Printer Friendly

Corporate payout policy, cash savings, and the cost of consistency: evidence from a structural estimation.

We develop a dynamic structural model to better understand how corporate payout policy is determined in conjunction with other corporate decisions. In a first-best model, a manager maximizes equity value by choosing the firm's optimal financing, investment, dividends, and cash holdings. By using simulated method of moments, we show that, on average, firms excessively smooth their payout while making corporate savings overly volatile and retaining excess cash. We then extend the model to capture the effect of a manager, who perceives a cost to cutting payouts. Estimating the model, we infer the magnitude of this cost. We find that a managerial preference for consistent payout explains the smooth payout and high volatility of cash holdings.

**********

Since Lintner (1956), it has been widely acknowledged that managers have a tendency to smooth corporate dividends. (1) While it is well understood that firms prefer to practice a consistent payout policy, the motivation for this behavior remains a contentious issue in the literature. One reason for the lack of consensus is because payout policy has close ties to other corporate decisions on corporate cash holdings, investments, and financing. Increasing payout diminishes resources available to increase investment and savings in current and future periods. This paper sheds light on the payout smoothing puzzle by studying the relationships between these policies simultaneously in a dynamic setting. We examine how manager-shareholder conflicts, more specifically a manager's perceived cost associated with payout reductions, explain observed payout patterns. We estimate the magnitude of this cost, as well as shareholder value loss due to this distortion.

The predominant explanations for payout smoothing in the literature are agency and information asymmetry. The agency explanations proposed by Jensen (1986) and Easterbrook (1984) suggest that paying a dividend that is both high and smooth forces firms to raise external capital to meet their financing needs. This exposure to the discipline of capital markets may reduce agency costs. However, this argument neglects the fact that in a dynamic world, a manager who perceives a cost to cutting dividends has an incentive to pile up excess cash to avoid future reductions in dividends. The joint determination of payout and cash savings by the manager and the dynamic effect of requiring smooth dividends, which, in turn, motivates even larger levels of savings, are ignored in this remedial view of dividend smoothing.

The endogenous choice of corporate policies may also underlie the conflicting and ambiguous results observed in the payout smoothing literature (Booth and Xu, 2008; Li and Zhao, 2008; Aivazian, Booth, and Cleary, 2009; Leary and Michaely, 2011). We suggest that these inconclusive findings may be due to disregarding the joint determination of variables such as payout ratio, operating profits, leverage, capital expenditures, and cash holdings.

Structural estimation is a reliable approach to avoiding these types of endogeneity problems. The model we propose captures both the manager's immediate trade-offs under a rich set of frictions and the manager's future considerations. We exploit this model to predict optimal levels of payout and cash holdings for firms that invest in capital, save cash, raise equity, issue/retire debt, and pay dividends, all contingent on uncertain future productivity.

We examine two models: 1) a first-best model where the manager maximizes the value of equity and 2) an agency model where the manager also considers a cost associated with any reduction in payout. We estimate a set of parameters for both of these models using a simulated method of moments (SMM). The SMM procedure estimates parameters by minimizing the error between simulated and empirical moments from corporate financing, payout, and investment choices. The empirical sample includes nonfinancial, unregulated US firms from 1988 to 2006 constructed from the annual 2006 Compustat industrial files.

We show that the first-best model fails to capture the dynamics of dividend policy. Similar to other dynamic structural models that include dividends such as Hennessy and Whited (2007), the best parameter fit results in a payout policy that is far more volatile than what is observed empirically. The simulated variance of payout is 0.0024, which is significantly larger than the empirical variance of 0.0016. Simulated variance of investment and cash holdings are significantly over and underestimated, respectively. The average cash-to-assets ratio from the simulated panel (0.0689) is smaller than the corresponding empirical moment (0.1631). These results also indicate that firms, on average, maintain more cash than can be explained through the dynamic tradeoff model alone. This is consistent with the empirical literature on cash holdings which argues that firms typically maintain too much cash (Opler, Pinkowitz, and Stulz, 1999; Dittmar and Mahrt-Smith, 2007).

The second model considers the maximization problem faced by a manager who perceives a downward adjustment cost from cutting payouts when making financial and real decisions within the firm. The manager's objective function is extended from solely maximizing the value of equity by adding a cost linear in the magnitude of any reduction in payout. This model is consistent with a survey study (Brav et al., 2005) that reports that 94% of managers of dividend paying firms strongly or very strongly agree that they actively try to avoid reducing dividends. This model provides a possible explanation for what motivates corporate payout smoothing, but we do not directly explore the sources of this disutility imposed on the manager. Through an estimation exercise, we find the managerial perceived cost for cutting payout that best explains observed payout, saving, financing, and investment dynamics.

The estimation of the agency model indicates a payout consistency cost parameter equal to 0.119 that translates to a typical firm behaving as if it has a manager who associates an average cost for cutting payout equal to $81,000 for a million dollars of shareholders' equity value. These parameter estimates support the view that: 1) on average, managers anticipate fairly large costs associated with cutting payouts, and 2) that firms are sensitive to this managerial agency parameter. Our estimates also determine a loss of approximately 6% in shareholders' equity value due to the perceived cost of cutting payouts.

In addition to reducing the variability of payouts, changes in the perceived cost from cutting payouts has a pronounced impact on cash and investment policies. These relationships are illustrated through a series of comparative statics exercises. An increase in the payout consistency cost produces smoother investments, while cash holdings largely absorb the volatility and become more variable. Higher cash levels are also maintained in order to decrease both the probability and magnitude of any future reduction in payout. While this may shed some light on the excess cash puzzle, our comparative statics and estimation results indicate that effect is relatively small and is likely only a mild contributor to the high observed levels of liquid assets.

These comparative statics results are not particularly supportive of the remedial view of payout smoothing suggested by Easterbrook (1984). From our results, it is not clear that imposing a large cost to cutting payouts onto a manager would enhance shareholder value. It is important to note that the empirical positive association between cash holdings and payout smoothing documented in Leary and Michaely (2011) may also be due to the endogenous relationship between these policies. This interpretation differs from the Leary and Michaely (2011) explanation that cash cows adopt payout smoothing to reduce agency costs. This notes the significance of the structural approach in enabling us to account for endogeneity among corporate policies.

We explore different motives for payout smoothing and possible heterogeneity on the impact of a managerial perceived cost to cutting payouts by performing estimations on subsamples of firms split by a variety of manager-firm characteristics. Through this exercise, we find some supporting results for both agency and information asymmetry explanations for payout smoothing.

To provide evidence pertaining to the agency explanation, the sample is split based upon Chief Executive Officer (CEO) incentive pay, proxied by pay-performance-sensitivity (PPS). Consistent with Easterbrook (1984), we find that managers who receive contracts with high PPS tend to associate lower costs with cutting payouts. Allen, Bernardo, and Welch (2000) argue that institutional investors, who may also lower agency costs through their monitoring activities, are attracted to firms paying larger dividends due to their tax status. These institutional investors, in the case of cutting dividends, may impose large costs on management. Supporting their theory, we find firms with larger institutional ownership behave as if they have managers with larger preferences for smooth payouts. Our estimates also result in larger managerial payout consistency costs for firms that pay larger fractions of their payout in the form of dividends rather than share repurchases. This is consistent with the empirical literature that suggests that managers do not smooth repurchases in the same manner as they smooth dividends (Skinner, 2008).

The information asymmetry explanation is studied through a sample split based on analyst forecasts dispersion. In the presence of information asymmetry, smooth dividends convey more information than erratic payouts. Studies such as Almeida, Campello, and Weisbach (2004) and Bates, Kahle, and Stulz (2009) argue that if future dividend cuts are viewed as costly, financially constrained firms will be reluctant to increase dividends even following a positive cash flow shock. This argument suggests that when investor information is poor, firms have a greater incentive to smooth their dividends. This prediction is consistent with signaling explanations such as Kumar (1988), Kumar and Lee (2001), and Guttman, Kadan, and Kandel (2001).

In support of this hypothesis, we find that firms suffering from greater information asymmetry tend to have managers with larger payout consistency costs. Interestingly, this result is masked when the empirical results are reviewed in isolation as the payout variances are similar between the subsamples of high and low information asymmetry firms.

While Leary and Michaely (2011) finds support only for the agency based explanations of payout smoothing, we determine that information asymmetry may also contribute to this practice. As in their paper, we also document that payout variance is not related to measures of information asymmetry. It is only by estimating the unobservable variable of the manager's disutility from cutting payouts that we are able to bring to light the effect of information asymmetry on the preference for payout smoothing.

Our cross-sectional results highlight the importance of structural estimation for corporate finance studies where endogeneity of certain key variables is a significant concern and also when model elements are unobservable. With this insight in mind, it is not surprising that the literature has produced such mixed evidence regarding the motives for payout smoothing.

Both information asymmetry and agency explanations of payout smoothing assume that payout smoothing is adopted to improve shareholder value. This may appear to conflict with our estimated perceived cost of cutting payouts and the associated equity loss to the shareholders from payout smoothing. However, our estimations are done in comparison to the first-best results where no conflict exists between the manager and the shareholders. Our paper does not propose de-emphasizing the possible value-enhancing effects of payout smoothing in the presence of other possible agency problems or information asymmetry.

The remainder of this article proceeds as follows. In Section I, we introduce the related studies and situate the paper within the existing literature. Section II presents the model. In Section III, we present sensitivity analyses and the estimation results. Finally, Section IV provides our conclusions. Explanations of the computational methodology used in this paper are included in the appendix.

I. Related Literature

This paper primarily relates to three major areas of corporate finance literature: 1) the empirical literature regarding corporate payouts, 2) literature on savings, as well as 3) the structural literature about dynamic corporate policies.

The literature regarding corporate payouts broadly supports the notion that dividend smoothing is prevalent. (2) Leary and Michaely (2011) examine different motivations and find support for agency explanations rather than asymmetric information justifications of smoothing behavior. (3) They note that younger, smaller firms and firms with higher earnings volatility smooth less. Firms with high cash holdings, low growth opportunities, and higher levels of institutional ownership smooth more. Aivazian et al. (2009) consider the relationship between dividend smoothing and access to public markets as proxied by bond ratings. They determine that firms smooth more when they raise debt in the public uninformed bond markets rather than in the private informed bank market. They conclude that the dividend smoothing decision is related to information asymmetry between the managers and the firm's creditors.

The empirical literature on how firms allocate their cash flow between investments, payouts, and cash savings is also relevant. Harford, Mansi, and Maxwell (2008) find that in the United States, firms with weaker corporate governance do not payout their excess cash through dividends. Rather, they choose to distribute excess cash by methods that do not impose any commitment on their future payouts, such as share repurchases. Moreover, these weakly controlled managers actually save less cash and choose to spend cash quickly on acquisitions and capital expenditures, rather than hoard it. Similarly, Francis et al. (2011), using a differences-in-differences approach, demonstrate that dividend payout ratios increase with the quality of corporate governance.

Our paper differs from all of the previous studies as we present the first dynamic structural estimation focused on payout smoothing. Our structural model does not directly tackle the question as to why firms smooth their payouts. Instead, it focuses on estimating the managerial perceived cost of cutting payouts. We contribute to the empirical cross-sectional studies by reestimating the managerial payout consistency cost for different subsamples of the data and find supporting evidence for both agency and asymmetric information explanations.

The dynamic structural models presented by Gamba and Triantis (2008), Riddick and Whited (2009), and Moyen and Boileau (2009) feature the trade-off between corporate policies subject to costly external financing. These studies highlight the role of cash savings in financing decisions. The model presented by Gamba and Triantis (2008) indicates a positive relationship between debt flotation costs and cash holdings. Riddick and Whited (2009) document a negative corporate propensity to save. They demonstrate how a firm counters increases in cash flow with a reduction in savings. Unlike our model, they do not include any sort of debt. Moyen and Boileau (2009) present a model in which precautionary savings may also arise due to a firm's liquidity constraints. They argue that the capital share of revenues has become smaller over time. Thus, the prudence motive is no longer empirically relevant. They find that the liquidity constraint motive may explain the observed increase in cash holdings. These papers exclude agency issues and do not broach the subject of payout policy.

Our paper is also closely related to Hennessy and Whited (2007) who utilize a structural model and SMM to study firm behavior when facing costly external financing. As in our paper, Hennessy and Whited (2007) find a suboptimal level of distribution variance indicating managerial preference for payout smoothing. However, unlike our model that addresses this discrepancy by estimating a payout consistency cost, Hennessy and Whited (2007) leave the payout smoothing question unanswered.

In addition to our paper, there exists only one other structural model that investigates the role of agency conflicts on corporate finance decisions. Nikolov and Whited (2010) estimate a dynamic model of firm investment and cash accumulation in the presence of shareholder-manager agency conflicts. In their study, the source of agency is different from ours. They model agency conflicts arising from limited managerial ownership, bonuses based on short-term profits, managerial empire building preferences, and managerial perquisite consumption. They find that agency problems following from perquisites are the most important agency conflict when explaining cash holdings.

The agency sources studied by Nikolov and Whited (2010) do not relate to the managerial payout consistency cost that is the focus of our paper. Moreover, they do not distinguish between debt and equity incorporating both into external financing. To their credit, Nikolov and Whited (2010) have the advantage of modeling compensation related agency which we do not touch on. Our study is, in many ways, complementary to Nikolov and Whited (2010), as we capture a different source of agency conflicts and its effect on corporate decisions such as payout policy.

II. The Model

A. The Base Case Model

The base case model represents a firm maximizing its equity value. We construct a discrete time, infinite horizon, partial equilibrium, stochastic model of payout, debt, investment, and cash holdings. The model allows the manager to maximize equity value by balancing the allocation of resources to payouts, investment, and cash holdings subject to debt and equity financing. Equity claimants are assumed to be risk neutral. Debt is chosen by trading off recapitalization costs with a tax benefit. Equity is subject to issuance costs. An individual firm is characterized by the particular sequence of stochastic shocks. The equity value [V.sub.t] takes the form:

[V.sub.t] = [D.sub.t] + T[D.sub.t]) + [LAMBDA]([D.sub.t]) + [1/1 + (1 - [[tau].sub.b])r] [E.sub.t][[V.sub.t+1]], (1)

where r is the discount rate, [[tau].sub.b] is the personal interest income tax rate, T([D.sub.t]) is the tax schedule on payout, [LAMBDA]([D.sub.t]) is the equity issuance cost, and [E.sub.t] is the conditional expectation at period t. Payouts are represented by the variable [D.sub.t].

Payouts are taxed according to the tax schedule T([D.sub.t]). In this model, we do not directly differentiate between dividends and share repurchases. The tax schedule is convex to reflect firms taking advantage of lower capital gains taxes by issuing smaller payouts in the form of share repurchases. (4) Following Hennessy and Whited (2007) and Moyen and Boileau (2009), the payout tax schedule is convex and features an increasing marginal tax rate:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (2)

where [phi] > 0 is the payout tax parameter, [[tau].sub.d] is the tax rate, and the function [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is equal to one if [D.sub.t] > 0, and zero otherwise. A larger (j) corresponds to a higher marginal tax rate.

A negative dividend is interpreted as an equity issuance. The cost associated with an equity issuance may result from underwriting fees and asymmetric information problems. To maintain tractability, these dynamics are captured in reduced form. As in Riddick and Whited (2009), the equity issuance cost is described by the following linear quadratic function:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (3)

where function [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is equal to one if [D.sub.t] < 0, and zero otherwise, [[lambda].sub.0], [[lambda].sub.1], and [[lambda].sub.2] are positive constants. The convexity of [LAMBDA]([D.sub.t]) is consistent with rising marginal costs observed by Altinkilic and Hansen (2000). As demonstrated in Equation (1), equity value is equal to the sum of the expected discounted stream of dividends net payout tax and equity issuance cost: [D.sub.t] + T([D.sub.t]) + [LAMBDA]([D.sub.t]). (5)

The capital stock is the depreciated capital stock of the previous period plus any new investment. Capital accumulation is presented as:

[K.sub.t+1] = (1 - [delta])[K.sub.t] + [I.sub.t]. (4)

In the above equation, the capital depreciation rate is [delta] [member of] (0, 1), [K.sub.t] is the capital stock, and investment is denoted by [I.sub.t].

Consistent with Riddick and Whited (2009), Moyen and Boileau (2009), and Nikolov and Whited (2010), while capital may be bought or sold at a price of one, an investment or divestment leads to adjustment costs:

A([K.sub.t], [I.sub.t]) = a/2 [([I.sub.t]/[K.sub.t]).sup.2][K.sub.t]. (5)

where a is the capital adjustment cost parameter that acts to smooth investment over time. (6)

The change in debt, [DELTA][B.sub.t+1], may represent a new issuance ([DELTA][B.sub.t+1] > 0) or retirement ([DELTA][B.sub.t+1] < 0) and is the difference between the debt level in the current period [B.sub.t+1] and the debt level [B.sub.t] from the previous period:

[DELTA][B.sub.t+1] = [B.sub.t+1] - [B.sub.t]. (6)

While debt matures each period, all or part of the obligation may be rolled over and new debt may be issued at the risk-free rate. (7)

We also assume a recapitalization cost for debt:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (7)

This functional form captures the flotation cost on new debt issued. There is no additional cost associated with paying down debt. (8) Parameter q [greater than or equal to] 0 captures the linear flotation cost of new debt issues. (9)

In this model, debt financing has a tax advantage as the firm may deduct interest payments at a rate [[tau].sub.c] which is greater than the individual tax rate on interest income ([[tau].sub.c] > [[tau].sub.b]). Thus, the recapitalization cost of debt plays an important role as it provides an implicit upper bound on the debt level. (10) The firm's payout is defined by the sources-and-uses of funds equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (8)

In the above equation, (1 - [[tau].sub.c])f([K.sub.t]; [[theta].sub.t]) is the firm's after-tax operating income from capital [K.sub.t] given underlying income shock [[theta].sub.t]. The depreciation tax shield is given by [[tau].sub.c][delta][K.sub.t] and the after-tax interest payment is (1 - [[tau].sub.c])r[B.sub.t].

It is assumed that the depreciation rate used in the depreciation tax shield is accurately set to the true capital stock depreciation rate. The stock of cash has to be nonnegative as we do not model lines of credit separately from debt issuance.

[C.sub.t+1] [greater than or equal to] 0. (9)

The firm's operating income before depreciation is presented as:

f([K.sub.t]; [[theta].sub.t]) = [[theta].sub.t][K.sup.[alpha].sub.t], (10)

where f([K.sub.t]; [[theta].sub.t]) = [[theta].sub.t][K.sup.[alpha].sub.t] features decreasing returns to scale when [alpha] [member of] (0, 1). The firm's income shock follows a first-order autoregressive process:

ln[[theta].sub.t+1] = [rho]ln[[theta].sub.t] + [sigma][[epsilon].sub.t+i], (11)

with persistence and volatility parameters [rho] and [sigma] and where [[epsilon].sub.t] ~ N(0, 1). The persistence parameter determines the degree to which the next period's income shock is predictable. So long as this parameter is nonzero, the firm can partially anticipate the next period's income shock and adjust their capital structure accordingly.

The manager maximizes the equity value of the firm such that Equations (2)-(11) are satisfied. (11) This forms the base case model where the manager acts in the interest of all equity holders. They do not extract any sort of private benefits that may cause an agency conflict between them and their shareholders. In the following section, we will extend the model to allow the manager to perceive a cost to cutting payouts.

In the base case model, the manager's choice of payout is similar to a residual payout policy in which the positive residual after choosing the new cash level, debt level, and the firm's investment is distributed to the shareholders. The manager selects optimal levels of payouts only through other decisions.

In a model with no cost of raising equity, there would be no need for a stash of cash. That is, if [LAMBDA]([D.sub.t]) = 0, then the firm can effectively manage its financing by buying and selling its capital stock and, when needed, issuing equity. However, due to costly equity issues, [LAMBDA]([D.sub.t]), the firm may save some cash to reduce expected future financing costs. The firm's optimal level of cash holdings also depends upon the firm's expected future financing needs. Thus, optimal cash levels depend upon operational income, f([K.sub.t]; [[theta].sub.t]), and, in particular, the uncertainty characterized by the shock [[theta].sub.t].

Optimal cash holdings also relate to the firm's debt and the recapitalization costs of that debt. If debt adjustments were not costly, the firm could simultaneously meet its financing needs and avoid costly equity issues by raising more debt. By including the recapitalization cost of debt hand in hand with the cost of issuing equity, we are able to address the simultaneous existence of debt, equity issues, and cash balances.

In summary, our dynamic model describes a firm that, at each period, chooses the payout [D.sub.t], the investment [I.sub.t], the change in debt [DELTA][B.sub.t+1], and how much cash [C.sub.t+1] to maintain. Constrained by Equations (2)-(11), the firm makes choices in order to maximize the equity value (Equation (1). These decisions are made by the firm using knowledge of the previous period's financial state, determined by capital stock [K.sub.t], debt [B.sub.t], and cash stock [C.sub.t], and following observations of this period value for the income shock [[theta].sub.t]. The firm's intertemporal problem is described by the following Bellman equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (12)

subject to Equations (2)-(11).

Decision rules cannot be determined analytically. To solve for optimal policies, numerical methods are employed. The approximate solution provides a set of decision rules that allows a sequence of firms to be simulated. Nonetheless, some intuition regarding optimal policies can be developed by considering the Euler equation relating the model's dynamics to changes in cash. Utilizing the envelope condition, the Euler equation can be presented as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (13)

The optimal interior financial policy has to satisfy this condition. The left-hand side of the equation is the marginal cost of cash holdings. This includes the immediate cost of saving one more dollar in terms of reduced payouts and increased issuance costs. Saving cash also influences both the probability of future payouts and the marginal cost of the payout tax in the next period. The right-hand side represents the expected marginal savings from reductions in future equity financing and corporate dividend taxes. In an optimal policy, a firm saves a dollar today only if the immediate costs, including taxes on savings, are compensated by reductions in equity issuance costs and taxes on payout. The firm will save until penalties from taxes on savings become greater than the future benefits from reductions in equity costs and taxes on payout.

Inspection of Equation (13) also reveals that optimal cash, investment, and debt policies are intertwined. In the base case model, the financing trade-offs are somewhat similar to the trade-offs in Riddick and Whited (2009). Although our base case model also incorporates risk-free debt and taxes on dividends, the fundamental trade-off is between the tax disadvantage of holding cash versus the financial flexibility provided by cash allowing the firm to avoid the cost of external equity and the recapitalization cost of debt.

B. The Payout Consistency Cost Model

In this section, we present a model where the objective function of the manager is modified to capture a perceived cost faced when dividends are cut. The maintenance of consistent dividends is a widespread practice generally taken as an article of faith. We also take this for granted as an empirical fact and focus on the estimation of the loss due to the prevalence of payout smoothing. The perceived cost of payout smoothing is present in the form of an attenuation of the manager's utility function that is linear in the magnitude of the payout reduction. (12) The manager's intertemporal problem is described by the following Bellman equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (14)

subject to Equations (2)-(11).

The constraints of the maximization problem remain unchanged from the base case model. The magnitude of the disutility function is captured by parameter [gamma]. To our knowledge, this model is the first study that estimates the perceived cost managers' associate with cutting dividends.

By construction, it is clear that the second model will lead to a lower payout variance. However, further investigation is required to determine whether in comparison to the base case results, incorporating this disutility results in higher levels of cash holding. Studies, such as Easterbrook (1984), suggest that dividend smoothing may be employed to make managers less flexible and more reliant upon external capital markets. They argue that this practice would compel managers pay out excess cash and face capital markets more often. Consistent with this view, Mansi and Wald (2011) find that firms that face legal limitations on debt, and cannot use high leverage to reduce the free cash flow problem, pay more dividends. In our model, the managers are aware of the possible costs they will face if the firm's dividends are cut. (13)

The perceived cost of dividend smoothing would affect both the marginal cost of retaining cash and the shadow value of cash balances. Keeping more cash today decreases the probability of having to cut dividends tomorrow. Conditional upon cutting dividends, higher cash levels decrease the magnitude of this reduction and, consequently, the manager's disutility. Earlier empirical studies generally do not account for this co-determination of cash holdings and payouts. While payout smoothing may be imposed by shareholders (e.g., institutional investors) with the goal of increasing payouts and reducing agency conflicts, it may lead to an increase in excess cash and further deviation from first-best corporate policies. In this model, we test for this effect by employing SMM to endogenously solve for the unknown parameters. Furthermore, we allow the persistence and volatility of the cash flow shocks ([rho], [sigma]) and also the equity issuance cost parameters, the depreciation rate, the debt recapitalization cost parameter, the distribution tax schedule parameter, and the capital adjustment cost parameter ([[lambda].sub.0], [[lambda].sub.1], [[lambda].sub.2], [delta], q, [phi], a) to be endogenously estimated alongside the agency parameter.

III. Estimated Optimal Corporate Policies

A. Estimation

Acquiring a solution to our base case model, details of which are explained in the appendix, requires values for the full set of parameters: r, [delta], [[tau].sub.c], [[tau].sub.b], q, [alpha], [phi], [rho], [sigma], [[lambda].sub.0], [[lambda].sub.1], and [[lambda].sub.2]. The majority of these parameters are estimated using SMM. Parameters r, [[tau].sub.c], [[tau].sub.b], [[tau].sub.d], and [alpha] are set exogenously based on their estimated values reported in the literature.

Following recent dynamic investment studies, we set the real interest rate r to 0.02. This lies between the values chosen by Moyen and Boileau (2009) and Riddick and Whited (2009). (14) Tax parameters are set to [[tau].sub.c] = 0.35 and [[tau].sub.b] = 0.25 to reflect actual US corporate and personal tax rates of 35% and 25%. The marginal dividend tax rate is set [[tau].sub.d] = 0.25, the approximate value of 0.2325 calculated by Moyen and Boileau (2009) for the time period 1995-2006. Furthermore, following Moyen (2004) and Gamba and Triantis (2008), we set the production return-to-scale parameter [alpha] at 0.45.

Optimal firm decisions for both models are derived numerically as described in Appendix A. Random outcomes of the income shock [[theta].sub.t] are used to generate the series [K.sub.t+1], [[DELTA].sub.B+1], [D.sub.t], [C.sub.t+1], and [V.sub.t]. A sample of 20,000 firms is generated, where each series in the panel defines a firm. (15) Since the initial state is selected at random, firms may begin with suboptimal capital structure. We drop the first 20 periods to avoid observing the firm before it has worked its way out of an unlikely initial state.

To mimic real world empirical variables, we define the following variables from the simulated panel:

[NetIncome.sub.t]/TotalAssets.sub.t] = (1 - [[tau].sub.c])(f([K.sub.t]; [[theta].sub.t]) - [delta] [K.sub.t] - r[B.sub.t])/[K.sub.t],

where f([K.sub.t]; [[theta].sub.t]) is from Equation (10).

[Investment.sub.t]/[TotalAssets.sub.t] = [K.sub.t+1] - (1 - [delta])[K.sub.t]/[K.sub.t],

[Debt.sub.t]/[TotalAssets.sub.t] = (1 + (1 - [[tau].sub.b])r)[B.sub.t]/[K.sub.t],

PayoutRatio = max{0, [D.sub.t]}/[NetIncome.sub.t] = max{0, [D.sub.t]}/(1 - [[tau].sub.c])(f([K.sub.t]; [[theta].sub.t] - [delta][K.sub.t] - r[B.sub.t]),

[EquityIssuance.sub.t]/[TotalAssets.sub.t] = -min{0, [D.sub.t]}/[K.sub.t],

[CashFlow.sub.t]/[TotalAssets.sub.t] = (1 - [[tau].sub.f])(f([K.sub.t]; [[theta].sub.t]) = r [B.sub.t] + [[tau].sub.c][delta][K.sub.t]/[K.sub.t].

The unknown parameters are estimated using SMM. This procedure chooses the payout consistency cost parameter, the volatility and persistence of the cash flow shocks, the equity issuance cost parameters, the debt recapitalization cost parameter, the capital adjustment cost parameter, the capital depreciation rate, and the distribution tax schedule parameter to minimize the distance between the moments generated from the simulation procedure and the corresponding empirical moments. We follow the SMM procedure exposited in papers such as Hennessy and Whited (2007), Nikolov and Whited (2010), and Moyen and Boileau (2009).

We use SMM to find a set of parameter values that generate simulated moments that most closely match empirical moments. The moments used in this study consist of first and second order moments of financial ratios, correlations, and regression coefficients. In order to estimate the simulated moments, we conduct simulations on the model using a given parameter vector generating rounds of simulated data sets. We then compute the selected moments from the simulated panels. The SMM estimator of the parameter vector minimizes the weighted distance between empirical and simulated moments. We employ the efficient weighting matrix equal to the inverse of the estimated covariance of the moments. The influence-function approach from Erickson and Whited (2000) is used to calculate this covariance matrix. Hennessy and Whited (2007) demonstrate that the indirect estimated vector of parameters is asymptotically normal for fixed rounds of simulations allowing an overidentification test of the model.

Minimizing the error involves a computational search through the space of parameters. We employ a gradient descent approach augmented by an Iterated Local Search metaheuristic (ILS). Lourengo, Martin, and Stutzle (2003) provide a detailed account of this stochastic local search algorithm.

B. Sensitivity Analysis

We perform a set of sensitivity analyses to gain a better understanding of how different simulated moments respond to changes to the key model parameters. This analysis also sheds light on the informativeness of the moments regarding model parameters. In particular, we are interested in how the payout consistency cost parameter affects payout, financing, and corporate savings. Figure 1 presents the results of the comparative statics exercises. These results focus on the two parameters that most directly affect corporate payout policy: [gamma] and [phi].

First, we examine the sensitivity of the payout moments (payout variance, payout ratio, and frequency of paying out), and a set of important financial and real moments (average cash-to-assets ratio, average equity issues-to-assets ratio, variance of investments-to-assets, and standard deviation of shocks to income-to-assets) to the payout consistency cost parameter [gamma]. Using the estimated parameter values from Table I, we solve the payout consistency cost model 12 times while incrementing [gamma] from 0 to 0.5. The first eight panels in Figure 1 illustrate the comparative statics for each chosen moment when changing [gamma].

The first comparative static indicates that the payout variance decreases as the managerial consistency cost parameter increases. Although the cost of cutting payouts is linear in the size of the payout reduction, the corresponding drop in payout variance is nonlinear. The payout variance drops sharply as [gamma] increases as the manager chooses to maintain a less volatile payout as she associates higher costs to cutting payouts. With larger values of [gamma], the perceived cost of cutting payouts becomes larger relative to the equity value resulting in the manager choosing an almost constant payout policy. This is illustrated by the payout variance approaching zero when [gamma] > 0.20.

The second panel of Figure 1 reports that the payout ratio decreases monotonically with [gamma]. This is particularly interesting as it indicates that a manager who associates larger costs with cutting payouts will maintain a lower level of payouts. On better days, a manager who foresees costs from future payout reductions will be more reluctant to increase the payouts. When the manager faces a greater realization of cash flow shocks, she would rather pay back debt and pile up cash than increase the payout when taking into account the probability of a reduction in future payouts. This cautious payout policy leads to a lower payout ratio. Unlike the suggestions of Easterbrook (1984) and Jensen (1986), a manager who takes into account the future disutility associated with cutting payouts would maintain a lower payout ratio. Therefore, shareholders imposing higher costs onto the manager will result in more consistent payouts while decreasing the level of the payouts.

The third panel presents the response of the frequency of paying out to changes in [gamma]. Interestingly, the payout consistency cost parameter has little effect on the payout frequency. We believe that there are two counteracting forces acting on this moment. In our model, a manager who has already committed to a positive payout, but receives low cash flow realization would not fully abandon the payout. This is different from the first-best results where a manager may fully avoid payouts when low shocks are realized. This force would lead to an increase in payout frequency as [gamma] increases. Alternatively, a manager who currently has zero payouts and experiences an intermediate or low cash flow shock would not initiate a payout. Due to the persistence of cash flow shocks, the manager would foresee future payout cuts and avoid this expected future disutility. This effect works in the opposite direction of the first force. It appears that, on average, these two effects cancel each other out leading to the payout frequency remaining relatively unchanged in response to changes in [gamma].

The cash-to-assets ratio increases with [gamma] as the manager has an additional motive for saving (Panel 4, Figure 1). The manager saves more cash to use as a cushion against future low realizations of cash flow, allowing her to maintain a more consistent payout. However, the cash-to-assets ratio increases slowly. A large value of [gamma] = 0.5 would increase the cash-to-assets ratio by 50% to 0.16. This suggests that the significantly larger cash holdings in the empirical sample are not primarily driven by the perceived cost of reducing payouts. (16)

An increase in [gamma] increases the variance of the cash-to-assets ratio. As the manager perceives greater costs to cutting payout, she smoothes payouts and investments. However, cash holdings soak up the variability that would be evident in payouts and investments in the absence of the perceived cost of cutting payouts.

In our structural model, the firm is also able to issue new equity that is mutually exclusive to paying out. The larger payout consistency cost indirectly leads to a smaller equity issues-to-assets ratio (Panel 5, Figure 1). As [gamma] increases, the firm becomes more cautious and less volatile in its payout and investment policies. Therefore, the firm relies more on its internal means of financing resulting in smaller costly equity issuances. The reduction in the equity issues-to-assets ratio is due to a lower magnitude of equity proceeds conditional upon issuing equity rather than facing the capital markets less frequently.

Panel 7 of Figure 1 demonstrates that the variance of investments decreases with [gamma]. Structural models such as Hennessy and Whited (2007) overshoot the variance of investments. Hennessy and Whited (2007) argue that incorporating irreversibility into the investment cost function may help explain the high level of investment variance. However, we explain the inconsistency between the simulated investment variance and its empirical value through a managerial perceived cost to cutting payouts. A manager who smoothes payouts will invest in smaller sizes, but more often leading to a lower variance of investment though the average investment is unchanged. In this manner, a preference for consistent payouts leads to a less volatile investment policy.

In Panel 8, we illustrate the relationship between the standard deviation of shocks to income-to-assets and the managerial payout consistency cost parameter [gamma]. A smoother investment policy, resulting from the preference for consistent payouts, leads to a less volatile income-to-assets ratio.

Overall, these results demonstrate how different financing and payout moments respond to changes in [gamma]. These moments are also all endogenously related to each other through their relationship with the managerial preference for consistent payouts. The larger cash balances, the lower payout ratio, and the smaller equity issuances associated with larger values of [gamma] do not support the remedial view of payout smoothing.

In the next three panels, we investigate the relationship between the three payout moments and the payout tax schedule parameter [phi] by solving the payout consistency cost model while incrementing [phi] from 0.1 to 0.5. Both [phi] and y intensify payout smoothing behavior. However, [phi] differs from [gamma] in that rather than penalizing reductions in payouts, [phi] simply penalizes large payouts.

Panel 9 indicates that the payout variance decreases as the payout tax schedule parameter increases. The payout tax parameter is positively related to the marginal tax rate on corporate payouts. Since the tax schedule is convex, as the payout tax parameter increases, a manager who cares about the smoothness of the firm's payouts avoids large and infrequent payouts. This effect results in a less volatile payout policy. However, as shown in Panel 10, the average payout ratio is not significantly affected by the tax schedule parameter [phi]. The last panel finds that as [phi] increases, the firm pays out more often. The more frequent payout, which is smaller in magnitude, leads to a similar payout ratio to the policy induced by lower [phi] featuring large and infrequent distributions. Together, these three panels illustrate that the payout tax schedule parameter [phi] also influences the payout moments of the model.

C. Identification and Matching Moments

The twinned goals of the SMM procedure are: 1) to generate a model that provides a useful and accurate predictor of the behavior of the sampled firms and 2) to provide meaningful estimates of the underlying parameters. To convincingly achieve these goals, these moments must be selected in a principled manner.

The first goal requires that the empirical moments are commonly employed in finance practice. If the moments are not interesting, then the predicted behavior will be of little value. We select moments that are used in the literature and provide a broad picture of the structure and dynamics of the firms' financing and productivity. We also select moments that provide a focused picture of payout policy. To correctly identify the parameter values, we ensure that there are a sufficient number of moments that are informative about each parameter.

We attempt to match the first and second moments of the ratio of cash holdings-to-assets, the payout ratio, the payout variance, the frequency of positive payout, the first and second moments of equity issuance-to-assets, the mean and the variance of the investment-to-assets ratio, the average of the debt-to-assets ratio, the second moment of the debt-to-assets ratio, the frequency of debt reduction, the covariance of cash and payout, and, finally, the ratios of the standard deviation of the shocks to income-to-assets and the serial correlation of income-to-assets. These moments are used to estimate parameters: [gamma], [delta], [[lambda].sub.0], [[lambda].sub.1], [[lambda].sub.2], a, [[sigma], [rho], [phi], q.

The payout ratio, the variance of payout, and the frequency of positive payout are all informative concerning the payout tax schedule [phi]. More importantly, these three moments help us to identify the agency parameter [gamma]. The perceived cost of cutting payouts clearly affects the variance of payouts. This parameter also influences the level of payouts as the manager considers the possibility of costly future reductions in payouts. The frequency of paying out is affected by both [gamma] and [phi] making it a good candidate for the SMM. The magnitude of the perceived cost of cutting dividends also influences corporate cash holdings. Thus, the mean and variance of cash-to-assets is also instructive concerning managerial payout consistency costs.

As discussed in the last section, the frequency of paying out and the average payout ratio react differently to changes in [phi] and [gamma]. Although both parameters negatively affect the variance of payouts, matching the other two moments of payout allows identification of both [phi] and [gamma]. However, given the similar effect of these parameters regarding punishing higher magnitudes of payouts, we estimate the managerial perceived cost of cutting payouts relative to a convex schedule.

The first and second moments of equity-issuance-to-assets are informative concerning the costs of issuing equity. They offer further data about the equity issuance cost parameters [lambda].sub.0], [lambda].sub.1], and [lambda].sub.2]. Costly equity issuance is part of the trade-off model and is one of the reasons firms keep cash balances. Therefore, the first and second moments of cash-to-assets are also provide information regarding these parameters.

The correlation between payouts and cash holdings provides insight regarding the trade-off between paying out and piling up cash that is influenced by [gamma]. This correlation sheds light as to how firms smooth their payouts. In good times, firms may take advantage of their higher cash flow to accumulate cash or to distribute it to their shareholders.

The variance of the investment-to-assets ratio is directly affected by the capital adjustment cost parameter a. Through the trade-off in the use of funds, this moment is indirectly informative regarding [gamma]. A manager who associates a disutility to fluctuations in corporate payouts will maintain cash as a cushion to avoid volatile distributions. This leads to more consistent and more frequent investments of smaller sizes. We also match the mean of the investment-to-assets ratio to its empirical value. This moment responds to the depreciation rate and allows us to estimate this parameter.

Two moments of debt policy are used to estimate the debt recapitalization cost parameter q: 1) the average debt-to-assets ratio and 2) the variance of the debt-to-assets ratio. The average debt-to-assets ratio also provides us with information about the position of debt in the financing hierarchy. The parameter of the recapitalization cost of debt also affects the level of equity issues and cash holdings. As debt recapitalization becomes more costly, equity issues become more attractive to the firm. Cash reserves, as internal financing resources, are another alternative to avoid the cost of issuing equity. As such, cash holdings are also influenced by changes in the debt recapitalization cost parameter. Thus, the means of equity issues-to-assets and cash-to-assets are both indirectly informative about q.

The last set of moments that we include is related to the production side of the firm. The persistence and volatility parameters, [rho] and [sigma], describe the stochastic process influencing the production function. To identify these parameters, we follow the methodology outlined in Holtz-Eakin, Newey, and Rosen (1988) by estimating a first-order panel autoregression of operating income on lagged operating income. This procedure results in the autoregressive coefficient and the standard deviation of the regression residual and provide data regarding [rho] and [sigma]. In the absence of payout smoothing, when managers follow a residual dividend policy, more volatile income leads to a higher payout variance. Therefore, these moments will also be affected by [gamma]. They also influence the precautionary motive for holding cash, as the greater volatility of shocks and the persistence of those shocks may lead to higher optimal levels of cash holdings.

As noted in Nikolov and Whited (2010), unobserved heterogeneity is present in the empirical sample. In contrast, simulated firms used in the SMM methodology are identically distributed. To ensure the moments from the empirical sample are comparable to the simulated moments, fixed firm and year effects are taken into account when estimating the empirical variance and empirical regression-based moments. (17)

D. Simulation and Estimation Results

The data and sample selection procedure are described in Appendix B. We present estimation results for the full sample of firm years. Table I reports the actual moments, the simulated moments for both models, and the t-statistics for differences between the empirical and the simulated moments. When comparing the actual moments with those from the simulations using the base case model, the largest inconsistency is the inability to match the payout variance. The simulated moment is 1.5 times larger than the empirical moment. This is consistent with the dividend smoothing puzzle in Hennessy and Whited (2007). We also note that the structural model predicts a lower average cash-to-assets ratio. These results support the empirical literature arguing that managers tend to smooth corporate payouts, and the high level of cash holdings observed in the data cannot be explained by a trade-off model. The variance of cash-to-assets and the variance of the investment-to-assets ratios are also lower and higher than their respective empirical moments.

The second moment of the investment-to-assets ratio and the cash-to-assets ratio are also inconsistent with their empirical moment. As suggested by the sensitivity analysis, these moments are also affected by managers' payout smoothing behavior. These results suggest the existence of some external parameter influencing the objective function of the firm's manager. When viewed in concert with the overestimation of the payout variance, this evidence suggests the importance of the cost managers' associate with cutting payouts.

In the third column, we report simulated moments using the model that includes the perceived cost of cutting payouts. Overall, these results indicate small differences between the simulated moments and the empirical moments. With the exception of two of the matched moments, the t-statistics are small indicating that the average empirical moments are not statistically different from the average simulated moments. The payout variance is matched closely to its empirical counterpart as are the payout ratio and the frequency of payouts. The mean of the cash-to-assets ratio is still slightly underestimated, but has been increased by a factor of three when compared to the results of the base case model. The payout consistency cost model also leads to a lower correlation between cash holdings and corporate payouts, matching this correlation to its empirical value. Without cost to payout reductions, both cash accumulation and payouts are closely correlated with production. During good days, firms utilize higher cash flows to accumulate cash and distribute dividends to their shareholders. However, in the payout consistency cost model, the perceived cost to cutting payouts results in a smaller payout ratio and larger cash balances. This second force acts on the association between cash holdings and payouts in the opposite direction of the first force, thereby decreasing the large positive correlation to a smaller, though still positive correlation. Overall, this result suggests that the proposed model, in which the manager associates a disutility with cutting dividends, has largely succeeded in matching the simulated moments with the empirical moments.

The second panel of Table I contains the estimated model parameters. Estimated persistence [rho] and the volatility of the cash flow shocks [sigma] using the base case model are 0.665 and 0.262, respectively, and are both statistically significant at the 10% level. Using the payout consistency cost model results in similar estimates for [rho] (0.682), but includes a smaller estimate for [sigma] (0.248). In the absence of payout consistency costs, in an attempt to match the mean and the variance of the cash-to-assets ratio, [sigma] is overestimated. Nevertheless, these estimates are consistent with those reported in Gamba and Triantis (2008) and Moyen (2004). They are also close to Hennessy and Whited (2007) and Nikolov and Whited (2010), which estimate the persistence to be 0.68 and 0.55 and the standard deviation of the shock to be 0.11 and 0.29, respectively.

There is little difference between the values of the depreciation rate [delta] and the debt recapitalization cost parameter q estimated using the base case model and those using the payout consistency cost model. The estimates of q and [delta] are statistically significant at the 10% and 1% level, respectively. Using the payout consistency cost model, the payout tax schedule parameter [phi] is estimated to be 0.320 and is statistically significant at 10%. This estimate is substantially less than its value when using the base case model and also less than the distribution tax schedule parameter found in Hennessy and Whited (2007). This outcome stems from our inclusion of the perceived cost of payout reduction. In the absence of this agency factor, the payout variance is primarily influenced by the payout tax schedule parameter. In an attempt to match the payout variance with its low empirical value, this tax parameter has to be very large. This is illustrated in the "Sensitivity Analysis" section, where larger values of [phi] result in smaller payout variances. However, this motive for payout smoothing alone is not enough to explain the substantially lower payout variance found within the data. Models that do not include managerial disutility for payout reduction, such as Hennessy and Whited (2007), are unable to match the second moment of the distributions precisely. Nevertheless, in an attempt to match this moment with its empirical value, they estimate a large value for [phi].

The base case model results in estimates for the equity issuance cost parameters that are abnormally large as the only motive for cash holding is the precautionary motive. Thus, the cost of equity issuance is overestimated to increase the mean of cash holdings. When using the payout consistency cost model, the estimates for the equity issuance cost parameters become smaller and consistent with those reported in Hennessy and Whited (2007). While the convex cost parameter [[lambda].sub.2] is statistically insignificant, [[lambda].sub.0] and [[lambda].sub.1] are statistically significant at 5% and 10%, respectively.

The managerial payout consistency parameter [gamma] is also statistically significant at the 5% level. The estimated [gamma] is equal to 0.119 indicating that managers who cut their payout, on average, perceive a cost roughly equal to 12% of the magnitude of the payout reduction. We discuss the value loss implications for shareholders in detail in later sections. To consider these estimates from an economic perspective, we also calculate the ratio of managerial disutility to equity value. The results illustrate that, on average, for every million dollars of shareholders' equity value, managers associate a cost equivalent to $81,000 in payout reductions.

E. Sample Splits and Cross-Sectional Variations

We now estimate our dynamic structural model on subsamples that have been split based on different corporate and managerial characteristics. This allows us to both explore possible heterogeneity in the perceived costs of payout reduction across firms and shed light on the motives underlying payout smoothing. For each variable of interest, we split the full sample into four quartiles. We report only estimates for the highest and lowest quartile subsamples. (18)

First, we divide the sample by firm size measured by book assets. Size provides a proxy for several underlying characteristics of the firm that may provide different motives for payout smoothing. Investigating the differences in the estimated agency parameter between the largest and smallest quartiles of the firms sheds light on the existing heterogeneity of managerial preferences.

Firm size is highly correlated with firm age. Larger and older firms usually have better and cheaper access to credit markets. As in Almeida et al. (2004), it has been argued that payout smoothing may arise from an effort to avoid costly external financing. This would lead us to expect more payout smoothing among smaller firms that have less access to external financing. Alternatively, larger and older firms are more prone to possess higher levels of free cash flow. For example, Leary and Michaely (2011) and DeAngelo et al. (2008) find a positive relationship between smoothing and the severity of the free cash flow problem.

Firm size may also proxy for information asymmetry and cash flow uncertainty. Small firms suffer from more information asymmetry and may also be subject to different real shocks as compared to large, mature firms. The empirical volatility of the shocks to income-to-assets is greater for smaller firms. Information asymmetry explanations for payout smoothing suggest that both uncertainty and greater information asymmetry lead firms to increase payout smoothing (see Kumar, 1988; Brennan and Thakor, 1990; Guttman et al. 2001). Uncertainty also affects corporate cash policy. Smaller firms require greater cash balances as a precautionary device for possible shortfalls in future cash flows. The variance of the cash-to-assets ratio is also larger for small firms. While in our sensitivity analysis we noted a positive relationship between cash holdings and payout consistency costs, it is not obvious that this drives the observed higher first and second moments of the cash-to-assets ratio for smaller firms. We begin by dividing the sample according to firm size to examine the differences between the estimated parameters across firms.

Columns 1 and 2 of Table II, respectively, report parameter estimates for large and small firms. Our results indicate a positive association between firm size and managerial payout consistency costs as managers of smaller firms tend to associate less costs with cutting their payouts. The estimated coefficient [gamma] is equal to 0.069 for small firms, which is less than the coefficient of 0.139 estimated for large firms. The smaller magnitude of [gamma] among smaller firms indicates that the higher mean and variance of cash holdings among these firms is not due to an agency problem. Rather, it can be attributed to the higher standard deviation of shocks to income and larger equity issuance costs.

As previously mentioned, firm size may proxy for different firm and managerial characteristics. These results may still mask substantial heterogeneity across firms. In the following splits, we select other proxies that attempt to measure relevant firm and managerial characteristics more directly.

We use the dispersion of analyst forecasts to measure the degree of information asymmetry between managers and investors. Larger forecast dispersion indicates a poorer information environment. Columns 3 and 4 of Table II report the results for subsamples of high and low analyst forecast dispersion. The estimated payout consistency parameter is larger for the subsample of high analyst forecast dispersion (0.126 vs. 0.093). This result suggests that managers of firms facing higher levels of information asymmetry perceive greater disutility from reductions in their payouts. It is interesting to note that the variance of corporate payouts does not differ significantly between the two subsamples (0.0017 vs. 0.0019). When compared to firms with low analyst forecast dispersion, firms with high analyst forecast dispersion experience much higher standard deviations of the shocks to income. The greater volatility of income leads to a higher variance of payouts. The larger estimated managerial payout consistency costs make an otherwise very large payout variance smaller, thereby leaving the two subsamples with similar payout variances.

Consistent with previous empirical studies, such as Leary and Michaely (2011), the volatility of payout is not significantly different among firms operating in environments featuring high and low levels of information asymmetry. However, the estimates of the managerial payout consistency cost parameter illustrates that information asymmetry is positively related to the perceived costs of payout reductions. In contrast to the conclusions of Leary and Michaely (2011), information asymmetry does have an impact on the preference for smooth payouts. This indicates that when examining different hypotheses, the importance of structural models that allow the estimation of unobservable managerial preferences.

The higher managerial payout consistency costs and larger volatility of the shocks to income, evident in the subsample of firms with a high dispersion of analyst forecasts, both act on the same direction and increase the variance of cash holdings. This is consistent with the observed higher volatility of the cash-to-assets ratio in the subsample of firms with greater information asymmetry.

Columns 5 and 6 contain the parameter estimates for the high and low PPS subsamples. PPS is a measure of managerial incentives that is estimated as the dollar value of the CEO's wealth change for a $1,000 change in shareholder value. The main component of the PPS is CEO ownership of stock and stock options. We employ PPS to measure how closely the CEO's wealth is tied to shareholder value. Studies, such as Easterbrook (1984), have suggested that enforcement of a smooth payout is an agency cost treatment to mitigate the agency costs of free cash flow. As all forms of controlling agency costs are themselves costly, we should expect to see substitution among agency cost treatments. Payout smoothing should then become less appealing to shareholders when the manager's incentives are better aligned with those of their shareholders.

Consistent with this explanation, we find that the estimated managerial payout consistency cost is larger for the subsample of low PPS when compared to the estimated parameter for the subsample of high PPS (0.130 vs. 0.075). The high PPS group has a payout variance that is twice as large as the payout variance in the low PPS group.

It is interesting to note that despite a lower payout consistency cost, the cash-to-assets ratio is larger in the high PPS subsample. We have demonstrated that a lower y will result in smaller cash holdings. Thus, to explain this discrepancy, we must examine the other parameter estimates. The standard deviation of shocks to income and the fixed and linear equity issuance cost parameters are all larger in the high PPS subsample, while the capital adjustment cost parameter is smaller. More uncertain cash flows alongside more costly access to equity markets and less smooth investment policies lead to higher precautionary motives for cash holdings. These forces outweigh the effect of the lower payout consistency costs, leading to a larger cash-to-assets ratio in the high PPS group.

The variance of the cash-to-assets ratio is not substantially different in the two subsamples. This is due to the countering effects of higher payout consistency costs and volatility of shocks to income on the variance of cash holdings in the subsample of firms with high PPS.

In Columns 7 and 8, we estimate our model parameters for subsamples of high and low institutional ownership. The importance of institutional investors in monitoring corporate management and influencing payout policy has been investigated in previous studies such as Grinstein and Michaely (2005). Other studies, such as Allen et al. (2000), argue that because of institutional investors, dividends may induce "ownership clientele" effects. The unique tax status of institutional investors leads them to favor firms that offer dividends. As a result, managers can attract these investors who are prized for their monitoring abilities by providing consistent dividends. The result is that firms with high levels of institutional ownership would be expected to maintain consistent payouts.

Consistent with Allen et al. (2000), we find that firms with lower institutional ownership have managers who associate smaller perceived costs to reductions in payouts. As expected, these firms experience higher payout variances. However, the cash-to-assets ratio is not significantly different between the two subsamples (0.148 vs. 0.138). This can be explained by noting that while the lower payout consistency costs and standard deviation of shocks to income induce lower cash balances, the effect of lower estimated investment adjustment costs and larger fixed equity issuance cost parameters counterbalance these forces giving rise to similar cash-to-assets ratios. The variance of cash-to-assets ratio is larger among firms with a higher level of institutional ownership. This may be explained by the impact of the larger estimated managerial payout consistency costs that outweigh the opposing effect of higher volatility of shocks to income.

Numerous studies, such as Leary and Michaely (2011), Jagannathan, Stephens, and Weisbach (2000), and Skinner (2008), document significant time series variation in share repurchases in recent years. This implies that firms do not typically smooth repurchases. While our model does not distinguish between dividends and repurchases, we are still able to test whether managers of firms that pay larger fractions of their payouts in the form of repurchases associate fewer costs to payout reductions. We construct the ratio of share repurchases to total payouts and investigate whether managerial payout consistency costs are sensitive to the type of payouts chosen by the firm.

Columns 9 and 10 report parameter estimates for high and low share repurchase-to-payout ratio subsamples. We find that managers associate less disutility from future payout reductions when they choose to distribute through more flexible share repurchases rather than dividends. This difference is highlighted by the large difference in the payout consistency cost parameter estimates of 0.145 for the low repurchase-to-payout ratio group and 0.050 for the high repurchase-to-payout ratio group.

In Panel B of Table II, we report the equity value loss by comparing the firm in which the manager has incentives to smooth payouts with the same firm where the manager's incentives are perfectly aligned with the shareholders ([gamma] = 0). We deflate the equity value by the firm's book assets. The reported average value loss is measured as the ratio of the deflated equity value of a firm with aligned incentives to the deflated equity value of the firm with parameters derived from our estimation. Using the full sample, the removal of the incentive to smooth dividends results in an increase of 6.1% in shareholders' equity value. On average, shareholders at large firms suffer a loss of 8.5% in equity value as compared to the 1.1% loss imposed on shareholders of smaller firms. The estimated value losses are larger for firms that suffer from greater information asymmetry, provide fewer incentives to their managers through compensation, have higher institutional ownership, and pay a higher fraction of their payout in the form of dividends.

IV. Conclusion

We develop a discrete time, infinite horizon, partial equilibrium, stochastic model of payouts, investment, debt, and cash holdings. In our model, firms trade off the tax benefit of debt against the recapitalization cost of debt. They also save cash in the presence of costly equity issuance and a tax penalty on the cash savings accounts. In the base case model, the firm pays out the residual cash flow after the manager chooses optimal debt, cash, and investment policies to maximize equity value in each period.

We employ SMM by matching a set of empirical moments with simulated moments, estimating parameters from two models: 1) the base case model and 2) the payout consistency cost model. In the payout consistency cost model, consistent with the empirical literature on payout smoothing and inspired by the low level of empirical payout variance, the objective function of the manager is extended to capture the perceived cost of payout reductions.

The main inconsistencies of the base case model are the overestimated payout and investment variances, and the underestimated mean and variance of the cash-to-assets ratio. The payout consistency cost model is able to explain the lower than optimal levels of payout variance and cash variance and higher levels of the investment variance concurrently. This model also helps to explain the high level of corporate cash holdings. As illustrated by comparative statics, the payout consistency cost can only partially explain the larger than optimal cash balances. This suggests that this cost is one of the many motives underlying the observed excess cash.

Our comparative statics results indicate that an increase in the payout consistency cost parameter is associated with reductions in the variance of payouts, the payout ratio, and the variance of investments, as well as an increase in the mean and the variance of cash balances. Empirical studies, such as Leary and Michaely (2011), attribute the positive correlation between cash and payout smoothing to the remedial view of Easterbrook (1984). The joint determination of payout and cash policies and the dynamic effects of requiring smooth dividends provide an alternative explanation for this correlation. While income, payouts, and investments all become smoother, cash holdings become more volatile to absorb the variability that would be shouldered by payouts and investments in the absence of payout smoothing.

We also determine that the managerial payout consistency cost accounts for a 6.1% loss in shareholder equity value. It is important to note that our model does not include any other agency conflicts, nor does it incorporate any sort of information asymmetry. The estimated shareholder loss in our paper is relative to the first-best equity values for a firm that faces no agency conflicts or information asymmetry. One interesting avenue for future research is to attempt to incorporate these other frictions within a structural model to quantify the possible benefits to payout smoothing.

We further explore the heterogeneity of the agency parameter across different firms by employing SMM on sample splits based on various firm-manager characteristics. We provide evidence that firms that are larger, have larger analyst forecast dispersion, have larger institutional ownership, that compensate their CEOs with low pay-performance packages, and that distribute larger fractions of their payouts in the form of dividends, have managers who associate larger costs to cutting their payouts. These results provide partial support for several common explanations of payout smoothing. Interestingly, like Leary and Michaely (2011), we find that differences in information asymmetry do not lead to significantly different payout variances; however, we find that this disguises a significantly larger payout consistency cost parameter estimate for firms with higher analyst forecast dispersion.

Our study focuses on payout smoothing rather than dividend smoothing. In our model, when estimating the managerial perceived cost to payout reductions, we do not differentiate between dividends and share repurchases. However, consistent with the empirical literature, our sample split results indicate that firms who distribute cash to their shareholders through repurchase programs are less inclined to smooth their payouts. The differences in the level of managerial commitments associated with each of these methods of payout, as well as other institutional differences such as taxes, suggest potentially productive avenues for future research.

Appendix A: Computational Method

In this section, we present a method for solving the Bellman equation described in Equation (12). To simplify the exposition, we consider a slightly abstracted version consistent with conventions in dynamic programming theory. In particular, Equation (12) can be represented in the following form:

V([S.sub.t]; [[theta].sub.t]) = D([S.sub.t], g([S.sub.t]; [[theta].sub.t])) + [1/[1 + r]] [E.sub.t] [V(tr([S.sub.t], g([S.sub.t]; [[theta].sub.t]); [[theta].sub.t]); [[theta].sub.t+1])]. (A1)

A solution consists of a policy g : S x [THETA] [right arrow] A that maps a nonstochastic state in S and stochastic state in [THETA] to an action in A. The nonstochastic state space S accounts for the capital [K.sub.t], cash [C.sub.t] and debt [B.sub.t]. The action space accounts for [K.sub.t+1], [C.sub.t+1], and [B.sub.t+1]. The function [D.sub.t] : S x A [right arrow] R is equal to the dividend payment described in Equation (8). The transition function tr(*) calculates the outstanding debt, and capital and cash on hand at period t + 1. For capital, cash and debt the transition function trivially maps to the respective action variables.

An approximate solution is found by discretizing the state and action space to generate a discrete Markov decision problem that is then solved for the optimal policy using a standard value function iteration algorithm. The state space is composed of an evenly spaced grid with 12 points per variable with the exception of capital stock. Since this is where the majority of the curvature occurs in the model, we use a more finely spaced grid composed of 24 points. The autoregressive process underlying the stochastic state is approximated by a discrete state Markov chain constructed with the method described in Tauchen (1986). The process is created with nine states spanning [e.sup.[+ or -]3[sigma]]. The discretized transition function selects the closest grid point to the output of the continuous transition function. Ex post, we verify that stable policies do not occur on any boundary points of the state space variables.

We use a slightly altered version of the above methodology in the appendix that incorporates risky debt into the base case model. In this model, equity holders have the choice to default if the expected equity value of the firm is negative. The resulting probability of default [PSI] is calculated for the product of states and actions, and results in a significant increase in computational complexity.

Together with the fair pricing of debt described in Equation (C2), [PSI] is used to endogenously calculate interest rate [i.sub.t]. Since fair pricing itself depends upon the value function, the constraint on interest does not allow the Bellman equation to fit into the standard dynamic programming paradigm. However, under reasonable restrictions on the size of interest payments, the Blackwell sufficient condition may be used to demonstrate that adding this penalty imposed by the interest charge to Equation A1 still results in a contraction mapping. This result is sufficient to demonstrate the uniqueness and existence of a stable optimal solution. The resulting recursive equation is discretized as above and then solved with an appropriately modified value function iteration algorithm.

Appendix B: Data and Sample Selection

Accounting data used in this study are drawn from the annual 2006 Compustat industrial files. We ensure that entries are omitted from regulated financial, and public service firms by deleting entries where the primary standard industrial classification (SIC) is within 4900 and 4999, within 6000 and 6999, or greater than 9000. From the remaining data, we remove any firm years where data are missing or for which total assets or sales are zero or negative. We also filter out any firms in the sample that have less than three consecutive years of complete data. We winsorize the top 1% of the variables in the remaining data entries. The resulting panel of firms is for years 1988-2006 and has between 1,966 and 3,183 observations per year.

The data variables correspond to items from Compustat as follows: total book assets is Item 6, capital stock is Item 7, cash flow is the sum of Items 18 and 14, investment is the difference between Items 30 and 107, equity issuance is Item 108, total debt is Item 9 plus Item 34 plus item 104, total payout is the sum of Item 19, Item 21, and Item 115, cash holdings is measured by Item 1, and revenue is Item 12. The market-to-book ratio is measured by total book assets minus book equity (Item 60) and the market value of equity (Item 199 times Item 25) less deferred taxes (Item 7) all divided by the total book value of assets.

We collect data on institutional holdings from Thomson Financial. These institutions include bank trusts, insurance companies, mutual funds, brokerage firms, pension funds, and endowments. We do not include individual blockholders. Institutional holdings are measured by the total stock ownership of all of the institutional investors. As a robustness check, following Grinstein and Michaely (2005) and Hartzell and Starks (2003), we also measure institutional ownership using the proportion of institutional ownership by the top five institutional investors in the firm. Our results are not sensitive to this alternative measure. To measure the information asymmetry of a firm, we calculate the standard deviation of analyst annual earnings forecasts from Institutional Brokers' Estimate System (IBES).

The executive compensation data are from ExecuComp. Following Jensen and Murphy (1990), PPS is the dollar value of the CEO's wealth change for a $1,000 change in shareholder value. Although managers can receive pay-performance incentives from a variety of sources, the majority are due to ownership of stock and stock options. Similar to Aggarwal and Samwick (2003) and Core and Guay (1999), we compute this sensitivity as the dollar value change of stock and options held by a CEO to a $1,000 shareholder return. For common stock, PPS is simply the fraction of the firm that the executive owns. PPS for options is the fraction of the firm's stock on which the options are written multiplied by the options' delta. We use the method developed by Core and Guay (2002) to estimate option deltas.

Appendix C: The Base Case Model with Risky Debt

We investigate how the introduction of risky debt would alter our base case results. The structure of our risky debt model is related to Moyen (2004, 2007). These papers investigate investment and financing decisions of a firm within an infinite horizon discrete time dynamic stochastic framework. In Moyen (2004), unlike our model, there is no need for cash since raising capital is not costly to the firm. In our model with risky debt, firms not only trade off a tax benefit of debt against the expected cost of bankruptcy and the recapitalization cost of debt, but also make decisions on the level of cash holdings in the presence of costly equity issuance.

Similar to the base case model, the firm maximizes equity value by choosing its dividend, investment, cash holdings, and debt policy. However, the firm's optimization is also subject to debt issues being fairly priced. Claimants on both equity and debt are risk neutral. The equity value [V.sub.t] takes the form:

[V.sub.t] = max{0, [D.sub.t] + T([D.sub.t]) + [LAMBDA]([D.sub.t]) + [1/1 + (1 - [[tau].sub.b])r]] [E.sub.t][[V.sub.t+1]]}. (C1)

This equation illustrates that equity holders have limited liability. Equity value equals the value of a call option. Equity claimants default whenever [D.sub.t] + T([D.sub.t]) + [LAMBDA]([D.sub.t]) + [1/[[1 + 1(1 - [[tau].sub.b])r]] [E.sub.t][[V.sub.t+1]] [less than or equal to] 0and otherwise receive the expected discounted stream of dividends [D.sub.t]. The firm may raise more external equity ([D.sub.t] < 0). However, the equity holders can choose to either add to their stake or give up their equity claim.

Similar to the risk-free debt model, the change in debt, [DELTA]Bt + 1, may represent a new issuance ([DELTA][B.sub.t+1] > 0) or retirement ([DELTA][B.sub.t+1] < 0). Unlike the risk-free debt model, the interest rate requested by creditors can change in each period. This interest rate becomes larger as the firm approaches the default boundary. Fair pricing of debt requires that:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (C2)

Again, debt claimants' face interest income tax rate [[tau].sub.b]. The deadweight default loss is the proportion [xi] of the debt face value. The indicator function [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is one if the firm does not default ([V.sub.t] > 0) and zero otherwise.

The fair pricing of debt implies that debt holders request an interest rate such that the value of their holdings in that period equals the expected discounted payoff in the following period. When equity claimants do not default, debt claimants receive their principal [B.sub.t+1] in addition to the after tax interest payment (1 - [[tau].sub.b])[i.sub.t+1][B.sub.t+1]. In the case of default, the debt claimants receive the net residual value R([K.sub.t+1]; [[theta].sub.t+1]) - [xi][B.sub.t+1].

The residual [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] going to the debt claimant upon default is the value of the firm after reorganization. [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and captures the optimal recapitalization:

R([K.sub.t+1], [C.sub.t+1]; [[theta].sub.t+1]) = (1 - [[tau].sub.c])f([K.sub.t];[[tau].sub.t]) + [[tau].sub.c][delta][K.sub.t] - [I.sub.t] + [B.sub.t+1] + [LAMBDA]([D.sub.t]) + T([D.sub.t])

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (C3)

The net residual value [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] going to debt claimants upon bankruptcy (i.e., when [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]) is always less than the debt holder's payment when there is no default, (1 + (1 - [[tau].sub.b])[i.sub.t])[B.sub.t]. (19) The recapitalization cost for new debt is similar to the risk-free debt model.

Using Equations (2)-(8), (10), (11), (C1), and (C2), we can implicitly express the income shock at which equity claimants trigger default, [bar.[theta]]([K.sub.t], [B.sub.t], [i.sub.t], [C.sub.t]). This is calculated by solving:

[D.sub.t] + T([D.sub.t]) + [LAMBDA]([D.sub.t]) + [1/1 + [(1 - [[tau].sub.b])r]][E.sub.t][[V.sub.t+1]] = 0,

and leads to the following expression for [bar.[theta]][K.sub.t], [B.sub.t], [i.sub.t], [C.sub.t]):

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (C4)

where the probability of default [PSI]([bar.[theta]]([K.sub.t], [B.sub.t], [i.sub.t], [C.sub.t])) follows a log-normal cumulative density function.

In this model, the manager maximizes the equity value of the firm such that Equations (2)-(11) and (C2) are satisfied. The firm's optimal level of cash holdings depends not only on the cost of issuing equity, but also on the firm's holdings of risky debt and the probability of facing costly default. If debt were not risky, the firm could avoid costly equity issues and raise more debt to finance its new investments. In this set up, the expected cost of default on the risky debt, as well as the flotation cost of new debt issues, in concert with the cost of issuing equity, enables us to address the simultaneous existence of debt and cash balances in the firm.

The manager chooses corporate policies that are also constrained by the bond pricing Equation (C2). The Bellman equation describing the firm's intertemporal problem is:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (C5)

subject to Equations (2)-(11) and (C2).

The model cannot be solved analytically. The solution is approximated using numerical methods. The employed numerical method, discussed in Appendix A, is similar to the method used to solve the base case model with risk-free debt. A firm is defined as a series of at least 20 consecutive periods without default. We drop the first 20 periods to avoid observing the firm before it has worked its way out of an unlikely initial state. Periods 21-40 are used to construct the simulated panel. Note that 0.55% of firms default within periods 21-40. When the next period's realized income shock is lower than expected, servicing the equity holder's chosen debt level may become too difficult.

We compare the Euler equation for optimal cash holdings from the base case model with risk-free debt (Equation (13)) with its analog when risky debt is incorporated. The following Euler equation relates the dynamics of interest and probability of default with changes in cash:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (C6)

The left-hand side represents the marginal cost of external equity finance, the marginal cost from the tax on payout, plus the marginal cost of default on the debt obligations, while the right-hand side represents the marginal benefit of cash balances. In addition to factors present in the risk-free model, saving a dollar today also influences both the probability of default and the interest rate promised to the debt claimants in the next period.

With risky debt, cash holdings affect both the interest rate requested by debt holders and the probability of default. The above equation confirms that the optimal savings decision considers both of these factors. Higher interest rates promised to debt claimants result in a larger tax benefit of debt, but it also increases the probability of default.

Setting the parameters to their respective values obtained from solving the base case model, we solve this model numerically. Similar to Moyen (2004), the dead weight cost of default is set to [xi] = 0.1 to compromise between Fischer, Heinkel, and Zechner (1989) and Kane, Marcus, and McDonald (1986), who use 5% and 15% of the debt face value for the dead weight cost. Table C1 reports the results. As the Euler equation suggested earlier, the average cash-to-assets ratio is higher than the same ratio in the base case model with risk-free debt (0.0813 vs. 0.0689). However, similar to the base case model, this model overestimates the variance of the investment-to-assets ratio and underestimates the variance of the cash-to-assets ratio. More importantly, with respect to the payout variance, the model significantly overshoots this moment. Although risky debt influences corporate policies such as cash holdings and leverage, it does not help to explain the smooth payout, investment, and the volatile cash holdings observed in the data. As in our analysis of the base case model, the results from the model including risky debt with endogenous default suggest a need to incorporate a perceived cost to cutting payouts into the manager's utility function.

Table C1. Exogenously Simulated and Empirical Moments: The Risky Debt
Model

This table reports the simulated and empirical moments using the base
case model with risky debt. The em-pirical moments are based on a
sample of nonfinancial, unregulated firms from the annual 2006
Compustat industrial files. The sample period is 1988-2006.
Estimation is done using the base case model including risky debt.
The parameters are exogenously set to their estimated values from
Column 2 of Table I using the base case model. The simulated panel of
firms is generated from the base case model with risky debt as
described in the appendix, and contains 20,000 firms over 40 time
periods, where only the last 19 time periods are kept for each firm.

Name of Moments                         Empirical   Simulated
                                         Moments     Moments

Average cash/assets                      0.1631      0.0813
Variance of cash/assets                  0.0059      0.0038
Average investment/assets                0.1101      0.1229
Variance of investment/assets            0.0057      0.0078
Average equity issuance/assets           0.0368      0.0425
Variance of equity issuance/assets       0.0113      0.0126
Payout ratio                             0.2072      0.2017
Variance of payout                       0.0016      0.0023
Frequency of paying out                  0.4511      0.4812
Average debt/assets                      0.2682      0.1802
Variance of debt/assets                  0.0089      0.0074
Correlation of payout and cash/assets    0.0543      0.2118
Standard deviation of the shock to       0.1121      0.1001
  income/assets
Serial correlation of income/assets      0.6782      0.5930


This paper benefited extensively from discussions with Jan Mahrt-Smith, Craig Doidge, Raymond Kan, Sergei Davydenko, Alexander Dyck, and Marcel Rindisbacher. We are also grateful for suggestions from Bill Christie (Editor) and an anonymous referee that have greatly improved the paper. We would like to thank the Rotman Finance Lab for providing computationalfacilities that made this study possible while we were at the University of Toronto. All errors are the authors' responsibility. This paper was previously circulated under the title "What Drives Corporate Excess Cash? Evidence from a Structural Estimation."

References

Aggarwal, R.K. and A.A. Samwick, 2003, "Why Do Managers Diversify Their Firms? Agency Reconsidered," Journal of Finance 58, 71-118.

Aivazian, V A., L. Booth, and S. Cleary, 2009, "Dividend Smoothing and Debt Ratings," Journal of Financial and Quantitative Analysis 41, 439-453.

Allen, F., A.E. Bernardo, and I. Welch, 2000, "A Theory of Dividends Based on Tax Clienteles," Journal of Finance 55, 2499-2536.

Almeida, H., M. Campello, and M.S. Weisbach, 2004, "The Cash Flow Sensitivity of Cash," Journal of Finance 59, 1777-1804.

Altinkilig, O. and R.S. Hansen, 2000, "Are There Economies of Scale in Underwriting Fees? Evidence of Rising External Financing Costs," Review of Financial Studies 13, 191-218.

Bates, T.W., K.M. Kahle, and R.M. Stulz, 2009, "Why Do US Firms Hold So Much More Cash Than They Used To?" Journal of Finance 64, 1985-2021.

Booth, L. and Z. Xu, 2008, "Who Smoothes Dividends?" University of Toronto Working paper.

Brav, A., J.R. Graham, C.R. Harvey, and R. Michaely, 2005, "Payout Policy in the 21st Century," Journal of Financial Economics 77, 483-527.

Brennan, M.J. and A.V Thakor, 1990, "Shareholder Preferences and Dividend Policy," Journal of Finance 45, 993-1018.

Cooper, R.W. and J.C. Haltiwanger, 2006, "On the Nature of Capital Adjustment Costs," Review of Economic Studies 73, 611-633.

Core, J. and W. Guay, 1999, "The Use of Equity Grants to Manage Optimal Equity Incentive Levels," Journal of Accounting and Economics 28, 151-184.

Core, J. and W. Guay, 2002, "Estimating the Value of Employee Stock Option Portfolios and Their Sensitivities to Price and Volatility," Journal of Accounting Research 40, 613-630.

DeAngelo, H., L. DeAngelo, and D.J. Skinner, 2008, "Corporate Payout Policy," Foundations and Trends in Finance 3, 95-287.

Dittmar A. and J. Mahrt-Smith, 2007, "Corporate Governance and the Value of Cash Holdings," Journal of Financial Economics 83, 599-634.

Easterbrook, F.H., 1984, "Two Agency-Cost Explanations of Dividends," American Economic Review 74, 650-659.

Erickson, T. and T.M. Whited, 2000, "Measurement Error and the Relationship Between Investment and q," Journal of Political Economy 108, 1027-1057.

Fama, E.F. and H. Babiak, 1968, "Dividend Policy: An Empirical Analysis," Journal of the American Statistical Association 63, 1132-1161.

Fischer, E.O., R. Heinkel, and J. Zechner, 1989, "Dynamic Capital Structure Choice: Theory and Tests," Journal of Finance 44, 19-40.

Francis, B.B., I. Hasan, K. John, and L. Song, 2011, "Corporate Governance and Dividend Payout Policy: A Test Using Antitakeover Legislation," Financial Management 40, 83-112.

Gamba, A. and A.J. Triantis, 2008, "The Value of Financial Flexibility," Journal of Finance 63, 2263-2296.

Grinstein, Y. and R. Michaely, 2005, "Institutional Holdings and Payout Policy," Journal of Finance 60, 1389-1426.

Guttman, I., O. Kadan, and E. Kandel, 2001, "A Theory of Dividend Smoothing," Stanford University, Working paper.

Harford, J., S.A. Mansi, and W.F. Maxwell, 2008, "Corporate Governance and Firm Cash Holdings in the US "Journal of Financial Economics 87, 535-555.

Hartzell, J.C. and L.T. Starks, 2003, "Institutional Investors and Executive Compensation," Journal of Finance 58, 2351-2374.

Hennessy, C. and T. Whited, 2007, "How Costly is External Financing? Evidence from a Structural Estimation," Journal of Finance 62, 1705-1745.

Holtz-Eakin, D., W. Newey, and H.S. Rosen, 1988, "Estimating Vector Autoregressions with Panel Data," Econometrica 1371-1395.

Jagannathan, M., C.P. Stephens, and M.S. Weisbach, 2000, "Financial Flexibility and the Choice Between Dividends and Stock Repurchases," Journal of Financial Economics 57, 355-384.

Jensen, M.C., 1986, "Agency Cost Of Free Cash Flow, Corporate Finance, and Takeovers," American Economic Review 76, 323-329.

Jensen, M.C. and K.J. Murphy, 1990, "Performance Pay and Top-Management Incentives," Journal of Political Economy 98, 225-264.

Kane, A., A.J. Marcus, and R.L. McDonald, 1986, "Debt Policy and the Rate of Return Premium to Leverage," NBER Working paper.

Kumar, P., 1988, "Shareholder-Manager Conflict and the Information Content of Dividends," Review of Financial Studies 1, 111-136.

Kumar, P. and B.S. Lee, 2001, "Discrete Dividend Policy with Permanent Earnings," Financial Management 30, 55-76.

Lambrecht, B.M. and S.C. Myers, 2010, "A Lintner Model of Dividends and Managerial Rents," NBER Working paper.

Leary, M.T. and R. Michaely, 2011, "Determinants of Dividend Smoothing: Empirical Evidence," Review of Financial Studies 24, 3197-3249.

Li, K. and X. Zhao, 2008, "Asymmetric Information and Dividend Policy," Financial Management 37, 673-694.

Lintner, J., 1956, "Distribution of Incomes of Corporations Among Dividends, Retained Earnings, and Taxes," American Economic Review 46, 97-113.

Lourengo, H.R., O.C. Martin, and T. Stutzle, 2003, "Iterated Local Search," International Series in Operations Research and Management Science 57, 321-354.

Mansi, S.A. and J.K. Wald, 2011, "Payout Policy with Legal Restrictions," Financial Management 40, 701-732.

Michaelides, A. and S. Ng, 2000, "Estimating the Rational Expectations Model of Speculative Storage: A Monte Carlo Comparison of Three Simulation Estimators," Journal of Econometrics 96, 231-266.

Morellec, E., B. Nikolov, and N. Schurhoff, 2012, "Corporate Governance and Capital Structure Dynamics," Journal of Finance 67, 803-848.

Moyen, N., 2004, "Investment-Cash Flow Sensitivities: Constrained versus Unconstrained Firms," Journal of Finance 59, 2061-2092.

Moyen, N., 2007, "How Big is the Debt Overhang Problem?" Journal of Economic Dynamics and Control 31, 433-472.

Moyen, N. and M. Boileau, 2009, "Corporate Precautionary Savings," University of Colorado Working paper.

Nikolov, B. and T.M. Whited, 2010, "Agency Conflicts and Cash: Estimates from a Structural Model," University of Rochester Working paper.

Opler, T., L. Pinkowitz, R. Stulz, and R. Williamson, 1999, "The Determinants and Implications of Corporate Cash Holdings," Journal of Financial Economics 52, 3-46.

Riddick, L.A. and T.M. Whited, 2009, "The Corporate Propensity to Save," Journal of Finance 64, 1729-1766.

Skinner, D.J., 2008, "The Evolving Relation Between Earnings, Dividends, and Stock Repurchases," Journal of Financial Economics 87, 582-609.

Tauchen, G., 1986, "Finite State Markov-Chain Approximations to Univariate and Vector Autoregressions," Economics Letters 20, 177-181.

(1) For evidence regarding dividend smoothing, see Fama and Babiak (1968) and Brav et al. (2005).

(2) For a complete survey on payout policy including agency and information asymmetry justifications, please see DeAngelo, DeAngelo, and Skinner (2008).

(3) See Allen et al. (2000), Easterbrook (1984), and Jensen (1986) for agency explanations. A differing agency explanation by Lambrecht and Myers (2010) explains dividend smoothing as a consequence of managerial rent smoothing. See Kumar (1988), Brennan and Thakor (1990), and Guttman et al. (2001) for dividend smoothing theories based on asymmetric information.

(4) The Securities and Exchange Commission's (SEC) concern over stock price manipulation and US Internal Revenue Service (IRS) regulations to prevent replacement of dividends with systematic share repurchases leads to smaller payouts in the form of repurchases, while larger distributions are done through tax disadvantaged dividends.

(5) In this section, we refer to any positive payout as dividends. [D.sub.t] > 0 represents the sum of repurchases and dividends.

(6) An alternative is to replace our quadratic capital adjustment cost function with a similar function that also includes a fixed cost to investment: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. This functional form, inspired by the Cooper and Haltiwanger (2006), has been used in Riddick and Whited (2009) and Nikolov and Whited (2010). Although the fixed cost leads to slightly higher optimal cash levels, it does not change the payout variance. Hence, we focus on the simpler quadratic cost function.

(7) We assume there are "perfect" debt covenants restricting the manager from asset sales, etc.

(8) In Appendix C, we replace the current simplified risk-free debt structure with more elaborate risky debt in which the requested interest rate by the creditors becomes an endogenous variable changing with how far the firm is from default. Here, we settle for the simpler debt structure as: 1) our focus of the paper is not firm capital structure, and 2) the risky debt model presents significant computational challenges resulting in problems for the SMM procedure. Nevertheless, in the appendix, we demonstrate that the base case model with risky debt does not lead to a dramatic difference in the payout ratio and payout variance. We thank the referee for pointing this out.

(9) The functional form of the recapitalization cost for debt is consistent with Gamba and Triantis (2008).

(10) We do not include a collateral constraint as it does not change our main results. Moreover, both the recapitalization cost of debt and a collateral constraint play a similar role by providing a ceiling on debt.

(11) This may be justified by simply assuming that the manager owns a portion of the firm's equity.

(12) This is one of the many possible approaches to model managerial preference for smooth payouts. We chose this special type of disutility after observing the suboptimal level of payout variance (relative to the base case results) in the empirical sample.

(13) We abstract from distinguishing among different types of disutility a manager may experience in the case of cutting dividends since the main focus of the paper is to estimate the magnitude of the cost and the consequent loss in equity value. The cost may arise from pressure from institutional investors requesting dividends as suggested by Allen et al. (2000) or not being able to convey credible information to the market as suggested by Kumar (1988), Kumar and Lee (2001), and Guttman et al. (2001).

(14) Moyen and Boileau (2009) use the average of the monthly annualized t-bill rate deflated by the consumer price index to determine the risk-free rate: r = 1.6091%.

(15) Michaelides and Ng (2000) find that a simulated sample of about 10 times the size of the actual data is required in order to produce reliable estimates from an indirect inference estimator.

(16) Other motives that are beyond the scope of our paper (e.g., liquidity motives captured in Moyen and Boileau, 2009) may contribute to the observed excess cash.

(17) As an alternative, one could add firm-specific heterogeneity to the simulations similar to Morellec, Nikolov, and Schurhoff (2012).

(18) Since our sample split variables are, at best, rough proxies of the true underlying constructs, we use only the high and low quartiles. This reduces the possibility that a firm is placed in an incorrect subsample.

(19) The residual is smaller than the principal and after tax interest that would be received by the debt holders when the firm does not default [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. This is because the tax on the interest payment is larger for the firm than the tax debt holders pay on interest income ([t.sub.c] > [t.sub.b]).

Hamed Mahmudi and Michael Pavlin *

* Hamed Mahmudi is an Assistant Professor in the Price College of Business at the University of Oklahoma in Norman. OK. Michael Pavlin is an Assistant Professor in the School of Business and Economics at Wilfrid Laurier University in Waterloo, ON, Canada.

Table I. Matching Moments and Estimated Model Parameters for the Full
Sample

This table presents empirical and simulated moments with associated
parameter estimates. Empirical moments are based on a sample of
nonfinancial, unregulated firms from 1988 to 2006 of the 2006
Compustat industrial files. The parameter estimation is done using SMM
to match moments of a simulated panel to corresponding empirical
moments. The simulated moments in Column 1 and 2 are from a data panel
generated from the base case model and the model with payout
consistency cost (Section IIIA) with estimated parameters,
respectively. Each panel includes 20,000 simulated firms for which the
last 19 of 40 time periods are retained. Empirical and simulated
moments are reported in Panel A. The r-statistics for the difference
between the empirical and the simulated moments are displayed in
parentheses. Panel B reports the estimated parameters, including p-
values in parentheses. Estimated parameters include the fixed, linear,
and quadratic equity issuance cost parameters [[lambda].sub.0],
[[lambda].sub.1], [[lambda].sub.2]; the payout consistency cost
parameter y; the depreciation rate, the flotation cost of new debt
issues q: the payout tax schedule parameter [gamma]; the capital
adjustment cost parameter a; and the standard deviation and serial
correlation of the cash flow shock parameters [sigma] and [rho]. The
final entry, [x.sup.2], reports the chi-squared statistic with four
degrees of freedom for the test of the overidentifying restrictions.
Its p-value is reported in parentheses.

Panel A. Moments

Name of Moments                (1)             (2)              (3)
                            Empirical   Simulated Moments    Simulated
                             Moments     Base Case Model      Moments
                                                              Payout
                                                            Consistency
                                                            Cost Model

Average cash/assets          0.1631          0.0689           0.1438
                                             (5.034)          (1.961)
Variance of cash/assets      0.0059          0.0046           0.0055
                                             (1.782)          (0.885)
Average investment/assets    0.1101          0.1231           0.1197
                                            (-0.636)         (-0.623)
Variance of investment/      0.0057          0.0073           0.0054
  assets                                    (-1.592)          (1.177)
Average equity issuance/     0.0368          0.0291           0.0311
  assets                                     (1.243)          (0.963)
Variance of equity           0.0113          0.0128           0.0104
  issuance/assets                           (-1.751)          (1.684)
Payout ratio                 0.2072          0.2216           0.1931
                                            (-1.782)          (1.646)
Variance of payout           0.0016          0.0024           0.0013
                                            (-2.879)          (0.934)
Frequency of paying out      0.4511          0.4834           0.4527
                                            (-0.432)         (-0.613)
Average debt/assets          0.2682          0.2851           0.2690
                                            (-1.790)         (-1.803)
Variance of debt/assets      0.0089          0.0078           0.0073
                                             (1.801)          (1.789)
Correlation of payout and    0.0543          0.1687           0.0597
  cash/assets                               (-2.154)         (-0.928)
Standard deviation of the    0.1121          0.1033           0.1104
 shock to income/assets                      (0.326)          (0.462)
Serial correlation of        0.6782          0.6870           0.6887
  income/assets                             (-1.849)         (-1.850)

Panel B. Estimated Parameters

              [[lambda].sub.0]   [[lambda].sub.1]   [[lambda].sub.2]

Base case           0.726             0.093              0.0004
Model              (0.089)           (0.044)            (0.121)
Consistency         0.485             0.074              0.0004
Cost model         (0.086)           (0.036)            (0.118)

                   [gamma]   [delta]      q       [phi]

Base case                     0.115     0.033     0.597
Model                        (0.008)   (0.082)   (0.112)
Consistency         0.119     0.113     0.038     0.320
Cost model         (0.046)   (0.007)   (0.087)   (0.095)

                      a       [rho]    [sigma]   [chi square]

Base case           0.782     0.665     0.262       8.02
Model              (0.076)   (0.063)   (0.080)     (0.091)
Consistency         0.581     0.682     0.248       8.13
Cost model         (0.089)   (0.061)   (0.065)     (0.087)

Table II. Estimated Model Parameters for Different Sample Splits

This table provides parameter estimates and equity value losses for
different sample splits. Empirical moments are based on a sample of
nonfinancial, unregulated firms from 1988 to 2006. The parameter
estimation is done using SMM to match moments of a simulated panel to
corresponding empirical moments. The simulated moments are from a
data panel generated from the model (Section IIB) with estimated
parameters. This panel includes 20,000 simulated firms for which the
last 19 of 40 time periods are retained. Panel A reports the
estimated model parameters for different sample splits, with
p-values in parentheses. The sample is split with respect to high and
low: 1) firm size, 2) analyst forecasts dispersion, 3) CEO
pay-performance sensitivity, 4) institutional ownership, and 5) share
repurchase to total payout ratio. High and low refer to the first and
last quartiles of the distribution, respectively. Estimated
parameters include: the fixed, linear, and quadratic equity issuance
cost parameters [[lambda].sub.0], [[lambda].sub.1], and
[[lambda].sub.2]; the payout consistency cost parameter [gamma]; the
depreciation rate [delta], the flotation cost of new debt issues q;
the payout tax schedule parameter [phi]; the capital adjustment cost
parameter a; and the standard deviation and serial correlation of the
cash flow shock parameters [sigma] and [rho]. The final entry,
[x.sup.2], reports the chi-squared statistic with four degrees of
freedom for the test of the overidentifying restrictions. Its
p-value is reported in parentheses. Panel B presents the equity value
loss due to the misalignment of incentives between managers and
shareholders. Equity value is deflated by the capital stock. Equity
value loss is measured as the ratio of the value of a firm with no
agency conflicts (i.e., [gamma] = 0) to the value of a firm described
by our model using the estimated parameters.

Panel A. Estimated Structural Parameters for Different Sample Splits

Estimated            (1)       (2)        (3)          (4)
Parameters          Large     Small    High Info.   Low Info.
                    Firms     Firms    Asymmetry    Asymmetry

[[lambda].sub.0]    0.389     0.890      0.460        0.375
                   (0.088)   (0.097)    (0.073)      (0.073)
[[lambda].sub.1]    0.059     0.097      0.058        0.089
                   (0.057)   (0.076)    (0.069)      (0.070)
[[lambda].sub.2]   0.0005    0.0005      0.0003      0.0004
                   (0.270)   (0.167)    (0.262)      (0.301)
[gamma]             0.139     0.069      0.126        0.093
                   (0.032)   (0.070)    (0.052)      (0.055)
[delta]             0.119     0.113      0.109        0.116
                   (0.009)   (0.015)    (0.011)      (0.015)
q                   0.023     0.062      0.058        0.012
                   (0.086)   (0.091)    (0.060)      (0.060)
[phi]               0.315     0.570      0.290        0.285
                   (0.074)   (0.162)    (0.075)      (0.090)
a                   0.391     0.696      0.628        0.481
                   (0.043)   (0.087)    (0.045)      (0.051)
[rho]               0.784     0.559      0.605        0.706
                   (0.061)   (0.123)    (0.130)      (0.121)
[sigma]             0.137     0.272      0.290        0.117
                   (0.089)   (0.073)    (0.073)      (0.094)
[chi square]        7.90      7.89        8.04        7.89
                   (0.095)   (0.096)    (0.090)      (0.086)

Panel B. Equity Value Loss

Full                Large     Small    High Info.   Low Info.
Sample              Firms     Firms    Asymmetry    Asymmetry

0.061               1.085     1.011      1.065        1.035

Panel A. Estimated Structural Parameters for Different Sample Splits

Estimated            (5)       (6)       (7)
Parameters          High       Low       High
                     PPS       PPS      Inst.
                                       Holdings

[[lambda].sub.0]    0.575     0.372     0.438
                   (0.040)   (0.066)   (0.067)
[[lambda].sub.1]    0.074     0.059     0.071
                   (0.063)   (0.075)   (0.086)
[[lambda].sub.2]   0.0004    0.0004     0.0003
                   (0.345)   (0.232)   (0.155)
[gamma]             0.075     0.130     0.124
                   (0.059)   (0.036)   (0.030)
[delta]             0.110     0.121     0.129
                   (0.012)   (0.014)   (0.014)
q                   0.039     0.047     0.021
                   (0.085)   (0.079)   (0.073)
[phi]               0.310     0.298     0.205
                   (0.071)   (0.088)   (0.075)
a                   0.457     0.675     0.692
                   (0.083)   (0.073)   (0.045)
[rho]               0.679     0.694     0.714
                   (0.092)   (0.076)   (0.093)
[sigma]             0.271     0.186     0.226
                   (0.652)   (0.069)   (0.060)
[chi square]        8.15      8.30       7.90
                   (0.079)   (0.081)   (0.095)

Panel B. Equity Value Loss

Full                High       Low       High
Sample               PPS       PPS      Inst.
                                       Holdings

0.061               1.026     1.069     1.080

Panel A. Estimated Structural Parameters for Different Sample Splits

Estimated            (8)        (9)      (10)
Parameters           Low       High       Low
                    Inst.      Share     Share
                   Holdings    Rep.      Rep.
                               Ratio     Ratio

[[lambda].sub.0]    0.570      0.730     0.405
                   (0.074)    (0.063)   (0.061)
[[lambda].sub.1]    0.073      0.081     0.055
                   (0.081)    (0.086)   (0.059)
[[lambda].sub.2]    0.0003    0.0004    0.0004
                   (0.290)    (0.193)   (0.153)
[gamma]             0.068      0.050     0.145
                   (0.057)    (0.071)   (0.040)
[delta]             0.105      0.112     0.118
                   (0.012)    (0.012)   (0.011)
q                   0.056      0.035     0.045
                   (0.071)    (0.094)   (0.075)
[phi]               0.424      0.215     0.416
                   (0.095)    (0.075)   (0.066)
a                   0.341      0.518     0.428
                   (0.072)    (0.047)   (0.043)
[rho]               0.665      0.639     0.774
                   (0.112)    (0.084)   (0.094)
[sigma]             0.281      0.305     0.163
                   (0.062)    (0.076)   (0.058)
[chi square]         7.88      9.12      8.08
                   (0.096)    (0.058)   (0.088)

Panel B. Equity Value Loss

Full                 Low       High       Low
Sample              Inst.      Share     Share
                   Holdings    Rep.      Rep.
                               Ratio     Ratio

0.061               1.029      1.020     1.086
COPYRIGHT 2013 Financial Management Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2013 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Mahmudi, Hamed; Pavlin, Michael
Publication:Financial Management
Geographic Code:1USA
Date:Dec 22, 2013
Words:17784
Previous Article:Lifting the veil on reverse leveraged buyouts: what happens during the private period?
Next Article:Do private placements turn around firms? Evidence from Taiwan.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters