 with CashP oor4. Table 1 Panel C compares the characteristics of cash-poor firms with those of cash-rich firms. We assign the sample firms into one of three groups based on the tercile cutoffs of the CashP oor4 distribution. For the purpose of this comparison, we call firms in the top tercile Cash-Poor firms, and those in the bottom tercile Cash-Rich firms. Firm size, investment, growth, leverage, dividend payout, and past external financing activities between these two groups are compared. The average value of each variable for each group and the Wilcoxon Z-statistics for the difference between the two groups are reported. Empirically cash-rich firms are larger as indicated by the Wilcoxon Z-statistics, even though their average asset sizes are smaller than those of cash-poor firms. Cash-rich firms have higher growth potentials (measured by Q) but do not necessarily invest more in capital expenditures. Cash-rich firms are also more likely to pay common dividend and 17 Bates et al. (2008) also include cash flows from operations in the regression. This is not included because this is related to the supply of cash, not the need for cash. 20

22 issue debt. Overall, the empirical characteristics of cash-rich firms and cash-poor firms are consistent with the theoretical definitions of these two types of firms. A.3. Control Variables The ΔX i,t in equation (12) contains the first differences of prominent factors from previous literature that influence investment. Specifically, following the findings in the investment-q-cash-flow literature, we control for the lagged change in a firm s Tobin s Q (items [#6 #60 + #25 #199]/#6) and the lagged change in its EBIT DA (item #13/#6]). We also control for the lagged investment ΔI i,t 1. As discussed in Eberly et al. (2008), lagged investment is empirically even more important than Q and cash flow. Besides ΔX i,t, we also control for the direct effects of a firm s cash-poorness so that we can properly estimate the interaction effects of cash-poorness and tax changes. Table 1 Panel B reports the pair-wise correlation between any two variables in the baseline regressions. B. Results B.1. Dividend Tax Effect on Investment Table 2 presents results for the main empirical model in equation (12). Separate results are reported for all the years with dividend tax cuts and for the years with the two largest dividend tax cuts. The full dividend-tax-cut sample includes the three periods mentioned before: from 1982 to 1983, from 1986 to 1988, and in The large tax-cut sample includes the two large single-year dividend tax cuts in 1982 (-5.5%) and in 2003 (-14.3%). Consider the direct effect of dividend taxation on capital expenditures. By the empirical design this is the dividend tax effect on cash-rich firms investments. Assuming dividend capitalization the model predicts that α > 0. If there is no dividend capitalization, then we expect to find α = 0. Table 2 shows that the coefficient estimate of Δτ d,t is positive and significant in all models. In model (1) the estimated α is and is significant at 1% confidence level. 21

23 This implies that a dividend tax cut (i.e., Δτ d,t < 0) leads to a decrease in investment by cash-rich firms. Model (3) focused on the years of the two largest dividend tax cuts, and the effect of dividend taxation on cash-rich firms investments becomes even stronger. The estimate of α is in those years, more than doubles the estimated effect in model (1) where all the dividend tax cuts are included. This is as it should be if the empirical identification strategy is correct. A positive and significant estimate for α provides support for the existence of the dividend-capitalization effect. Dividend taxation does affect the investment incentives of cash-rich firms, but not in the way expected by policy makers. The empirical results do not support the prediction from a pure tax-overhang effect that dividend taxes are irrelevant to cash-rich firms marginal investment incentives. The theoretical model predicts that the dividend tax effect on cash-poor firms investment is positive, which means that β should be sufficiently negative to outweigh the effect of a positive α. This is exactly what we find in the data. The coefficient estimate of the interaction term between the change in the dividend tax rate and a firm s cash-poorness is negative and statistically significant in all models. Models (1) and (3) show that the estimates of β are larger in absolute magnitudes than the estimates of α. Accordingly (α + β) is significantly below zero. These results imply that cash-poor firms and cash-rich firms respond in opposite ways to dividend tax changes. Now let us consider the economic significance of the dividend tax effect on both cashrich firms and cash-poor firms. To be concrete, consider model (3) and the 2003 dividend tax cut (Δτ d = 0.143). The average firm in the sample has about $1 billion in assets. First suppose that the average firm is a cash-poor firm in the top tercile of the CashPoor distribution. The average cash-poorness in the top tercile is Then the average firm will increase its capital expenditures by $13.2 million (= ( ) ( 0.143) $1 billion). Then suppose that the average firm is a cash-rich firm that has a cash-poorness of 0.64, the average level in the bottom tercile of the CashPoor distribution. Then the average firm will decrease its investment by about $15 million ( = ( ) ( 0.143) $1 billion). 22

24 The above calculations show that the distinction between a cash-poor firm and a cashrich firm is economically significant. Further, in our sample the average annual capital expenditures of firms with about one billion dollars of assets is $73 million. Therefore, the estimated changes in both types of firms capital expenditures due to the 2003 dividend tax cut are also economically significant. The heterogeneity in the dividend tax effect on cash-rich firms and cash-poor firms implies that assessing the effect of a tax cut using changes in the average investment or in the aggregate investment can be misleading. Following the 2003 dividend tax cut, the average corporate investment increased by 0.7 percentage point or $7 million in a firm with $1 billion in assets. 18 This is much smaller than the estimated change in either the cash-poor firm or the cash-rich firm. The effect of the dividend tax cut on the aggregate investment depends on the distribution of cash-rich firms and cash-poor firms in the economy. The estimated change in the aggregate investment in 2003 due to the dividend tax cut is only 0.8% of the aggregate corporate assets, consistent with the flat investment response documented in other studies. 19 Significant dividend tax rate changes may be accompanied by other macroeconomic events. For example, the 2003 dividend tax cut was only a part of the fiscal stimulus policy the Jobs and Growth Tax Relief Reconciliation Act. To control for the effects of confounding events, year fixed effects are included in models (2) and (4). The direct effect of a dividend tax change drops out because the change applies to all firms in a given year. However, the rest of the results continue to hold. The negative estimate of β implies that cash-poor firms tend to increase investment following a dividend tax cut. Even the magnitude of the effect is similar to that in models (1) and (3) where we do not control 18 To calculate the average effect we run the following regression: ΔI i,t = φδτ d,t + γδx i,t + ε i,t in years with large tax cuts (i.e., 1982 and 2003). That is, we do not distinguish between cash-rich firms and cash-poor firms. The coefficient estimate of Δτ d is Multiplying this coefficient estimate with Δτ d in 2003 (-0.143), gives the 0.7 percentage-point average tax effect on a firm s investment. 19 The aggregate effect is computed as follows. For each firm in year 2003 we compute the change in its investment ratio due to the tax cut as Δτ d,t (CashP oor i,t 1 Δτ d,t ), where t is The change in the level of investment is the change in investment ratio times the firm s total book assets at the beginning of year We then sum up the change in the investment level of all firms and divide it by the sum of the book assets of all firms, and this ratio is the estimated effect of the tax cut on the aggregate investment. 23