Chapter 4 Dividend Smoothing, Financial Flexibility and Capital Structure
4.3 Data and methodology
We use data for all firms in both the CRSP and Compustat databases over the period 1986- 2008. We exclude financial firms (SIC codes 6000-6999) and utilities (4000-4999). For the analysis of corporate smoothing behaviour we require firms to have sufficient data to calculate smoothing metrics. Like Leary and Michaely (2011), we limit the sample to those firms that pay a dividend in at least ten of the years during our sample. Additionally, we reject firms that did not pay a dividend in at least three out of ten years. We do this to ensure that we select only firms with a history of reliable dividend payouts. We subsequently require firms to have information on the control variables used (discussed below). After applying all requirements, we have 517 firms from which we derive 5,159 firm-year observations (an average of approximately ten years of data per firm).
Measuring dividend smoothing
We use Linter’s (1956) model to calculate firm specific speed of adjustments (formula 3). By using rolling window regressions we are able to calculate speed of adjustments that change over time. The partial adjustment hypothesis, as introduced by Lintner (1956), holds that managers recognise the transitory nature of current earnings. Moreover Lintner shows that firms need a dividend discipline in which the changes in dividends are determined by prior dividend levels and current earnings. The firm’s desired level of dividends (ܦ݅ݒ௧כ) is given by:
ܦ݅ݒ௧כൌ ݎܧ௧ (1)
where ܧ௧ are the firm’s current earnings and ݎis the firm’s target payout ratio and can be expressed as a function of the firm’s investment and borrowing opportunities (Ang, 1975).50
Additionally, Lintner (1956) argues that when earnings increase, a firm will not adjust its dividends completely if there is uncertainty about the firms’ ability to keep the dividends at the higher level. Moreover, the partial adjustment hypothesis shows that firms are reluctant to cut the amount of dividends, whereby the changes in dividends will be only gradual. In light of this, Lintner’s
proposed partial adjustment process of the firm’s dividends is given by:
οܦ݅ݒ௧ൌ ߙ ߚଵܧ௧ ߚଶܦ݅ݒ௧ିଵ ߝ௧Ǣ݂݅ݎ݉ݏܽݐݐ݅݉݁ݐ (2)
Where the target payout ratio (TPR) is then given by ఉభ
ఉమ and the speed with which firms adjust their
dividends (SOA) is given by െߚଶ.51 Table I provides the definitions of the variables used in the
subsequent analysis.
Explaining cross-sectional variations in dividend smoothing
To test our hypotheses we explain cross-sectional variations in dividend smoothing for large and mature firms with a history of dividend payments using proxies for the firms’ financial flexibility and agency costs. Our first variable of interest is the firms unused debt capacity, as estimated by De Jong et al. (2011). The firms’ unused debt capacity captures the firms relative ability to increase the amount of on-balance sheet debt, without losing its investment grade rating. The second variable
50 Moreover, Ang (1975) lists investors’ preferences, marginal tax rates and transaction costs as potential determinants of the target payout ratio. This is, however, beyond the scope of our argument.
51 Note that SOAi,t is an inverse metric; put differently, firms with stable dividends have low adjustment speeds. Since SOAi,t is estimated for each firm separately we require each firm to have at least ten consecutive observations to allow for a robust rolling window estimation of the dividend adjustment speed at time t. The mid-points of these estimations are then matched with control variables and variables of interest at that point in time.
of interest is the firms’ capital structure adjustment speed, as estimated by Fama and French (2002) and Flannery and Rangan (2006). The firms’ capital structure adjustment includes the relative costs against which firms can quickly adjust their capital structure. Third we calculate the WW- index, as estimated by Whited and Wu (2006), to capture agency costs induced by underinvestment. Fourth, we calculate the extent to which firms experience a shock to net income (scaled by total assets). We define a shock as change of the firms’ net income that is economically relevant to the firms’ dividend payment. Since the firms in our sample pay out approximately 2% of their asset value each period, we define a shock to net income of more than 2% to be economically relevant to the firms’ dividend policy.
To explain corporate smoothing behaviour we then estimate the following base equation:
ܱܵܣǡ௧ൌ ߙ ߚଵܥǡ௧ ߚଶܫǡ௧ ߝǡ௧ (3)
ܱܵܣǡ௧ is the speed of a firm’s dividend adjustment obtained from Equation 2. ܥǡ௧ is a
vector of the firm specific control variables, which include leverage, m/b, firm size, profitability, tangibility, cash and earnings volatility. ܫǡ௧ are the variables of interest, including unused debt capacity and capital structure adjustment speed and the WW-index (appendix A and B on the specifications). We specify our models as Feasible Least Squares estimations, with an identity link function (ܺߚ ൌ ߤ). We correct for a heteroskedastic error structure with no cross-sectional correlation. Because our dependent variable is estimated using a rolling window regression, we employ a panel-specific AR1 autocorrelation structure (similar to Byoun (2008)), where the autocorrelation parameter is specified by the Durbin-Watson statistic (݀ ൌ൫σసమሺఌିఌషభ൯
మ σ ఌమ
సమ ).
52 Since
SOA is an estimated variable we weigh the dependent variable by the inverse of its estimation
error, which allows us to put more weight on more accurately measured speed of adjustments. All models include time, industry and age decile fixed effect to ensure our findings are not driven by an omitted variable.53