liquidity and dividends

Upload: 10164

Post on 07-Apr-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/6/2019 Liquidity and Dividends

    1/47

    Stock Market Liquidity and Firm Dividend Policy

    Suman BanerjeeA. B. Freeman School of Business

    Tulane University7 McAlister Drive

    New Orleans, LA [email protected]

    (504 ) 865 -5558

    Vladimir A. GatchevA. B. Freeman School of Business

    Tulane University7 McAlister Drive

    New Orleans, LA [email protected]

    (504 ) 865 -5534

    Paul A. SpindtA. B. Freeman School of Business

    Tulane University7 McAlister Drive

    New Orleans, LA [email protected]

    (504 ) 865 -5413

    February 2005

    JEL classi cation : G35 , G33Keywords : Dividends; Payout policy; Liquidity; Trading friction

    Please address correspondence to Vladimir A. Gatchev, A. B. Freeman School of Business, Tulane Uni-versity, 7 McAlister Drive, New Orleans, LA 70118. Tel. (504) 865-5534; E-mail: [email protected].

    For helpful discussions and comments we thank Vladimir Atanasov, David Blackwell, Michael Brandt, ValentinDimitrov, Chitru Fernando, John Graham, Joel Horowitz, Kose John, David Lesmond, Neal Maroney, DavidMauer, Tom Noe, Nagpurnanand Prabhala, Michael Rebello, Bill Reese, Ramana Sonti, Venkat Subramaniamand Sheri Tice. The paper has also bene ted from comments by seminar participants at Tulane University,University of New Orleans, Southern Methodist University, Kennesaw State University, University of Mis-souri, Georgia State University, University of Central Florida, Auburn University, University of Cincinnati,University of Oklahoma, the 2003 EFMA meetings, and the 2004 WFA meetings. We would like to thankMichael Lemmon in particular, our discussant at the 2004 WFA meetings, for his comments and insights. Weremain responsible for any errors.

  • 8/6/2019 Liquidity and Dividends

    2/47

    Stock Market Liquidity and Firm Dividend Policy

    ABSTRACT

    We provide evidence of a link between rm dividend policy and stock market liquidity. Inthe cross-section, owners of less (more) liquid common stock are more (less) likely to receive

    cash dividends. Over time, the notable increase in US stock market liquidity explains most of

    the declining propensity of rms to pay dividends documented by Fama and French ( 2001 ).

    We further show that past liquidity is an important determinant of dividend initiations and

    omissions for individual rms. Extending our analysis, we nd evidence that sensitivity of

    rm value to innovations in aggregate liquidity declines after dividend initiations.

    JEL classi cation : G35 , G33Keywords : Dividends; Payout policy; Liquidity; Trading friction

  • 8/6/2019 Liquidity and Dividends

    3/47

    Firms dividend policies continue to puzzle nancial researchers. In this paper, we argue

    that investor demand for stocks paying cash dividends is positively related to the trading

    friction that investors face when creating homemade dividends. We further hypothesize that

    the likelihood a rm will pay cash dividends is positively related to investor demand for

    dividend payments and therefore inversely related to the market liquidity of the rms stock.

    Examining the empirical evidence, we nd strong support for our hypothesis.

    In their seminal work, Miller and Modigliani ( 1961 ) formally developed the dividend

    irrelevance hypothesis. In perfect capital markets populated by rational investors, a rms

    value is solely a function of the

    rms investment opportunities and is independent of the rms payout policy. A large body of theoretical work has tried to evaluate the importance

    that managers and investors attach to dividend policy in light of the irrelevance proposition.

    The starting point of these studies is to question some of the assumptions that characterize

    the perfect capital markets hypothesized by Miller and Modigliani. 1

    One notable assumption of the dividend irrelevance proposition, and one central to this

    paper, is that trading is frictionless. In perfect markets, investors can instantaneously invest

    or liquidate their investment in any stock without incurring any direct or indirect costs of trading and without changing the price of the underlying security. In markets with no trading

    friction, rational investors with liquidity needs can create homemade dividends at no cost by

    selling an appropriate amount of their holdings in the rm. As a result, they will be indi ff erent

    between receiving a dollar of dividend and selling a dollars worth of their investment.

    In markets with trading friction, stocks that pay cash dividends allow investors to satisfy

    their liquidity needs with little or no trading in the stock and thus enable them to avoid

    trading friction. As a result, investors with current or anticipated future liquidity needs

    may have a preference for dividend paying stocks. This preference will be positively related

    to the level of trading friction so that higher (lower) trading friction will lead to higher

    (lower) demand for cash dividends relative to homemade dividends. Dong, Robinson, and1 Allen and Michaely (2001) provide a survey on the literature.

    1

  • 8/6/2019 Liquidity and Dividends

    4/47

    Veld ( 2003 ) present survey evidence that retail investors want dividends, partly because their

    costs of cashing in dividends are lower than the transaction costs involved in selling shares. 2

    It is important to address the question of how investor demand for dividends translates

    into actual dividend policy. On the one hand, existing literature argues that stock market

    liquidity a ff ects the valuation of rms both in the cross-section and through time. 3 In this

    literature, stocks with higher liquidity levels (i.e., lower trading friction) trade at a premium

    and have lower expected returns relative to stocks with lower liquidity levels (i.e., higher

    trading friction). Firms, however, can pay cash dividends, reduce investor dependence on

    the liquidity of the market, and therefore raise their valuations an option more valuablefor rms with higher discount rates due to lower liquidity levels. Indeed, Baker and Wurgler

    (2004 a, 2004 b) present signi cant evidence that rms consider valuation e ff ects when choosing

    a dividend policy. On the other hand, it is also possible that investors directly enforce the

    desired dividend policy, as suggested by La Porta, Lopez-de-Silanes, Shleifer, and Vishny

    (2000 ).

    While the possibility of a link between stock market liquidity and the dividend policy of

    the rm dates at least back to Miller and Modigliani ( 1961 ), current literature provides littledirect empirical evidence on that issue. Some indirect evidence, however, is consistent with

    our hypothesis. For example, Long ( 1978 ) documents that between 1956 and 1976 the cash

    dividend class of shares of Citizens Utilities Company on average sold at a premium to the

    stock dividend class. Subsequent work by Poterba ( 1986 ) shows that the two classes of shares

    trade at similar prices for the 1976 -1984 period. The disappearing premium on the cash

    dividend shares is consistent with an increase in the liquidity of the market in that period.

    2 Dividend reinvestment, if needed, can result in additional trading costs for investors. In 1954 NYSEimplemented the Monthly Investment Plan (MIP) that, among other things, allowed reinvestment of dividends.This program was terminated in 1976. Meanwhile, in 1968, Citibank (then First National City Bank of NewYork) introduced the rst dividend reinvestment program (DRIP). DRIPs increased in popularity and sincethe mid-1970s most rms have such programs (see Davey (1976) and Carlson (1996)). One of the majorobjectives of such programs is to allow investors to reinvest dividends.

    3 See, for example, Amihud and Mendelson (1986), Brennan and Subrahmanyam (1996), Brennan, Chordia,and Subrahmanyam (1998), Amihud (2002), Jones (2002), and Pstor and Stambaugh (2003).

    2

  • 8/6/2019 Liquidity and Dividends

    5/47

    Nevertheless, the question of whether stock market liquidity has an incremental impact

    on the dividend policy of the rm remains largely an empirical one and its investigation is the

    focus of the current study. We perform our analysis while taking into consideration rm size,

    pro tability, and growth opportunities. The necessity to control for these variables arises for

    at least two reasons. First, their use as determinants of dividend policy is consistent with the

    role of dividends in controlling the agency costs of free cash ow (Easterbrook, 1984 ; Jensen,

    1986 ) and with a pecking-order model where rms avoid issuing securities due to asymmetric

    information costs (Myers and Majluf, 1984 ; Myers 1984 ) and other otation costs. The

    empirical importance of these variables for the

    rms decision to pay dividends is examinedin Fama and French ( 2001 ) and is further con rmed in our study. Second, the liquidity of the

    rms common stock can also be related to the size, pro tability, and growth opportunities

    of the rm. Therefore, it is important to examine the link between rm dividend policy

    and liquidity after controlling for the possibility of such a relation. For the remainder of the

    paper, we refer to these variables as rm characteristics and to their collective explanatory

    power over the dividend policy of the rm as the rms ability to pay dividends.

    The main results of the paper can be summarized as follows. First, we document that rms with less liquid markets (characterized by low trading activity, high proportion of zero

    trading days, and high price impact of order- ow) are more likely to pay dividends. These

    results persist after we control for the characteristics of the rm discussed above and provide

    direct support for our hypothesis. Second, we present evidence that market liquidity and

    rm likelihood to pay dividends are negatively related over time. The past four decades are

    characterized by declining commission rates, declining bid-ask spreads, and a ten-fold increase

    in market activity measures frequently used to quantify the liquidity of the stock market.

    When we apply our 1963 -1977 estimates to predict the proportion of dividend payers in more

    recent years, we nd that increased market liquidity explains most of the lower propensity of

    rms to pay dividends documented by Fama and French ( 2001 ). Furthermore, the predictive

    accuracy of a model that controls for stock market liquidity, versus a model that does not, is

    3

  • 8/6/2019 Liquidity and Dividends

    6/47

    more pronounced for rms more likely to pay dividends based on their size, pro tability, and

    growth opportunities (i.e., rms with higher ability to pay) and for rms with more liquid

    stocks. 4

    We further address the question of whether dividend policy determines stock market

    liquidity and not vice versa. We now perform our analysis conditional on the past dividend

    policy of rms while at the same time we use a historic measure of liquidity rather than a

    contemporaneous one. We nd that past year market liquidity is an important determinant of

    dividend initiations and of dividend omissions. Less (more) liquid rms that have never paid

    dividends are more (less) likely to initiate dividend payments. Similarly, less (more) liquid rms that have paid dividends for the past ve years are more (less) likely to continue paying

    dividends in the future. For dividend initiations, the predictive accuracy of a model that

    controls for market liquidity, versus a model that does not, is higher and the improvement

    is comparable to our results for all rms. For dividend omissions, stock market liquidity

    has no economic power in explaining the dividend omission rates of rms. In fact, we do

    not nd lower propensity to pay dividends for rms with long history of dividend payments.

    Models based on rms ability to pay dividends and models based on ability and stock marketliquidity equally well explain more recent dividend omission rates of rms. In other words,

    we do not observe lower propensity to pay (i.e., higher propensity to omit dividends) for

    dividend paying rms.

    Up to this point of our discussion we have focused on the relation between dividend

    policy and liquidity at the rm level. Recent studies, however, present evidence of a common

    liquidity factor across rms. Chordia, Roll, and Subrahmanyam ( 2000 ), for example, nd that

    several measures of liquidity co-move with market- and industry-wide liquidity. Pstor and4 DeAngelo, DeAngelo, and Skinner ( 2004 ) nd that the reduction in payers documented by Fama and

    French ( 2001 ) occurs almost entirely among rms that paid or would have paid very small dividends. Thisevidence is consistent with our hypothesis since in more liquid markets investors can hold portfolios with morestocks and replicate small dividend payouts by combining rms with high payouts and rms with no payoutsin their portfolios. This would e ff ecively reduce the demand for low dividend payout stocks.

    4

  • 8/6/2019 Liquidity and Dividends

    7/47

    Stambaugh ( 2003 ) propose that assets with high positive sensitivity of returns to aggregate

    liquidity result in disproportionate decrease of investor welfare when aggregate liquidity is

    low. They nd signi cant evidence that investors price this liquidity risk so that stocks with

    high sensitivities of returns to aggregate liquidity have higher expected returns than stocks

    with low or negative sensitivities.

    Extending our previous arguments, we now suggest that the demand of investors for

    dividend paying stocks, and thus the value of such stocks relative to non-paying stocks, is

    higher in states characterized by low aggregate liquidity. We therefore expect that dividend

    initiating

    rms will reduce their return sensitivity to innovations in aggregate liquidity. Webuild upon the work of Pstor and Stambaugh ( 2003 ) and indeed nd that, after rms

    initiate dividend payments, their stock returns become less sensitive to aggregate liquidity.

    This result further suggests that investors, when valuing rms, view cash dividends and stock

    market liquidity as substitutes.

    The rest of the paper is organized as follows. Section I describes the data and the variables

    of this study. Section II provides our cross-sectional results. Section III outlines the changes

    in the qualities of US security markets for the period of 1963 -2003 and the changes in rmdividend policy. Section IV investigates the e ff ectiveness of liquidity in explaining the changes

    in rm dividend policy over time. Section V reports separate results for past payers and past

    non-payers. Section VI describes our robustness tests. Section VII examines the changes in

    systematic liquidity risk around dividend initiations, and Section VIII concludes.

    5

  • 8/6/2019 Liquidity and Dividends

    8/47

    I. Sample and Variables

    A. Sample

    Our study covers NYSE and AMEX rms for the years from 1963 to 2003 .5 We gather data

    from the COMPUSTAT annual les, and the Center for Research in Security Prices (CRSP)

    monthly and daily les. We exclude rms with CRSP Standard Industrial Classi cation

    (SIC) codes between 6000 -6999 ( nancials) and between 4900 -4949 (utilities) and restrict our

    main sample to rms with publicly traded common stock with CRSP share codes of 10 or 11 .

    Our main sample consists of all rms for which we can obtain the earnings-to-assets ratio,

    the market capitalization, the market-to-book ratio, the growth in assets from the previous

    year, and share turnover. Data requirements on additional variables used in some of the tests

    dictate the actual sample sizes of these tests.

    B. Variables

    In this section we present and motivate the variables that we use in our empirical tests.

    The precise computation of these variables is outlined in Table I.

    A rm is de ned as a dividend payer in year t whenever COMPUSTAT reports positive

    dividends per share for scal year t . Our results, however, do not change if we use CRSP

    data to identify dividend-paying rms by comparing returns including distributions to returns

    excluding distributions.

    The rst set of variables that we use to explain the dividend decision of rms is based

    on the size, pro tability, and growth opportunities of the rm. We construct these variables

    as in Fama and French ( 2001 ). For a given year t and for every rm i the measure of rm

    size is equal to the percentage of NYSE rms with market capitalization lower than the

    market capitalization of rm i . The rms market capitalization for year t is equal to the5 Nasdaq trading volume is overstated relative to NYSE and AMEX trading volume and this does not allow

    us to use all rms. However, we have performed our analysis also separately for Nasdaq stocks and obtainsimilar results, which are available on request.

    6

  • 8/6/2019 Liquidity and Dividends

    9/47

    product of its share price and shares outstanding for June of year t as reported in the CRSP

    monthly les. This measure of rm size is constructed under the assumption that the NYSE

    market capitalization percentiles have constant implications for the dividend policy of the

    rm throughout the examined period. The pro tability and growth opportunities proxies are

    calculated using COMPUSTAT data for scal year t . Firm pro tability for year t we measure

    as earnings divided by assets for that year (E t /A t ). To proxy for growth opportunities we

    use the value-to-assets ratio of the rm for year t (V t /A t ) and the proportionate change in

    assets for year t (dA t /A t ).

    The second set of variables that we use to explain the dividend decision of

    rms is aimed atcapturing the market liquidity of the rms common stock. It is unlikely that a single empirical

    measure can capture all aspects of market liquidity. Therefore, in our cross-sectional analysis

    we use several proxies for stock market liquidity. Three of the proxies are directly related

    to the trading activity in a rms common stock, and one is related to the price impact of

    trades.

    The trading activity in the stock of the rm has both theoretical as well as empirical appeal

    as a measure of liquidity. Constantinides ( 1986 ) shows that larger xed transaction costsbroaden the region of no transaction while Amihud and Mendelson ( 1986 ) develop a model

    where assets with higher bid-ask spreads have longer holding periods, thus lower trading

    activity. Atkins and Dyl ( 1997 ) provide empirical support for these models. Additionally,

    the combined evidence of Stoll ( 1978 ) and Stoll ( 2000 ) suggests that a measure of trading

    activity plays an important role in explaining the cross-sectional variation in bid-ask spreads

    both in historic and current data. Trading activity also has implications for the execution

    risk of an investor where rms with higher trading activity have lower execution risk, all else

    equal. Finally, trading activity may also have a more direct impact on investor demand for

    cash dividends. When there are economies of scale in trading, the marginal cost of creating

    homemade dividends is lower when investor trading activity is higher. As a result, investor

    demand for cash dividend paying stocks should decline when trading activity is high. Our rst

    7

  • 8/6/2019 Liquidity and Dividends

    10/47

    measure of trading activity is the annual share turnover the ratio of shares traded to shares

    outstanding for calendar year t from COMPUSTAT (TURN t ).6 Existing research has widely

    used share turnover as a proxy for liquidity (see, for example, Datar, Naik, and Radcli ff e

    (1998 ) and Chordia, Subrahmanyam, and Anshuman ( 2001 )). Because of its theoretical and

    empirical appeal, we use share turnover to proxy for liquidity when we analyze the relation

    between liquidity and dividend payers over time. In these tests we assume that share turnover

    has relatively constant implications for the dividend policy of rms over time. We construct

    two additional proxies for the trading activity in a stock using the annual traded dollar

    volume in the stock (DVOL t ) (Brennan, Chordia, and Subrahmanyam (1998

    ) and Chordia,Subrahmanyam, and Anshuman ( 2001 )) and the proportion of days with zero traded volume

    as an inverse measure of trading activity (NOTRD t ) (Glosten and Milgrom ( 1985 ), Kyle

    (1985 ), Constantinides ( 1986 ), Dumas and Luciano ( 1991 )).

    Our nal proxy for liquidity is the illiquidity ratio (ILLIQ t ). This measure, or its inverse

    (the Amivest measure of liquidity), is used in existing research to proxy for the depth of

    the market and the impact of order- ow on stock prices as analyzed by Kyle ( 1985 ).7 It is

    calculated as the average ratio of absolute daily return to daily dollar volume using data fromthe CRSP daily les.

    In order to ensure that outliers do not drive our results, we winsorize all variables based on

    their annual 0 .5th and 99 .5th percentiles, excluding the proxy for size, which by construction

    is bounded between 0 .00 and 1 .00 .

    [Insert Table I about Here]

    6 Using CRSP data gives similar results.7 See, for example, Amihud, Mendelson, and Lauterbach ( 1997 ), Amihud ( 2002 ), and the references therein.

    8

  • 8/6/2019 Liquidity and Dividends

    11/47

    II. Empirical Evidence in the Cross-section

    In this section we rst perform annual cross-sectional logistic regressions between 1963 and2003 to explain the dividend policy of the rm. We report the average coe ffi cient estimates

    for several time periods to assess the importance of the variables and their robustness over

    time.

    In Table II we present the average coe ffi cient estimates for di ff erent speci cations for three

    sub-periods ( 1963 -1977 , 1978 -1992 , and 1993 -2003 ).

    Panel A of Table II uses only share turnover to predict the probability of dividends. The

    results suggest that there is a signi cant negative relation between a rms stock market

    liquidity and its likelihood to pay dividends. This relationship is signi cant at the 0 .01 level

    for all examined sub-periods. Further investigation reveals that the impact of liquidity on

    dividends is nontrivial. The likelihood of a dividend for the average rm for the 1963 -1977

    period is 71 .50 percent. Our estimates suggest that one standard deviation increase (decrease)

    in liquidity leads to a decrease (increase) in this probability to 59 .55 (81 .04 ) percent. The

    additional analyses for 1978 -1992 and for 1993 -2003 also reveal signi cant sensitivities of

    dividends to the liquidity of the rms stock. Using the estimates for 1993 -2003 we nd that

    one standard deviation increase (decrease) in liquidity leads to a decrease (increase) in the

    average probability of 48 .12 to 41 .84 (54 .46 ) percent.

    We now turn to multivariate tests where we also control for several rm characteristics

    that existing research relates to rm dividend policy. As explanatory variables, we rst use

    proxies for rm size, rm pro tability, and rm growth opportunities. Results are presented

    in Panel B of Table II, columns ( 1), (4) and ( 7). The estimates are similar to the ones

    reported by Fama and French ( 2001 ). Larger and more pro table rms are more likely to

    pay dividends, while rms with higher growth opportunities are less likely to do so.

    We also add several measures of liquidity to the set of explanatory variables. The results

    with share turnover are presented in Panel B while the results with all other measures are

    9

  • 8/6/2019 Liquidity and Dividends

    12/47

    presented in Panel C of Table II. Columns ( 2), (5), and ( 8) of Panel B, Table II show that

    share turnover is again negatively and signi cantly (at the 0 .01 level) related to the likelihood

    of dividends for the three examined sub-periods. The 1963 -1977 coeffi cient estimate suggests

    that one standard deviation increase (decrease) in liquidity leads to a decrease (increase) in

    rm probability to pay dividends from 71 .49 percent to 58 .60 (81 .64 ) percent. The results

    for 1963 -1977 are very similar to the univariate results presented in Panel A. In later periods,

    however, the impact of liquidity on the likelihood of dividends is more signi cant when we

    control for the rm characteristics discussed above. For 1993 -2003 , for example, one standard

    deviation increase (decrease) in liquidity leads to a decrease (increase) in the probability of dividends from 48 .12 percent to 30 .04 (66 .70 ) percent. The increased importance of liquidity,

    after controlling for the growth opportunities of the rm, is not consistent with the idea that

    the impact of liquidity on rm dividend policy is mainly driven by a possible positive link

    between liquidity and growth opportunities in the cross-section.

    The tests that we perform above assume that liquidity has the same impact on the divi-

    dend decision of rms regardless of their characteristics. In general, this need not be the case.

    In particular, liquidity should be more relevant for the dividend decision of rms with higherability to pay dividends (i.e., large, pro table rms and rms with low growth opportuni-

    ties) since such rms have more exibility in their decision to pay or not to pay dividends.

    Alternatively, if it is prohibitively costly for the rm to provide dividends (e.g., small rms

    with no pro t and high growth opportunities), then stock market liquidity may have little

    or no eff ect on the dividend policy of the rm. To analyze whether liquidity has di ff erential

    impact on the dividend policy of rms depending on their rm characteristics, we allow for

    diff erent coeffi cient estimates of share turnover for two portfolios of rms. The rst portfolio

    consists of rms that are less likely to pay dividends (probability less than 70 percent) based

    on rm characteristics while the second portfolio consists of rms that are more likely (prob-

    ability more than 70 percent) to pay dividends based on rm characteristics. The estimates

    are given in columns ( 3), (6), and ( 9). We nd that the probability of dividend payments

    10

  • 8/6/2019 Liquidity and Dividends

    13/47

    is more sensitive to liquidity for the portfolio of rms that are more likely to pay based on

    rm characteristics. For 1963 -1977 , one percentage point increase in share turnover results

    in approximately 0 .33 percentage points decrease in the likelihood of dividend payments for

    rms that are less likely to pay based on rm characteristics. In contrast, one percentage

    point increase in share turnover for rms that are more likely to pay, based on their charac-

    teristics, results in a decrease in the likelihood to pay dividends by 0 .41 percentage points.

    This evidence is consistent with the notion that liquidity is more relevant for the dividend

    policy of rms with lower costs of issuing dividends.

    The rest of our liquidity measures are also related to the likelihood of dividends in linewith our hypothesis. Columns ( 1), (4), and ( 7) of Panel C, Table II show that rms with

    higher illiquidity ratios are more likely to pay dividends. Similarly, rms with lower trading

    volume and rms with higher proportion of days with no trading are also more likely to pay

    dividends. All of the examined relations are signi cant at the 0 .01 level for all sub-periods

    except the illiquidity ratio in the 1993 -2003 sub-period when it is signi cant at the 0 .05 level.

    We also analyze the impact of one standard deviation change in the di ff erent liquidity

    measures on the probability of dividends. In 1963 -1977 , for example, one standard deviationdecrease in the illiquidity ratio (i.e., increase in liquidity) results in a decline in the probability

    of dividends from 71 .41 percent to 38 .02 percent. In the same period, one standard deviation

    increase in dollar volume results in a decline in the probability of dividends from 71 .49 percent

    to 59 .61 percent while one standard deviation decrease in the proportion of days with no

    trading leads to a decline in the probability to pay from 71 .40 percent to 58 .44 percent.

    While all liquidity variables have a nontrivial impact on the probability of dividends, the

    illiquidity ratio (for the rst two periods) and share turnover (for the last period) have the

    most notable impact on the estimated probability of dividends.

    [Insert Table II about Here]

    In additional tests we have investigated the possibility that measurement error in value-

    11

  • 8/6/2019 Liquidity and Dividends

    14/47

    to-assets is driving our cross-sectional results (Erickson and Whited ( 2000 )). We have used

    an instrumental variable approach where as instruments for the value-to-assets of the rm we

    use research and development expense divided by assets and our results remain unchanged.

    Further, we have examined whether share turnover is capturing rm growth rather than

    liquidity. When we regress each year share turnover on value-to-assets, change in assets,

    and research and development to assets, the R 2 of the regression is around 5% and rarely

    above 10% . Further, we have also included additional proxies for growth to explain the rms

    decision to pay dividends in the attempt to better control for growth. These proxies include

    research and development-to-assets, change in assets in year t+1

    , and change in assets in yeart+ 2. The qualitative as well as the quantitative results pertaining to liquidity are robust to

    these tests and the magnitude of the coe ffi cients and the predictive ability are insensitive to

    these inclusions. Thus, even though we cannot rule out the above-mentioned problems, we

    do not nd any evidence that such problems are responsible for our ndings.

    The evidence presented in this section provides signi cant direct support for our hypoth-

    esis. After controlling for the impact of rm characteristics on rm dividend policy, we nd

    that holders of common stock with less liquid market are also more likely to receive dividends.This link is robust across the 41 years of data we have gathered and across di ff erent measures

    of market liquidity.

    III. Stock Market Liquidity and Dividend Payers over Time

    A. Changes in Market Liquidity from 1963 to 2003

    In this section we brie y outline the signi cant changes in the features of US stock mar-

    kets between 1963 and 2003 . Prior to 1975 the cartel on NYSE was characterized by xed

    commission rates, limited entry, and rules that prohibited price-cutting and that limited bro-

    kerage services per seat. Potential competition from other exchanges in the trading of NYSE

    listed stocks was reduced through additional regulations.

    12

  • 8/6/2019 Liquidity and Dividends

    15/47

    The Securities Acts Amendments of 1975 and Rule 19 b-3 became eff ective on May 1 ,

    1975 . These amendments resulted in the abolition of xed commission rates and mandated

    a national market system for securities in which competitive forces would play a much more

    signi cant role. The deregulation of the industry was accompanied by the emergence and

    expansion of discount brokers. In more recent years, the emergence of Internet brokers has

    led to even higher competition in the industry.

    By the rst quarter of 1976 , commissions of institutional rms have declined by 31 .6%

    (see Stoll ( 1979 )). The decline in overall commission rates continued for the remainder of the

    century. Evidence in Jones (2002

    ) shows a steady decline of average commission rates fromaround 0 .80% in the 1960 -1980 period to around 0 .10% in 2000 . Furthermore, Jones ( 2002 )

    nds that the average proportionate quoted bid-ask spread for the 30 Dow Jones Industrial

    Average (DJIA) stocks has declined from around 0 .60% in the 1960 -1980 period to around

    0 .20% by the end of the 1990 s. Combining commissions and bid-ask spread costs, Jones

    (2002 ) argues that total one-way costs have decreased from around 1 .30% in the 1960 -1980

    period to around 0 .20% in 2000 . The above outlined changes in the competitive environment

    of US security markets and the direct costs of trading were accompanied by a dramaticincrease in trading activity. Average (median) annual share turnover has increased from

    approximately 25% (17% ) in 1963 to around 101% (82% ) in 2000 . The decline in trading

    costs and the increase in trading activity suggest that the liquidity of the stock market has

    improved signi cantly over time.

    [Insert Figure I about Here]

    B. Stock Market Liquidity and Dividend Payers a Graphic Interpretation

    In this section we look at the time trends in stock market liquidity, measured by the

    average share turnover, and the proportion of dividend paying rms. Figure II shows that

    the steady increase in liquidity after 1978 is accompanied by a steady decline in the proportion

    13

  • 8/6/2019 Liquidity and Dividends

    16/47

    of dividend payers. We further nd that improved liquidity in the late 1960 s is also followed

    by a decline in the proportion of dividend payers.

    [Insert Figure II about Here]

    To further investigate this issue, we examine dividend decisions of rms based on their

    past dividend policies. Figure III, Panel A examines rms that did not pay dividends in

    year t -1 and reveals that dividend initiation (and resumption) rates are negatively related

    to the liquidity of the stock market. We also observe that changes in stock market liquidity

    and dividend initiation rates are closely aligned over time. When we analyze the dividend

    decisions of rms that pay dividends in year t -1 (Panel B), we see that dividend omission

    rates are higher after 1978 . However, this result does not seem to be as pronounced as the

    decline in dividend initiation rates for the sample of non-payers and former payers.

    [Insert Figure III about Here]

    Baker and Wurgler ( 2004 a) document that rms cater to the preferences of investors so

    that when dividend payers sell at a premium (discount) more

    rms tend to pay (not pay)dividends. Baker and Wurgler ( 2004 b) further use the catering theory to explain the declining

    propensity of rms to pay dividends. Our paper provides one possible explanation for the

    variation in the dividend premium over time - namely changes in the liquidity of the stock

    market. We use the dividend premium reported by Baker and Wurgler ( 2004 a) to compare

    its variation to the variation in stock market liquidity over time. Figure IV suggests that

    when liquidity levels are high dividend payers tend to sell at a discount while the opposite

    is true for low levels of liquidity. Also, we nd that the standardized returns to dividendinitiation announcements (again obtained from Baker and Wurgler ( 2004 a)) are low when

    stock market liquidity is high and vice versa. This evidence is consistent with our hypothesis

    that rms cater (possibly through market valuation) to the liquidity preferences of investors.

    [Insert Figure IV about Here]

    14

  • 8/6/2019 Liquidity and Dividends

    17/47

    The initial results of this section are consistent with the notion that stock market liquidity

    is related to the proportion of dividend payers over time and suggest that the declining

    proportion of dividend payers is related to improved market liquidity. The link seems to be

    more pronounced for rms that do not pay dividends in year t -1, i.e. those rms that are the

    main source of the decline of dividend payers.

    IV. Predicting Dividend Payers over Time

    A. The Predictive Ability of Liquidity

    In this section we further extend the tests of our hypothesis by analyzing the ability

    of improved market liquidity to predict the proportion of dividend payers for the period of

    1978 -2003 . In the base model we use the estimated coe ffi cients from column ( 1) of Table II,

    Panel B to predict the proportion of dividend payers based on the ability of the rm to pay

    dividends. In the second model we add share turnover as an explanatory variable so that we

    use the estimated coe ffi cients from column ( 2) of Table II, Panel B to predict the proportion

    of dividend payers. In the third model we use the coe ffi cient estimates from column ( 3) of Table II, Panel B where we allow for di ff erential impact of market liquidity on rms that are

    less able and rms that are more able to pay based on their size, pro tability and growth

    opportunities. The actual and the predicted proportions of dividend payers for the three

    models are presented in Table III. 8

    [Insert Table III about Here]

    When we analyze the predictive ability of the three models we see that the model that

    takes into account market liquidity signi cantly decreases the di ff erence between the predicted

    payers estimated from the base model and the actual payers. In 1998 -2003 , for example, the8 We have also used variability in earnings as another predictor of the probability to pay dividends. While

    the results in terms of prediction of payers over time improve, lower propensity of rms to pay is still evidentand liquidity still explains most of it.

    15

  • 8/6/2019 Liquidity and Dividends

    18/47

    diff erence between predicted and actual payers is 22 .20 percentage points using the original

    model based only on rm characteristics. This di ff erence declines to 6 .36 percentage points

    when liquidity is taken into account. There is even further improvement in predictive ability

    when we allow for diff erential impact of liquidity on the portfolios of rms with higher ability

    to pay and for rms with lower ability to pay - the predictive error in 1998 -2003 is reduced

    to 4 .69 percentage points.

    The results in this section provide evidence over time consistent with our hypothesis that

    the likelihood of dividend payments is negatively related to the liquidity of common stocks.

    Improved liquidity is one of the reasons (though not the only one) for the decline in dividendpaying rms for the past quarter century. In more recent work, Julio and Ikenberry ( 2004 )

    nd that during 2003 and 2004 the propensity of rms to pay dividends has increased. They

    attribute this increase to the maturing of rms, the desire of rms to signal con dence in the

    wake of corporate governance scandals, and the dividend tax cut of 2003 . It is important to

    note that our hypothesis and the hypotheses advanced by Julio and Ikenberry ( 2004 ) are not

    mutually exclusive. As a result, the evidence in Julio and Ikenberry ( 2004 ) is not inconsistent

    with our hypothesis.

    B. The Predictive Ability of Liquidity for Several Portfolios

    In this section we rst analyze the predictive ability of the di ff erent models for two portfo-

    lios based on the market liquidity of the rms common stock. We then analyze the predictive

    ability of the di ff erent models for two portfolios based on the likelihood of dividend payments

    as predicted only by rm characteristics. We expect that improved market liquidity should

    explain the declining propensity of rms to pay dividends better for more liquid rms and

    for rms with higher ability to pay dividends.

    In Panel A of Table IV we create two portfolios based on the median share turnover

    for the 1963 -2003 period. The sample size of rms with low share turnover decreases over

    16

  • 8/6/2019 Liquidity and Dividends

    19/47

    time due to the increase in share turnover for the average rm. The model based only on

    rm characteristics (model ( 1) of Table II, Panel B) results in higher prediction error for the

    portfolio of high turnover stocks. For the six-year period of 1998 -2003 the average predictive

    error for the low turnover stocks is 14 .53 percentage points while the average error for the high

    turnover stocks is 25 .97 percentage points. When we add share turnover as an explanatory

    variable (model ( 2) of Table II, Panel B) we see that the predictive ability improves mostly

    for the portfolio of more liquid (high turnover) stocks. More important, there is a reversal in

    the predictive error the predictive error now is lower for the portfolio of stocks with higher

    share turnover (0

    .98

    percentage points) as opposed to the portfolio with low share turnover(16 .83 percentage points).

    We now create two portfolios of rms based on the predicted probability of dividend

    payment using only rm characteristics. Firms that are more likely to pay dividends (a

    predicted probability above 70% ), based on rm characteristics and average estimates from

    the 1963 -1977 period enter into the rst portfolio. The second portfolio consists of rms

    that are less likely to pay dividends (a predicted probability below 70% ) based on rm

    characteristics. In the tests to follow, we allow stock market liquidity to di ff erentially in uencethe dividend choice of rms in the two portfolios. The results are presented in Panel B of

    Table IV.

    Trying to predict dividend payers based only on rm characteristics (model ( 1) of Table

    II, Panel B) results in higher predictive error for the portfolio of rms that are more likely to

    pay - larger, more pro table rms, and rms with fewer growth opportunities. The predictive

    error for such rms is 28 .76 percentage points as opposed to a predictive error of 13 .04 for

    rms with the characteristics of non-payers. This result is to be expected in view of the

    ndings of Fama and French ( 2001 ) that rms with the characteristics of payers are less

    inclined to pay dividends in more recent years. Adding share turnover as an explanatory

    variable (Model ( 3) of Table II, Panel B) produces a reversal in predictive ability so that now

    the predictive error is lower for the portfolio of rms that are more likely to pay based on

    17

  • 8/6/2019 Liquidity and Dividends

    20/47

    rm characteristics. The reduction in the predictive error for the portfolio of rms more able

    to pay dividends based on their characteristics is noteworthy. In the 1998 -2003 period, for

    example, the predictive error when we control for liquidity declines to 3 .27 percentage points.

    For the portfolio of rms that are less able to pay dividends the predictive error declines to

    6 .60 percentage points.

    [Insert Table IV about Here]

    We summarize the above discussion by concluding that a model that attempts to predict

    dividend payers based on

    rm characteristics fails to perfectly predict the proportion of dividend payers in more recent years. Further examination shows that such a model is

    less accurate when applied to rms that have more liquid stock markets and rms with

    higher ability to pay dividends. Controlling for stock market liquidity improves the overall

    predictability of the model. The predictability improves more signi cantly for rms with more

    liquid stock markets and rms that, for a given level of stock market liquidity, are more able

    to pay dividends. These ndings further support our hypothesis that stock market liquidity

    is relevant for rm dividend policy and that improved market liquidity has in uenced rms

    with the characteristics of dividends payers to not pay dividends in more recent years. 9

    V. Dividend Initiations and Omissions

    Existing research has found that the dividend decisions of rms that do not pay dividends

    are not the same as the dividend decisions of rms that already do. To address this issue we

    perform our tests separately for past payers and past non-payers.Furthermore, di ff erences in the present and past dividend policy across rms may lead to

    diff erential trading of investors in those rms. In other words, there is a possibility that our9 When we use the portfolio approach of Fama and French (2001) we obtain similar patterns in the overall

    predictive ability of the two models under consideration. The overall di ff erence between predicted and actualpayers, however, is larger than in the logistic predictive model.

    18

  • 8/6/2019 Liquidity and Dividends

    21/47

  • 8/6/2019 Liquidity and Dividends

    22/47

    characteristics makes an error of 4 .5 percentage points while including share turnover reduces

    the error to 1 .75 percentage points.

    [Insert Table VI about Here]

    The evidence presented in this section is consistent with our hypothesis that stock market

    liquidity is an important (although not the only) determinant of the rms decision to pay

    or not pay dividends. The results further show that liquidity is an economically important

    variable for dividend initiations but not for dividend omissions.

    VI. Robustness Analyses 10

    A. Share Repurchases

    Since the mid- 1980 s the share repurchase activity of rms, especially open market share

    repurchase activity, has increased signi cantly (see, for example, Jagannathan, Stephens,

    and Weisbach ( 2000 ) and Grullon and Michaely ( 2002 )). Share repurchases constitute an

    alternative means through which rms can distribute cash to shareholders and rms that

    repurchase shares may have lower ability to also pay dividends. Furthermore, share turnover,as a measure of liquidity, can be a ff ected by open market share repurchases performed by the

    rm. We address these issues in the current section.

    Before we continue, it is important to note that dividends and share repurchases are not

    necessarily perfect substitutes. On the one hand, dividends are usually taxed at higher rates

    than capital gains and they are less exible as a payout policy (see Jagannathan, Stephens and

    Weisbach ( 2000 )). On the other hand, repurchases can impose additional costs on investors

    (see Barclay and Smith ( 1988 ), Brennan and Thakor ( 1990 ), and Chowdhry and Nanda

    (1994 )) because they can lead to transfer of wealth among investors. Finally, open market

    repurchases do not lead to investor avoidance of trading friction since investors still have to

    trade in order to create homemade dividends.10 For the sake of brevity we do not tabulate the results of this section. All tables are available upon request.

    20

  • 8/6/2019 Liquidity and Dividends

    23/47

    Fama and French ( 2001 ) show that share repurchases are largely performed by dividend

    paying rms and that repurchases do not account for the declining propensity of rms to

    pay dividends. In this section, we argue that our own results are not driven by the increased

    share repurchase activity of rms.

    First, we nd that stock market liquidity is negatively related to the dividend policy of the

    rm in the cross-section for a portfolio of rms that do not repurchase shares. This relation

    is signi cant at the 0 .01 level in all the sub-periods and is economically important. When we

    predict the proportion of dividend payers for 1993 -2003 , a model based only on rm ability

    to pay makes a predictive error of 23

    .26

    percentage points while a model including liquidityreduces the error to 11 .73 percentage points.

    Second, to take into account the e ff ect of share repurchases on the ability of the rm to

    pay dividends we subtract the repurchased amounts from the earnings available to common

    shareholders and use this adjusted measure of pro tability to predict dividend payers. We

    additionally adjust share turnover for the impact that open market share repurchases may

    have on it from the traded shares in a given year we subtract the repurchase dollar amounts

    divided by the share price of the rm. Since we use total repurchases and not only openmarket repurchases, this adjustment overestimates the total number of repurchased shares.

    However, the error should be smaller in the 1970 s11 when rms rarely repurchased shares and

    for the end of our sample when most of the share repurchases are open market repurchases

    (Stephens and Weisbach ( 1998 ); Jagannathan, Stephens, and Weisbach ( 2000 )). As a result,

    our predictions for later years based on 1963 -1978 estimates should be relatively una ff ected. 12

    We nd that the coe ffi cients on the adjusted share turnover are very similar to the coe ffi cients

    on the unadjusted share turnover presented in Table II, Panel B. We also nd that, based on

    rm characteristics, there is a lower propensity of rms to pay dividends over time. Including11 COMPUSTAT has repurchase data from 1971, so we do not adjust values prior to that year. Using only

    years for which we have share repurchase data (1971 to 1978) as the estimation period leads to similar results.12 We calculate share repurchase amounts as in Grullon and Michaely (2002). Their measure is similar to

    the one used by Jagannathan, Stephens, and Weisbach (2000). Using change in treasury stock as in Fama andFrench (2001) gives similar results.

    21

  • 8/6/2019 Liquidity and Dividends

    24/47

    adjusted share turnover as a predictor again decreases the predictive error signi cantly. In

    the 1998 -2003 period, for example, the model based on rm characteristics has a predictive

    error of 21 .91 percent on average. Including share turnover reduces that error to 5 .93 percent.

    The overall results exhibit little di ff erences from our previous ndings.

    B. Institutional Clienteles

    It is possible that changes in institutional clienteles have resulted in changes in the div-

    idend policies of rms. In this section we control for institutional ownership. We collect

    institutional ownership data from SDC Thomson Financial and we create a variable that

    is equal to the proportion of rm shares held by institutional investors. We then use this

    variable as another predictor of rm dividend policy.

    We nd that the estimated coe ffi cients for share turnover, after controlling for institutional

    ownership, are still signi cantly negative (at the 0 .01 level) and even larger in magnitude.

    For 1993 -2003 , for example, the coe ffi cient for share turnover is 1 .15 as opposed to 1 .00

    when we do not control for institutional ownership. Therefore, even after we control for

    institutional ownership, we still nd that stock market liquidity is negatively related to the

    probability of dividend payments.

    C. Stock Option Compensation

    The past two decades have also witnessed a signi cant growth in stock option compensa-

    tion to rm top management. Since managerial stock options are not protected for dividend

    payments, they may provide incentives to managers to not pay cash dividends.

    We collect data from COMPUSTAT on shares reserved for conversion for the exercise of

    stock options (data item 215 ) and on total shares reserved for conversion (data item 40 ). We

    express these variables as a proportion of total shares outstanding and use them as additional

    control variables in our cross-sectional regressions.

    Our ndings suggest that rms with more shares reserved for conversion (either for the

    exercise of stock option of for other reasons) are less likely to pay dividends. We further nd

    22

  • 8/6/2019 Liquidity and Dividends

    25/47

    that the coe ffi cient estimates of share turnover are mostly una ff ected by the inclusion of these

    new control variables. For the 1963 -1977 period for example the coe fficient estimate of share

    turnover is 1 .63 when we control for total shares reserved for conversion. This estimate is

    very close to our previous estimate of 1 .74 when we do not control for shares reserved for

    conversion.

    From 1996 onward COMPUSTAT stopped reporting the number of shares reserved for

    conversion. Therefore our last predicted proportion of dividend payers based on a model that

    controls for shares reserved for conversion is for 1995 . The evidence suggests that, even after

    we control for shares reserved fro conversion, there is a lower propensity to pay dividends inmore recent year. Furthermore, including share turnover again explains most of this lower

    propensity to pay. 13

    VII. Dividend Policy and Liquidity Risk

    In this section we deviate from our previous framework. Until now we have focused our

    analysis on the relation between dividend policy and liquidity at the rm level. At this

    point we address the possibility of a link between dividend policy and aggregate liquidity.

    Pstor and Stambaugh ( 2003 ) propose that assets with high positive sensitivity of returns

    to aggregate liquidity result in disproportionate decline of investor welfare when aggregate

    liquidity is low. This is because liquidation is costlier when liquidity is lower and because

    investors have higher marginal utility of wealth when their wealth declines.

    Extending our arguments presented in the previous sections of the paper, we argue that

    investor demand for dividend paying stocks, and therefore the value of such stocks relative

    to non-paying stocks, is higher in states characterized by low aggregate liquidity. One impli-cation of our argument is that dividend initiations will lead to a reduction in the sensitivity

    of stock returns to aggregate liquidity.13 In unreported results we also control for the potential impact of dividend taxation relative to the taxation

    of capital gains, the debt level of rms, and the Nixon era. The qualitative results of the paper pertaining toliquidity persist in all of our robustness tests. The results are available on request.

    23

  • 8/6/2019 Liquidity and Dividends

    26/47

    In order to test the above implication, we identify a sample of rms that initiate divi-

    dend payments and analyze the changes in sensitivity of their returns to aggregate liquidity.

    Our methodology for estimating sensitivities to aggregate liquidity is based on Pstor and

    Stambaugh ( 2003 ).

    We collect a sample of rms that initiate dividends between 1966 and 1999 and use only

    rms that do not pay dividends in years t -3 , t-2, and t -1 and pay dividends in years t+ 1, t+ 2 ,

    and t+ 3. We then create two value-weighted portfolios. 14 The rst portfolio includes rms in

    the three years prior to dividend initiation. The second portfolio consists of the same rms

    but after dividend initiation. For example, the post-dividend portfolio in1985

    consists of all rms that initiated dividend payments in 1982 , 1983 , or 1984 . The pre-dividend portfolio in

    1985 consists of all rms that initiated dividends in 1986 , 1987 , or 1988 . As a result of our

    sample selection methodology, the post-dividend portfolio in 1989 contains the same rms as

    the pre-dividend portfolio in 1985 .

    We then estimate the market model, the three-factor model of Fama and French ( 1993 ),

    and the four-factor model (the three Fama-French factors plus a momentum factor) for each

    portfolio while also including the market-wide liquidity factor of Pstor and Stambaugh(2003 ). The three-factor model includes the market factor, the size factor, and the book-

    to-market factor. The market factor is the return of the value-weighted CRSP portfolio

    minus the risk-free rate, the size factor is the di ff erence in returns between small and large

    stocks, and the book-to-market factor is the di ff erence in returns between stocks with high

    and low book-to-market ratios. We use the momentum factor to account for the evidence

    presented by Jegadeesh and Titman ( 1993 ) that past performance is positively related to

    future performance. 15 The liquidity factor of Pstor and Stambaugh ( 2003 ) is based on the

    idea that order ow induces greater return reversals when liquidity is lower. The construction14 Pstor and Stambaugh (2003) show that estimates of the liquidity beta are highly imprecise for small

    stocks and that the problem is mitigated in value-weighted portfolios.15 We obtain the monthly Fama-French factors and the momentum factor from the Web site of Kenneth

    French (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/).

    24

  • 8/6/2019 Liquidity and Dividends

    27/47

    of the liquidity factor is outlined in the Appendix of the paper.

    The pre-dividend portfolio has 444 monthly returns and extends from 1963 to 1999 .

    The post-dividend portfolio includes the same rms as the pre-dividend one but covers 444

    monthly returns from 1967 to 2002 . Both portfolios have a minimum of 10 stocks and a

    maximum of 88 stocks.

    Panel A of Table VII reports the estimated pre-dividend and post-dividend liquidity betas

    for each of the speci cations. We nd that the pre-dividend liquidity beta is equal to 3 .49 in

    the four-factor model while the post-dividend liquidity beta for the same portfolio declines to

    3.35

    . These two estimates are statistically diff

    erent (at the0

    .10

    level) under the assumptionof independence. Pstor and Stambaugh ( 2003 ) report that, for individual stocks, liquidity

    betas are signi cantly and positively correlated over time. Positive serial correlation of the

    liquidity beta of our portfolio will lead to higher than reported signi cance levels of the

    estimated di ff erence. Using the distribution of liquidity betas for the ten portfolios reported

    by Pstor and Stambaugh ( 2003 ), our results suggest that dividend-initiating rms move

    from the highest liquidity beta portfolio (ten) to liquidity beta portfolio four (see their Table

    3). This leads to a reduction in the expected rate of return of the rm of 3 .12 percentagepoints per annum (see their Table 4).

    We further investigate the possibility that a similar trend is also observed for rms that

    do not initiate dividends. If we indeed observe such a trend, this would suggest that our

    results are not related to the dividend initiation decision of rms. Panel B replicates the

    analysis for rms that do not pay dividends in any year between year minus three to year

    plus three. The portfolios of non-initiating rms include signi cantly more stocks with a

    minimum of 152 and a maximum of 2 , 157 stocks per month. We nd that both before and

    after the measurement year this portfolio has negative liquidity betas that are relatively high

    in magnitude. Prior to the selection year this portfolio of dividend non-payers has a liquidity

    beta of 2 .64 while after the selection year the portfolios liquidity beta is 3 .13 and the two

    betas are indistinguishable from each other. We conclude that the results reported in Panel

    25

  • 8/6/2019 Liquidity and Dividends

    28/47

    A are speci c to dividend initiating rms.

    [Insert Table VII about Here]

    In summary, this section provides preliminary evidence that sensitivity of stock returns

    to aggregate liquidity declines after dividend initiations. One possible interpretation of this

    result is that, after dividend initiations, the value of the rm increases in states characterized

    by low aggregate liquidity and high marginal utility of wealth. Combined with existing

    evidence by Pstor and Stambaugh ( 2003 ), the results further suggest that reduced liquidity

    risk lowers expected returns by economically signi

    cant amounts. The overall results areconsistent with the idea that stock market liquidity and cash dividends act as substitutes

    from the perspective of investors.

    VIII. Conclusion

    We hypothesize that, all else equal, the payout policy of the rm is related to the liquidity

    of its common stock. In illiquid markets investors will have greater demand for cash dividends

    from the stocks they hold. In highly liquid markets, however, investors can cheaply create

    homemade dividends. As a result, rms with more (less) liquid stock markets will have lower

    (higher) incentives to distribute cash dividends to their shareholders.

    We nd a strong empirical relation between the dividend policy of the rm and the

    liquidity of its common stock. These results are prevalent throughout the analyzed period and

    remain after we control for the ability of rms to pay dividends. The documented declining

    propensity of rms to pay dividends over time is signi cantly related to the signi cant changesin the liquidity of US security markets. A period of fewer dividend payers is characterized

    by lower trading costs and increased market activity. We further present evidence that our

    results are more relevant for more liquid rms and rms that have the ability to pay cash

    dividends to their shareholders.

    26

  • 8/6/2019 Liquidity and Dividends

    29/47

    In this paper we take the repurchase policy of the rm as given and we show that, while

    share repurchases consume cash that can otherwise be distributed as dividends, our results

    are not driven by the increased popularity of open market share repurchases. It is also

    possible, however, that the proliferation of market share repurchase programs is, at least to

    some extend, stimulated by the improved liquidity of the stock markets over time.

    We additionally nd that liquidity is a signi cant determinant of dividend initiations and

    dividend omissions. Our results suggest that rms with less (more) liquid stocks are more

    (less) likely to initiate dividend payments. Similar to our overall results, improvements in

    stock market liquidity over time account for most the declining propensity of

    rms to initiatedividends as well.

    Finally, we investigate one possible extension of our analysis. Our ndings suggest that

    cash dividends and stock market liquidity act as substitutes from investors standpoint. Firms

    that initiate dividend payments reduce the sensitivity of their returns to aggregate liquidity,

    possibly because they lower investor exposure to systematic liquidity risk. This link is im-

    portant since it further suggests that dividend policy can have an impact on rm value due

    to market imperfections.

    27

  • 8/6/2019 Liquidity and Dividends

    30/47

    Appendix

    For each rm i , for a given month t we estimate the regression:

    r ei,d +1 ,t = i,t + i,d,t + i,t sign r ei,d,t i,d,t + i,d +1 ,t , d = 1 ,...,D, (1)where r i,d,t is the return of stock i on day d in month t , r ei,d,t is r i,d,t minus the CRSP

    value-weighted market return, and sign r ei,d,t i,d,t is the signed dollar volume for stocki on day d in month t in millions of dollars. We use only common shares with at least 15

    daily returns ( D > 15 ), that are listed on NYSE or AMEX at the end of year t -1 , that arepriced between $5 and $1000 at the end of the previous month, and for which t is not the

    rst or the last month of listing. Using the estimates of i,t for all rms in a given month

    that satisfy these conditions, we compute their equally-weighted average as b t =1N

    N

    Pi =1 b i,t .We then de ne:4

    b t = m tm 1 1N N

    Pi =1

    b i,t

    b i,t 1

    , (2)

    where m t is the market value of all rms used in the estimation for month t and m t

    corresponds to August of 1962 .

    Finally, we estimate:

    4 b t = a + b4 b t 1 + cm t 1m 1 b t 1 + u t , (3)

    and set our measure of aggregate liquidity equal to

    L t =1

    100 bu t . (4)

    28

  • 8/6/2019 Liquidity and Dividends

    31/47

    REFERENCES

    Allen, Franklin, and Roni Michaely, 2003 , Payout Policy, in Constantinides, George, MiltonHarris, and Rene Stulz eds. Handbook of the Economics of Finance (Amsterdam: North-Holland).

    Amihud, Yakov, 2002 , Illiquidity and stock returns: Cross-section and time-series e ff ects,Journal of Financial Markets 5, 31 -56 .

    Amihud, Yakov, and Haim Mendelson, 1986 , Asset pricing and the bid-ask spread, Journalof Financial Economics 17 , 223 -249 .

    Amihud, Yakov, Haim Mendelson, and Beni Lauterbach, 1997 , Market microstructure andsecurities values: Evidence from the Tel Aviv Stock Exchange, Journal of Financial Economics45 , 365 -390 .

    Atkins, Allen B., and Edward A. Dyl, 1997 , Transactions costs and holding periods forcommon stocks, Journal of Finance 52 , 309 -325 .

    Baker, Malcolm P., and Je ff rey Wurgler, 2004 a, A catering theory of dividends, Journal of Finance 59 , 1125 -1165 .

    Baker, Malcolm P., and Je ff rey Wurgler, 2004 b, Appearing and disappearing dividends: The

    link to catering incentives, Journal of Financial Economics 73 , 271 -288 .Barclay, Michael J., and Cli ff ord W. Smith, Jr., 1988 , Corporate payout policy: Cash Divi-dends versus Open-Market Repurchases, Journal of Financial Economics 22 , 61 -82 .

    Brennan, Michael J., and Avanidhar Subrahmanyam, 1996 , Market microstructure and assetpricing: On the compensation for illiquidity in stock returns, Journal of Financial Economics41 , 441 -464 .

    Brennan, Michael J., Tarun Chordia, and Avanidhar Subrahmanyam, 1998 , Alternative factorspeci cations, security characteristics, and the cross-section of expected stock returns, Journalof Financial Economics 49 , 345 -373 .

    Brennan, Michael J., and Anjan V. Thakor, 1990 , Shareholder preferences and dividendpolicy, Journal of Finance 45 , 993 -1019 .

    Carlson, Charles B., 1996 , Buying stocks without a broker, 2nd ed., (New York: McGraw-Hill).

    29

  • 8/6/2019 Liquidity and Dividends

    32/47

    Chordia, Tarun, Lakshmanan Shivakumar, and Avanidhar Subrahmanyam, 2003 , Liquiditydynamics across small and large rms, UCLA working paper.

    Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2000 , Commonality in liquid-ity, Journal of Financial Economics 56 , 3-28 .

    Chordia, Tarun, Avanidhar Subrahmanyam, and V. Ravi Anshuman, 2001 , Trading activityand expected stock returns, Journal of Financial Economics 59 , 3-32 .

    Chowdhry, Bhagwan, and Vikram Nanda, 1994 , Repurchase premia as a reason for dividends:A dynamic model of corporate payout policies, Review of Financial Studies 7, 321 -350 .

    Constantinides, George M., 1986 , Capital market equilibrium with transaction costs, Journalof Political Economy 94 , 842 -862 .

    Datar, Vinay T. Y., Narayan Naik, and Robert Radcli ff e, 1998 , Liquidity and stock returns:An alternative test, Journal of Financial Markets 1, 203 -219 .

    Davey, Patrick J., 1976 , Dividend reinvestment programs, Conference Board Report No. 699 ,The Conference Board Inc.

    DeAngelo, Harry, Linda DeAngelo, and Douglas J. Skinner, 2004 , Are dividends disappear-ing? Dividend concentration and the consolidation of earnings, Journal of Financial Eco-nomics 72 , 425 -456 .

    Dong, Ming, Chris A. Robinson, and Chris Veld, 2003 , Why individual investors want divi-dends, Working Paper.

    Dumas, Bernard, and Elisa Luciano, 1991 , An exact solution to a portfolio choice problemunder transactions costs, Journal of Finance 46 , 577 -595 .

    Easterbrook, Frank H., 1984 , Two agency-cost explanations of dividends, American EconomicReview 74 , 650 -659 .

    Erickson, Timothy, and Toni M. Whited, 2000 , Measurement error and the relationshipbetween investment and q, Journal of Political Economy 108 , 1027 -1057

    Fama, Eugene F., and Kenneth R. French, 1993 , Common risk factors in the returns on stocksand bonds, Journal of Financial Economics 33 , 3-56 .

    Fama, Eugene F., and Kenneth R. French, 2001 , Disappearing dividends: changing rmcharacteristics or lower propensity to pay? Journal of Financial Economics 60 , 3-43 .

    Glosten, Lawrence R., and Paul R. Milgrom, 1985 , Bid, ask and transaction prices in aspecialist market with heterogeneously informed traders, Journal of Financial Economics 14 ,

    30

  • 8/6/2019 Liquidity and Dividends

    33/47

    71 -100 .

    Grullon, Gustavo, and Roni Michaely, 2002 , Dividends, share repurchases, and the substitu-tion hypothesis, Journal of Finance 57 , 1649 -1684 .

    Jagannathan, Murali, Cli ff ord P. Stephens, and Michael S. Weisbach, 2000 , Financial exibil-ity and the choice between dividends and stock repurchases, Journal of Financial Economics57 , 355 -384 .

    Jegadeesh, Narasimhan and Sheridan Titman, 1993 , Returns to buying winners and sellinglosers: Implications for stock market e ffi ciency, Journal of Finance 48 , 65 -91 .

    Jensen, Michael C., 1986 , Agency costs of free cash ow, corporate nance, and takeovers,American Economic Review 76 , 323 -329 .

    Jones, Charles M., 2002 , A century of market liquidity and trading costs, Working Paper,Columbia University.

    Julio, Brandon, and David L. Ikenberry, 2004 , Reappearing dividends, Working Paper, Uni-versity of Illinois.

    Kyle, Albert S., 1985 , Continuous auctions and insider trading, Econometrica 53 , 1315 -1335 .

    La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny, 2000 ,Agency problems and dividend policy around the world, Journal of Finance 55 , 1-33 .

    Long, John B., Jr., 1978 , The market valuation of cash dividends: A case to consider, Journalof Financial Economics 6, 235 -264 .

    Miller, Merton H., and Franco Modigliani, 1961 , Dividend policy, growth, and the valuationof shares, Journal of Business 34 , 411 -433 .

    Myers, Stewart C., 1984 , The capital structure puzzle, Journal of Finance 39 , 281 -307 .

    Myers, Stewart C., and Nicholas S. Majluf, 1984 , Corporate nancing and investment deci-sions when rms have information the investors do not have, Journal of Financial Economics13 , 187 -221 .

    Pstor, Lubo, and Robert F. Stambaugh, 2003 , Liquidity risk and expected stock returns,Journal of Political Economy 111 , 642 -685 .

    Poterba, James M., 1986 , The market valuation of cash dividends: The Citizens Utilities casereconsidered, Journal of Financial Economics 15 , 395 -406 .

    31

  • 8/6/2019 Liquidity and Dividends

    34/47

    Routledge, Bryan, and Stanley E. Zin, 2001 , Model uncertainty and liquidity, Working paper,Carnegie Mellon University.

    Rozeff , Michael S., 1982 , Growth, beta and agency costs as determinants of payout ratios,Journal of Financial Research 5, 249 -259 .

    Shleifer, Andrei, and Robert W. Vishny, 1986 , Large shareholders and corporate control,Journal of Political Economy 94 , 461 -488 .

    Stoll, Hans R., 1978 , The supply of dealer services in securities markets, Journal of Finance33 , 1133 -1151 .

    Stoll, Hans R., 1979 , Regulation of securities markets: an examination of the e ff ects of increased competition, Monograph Series in Finance and Economics 1979 -2, New York Uni-versity Salomon Center.

    Stoll, Hans R., 2000 , Friction, Journal of Finance 55 , 1479 -1514 .

    32

  • 8/6/2019 Liquidity and Dividends

    35/47

    1.50%

    1.00%

    0.50%

    0.00%

    2.00%

    2.50%

    1963 1968 1973 1978 1983 1988 1993 1998 20030%

    40%

    80%

    120%

    160%

    200%

    Figure 1. NYSE Commission Rates and Stock Market Activity. Annual commission revenues of NYSEmembers and annual dollar volume on the exchange between 1966 and 2003 are obtained from theNYSE Fact Book. The annual NYSE commission rate (dashed line; left axis) is equal to total

    commission revenues divided by total dollar volume on NYSE for a given year. The sample used tocalculate the average share turnover (solid line; right axis) consists of all firms with publicly tradedcommon stock on NYSE with available information in the CRSP monthly files and the COMPUSTATannual files. Share turnover is the ratio of shares traded to shares outstanding for calendar year t .

    33

  • 8/6/2019 Liquidity and Dividends

    36/47

    0%

    20%

    40%

    60%

    80%

    100%

    1963 1968 1973 1978 1983 1988 1993 1998 20030%

    40%

    80%

    120%

    160%

    200%

    Figure 2. Dividend Payers and Stock Market Liquidity. The sample consists of all firms with publiclytraded common stock on NYSE or AMEX with available information in the CRSP monthly files andthe COMPUSTAT annual files. Dividend payers (%) is the proportion of firms that paid dividends infiscal year t (bars; left axis). Average share turnover (%) is the average percentage ratio of shares

    traded to shares outstanding for calendar year t (line; right axis).

    34

  • 8/6/2019 Liquidity and Dividends

    37/47

    Panel A. Dividend initiations (bars) and share turnover (solid line)

    0%

    4%

    8%

    12%

    16%

    20%

    1963 1968 1973 1978 1983 1988 1993 1998 20030%

    40%

    80%

    120%

    160%

    200%

    Panel B. Dividend omissions (bars) and share turnover (solid line)

    0%

    2%

    4%

    6%

    8%

    10%

    1963 1968 1973 1978 1983 1988 1993 1998 20030%

    20%

    40%

    60%

    80%

    100%

    Figure 3. Dividend Payers and Stock Market Liquidity Conditional on Dividend Policy in Year t -1. The sample consists of all firms with publicly traded common stock on NYSE or AMEX with availableinformation in the CRSP monthly files and the COMPUSTAT annual files. Panel A uses firms thatdid not pay dividends in fiscal year t -1. The figure plots the proportion of firms that pay dividends inyear t (bars; left axis) and the average percentage ratio of shares traded to shares outstanding for thesample for calendar year t (solid; right axis). Panel B uses firms that paid dividends in fiscal year t -1.The figure plots the proportion of firms that do not pay dividends in year t (bars; left axis) and theaverage percentage ratio of shares traded to shares outstanding for the sample for calendar year t (line; right axis).

    information in the CRSP monthly files and the COMPUSTAT annual files. Panel A uses firms thatdid not pay dividends in fiscal year t -1. The figure plots the proportion of firms that pay dividends inyear t (bars; left axis) and the average percentage ratio of shares traded to shares outstanding for thesample for calendar year t (solid; right axis). Panel B uses firms that paid dividends in fiscal year t -1.The figure plots the proportion of firms that do not pay dividends in year t (bars; left axis) and theaverage percentage ratio of shares traded to shares outstanding for the sample for calendar year t (line; right axis).

    35

  • 8/6/2019 Liquidity and Dividends

    38/47

    0%

    50%

    36

    Figure 4. Stock Market Liquidity and the Relative Valuation of Non-Payers. The sample consists of allfirms with publicly traded common stock on NYSE, AMEX, or Nasdaq with available information inthe CRSP monthly files and the COMPUSTAT annual files. The figure plots the average percentageratio of shares traded to shares outstanding for calendar year t (bold solid; left axis), the valuation of non-payers relative to dividend payers (solid; left axis), and the average price reaction to dividendinitiations (dash; right axis). Share turnover for Nasdaq firms is scaled by 50%. The valuation of non-

    payers relative to payers for year t is equal to the exponent of the negative value-weighted dividendpremium for year t reported by Baker and Wurgler (2004). The average price reaction to dividendinitiations for year t is equal to the standardized announcement returns reported by Baker andWurgler (2004) and averaged over years t -1, t , and t +1.

    100%

    %

    %

    1963 1968 1973 1978 1983 1988 1993 1998-0.50%

    0.00%

    0.50%

    1.00%

    1.50%200

    150

  • 8/6/2019 Liquidity and Dividends

    39/47

    37

    Table IVariable Definitions

    We gather all data from the COMPUSTAT annual files except where indicated. COMPUSTAT itemnumbers are presented in parenthesis. For COMPUSTAT variables, t refers to fiscal year t of the firm.

    Variable Definition

    Dividend

    Payer

    A firm is defined as a dividend payer in year t whenever COMPUSTAT reportspositive Dividends per Share (26) for fiscal year t . Otherwise a firm is defined as anon-payer.

    NYP tThe proportion (in %) of NYSE firms with lower market capitalization of commonstock in June of year t . Share price and number of shares outstanding for June of year t are obtained from the CRSP monthly files.

    E t/A t[Earnings before Extraordinary Items (18) + Interest Expense (15) + IncomeStatement Deferred Taxes (50) if available] divided by Assets (6).

    PreferredStock

    Preferred Stock Liquidating Value (10) [or Preferred Stock Redemption Value (56), orPreferred Stock Par Value (130)].

    BookEquity

    Stockholder s Equity (216) [or Common Equity (60) + Preferred Stock Par Value(130), or Assets (6) Liabilities (181)] Preferred Stock + Balance Sheet DeferredTaxes and Investment Tax Credit (35) if available Post Retirement Assets (330) if available.

    Vt/A t[Assets (6) - Book Equity + Stock Price (199) times Common Shares Outstanding(25)] divided by Assets (6).

    dA t/A t [Assets t (6) Assets t-1 (6)] divided by Assets t (6).

    TURNt

    Common Shares Traded (28) divided by Common Shares Outstanding (25).

    DVOL tNatural logarithm of [1.00 + Share Price (24) times Common Shares Traded (28)divided by the 1996 Consumer Price Index].

    ILLIQ t

    We use the CRSP daily files. For every trading day in year t , we divide the absolutereturn by the dollar volume. Dollar volume is expressed in 1996 US dollars using theConsumer Price Index. We then create the illiquidity variable by averaging the dailyestimates over all trading days of year t and then taking natural logarithm. If thereare less than 30 daily observations for a given year this variable is missing.

    NOTRD tThis variable is equal to the proportion of days with no trading volume on the CRSPdaily files for year t . The variable is missing if there are fewer than 30 dailyobservations.

  • 8/6/2019 Liquidity and Dividends

    40/47

    38

    Table IIEstimates from Logistic Regressions to Explain Dividend Payers

    For each year t between 1963 and 2003 we estimate logistic regressions to explain whether a firm paysdividends in year t . The sample contains all NYSE/AMEX firms with available information. Panel Auses only share turnover (TURN t) as an explanatory variable. As explanatory variables for the rest of the models we use the proportion of NYSE firms with lower market capitalization (NYP t), theearnings-to-assets ratio (E t/A t), the market-to-book ratio (V t/A t), and the growth rate of assets(dA t/A t). We refer to these four variables as firm characteristics. We then add several variables thatproxy for the market liquidity of the firm s stock. In Panel B we add share turnover (TURN t). Wealso allow for different coefficients of share turnover for firms that are less likely (TURN _Lt) and forfirms that are more likely (TURN _Mt) to pay dividends based on firm characteristics. Firms that havean estimated probability below 70% (the average proportion of payers for 1963-1977) to pay, based ontheir firm characteristics and the average coefficient estimates from the 1963-1977 period, are classifiedas less likely to pay. Firms for which this probability is above 70%, are classified as more likely to pay.In Panel C we further use the illiquidity ratio (ILLIQ t), the traded dollar volume (DVOL t), and theproportion of non-trading days (NOTRD t) as explanatory variables. When we use the illiquidity ratiowe also include the natural log of the monthly average of the share price for year t as a control formicrostructure effects resulting from minimum tick size rules. The table reports the average estimatedcoefficient for a given period and the t -statistic of whether the average estimate is significantlydifferent from zero. Firms is the average number of firms used in the regressions. Payers is the averagepercent of payers for a given period. Payers (+) is the estimated percent of payers if the examinedliquidity measure improves by one standard deviation while Payers ( ) is the estimated percent of payers if the examined liquidity measure deteriorates by one standard deviation. The effects of

    liquidity changes on the percent of payers are examined at the presented average percent of payers.a, b,

    c indicate significance respectively at the 1%, 5%, and 10% levels from a two-tailed t -test.

    Panel A. Dependent variable is whether a firm pays dividends in year t 1963-1977 1978-1992 1993-2003

    (1) (2) (3)

    Intercept 1.60 a 1.00a 0.18b

    (10.28) (6.23) (2.40)

    TURN t 1.62a 0.50a 0.33a

    ( 5.93) ( 3.66) ( 5.53)

    Firms 1,374 1,389 1,432Payers 71.50 67.06 48.12Payers (+) 59.55 62.29 41.84Payers ( ) 81.04 71.51 54.46

  • 8/6/2019 Liquidity and Dividends

    41/47

    39

    Panel B. Dependent variable is whether a firm pays dividends in year t

    1963-1977 1978-1992 1993-2003(1) (2) (3) (4) (5) (6) (7) (8) (9)

    Intercept 0.22 0.33c 0.28 0.18 0.29 0.24 1.03a 0.58a 0.62a

    ( 1.62) (1.99) (1.70) ( 0.91) ( 1.47) (1.24) ( 15.33) ( 7.91) ( 8.70) NYP t 4.17a 4.13a 4.40a 4.28a 5.08a 5.26a 3.19a 4.17a 4.30a

    (32.11) (18.71) (22.16) (73.06) (42.67) (48.79) (25.68) (34.36) (32.20) Vt/A t 0.84a 0.74a 0.79a 0.73a 0.66a 0.70a 0.34a 0.35a 0.37a

    ( 9.44) ( 8.90) ( 10.02) ( 6.35) ( 7.29) ( 7.21) ( 10.54) ( 12.28) ( 13.08) dA t/A t 0.54 0.05 0.09 0.66a 0.34b 0.39b 1.32a 1.05a 1.07a

    ( 1.25) ( 0.13) ( 0.22) ( 4.21) ( 2.24) ( 2.62) ( 7.31) ( 5.08) ( 5.18)

    E t/A t 16.57a 15.87a 16.56a 8.18a 7.97a 8.58a 5.11a 4.72a 5.16a(13.10) (14.62) (14.40) (11.61) (10.77) (10.36) (27.73) (23.85) (24.05)

    TURN t 1.74a 1.45a 1.00a

    ( 14.91) ( 23.42) ( 12.21)

    TURN _Lt 1.59a 1.26a 0.87a

    ( 9.50) ( 24.09) ( 10.96) TURN _Mt 1.99a 1.56a 1.05a

    ( 15.22) ( 20.27) ( 11.94)

    Firms 1,374 1,389 1,433Payers 71.49 67.06 48.12Payers (+) 58.60 52.61 30.04Payers ( ) 81.64 78.88 66.70

  • 8/6/2019 Liquidity and Dividends

    42/47

  • 8/6/2019 Liquidity and Dividends

    43/47

    41

    Table IIIActual and Predicted Percent of Dividend Payers Using Average Logistic Regression Estimates from

    1963-1977For each year between 1963 and 1977 we estimate logistic regressions to explain whether a firm paysdividends in year t . As explanatory variables in model (1) we use the proportion of NYSE firms withlower market capitalization (NYP t), the earnings-to-assets ratio (E t/A t), the market-to-book ratio(V t/A t), and the growth rate of assets (dA t/A t). We refer to these four variables as firmcharacteristics. In Model (2) we add share turnover (TURN t) as an explanatory variable. In Model

    (3) we estimate different slope coefficients for share turnover for firms that are less likely (TURN_

    Lt)and for firms that are more likely (TURN _Mt) to pay dividends based on firm characteristics. Firmsthat have an estimated probability below 70% (the average proportion of payers for 1963-1977) to pay,based on their firm characteristics and the average coefficient estimates from the 1963-1977 period, areclassified as less likely to pay. Firms for which this probability is above 70%, are classified as morelikely to pay. Firms is the number of firms for a given year. Actual Percent is the average actualpercent of payers in a given period. For each model, we estimate the individual firm s probability topay dividends using the average coefficients from 1963-1977 and the values of the explanatoryvariables for year t in the logistic regression. We then average these probabilities across firms toestimate the Predicted Percent of payers in a given year t . Finally, we average our annual predictionsfor a given period.

    (1) (2) (3)

    Actual Predicted Predicted Predicted Predicted Predicted PredictedFirms Percent Percent Actual Percent Actual Percent Actual

    1978-1982 1,545 77.24 76.27 0.97 73.76 3.48 73.31 3.92

    1983-1987 1,338 65.97 69.69 3.72 64.42 1.55 63.82 2.14

    1988-1992 1,283 57.98 68.58 10.60 64.26 6.28 63.62 5.65

    1993-1997 1,485 51.16 67.36 16.20 59.50 8.34 58.63 7.47

    1998-2003 1,390 45.58 67.78 22.20 51.95 6.36 50.27 4.69

  • 8/6/2019 Liquidity and Dividends

    44/47

    42

    Table IVActual and Predicted Percent of Dividend Payers for Different Portfolios Using Average Logistic

    Regression Estimates from 1963-1977For each year between 1963 and 1977 we estimate a logistic regression to explain whether a firm paysdividends in year t . We present the actual proportion of payers and the difference between thepredicted and the actual proportion of payers for several periods. As explanatory variables in Models(1) and (3) we use the proportion of NYSE firms with lower market capitalization (NYP t), theearnings-to-assets ratio (E t/A t), the market-to-book ratio (V t/A t), and the growth rate of assets(dA t/A t). We refer to these four variables as firm characteristics. In Models (2) and (4) we also addshare turnover for year t (TURN t) as an explanatory variable. In Panel A we create two portfoliosbased on share turnover (TURN t) using the average of its 50 th percentile for the 1978-2003 period. InPanel B we create two portfolios based on the predicted probability of dividend payment. Firms withan estimated probability below 70% (the average proportion of payers for 1963-1977) to pay, based ontheir firm characteristics and the average coefficient estimates from the 1963-1977 period (Table I,Panel B, model 1), are classified as less able to pay while firms, for which this probability is above70%, are classified as more able to pay dividends. In models (2) and (4) of Panel B, we estimatedifferent slope coefficients for share turnover for firms that are less likely (TURN Lt) and for firms thatare more likely (TURN Mt) to pay dividends based on the above classification (Table II, Panel B,model 3). Firms is the average number of firms in a portfolio for a given period. Actual Percent is theaverage percent of payers in a given portfolio for a given period. We estimate the individual firm sprobability to pay dividends using the average coefficients from 1963-1977 and the values of theexplanatory variables for year t in the logistic model. We then average these probabilities across thefirms in a given portfolio for a given year t to estimate the Predicted Percent of payers for thatportfolio in that year and then obtain the value of Predicted minus Actual payers. For each period,the table presents the averages of the annual values.

    Actual Predicted Predicted Actual Predicted PredictedFirms Percent Actual Actual Firms Percent Actual Actual

    (1) (2) (3) (4)

    Panel A

    Less liquid (low share turnover) More liquid (high share turnover)

    1978-198