A time-varying dividend premium across
the business cycle
By D.S. Smidt
Student number 1399799
Date June 21, 2013
Supervisor Dr. J.O. Mierau
Master thesis MSc Finance Faculty of Economics and Business
University of Groningen
Abstract
Baker and Wurgler (2004a) were the first to observe evidence concerning the existence of a time-varying dividend premium. Fuller and Goldstein (2011) concurrently, found a higher (lower) dividend premium during down (up) markets, but could not elicit a fundamental explanation for its occurrence. This paper uses leading indicators as an a priori proxy for market expectations (investor sentiment) and finds a time-varying dividend premium across the business cycle. There are indications that it is not the height of relative market returns, but rather investor sentiment that produces a more significant impact on investor demand for dividend paying stocks, and thusly, the dividend premium.
JELclassification
G35
Keywords
1
1. Introduction
A lot of time has been spent evaluating whether or not the distribution of dividends adds a premium to a firm’s value. Despite the fact that during the 1970’s, numerous studies alluded to a positive relationship between the payout of dividends and a firm’s valuation, recent studies have difficulty attesting such a relationship. One explanation for these mixed results could conceivably be that the pay out of dividends does impact on a firm’s valuation, but at different directions at different times.
Baker and Wurgler (2004a) were the first to acknowledge the existence of a time-varying dividend premium. They suggest that managers cater to investor demand for dividends by distributing more cash when the dividend premium is high. Furthermore, they suggest that the dividend premium is largely driven by investor sentiment. The demand for dividend paying stocks seems to be low during times when the sentiment for growth stocks is high. On the other hand, the dividend premium seems to be high after a crash.
Not only firms, but also mutual funds seem to be well aware of this time-varying investor demand for dividend paying stocks. To illustrate; a fund manager1 from Kempen Capital Management writes in a recent column2 that, in the search for relatively safe stock exposure, the “high dividend yield strategy” has received competition from the “low volatile” strategy. Noteworthy is that Kempen Global Dividend Fund only started operations on the 24th of October 2007, just after a large market crash. Do mutual funds therefore also cater to investor demand to maximize cash inflows? Cooper
et al. (2005) find evidence that mutual funds altered their names or professed strategy
to what they perceived might be the flavor of the month, in order to maximize the influx of funds.
As of yet, only one other study by Fuller and Goldstein (2011) has empirically investigated whether the dividend premium is indeed time-varying. They investigated whether or not there was a difference regarding the dividend premium in up and down market months and find a larger dividend premium during down markets than during up markets. After controlling for a varying number of firm fundamentals they can’t seem
1 Jorik van den Bos manages Kempen Capital Management’s Kempen European High Dividend Fund and
Kempen Global Dividend Fund.
2
to find a clear explanation of why there is a time-varying dividend premium. So the question remains: does investor sentiment as proposed by Baker and Wurgler (2004a) explain a time-varying dividend premium or are it firm fundamentals that increase the value of dividend paying stocks during times of distress, and hence increase investor demand for dividend paying stocks? This paper will try to attempt to provide new insights regarding this question.
It is a daunting task to measure investor sentiment. Therefore I shall employ a priori market expectations as a proxy for investor sentiment. If investors believe markets will rise, sentiment will be(come) (more) positive, whereas if investors believe markets will go down, sentiment will be(come) (more) negative. I base these a priori market expectations on so-called leading indicators. Leading indicators are used by investors as prognosticators of future macro-economic environments. Siegel (1998) shows that investors who are knowledgeable with regards to the macro-economic environment and its immediate circumstance are able to generate significant excess returns over a buy-and-hold strategy. Therefore it might be attractive for investors to allocate both time and effort in order to gain insight into future macro-economic developments. One might question whether investors are indeed using leading indicators for trading. When looking at recent3 stock market movements you might think so. Stock exchanges4 were breaking all-time highs in May, even though newspapers are still writing economic growth to be in recovery. Investors believe economic activity to increase in the near-term future and equities are therefore rising solely based upon these expectations!5
Furthermore, by assigning each monthly leading indicator variable into a quadrant of the business cycle: (1) slowdown, (2) recession, (3) recovery, (4) expansion, it can be determined whether the dividend premium varies across the business cycle. If investor demand for dividend stocks does indeed change “predictably” over the business cycle (as defined by the leading indicators), these changes in investor demand will cause the dividend premium to vary “predictably” across the business cycle. Therefore, an investor might be able to utilize this knowledge by means of implementation into
3 Time of writing: 06-21-2013
4 The S&P 500 and Dow Jones Industrial Average
5 Ignoring the Quantitative Easing (QE) program the Federal Reserve is currently undertaking, which
3
his/her asset allocation model and thereby observe improvement of their respective portfolio performance. Brocato and Steed (1998) show that using a broadly diversified collection of debt, equity, and commodities, portfolio weights should be modified to accommodate cyclical shifts in the macro-economic environment if mean variance efficiency is to be maintained over the business cycle.
The paper is organized as follows. Section 2 provides a short overview of the history regarding empirical research related to the dividend premium. Furthermore, theories are formulated reasoning why the dividend premium is or should be time-varying. Section 3 provides a description of the sample and dataset used, a description of which regressions are to be used and why, as well as a firm description of how the regressions are estimated. Section 4 provides us with main results, a discussion, and robustness checks. Section 5 provides summaries and final conclusions.
2. Literature review
The question whether dividends add a premium to the value of the company and thereby on its company stock prices has been researched quite intensively in the past. The first empirical research surrounding this question can be traced back to the 1950’s. The most significant contribution of this period is undoubtedly the work of Modigliani and Miller (1961). They prove that the value of the firm is unaffected by current and/or future dividend decisions in a perfect, frictionless market. Higher dividends now will lead to decreased dividends in the future, as surely as investments here-and-now are needed to obtain higher earnings in the future. Therefore changes in dividend policy will only alter the distribution of the company’s total returns. Only irrationality would command a premium on high dividends yield stocks, whereas a discount should be expected when taking taxes into consideration since dividends are more heavily taxed than capital gains.
4
Blume (1980), Ball et al. (1979), Litzenberger and Ramaswamy (1979, 1982) and Rosenberg and Marathé (1979).
Formerly, the practical positive relationship between dividends and common stock returns was a somewhat surprising result which is why Black (1976) formulated the dividend puzzle: “Dividends are irrational when they are taxed at higher rates than capital appreciation and cutting dividends is the cheapest way of new source of capital. Then why do investors demand dividends?”
Different hypotheses were formed attempting to explain this “irrational” dividend premium. Among them the free cash flow hypothesis of Jensen (1986) and Easterbrook (1984). Conflicts of interest between shareholders and managers are more likely to occur in firms with large free cash flows. Payouts to shareholders reduce the resources under manager’s control, thereby reducing manager’s power and making it more likely they will incur the monitoring of the capital markets which occurs when the firm must obtain new capital.
A second hypothesis allows firm’s managers (insiders) to know more than outside investors about the true state of the firm’s future earnings potential (Miller and Rock, 1985). The dividend payout decision signals the firm’s future earning power. Therefore an increase in payouts is associated with higher (future) profitability and is therefore followed by a positive stock price reaction. According to Fuller and Blau (2010) the signal observed by outsiders regarding dividends is nonmonotonic. Higher quality firms pay dividends to eliminate the free cash flow problem, whereas firms with the lowest prior performance pay the lowest dividends because they are correctly identified. They both have no need to signal their profitability by paying out (more) dividends. It is the firms with mediocre prior performance that pay dividends, both to solve their free cash flow problem and as to distinguish themselves from others with similar prior performance. Essentially, firms with intermediate prior performance have the highest abnormal stock price reaction to a dividend change and these firms are the ones with the highest dividend yields.
5
Grullon et al. (2005) do not find supporting evidence. Regarding the free cash flow hypothesis, Lang and Litzenberger (1989) find that excess returns around the announcement of a dividend increase is positively related to a firm’s potential to overinvest, whereas Denis et al. (1994) and Yoon and Starks (1995) find only little evidence of such a relationship.
A third explanation is the catering explanation proposed by Baker and Wurgler (2004a). The decision to pay dividends is driven by prevailing investor demand for dividend payers. Managers cater to investors by paying dividends when investors put a stock price premium on payers and vice versa. They find that dividends are highly relevant to share value, but in different directions at different times. Managers cater to changing investor demands regarding dividend payments. They suggest that investor sentiment impacts investor demand for dividend stocks. Baker and Wurgler (2004b) studied one hundred and three abstracts of New York Times articles published between 1969 and 2001. Their findings allude to a large number of stories to be suggestive of time-varying catering incentives. Based on these articles they conclude that the demand for dividend paying stocks seems to be low at times when the sentiment for growth stocks is high. On the other hand, the dividend premium seems to be high after a crash. Dividend paying stocks have a safe image, thereby increasing investor demand for dividend paying stocks during times of distress. If sentiment is responsible for the time-varying dividend premium, then what explains investor sentiment?
I pose that the state of the business cycle is largely responsible for changes in investor sentiment. The term business cycle refers to the system-wide fluctuation in economic activity observed over several months or years. Typically these economic fluctuations alternate around a long-term growth trend and are characterized by periods of relatively rapid economic growth and periods of relative stagnation or decline. The economic cycle can be divided into four quadrants: expansion, slowdown, recession, and recovery. In an expansion growth is above trend and rising, in a slowdown growth is above trend but declining, in a recession growth is below trend and declining, and in a recovery growth is below trend but rising.
6
differences in performance across different parts of the business cycle. So while bonds are expected to do best during recessions due to decreasing interest rates, equity is expected to do best during recoveries due to newly arising growth opportunities and an increase in demand for firm products. Commodities are expected to do best during expansions due to an increase in demand for production materials.
Now, what causes these changes in investor demand across the business cycle? Not only asset classes have differing risk characteristics and attributes. There are also differences within asset classes. For example, Perez-Quiros and Timmermann (2000) find that small, less liquid firms with little collateral are more strongly affected by tighter credit market conditions in a recession state than large, better-collateralized ones. Small firms depend more heavily on information-intensive sources of credit such as bank loans. Bernanke and Gertler (1989) hypothesize that a recession may result in a “flight to quality”, causing investors to stay away from the high-risk small firms and switch towards better collateralized and hence, safer, larger firms. It are these quality firms which have the highest propensity to payout dividends. For example, Denis and Osobov (2008) and DeAngelo et al. (2003) found that the propensity to pay out dividends is higher among larger, more profitable firms, and for those which retained earnings comprise a large fraction of total equity. This “flight to quality” must thus in part be a “flight to dividends”. This increase in demand for dividend-paying stocks should increase the dividend premium during times of distress on the stock markets.
7
This theory of rational investors contradicts the findings of Li and Lie (2006). They find the probability that a firm increases its dividends to be higher when the dividend premium is high and vice versa. If indeed the dividend premium is largest during times of distress, this means that firms pay out more dividends during times of distress. This does not make sense from an economic standpoint and if this is indeed true, investors are irrational. It would mean that the decision to pay out dividends is at least partly driven by the catering theory of Baker and Wurgler (2004a) and is therefore based on investor demand for dividend paying stocks and not on firm fundamentals. This is exactly what Ali and Urcan (2012) find. When the dividend premium is high an increase in dividends does not signal future earnings. The catering incentive increases misinformation. If managers are indeed maximizing stock prices, firms pay out money they can’t miss. However, when the dividend premium is low and therefore the catering incentive is low, an increase in dividend payouts does provide information about future firm profitability.
A few other related studies have also found evidence pointing towards a time-varying dividend premium across the business cycle. For example, Davis and Madura (2012) have researched what ex-ante factors influenced the stock price movements the most during the financial crisis from 2007-2008. Winner stocks during the crisis represented larger firms, offered a higher dividend yield and exhibited relatively lower volatility before the financial crisis. They found a dramatic shift in sentiment away from risk and a “flight to quality”. They suggest that their results provide strong support for the theories of projection bias, risk aversion and regret avoidance. These behavioral explanations are very likely to magnify a shift in investor preferences during a severe crisis like the one in 2007-2008.
8
and vice versa. An up market month might occur after a severe down market month because of a tendency from investors to overreact to bad news. When investors become aware of their overreaction an up month might follow, even though we are still in a recession. Therefore it is very hard to predict whether next month will be an up or a down market month. Timing when to shift your portfolio towards more risky investments, or when to shift towards quality, will be very hard. Basing the timing decision on where we find ourselves on the business cycle might be a better idea.
This paper investigates whether or not there is a time-varying dividend premium and what the sign and magnitude of this premium is across different states of the business cycle. Furthermore, the question whether or not there is a relationship between the relative height of market returns and the dividend premium will be researched.
3. Data and methodology 3.1 Dataset
The dataset consists of company data for the constituents of the Dow Jones Industrial Average, the NASDAQ Composite and the S&P 500, which consist of 30, 2435 and 500 companies respectively. Due to the fact that some companies are traded on more than one index, doubles have to be removed. A sample of 2825 stocks remains. The sample period is bounded from 03-01-1988 to 01-01-2013 and exists of 299 monthly observations.
First, monthly stock return data is obtained from Thomson Reuters datastream in the form of Total Return Indices (TRI). A TRI reinvests paid out dividends.
Also, the dividend yield (DY), the money amount of yearly sales (SAL), the market-to-book ratio (M/B) the amount of available free cash flow (FCF) and the return on equity (ROE) are acquired from datastream. All data are monthly distributed, except for FCF and ROE, which are both distributed on a yearly basis and SAL, which is either distributed on a yearly or quarterly basis.
Furthermore, data about the United States amplitude adjusted Composite Leading Indicators (CLI)6 are downloaded from the website of the OECD7. Each monthly data
6 http://www.oecd.org/std/clits/41629509.pdf
9
point is assigned to one of the following four dummy variables: recession, recovery, expansion or slowdown. The US 3-month Treasury bill is taken as a proxy for the risk-free rate (Rf) and the TRI of the S&P 500 is taken as a proxy for market returns (Rm). The data form an unbalanced panel. See Table 1 for descriptive statistics.
Table 1 - Descriptive statistics
Description of 2,825 S&P 500, Dow Jones Industrial Average and NASDAQ Composite listed firms for a total of 299 monthly observations from March 1988 till January 2013. The up - down dummy is a dummy variable indicating either an up market month or a down market month for the S&P 500. A value of 0 indicates a down market and a value of 1 indicates an up market. The dividend dummy has a value of 0 when a firm is identified as a non-dividend paying firm and a value of 1 when a firm is identified as a dividend paying firm. Monthly data for the United States amplitude adjusted Composite Leading Indicators (CLI) are downloaded from the website of the OECD. Each monthly data point is assigned to one of the four dummy variables: recession, recovery, expansion or slowdown. In a recession future expected growth is below trend and decreasing, in a recovery expected future growth is below trend but increasing, in an expansion future growth is expected to be above trend and increasing, and in a slowdown future expected growth is above trend but decreasing.
Avg. Median Min. Max. Std. Dev. Monthly returns Sample 0.0084 -0.0007 -0.9642 1.2500 0.1570 S&P 500 0.0056 0.0106 -0.1667 0.1597 0.0443 NASDAQ 0.0055 0.0065 -0.1888 0.1736 0.0582 Up - down dummy 0.6254 1.0000 0.0000 1.0000 0.4840 Dividends Total sample 0.0108 0.0000 0.0000 0.2494 0.0190 Dividend payers 0.0270 0.0225 0.0001 0.2494 0.0217 Dividend dummy 0.4008 0.0000 0.0000 1.0000 0.4900 Leading indicators Recession 0.2508 0.0000 0.0000 1.0000 0.4335 Recovery 0.2508 0.0000 0.0000 1.0000 0.4335 Expansion 0.2676 0.0000 0.0000 1.0000 0.4427 Slowdown 0.2308 0.0000 0.0000 1.0000 0.4213
10
out dividends. Technology firms are expected to be growth stocks and paying out dividends might signal a lack of growth opportunities to investors, which in reaction put a discount on these stocks. But, one can argue that growth possibilities during a recession are worth less than during a boom, therefore increasing the relative value of dividends during recessions.
There might also be a survivorship bias in the sample. The sample consists of the stocks that are the constituents of the aforementioned indices right now. That means that stocks that were constituents of these indices in the past, but are delisted during the sample period, are left out of the sample. Delisting is most likely to occur due to one of three reasons: bankruptcy, going private and takeovers. Firms that went bankrupt are most likely to have had the lowest stock returns and, theorizing they probably had low free cash flow, were least likely to pay out dividends. Firms that went private or were taken over are likely to not be the biggest companies out there. It’s hard to imagine an investor group taking Apple and/or Google private or for another company to overtake them. For these reasons the sample is most likely biased towards larger companies and since these companies are more likely to pay out dividends, towards dividend paying stocks.
Lastly, the sample contains more cross-sections for later years than for earlier years. This is mostly due to the fact that not all companies in the indices today were founded before March 1988, but might also be due to some data non-availability during the earlier years of the sample. For example, in March 1988 the sample contains 446 cross-sections compared to 2586 cross-cross-sections in December 2012. Although this is a big difference, I think it is still possible to get robust results using 446 cross-sections. A robustness check will be performed later on to check whether this is indeed true.
11
monthly returns lower than -100% were removed from the sample, simply because this is not possible. Similar actions were taken for the variables ‘DY’, ‘M/B’ and ‘ROE’.
3.2 Methodology
The regression used to estimate the influence of the state of the business cycle on the dividend premium is an extension of the three factor model originally designed by Fama and French (1993), first introducing size and the market-to-book ratios as additional risk factors.
, (1)
where is the simple return for company at time calculated as in Brooks (2008):
, (2)
where is the value of the TRI index for stock at time and is the value of the TRI index at time . Simple returns are used instead of continuously compounded returns since these are most appropriate for monthly stock returns. The skew of lognormal returns is positive, whereas actual monthly market returns are negatively skewed. One problem of using simple returns is that it biases the sample average returns upwards. Consider a stock with a value of $20. The first month, the stock’s value drops by 90%. The next month the same stock rises with 200%. Simple returns now indicate a positive return of 55%, while the absolute value has dropped to $4. A negative return of 80%! For this reason average sample returns are larger than the average market returns of the S&P 500 and the NASDAQ (Table 1). is the risk-free rate which has been estimated by the US 3-month Treasury bill. is the regression intercept. is estimated by transforming the TRI of the S&P 500 according (2). is taken as a proxy for size. Its natural logarithm is taken to transform the exponentially distributed size factor to a (more) normally distributed variable.
12
The coefficients , and are proxies for the risk-factors as determined in Fama and French (1993), where is the sample Beta. The value of is the risk factor placed on size. This coefficient is expected to be negative since smaller firms tend to outperform larger firms (over a whole business cycle that is). The value of is the risk factor placed on higher market-to-book ratios. Fama and French (1993) find a higher market value of equity relative to the book value of equity to be related to higher earnings on assets and that this is persistent for the following five years. Therefore this coefficient is expected to be positive.
As in Black and Scholes (1974), a stock is defined as dividend-paying the first month after it has initialized paying out dividends and will be identified as such as long as it is expected to keep paying out dividends on a regular basis. This can be either quarterly, semi-annually or yearly. The is the last paid dividend, normalized for a yearly distribution, divided by the stock price at time . This means that the variable will change monthly, even though it only pays dividends quarterly/yearly, just due to changes in the stock price.
The coefficient is expected to be positive. Earlier empirical works have found enough evidence to support this expectation, as for example, Blume (1980), Ball et al. (1979), Litzenberger and Ramaswamy (1979, 1982), Rosenberg and Marathé (1979) and Fuller and Goldstein (2011).
13
smoother indicator of the future state of the economic business cycle than up and down markets. Up and down markets have less serial correlation and seem to occur more randomly. Also, as can be seen in Table 1, there seems to be more up markets than was predicted by the CLI data. 187 up market months vs. 155 predicted by the CLI data (up markets are expected during recoveries and expansions whereas down markets are expected during slowdowns and recessions). Furthermore, the up and down market dummy variable as used by Fuller and Goldstein (2011) is a hindsight indicator of market expectations, since it is based on past market returns. This differs from the OECD CLI, which is an a priori proxy of market expectations.
I transform this CLI into four dummy variables, , where can take four different values. Each of these values represents a forecast of the future state of the economic business cycle, namely expansion, slowdown, recession and recovery. They are defined as follows: in an expansion growth is above trend and upwards, in a slowdown growth is above trend but downwards, in a recession growth is below trend and downwards, and in recovery growth is below trend but upwards. The can only take the value of either a 0 or a 1.
A data point with a value of above 100 indicates an above trend GDP growth, whereas a value of below 100 indicates a below trend GDP growth. The trend is established on the basis of taking previous month’s data point and comparing this to next month’s data point. If this month’s data point is larger (smaller) than previous month’s data point, the trend is upwards (downwards). See Figure 1 for a graphical representation.
14
smaller during recoveries and expansions. The mirror image of the states of the economic cycle in which negative returns, respectively positive returns are expected.
Figure 1. The different states of the business cycle. In an expansion GDP growth is above trend and
rising. In a slowdown GDP growth is above trend, but declining. In a recession GDP growth is below trend and declining, and in a recovery GDP growth is below trend, but rising. The x-axis represents the trend line or the average normalized GDP growth.
3.3 Free cash flow
According to the free cash flow hypothesis conflicts of interest between shareholders and managers are more likely to occur in firms with large free cash flows. The companies with the largest free cash flows are in most need to pay out dividends in order to diminish possible agency costs which may arise when managers start overinvesting. During recessions profitable investment opportunities have most likely been decreasing, thereby increasing the chance of overinvestment among managers. This should put an increasing value on the distribution of cash during economic downturns.
15
one can only make such a decision when one has enough cash lying around. Therefore, an increased dividend premium during a recession might simply be related to the fact that dividend paying firms have, on average, higher free-cash flow available and are able to make this before mentioned optimal payout decision. To check whether this is the case, is added to regression (1):
(3)
It can be the case that the results found for the relative dividend premiums, , in (3) differ significantly from the results found for the in (1). If so, the addition of FCF to the regression is responsible. Then it is not the payment of dividends that explains the time-varying dividend premium, but the amount of free cash flow available inside a company. This can be interpreted as further proof for the findings of Latham and Braun (2008), since free cash flow and current ratios are highly correlated.
3.4 Quality
As mentioned earlier, larger, more profitable and more mature firms have a higher propensity for paying out dividends. Allen et al. (2000) argue that dividend paying firms are of better quality. Perez-Quiros and Timmermann (2000) find that smaller, less liquid firms and with less collateral are most strongly affected by economic downturns. Therefore the more liquid, larger and more profitable firms are most likely to keep continuing paying dividends during these economic downturns. To control for this quality effect, is added to regression (1):
(4)
16
another quality variable in the regression. If indeed quality differences are responsible for valuation differences during economic downturns, the slope dummy variables in (1) might have been proxies for the premiums placed on quality differences by investors. In this case the coefficients for the slope dummy variables found in (4) will become insignificant.
3.5 Estimating the regressions
All regressions are estimated using ordinary least squares (OLS). Furthermore, a redundancy fixed effects test is employed to check for fixed effects. This test restricts the fixed effects to zero. When the p-value found in this test is significant (P<0.05), that means that the restrictions are not supported by the data and that a pooled sample cannot be employed (Brooks, 2008). The regressions can’t be estimated using time fixed effects (or both time and cross-section fixed effects) due to a near singular matrix. This is caused by the regression dummy’s and should not have an impact on results, since the dummy’s used already control for time-varying effects.
Next, the regressions are estimated using random effects. Due to an unbalanced panel I can’t both check for cross-section and period random effects together, but I can estimate the regressions using either one. Now, a Hausman test is employed to check whether the random effects are uncorrelated with the explanatory variables.
Although some period random effects were found estimating regressions (1), (3) and (4), evidence was not convincing. Furthermore, the redundancy fixed effects test shows strong evidence for cross-section fixed effects. Again, due to an unbalanced panel I was not able to estimate the regressions using both cross-section fixed and time random fixed effects. Therefore all regressions are estimated using cross-section fixed effects only (except during the robustness checks where some period random effects are found). According to Brooks (2008) this makes sense. He states that when the entities in the sample effectively constitute the entire population (for example, when the sample comprises all of the stocks traded on a particular exchange) a fixed effect model is more appropriate. The variance-covariance matrix is calculated using White’s period errors, allowing for correlation over periods for each individual stock.
17
Findings from this step can be interpreted as the sample coefficients. Step 2 runs the regressions including both the dummy variables as well as the slope dummy variables and thereby determining (1) the relative stock performance during the different states of the business cycle and (2) the dividend premium across the different states of the business cycle. For example, regression (1) will be estimated as follows:
Step 1:
(1a)
Step 2:
(1b)
The reason for this vast methodology is that Eviews doesn’t allow estimating the regressions without a constant. Therefore one of the dummy’s, , has to be left out of the regression in order to avoid the dummy variable trap; if the intercept and the four dummy variables were all included in the same regression a case of perfect multicollinearity would occur and none of the coefficients could be estimated. I decided to leave the dummy variable for slowdown out of the regression. This has a few consequences: (1) the values found for the coefficients in the regressions estimated during step 2 can be interpreted as the coefficients during a slowdown, (2) the values found for the included dummy variable coefficients (recession, recovery, and expansion) are to be interpreted as relative to a slowdown and (3) the constant, , found in step 2 has to be subtracted from the constant, , found in step 1 to determine the relative stock performance during a slowdown.
18
not to be confused with the value found for , the sample dividend premium, during step 1, (2) the values found for the three coefficients, , are the relative deviations to this slowdown dividend premium during either a recession, a recovery or an expansion and (3) the slowdown dividend premium, , has to be subtracted from the sample dividend premium, , found in step 1 in order to determine the relative average dividend premium during a slowdown. This procedure makes sure that I now have four coefficient values for both the dummy variables as well as for the slope dummy variables.
Now, to test whether these (slope) dummy variables have any statistical power t-tests are performed. This is done by (1) subtracting the relative stock performance over the business cycles from the average sample stock returns and dividing each value by the individual dummy standard errors, and (2) subtracting the average sample dividend premium from the values found for the slope dummy variables in step 3 and dividing each by the individual slope dummy standard errors found during the estimation of step 2:
, (5)
where is the samples average stock return, are the four dummy variable coefficients, each representing one state of the business cycle, and are the four individual dummy variable standard errors during each state of the business cycle.
, (6)
19
4. Results and discussion
Baker and Wurgler (2004a) found that dividends are highly relevant for share value but at different directions at different times. They suggest that their finding is due to catering to investor demands and that investor demand is time-varying due to changes in investor sentiment. I suggest that investor sentiment is most likely to change due to fluctuations in economic activity as indicated by the business cycle. In order to investigate whether the dividend premium varies across different states of the economy, I calculate the average dividend premium of the sample as a whole and over the different states of the business cycle as defined by the OECD CLI. Furthermore I calculate the relative stock performance across different states of the business cycle. Do higher (lower) stock returns result in a relatively lower (higher) dividend premium?
4.1 Main results
During step 1 all regressions are run without any dummy variables. Results are presented in Table 2. All regressions show similar results. All coefficients are highly significant, except for in (3), which indicates that, on average, free cash flow does not have an impact on stock returns.
Table 2 - Step 1: Estimating the regressions without dummy variables
Regressions (1), (3) and (4) are estimated without any dummy variables to determine the sample regression coefficients.
(1a) -
(3a) -
(4a) -
where is the monthly stock return for company at time , is the risk-free rate at time , is the market risk premium at time , is the natural logarithm of the
quarterly/yearly sales for company at time , is the market-to-book ratio for company at time , is the dividend yield for company at time , is the free cash flow available for company at
time , and is return on equity for company at time .
Panel A: Step 1 regression coefficients Regression
(1a) 0.0534* 1.0813* -0.0037* 0.0010* -0.0048* - -
(3a) 0.0571* 1.0856* -0.0040* 0.0010* -0.0048* 0.0000 -
(4a) 0.0602* 1.0709* -0.0043* 0.0018* -0.0050* - -0.0143*
20
A few remarkable observations can be made. For example, , the dividend premium is negative in all regressions. This contradicts with the findings of Blume (1980), Ball et al. (1979), Litzenberger and Ramaswamy (1979, 1982), Rosenberg and Marathé (1979), and Fuller and Goldstein (2011). It does coincide with the findings from Baker and Wurgler (2004a). They have found the dividend premium to be negative from 1978 till 2000. They use the difference between the logs of the average market-to-book ratios between dividend payers and non-payers in order to determine the dividend premium, whereas the aforementioned studies have used market returns. The negative dividend premium found can also be a consequence of the sample used in this paper. The constituents of the NASDAQ Composite are a large part of the sample and mainly consist of technology stocks. Technology firms are expected to be growth stocks and paying out dividends might signal a lack of growth opportunities to investors, which in reaction put a discount on these stocks. The fact whether the dividend premium is on average positive or negative should not have an impact on the question I really want to answer: does the dividend premium vary across different states of the economic cycle?
The coefficient is insignificant. This can be interpreted as free cash flow having no effect on monthly stock returns when all other variables are included. The coefficient, , in (4) is negative, which indicates that a higher ROE results in a lower monthly stock return. This seems counterintuitive, since a higher ROE is likely to result in a more profitable firm, which should in turn lead to above average stock returns. Maybe the positive effect of ROE on monthly stock returns is already incorporated by the M/B variable?
All other coefficients seem to have the appropriate sign. is larger than one, which makes sense, since average monthly sample returns are larger than average monthly market returns. is negative, which indicates that stock returns decline when size increases. A higher M/B-ratio increases stock returns, as indicated by a positive sign for .
21
Table 3 - Step 2: Interpreting the slope dummy variables
Full regressions (1), (3) and (4) are estimated to determine: (1) the relative stock performance over the business cycle and (2) the interaction effect between dividend yield and the state of the economic business cycle. A t-test is executed to determine whether there is a statistical difference between (1) the average sample stock returns and the stock returns over the different states of the business cycle, and (2) the sample dividend premium and the dividend premium found for the different states of the business cycle. The following regressions are estimated:
(1b) -
(3b) -
(4b)-
where is the monthly stock return for company at time , is the risk-free rate at time , is the market risk premium at time , is the natural logarithm of the yearly sales
for company at time , is the market-to-book ratio for company at time , is the dividend
yield for company at time , is the free cash flow available for company at time , is
return on equity for company at time , are dummy variables which can take a value of 0 or 1,
indicating the expected future state of the business cycle at time as predicted by the OECD Composite Leading Indicators. can take four values: slowdown, recession, recovery and expansion.
are slope dummy variables for which its coefficient indicates the relative dividend premium across different states of the business cycle.
Panel A shows the step 2 regression coefficients, which are the result from estimating the regressions as described in section 3.5 and can be interpreted as the regression coefficients during a slowdown. The dividend coefficient is left out of Panel A since this variable can be interpreted as the dividend premium during a slowdown.
Panel B shows the relative dividend premiums across the different states of the business cycle. A positive (negative) number represents a higher (lower) dividend premium during that state of the business cycle as opposed to the samples average dividend premium. The significance of the relative dividend premiums during each state of the cycle is tested as indicated in (6).
Panel C shows the relative market returns across the different states of the business cycle. A positive (negative) number represents a higher (lower) market return during that state of the business cycle as opposed to the samples average market returns. The significance of the relative market returns is tested as in (5).
Panel A: Step 2 regression coefficients
Regression
(1b) 0.0382* 1.0448* -0.0035* 0.0010* - -
(3b) 0.0404* 1.0489* -0.0037* 0.0010* 0.0000 -
(4b) 0.0439* 1.0392* -0.0039* 0.0019* - -0.0132*
Panel B: Relative dividend premiums ( )
Regression Slowdown Recession Recovery Expansion
(1) 0.0016* 0.0025* -0.0032* -0.0031*
(3) 0.0017* 0.0026* -0.0032* -0.0034*
(4) 0.0016* 0.0025* -0.0030* -0.0031*
22
represents the relative stock performance during slowdowns, whereas the difference between and indicates a slowdown beta differing from the sample beta. The lower value of is most likely a result of the relative lower stock returns during slowdowns.
Findings are consistent across the different regressions. In order from worst performing to best performing: 1. slowdown, 2. recession, 3. expansion and 4. recovery. Performance is a staggering -1.6% a month during slowdowns compared to the sample average. During a recession performance is about 0.9% a month worse. During recovery, stocks outperform by about 1% and during an expansion stocks outperform by about 0.1-0.2%. This is in line with the findings of Siegel (1998). All t-tests are significant at the 1% level, except for the relative stock performance during expansions for regression (4). The dividend premium is above the sample average during slowdowns and recessions, whereas the dividend premium is below the sample average during both recoveries and expansions. Dividend paying stocks perform on average 0.16% and 0.25% a month better during slowdowns and recessions respectively for every one percent increase in dividend yield. However, dividend paying stocks underperform on average by -0.32% and -0.31% a month during recoveries and slowdowns for every one percent increase in dividend yield. This is in line with formulated expectations. When performing the t-test, the dividend premiums all significantly differ from the average sample dividend premium. Furthermore, a Wald-test is performed to check whether the relative dividend premium during both slowdowns and recessions, as well during recoveries and expansions, are significantly different from each other. Results are presented in Table 4. No statistical difference between the relative dividend premiums during expansions and recoveries was found. The difference between the relative dividend premium during slowdowns and recoveries is statistically significant at the 1% level though.
Table 3 - continued
Panel C: Relative market returns ( )
Regression Slowdown Recession Recovery Expansion
(1) -0.0153* -0.0085* 0.0108* 0.0025*
(3) -0.0167* -0.0095* 0.0097* 0.0014*
(4) -0.0162* -0.0089* 0.0089* 0.0007*
23
Table 4 - Wald Tests
A Wald Test is performed to check whether there is a significant statistical difference between the relative dividend premiums of (1) slowdowns and recessions, and (2) recoveries and expansions. Panel A shows the results from the Wald Test. A statistical significant finding indicates a difference between the relative dividend premiums during (1) slowdowns and recessions, and/or during (2) recoveries and expansions.
Panel A : Results Wald Test
Regression Slowdown:Recession Recovery:Expansion
(1) 0.0061* 0.8511
(3) 0.0069* 0.7098
(4) 0.0091* 0.8705
*Significant at the 1% level
The relative dividend premiums across the business cycle found for regressions (1), (3) and (4) are similar, indicating that it is not firm quality (as proxied by ROE) and/or higher available free cash flow that is responsible for the results found. Two firm characteristics often associated with dividend paying firms. The regressions control for size as well, empirically shown to be related to the propensity of paying out dividends. If firm fundamentals are not able to explain the time-varying dividend premium, than what can?
The dividend premium is larger during recessions than during slowdowns, even though stock returns during slowdowns are worse than during recessions. Also, the dividend premium found during recoveries and expansions is about the same size, even though stock returns are significantly larger during recoveries than during expansions. This does not support the findings of Fuller and Goldstein (2011), who find that large negative (positive) market movements result in a higher (lower) dividend premium. Findings are not a 100% contradiction per se though. Stock returns during slowdowns may be lower on average, but smoother. Whereas stock returns during recessions might have more variability, resulting in larger price movements. Although farfetched, it is not impossible that this is the case.
24
Davis and Madura (2012) conclude that during the financial crisis of 2008 investor risk-aversion indeed led to a herd mentality to quality. A mirror image occurs for more risky firms during booms. Perez-Quiros and Timmermann (2000) indeed find that smaller, more risky firms outperform during periods of low distress.
Furthermore, my findings are fully in line with the findings of Massa and Simonov (2005). They show that, on a yearly horizon, investors that have had good returns previously become more risk seeking - the house money effect. On the other hand, investors which have had bad returns previously become more risk averse – a snakebite reaction. Investors have to lose money before their risk attitudes change. Therefore investors might be more risk averse during recessions than during slowdowns, even though losses are smaller during recessions. Also, investors adjust their level of risk seeking upwards only slowly after previously good results, hence causing the dividend premium to be lower during expansions than during recoveries, even though market returns during a recovery are larger.
4.2 Robustness checks
Findings for regressions (1), (3) and (4) are consistent. The dividend premium indeed seems to be time-varying across different states of the business cycle. In this section some robustness checks are performed to challenge and/or strengthen previous findings. Results can be found in Table 5.8
During the estimation of regression (4), the sign found for the coefficient of ROE was counterintuitive. Therefore I replace ROE by another variable: Net Profit (NP). Both are measures for firm quality. Results from regression (7) show that the coefficient for NP does have the appropriate sign. The coefficient for NP is not significant during the estimation of step 1, but is significant when estimating step 2. Profitability seems to be more important during a slowdown. Another indication that firm quality is appreciated more during times of distress. Results regarding the dividend premium are not different from regression (4).
8 Full robustness check regression coefficient estimations are available on demand:
25
Table 5 - Robustness checks
Regressions are run with small deviations in variables to determine: (1) the relative stock performance over the business cycle and (2) the interaction effect between dividend yield and the state of the economic business cycle. A t-test is executed to determine whether there is a statistical difference between (1) the average sample stock returns and the stock returns over the different states of the business cycle, and (2) the sample dividend premium and the dividend premium found for the different states of the business cycle. The following regressions are estimated:
(7) As in (4) but the variable is replaced by Net Profit, .
(8) As in (1) but now the variable is estimated using NASDAQ market returns instead of S&P 500
market returns
(9) As in (1), but now the dummy variables, , are created using OECD data for leading indicators from the G7,
instead of US data.
(10) As in (1), but now the variable is replaced by . A dividend dummy variable with value 0 or 1
indicating either a non dividend-paying or a dividend-paying firm respectively.
(11) As in (1), but now the dummy variables, , are replaced by a dummy which represents up and down
market months in the S&P 500, .
(12) As in (1), but now replacing both the variable by , and the dummy variables by .
(13) As in (1) but now dividing the sample in two time frames. t=1: 1988m03-1999m01 and t=2: 1999m02-2012m12.
Panel A shows the relative dividend premiums across the different states of the business cycle. A positive (negative) number represents a higher (lower) dividend premium during that state of the business cycle as opposed to the samples average dividend premium. Regression (11) and (12) use an up and down market dummy variable. Their values represent the relative dividend premium during either a down or an up market. The significance of the relative dividend premiums during each state of the cycle is tested as indicated in (6).
Panel B shows the relative market returns across the different states of the business cycle. A positive (negative) number represents a higher (lower) market return during that state of the business cycle as opposed to the samples average market returns. Regression (11) and (12) use an up and down market dummy variable. Their values represent the relative dividend premium during either a down or an up market. The significance of the relative market returns is tested as in (5).
Panel A: Relative dividend premiums ( )
Regression Slowdown Recession Recovery Expansion
(7) 0.0016* 0.0025* -0.0032* -0.0031* (8) 0.0014* 0.0022* -0.0028* -0.0023* (9) 0.0025* 0.0014* -0.0024* -0.0028* (10) 0.0105* 0.0167* -0.0158* -0.0106* (11) (Down) 0.0040* (Up) -0.0029* (12) (Down) 0.0198* (Up) -0.0129* (13)(t=1) 0.0013* 0.0010* -0.0010* -0.0013* (13)(t=2) 0.0013* 0.0032* -0.0042* -0.0040*
26
Table 5 - continued
Panel B: Relative market returns ( )
Regression Slowdown Recession Recovery Expansion
(7) -0.0150* -0.0081* 0.0112* 0.0029* (8) -0.0063* -0.0093* 0.0134* 0.0084* (9) -0.0174* -0.0066* 0.0066* 0.0007* (10) -0.0138* -0.0112* 0.0171* 0.0078* (11) (Down) -0.0056* (Up) 0.0019* (12) (Down) -0.0091* (Up) 0.0059* (13)(t=1) -0.0102* -0.0072* 0.0044* 0.0039* (13)(t=2) -0.0111* -0.0031* 0.0195* 0.0062*
* Significant at the 1% level
A large part of the stock sample consists of constituents of the NASDAQ Composite. Therefore I check whether results differ from regression (1) when changing the benchmark market returns. Instead of using the S&P 500 market returns, regression (8) is estimated using the NASDAQ market returns for calculating . The time period, November 2003 till December 2012, is shorter and contains 110 monthly observations. There was no TRI available before this period. Results do not differ from the findings of regression (1).
The CLI dummy variable is based upon US data. Since the United States is a very large and open economy, it might be possible that a variable based upon the leading indicators for the largest seven economies is more appropriate. Therefore a G7 CLI dummy variable is created, which consists of 82 slowdowns, 65 recessions, 64 recoveries and 88 expansions. Results for (9) do differ slightly from earlier findings. The dividend premium is now largest during slowdowns instead of during recessions. This coincides with the largest negative returns delivering the largest dividend premium as was found in Fuller and Goldstein (2011), but the dividend premium is lower during expansions than during recoveries, contradicting the aforementioned finding. I personally believe these results are most likely caused by a more noisy proxy for market expectations. The G7 CLI data might simply not be the best estimator for US market expectations.
27
value of 1, when a company is determined to be dividend paying. The determination of whether a firm is dividend paying or not is the same as explained in the methodology section. No cross-section fixed effects were found, but there were period random effects found to be present and regression (10) is estimated accordingly.
The sign found for the coefficient of sales is now positive, which is unexpected. Also in (10) is less negative than in (1). This might very well be caused by the positive correlation found for size and stock returns in this regression. As I have mentioned before, firm size and the propensity to pay dividends are highly correlated. Furthermore, the time-varying deviations found in the dividend premium across the business cycle are larger than in (1). This makes sense. The average sample dividend yield for dividend payers is 2.7%, where the maximum value of the dividend dummy variable is 1. Therefore the expected relative dividend premium is 2.7 times larger in (10) than in (1).
But all relative premiums from (1) multiplied by 2.7 are significantly smaller than the relative premiums of (10). This indicates that higher yield stocks underperform relatively to lower yield stocks during times of distress and outperform lower yield stocks during booms. This is in accordance with Fuller and Blau (2010), who state that it are firms with the highest yields that had mediocre prior performance. I pose that these firms are also more likely to have larger variability in future returns and are therefore more risky. A firm characteristic appreciated more during booms. Although complying with Fuller and Blau (2010), these findings contradict the findings of Fuller and Goldstein (2011), who find that, after separating dividend paying stocks in five quintiles, stocks in the quintile with the highest yields are rewarded with relatively higher returns during down markets, implying that the dividend premium is higher for firms with the largest yields during times of distress. Another explanation for the results found can be that the regression is estimated using period random effects instead of cross-section fixed effects.
28
down market months. There are more up market months than was predicted by the CLI dummy variable (155 up and 144 down market months predicted if you expect up markets during recoveries and expansions, and down markets during slowdowns and recessions). The regression is estimated with period random effects instead of cross-section fixed effects. This results, again, in the coefficient for sales becoming positive and the sample dividend premium becoming less negative.
The difference relative to the sample dividend premium is more severe during down than during up markets months. This is simply due to the fact that there are more up market months. Results regarding the time-varying dividend premium do not seem to be different from the findings in (1). The premiums found during both up and down market months are somewhat larger in value than found when using the CLI variables. This makes perfect sense, since hindsight is used here. When the market indeed went up, the dividend premium was lower and vice versa.
Regression (12) is run after replacing both the CLI dummy variables as well as the dividend yield variable by the dividend dummy variable as in (10) and the up and down market month dummy as in (11). Period random effects are used and again, the coefficient for sales becomes positive. Results are similar as found in (10). The expected relative dividend premiums are 2.7 times larger in (12) than in (11). But both relative premiums found in (11) multiplied by 2.7 are significantly smaller than the relative premiums found in (12). This might be an indication that higher yield stocks underperform relatively to lower yield stocks during times of distress and outperform lower yield stocks during booms. Investors seem to place a relatively higher value on quality during down than during up market months. Both regressions (11) and (12) are run using period random effects, thereby eliminating the explanation of a differing methodology. One remark to be made here is that the relative market returns in (12) during either up and down markets are also larger than the relative market returns during (11). Therefore one should be careful giving too much weight to these findings. There seems to be no conclusive evidence supporting either Fuller and Blau (2010) or Fuller and Goldstein (2011).
29
differences are most likely caused both by investors changing their preferences or the fact that there is less information available during the earlier years of the sample (or simply biased towards larger firms). Period one dates from March 1988 till January 1999 and includes 131 monthly observations and 1400 cross-sectional observations. Period 2 dates from February 1999 till December 2012 and includes 167 monthly observations and 2625 cross-sectional observations. Both regressions provide evidence for a time-varying dividend premium across the business cycle and give similar results, although the evidence is more prominent during period 2.
5. Summary and conclusion
The goal of this paper was to investigate whether or not the dividend premium is time-varying as was first suggested by Baker and Wurgler (2004a). They suggested that the demand for dividend paying stocks was dependent on investor sentiment, whereas Fuller and Goldstein (2011) relate the size of an up or down market towards the height of the dividend premium. My results indicate that the dividend premium differs depending on the state of the business cycle. The dividend premium is higher during slowdowns and recessions, where it is lower during periods of recovery and expansion. Results indicate that it is not, on average, the size of a market up or down movement that has the largest impact on the dividend premium, but the state of the business cycle. This contradicts with the findings of Fuller and Goldstein (2011), but seems to be in line with the prediction of Baker and Wurgler (2004a). Namely, that investor sentiment is responsible for the height of the dividend premium. Negative sentiment occurs after slowdowns and recessions (down markets), whereas investors tend to become more risk seeking during recoveries and expansions (up markets). Massa and Simonov (2005) show that, on a yearly horizon, investors that have had previously good returns become more risk seeking, whereas investors who previously have had bad returns become more risk averse. This is an explanation for the higher (lower) dividend premium found during recessions (expansions) than during slowdowns (recoveries), even though returns are lower (higher).
30
dividend premium, but rather the payment of dividends itself. This is in accordance with Fuller and Goldstein (2011) who could not find an explanation for the outperformance of dividend paying stocks during down markets based on firm fundamentals.
Further, the results hold for both the use of the CLI dummy variables as well as the use of an up and down market month dummy variable, and the use of a dividend dummy variable. Also, results hold for different time periods, although results are stronger in later years. I also find indications (although not conclusive) that lower yield stocks outperform higher yield stocks during times of distress and vice versa during booms. I hypothesize that is due to an additional risk factor placed upon higher yield stocks. According to Fuller and Blau (2010) higher yield stocks tend to have mediocre prior performance. Extending this to the future, they also have higher variability in future earnings and are therefore more risky. Jain (2007) finds that relatively lower-taxed institutional investors tend to prefer lower yield stocks. This makes sense if you indeed believe that these lower yield stocks incorporate lower risks. Institutional investors do not only care about upward potential, but also very much about downside risk. Taking this altogether, I suggest that it is investor’s perceived risk and perceived quality of dividend paying stocks that is (at least partly) responsible for results found.
In order to capitalize on my findings you should be a long term investor. One that plans ahead and keeps track of the macro-economic environment. When using Tactical Asset Allocation (TAA) my findings might help an investor time their shift towards less and more risky investments. Brocato and Steed (1998) show that portfolio weights should be modified to accommodate cyclical shifts in the macro-economic environment if mean variance efficiency is to be maintained over the business cycle. Therefore it might be interesting to investigate whether other risk attributes and firm characteristics might have time-varying premiums across the business cycle. For example, Perez-Quiros and Timmermann (2000) find that smaller, more risky firms tend to outperform during booms and returns deteriorate rather rapidly during downturns. Furthermore, stocks with relatively higher leverage are expected to suffer more pronounced price declines during times of distress as found by George and Hwang (2010).
31
32
References
Ali. A. and Urcan, O., 2012. Dividend increases and future earnings, Asia-Pacific Journal of
Accounting and Economics, 19, 1, 12-25.
Allen, F., Bernardo, A.E. and Welch, I., 2000. A theory of dividends based on tax clienteles, Journal of Finance, 55, 2499-2536.
Baker, M. and Wurgler, J., 2004a. A catering theory of dividends, Journal of Finance, 59, 1125-1165.
Baker, M. and Wurgler, J., 2004b. Appearing and disappearing dividends: the link to catering incentives, Journal of Financial Economics, 73, 271-288.
Ball, R., Brown, P., Finn, F.J. and Officer, R.R., 1979. Dividends and the value of the firm: evidence from the Australian equity market, Australian Journal of Management, 4, 13, 13-26.
Barberis, N. and Shleifer, A., 2003. Style investing, Journal of Financial Economics, 68, 161-199.
Benartzi, S., Michaely, R. and Thaler, R., 1997. Do changes in dividends signal the future or the past?, Journal of Finance, 52, 1007-1034.
Bernanke, B. and Gertler, M. 1989. Agency costs, net worth, and business fluctuations,
American Economic Review, 79, 14-31.
Black, F., 1976. The dividend puzzle, Journal of Portfolio Management, Winter.
Black, F. and Scholes, M., 1974. The effects of dividend yield and dividend policy on common stock prices and returns, Journal of Financial Economics, 1, 1-22.
Blume, M.E., 1980. Stock returns and dividend yields: some more evidence, Review of
Economics and Statistics, 62, 4, 567-577.
Brocato., J. and Steed, S., 1998. Optimal asset allocation over the business cycle, Financial
33
Brooks, C., 2008. Introductory econometrics for finance, 2nd edition, Cambridge
University Press.
Christie, W., 1990. Dividend yield and expected returns: the zero-dividend puzzle,
Journal of Financial Economics, 28, 95-125.
Cooper, M.J., Gulen, H. and Rau, P.R., 2005. Changing names with style: mutual fund name changes and their effects on fund flows, Journal of Finance, 60, 2825-2858.
Davis, S.M. and Madura, J., 2012. How the shift to quality distinguished winners from losers during the financial crisis, Journal of Behavioral Finance, 13, 2, 81-92.
DeAngelo, H., DeAngelo, L. and Skinner, D. J., 2004. Are dividends disappearing? Dividend concentration and the consolidation of earnings, Journal of Financial
Economics, 72, 425-456.
DeAngelo, H., DeAngelo, L. and Stulz, R.M., 2006. Dividend policy and the earned/contributed capital mix: a test of the life-cycle theory, Journal of Financial
Economics, 81, 2, 227-254.
Denis, D.J., Denis, D.K. and Sarin, A., 1994. The information content of dividend changes: cash flow signaling, overinvestment, and dividend clienteles, Journal of Financial and
Quantitative Analysis, 29, 567-587.
Denis, D.J. and Osobov, I., 2008. Why do firms pay dividends? International evidence on the determinants of dividend policy, Journal of Financial Economics, 89, 62-82.
Easterbrook, F.H., 1984. Two-agency cost explanations of dividends, American Economic
Review, 74, 4, 650-659.
Fama, E.F. and French, K.R., 1993. Common risk factors in the returns on stocks and bonds, Journal of Financial Economics, 33, 3-56.
Fuller, K.P. and Blau, B.M., 2010. Signaling, free cash flow and “nonmonotonic“ dividends, Financial Review, 45, 21-56.
Fuller, K.P. and Goldstein, M.A., 2011. Do dividends matter more in declining markets?,
34
George, T.J. and Hwang, C., 2010. A resolution of the distress risk and everage pebbles in the cross section of stock returns, Journal of Financial Economics, 96, 56-79.
Grinstein, Y. and Michaely, R., 2005. Institutional holdings and payout policy, Journal of
Finance, 60, 3, 1389-1426.
Grullon, G., Michaely, R., Benartzi, S. and Thaler, R., 2005. Dividend changes do not signal changes in future profitability, Journal of Business, 78, 5, 1659-1682.
Dynamics and Control, 32, 2939-2970.Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers, American Economic Review, 76, 2, 323-329.
Jain, R., 2007. Institutional and individual investor preferences for dividends and share repurchases, Journal of Economics and Business, 59, 406-429.
Lang, L. and Litzenberger, R.H., 1989. Dividend announcements: cash flow signaling vs. free cash flow hypothesis?, Journal of Financial Economics, 24, 181-191.
Latham, S.F. and Braun, M.R., 2008. The performance implications of financial slack during economic recession and recovery: observations from the software industry (2001-2003), Journal of Managerial Issues, 20, 1, 30-50.
Li, W. and Lie, E., 2006. Dividend changes and catering incentives, Journal of Financial
Economics, 80, 293-308.
Litzenberger, R.H. and Ramaswamy, K., 1979. The effects of personal taxes and dividends on capital asset prices: theory and empirical evidence, Journal of Financial
Economics, 7, 163-195.
Litzenberger, R.H. and Ramaswamy, K., 1982. The effects of dividends on common stock prices tax effects or information effects?, Journal of Finance, 37, 2, 429-443.
Massa, M. and Simonov, A., 2005. Behavioral biases and investment, Review of Finance, 9, 483-507.
