Dividend behaviour of US equity REITs

Abstract

A new era began for US equity Real Estate Investment Trusts (REITs) around 1992. This study is the first to document the dividend smoothing, determinants of dividend payouts, and the market reaction to dividend announcements of the modern REIT. We find the Lintner (1956) smoothing parameter suggests a high level of dividend smoothing in quarterly data and, although we observe substantially less smoothing in the annual data, modern REITs smooth dividends at least as much as their industrial counterparts. Beyond the dividend paid last period and contemporaneous earnings, usual determinants of dividend behaviour in non‐REIT firms are not economically significant, which is a departure from old‐era REITs. Additionally, as compared with non‐REIT firms, we find muted excess stock returns upon announcements of dividend changes. Overall, our findings suggest that managers set a stable (smooth) dividend policy even in the relative absence of market imperfections.

Keywords:

• payout policy
• dividend smoothing
• REITs

Introduction

In this paper, we examine the dividend behaviour of modern equity real estate investment trusts (REITs) in the United States. Specifically, we document the cycle of dividend payments, the market reaction to the amount and dividend smoothing. The usual explanations for a firm’s strategic dividend policy and the market reaction are market imperfections. The transparent nature and limited investment set of REITs mitigates the effect of market imperfections, making REITs a unique experimental setting to examine dividend behaviour.

At least since Lintner’s (1956) seminal work, the literature has found that industrial firms are reluctant to cut dividends and that they only gradually increase dividends with increases in permanent earnings. Consistent with this notion, the market greets dividend increases positively, while dividend decreases are punished severely. Denis, Denis and Sarin (1994) demonstrate that the reactions to dividend increases and dividend decreases are asymmetric: dividend increases are associated with a small positive abnormal returns (approximately 1%) while dividend decreases result in large negative abnormal returns (about –5%). Clearly, firms have an incentive to smooth dividends. More recently, Brav, Graham, Harvey and Michaely (2005) survey the financial officers of US industrial firms and confirm this incentive to smooth dividends. They find that ‘maintaining a smooth dividend stream’ is an important consideration in determining payout policy and firms only slowly adjust to deviations from their target payout ratio caused by changes in income – essentially, firms do smooth dividends.

Although there are many potential reasons firms might smooth dividends, each is the result of market imperfections. Allen and Michaely (2003) detail five market imperfections that make dividend policy relevant. These imperfections are summarised as taxes, asymmetric information, incomplete contracts (agency conflicts), institutional constraints and transaction costs. Since US REITs are relatively transparent compared with industrial firms, the effects of asymmetric information and other market imperfections are largely mitigated (see Appendix for a detailed discussion of the application of the market imperfections to REITs). Consequently, REITs dividends should be relatively uninformative and we would expect little market reaction to changes in dividends. Absent the market reaction, there is little incentive for REITs to smooth their dividend payouts.

In addition to being relatively transparent, US Internal Revenue Service (IRS) statutes allow REITs to exclude dividend payments from taxable income provided 90% of their taxable earnings are distributed to shareholders in the form of dividends. ¹ Consequently, regulatory requirements potentially constrain a REIT’s ability to engage in any strategic dividend behaviour. Thus, we might expect REITs to have a residual dividend policy – essentially a static payout ratio of 90% or slightly greater.

Conversely, US REITs are known to be a cash‐flow business and, due to depreciation and other non‐cash charges, REIT cash flows are generally in excess of their taxable earnings. Prior studies by Wang, Erickson and Gau (1993), Kallberg, Liu and Srinivasan (2003), and Ghosh and Sirmans (2006) report average dividend payments by US REITs well in excess of earnings. This fact suggests that the IRS dividend requirement is manageable for most US REITs. Whether meeting the requirement is burdensome or not, it presents a further complication that hinders a US REIT in following a prescribed strategic dividend policy.

The mitigation of market imperfections along with the high dividend‐payout requirement provide meaningful reasons for REITs to forego a systematic dividend policy. Thus, results of REIT dividend payouts should be biased in favour of finding: (1) insignificant explanatory variables in models of dividend payments due to transparency; (2) no market reaction to dividend announcements, either positive or negative; and (3) a lack of dividend smoothing. Explicit examination of the relation of these three aspects of dividend behaviour is our first contribution to the REIT literature.

Of course we are not the first to note that REIT dividend behaviour may be unique to the organisational form and therefore an interesting experimental environment. However, the extant literature examines REIT dividend determinants prior to a structural break that occurred in the US REIT industry around 1992. Hence, our study’s second contribution to the REIT literature is the first investigation of dividend behaviour of the modern equity REIT.

The change in US REITs was facilitated by two events. First, a US tax code revision in 1986 allowed REITs to own and manage properties, and the subsequent offering of Kimco Realty Corporation in 1991 under the new tax code facilitated vertically‐integrated REITs. Second, the 1992 offering of Taubman Properties launched the public Umbrella Partnership (UPREIT) structure. The UPREIT structure provides property investors with a tax‐protected method for transferring ownership out of relatively illiquid real estate. Taken together, these two changes effectively launched a new era for US REITs. Capozza and Seguin (2003), Ling and Naranjo (2003), and Ott, Riddiough, and Yi (2005) note some of the differences between the old‐ and new‐era REITs.

Our last contribution is to the broader dividend‐behaviour literature. To the extent that our results can be applied in a general context, empirical findings with respect to REIT dividend smoothing can provide additional insight since REITs help disentangle competing rationales in industrial firms.

Our results can be summarised as follows. First, we observe that the Lintner (1956) smoothing parameter suggests a high level of smoothing in the quarterly dividend specifically set by firm management. We find less smoothing in annual dividends – computed smoothing parameters are similar to those observed in US industrial firms. Second, consistent with mitigation of market imperfections, we find determinants and proxies of dividend behaviour in non‐REIT firms are generally not significant predictors of modern REIT dividend behaviour. These findings suggest a departure by modern REITs when compared with old‐era REITs. Third, the results demonstrate that generally accepted accounting principles (GAAP) net income and funds from operations (FFO) are both significant in explaining dividend payments, individually and in combination. They are not multicollinear regressors. Last, we find the market’s reaction to dividend changes is considerably less than in industrial firms.

Collectively, these findings can provide additional insight into the broader dividend behaviour of corporations, which we discuss in the concluding remarks of this paper. Prior to that discussion, though, we review the extant literature in the next section and follow with explanation of the data and empirical design in the third section.

Extant literature

In his seminal work, Lintner (1956) showed that US dividends are sticky. ² Not only are US firms reluctant to cut dividends or, perhaps because firms are reluctant to cut dividends, firms only gradually increase dividends when earnings increase. Although firms may have a target payout ratio, they deviate from this target and only partially adjust dividend payments to new higher levels of income over time. In what is regarded as the first direct test of Lintner’s (1956) partial‐adjustment model, Fama and Babiak (1968) show that the restricted model does a good job of explaining dividend behaviour. Recently, Brav et al. (2005) confirm the importance of dividend smoothing in a survey of financial officers in 384 US firms. Of the executives, 89.6% report that ‘maintaining a smooth dividend stream’ is an important consideration in determining payout policy. Consistent with Lintner (1956), they also report that firms have a relatively low target payout ratio (0.22) and only slowly adjust to deviations from this target ratio.

Since the time of Lintner’s study, financial economists have conducted an abundance of dividend studies and debated the relevance of dividend policy and the economic impact of dividends paid in less than perfect markets. ³ Beginning with Fama, Fisher, Jensen and Roll (1969), numerous researchers have also examined the market reaction to changes in dividends as well as dividend initiations and omissions (see Pettit, 1972 and Watts, 1973). Overall, the evidence is overwhelming that dividend changes by industrial firms convey information to the market over‐and‐above that provided by earnings.

But does the conveyance of information hold true for REITs? In much of the dividend policy research in the finance literature, REITs are specifically excluded. The rationale often given is that financial firms have different capital structures. Presumably another reason more specific to US REITs is due to the 90% payout requirement and its potential to constraint a management’s ability to manage a specific dividend policy. For instance, Baker and Smith (2006) exclude REITs ‘because REITs have little discretion in setting dividend policy’.

There are four existing studies of REIT dividend behaviour. Lee and Kau (1987) are the first to examine REIT dividend payments and smoothing. They report general smoothing in 29 US REITs from 1971 to 1981. Wang et al. (1993) study the determinants of US REIT dividend policy as well as dividend announcement effects. They examine the correlation of return‐on‐assets, growth, Tobin’s Q, and leverage on the natural logarithm of dividends scaled by net income. Their results indicate a lack of one consistent determinant across the years from 1985 to 1988. Bradley, Capozza and Seguin (1998) extend the sample period from 1985 to 1992. They examine dividends per share as a function of lagged FFO, the change in FFO, leverage, market value of assets, a fitted FFO volatility measure, and Herfindahl measures for the region and property‐type. They find that leverage and market assets are significant determinants. In a recent study that examines the modern REIT period from 1999 to 2005, Hardin and Hill (2008) take a different approach by modelling the excess dividend payment above the IRS requirement. They find a stock repurchase plan and the access to short‐term bank debt are determinants of the excess dividend payments.

Beyond these US studies, there does not appear to be any papers specifically investigating dividend behaviour in non‐US REITs. This is due in large part to many countries recently opening up regulation and markets to REIT‐type structures; the exception is Australia, which launched the REIT concept or Listed Property Trusts (LPT) in 1971. Canadian REITs began in 1993. Asian REITs emerged from 2000 to 2002 in Japan, Singapore, South Korea and Taiwan markets. Hong Kong established its REITs rules in 2003 with the first REIT IPO by Link REIT in late 2005. Israel established REIT regulation in the beginning of 2006. The United Kingdom, Germany and Italy introduced the REIT structure in 2007.

Although they do not specifically examine dividend payouts, there exists additional literature using REIT dividends to investigate other financial issues. Kallberg et al. (2003) examine dividend pricing models on a REIT index. Ghosh and Sirmans (2006) examine US REIT dividend payout and yield with respect to CEO, board and financial characteristics. They find that dividend payout and yield are functions of corporate performance, board structure, CEO tenure and CEO ownership of company shares. Li, Sun and Ong (2006) examine the relation between dividends and splits by investigating REITs’ dividend streams around stock split announcements. They contend that stock splits by US REITs provide useful information about future dividend changes. Dimovski (2006) examines Australian LPT IPOs for a bias in the dividend forecasts and finds evidence that the projections are not optimistically biased.

Overall, there is not an overwhelming number of studies regarding REIT dividend behaviour, especially when compared with the hundreds of studies in the finance literature and the importance of the high‐dividend payout by REITs to income investors. We endeavour to address the need with the empirical design detailed in the next section and empirical results in the subsequent section.

Sample description, the partial adjustment model and the empirical design

Sample description

Due to the structural break in US REITs, we use a sample period from 1992 to 2003. We collect all dividend‐paying US REITs listed in the Center for Research in Securities Prices (CRSP) and COMPUSTAT databases. For quarterly data, we investigate only regular dividends (CRSP dividend distribution code 1232) and excluding any special or one‐time dividends. By excluding special dividends, we focus solely on the routine dividend policy set by firm management. In both the annual and quarterly data, we remove hybrid and mortgage REITs as well as captive and liquidating trusts. ⁴

The partial adjustment model

We begin our analysis by applying the Lintner (1956) partial‐adjustment model to both the annual and quarterly data samples. Lintner (1956) finds that managers believe that shareholders prefer a steady stream of dividends; thus, firms tend to make periodic partial adjustments toward a target payout ratio rather than dramatic changes. This relation is initially reflected as:

(1)

where D^a _i,t is the target dividends for firm i in year t and r _i is the firm’s target ratio of dividends to after‐tax earnings. According to the model, firms do not move immediately to a new target dividend, but instead smooth out changes in their dividends by moving part of the way to the target dividend each period. The smoothing behaviour leads to the following relation:

(2)

where a _i is a constant, 0 < c < 1 is a speed‐of‐adjustment (SOA) coefficient, and µ _i,t is a normally distributed random error term. Substituting Equation (1) into Equation (2) yields:

(3)

The estimating equation in a multivariate regression model is then:

(4)

where β _1,i = 1 − c and β _2,i = c _i r _i. The SOA parameter provides a dividend smoothing measure because it is an indication of the level of the change in dividends that occurs in the short run in response to a change in the level of earnings. If a firm pays the same or slightly increasing dividend from period to the next period c will trend toward zero. Another output from the partial adjustment model is a measure of a firm’s target payout. We compute and report this measure when we examine the partial‐adjustment model results.

Empirical design

The Lintner (1956) model does not account for other determinants of dividend policy beyond lagged dividends and earnings. Over the past five decades a number of theories and hypotheses have been proposed and examined as further explanations of dividend payouts. Our approach is to control for these other determinants detailed in the finance and economic literature. As we discuss in the Appendix, the transparent nature of REITs mitigates the importance of asymmetric information and other market imperfections. Consequently, we expect that many, if not most, of covariates will be statistically and/or economically unimportant, however, this is an empirical question that has not been tested.

We model contemporaneous dividend payments as a function of the following variables of interest using a panel of data for every i firm at time t.

• Dividends _i,t = ß ₀ + ß _1· Dividends _i,t‐1 + ß ₂ Earnings _i,t + ß _3· Growth _i,t + ß ₄ Tobin’s Q _i,t + ß _5· MTB _i,t + ß _6· Slack _i,t + ß _7· FCF _i,t + ß _8· Leverage _i,t + ß _9· Ln(Assets) _i,t + ß _10· Ln(Shareholders) _i,t + ß _11· Asset turnover _i,t + ß _12· Change in debt _i,t + ß _13· Liquidity ratio _i,t + ß _14· Tax code change _i,t + ε _i,t

• Dividends _i,t = dividends levels scaled by total assets for annual data and DPS or dividends per share for quarterly data

• Earnings _i,t = GAAP net income scaled by total assets and FFO scaled by total assets for annual data and EPS or earnings per share for the quarterly data

• Growth _i,t = (total assets_t/total assets_t‐1)‐1∗100

• Tobin’s Q _i,t = Chung–Pruitt’s (1994) measure of Tobin’s Q, a common growth proxy

• MTB _i,t = market value of equity scaled by book value of equity

• Slack _i,t = cash scaled by total assets

• FCF _i,t = Lehn and Poulsen (1989) definition of free cash flow

• Ln(Assets) _i,t = natural logarithm of total assets

• Leverage _i,t = total liabilities divided by total book assets

• Ln(Shareholders) _i,t = natural logarithm of the number of shareholders

• Asset turnover _i,t = sales divided by total assets

• Change in debt _i,t = long‐term debt_t ‐ long‐term debt_t‐1

• Liquidity ratio _i,t = ratio of current assets to current liabilities

• Tax code change _i,t = dichotomous variable with value equal to 0 if the observation year is before the tax code change in 2001 or equal to 1 if the year is post 2000.

We use two measures of earnings – FFO and GAAP net income. Clearly, the 90% IRS payout requirement should cause a high correlation between taxable income and dividend payments. However, taxable income is typically not reported publicly, therefore, we use FFO and GAAP net income to measure the unobservable dividend‐to‐taxable income relation. GAAP net income is not the same as taxable income, thus, there is not a tautological relation. The fact that REITs report GAAP net income quarterly facilitates additional analysis beyond annual results.

The other measure of earnings is FFO. Since we are examining the more common equity REIT, the sample firms will realise depreciation, which reduces reported GAAP net income. Depreciation is not an actual cash flow, thus net income is not necessarily an accurate measure of the net cash flow available to distribute to investors as dividends. Consequently, US REITs compute FFO as a supplemental earnings measure that adds back depreciation and amortisation expenses to GAAP net income. We initially examine FFO instead of GAAP net income. Subsequently, we examine both earnings measures in combination, which has not been examined in prior dividend studies.

We trim all of the annual variables at the 1% level as we find that extreme outliers affect the results. Table 1 provides the descriptive statistics for our sample. We report annual values in column 1 and quarterly values in column 2. Both the mean and median values of Dividends and DPS are greater than the mean and median values of our primary earnings measures, Net income and EPS, which suggests payout ratios greater than 100%. We do not model Payout ratio (i.e. dividends divided by net income) in the multivariate analysis but include it in our summary statistics for reference. The mean payout ratio in our sample is 1.78 with a median of 1.17; if preferred dividends are included, which can be used to meet the IRS dividend requirement, the mean and median payout ratios increase to 1.98 and 1.26, respectively. The vast majority of REITs in our sample payout considerably more than the IRS requirement and thus provides the flexibility in their dividend policy.

Table 1. Descriptive statistics.

	Annual	Quarterly
Variable	Mean (median)	Obs Std dev.	Mean (median)	Obs Std dev.
Dividends	0.040 (0.038)	1050 0.021
Net income	0.031 (0.032)	1050 0.025
FFO	0.640 (0.611)	756 0.262
DPS			0.382 (0.389)	3328 (0.140)
EPS			0.344 (0.323)	3328 (0.364)
Payout ratio – common dividends only	1.798 (1.167)	960 3.674
Payout ratio – common and preferred dividends	2.002 (1.264)	960 3.962
Growth	22.200 (11.054)	1050 55.513	5.415 (1.010)	3328 24.360
Tobin’s Q	1.040 (1.022)	1050 0.249
MTB	1.497 (1.375)	1050 0.763	1.356 (1.439)	3328 5.609
Slack	0.022 (0.012)	1050 0.032	0.024 (0.012)	3328 0.049
FCF	0.016 (0.017)	1050 0.036
Leverage	0.559 (0.568)	1050 0.186	0.568 (0.581)	3328 0.203
Ln(Assets)	6.332 (6.482)	1050 1.293	6.632 (6.665)	3328 1.254
Ln(Shareholders)	0.648 (0.362)	1050 1.667
Asset turnover	0.149 (0.144)	1050 0.047	0.041 (0.039)	3328 0.012
Change in debt	27.090 24.993	1050 288.621	30.064 (7.021)	3328 157.920
Liquidity ratio	2.889 (0.723)	1050 9.893
Tax code change	0.691 (1.000)	1050 0.462	0.662 (1.000)	3328 0.473
The table reports summary statistics for a sample of US equity REITs from 1992 to 2003. The definitions of the variables are as follows: Dividends = annual measure of dividend payments scaled by total assets; Net income = annual measure of GAAP net income scaled by total assets; FFO = funds from operations scaled by total assets (an annual measure); EPS = quarterly measure of GAAP net income scaled by total shares outstanding; DPS = quarterly measure of dividends per share; Payout ratio = annual measure of common dividends and common + preferred dividends scaled by net income; Tobin’s Q = annual measure of Chung–Pruitt’s (1994) Tobin’s Q; Slack = cash scaled by total assets; Asset turnover = sales divided by total assets; Ln(Assets) = natural logarithm of total assets; Leverage = total liabilities divided by total book assets; Liquidity ratio = current assets scaled by current liabilities; Growth = change in assets scaled by previous year total assets; Change in debt = long‐term debt current year minus long‐term debt prior year; MTB = market value of equity scaled by book value of equity; FCF = Lehn and Poulsen (1989) measure of free cash flow scaled by total assets; Ln(Shareholders) = the natural logarithm of the number of shareholders (an annual measure); and Tax code change = dummy variable where 0 is pre‐year 2001 and 1 is post‐year 2000. Payout ratios withhold 90 net‐loss observations. Annual variables trimmed at the 1% level.

Regarding the other covariates, the mean and median values of Tobin’s Q are approximately 1.00 suggesting these REITs are neither distressed nor high growth firms. The MTB ratios, however, are above 1.0; likely a result of both asset appreciation increasing market values and accumulated depreciation reducing book values. Finally, the REITs are fairly highly levered compared to their industrial counterparts – the average leverage ratio is 0.56.

In our multivariate analysis, we lag all regressors one period with the exception of Net income and FFO since we are interested in their contemporaneous relation with dividends. When using least squares models, the assumption is made that observations on the independent variables are fixed in repeated samples. This may be a concern when applied to dividend payout models since modelling contemporaneous independent variables against the dependent variable of dividend payouts can cause the simultaneous determination of dividend policy with a number of the explanatory variables. For example, firms may take on more debt to maintain a consistent dividend policy. Due to the endogeneity of dividend policy, the assumption that the explanatory variables and the error term are independent may not hold. We use instrumental variables estimation by lagging each variable one period – each instrumental variable is correlated with the contemporaneous determinant, but not with the contemporaneous error term.

In addition to the multivariate models, we examine the impact of announcements of dividend changes. We employ standard event study methods to examine the effects. We measure the cumulative abnormal returns (CARs) for the three‐day period beginning the day prior to the announcement and ending the day subsequent to the announcement. We measure the abnormal returns using both simple market adjustments subtracting the market proxy from the firm return on any given day and the market model residuals. The abnormal returns are then geometrically cumulated over the three‐day period to obtain CAR estimates.

Equity REITs, in particular, are subject to non‐synchronous trading potentially resulting in serial autocorrelation in abnormal returns. Therefore, standard parametric tests may result in incorrect inferences. Seiler (2000) specifically studies the non‐synchronous trading issue along with event‐induced variance in equity REITs to determine which event study methods are the best. He finds that Corrado’s (1989) Nonparametric Rank Test and a hybrid of Patell’s (1976) standardised residual test and Charest’s (1978) cross‐sectional test, which is termed the standardised cross‐sectional test, consistently outperform four other procedures. Consequently, we report these tests.

Empirical results

Quarterly dividends

Using the Lintner (1956) partial‐adjustment model, column 1 of Table 2 reports the OLS estimates and Column 3 details Tobit parameters. The results of both models demonstrate a strong correlation between the contemporaneous and lagged dividends paid per share. For every 1 unit change in current dividends, the dividends paid last quarter account for 97.2% (OLS) and 96.0% (Tobit). The coefficient on contemporaneous earnings is about 0.015 using either specification. Given these coefficients, the SOA is 0.028 using an OLS specification and 0.040 using a Tobit model. These SOA parameters imply that US equity REITs are quite slow to adjust to changes in earnings; instead basing their contemporaneous dividend payout almost entirely on the dividend paid last quarter.

Table 2. Quarterly dividends – Linter (1956) partial adjustment model.

Variable	OLS	Tobit
Intercept	0.009∗∗∗ (0.002)	0.013∗∗∗ (0.002)
Lagged DPS	0.972∗∗∗ (0.004)	0.960∗∗∗ (0.005)
EPS	0.015∗∗∗ (0.002)	0.014∗∗∗ (0.002)
Adjusted R^2	0.936
Log‐likelihood		6265
Speed of adjustment	0.028	0.040
Target payout ratio	0.533	0.357
Observations	3328	3328
The table reports the determinants of quarterly dividends payments using the Lintner (1956) partial adjustment model. The independent variable is the routine quarterly dividend (CRSP dividend distribution code 1232) made by a sample of US equity REITs from 1992 to 2003. Standard errors are in parentheses. ∗∗∗ indicates statistical significance at the 1% level.

When we expand the Lintner (1956) model to include other predictors of REIT dividend behaviour, the results in Table 3 are largely consistent with the findings in the restricted models. Using an OLS model that controls for serial correlation, we observe that the coefficient on Lagged DPS is 0.971 in column 1. The marginal slope on EPS is 0.014. Combined, these coefficients result in a SOA at the far end of the spectrum of 0.029. Similarly, the coefficient of 0.953 and 0.013 on Lagged DPS and EPS of the Tobit model compute a SOA of 0.047. ⁵ Overall, the results of both the restricted and unrestricted dividend models yield SOA parameters that imply a remarkable amount of smoothing in quarterly dividends.

Table 3. Quarterly dividends – unrestricted model.

Variable	OLS	Tobit
Intercept	−0.002 (0.004)	−0.005 (0.005)
Lagged DPS	0.971∗∗∗ (0.004)	0.953∗∗∗ (0.005)
EPS	0.014∗∗∗ (0.002)	0.013∗∗∗ (0.002)
Growth (∗10^−5)	0.017∗∗∗ (0.004)
MTB (∗10^−2)	0.017∗∗∗ (0.001)	0.000 (0.000)
Slack	0.005 (0.012)	−0.008 (0.013)
Leverage	0.000 (0.002)	−0.001 (0.003)
Ln(Assets)	0.001∗∗ (0.001)	0.003∗∗∗ (0.001)
Asset turnover	−0.003 (0.048)	0.019 (0.055)
Change in debt (∗10^−3)	−0.019∗∗∗ (0.007)
Tax code change	0.005∗∗∗ (0.001)	0.006∗∗∗ (0.001)
Adjusted R^2	0.937
Log‐likelihood		6280
Speed of adjustment	0.029	0.047
Observations	3328	3328
The table reports the predictor variables of quarterly dividends payments by a sample of US equity REITs from 1992 to 2003. The independent variable is the routine quarterly dividend (CRSP dividend distribution code 1232) scaled by shares outstanding. Standard errors are in parentheses. ∗∗∗ and ∗∗ indicate statistical significance at the 1 and 5% levels, respectively.

While the results thus far are quarterly, we can still compare them with the following annual studies to gauge the degree of smoothing. Lee and Kau (1987) report a SOA for REITs from 1971 to 1981 of 0.24. Lintner’s original study using industrial firms finds a SOA coefficient of 0.30. Other dividend smoothing studies of non‐REIT firms include Fama and Babiak (1968), who find an average SOA of 0.37, Fama and French (2002), who report a parameter of 0.33, and Brav et al. (2005), who find a mean of 0.33 for all Compustat firms with valid data from 1984–2002.

Some of the other determinants of REIT dividend behaviour in Table 3 are statistically significant but they do not exhibit economic importance. Parameter estimates on Ln(Assets), MTB, Change in debt, Growth, and Tax code change have a p‐value less than 0.01; however, the coefficient values are 0.006 or less. The significance of these quarterly factors is a function of reduced standard errors, which seems to be the result of the impact of the dividend paid last quarter and the strength of the model – adjusted R² of 0.937. Alternatively, the intercepts on the unrestricted models are not significant, which implies the covariates beyond net income and lagged dividends are somewhat influential.

Annual dividends

Using the same empirical method as with the quarterly payments, we next apply the restricted and unrestricted models to the annual dividend data. We report the results of the Lintner (1956) restricted model in Table 4. Again, we observe that the level of dividends paid last year along with contemporaneous net income explain a considerable portion of the variability in REIT dividend payments. The adjusted R² using OLS is 0.695 and each marginal slope is significant with p‐values less than 0.001.

Table 4. Annual dividend payments – Lintner (1956) partial adjustment model.

Variable	OLS	Tobit
Intercept	0.007∗∗∗ (0.001)	0.006∗∗∗ (0.001)
Lagged dividends	0.621∗∗∗ (0.017)	0.629∗∗∗ (0.191)
Net income	0.245∗∗∗ (0.017)	0.264∗∗∗ (0.017)
Adjusted R^2	0.695
Log‐likelihood		3037
Speed of adjustment	0.379	0.371
Target payout ratio	0.646	0.711
Observations	1050	1050
The table reports the application of the Lintner (1956) partial adjustment model to a sample of US equity REITs for the sample period from 1992–2003. The dependent variable is the annual level of dividend payments scaled by total book assets. Standard errors are in parentheses. ∗∗∗ indicates statistical significance at the 1% level.

In contrast with the quarterly results though, the coefficient estimates are different using annual dividend payments. Lagged dividends decreases from a value around 0.95 with quarterly data to approximately 0.62 using annual data. Net income increases from about 0.01 using quarterly data to annual estimates of 0.245 (OLS) and 0.264 (Tobit). Given these coefficients, the SOA parameter computes to approximately 0.37 using either model, which is largely consistent with prior empirical findings on non‐REIT US firms.

We next expand the partial‐adjustment model to an unrestricted specification and report the results in Table 5. The OLS coefficients on Lagged dividends and Net income are 0.592 and 0.230, respectively. The Tobit coefficients are 0.608 for Lagged dividends and 0.218 for Net income. The SOA given these coefficients is about 0.40. Consistent with the quarterly unrestricted findings and mitigation of market imperfections, we observe that the other control variables are not economically significant with the slight exception of Asset turnover. The coefficient estimate is about −0.02 using either OLS or Tobit.

Table 5. Annual dividend payments – unrestricted model.

Variable	OLS	Tobit
Intercept	0.011∗∗∗ (0.004)	0.017∗∗∗ (0.004)
Lagged dividends	0.592∗∗∗ (0.022)	0.608∗∗∗ (0.022)
Net income	0.230∗∗∗ (0.019)	0.218∗∗∗ (0.018)
Growth	−0.000 (0.000)
Tobin’s Q	0.002 (0.002)	0.001 (0.002)
MTB	0.001 (0.001)	0.001 (0.001)
Slack	−0.014 (0.012)	−0.021∗ (0.013)
FCF	−0.014 (0.011)	0.013 (0.012)
Leverage	−0.001 (0.003)	−0.004 (0.003)
Ln(Assets) (∗10^−2)	−0.036 (0.033)	−0.101∗∗∗ (0.032)
Ln(Shareholders)	0.000 (0.000)	0.000 (0.000)
Asset turnover	−0.019∗∗ (0.009)	−0.018∗∗ (0.008)
Change in debt ∗(10^−3)	0.003∗∗ (0.002)
Liquidity ratio (∗10^−2)	0.009∗∗ (0.004)	0.005 (0.004)
Tax code change ∗(10^−2)	−0.014 (0.086)	−0.000 (0.001)
Adjusted R^2	0.700
Log‐likelihood		3124
Speed of adjustment	0.408	0.392
Observations	1050	1050
The table reports the determinants of dividend payments by US equity REITs for the period from 1992 to 2003. The regressand is the level of annual dividend payments scaled by total book assets. Standard errors are in parentheses. ∗∗∗ and ∗∗ denote statistical significance at the 1 and 5% levels, respectively.

The other statistically significant variable that does not exhibit a high magnitude of economic affect is the intercept. We note that, unlike the quarterly data, the addition of the other control variables does not remove the systematic effect in the intercept. One possible explanation is an omitted variable.

Based upon Bradley et al. (1998), we investigate FFO as the omitted variable. Table 6 reports three specifications that include FFO. In Model 1, we examine FFO in the unrestricted specification but without Net income as a covariate. Model 2 includes both FFO and Net income. We initially question whether both earnings measures will display multicollinearity in a multivariate model but we find that all variance inflation factors using OLS in Model 2 are less than 2.0. The last model in Table 6 reports the findings of splitting contemporaneous FFO into the level of FFO last period and the Change in FFO from last year to this year, which is consistent with Bradley et al. (1998). We use the Tobit specification for each of these models due to the truncation of the dependent variable and the lack of need to correct for serial correlation using OLS in the annual data. In every specification in Table 6, FFO is significant.

Table 6. Annual dividend payments – unrestricted net income and FFO model.

Variable	Model 1	Model 2	Model 3
Intercept	0.007∗∗ (0.003)	0.008 ∗∗ (0.003)	0.005 (0.003)
Lagged dividends	0.499∗∗∗ (0.026)	0.484∗∗∗ (0.025)	0.624∗∗∗ (0.026)
Net income		0.131∗∗∗ (0.022)	0.134∗∗∗ (0.021)
FFO	0.030∗∗∗ (0.002)	0.023∗∗∗ (0.002)
Lagged FFO			0.013∗∗∗ (0.002)
Change in FFO			0.031∗∗∗ (0.003)
Tobin’s Q	0.006∗∗∗ (0.002)	0.004∗∗∗ (0.002)	0.003∗∗ (0.002)
MTB (∗10^−2)	−0.008∗∗ (0.004)	−0.002 (0.004)	0.006 (0.004)
Slack	−0.001 (0.010)	0.006 (0.010)	0.001 (0.010)
Leverage	−0.003 (0.002)	−0.001 (0.002)	0.001 (0.002)
Ln(Assets) (∗10^−2)	−0.061∗ (0.033)	−0.080∗∗ (0.032)	−0.047 (0.030)
Ln(Shareholders) (∗10^−2)	0.041∗∗ (0.019)	0.034∗ (0.019)	0.032∗∗ (0.018)
Asset turnover (∗10^−2)	−0.043∗∗∗ (0.007)	−3.671 ∗∗∗ (0.743)	−1.772∗∗ (0.718)
Liquidity ratio (∗10^−2)	0.002∗∗∗ (0.001)	0.002∗∗ (0.001)	0.002∗∗∗ (0.001)
Tax code change ∗(10^−2)	−0.060 (0.071)	−0.054 (0.070)	−0.059 (0.066)
Log‐likelihood	2377	2394	2358
Speed of adjustment		0.516	0.376
Observations	756	756	716
The table reports the determinants of dividend payments made by a sample of US equity REIT from 1992 to 2003. The dependent variable is the annual level of dividends scaled by total book assets. All models use the Tobit specification due to truncation of the dependent variable. Standard errors are in parentheses. ∗∗∗, ∗∗ and ∗ denote statistical significance at the 1, 5 and 10% levels, respectively.

In Model 1 we observe that using FFO partially reduces the impact of Lagged dividends. Comparing the value of Lagged dividends in Model 1 to the unrestricted Tobit in Table 5, the coefficient decreases from 0.608 to 0.499. We also observe that FFO does not explain as much of the variability of dividend payments as Net income in previous models. The Net income coefficient estimate is approximately 0.22 in Table 5. FFO in Model 1 of Table 6 is 0.03.

The slope coefficients on the other determinants in Model 1 are consistent with prior results. There are some instances of statistical significance but the greatest economic impact is Tobin’s Q with a value of 0.006. The general lack of economic significance is also true of the control variables in the other two models using FFO. Some covariates exhibit p‐values less than 0.05 but the accompanying marginal slope coefficients are less than 0.01.

Model 2 combines Net income and FFO. The findings suggest that neither earnings measure subsumes the other. The coefficient on Net income is a significant 0.131, which is a reduction from 0.218 in Table 5. The importance of FFO does not change considerably from Model 1; the value is 0.023 when we introduce Net income versus 0.030 without Net income. Finally, the marginal slope on the level of dividends paid last year is 0.484.

Model 3 combines Net income with Lagged FFO and the Change in FFO. The findings in last columns of Table 6 demonstrate that Lagged FFO and the Change in FFO are significant determinants of the dividend behaviour of the modern US equity REIT. Indeed, the coefficient on the level of FFO last year is as significant as contemporaneous FFO in Model 1. The other revenue measure of Net income maintains its significance to dividend payouts with a coefficient of 0.134. The slope on Lagged dividend increases in Model 3 to similar levels as those we find in Tables 4 and 5; a coefficient of 0.624. Given that Net income is included in the specification, we can compute the SOA as 0.376 using a more complete model that includes FFO. Note also that the Intercept is insignificant in Model 3.

In summary, the empirical results thus far suggest that US REITs display at least as much dividend smoothing as US industrial firms. The parameter observed in the quarterly data suggests a high degree of smoothing. Given that the empirical evidence demonstrates confirmation of the hypothesis that REITs largely mitigate the usual market imperfections that affect dividend policy in non‐REIT firms, we wonder why REITs smooth at least as much as non‐REIT firms. Therefore, we next examine the market reaction to dividend increases and decreases by REITs to determine if this provides an incentive for REITs to smooth dividends.

Announcements of dividend changes

Wang, et al. (1993) previously examine US REIT announcements of dividend changes; however, (i) the sample period is 1985–1988 and (ii) the number of dividend decreases is 23 and 29, depending upon the model. The low number of observations may impact the power of the tests. To increase statistical power and to coordinate with the modern REIT period of our sample, we conduct an event study of equity‐REIT dividend announcements over the sample period from 1992 to 2003.

Panel A in Table 7 details the findings of the stock return for announcements of dividend increases for equity REITs. Overall, the results show that the mean CAR for a dividend increase is statistically significant at the 1% level. The mean CAR for equity REITs of approximately 0.45% is lower than the Wang et al. (1993) estimates of 0.66 and 0.80%. One justification for this reduction is further transparency in the modern REIT.

Table 7. Stock price reactions to the announcement of a change in dividends.

	N	Mean CAR %	Standardised cross‐sectional test	Non‐parametric rank test
Panel A: Dividend increase announcements
EW‐market model	742	0.47	5.25∗∗∗	3.74∗∗∗
VW‐market model	742	0.45	4.51∗∗∗	3.51∗∗∗
EW‐market adjusted	742	0.43	3.90∗∗∗	3.38∗∗∗
VW‐market adjusted	742	0.49	4.18∗∗∗	2.12∗∗
Panel B: Dividend decrease announcements
EW‐market model	76	−0.85	−1.66∗∗	−1.19
VW‐market model	76	−1.01	−1.77∗∗	−1.19
EW‐market adjusted	76	−1.05	−1.93∗∗	−0.94
VW‐market adjusted	76	−0.84	−1.59∗	−0.75
This table reports three‐day cumulative abnormal returns (CARs) around the announcement of changes in dividends. Panel A details dividend increases and Panel B cites dividend decreases. We report both market‐adjusted CARs and residual market‐model CARs. We use both the CRSP equally‐weighted index and the value‐weighted index as market. Due to non‐synchronous trading, we use Charest’s (1978) standardised cross‐sectional test and Corrado’s (1989) nonparametric rank test to determine statistical significance. ∗∗∗, ∗∗ and ∗ denote statistical significance at the 1, 5 and 10% levels, respectively.

Panel B in Table 5 details the findings of the stock return for announcements of dividend decreases. The mean CARs are not entirely statistically significant – this is probably due to a lack of statistical power. ⁶ The coefficients, ranging from –1.05 to –0.84, are greater in absolute value than the mean CARs for dividend increases. These higher coefficients may yield statistical significant with more observations.

The asymmetrical reaction of the market to REIT dividend announcements is consistent with non‐REIT firms but the magnitude is considerably less. Denis et al. (1994) find that US industrial firms experience an excess stock return of about 1.25% upon announcement of a dividend increase. Alternatively, they also find an excess stock return of −5.71% when a US firm announces a dividend decrease. We observe a similar asymmetric response of almost 0.05% for an increase and about 1.00% for a decrease. Clearly, though, the magnitude is considerably less.

The muted stock return reaction to a dividend change leads to two additional thoughts when reconciling the REIT and non‐REIT market reactions. On the one hand, the fact that the market reaction is subdued is not entirely surprising given the relative transparency of REITs to non‐REIT firms – general lack of market imperfections begets a general lack of extreme market reaction, especially on dividend decreases. On the other hand, given the restrained market reaction, we question whether there is such a need to smooth the quarter dividend. Clearly, ceteris paribus, a REIT manager will not desire to suffer a 1% loss in excess stock return and will be disinclined to announcement a dividend decrease. However, does a –1% excess stock return cause a REIT manager to incur a SOA factor near 0.05 given more opaque firms smooth at SOA levels around 0.35?

Concluding remarks

We provide empirical evidence of the dividend payout behaviour of US REITs, an organisational form generally excluded from samples of other empirical dividend payout research. We examine the dividend feedback‐loop process where (i) a firm announces a dividend payment which, (ii) due to market imperfections, the market views as informational and increases the stock price upon the announcement of a dividend increase or punishes the stock value if a firm announces a dividend decrease. Since, (iii) managers of a firm understand the market’s reaction to dividend announcements and the asymmetric nature of the response, they will not increase dividends in the absence of a permanent increase in earnings, which leads to (iv) dividend smoothing as a specific level of dividend payments. Given that REITs largely mitigate market imperfections, we investigate whether the REIT organisational form halts the dividend cycle and restricts the stylised facts of smoothing and market reactions found in non‐REIT firms.

We observe a general lack of economic significance for a comprehensive set of explanatory variables used in the finance literature to interpret dividend behaviour. These include the determinants of leverage and market value of assets, which is a divergence from previous studies of the ‘old‐REIT’ era. Moreover, our findings are consistent with the transparency of the REITs organisational form. REIT dividends generally lack the informational value that is used by the literature to explain the market reaction to dividend announcements.

Given the lack of informational value, we examine the excess stock return of REIT dividend announcements. We observe a muted market reaction: on average, excess returns are about 0.5% for a dividend increase and about 1.0% for a dividend decrease. These are considerably less than the excess returns in non‐REIT studies. However, these results are not entirely surprising given the relative transparency in REITs.

Based upon the subdued market reaction, we investigate the level of dividend smoothing. We find that the Lintner (1956) measure of dividend smoothing suggests a high degree of smoothing by REITs in the quarterly data; the routine dividend set by firm management. The smoothing in annual dividends is less dramatic, but still comparable with the amount of dividend smoothing observed in industrial firms. These results could be expected since there exists a market reaction to increases and decreases and the reaction is asymmetric. In contrast, given the muted market reaction, the degree of smoothing in quarterly payments is high.

Some care should be taken in the interpretation of the quarterly results though, as the findings may be an operational remnant of managers simply paying the same dividend as last quarter. Indeed, the dividend paid last quarter may be a reference point that causes an inertia that may not be due to a strategic dividend policy; firm managers simply pay the same dividend every quarter and account for the IRS requirement or a target payout ratio at year end. Nevertheless, the annual data display about the same amount of dividend smoothing as their industrial counterparts, which does not appear to be necessary given the mild stock market reaction.

On a broader scale, our analysis is the first examination of the dividend behaviour of the modern US equity REIT. Additionally, to the extent the results can be applied to general corporate dividend policy, our findings suggest that firms will not follow a residual payout policy even in the relative absence of the standard explanations for a systematic dividend policy. REITs largely interrupt the feedback loop between firm and market; nevertheless, firm management follows a stable or smoothing payout policy.

Notes on contributors

Darren Hayunga is an Assistant Professor at the University of Texas at Arlington. His research interests include all areas of real estate and specific areas in finance including investments and corporate finance. He has published papers regarding spatial diversification, commercial real estate and land use.

Clifford P. Stephens is an Assistant Professor at Louisiana State University and has published numerous articles on corporate payout policy. His research interests include all areas of corporate finance, but, specifically, payout policy, capital structure and corporate governance issues.

Notes

1. Prior to 2001 the required payout was 95%.

2. Over 300 studies have referenced the seminal work of the Lintner’s (1956) survey of dividend behaviour.

3. See, for example, Miller and Modigliani (1961), Bhattacharya (1979), Miller and Rock (1985), John and Williams (1985), and Healy and Palepu (1988).

4. Hsieh and Sirmans (1991) find that REITs that are captive financing vehicles for their sponsors demonstrate financial performance significantly different than non‐captive REITs.

5. The Growth and Change in debt regressors in the unrestricted Tobit model do not converge in Tables 3 and 5.

6. We add 97 mortgage and hybrid REITs to the sample of dividend decreases to increase the statistical power but still find a lack of significance using the non‐parametric rank test.

Appendix

Whereas market imperfections impact the dividend policy of industrial firms, the organisational structure, regulatory requirements, and investment opportunities of REITs largely mitigate the effects of these market imperfections. Allen and Michaely (2003) detail five market imperfections that make dividend policy an important consideration for non‐REIT officials. They are taxes, asymmetric information, incomplete contracts (agency conflicts), institutional constraints, and transaction costs. We discuss the implications of each of these using the REIT structure.

Taxes

Elton and Gruber (1970) show that income taxes induce dividend clienteles. To qualify as a REIT, a firm must meet a number of operational, distribution and compliance requirements. If the firm satisfies these requirements and is classified as a REIT, it will avoid paying taxes on any dividends paid from its taxable income. Hence, a REIT that distributes 100% of its taxable income will not have a federal income tax liability. Given that REITs must pay at least 90% of their taxable income in dividends, and that many REITs pay 100% or more, taxes are largely eliminated within a REIT.

Additionally, due to the required dividend payments and high yields, REITs create a subset of dividend‐paying stocks that should result in a homogenous investor clientele. Investors purchase REIT stocks knowing that they will receive a high dividend yield, and REIT shareholders self select based upon their preference for current yields. Thus, REITs largely hold constant tax‐clientele effects since there is not a great distinction between REITs that pay a high dividends versus firms that pay out low dividends. Contrasted to any other firm, REITs pay out a considerably high dividend.

Another aspect of the market imperfection of taxes is how repurchases relate to dividend policy as part of the firm’s total payout policy. When a firm repurchases shares, self‐selecting investors desiring liquidity can sell shares and receive preferential tax treatment through capital gains instead of being taxed on dividend payment as ordinary income. The historic challenge in financial economics has been explaining why firm pay dividends in lieu of repurchasing shares given the tax impacts. Again, with regards to REITs, the issue is largely mitigated. Once a REIT has distributed all of its taxable income as dividends, any additional distribution is considered to be the return of capital. The additional payments reduce the investor’s tax basis. With a reduced tax basis, the return of capital is taxed as a capital gain. In sum, the tax incentive for share repurchases is lost – the US tax law requires a certain dividend payment and the excess payments are ultimately taxed as a capital gain.

Asymmetric information

The asymmetric information or signalling models developed by Bhattacharya (1979), John and Williams (1985), and Healy and Palepu (1988) posit that managers choose dividend payouts based upon their inside knowledge of the true worth of their firm. Managers convey private information to investors through dividends, which is a costly signal since an increased dividend payment cannot be easily mimicked by other firms without increasing the probability of later incurring a dividend decrease – an event that is generally punished by the market through a reduction in stock price.

For non‐REIT firms, project confidentiality, adverse selection and moral hazard can hinder the direct transfer of information between market participants. Conversely, REIT investors and analysts are well aware of the property assets and mortgages held by publicly traded REITs. Given (i) the ability to witness the holdings of a REIT, (ii) that there exists an active market for assets similar to those owned by REITs, and (iii) that the REIT’s income is not based upon technologically advanced processes, investors have great insight into the investments in place, current cash flows and earnings. These facts leave little private information with REIT management.

Equity REITs offer another reduction in asymmetric information when assessing firm value. Instead of discounting long‐term dividend projections, REIT analysts estimate the aggregate net operating income of the REIT’s holdings for the next year. The aggregate net operating income is then capitalised by a weighted average capitalisation rate for the portfolio. The capitalisation rates for REITs are quite specific and standardised within each particular REIT sub‐industry (e.g. apartment, industrial and retail). As detailed by Ling and Archer (2005, p. 402), capitalisation rates are determined directly from comparable sales transactions, appraisers, and institutional investors. The capitalisation process also uses information from the private property market to perform a mass appraisal of the REIT’s properties. The ability to use information from the well‐functioning private property market to value shares of a publicly traded REIT is unique and distinguishes the REIT market from other industries where assets are not separately traded in a private market.

One area that REITs could differentiate themselves through signalling is the value of future investment opportunities. A firm with confidence in their ability to capitalise on better future projects than competitors can signal itself as a high‐quality firm using dividends. A lower‐quality firm with less confidence concerning future prospects may not be able to mimic a higher than normal dividend payout.

Incomplete contracts/agency costs

The focused nature of REITs reduces asymmetric information and potentially agency costs. One of the conditions of REIT status is that at least 75% of a REIT’s assets must consist of real estate assets, cash, and government securities. A further requirement is that at least 75% of the REIT’s quarterly gross income must be obtained from real estate assets. These conditions aid in our examination of dividend policy by restricting firms in the study to the real estate industry. Demsetz and Lehn (1985) argue that regulation restricts the investment options available to managers. Smith and Watts (1992) extend the argument by predicting that a restricted investment set assists in mitigating agency problems.

The disciplining mechanism proposed by Easterbrook (1984) should further mitigate agency concerns for REITs. Easterbrook (1984) posits that when firms pay dividends, managers must raise funds more frequently in the capital markets. Once in the capital markets, firms are subjected to the monitoring by investment professionals. Therefore, the paying of dividends provides monitoring that should reduce agency conflicts. Since REITs must pay a relatively high level of dividends, they are active in the capital markets. Further, these high payout ratios should result in REITs disgorging material amounts of free cash flow (FCF). Jensen’s (1986) FCF hypothesis suggests that firms disgorging large amounts of FCF in the form of dividends are less likely to engage in negative NPV projects. Consistent with this argument, Mooradian and Yang (2001) find that equity REITs within the hotel industry have significantly smaller amounts of FCF than non‐REIT hotel firms.

Other market imperfections

The last two market imperfections discussed by Allen and Michaely (2003) are institutional constraints and transaction costs. Regarding institutional constraints, Allen and Michaely (2003) detail how an institution may avoid investing in low‐dividend paying stocks due to constraints such as the ‘prudent man rules’. Since ERISA plans and private trusts have a fiduciary responsibility, these institutions are theorised to want to hold investments in firms that pay a higher dividend. Since all REITs pay at least 90% of their taxable income, the entire class of REITs is distinctly different from non‐REIT firms. Institutional constraints are largely mitigated by REITs since the entire asset class pays out a substantial dividend.

Similarly, the naturally high dividend payout mitigates transaction costs as a viable hypothesis affecting REIT dividend policy. The theory propagates that for those investors that desire a steady flow of income, dividend payments may be the cheapest method of distribution. This is in contrast to selling firm shares and generating capital gains. The transaction costs can be the actual buying and selling costs, the time and effort spent on the transaction, or the effort of self control along the lines of Shefrin and Statman (1984). Within the REIT industry, investors have presumably already self‐selected to receive a high dividend payout. There is not a contrasting difference between dividends and selling shares within REITs since a high dividend payout is mandatory.