Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Rate of return wikipedia , lookup
Negative gearing wikipedia , lookup
Modified Dietz method wikipedia , lookup
Financial economics wikipedia , lookup
Pensions crisis wikipedia , lookup
Luxembourg Leaks wikipedia , lookup
Stock selection criterion wikipedia , lookup
Interest rate wikipedia , lookup
Liquidity, Investor-Level Tax Rates, and Expected Rates of Return Stephanie A. Sikes† The Wharton School University of Pennsylvania [email protected] Robert E. Verrecchia The Wharton School University of Pennsylvania [email protected] July 2012 Abstract We provide new theory that predicts that lower liquidity amplifies the positive relation between expected rates of return and the dividend tax penalty (i.e., extent to which the tax rate on dividend income exceeds that on capital gain income) documented in prior literature. We empirically test our prediction using the Jobs and Growth Tax Relief Reconciliation Act of 2003 (JGTRRA03), which eliminated the penalty. We find that lower liquidity amplifies the decrease in expected rates of return following JGTRRA03. We confirm that expected rates of return decrease with dividend yield following JGTRRA03 and show that lower liquidity amplifies the decrease. Two implications for prior studies are that institutional ownership could mitigate tax capitalization due to firms with greater institutional ownership being more liquid and that prior studies’ finding that expected rates of return fall more for non-dividend paying firms following JGTRRA03 could be due to non-dividend-paying firms being less liquid. Key Words: Taxes, Liquidity, Expected Return The authors thank Peter Easton (discussant), Luzi Hail, Richard Sansing (discussant), Pavel Savor, and participants at the 2012 UNC Tax Symposium and 2012 American Taxation Association Mid-Year Meeting for helpful comments. † Corresponding Author. Mailing Address: Accounting Department, The Wharton School, University of Pennsylvania, 1300 Steinberg Hall-Dietrich Hall, 3620 Locust Walk, Philadelphia, PA 19104-6365; Phone: 215-898-7783 1 Introduction For decades researchers in economics, …nance, and accounting have debated and sought evidence that dividend taxes are impounded into the value of equity shares (e.g., Black and Scholes 1974; Litzenberger and Ramaswamy 1979, 1980, 1982; Miller and Scholes 1978, 1982). We contribute to this debate by o¤ering evidence that liquidity a¤ects the positive relation between expected pretax rates of return and dividend taxes that is predicted in many prior studies (e.g., Ayers, Cloyd, and Robinson 2002; Dhaliwal, Li and Trezevant 2003; Dhaliwal, Li, Moser, and Krull 2005; and Scholes, Wolfson, Erickson, Maydew and Shevlin 2009). Our evidence arises from two sources: a theory-based analysis, and empirical …ndings that are consistent with the predictions of our analysis. Both sources suggest that lower liquidity ampli…es and higher liquidity attenuates the general positive relation between the spread between the dividend tax rate and the capital gains tax rate and …rms’expected pretax rates of return, hereinafter referred to as “expected rates of return.” Although the prediction that tax capitalization is ampli…ed when liquidity is lower and attenuated when liquidity is higher may not be surprising to some readers, to our knowledge no one has posited or tested this prediction. We suspect that this is the result of prior theory-based studies of tax capitalization having been motivated by an after-tax Capital Asset Pricing Model (CAPM) (e.g., Brennan 1970; Gordon and Bradford 1980; Guenther and Sansing 2010). The CAPM is a model of perfect competition; as such, it harbors no notion of liquidity (see, e.g., Lambert, Leuz, and Verrecchia 2012). In contrast, our analysis is motivated by imperfect competition, and liquidity plays a central role in imperfectly 1 competitive markets. The intuition for our analysis is very simple, and compelling in its simplicity. As the market for a …rm’s shares becomes less liquid, share price becomes more sensitive to changes in features of the economy. Share price is more sensitive in a less liquid market because less liquidity implies that there is less “cushion”to absorb shocks to, or changes in, the economy. Features of an economy whose change might result in price shocks include investor-level tax rates, investors’ aversion to risk, the quality of information about the …rm, etc. As these features change, and as the market becomes less liquid, heightened price sensitivity implies that the absolute value of the percentage change in price increases. For example, as the market for a …rm’s shares becomes increasingly less liquid, a reduction in an investorlevel tax rate leads to an increasingly larger percentage increase in price. Similarly, as the market becomes increasingly less liquid, an increase in an investor-level tax rate leads to an increasingly larger percentage decline in price. Because the absolute value of the percentage change in price is associated negatively with liquidity, lower liquidity will amplify the e¤ect of a tax rate change on a …rm’s expected rate of return, whereas higher liquidity will attenuate the e¤ect. This implies that when one assesses the behavior of a …rm’s expected rate of return, the e¤ect of a change in a tax rate cannot be divorced from the level of liquidity in the market for the …rm’s shares. We center the empirical test of our prediction about the role of liquidity around the Jobs and Growth Tax Relief Reconciliation Act of 2003 (JGTRRA03), which decreased the maximum statutory individual-level dividend tax rate from 38.1 percent to 15 percent, and 2 the maximum statutory individual-level capital gains tax rate from 20 percent to 15 percent. In other words, JGTRRA03 eliminated the tax penalty on dividend income.1 Prior studies (Dhaliwal, Krull, and Li 2007; Auerbach and Hassett 2007) …nd that expected rates of return (i.e., cost of equity capital) fell for both dividend-paying and non-dividend-paying …rms following JGTRRA03. We begin by replicating prior studies’ …nding that expected rates of return decrease following JGTRRA03. Our primary proxy for expected rates of return are annual estimates of cost of equity capital, hereinafter referred to simply as “cost of capital.”We use four di¤erent models to estimate the implied cost of capital. These models are those suggested in Claus and Thomas (2001); Gebhardt, Lee, and Swaminathan (2001); Ohlson and Juettner-Nauroth (2005), as implemented in Gode and Mohanram (2003); and Easton (2004). We also use a measure that equals the average of the estimates from the four above models. Our sample includes annual measures of cost of capital estimated over the period 2000-2006, excluding the second quarter of 2003 since JGTRRA03 was enacted in May 2003. Using each of our measures of implied cost of capital, we …nd that expected rates of return decrease following JGTRRA03, consistent with prior studies. We next test our prediction that expected rates of return decrease more for less liquid …rms following JGTRRA03. We use Amihud’s (2002) measure of price impact as our measure of illiquidity. Amihud’s ratio gives the absolute percentage price change per dollar of daily trading volume, or the daily price impact of the order ‡ow. In this sense, the measure is 1 Following Poterba and Summers (1985), prior literature de…nes the "dividend tax penalty" as the dividend tax rate minus the capital gains tax rate, divided by one minus the capital gains tax rate. In this paper, we simply refer to the dividend tax penalty to describe that, with the exception of the period since May 2003, dividend income has historically been taxed at a higher rate than capital gain income. 3 consistent with Kyle’s (1985) concept of illiquidity ( ), or the response of price to order ‡ow. Amihud’s (2002) measure varies inversely with a …rm’s level of liquidity. Consistent with our prediction, we …nd that the decrease in each of our measures of implied cost of capital following JGTRRA03 is signi…cantly greater for less liquid …rms. Next, we examine the e¤ect of liquidity on the relation between dividend yield, the dividend tax penalty, and expected rates of return. Consistent with Dhaliwal, Krull and Li (2007) and Auerbach and Hassett (2007), we …nd that expected rates of return decrease with dividend yield following JGTRRA03. Moreover, consistent with our prediction, we …nd that this decrease is ampli…ed by lower liquidity. In addition to presenting a new cross-sectional prediction related to dividend tax capitalization, our paper has two implications for prior work on tax capitalization. First, our results have an important implication for prior studies that conclude that expected rates of return decrease more for non-dividend-paying …rms than for dividend-paying …rms following JGTRRA03. Some view this result in prior studies as surprising based on the expectation that expected rates of return of dividend-paying …rms would be a¤ected by both tax rate cuts, whereas expected rates of return of non-dividend-paying …rms would only be a¤ected by the capital gains tax rate cut. One possible explanation for the surprising result is that non-dividend-paying …rms are signi…cantly less liquid than dividend-paying …rms, and prior studies do not control for the impact of liquidity on the relation between investor-level tax rates and expected rates of return. We replicate the …nding in prior studies that expected rates of return decrease more for non-dividend-paying …rms than for dividend-paying …rms, 4 but we are only able to do so when we use the implied cost of capital estimate generated from the model suggested in Claus and Thomas (2001). However, once we control for the e¤ect of liquidity on the relation between investor-level tax rates and expected rates of return, we no longer …nd that the change in expected rates of return of non-dividend-paying …rms is signi…cantly di¤erent than the change in expected rates of return of dividend-paying …rms. Second, this paper sheds light on the debate regarding whether the degree of dividend tax capitalization is in‡uenced by the tax status of the marginal investor. Several prior studies hold that dividend tax capitalization is stronger the more likely it is that the marginal investor is a taxable investor. These studies use institutional investor ownership to proxy for tax-exempt and corporate ownership and predict that tax capitalization is weaker the greater a …rm’s institutional investor ownership (e.g., Ayers, Cloyd, and Robinson 2002; Dhaliwal, Li and Trezevant 2003; Dhaliwal, Li, Moser, and Krull 2005; Dhaliwal, Krull, and Li 2007; Campbell, Chyz, Dhaliwal, and Schwartz 2011).2 Our theoretical and empirical analyses suggest that the mitigating force of institutional investor ownership on dividend tax capitalization documented in prior studies could be the result of an omitted correlated variable: liquidity. Consistent with prior studies, we …nd that institutional investor ownership mitigates the positive relation between dividend yield, the dividend tax rate and expected rates of return. However, once we control for the e¤ect of liquidity, we no longer …nd that institutional ownership serves a mitigating role. The paper proceeds as follows. We present our theoretical analysis and its predictions 2 Miller and Scholes (1982) also hold that the equilibrium price of the …rm is determined by a single marginal investor whose identity determines the pricing of taxes. 5 in Section 2. In Section 3, we present our empirical tests of the prediction that results from our theoretical analysis. Section 4 concludes. 2 The E¤ect of Liquidity on Tax Rate Changes Our goal is to understand how liquidity a¤ects the relation between investor-level tax rates and expected rates of return. Toward that goal, …rst we describe a market setting where liquidity and tax rates play salient roles, and then derive the price of a …rm’s shares in this setting. Next we study how liquidity a¤ects the impact of tax rate changes on expected rates of return. The results of our analysis suggest that lower liquidity will amplify the impact of tax rate changes, whereas higher liquidity will attenuate the impact of tax rate changes. In Section 3, using changes in the maximum statutory tax rates applied to dividend income and capital gains, we empirically test our prediction that lower liquidity ampli…es the e¤ect of tax rate changes on expected rates of return. 2.1 Share price in an imperfectly competitive market We consider a one-period economy that consists of one …rm, a risk-free asset, and some number of investors. The risk-free asset is a risk-free bond that yields a (pre-tax) return of $1 + Rf , where Rf 0 is the (pre-tax) risk-free rate. Let S > 0 represent the number of shares of stock the …rm supplies to the economy. Each share of stock generates uncertain cash ‡ow of V~ at the end of the period, where V~ = V represents the (per-share) realization of the cash ‡ow. This implies that the …rm generates total (uncertain) cash ‡ow of S V~ . In our analysis the role of S is not benign because as S increases investors in the economy 6 have to bear increasingly more risk. As we show below, this depresses the price of the …rm’s shares despite the fact that it has no e¤ect on per-share cash ‡ow (i.e., no e¤ect on V~ ). We summarize the notation we employ in Table 1. [INSERT TABLE 1 HERE] Let P represent the per-share price of the …rm’s shares at the beginning of the period. We assume there are N 3 investors in the economy, each of whom has identical informa- tion about the distribution of end of period cash ‡ow.3 Because investors have homogeneous beliefs, we suppress the speci…cs of their information and simply assume they believe that h i ~ the …rm’s cash ‡ow, V , has a normal distribution with expected value E V~ and variance h i V ar V~ . We assume that each investor has a negative exponential (or CARA) utility function for an amount w given by r (1 exp [ w=r]), where r is the an investor’s constant absolute risk tolerance. We assume that all investors are subject to tax and pay tax at the end of the period. The realization of the …rm’s end-of-period cash ‡ow, V~ = V , is paid out to investors at the end of the period in two ways: 1) as a dividend d; and 2) as a residual, net-of-dividend, cash payment V d. This implies that investors pay tax on: 1) dividends, d, at a dividend tax rate of ; 2) the residual cash payment, V d, at a capital gains tax rate of t; and 3) the cash proceeds from their investment in the risk-free bond (interest income) at the tax rate of , where 0 , t, < 1. In other words, investors’after-tax proceeds on their 3 Imperfect competition settings of the type we study require the participation of at least three investors so as to eliminate the possibility of one investor, or a pair of investors, having too much monopoly power: see the discussion on p. 329 of Kyle (1989). 7 investments are: 1) (1 ) d on their dividend payout; 2) (1 t) V~ d P + P on their residual cash payment (this expression follows from the discussion of capital gains tax in Sikes and Verrecchia 2011); and 1 + (1 ) Rf on the cash proceeds from their investment in the risk-free bond. To ensure that the analysis is tractable, we make two “facilitating” assumptions. First, at the appropriate time we set the capital gains tax rate to 0: that is, t = 0. We do not do so immediately, however, but instead defer until we can show that setting t = 0 is without loss of generality. Second, by assuming that all investors are subject to tax we rule out the possibility that only some proportion of investors are subject to tax (as in, e.g., Guenther and Sansing 2006, 2010; Sikes and Verrecchia 2011). Assuming that all investors are subject to tax has no qualitative e¤ect on the results we report below. The reason for this is that changing the mix or proportion of investors who are subject to tax will a¤ect the magnitude of a tax rate change on share prices, but our results concern the marginal e¤ect of changes in liquidity and tax rates on expected rates of return and here the proportion plays no role. In other words, increasing the proportion of investors who are not subject to tax will reduce the magnitude of a tax rate change on share prices, but it will not a¤ect the direction of the comparative static that is associated with that change. Before proceeding with our analysis, we attempt to describe in broad terms the market process that leads to a determination of …rm share price, P . As in Kyle (1989), our analysis characterizes a market process where at the beginning of the period each investor submits a demand function for the …rm’s shares to a Walrasian auctioneer. The auctioneer aggregates together investors’ demands, and then determines a single price for the shares such that 8 at that price demand for shares equals the supply of those shares. As is standard in any rational expectations setting, each investor’s demand for shares maximizes an investor’s expected utility as a function of the price that clears the market. In this sense our marketclearing process is identical to the vast literature on rational expectations equilibria in market settings.4 What distinguishes our market-clearing process from the rational expectations literature based on perfect competition, however, is that we assume that each investor’s demand function incorporates a belief as to how his demand a¤ects prices, and this belief is self-sustaining in equilibrium. To characterize the market-clearing process in greater detail, we start by letting Di represent investor i’s demand for shares of the …rm, Bi represent investor i’s demand for the risk-free bond, and F undsi represent investor i’s funds available to invest in shares of the …rm and the risk-free bond. The after-tax proceeds to investor i for holding Di shares of the …rm is Di (1 ) d + (1 t) V~ d P +P = Di (1 t) V~ ( t) d + tP : Thus, in determining her demand, Di , for shares in the …rm and the risk-free bond, Bi , each investor solves the following objective function max E r 1 Di ; Bi exp 1 Di (1 r t) V~ ( t) d + tP + (1 + (1 ) Rf ) Bi ; subject to the budget constraint Di P + Bi = F undsi : 4 See, for example, Grossman and Stiglitz (1980), Hellwig (1980), Diamond and Verrecchia (1981), Admati (1985), etc. 9 Substituting Bi = F undsi max E r 1 Di ; Bi Di P allows us to re-express the objective function as 1 Di (1 r exp t) V~ ( t) d + tP + (1 + (1 ) Rf ) (F undsi Di P ) simplifying this expression yields max E[r(1 Di ; Bi exp[ 1 (Di (1 r + (1 + (1 t) V~ ( t) d (1 t + (1 ) Rf ) P (1) ) Rf ) F undsi ))]]: Taking the expectation of eqn. (1) using standard techniques for evaluating the moment generating function yields max r(1 Di ; Bi 1 exp[ (Di (1 r + (1 + (1 h i t) E V~ ( t) d (1 11 ) Rf ) F undsi ) + 2 Di2 (1 2r t + (1 ) Rf ) P h i t) V ar V~ ]): 2 (2) In contrast to a model of perfect competition, a model of imperfect competition posits that investors take into consideration the e¤ect of their demand on price. Speci…cally, and following Kyle (1989) and Lambert et al. (2012), we characterize imperfect competition as the self-sustaining belief by each investor that he faces an upwardly-sloping price curve for …rm shares. In particular, we assume each investor believes that his demand is related to price as follows: P = p0 + Di ; where p0 is the intercept in the determination of P that is unrelated to an investor’s demand and is the coe¢ cient applied to an investor’s demand. We interpret illiquidity associated with an individual investor’s demand. When 10 as the degree of is small, an investor’s ; demand moves price less, and thus the market for …rm shares is more liquid with respect to demand; when is large, an investor’s demand moves price more, and thus the market is less liquid for …rm shares. Each investor treats as …xed when he determines the demand that maximizes his expected utility; our goal below is to determine a self-sustaining . Returning to an investor’s optimization problem, an investor solves for Di based on his belief that his demand a¤ects …rm share price through the relation P = p0 + Di . Based on this belief, an investor solves for Di in eqn. (2) by substituting into this equation the relation P = p0 + Di , taking the derivative of the expression that results with respect to Di , and then setting the result equal to 0. In other words, an investor takes the derivative of max r(1 Di ; Bi h i t) E V~ 1 (Di (1 r exp[ + (1 + (1 ( t) d ) Rf ) F undsi ) + (1 11 2 D (1 2 r2 i t + (1 ) Rf ) (p0 + Di ) h i t)2 V ar V~ ]) with respect to Di and sets the result equal to 0: this yields the …rst-order condition 0 = (1 h i t) E V~ ( ((1 t) d (1 t + (1 t) + (1 ) R f ) Di ) Rf ) (p0 + Di ) 1 Di (1 r h i t)2 V ar V~ : (3) Substituting P for the relation p0 + Di back into eqn. (3) and solving for the investor’s optimal demand yields Di = 1 (1 r (1 h i t) V ar V~ + ((1 2 h i t) E V~ ( t) d 11 1 t) + (1 (1 ) Rf ) t + (1 ) Rf ) P : (4) Market clearing requires that investors’demand for …rm shares equals the supply of those shares. That is, P must satisfy N X Di (P ) S = 0; i=1 where here we characterize an investor’s demand, Di ( ), as a function of the price of the …rm’s shares, P . Substituting for Di (P ) from eqn. (4) and simplifying yields the following result: (1 P = 1 h i t) E V~ t + (1 ( ) Rf t) d 0 @ 1 r 1 (1 h i t)2 V ar V~ t + (1 1 S + A : ) Rf N (5) We codify our analysis to this point in the following proposition. Proposition 1. In an imperfectly competitive economy where N identically informed investors compete to hold shares in a …rm, the price of the …rm’s shares is h i h i 0 1 2 1 ~ (1 t) V ar V (1 t) E V~ ( t) d S @r + A : P = 1 t + (1 ) Rf 1 t + (1 ) Rf N Proposition 1 implies that the …rm’s share price can be expressed as three components. The …rst component, ponent, (1 t)E [V~ ] ( t)d , 1 t+(1 )Rf t)2 V ar [V~ ] S , 1 t+(1 )Rf N 1 (1 r is the …rm’s expected after-tax cash ‡ow. The second com- is the discount in price that manifests in a setting where competition among risk averse investors to hold …rm shares is perfect. The third component, S , N is an additional discount in the expression for price that measures the extent to which the market is illiquid (i.e., the extent to which the market is not perfectly competitive). As we alluded to above, both discounts increase as S increases because investors in the economy have to bear increasingly more risk. To expand on this issue brie‡y, consider the possibility that the economy is perfectly competitive. Perfect competition is tantamount to assuming 12 = 0, and here the price of the …rm’s shares in eqn. (5) reduces to h i (1 t) E V~ ( t) d P = 1 t + (1 ) Rf 1 r 1 (1 h i t)2 V ar V~ S : t + (1 ) Rf N Alternatively, if competition is imperfect then this is tantamount to assuming S : N thus price absorbs the additional discount > 0 and see, for example, the discussion in Lambert et al. (2012). 2.2 Solving for The next step in the analysis is to solve for . Although the solution follows directly from Kyle (1989) (see also Lambert et al. 2012) and thus is straightforward, nonetheless it is complicated in detail. Thus, we relegate the solution to the Appendix and here simply state the result: solving for yields = Combining the expressions for 1 N (1 2 r (1 h i t)2 V ar V~ t + (1 ) Rf ) (6) : in eqn. (6) and P in Proposition 1 yields the following result. Proposition 2. In an imperfectly competitive economy where N identically informed investors compete to hold shares in a …rm, one can show that h i 2 (1 t) V ar V~ 1 ; = N 2 r (1 t + (1 ) Rf ) and thus the price of shares in the …rm reduces to h i (1 t) E V~ ( t) d (N 1) (1 P = 1 t + (1 ) Rf N (N 2) r (1 13 h i t)2 V ar V~ t + (1 ) Rf ) S: Note that the expression for price in Proposition 2 is unique and well-de…ned, provided that N 3. Combining the expressions for Di in eqn. (4), in eqn. (6), and P in Proposition 2, one can show that in equilibrium each investor holds Di = S N shares in the …rm. This result is intuitive: in equilibrium, each of N identical investors holds 1=N th of the total number of the …rm’s shares, S. This implies that each investor bears the risk associated with uncertain cash ‡ow of S N V~ , and in aggregate investors bear the risk of the …rm’s total cash ‡ow of S V~ . In the expression for P in Proposition 2, note that the e¤ect of a dividend on …rm share price is to reduce the …rm’s (per-share) expected cash ‡ow net of capital gains tax, h i (1 t) E V~ , by the dividend multiplied by the di¤erence between the dividend tax rate and the capital gains tax rate, ( t) d. In other words, (per-share) after-tax cash ‡ow adjusts for the fact that dividends and capital gains are taxed at di¤erent rates. We represent the di¤erence in tax rates as T : that is, T = t. The focus of our attention is on the impact of an increase in the di¤erence between the dividend tax rate and the capital gains tax rate t on the …rm’s expected rate of return: in other words, the impact of an increase in T . For example, T can increase if: 1) the dividend tax rate increases holding the capital gains tax rate t …xed; 2) the capital gains tax rate t falls holding the dividend tax rate 3) the dividend tax rate …xed; or simultaneously increases while the capital gains tax rate t falls. Because the focus of our attention is on T and not on or t per se, henceforth we assume without loss of generality that the capital gains tax rate t is 0: that is t = 0. Note that by “without loss of generality,” we mean that all of our results, and in particular Proposition 14 3 below, hold for t > 0: the role of setting t = 0 is simply to ease the burden of expressing …rm share price and expected rates of return in terms of t. The following expression for …rm share price results from setting t = 0. Corollary to Proposition 2. Setting the capital gains tax rate t to 0 (i.e., t = 0) in Proposition 2 yields = 1 N h i V ar V~ 2 r (1 + (1 ) Rf ) ; and thus the price of shares in the …rm becomes h i h i E V~ T d (N 1) V ar V~ P = S: 1 + (1 ) Rf N (N 2) r (1 + (1 ) Rf ) Our next step is to use the results of this section to study expected rates of return. 2.3 Comparative statics Using the expression for the price of the …rm’s shares in the Corollary to Proposition 2, the …rm’s expected rate of return is h i " # E V~ V~ P E = P P 1= E [V~ ] T d 1+(1 )Rf h i E V~ (N 1)V ar [V~ ] N (N 2)r(1+(1 )Rf ) 1: (7) S Note that tax rates a¤ect the …rm’s expected rate of return in two ways: 1) through the di¤erence between the dividend tax rate and the capital gains tax rate, T ; and 2) through the tax rate on the risk-free bond (interest income) of . Our goal is to understand how liquidity a¤ects the impact of a change in T on the …rm’s expected rate of return. Toward achieving that goal, …rst we have to address three issues. The …rst issue is that in our analysis, which is sometimes referred to as “Kyle’s ” in reference to Kyle (1985, 1989), is a measure of illiquidity. Consequently, we re-state our goal 15 as one of determining the marginal e¤ect of a change in illiquidity on the marginal e¤ect of a change in the tax rate on the …rm’s expected rate of return. This goal can be expressed mathematically as one of determining the sign of the cross-partial derivative " # @ @ V~ P E : @ @T P The second issue is that in our analysis is an endogenous variable, and so properly we should not be using it to take a derivative. A technique for addressing this issue is to reverse the roles of and N . Speci…cally, N is ostensibly an exogenous parameter, but by appealing to eqn. (6) and setting t = 0 we can de…ne N in terms of , as opposed to de…ning in terms of N : in other words, eqn. (6) implies N= h i V ar V~ r (1 + (1 ) Rf ) + 2: This technique treats N as an endogenous variable and (8) as an exogenous parameter. In turn, this allows us to re-express the price of the …rm’s shares in the Corollary to Proposition 2 in terms of as P = h i E V~ 1 + (1 T d r (1 + (1 ) Rf 2r (1 + (1 h i ) Rf ) + V ar V~ h i S; ) Rf ) + V ar V~ (9) and then re-express the …rm’s expected rate of return in eqn. (7) in terms of as h i " # E V~ V~ P E = 1: (10) E [V~ ] T d r (1+(1 )Rf )+V ar [V~ ] P S 1+(1 )Rf 2r (1+(1 )Rf )+V ar [V~ ] h i h i ~ ~ @ This allows us to compute @@ @T E V P P by representing E V P P by the right-hand-side of eqn. (10). 16 To digress brie‡y, note that if we require that N 3 for the expression for price in Proposition 2 to be unique and well de…ned, then in eqn. (8) it must be the case that V ar [V~ ] r(1+(1 )Rf ) 1, or h i V ar V~ r (1 + (1 Thus, one feature to the approach of treating ) Rf ) : (11) as an exogenous parameter is that it restricts to levels of illiquidity that are not “excessively large,” where the latter is de…ned by the right-hand-side of eqn. (11). That said, this restriction is simply the reciprocal of the requirement that at least three investors compete to hold …rm shares (i.e., N 3), which, arguably, is not an especially burdensome, real-world, institutional assumption. The third issue is that we would like to eliminate from consideration potentially perverse cases that arise from the possibility that the price of the …rm’s shares is negative. We interpret a negative share price as a circumstance where investors have to be subsidized to hold a …rm’s shares - once again, an unlikely outcome in real-world institutional settings. A necessary condition that share price be nonnegative is that end-of-period expected cash h i ‡ow is nonnegative: that is, E V~ 0. Even in the presence of nonnegative expected cash ‡ow, however, share price can be negative if, for example: 1) the …rm’s expected cash ‡ow h i h i is too low in relation to the variance of its cash ‡ow (i.e., E V~ is low and V ar V~ high); 2) investors’tolerance for risk is too low (i.e., r is low); and/or 3) the supply of the …rm’s shares is too high (i.e., S is high). Recall that we assume that S > 0. Thus, an inequality that restricts the supply of the …rm’s shares in relation to other (exogenous) parameters so 17 as to ensure that share price is nonnegative is 0<S E [V~ ] T d 1+(1 )Rf (1+(1 2r (1+(1 r )Rf )+V ar [V~ ] )Rf )+V ar [V~ ] (12) : Henceforth we assume this to be the case. Having addressed these three issues, next we calculate @ @ E @ @T h V~ P P i under the assump- tion …rm share price is nonnegative: that is, eqn. (12) holds. The calculation is tedious, and so here we simply report the result: the marginal e¤ect of an increase in illiquidity on the marginal e¤ect of an increase in the di¤erence between the dividend tax rate and the capital gains tax rate on expected rates of return is positive. We state this formally as our third proposition and provide a proof in the Appendix. Proposition 3. In an imperfectly competitive economy where N identically informed (and endogenously determined) investors compete to hold shares in a …rm and the resulting price of those shares is nonnegative, the marginal e¤ect of an increase in illiquidity, , on the marginal e¤ect of an increase in the di¤erence between the dividend tax rate and the capital gains tax rate, T , on expected rates of return is positive. The economic intuition for Proposition 3 is that as the market for a …rm’s shares becomes less liquid, share price becomes more sensitive to changes in features of the economy. Share price is more sensitive in a less liquid market because less liquidity implies that there is less “cushion” to absorb shocks to, or changes in, the economy. Consequently, lower liquidity will amplify the e¤ect of tax rate changes, whereas higher liquidity will attenuate the e¤ect of tax rate changes. This implies that in assessing the behavior of expected rates of return, the e¤ect of a change in a tax rate cannot be divorced from the level of liquidity in trade in 18 a …rm’s shares. 3 Empirical Tests In this section, we empirically test our prediction that lower liquidity ampli…es and higher liquidity attenuates the positive relation between the expected rates of return and the excess of the dividend tax rate over the capital gains tax rate. Our tests are centered around the Jobs and Growth Tax Relief Reconciliation Act of 2003 (JGTRRA03), which decreased the maximum statutory individual-level dividend tax rate from 38.1 percent to 15 percent and the maximum statutory individual-level long-term capital gains tax rate from 20 percent to 15 percent, and thereby reduced the spread between the dividend tax rate and the capital gains tax rate from 18.1 percent to zero. Prior studies (Dhaliwal, Krull, and Li 2007; Auerbach and Hassett 2007) …nd that expected rates of return (i.e., cost of capital) decrease for both dividend-paying and non-dividend-paying …rms.5 We begin by …rst testing whether we …nd the same relation using our measures of expected rate of return. We estimate the following 5 Dhaliwal, Krull and Li (2007) compare quarterly implied cost of equity capital estimates for the six quarters before to the estimates for the six quarters after JGTRRA03 (where they include the enactment quarter in the post period). Auerbach and Hassett (2007) conduct an event study in which they analyze the market reaction over eight …ve-day windows within which signi…cant news concerning the likelihood of passage of the dividend tax rate cut was made public. 19 Ordinary Least Squares (OLS) regression: EXP ECT ED RET U RNi;y = 4 SIZEi;y 1 8 LEVi;y 1 + + 5 BMi;y 1 + 2 P OST ACTy + 3 IN STi;y 1 6 DISP ERSIONi;y 1 + 7 ROEi;y 1 1 + 9 BET A_M KT RFi;y 1 11 BET A_HM Li;y 1 + + 10 BET A_SM Bi;y 1 12 BET A_U M Di;y 1 + + + + 13 42 IN DU ST RYi;y 1 + ": (13) Our proxies for expected rates of return are annual estimates of implied cost of capital. We use four di¤erent models to estimate the implied cost of capital measures. These models are those suggested in Claus and Thomas (2001); Gebhardt, Lee, and Swaminathan (2001); Ohlson and Juettner-Nauroth (2005), as implemented in Gode and Mohanram (2003); and Easton (2004).6 The basic idea of all four models is to substitute price and analyst forecasts into a valuation equation and to back out the cost of capital as the internal rate of return that equates current stock price with the expected future sequence of residual incomes or abnormal earnings. The individual models di¤er with respect to the use of analyst forecast data, the assumptions regarding short-term and long-term growth, the explicit forecasting horizon, and whether and how in‡ation is incorporated into the steady-state terminal value. We also use a measure that equals the average of the estimates generated by the four models. Our sample includes all …rm-year observations associated with …scal years ending in 1999 6 Dhaliwal, Krull and Li (2007) use quarterly estimates based on the models in Claus and Thomas (2001); Gebhardt, Lee, and Swaminathan (2001); and Ohlson and Juettner-Nauroth (2005), as implemented in Gode and Mohanram (2003). We use annual as opposed to quarterly estimates to avoid the error in quarterly forecasts (which tends to average out if aggregated over a year) and because of the seasonality in quarterly data. We thank Luzi Hail and Christian Leuz for sharing these annual measures of cost of capital with us and refer readers to the appendix of Hail and Leuz (2006) for a detailed explanation of the calculation of each measure. 20 through 2005, with the exception that we exclude …rm-year observations with …scal yearends within the window of June 2002 through August 2002. We exclude these …rm-year observations because cost of capital is estimated ten months after …scal year-end; thus, the implied cost of capital for these observations is estimated in the second calendar quarter of 2003, which is the enactment quarter of JGTRRA03.7 We set POSTACT equal to one if implied cost of capital is estimated after the second quarter of 2003 and equal to zero if implied cost of capital is estimated before the second quarter of 2003.8 Consistent with prior literature, we control for the impact of information asymmetry on a …rm’s expected rate of return. We include several controls for information asymmetry. The …rst is analyst forecast dispersion, measured as the standard deviation of analysts’ forecasts of …rm i’s one-year-ahead earnings per share divided by the absolute value of the mean forecast of …rm i’s one-year-ahead earnings per share (with zero mean forecast observations excluded from the sample) (Diether, Malloy, and Scherbina 2002; Johnson 2004; Sadka and Scherbina 2007). The measure is right-skewed; thus, we use the natural logarithm of the measure (DISPERSION ). Other control variables that prior studies …nd are potentially associated with information asymmetry are the percent of outstanding shares owned by institutional investors (INST ), …rm size measured as the natural logarithm of market capitalization (SIZE ), book-to-market ratio (BM ), and leverage measured as the 7 The results reported in the paper are the same if we instead only eliminate observations associated with …scal years ending in July 2002 (i.e., observations for which the cost of capital is estimated in May 2003). 8 The subscript y on the dependent variable and the subscript y-1 on all of the independent variables other than POSTACT denote that the implied cost of capital measures are estimated ten months after the …scal year-end. The year y coincides with the year in which implied cost of capital is estimated. 21 sum of current and long-term liabilities scaled by total assets (LEV ). One might expect for information asymmetry to be lower among larger …rms and …rms with greater institutional investor ownership and higher among …rms with higher book-to-market ratios and higher leverage (Sadka and Scherbina 2007). We control for pro…tability with the ratio of net income before extraordinary items divided by book value of equity (ROE ). We control for risk by including the coe¢ cient estimates from estimating a four-factor Fama-French-Carhart model (Fama and French 1993; Carhart 1997) using return data from the 48 months prior to year y-1 (BETA_MKTRF, BETA_SMB, BETA_HML, and BETA_UMD). We control for industry e¤ects by including an indicator variable for each of Fama and French’s 30 industry portfolios (Fama and French 1997). Consistent with Hail and Leuz (2006), we truncate the …ve cost of capital measures (r_ave, r_ct, r_oj, r_mpeg, r_gls) at the bottom and top one percent.9 We winsorize all continuous independent variables used in our analysis at the 1st and 99th percentiles. Table 2 includes variable de…nitions. [INSERT TABLE 2 HERE] Our sample includes …rms that are listed on either the New York Stock Exchange (NYSE), American Stocks Exchange (AMEX), or NASDAQ; are in the Center for Research in Security Prices (CRSP)/Compustat Merged Database; have data on quarterly institutional holdings reported on Form 13F available in the Thomson-Reuters database; and are followed by three for more analysts and have analyst forecast data in I/B/E/S. Panel A of Table 3 reports the descriptive statistics for the variables used in the analysis. All values are reported in decimal 9 The average of the four measures (r_ave) is calculated prior to the truncation. 22 format. The mean implied cost of equity capital over the sample period estimated using the models in Claus and Thomas (2001), Ohlson and Juettner-Nauroth (2005) as implemented in Gode and Mohanram (2003), Easton (2004), and Gebhardt, Lee, and Swaminathan (2001) equal 9.6 percent, 11.8 percent, 12 percent, and 5.3 percent, respectively. The mean implied cost of equity capital using the average of the four estimates equals 9.6 percent. Panel B reports the mean and median values for each of the …ve implied cost of capital measures in the period before and in the period after JGTRRA03. The mean and median value for each of the …ve cost of capital measures is signi…cantly lower in the period following JGTRRA03 than in the period preceding it, consistent with the general positive relation between investor-level tax rates and expected rates of return. [INSERT TABLE 3 HERE] Table 4 presents the results of estimating equation (13). Consistent with prior studies, using each of the implied cost of capital measures, we …nd that the implied cost of equity capital is signi…cantly lower in the period following JGTRRA03 than in the period preceding it. Focusing on column (1) in Table 4, the coe¢ cient on POSTACT suggests that cost of equity capital declines 118 basis points following JGTRRA03. This estimate is consistent with Dhaliwal, Krull and Li (2007), who estimate that the cost of capital is 160 basis points lower in the six quarters following JGTRRA03 than in the six quarters preceding it. [INSERT TABLE 4 HERE] We next test our prediction that expected rates of return decrease more for less liquid 23 …rms following JGTRRA03 by estimating the following OLS regression: EXP ECT ED RET U RNi;y = 4 ILLIQU IDIT Yi;y 1 6 SIZEi;y 1 10 LEVi;y 1 + + 1 + 2 P OST ACTy P OST ACTy + 7 BMi;y 1 + + 5 IN STi;y 1 8 DISP ERSIONi;y 1 11 BET A_M KT RFi;y 1 13 BET A_HM Li;y 1 + + 3 ILLIQU IDIT Yi;y 1 + + 9 ROEi;y 1 12 BET A_SM Bi;y 1 14 BET A_U M Di;y 1 + + + + 15 44 IN DU ST RYi;y + ": (14) Our proxy for lower levels of liquidity is Amihud’s (2002) measure of price impact. It equals the ratio of the daily absolute return to the dollar trading volume on that day, averaged over the trading days in year y-1 for which there is return and volume data, or s ! Di;y 1 X 1 jRi;y 1;d j ILLIQU IDIT Yi;y 1 = 1; 000 Di;y 1 d=1 V OLDi;y 1;d where jRi;y 1 ;d j is …rm i’s absolute return on day d of year y-1, VOLD i;y daily volume in dollars, and D i;y 1 1;d is the respective is the number of days for which data are available for …rm i in year y-1. Amihud’s ratio gives the absolute percentage price change per dollar of daily trading volume, or the daily price impact of the order ‡ow. The measure is consistent with Kyle’s (1985) concept of illiquidity ( ), or the response of price to order ‡ow. Similar to Chen, Goldstein, and Jiang (2010), we use the square root version of the Amihud (2002) measure. Consistent with Amihud (2002), we apply several restrictions when calculating the ratio. First, we require for there to be stock return and volume data for more than 200 days of year y-1 for …rm i in order to calculate its ratio. Second, we require for the stock price to be greater than $5 at the end of year y-1. Third, data to calculate …rm i’s 24 market capitalization must be available at the end of year y-1. Fourth, after applying the restrictions above, we winsorize observations for the measure at the 1st and 99th percentiles for each year. All other variables are previously de…ned. A negative 4 will be consistent with our prediction. Table 5 presents the results. The coe¢ cient on ILLIQUIDITY is positive and significant in all …ve columns, consistent with expected rates of return being positively related to illiquidity (see, e.g., Amihud and Mendelson 1986; Brennan and Subrahmanyam 1996; Brennan, Chordia, and Subrahmanyam 1998; Datar, Naik, and Radcli¤e 1998; Haugen and Baker 1996, and Hu 1997 with respect to …rm-level liquidity, and Pastor and Stambaugh 2003 with respect to market-wide liquidity). Consistent with our prediction, we …nd that the decrease in each of the implied cost of capital estimates following JGTRRA03 is signi…cantly greater for less liquid …rms. In terms of economic magnitude, when the analysis is conducted using the average of the four estimates as the dependent variable, the results suggest that a one standard deviation increase in Amihud’s (2002) measure of price impact results in a 0.3 percentage point (i.e., 30 basis point) larger decrease in the expected rate of return following JGTRRA03. This represents a three percent change for the mean …rm. 10 [INSERT TABLE 5 HERE] We next examine 1) whether expected rates of return decrease with dividend yield following JGTRRA03, and 2) if so, whether the decrease is greater among less liquid …rms. To 10 We calculate the 0.30 percentage point as the coe¢ cient estimate on ILLIQUIDITY*POSTACT in column (1) (-0.0270) multiplied by standard deviation of ILLIQUIDITY (0.11). The three percent equals the 0.30 percentage point divided by the mean value of r_ave (0.096). 25 test 1), we estimate equation (14) below: EXP ECT ED RETURN i;y = 4Y IELDi;y 7 BMi;y 1 + 1 P OST ACTy + 1 + 2 P OST ACTy 5 IN STi;y 1 8 DISP ERSIONi;y 1 + + + 14 BET A_U M Di;y 1 15 44 IN DU ST RYi;y + 3Y IELDi;y 6 SIZEi;y 1 9 ROEi;y 1 11 BET A_M KT RFi;y 1 + 12 BET A_SM Bi;y 1 + + + ": 1 + + 10 LEVi;y 1 + 13 BET A_HM Li;y 1 + (15) The variable YIELD equals the amount of dividends paid to common shareholders in year 1 scaled by …rm i’s market capitalization, both collected from Compustat (Ayers, Cloyd, y and Robinson 2002). We set YIELD equal to zero for non-dividend-paying …rms. All other variables are previously de…ned. We demean the value of YIELD used in the interaction term by subtracting the sample mean.11 Table 6 presents the results. Using all of the implied cost of capital measures, the coe¢ cient on the interaction YIELD*POSTACT is negative and signi…cant, consistent with expected rates of return decreasing with dividend yield following JGTRRA03. [INSERT TABLE 6 HERE] We next estimate the following OLS regression to test whether the change in the relation between dividend yield and expected rates of return following JGTRRA03 is greater among 11 Demeaning, or centering, continuous variables used in interactions reduces the amount of multicollinearity that is induced by multiplying together two independent variables. See Aiken and West (1991) for a discussion of the bene…ts of demeaning variables. 26 less liquid …rms: EXP ECT ED RETURN i;y = 4Y IELDi;y 1 1 P OST ACTy + + 2 P OST ACTy + 3Y 5 HIGHILLIQU IDIT Yi;y 1 6 HIGHILLIQU IDIT Yi;y 1 P OST ACTy + 7Y HIGHILLIQU IDIT Yi;y IELDi;y 1 9 SIZEi;y 1 + 13 LEVi;y 1 + P OST ACTy 10 BMi;y 1 + 11 DISP ERSIONi;y 1 14 BET A_M KT RFi;y 1 16 BET A_HM Li;y 1 + IELDi;y + + 1 + + + + 8 IN STi;y 1 12 ROEi;y 1 15 BET A_SM Bi;y 1 17 BET A_U M Di;y 1 1 + + + 18 47 IN DU ST RYi;y + ": (16) HIGHILLIQUIDITY equals one if a …rm is in the top two quintiles when …rms are sorted into quintiles each year based on their values of ILLIQUIDITY, and equals zero otherwise. We demean the value of YIELD used in the interaction terms by subtracting the sample mean. Table 7 presents the results. With the exception of when the dependent variable is r_gls, the coe¢ cient on YIELD*POSTACT*HIGHILLIQUIDITY is negative and signi…cant, consistent with the change in the relation between dividend yield and expected rates of return following JGTRRA03 being greater among less liquid …rms. [INSERT TABLE 7 HERE] In summary, the results discussed above are consistent with the prediction generated from our theoretical analysis in Section 2. We …nd that lower liquidity ampli…es the decrease in expected rates of return following the 2003 cuts to investor-level tax rates, which reduced 27 the dividend tax penalty to zero. In addition to presenting a new cross-sectional prediction related to tax capitalization, our paper has two important implications for prior work on tax capitalization. We address each of the implications below. 3.1 Dividend vs. non-dividend paying …rms Auerbach and Hassett (2007) and Dhaliwal, Krull, and Li (2007) document that expected rates of return decrease more for non-dividend-paying …rms than for dividend-paying …rms following JGTRRA03. Some view this result as surprising based on the expectation that expected rates of return of dividend-paying …rms would be a¤ected by both tax rate cuts, whereas expected rates of return of non-dividend-paying …rms would only be a¤ected by the capital gains tax rate cut. Auerbach and Hassett (2007) and Chen, Dai, Shackelford and Zhang (2011) o¤er possible explanations for the unexpected result. Auerbach and Hassett (2007) explain that if investors viewed the dividend tax rate cut as permanent (or at least semi-permanent), then investors could have incorporated expected future dividend taxes when pricing the shares of …rms that currently were not paying a dividend but which investors expected to pay a dividend prior to a potential reversal in the dividend tax rate cut. Chen, Dai, Shackelford, and Zhang (2011) provide empirical evidence suggesting that expected rates of return fell more for non-dividend-paying …rms than for dividend-paying …rms because the former are more …nancially constrained and thus have the most pressing need for capital. We o¤er a third possible explanation for the result. Consistent with the prediction from our theoretical analysis, we posit that expected rates of return could have decreased more for non-dividend-paying than for dividend-paying …rms following JGTRRA03 due to the fact 28 that non-dividend-paying …rms are generally less liquid. In other words, prior studies that conclude that expected rates of return decrease more for non-dividend-paying …rms following JGTRRA03 could su¤er from an omitted correlated variable problem. In untabulated tests, we …nd that the non-dividend-paying …rms are signi…cantly less liquid than the dividendpaying …rms in our sample. The mean (median) ILLIQUIDITY equals 0.11 (0.06) for the non-dividend paying …rms and 0.07 (0.04) for the dividend-paying …rms, and the di¤erences between the means and medians are statistically signi…cant at the one percent level. We …rst examine whether, like Auerbach and Hassett (2007) and Dhaliwal, Krull, and Li (2007), we …nd that expected rates of return fall more for non-dividend-paying …rms than for dividend-paying …rms following JGTRRA03. We estimate the following OLS regression: EXP ECT ED RETURN i;y = + 4 DIVi;y 1 6 SIZEi;y 1 10 LEVi;y 1 P OST ACTy + + + 7 BMi;y 1 + 1 + 2 P OST ACTy 5 IN STi;y 1 + + 3 DIVi;y 1 + 8 DISP ERSIONi;y 1 11 BET A_M KT RFi;y 1 13 BET A_HM Li;y 1 + + 9 ROEi;y 1 12 BET A_SM Bi;y 1 14 BET A_U M Di;y 1 + + + 15 44 IN DU ST RYi;y + ": (17) The variable DIV is an indicator variable that equals one if …rm i pays a dividend to common shareholders in year y-1; zero otherwise. All other variables are previously de…ned. A positive 4 will be consistent with the …nding in Auerbach and Hassett (2007) and Dhaliwal, Krull, and Li (2007). Panel A of Table 8 reports the results. The coe¢ cient on DIV*POSTACT is positive and signi…cant only when r_ct is the dependent variable. It is insigni…cant when 29 the dependent variable is either r_oj, r_mpeg, or r_ave, and is negative and signi…cant when r_gls is the dependent variable. We next re-estimate equation (17) and include ILLIQUIDITY as an additional independent variable. The results are reported in Panel B of Table 8 and are very similar to those in Panel A. Then we re-estimate equation (17) and include ILLIQUIDITY as an additional independent variable as well as the interaction ILLIQUIDITY*POSTACT. Panel C of Table 8 reports the results. If the …nding in prior studies that expected rates of return decrease more for non-dividend-paying …rms than for dividend-paying …rms following JGTRRA03 is due to an omitted correlated variable, then once we include ILLIQUIDITY*POSTACT in the estimation, the coe¢ cient on DIV*POSTACT should no longer be positive. Indeed, this is what we …nd. The coe¢ cient on DIV*POSTACT when r_ct is the dependent variable is now insigni…cant. Moreover, the magnitude and statistical signi…cance of the negative coe¢ cient on DIV*POSTACT when r_gls is the dependent variable increase. Consistent with our prediction, the coe¢ cient on ILLIQUIDITY*POSTACT is negative and signi…cant in all columns. [INSERT TABLE 8 HERE] In summary, the results in this section suggest that the puzzling result in prior studies that expected rates of return decrease more for non-dividend-paying …rms than for dividendpaying …rms following JGTRRA03 could be due to prior studies not controlling for di¤erences in liquidity between dividend-paying and non-dividend-paying …rms. 30 3.2 Marginal investor Several prior studies predict that the dividend tax capitalization e¤ect is stronger the more likely it is that the marginal investor is a taxable investor (e.g., Dhaliwal, Li and Trezevant 2003; Ayers, Cloyd, and Robinson 2002; Dhaliwal, Krull, Li and Moser 2005; Dhaliwal, Krull, and Li 2007; Campbell, Chyz, Dhaliwal, and Schwartz 2011). These studies use the percent of a …rm’s outstanding shares owned by institutional investors to proxy for tax-exempt and corporate ownership and predict that institutional investor ownership mitigates dividend tax capitalization (i.e., dividend tax capitalization is weaker the more likely it is that a tax-exempt or corporate investor is the marginal investor setting price). Such a prediction is inconsistent with prior theoretical studies that rely upon the after-tax CAPM. The after-tax CAPM predicts that the weighted average tax rate of all investors in the economy (where the weight depends on investors’ risk tolerances) rather than the tax rate of the marginal investor in the …rm is the relevant tax rate in determining the extent of tax capitalization (see Guenther and Sansing (2010) and Hanlon and Heitzman (2010) for a discussion of the issue). Our theoretical analysis and empirical …ndings suggest that prior studies’ …nding that institutional investor ownership mitigates dividend tax capitalization could be due to an omitted correlated variable problem: stocks with greater institutional investor ownership are generally more liquid. Over our sample period, we …nd that the correlation between INST and ILLIQUIDITY is negative and statistically signi…cant (untabulated Pearson correlation = 0:43; p < 0:001). To investigate the issue further, we next estimate a modi…ed 31 version of equation (16) where we replace HIGHILLIQUIDITY with HIGHINST, which equals one if a …rm is in the top two quintiles when …rms are sorted into quintiles annually based on their value of INST and equals zero otherwise. As with the original estimation, we demean the value of YIELD used in the interactions. A positive coe¢ cient on YIELD*POSTACT*HIGHINST will be consistent with the …nding in prior studies that institutional ownership mitigates dividend tax capitalization. Panel A of Table 9 presents the results. The coe¢ cient on YIELD*POSTACT*HIGHINST is positive and signi…cant at the ten percent level when the dependent variable is r_ave, r_oj, or r_mpeg, and is insigni…cant when the dependent variable is r_ct or r_gls. Next, we examine whether the …nding in prior studies that institutional ownership mitigates dividend tax capitalization is due to an omitted correlated variable (i.e., liquidity). We estimate equation (16) as originally outlined and also include HIGHINST, HIGHINST*POSTACT, and YIELD*POSTACT*HIGHINST. If institutional ownership proxies for greater liquidity rather than the marginal investor being tax-insensitive to dividends, then once we control for liquidity, we should no longer …nd that institutional ownership mitigates dividend tax capitalization. reports the results. This is indeed what we …nd. Panel B of Table 9 The coe¢ cient on YIELD*POSTACT*HIGHINST is insigni…cant in all …ve columns. Consistent with our prediction and the results in Table 7, the coe¢ cient on YIELD*POSTACT*HIGHILLIQUIDITY is negative and signi…cant in all columns except when r_gls is the dependent variable, in which case it is insigni…cant. [INSERT TABLE 9 HERE] 32 In summary, the results in Table 9 suggest that prior studies that attribute the mitigating force of institutional ownership on dividend tax capitalization to the marginal investor being tax-insensitive su¤er from an omitted correlated variable problem. Once we control for the e¤ect of liquidity on the relation between investor-level tax rates and expected rates of return, we no longer …nd that institutional ownership mitigates dividend tax capitalization. We are not the …rst to disagree with the marginal investor interpretation. Prior theoretical studies posit that dividend tax capitalization is a function of the weighted-average tax rate across all investors, where the weight depends on investors’tolerances for risk (e.g., Brennan 1970; Gordon and Bradford 1980; Michaely and Vila 1995; Guenther and Sansing 2010; Bond, Devereux, and Klemm 2007). According to Guenther and Sansing (2010), one reason that prior empirical studies …nd that institutional holdings mitigate dividend tax capitalization is that institutional holdings are negatively correlated with individual (i.e., taxable) investors’ tolerances for risk. Guenther and Sansing’s (2010) suggestion that differences in investors’risk tolerances explain the relation between institutional holdings and dividend tax capitalization and our suggestion that liquidity explains this relation are not mutually exclusive explanations.12 12 The results presented in the paper are robust to replacing Amihud’s (2002) measure of illiquidity with a …rm’s e¤ective bid-ask spread (Chordia, Roll, and Subrahmanyam 2000; Korajczyk and Sadka 2008) and replacing the measures of implied cost of capital with a measure that is simultaneoulsy estimated with a …rm’s growth in residual earnings based on the model in Easton, Taylor, Shro¤, and Sougiannis (2002). For the sake of brevity, we omit these results from the manuscript. They are available upon request. 33 4 Conclusion This paper o¤ers a new theory related to cross-sectional variation in dividend tax capitalization. Unlike prior theory studies on dividend tax capitalization that rely on the after-tax CAPM, our analysis features an imperfectly competitive market in which liquidity plays a role in determining the e¤ect of investor-level tax rates on …rms’expected rates of return. The prediction generated from our analysis is that lower liquidity ampli…es the e¤ect of a change in the dividend tax penalty (i.e., the excess of the dividend tax rate over the capital gains tax rate) on expected rates of return, whereas higher liquidity attenuates the e¤ect. We empirically test our prediction using the cuts to the maximum statutory tax rate on dividend income (from 38.1 to 15 percent) and on capital gain income (from 20 percent to 15 percent), enacted by the Jobs and Growth Tax Relief Reconciliation Act of 2003. Combined, these cuts reduced the spread between the dividend tax rate and the capital gains tax rate from 18.1 percent to zero. We begin by con…rming the result in prior studies that expected rates of return decrease signi…cantly following JGTRRA03. Then, we show that expected rates of return decrease more for less liquid …rms following JGTRRA03, consistent with the prediction generated from our theoretical analysis. In addition to providing a new theory related to cross-sectional variation in tax capitalization, this paper o¤ers a couple of important implications for prior studies. First, prior studies conclude that expected rates of return decrease more for non-dividend-paying …rms than for dividend-paying …rms following JGTRRA03. Some view this result as puzzling based on the expectation that expected rates of return of dividend-paying …rms would be a¤ected by both 34 tax rate cuts, whereas expected rates of return of non-dividend-paying …rms would only be a¤ected by the capital gains tax rate cut. We posit a possible explanation for the puzzling result and provide empirical results to support our explanation. Our explanation is that, on average, non-dividend-paying …rms are signi…cantly less liquid than dividend-paying …rms. Once once controls for the e¤ect of liquidity on the relation between investor-level tax rates and expected rates of return, expected rates of return no longer appear to decrease more for non-dividend-paying …rms. Second, this paper contributes to the recent debate regarding whether the identity of the marginal investor determines the extent to which investor-level taxes are priced. Our results suggest that prior studies that attribute the attenuating force of institutional ownership on dividend tax capitalization to the marginal investor being tax-exempt (or tax-insensitive with respect to dividends) could su¤er from an omitted correlated variable problem. Once we control for the e¤ect of liquidity on the relation between investor-level tax rates and expected rates of return, we no longer …nd that institutional ownership attenuates dividend tax capitalization. 35 Appendix . To proceed, …rst we provide an alternative characterization of the The solution for investor’s demand strategy. In particular, in competing with other informed investors, we conjecture that each investor adopts a strategy Di = where is an intercept term and P; is the weight an investor places on P . For this conjectured strategy to be rational based on the computation of Di in eqn. (4), it must be the case that = = 1 (1 r 1 (1 r h i t)2 V ar V~ + ((1 1 t) + (1 h i t)2 V ar V~ + ((1 ) Rf ) (1 t + (1 (1 h i t) E V~ 1 t) + (1 ) Rf ) ) Rf ) and ( t) d : Market clearing requires that investors’demand for …rm shares equals the supply of those shares. Thus, P must satisfy N Di (P ) S = 0; where once again Di (P ) represents an investor’s demand as a function of P . Substituting Di = P into this equation and solving for the P that clears the market yields P = where (N S) ; is given by = 1 N 1 = 0 1 r 1 @ N 1 (1 h i t)2 V ar V~ t + (1 36 ) Rf 1 + A: Here, represents the marginal impact on price of an additional share brought to (or withdrawn from) the market by the …rm. To ensure the market clearing condition P = jecture that Di = P , p0 and p0 = S) in conjunction with the con- (N in the expression P = p0 + Di must be of the form ((N S) and 1) 1 = N 1 1 ; respectively. The rationale for this claim is that when this is the case P = p 0 + Di which implies P (1 + )= = ((N 1) = (N S) in eqn. (A1) yields 1 N P; = = 1 1 = 1 N 1 1 1 r This yields the result: = N = 1 1 N 1 in the identity = P) S) and thus (N 1+ solving for S) + ( 21 1 N 1 (A1) ; . We further solve for 0 @ 1 r (1 1 h i t) V ar V~ 2 t + (1 h i t)2 V ar V~ (1 t + (1 ) Rf ) Rf by substituting in for 1 + A: : Proof of Proposition 3. To start, we restate the expression for the …rm’s expected rate of return in eqn. (10) " E V~ P P # = E [V~ ] T d 1+(1 )Rf h i E V~ (1+(1 2r (1+(1 r 37 )Rf )+V ar [V~ ] S )Rf )+V ar [V~ ] 1: (10) Taking the derivative of the …rm’s expected rate of return with respect to T yields h i " # d E V~ @ V~ P E = ; @T P (1 + (1 ) Rf ) P 2 where P is de…ned as in eqn. (9): h i E V~ T d P = 1 + (1 ) Rf h i ) Rf ) + V ar V~ h i S: ) Rf ) + V ar V~ r (1 + (1 2r (1 + (1 It should be clear from the expression in eqn. (A2) that h i @ V~ P derivative of @T E P with respect to : this yields " @ @ V~ P E @ @T P # = (1 + (1 h V~ P P i ) Rf ) 2r (1 + (1 h i h i V ar V~ V ar V~ + 2r (1 + (1 h i V ar V~ + 2r (1 + (1 It should be clear from the expression in eqn. (A3) that the price of the …rm’s shares, P , is non-negative. Q.E.D. 38 h i ) Rf ) + V ar V~ ) Rf ) + 2r2 2 (9) > 0. Now take the h i 2K Sd E V~ where K is de…ned as K @ E @T (A2) (1 + (1 3 ; (A3) P3 ) Rf )2 ) Rf ) : @ @ E @ @T h V~ P P i > 0 provided that Table 1 Table of Notation N r Di number of investors risk tolerances of investors investor i’s demand for shares in the …rm Bi investor i’s demand for a risk-free bond V~ per-share cash ‡ow of the …rm d per-share dividend of the …rm Rf risk-free rate tax rate on dividends t tax rate on capital gains T the di¤erence between and t tax rate on risk-free investments P per-share price of shares in the …rm S supply of shares in the …rm illiquidity: the extent to which an investor’s demand moves P p0 intercept in P unrelated to an investor’s demand 43 Table 2 Variable Definitions = Average of r_ct, r_oj, r_mpeg, and r_gls = Implied cost of capital estimate based on the model suggested in Claus and Thomas (2001) = Implied cost of capital estimate based on the model suggested in Ohlson and JuettnerNauroth (2005), as implemented in Gode and Mohanram (2003) = Implied cost of capital estimate based on the model suggested in Easton (2004) = Implied cost of capital estimate based on the model suggested in Gebhardt, Lee, and Swaminathan (2001) POSTACT = 1 if the expected rate of return (i.e., cost of capital) is estimated after the second calendar quarter of 2003; 0 if the expected rate of return is estimated prior to the second calendar quarter of 2003 ILLIQUIDITY = square root version of Amihud's (2002) measure of price impact. See text for calculation. HIGHILLIQUIDITY = 1 if firm i is in the top two quintiles for the year when firms are sorted annually into quintiles according to their values of ILLIQUIDITY; 0 otherwise YIELD = firm i's dividends paid to common shareholders (DVC from Compustat) divided by firm i's market capitalization (CSHO*PRCC_F from Compustat); equal to zero for non-dividend paying firms DIV INST = = 1 if YIELD > 0; 0 otherwise % of firm i's outstanding shares owned by institutional investors HIGHINST = 1 if firm i is in the top two quintiles for the year when firms are sorted annually into quintiles according to their values of INST; 0 otherwise SIZE = BM = DISPERSION = ROE = LEV = BETA_MKTRF = firm i's loading on MKTRF from estimating Fama-French-Carhart 4-factor model using return data for the prior 48 months BETA_SMB = firm i's loading on SMB from estimating Fama-French-Carhart 4-factor model using return data for the prior 48 months BETA_HML = firm i's loading on HML from estimating Fama-French-Carhart 4-factor model using return data for the prior 48 months BETA_UMD = firm i's loading on UMD from estimating Fama-French-Carhart 4-factor model using return data for the prior 48 months r_ave r_ct r_oj r_mpeg r_gls natural log of firm i's market capitalization firm i's book equity (CEQ from Compustat) divided by market capitalization (CSHO*PRCC_F from Compustat) natural logarithm of the standard deviation of analysts' forecasts of firm i's oneyear-ahead earnings per share divided by the absolute value of the mean forecast of one-year-ahead earnings per share (with zero mean forecast observations excluded from the sample) firm i's net income before extraordinary items (IBCOM from Compustat) divided by book value of equity (CEQ from Compustat) sum of firm i's current and long-term liabilities (DLC+DLTT from Compustat) scaled by total assets (AT from Compustat) 44 Table 3 Descriptive Statistics Panel A: Descriptive Statistics for Variables in Implied Cost of Capital Regressions Variable r_ave r_ct r_oj r_mpeg r_gls POSTACT ILLIQUIDITY YIELD INST SIZE BM DISPERSION ROE LEV BETA_MKTRF BETA_SMB BETA_HML BETA_UMD N 10,275 10,964 10,316 10,245 10,696 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 11,085 Mean 0.096 0.096 0.118 0.120 0.053 0.594 0.089 0.012 0.627 21.059 0.490 -3.825 0.126 0.224 1.048 0.543 0.400 -0.143 Std. Dev. 0.028 0.029 0.032 0.046 0.031 0.491 0.110 0.018 0.227 1.508 0.315 1.129 0.281 0.187 0.756 0.833 1.064 0.598 5th Pctile 0.061 0.057 0.080 0.067 0.009 0.000 0.010 0.000 0.201 18.818 0.118 -5.313 -0.077 0.000 0.021 -0.537 -1.485 -1.152 25th Pctile 0.078 0.078 0.097 0.090 0.028 0.000 0.025 0.000 0.473 19.969 0.281 -4.644 0.066 0.055 0.565 0.024 -0.056 -0.398 Median 0.091 0.091 0.111 0.108 0.052 1.000 0.051 0.002 0.657 20.894 0.440 -4.007 0.120 0.206 0.970 0.417 0.520 -0.078 75th Pctile 0.109 0.107 0.131 0.137 0.075 1.000 0.108 0.019 0.802 22.007 0.625 -3.205 0.174 0.344 1.443 0.939 1.009 0.164 95th Pctile 0.149 0.149 0.178 0.210 0.104 1.000 0.301 0.048 0.963 23.936 1.024 -1.660 0.327 0.559 2.371 2.052 1.832 0.641 Panel B: Implied Cost of Capital Pre- Vs. Post-JGTRRA03 Pre-JGTRRA03 Mean Post-JGTRRA03 Mean Difference r_ave 0.105 0.090 0.014*** r_ct 0.105 0.089 0.016*** r_oj 0.127 0.112 0.015*** r_mpeg 0.130 0.113 0.018*** r_gls 0.056 0.051 0.005*** Pre-JGTRRA03 Median Post-JGTRRA03 Median Difference 0.100 0.087 0.013*** 0.100 0.088 0.012*** 0.120 0.107 0.013*** 0.119 0.103 0.016*** 0.053 0.051 0.001*** See Table 2 for variable definitions. *** denotes statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. 45 Table 4 Effect of the Jobs and Growth Tax Relief & Reconciliation Act of 2003 on Expected Rates of Return (1) (2) (3) (4) (5) r_ave r_ct r_oj r_mpeg r_gls -0.0118*** -0.0144*** -0.0130*** -0.0135*** -0.0040*** (-21.269) (-23.764) (-19.941) (-14.938) (-8.877) INST 0.0023 0.0004 0.0037* 0.0059** 0.0020 (1.352) (0.228) (1.916) (2.285) (1.364) SIZE -0.0024*** -0.0023*** -0.0030*** -0.0037*** -0.0009*** (-9.807) (-8.682) (-10.246) (-9.320) (-3.839) BM 0.0154*** 0.0026** 0.0114*** 0.0194*** 0.0263*** (11.322) (2.013) (7.560) (9.044) (21.295) DISPERSION 0.0024*** -0.0028*** 0.0036*** 0.0080*** -0.0009*** (7.182) (-8.555) (9.646) (15.271) (-3.637) ROE -0.0019* 0.0035*** -0.0044*** -0.0106*** 0.0036*** (-1.713) (3.609) (-3.354) (-5.164) (2.987) LEV 0.0190*** 0.0255*** 0.0176*** 0.0225*** 0.0097*** (8.921) (10.859) (7.123) (6.703) (5.299) BETA_MKTRF 0.0025*** 0.0024*** 0.0029*** 0.0039*** 0.0017*** (6.142) (5.450) (5.972) (5.646) (4.471) BETA_SMB 0.0018*** 0.0021*** 0.0023*** 0.0021*** 0.0004 (4.408) (4.646) (4.903) (3.213) (1.022) BETA_HML 0.0010*** 0.0011*** 0.0008** 0.0015*** 0.0006** (3.196) (3.603) (2.330) (2.969) (2.284) BETA_UMD -0.0012** -0.0008 -0.0015** -0.0032*** 0.0001 (-2.242) (-1.353) (-2.493) (-3.663) (0.146) Constant 0.1509*** 0.1354*** 0.1932*** 0.2249*** 0.0451*** (26.354) (22.860) (29.511) (24.617) (8.522) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.306 0.207 0.246 0.256 0.524 This table presents results of estimating equation (13). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. Variables POSTACT 46 Table 5 Effect of Liquidity on the Relation between Tax Rates and Expected Rates of Return (1) (2) (3) (4) (5) r_ave r_ct r_oj r_mpeg r_gls -0.0113*** -0.0138*** -0.0124*** -0.0130*** -0.0039*** (-20.437) (-22.848) (-19.088) (-14.271) (-8.472) ILLIQUIDITY 0.0340*** 0.0415*** 0.0384*** 0.0331*** 0.0151*** (6.400) (7.088) (5.980) (3.970) (3.791) ILLIQUIDITY*POSTACT -0.0270*** -0.0327*** -0.0266*** -0.0159* -0.0203*** (-5.158) (-5.897) (-4.183) (-1.796) (-4.911) INST 0.0050*** 0.0038** 0.0072*** 0.0094*** 0.0026* (2.823) (2.033) (3.483) (3.443) (1.691) SIZE -0.0015*** -0.0012*** -0.0019*** -0.0026*** -0.0006** (-5.150) (-3.820) (-5.324) (-5.522) (-2.357) BM 0.0142*** 0.0013 0.0102*** 0.0184*** 0.0257*** (10.430) (0.964) (6.726) (8.569) (20.826) DISPERSION 0.0023*** -0.0029*** 0.0035*** 0.0079*** -0.0009*** (6.968) (-8.866) (9.409) (15.052) (-3.652) ROE -0.0021* 0.0033*** -0.0046*** -0.0108*** 0.0035*** (-1.895) (3.371) (-3.470) (-5.200) (2.876) LEV 0.0184*** 0.0248*** 0.0170*** 0.0221*** 0.0094*** (8.675) (10.614) (6.888) (6.571) (5.129) BETA_MKTRF 0.0027*** 0.0026*** 0.0031*** 0.0041*** 0.0017*** (6.494) (5.937) (6.366) (5.919) (4.521) BETA_SMB 0.0019*** 0.0022*** 0.0024*** 0.0022*** 0.0003 (4.527) (4.787) (5.058) (3.347) (0.982) BETA_HML 0.0010*** 0.0011*** 0.0008** 0.0015*** 0.0006** (3.145) (3.448) (2.248) (2.883) (2.276) BETA_UMD -0.0014** -0.0010* -0.0017*** -0.0033*** -0.0000 (-2.568) (-1.700) (-2.821) (-3.827) (-0.026) Constant 0.1269*** 0.1060*** 0.1641*** 0.1963*** 0.0384*** (17.379) (13.948) (19.457) (17.009) (5.820) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.313 0.216 0.252 0.259 0.525 This table presents results of estimating equation (14). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. Variables POSTACT 47 Table 6 Tax Rates, Dividend Yield, and Expected Rates of Return (1) (2) (3) (4) (5) r_ave r_ct r_oj r_mpeg r_gls -0.0118*** -0.0143*** -0.0130*** -0.0135*** -0.0039*** (-21.401) (-23.807) (-20.075) (-14.995) (-8.685) YIELD -0.0114 -0.0220 0.0241 0.0341 0.0021 (-0.349) (-0.646) (0.604) (0.660) (0.085) YIELD*POSTACT -0.0821** -0.0807** -0.0842** -0.1123** -0.1262*** (-2.555) (-2.448) (-2.091) (-2.166) (-5.181) INST 0.0011 -0.0010 0.0032 0.0052** 0.0005 (0.669) (-0.567) (1.641) (1.975) (0.345) SIZE -0.0023*** -0.0022*** -0.0029*** -0.0037*** -0.0008*** (-9.437) (-8.259) (-10.085) (-9.170) (-3.376) BM 0.0156*** 0.0029** 0.0115*** 0.0194*** 0.0265*** (11.408) (2.195) (7.528) (9.000) (21.367) DISPERSION 0.0024*** -0.0028*** 0.0036*** 0.0080*** -0.0009*** (7.217) (-8.566) (9.654) (15.286) (-3.542) ROE -0.0019* 0.0035*** -0.0044*** -0.0107*** 0.0036*** (-1.730) (3.664) (-3.390) (-5.192) (3.019) LEV 0.0201*** 0.0270*** 0.0182*** 0.0233*** 0.0115*** (9.428) (11.499) (7.364) (6.845) (6.130) BETA_MKTRF 0.0023*** 0.0022*** 0.0028*** 0.0038*** 0.0014*** (5.682) (4.938) (5.736) (5.419) (3.717) BETA_SMB 0.0017*** 0.0019*** 0.0023*** 0.0021*** 0.0002 (4.070) (4.263) (4.725) (3.063) (0.486) BETA_HML 0.0010*** 0.0012*** 0.0008** 0.0015*** 0.0007** (3.243) (3.680) (2.302) (2.946) (2.334) BETA_UMD -0.0011** -0.0007 -0.0015** -0.0031*** 0.0002 (-2.093) (-1.189) (-2.407) (-3.585) (0.367) Constant 0.1499*** 0.1342*** 0.1925*** 0.2239*** 0.0437*** (26.040) (22.529) (29.307) (24.437) (8.245) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.308 0.209 0.246 0.257 0.526 This table presents results of estimating equation (15). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. Variables POSTACT 48 Table 7 Liquidity, Tax Rates, Dividend Yield, and Expected Rates of Return (1) (2) (3) (4) (5) r_ave r_ct r_oj r_mpeg r_gls -0.0097*** -0.0115*** -0.0113*** -0.0122*** -0.0023*** (-14.654) (-15.832) (-14.448) (-11.302) (-4.088) YIELD -0.0004 -0.0059 0.0361 0.0445 0.0081 (-0.012) (-0.170) (0.888) (0.846) (0.317) YIELD*POSTACT -0.0549 -0.0602 -0.0269 -0.0535 -0.1395*** (-1.481) (-1.573) (-0.594) (-0.929) (-5.114) HIGHILLIQUIDITY 0.0025** 0.0049*** 0.0024* 0.0020 0.0014 (2.048) (3.665) (1.715) (1.007) (1.377) HIGHILLIQUIDITY*POSTACT -0.0053*** -0.0075*** -0.0047*** -0.0039** -0.0040*** (-4.724) (-6.208) (-3.572) (-2.084) (-4.440) YIELD*POSTACT*HIGHILLIQUIDITY -0.1068** -0.1134*** -0.1763*** -0.1761** 0.0065 (-2.351) (-2.625) (-3.005) (-2.395) (0.209) INST 0.0009 -0.0006 0.0032 0.0052* 0.0001 (0.535) (-0.343) (1.561) (1.867) (0.077) SIZE -0.0025*** -0.0021*** -0.0030*** -0.0037*** -0.0010*** (-7.961) (-6.387) (-8.322) (-7.532) (-3.449) BM 0.0150*** 0.0021 0.0109*** 0.0189*** 0.0261*** (10.925) (1.581) (7.147) (8.728) (21.130) DISPERSION 0.0024*** -0.0028*** 0.0036*** 0.0080*** -0.0009*** (7.268) (-8.685) (9.654) (15.251) (-3.425) ROE -0.0020* 0.0034*** -0.0045*** -0.0107*** 0.0035*** (-1.828) (3.489) (-3.418) (-5.191) (2.924) LEV 0.0199*** 0.0267*** 0.0180*** 0.0232*** 0.0114*** (9.402) (11.472) (7.353) (6.825) (6.082) BETA_MKTRF 0.0023*** 0.0021*** 0.0027*** 0.0037*** 0.0014*** (5.568) (4.924) (5.643) (5.347) (3.618) BETA_SMB 0.0017*** 0.0020*** 0.0023*** 0.0021*** 0.0002 (4.144) (4.302) (4.765) (3.082) (0.619) BETA_HML 0.0010*** 0.0011*** 0.0008** 0.0015*** 0.0007** (3.155) (3.487) (2.153) (2.831) (2.344) BETA_UMD -0.0012** -0.0008 -0.0015** -0.0031*** 0.0001 (-2.127) (-1.328) (-2.408) (-3.563) (0.287) Constant 0.1521*** 0.1304*** 0.1931*** 0.2245*** 0.0475*** (20.607) (16.879) (22.858) (19.089) (7.114) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.310 0.213 0.249 0.258 0.527 This table presents results of estimating equation (16). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. Variables POSTACT 49 Table 8 Effect of the Jobs & Growth Tax Relief Reconciliation Act of 2003 on Dividend-Paying vs. Non-Dividend-Paying Firms Panel A: Without liquidity control (1) (2) (3) (4) (5) Variables r_ave r_ct r_oj r_mpeg r_gls POSTACT -0.0127*** -0.0159*** -0.0138*** -0.0140*** -0.0032*** (-14.873) (-16.929) (-14.125) (-10.266) (-4.736) DIV -0.0016 -0.0028** -0.0009 -0.0005 0.0005 (-1.454) (-2.297) (-0.675) (-0.272) (0.502) DIV*POSTACT 0.0018 0.0030** 0.0017 0.0009 -0.0015* (1.618) (2.536) (1.309) (0.531) (-1.698) INST 0.0022 0.0003 0.0039* 0.0060** 0.0017 (1.298) (0.168) (1.960) (2.277) (1.193) SIZE -0.0024*** -0.0022*** -0.0030*** -0.0037*** -0.0008*** (-9.315) (-8.131) (-9.931) (-8.935) (-3.598) BM 0.0154*** 0.0026** 0.0114*** 0.0193*** 0.0263*** (11.325) (2.053) (7.536) (9.017) (21.380) DISPERSION 0.0024*** -0.0028*** 0.0036*** 0.0080*** -0.0010*** (7.154) (-8.539) (9.656) (15.279) (-3.685) ROE -0.0018* 0.0035*** -0.0044*** -0.0106*** 0.0036*** (-1.684) (3.638) (-3.328) (-5.155) (2.987) LEV 0.0191*** 0.0255*** 0.0176*** 0.0225*** 0.0098*** (8.937) (10.887) (7.126) (6.709) (5.312) BETA_MKTRF 0.0025*** 0.0024*** 0.0029*** 0.0039*** 0.0016*** (6.018) (5.312) (5.951) (5.574) (4.167) BETA_SMB 0.0018*** 0.0020*** 0.0023*** 0.0022*** 0.0003 (4.259) (4.459) (4.845) (3.159) (0.921) BETA_HML 0.0010*** 0.0012*** 0.0009** 0.0015*** 0.0007** (3.322) (3.853) (2.339) (2.951) (2.272) BETA_UMD -0.0012** -0.0008 -0.0016** -0.0032*** 0.0001 (-2.261) (-1.374) (-2.561) (-3.690) (0.254) Constant 0.1508*** 0.1353*** 0.1938*** 0.2252*** 0.0443*** (25.938) (22.547) (29.216) (24.222) (8.323) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.307 0.208 0.246 0.256 0.524 This table presents results of estimating equation (17). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. 50 Table 8 (continued) Effect of the Jobs & Growth Tax Relief Reconciliation Act of 2003 on Dividend-Paying vs. Non-Dividend-Paying Firms Panel B: With liquidity control (1) (2) (3) (4) (5) Variables r_ave r_ct r_oj r_mpeg r_gls POSTACT -0.0121*** -0.0152*** -0.0132*** -0.0134*** -0.0030*** (-14.282) (-16.141) (-13.525) (-9.790) (-4.409) DIV -0.0014 -0.0025** -0.0007 -0.0003 0.0006 (-1.311) (-2.098) (-0.526) (-0.185) (0.568) DIV*POSTACT 0.0016 0.0027** 0.0014 0.0007 -0.0016* (1.452) (2.346) (1.128) (0.424) (-1.776) ILLIQUIDITY 0.0240*** 0.0298*** 0.0286*** 0.0272*** 0.0084** (5.367) (5.815) (5.280) (3.827) (2.388) INST 0.0057*** 0.0045** 0.0080*** 0.0098*** 0.0029* (3.164) (2.397) (3.868) (3.583) (1.891) SIZE -0.0014*** -0.0010*** -0.0018*** -0.0026*** -0.0005* (-4.399) (-2.905) (-4.794) (-5.208) (-1.743) BM 0.0148*** 0.0020 0.0108*** 0.0188*** 0.0262*** (10.961) (1.580) (7.146) (8.759) (21.271) DISPERSION 0.0023*** -0.0030*** 0.0035*** 0.0079*** -0.0010*** (6.796) (-9.026) (9.284) (15.000) (-3.830) ROE -0.0018* 0.0035*** -0.0044*** -0.0106*** 0.0036*** (-1.694) (3.627) (-3.345) (-5.164) (2.988) LEV 0.0188*** 0.0253*** 0.0173*** 0.0223*** 0.0097*** (8.848) (10.821) (7.020) (6.635) (5.285) BETA_MKTRF 0.0027*** 0.0026*** 0.0032*** 0.0042*** 0.0017*** (6.527) (5.961) (6.476) (5.899) (4.341) BETA_SMB 0.0019*** 0.0022*** 0.0025*** 0.0023*** 0.0004 (4.529) (4.759) (5.128) (3.346) (1.033) BETA_HML 0.0010*** 0.0011*** 0.0008** 0.0014*** 0.0006** (3.106) (3.532) (2.119) (2.806) (2.152) BETA_UMD -0.0013** -0.0009 -0.0017*** -0.0033*** 0.0001 (-2.446) (-1.575) (-2.761) (-3.806) (0.198) Constant 0.1242*** 0.1023*** 0.1621*** 0.1951*** 0.0349*** (16.738) (13.120) (18.849) (16.659) (5.318) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.310 0.214 0.250 0.258 0.524 This table presents results of estimating equation (17) with ILLIQUIDITY as an additional control variable. The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. Tstatistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. 51 Table 8 (continued) Effect of the Jobs & Growth Tax Relief Reconciliation Act of 2003 on Dividend-Paying vs. Non-Dividend-Paying Firms Panel C: With liquidity control and interaction (1) (2) (3) (4) (5) Variables r_ave r_ct r_oj r_mpeg r_gls POSTACT -0.0117*** -0.0146*** -0.0127*** -0.0131*** -0.0026*** (-13.858) (-15.603) (-13.137) (-9.632) (-3.878) DIV -0.0009 -0.0019 -0.0002 -0.0000 0.0010 (-0.872) (-1.592) (-0.146) (-0.026) (0.990) DIV*POSTACT 0.0007 0.0017 0.0006 0.0003 -0.0023** (0.697) (1.471) (0.493) (0.151) (-2.546) ILLIQUIDITY 0.0337*** 0.0408*** 0.0382*** 0.0330*** 0.0159*** (6.353) (6.943) (5.947) (3.947) (3.962) ILLIQUIDITY*POSTACT -0.0265*** -0.0314*** -0.0262*** -0.0158* -0.0220*** (-5.075) (-5.630) (-4.110) (-1.766) (-5.259) INST 0.0049*** 0.0037* 0.0073*** 0.0094*** 0.0023 (2.729) (1.923) (3.488) (3.413) (1.491) SIZE -0.0015*** -0.0011*** -0.0019*** -0.0027*** -0.0006** (-4.875) (-3.533) (-5.209) (-5.359) (-2.205) BM 0.0142*** 0.0013 0.0102*** 0.0184*** 0.0257*** (10.453) (1.030) (6.714) (8.553) (20.877) DISPERSION 0.0023*** -0.0029*** 0.0035*** 0.0079*** -0.0010*** (6.921) (-8.869) (9.403) (15.055) (-3.704) ROE -0.0021* 0.0033*** -0.0046*** -0.0108*** 0.0035*** (-1.877) (3.405) (-3.460) (-5.197) (2.863) LEV 0.0185*** 0.0249*** 0.0170*** 0.0221*** 0.0094*** (8.687) (10.640) (6.887) (6.574) (5.126) BETA_MKTRF 0.0027*** 0.0026*** 0.0031*** 0.0041*** 0.0016*** (6.338) (5.764) (6.313) (5.832) (4.186) BETA_SMB 0.0018*** 0.0021*** 0.0024*** 0.0023*** 0.0003 (4.385) (4.613) (5.005) (3.294) (0.880) BETA_HML 0.0010*** 0.0011*** 0.0008** 0.0015*** 0.0006** (3.224) (3.638) (2.217) (2.849) (2.231) BETA_UMD -0.0014** -0.0010* -0.0017*** -0.0033*** 0.0000 (-2.550) (-1.676) (-2.855) (-3.839) (0.103) Constant 0.1267*** 0.1057*** 0.1645*** 0.1965*** 0.0374*** (17.128) (13.819) (19.262) (16.797) (5.641) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.313 0.217 0.252 0.259 0.526 This table presents results of estimating equation (17) with the inclusion of ILLIQUIDITY and the interaction ILLIQUIDITY*POSTACT. The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. 52 Table 9 Effect of Institutional Ownership on the Relation between Tax Rates and Expected Rates of Return Panel A: Without liquidity control (1) (2) (3) (4) (5) r_ave r_ct r_oj r_mpeg r_gls -0.0125*** -0.0147*** -0.0135*** -0.0143*** -0.0048*** (-17.177) (-18.813) (-15.968) (-12.202) (-8.140) YIELD -0.0242 -0.0279 0.0057 0.0081 -0.0062 (-0.723) (-0.801) (0.139) (0.152) (-0.241) YIELD*POSTACT -0.0909** -0.0856** -0.0982** -0.1259** -0.1050*** (-2.564) (-2.397) (-2.213) (-2.209) (-3.986) HIGHINST -0.0015 -0.0019 -0.0014 -0.0015 -0.0012 (-1.487) (-1.621) (-1.240) (-0.923) (-1.354) HIGHINST*POSTACT 0.0026** 0.0009 0.0027** 0.0042** 0.0021** (2.218) (0.703) (1.979) (2.172) (2.071) YIELD*POSTACT*HIGHINST 0.1129* 0.0451 0.1289* 0.1699* -0.0389 (1.908) (0.751) (1.781) (1.761) (-0.915) SIZE -0.0023*** -0.0022*** -0.0028*** -0.0035*** -0.0007*** (-9.334) (-8.472) (-9.704) (-8.761) (-3.355) BM 0.0156*** 0.0029** 0.0117*** 0.0196*** 0.0265*** (11.473) (2.216) (7.668) (9.132) (21.473) DISPERSION 0.0023*** -0.0029*** 0.0035*** 0.0079*** -0.0009*** (7.061) (-8.780) (9.438) (15.109) (-3.531) ROE -0.0019* 0.0035*** -0.0045*** -0.0108*** 0.0036*** (-1.774) (3.634) (-3.407) (-5.208) (3.020) LEV 0.0203*** 0.0272*** 0.0185*** 0.0236*** 0.0115*** (9.544) (11.638) (7.537) (6.962) (6.092) BETA_MKTRF 0.0024*** 0.0022*** 0.0028*** 0.0039*** 0.0014*** (5.748) (4.953) (5.870) (5.560) (3.740) BETA_SMB 0.0017*** 0.0019*** 0.0023*** 0.0022*** 0.0002 (4.196) (4.303) (4.913) (3.252) (0.495) BETA_HML 0.0010*** 0.0012*** 0.0008** 0.0015*** 0.0007** (3.236) (3.705) (2.325) (2.954) (2.348) BETA_UMD -0.0011** -0.0007 -0.0014** -0.0031*** 0.0001 (-2.097) (-1.161) (-2.382) (-3.580) (0.283) Constant 0.1498*** 0.1341*** 0.1917*** 0.2229*** 0.0438*** (25.972) (22.609) (29.058) (24.261) (8.290) Observations 10,275 10,964 10,316 10,245 10,696 R-squared 0.308 0.210 0.247 0.257 0.526 This table presents results of estimating a modified version of equation (16) where we replace HIGHILLIQUIDITY with HIGHINST. The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. Variables POSTACT 53 Table 9 (continued) Effect of Institutional Ownership on the Relation between Tax Rates and Expected Rates of Return Panel B: With liquidity control Variables POSTACT YIELD YIELD*POSTACT HIGHINST HIGHINST*POSTACT YIELD*POSTACT*HIGHINST HIGHILLIQUIDITY HIGHILLIQUIDITY*POSTACT YIELD*POSTACT*HIGHILLIQUIDITY SIZE BM DISPERSION ROE LEV BETA_MKTRF BETA_SMB BETA_HML BETA_UMD Constant Observations R-squared (1) r_ave -0.0101*** (-12.062) -0.0091 (-0.266) -0.0718* (-1.707) -0.0009 (-0.841) 0.0014 (1.181) 0.0934 (1.504) 0.0021* (1.761) -0.0050*** (-4.465) -0.0939** (-1.983) -0.0025*** (-7.878) 0.0150*** (10.983) 0.0023*** (7.160) -0.0020* (-1.862) 0.0201*** (9.526) 0.0023*** (5.609) 0.0018*** (4.274) 0.0010*** (3.178) -0.0011** (-2.096) 0.1530*** (20.812) 10,275 0.310 (2) r_ct -0.0111*** (-11.894) -0.0057 (-0.159) -0.0766* (-1.799) -0.0006 (-0.550) -0.0009 (-0.650) 0.0253 (0.405) 0.0047*** (3.511) -0.0077*** (-6.208) -0.1043** (-2.321) -0.0022*** (-6.524) 0.0021 (1.590) -0.0029*** (-8.803) 0.0033*** (3.477) 0.0270*** (11.617) 0.0021*** (4.926) 0.0020*** (4.367) 0.0011*** (3.551) -0.0007 (-1.254) 0.1315*** (17.027) 10,964 0.214 (3) r_oj -0.0116*** (-11.799) 0.0205 (0.490) -0.0426 (-0.824) -0.0009 (-0.772) 0.0017 (1.231) 0.0892 (1.178) 0.0017 (1.214) -0.0043*** (-3.267) -0.1648*** (-2.688) -0.0030*** (-8.127) 0.0112*** (7.284) 0.0035*** (9.524) -0.0046*** (-3.442) 0.0184*** (7.530) 0.0027*** (5.729) 0.0024*** (4.963) 0.0008** (2.223) -0.0014** (-2.346) 0.1949*** (23.143) 10,316 0.248 (4) r_mpeg -0.0130*** (-9.383) 0.0197 (0.365) -0.0663 (-1.013) -0.0012 (-0.694) 0.0035* (1.762) 0.1304 (1.298) 0.0010 (0.499) -0.0031* (-1.665) -0.1644** (-2.141) -0.0037*** (-7.264) 0.0192*** (8.873) 0.0079*** (15.127) -0.0108*** (-5.218) 0.0235*** (6.951) 0.0038*** (5.446) 0.0022*** (3.274) 0.0015*** (2.874) -0.0031*** (-3.532) 0.2266*** (19.283) 10,245 0.258 (5) r_gls -0.0029*** (-4.053) 0.0032 (0.122) -0.1217*** (-4.092) -0.0008 (-0.849) 0.0011 (1.100) -0.0362 (-0.841) 0.0012 (1.202) -0.0037*** (-4.013) -0.0004 (-0.012) -0.0009*** (-3.405) 0.0261*** (21.246) -0.0009*** (-3.401) 0.0035*** (2.938) 0.0114*** (6.065) 0.0014*** (3.623) 0.0002 (0.607) 0.0007** (2.355) 0.0001 (0.251) 0.0477*** (7.103) 10,696 0.527 This table presents results of estimating a modified version of equation (16). The dependent variable is r_ave, r_ct, r_oj, r_mpeg, and r_gls in columns (1)-(5), respectively. See Table 2 for variable definitions. Industry indicator variables are included in the estimation but suppressed in the table. T-statistics are calculated using standard errors clustered by firm and appear in parentheses below coefficient estimates. ***,**, * denote statistically significant at p < 0.01, p < 0.05, and p < 0.10, respectively, using a two-tailed test. 54