Document Sample
                 OTHERS DON'T

                                             November 2002

                    Varouj Aivazian and Laurence Booth (University of Toronto)
                               Sean Cleary (St. Mary's University)*

JEL Classification Codes: G35, G32

Keywords: Dividends, information asymmetries, agency costs, debt ratings

* The authors thank the Social Sciences and Humanities Research Council of Canada (SSHRC) for
financial support provided for this project. They also would like to thank participants at the Northern
Finance Association 2001 meetings and Queens University for their comments

Corresponding author: Professor Laurence Booth, booth@rotman.utoronto.ca

One of the main arguments supporting a traditional Lintner style dividend smoothing policy is
that high stable dividend payments solve agency and signalling problems. However, it is apparent
that whether such policies are optimal or not depends crucially on the underlying contractual
environment. We argue that differences in dividend policy reflect the type of debt that the firm
has outstanding. If the firm accesses "uninformed" public debt markets (bonds) it is more likely
both to pay dividends and to follow a Lintner (1956) style dividend smoothing policy than if it
uses private informed (bank) debt. Our empirical results confirm these hypotheses, where we
study the probability of paying a dividend, the probability of having a bond rating and the extent
of dividend smoothing by US firms from 1981 to 1999. Our most significant result is that by
including interaction terms we demonstrate that dividend smoothing is almost completely a
function of whether or not the firm has a bond rating. Firms without bond ratings flow through
more of their earnings as dividends and display very little dividend smoothing behaviour. In
effect, they seem to follow a residual dividend policy. In contrast, firms with bond ratings follow
a traditional Lintner style smoothing policy, where the influence of the prior dividend payment is
very strong and the current dividend is relatively insensitive to current earnings. These results are
robust to alternative econometric specifications and the results for firms with bond ratings mimic
those of Lintner (1956) and contradict those contained in more recent research.

1.      Introduction

There is no simple economic rationale for the payment of dividends or the adoption of a particular
dividend policy, for example a residual versus a dividend smoothing policy. While the costs of a
dividend payment are relatively easy to identify, for example the excess tax burden, or the higher
transaction costs of external equity, the benefits are less obvious. It is commonly accepted that there
are two major advantages to a firm adopting a policy of smoothing annual dividend payments. The
first is that a policy of regular dividend payments plays a signaling role in reducing information
asymmetries. The second is that dividend payments help to reduce agency costs. In both cases, the
normal assumption is that it is the equity holders who benefit from this dividend policy. However, in
this paper we examine the role of dividend policy as a signalling and agency reduction tool from the
point of view of the firm's debt holders. Specifically we examine the interaction between the firm's
debt decision and its dividend policy.

It has long been recognized that debt and dividend decisions are joint solutions to both signalling and
agency problems. For example, a firm can pre-commit to making large dividend payments or interest
payments; both will reduce the firm's free cash flow and the attendant equity agency costs
(Easterbrook (1984)). Similarly, a firm can signal its higher quality, either by increasing its debt ratio
(Ross (1977)) or by paying higher dividends (Bhattacharya (1979)). In both cases, the shareholders
notice the firms' decision and bid up its value accordingly. In the former case due to the lower risk of
a dissipation of corporate resources and in the latter because the firm has signaled its higher quality.
In this sense both dividend payments and the debt decision serve as substitutes in the signaling and
agency reduction roles.1

However, the efficiency of the two decisions depends on the organizational structure of the capital
market. For example, La Porta et al (2000) show that the effectiveness of securities laws and the
extent of monitoring by insiders, that is, how closely held the firm is, affects the usefulness of debt
and dividend decisions in these roles. For example, a closely held firm does not need to increase its
dividend, or take on more debt, to signal to insiders the higher quality of its earnings. Similarly,

managers who waste corporate resources in personal perquisite consumption can easily be removed
when there is an inside control block. In both cases, the higher proportion of inside informed
shareholders changes the efficiency of the dividend and debt decisions.

We argue that a similar constraint attaches to debt markets. The firm may have a choice between
private "informed" debt (bank debt) and public "uninformed" debt (bonds). We argue that similar to
closely held firms, the choice of private bank debt reduces the value of the signalling and agency
reduction role played by dividend policy. In contrast, we argue that firms with public market debt
have a greater incentive to reduce information asymmetries and agency problems to induce investors
to hold their debt. They do this by adopting dividend policies that signal their higher quality and
reduce agency costs. Consequently, we argue that the probability of a firm paying a dividend and
subsequently following a dividend smoothing policy depends on the underlying factors that cause it
to have public market debt outstanding.

In this context, it is important to remember the legal distinction between debt and equity. Equity
holders are protected by securities law, while debt holders are protected by contract law, that is, the
contractual features of the specific debt contract. For example, debt holders cannot appeal to the
provisions of state and federal securities law mandating fair disclosure, protection of minority rights,
prohibition on insider trading, non arms length transactions etc. Instead, their protection lies in the
specifics of the debt contract. If a firm reduces the value of its debt by doing something that is not
prohibited in the debt contract, the debt-holders have no protection, whereas the equity holders can
seek redress under securities legislation. Given their narrower legal basis it is clear that debt-holders,
similar to equity-holders have a vested interest in resolving the informational and agency problems
facing the firm. However, unlike equity markets this interest depends critically on the nature of the
debt contract, that is whether it is private informed (bank) debt or public uninformed (bond) debt.

In addressing these questions, we look at the dividend decisions of US firms over the period 1981 to
1999. We examine the firm's decision to pay a dividend by analyzing a variety of fundamental firm
characteristics. We extend these characteristics to include factors that we argue cause the firm to seek

public market rather than bank debt. Conditional on these factors we can then estimate the classic
Lintner (1956) dividend adjustment model, where we argue that the decision to smooth dividends or
adopt a residual dividend policy depends on public market access. As far as we are aware, no one has
previously looked at the direct impact of public market debt decisions on corporate dividend policy.

The rest of this paper is organized as follows: Section 2 discusses some fundamentals and important
firm characteristics; Section 3 discusses in detail the role of public versus private debt markets;
Section 4 describes the data and summary statistics; Section 5 estimates logistic regressions for
predicting whether or not the firm pays a dividend; Section 6 presents Lintner coefficient estimates
conditioned on firm characteristics and debt ratings; and finally Section 7 adds some conclusions and
suggestions for further research.

2.      The Importance of Dividend Policy for Debt Markets

Miller and Modigliani (1961) introduced the residual theory of dividends based on the firm's
sources and uses of funds, that is if cash is constant

                        MP  X  Capex  Div
A debt-free firm's cash flow from operations (X) minus its capital expenditures (Capex),
commonly referred to as its free cash flow, plus the proceeds of any new share issues (shares
issued (M) times net price (P)) has to equal aggregate dividends (Div). For a 100% equity
financed firm this is an identity.

However, assuming that the firm undertakes all positive net present value projects makes Capex
exogenous, as is X. Consequently dividends are a residual: any change in aggregate dividends
necessitates an offsetting change in the number of shares. If the dividend is too low then shares
are repurchased and if a dividend is too large, then new shares are issued causing share price
declines. In the M&M approach the fundamental is the firm's free cash flow (X-Capex).
Assuming that the constraint is binding, that is, that cash is not indefinitely increasing or
decreasing implies that dividend payments are determined by the fundamental factors affecting

free cash flow.

In predicting a firm's free cash flow we take the firm's profits (prof) as a proxy for the firm's cash
flow from operations. Profits is calculated as net income before extraordinary items, to abstract
from items like restructuring costs or one-time gains and losses, scaled by total assets. This is to
avoid the problem of negative book equity that exists for some firms. Consistent with M&M we
would expect firms with high profits to pay dividends. To proxy for the firm's capital
expenditures we use the firm's capital expenditures scaled by the firm's net fixed assets, which
we term invest. Consequently a high value for invest indicates a firm with relatively high capex
and low free cash flow. We would expect firms with high values for invest to have low or zero
dividend payments. Finally, we use the firms' market to book (M/B) ratio as a proxy for the firm's
investment opportunities. The M/B ratio, defined as the closing stock market price divided by the
book value per share, is a proxy for Tobin's Q. We would expect firms with high M/B to have
significant growth opportunities or intangibles. Consequently, even though a firm's free cash flow
in the current year may be positive, we would argue that high M/B firms would build up cash for
future investments. As a result, they will also tend to have low dividend payments.

The M&M model applies to a 100% equity-financed firm; however to consider the impact of
different types of debt we have to extend the cash flow constraint. The accounting identity
indicates that cash flow from operations (X) plus cash flow from investing (Capex), plus cash
flow from financing equals the change in cash. From our point of view we only include debt
financing as cash flow from financing, since we are interested in the share issue/dividend
decision. We use the firm's debt ratio, which is its total debt scaled by total assets as a proxy for
the firms' debt market constraint. Firms with higher debt ratios (debtta) have less financial
flexibility, since they have used up more of their debt capacity, consequently we would expect
them to have lower or zero dividend payments.

Our steady-state cash flow identity (that is no change in cash) is

                  X  Capex  Debt  Div  MP
Our proxies for the fundamentals that affect dividend payments are: profits (prof), the investment
rate (invest), the market to book ratio (MB) and the debt ratio (debtta). These generate
predictions about whether the firm is generating or using cash and thus able to pay dividends.
However, the fundamentals do not indicate the nature of the dividend policy chosen by the firm.

Lintner (1956) was the first to observe that firms tended to follow a slow adaptive process in
setting their dividend per share. Lintner hypothesized that firm's actual dividend per share was an
adjustment of their existing dividend per share (di,t-1) to their target dividend per share (di,t*), that

                               d i ,t  ai  ci (d i*,t  d i ,t 1 )   i
where ci is the adjustment coefficient, ai is a fixed time series intercept and εi a random error
term. Lintner further argued that the firm's target dividend is the product of their target payout
rate and their normalized earnings, so he estimated the following equation

                                 d i ,t  ai  bi ei ,t  ci d i ,t 1   i

where the coefficient on normalized earnings is the product of the optimal payout and the
adjustment coefficient and that on the lagged dividend one minus the adjustment coefficient.

Lintner estimated his equation for 28 carefully selected US firms from 1918-1941 and the
earnings data for each firm was smoothed prior to his analysis. Lintner estimated the coefficient
on lagged dividends to be 0.70 indicating an adjustment speed coefficient of 0.30 and a
coefficient on current earnings of 0.15. This produced an optimal payout rate of 50%, Further
Lintner found "that the relationship between current earnings and the existing dividend rate was
very generally much the most important single factor determining the amount of any change in
dividends decided upon.” Using a slight variation of the Lintner approach, Fama and Babiak
(1968) examined data for two samples of 201 and 191 U.S. firms over the 1947 to 1964 period.
Their results implied a slightly higher speed of adjustment of 0.366. The Lintner model of
dividend smoothing with a slow adjustment to an equilibrium rate is still widely regarded as the

standard model of dividend policy.

The Lintner model is normally viewed as a solution to the agency and signalling problems. A
pattern of consistent and increasing dividends convinces the external shareholders that excess
cash will be disgorged, thus mitigating the agency costs of equity, while also signalling the
consistent quality of the firm's earnings. However, as La Porta et al (2000) reminded us this
solution depends on the characteristics of the firm's investors. It may be an optimal solution for a
firm with a dispersed share ownership, but closely held firms may not need to signal earnings
quality or disgorge cash to discipline managers, since the "internal" stock holders can observe
them directly. Dewenter and Warther (1998) used this argument to estimate the Lintner model for
Japanese firms that were members of a Keiretsu where there were interlocking share ownership
patterns with non-Keiretsu Japanese and US firms. We take this insight a step further and argue
that the Lintner dividend policy is also a solution to debt market dispersion. In this case, the
distinction is between "informed" private (bank) debt held by one institution and "uninformed"
public market debt (bonds) with multiple investors.

3.     Private versus Public Debt Markets

The contrast between bank debt and the bond market has been analyzed in an extensive literature
dating at least to Leland and Pyle (1977) and Diamond (1984) who both argued that banks have a
comparative information advantage over the bond market. According to Diamond, this stems
from the reduced monitoring costs faced by banks. Private "bank" debt holders have better access
to both the senior managers of the firm and confidential information that can help to alleviate
signalling and agency problems. In this context managers can reveal information to their bankers
that they may not want or are not allowed to reveal to securities markets, including the public
debt markets.

The banking relationship usually requires the filing of quarterly financial information in a
standardized form, as well as regular site visits by the lending officer, so that the lending officer

is familiar with the company (Fama (1985)). Much of bank debt is either short term or involves
“material adverse conditions” clauses that effectively give the bank an almost continuous call
option on its loan. Risk is also reduced via the practice of recovering the principal of the loan
through monthly mortgage type payments, which in effect serves a similar pre-commitment
function as a dividend. Finally, the bank loan is simply one part of a services package offered by
the bank, including check processing, credit cards, as well as a flexible lending package of
operating lines and term loans (Sharpe (1990)).

All of these aspects of the bank relationship serve to control agency costs and reduce signaling
problems. It is not surprising therefore, that James (1987) found that the announcement of a
credit facility was accompanied by a 1.7% two-day abnormal equity return. Initiating a banking
relationship divulges information to the capital market in the same way as the initiation of a
dividend. Lummer and McConnell (1989) took James‟ results a step further and showed that a
renewal of a credit facility, rather than the signing of a first credit facility, accounted for the
favorable stock market reaction and within this group, expanded facilities were associated with
increased stock prices, whereas more restrictive agreements such as loan reductions or
cancellations were associated with stock price declines.

One implication is that the cost of bank debt relative to public debt will be lower for firms with
strong, but risky, growth opportunities, low current profits and firms facing high information
asymmetries. These considerations point to size as a factor in the type of debt used by firms.
Information acquisition has scale economies. For example, larger firms have more coverage by
security analysts, which would tend to reduce the information asymmetries and reduce the
relative importance of growth options. In contrast, less is publicly known about smaller firms,
while they would also tend to have a higher proportion of growth options and face larger
transactions in accessing public markets. For these firms the "know your client" advantage of
informed bank debt reduces the longer term rescheduling risk.

In contrast, the public bond markets are dominated by a dispersed group of institutional

investors. For example Table 1 provides aggregate data on bond holdings for 1990 and 1999.
Since bonds tend to have longer maturity dates than bank debt, they are important assets for
institutions with long-term liabilities. In 1990 approximately 62% of all outstanding bonds were
held by insurance companies, savings institutions, retirement funds, and private pension funds.
By 1999 this share had dropped to 47%, mainly due to the increase in foreign holdings from
12.7% to 18.0%, and of bond funds from 3.5% to 8.1%. Over the same period foreign investors
and bond funds increased their importance as net purchasers of bonds, however, they in turn are
likely to be institutional purchasers. What is striking about this data is the "institutionalization"
of the bond market. In both 1990 and 1999 the household sector was a marginal player in the
bond market holding only 12.7% (1990) and 13.1% (1999) of the outstanding bonds.

The fact that the bond market is an institutional market with multiple investors has important
implications for the criteria used to buy bonds, since these institutions are heavily regulated to ensure
that they can meet their liabilities, such as insurance and pension commitments. This can be seen in
the typical prospectus for a bond issue where one of the issues addressed is always who is eligible to
hold these bonds, based on criteria that have traditionally been referred to as the "legal for life" rules.
The rules have varied across both time and jurisdiction, but all state regulators are concerned that
"their" companies can meet their commitments.

For example, 33 states and the District of Columbia have adopted the Uniform Prudent Investor Act,
while the Uniform Management of Institutional Funds Act (which governs endowment funds) has
been adopted by 46 states and the District of Columbia. Most private sector pension plans are
regulated by Chapter 29 of the US Labor Code under the Employee Retirement Income Savings
Program (ERISA). The common feature of these regulations is an attempt to define a set of
investments that pass a "reasonableness" test. The decided advantage of these regulations for fund
managers is that almost by definition such bonds are regarded as prudent and protect the manager
from charges of irresponsibility.

These criteria serve to differentiate the private informed bank market from the public "uninformed"
bond market. Public bond markets suffer from Akerlof's (1970) classic lemons problem, which
necessitates screening and signaling mechanisms; in effect inducing quality selection based on self
imposed firm constraints. Consequently, the dispersed institutional investors are more concerned
with broad financial ratios, such as interest coverage ratios, debt ratios, cash-flow to debt ratios and
above all the firm‟s bond rating than is an informed bank lender. In this context dividend policy has a
role to play in signalling quality and reducing agency costs.

The importance of dividend policy for the public bond markets is not just a theoretical proposition,
but a fact long enshrined in legislation. For example, Section 63 of Chapter 175 of the General Laws
of Massachusetts concerning insurance states:

       The capital of any domestic company, other than life, and three fourths of the reserve of any
       domestic stock or mutual life company, shall be invested only as follows…
       In the bonds, notes or other evidences of indebtedness of any corporation primarily engaged in public
       transportation which is incorporated or located wholly or in part in the commonwealth, or in the bonds,
       notes or other evidences of indebtedness of any corporation primarily engaged in public transportation
       which is located wholly in or in part in any state of the U.S., whose capital stock equals at least one
       third of its funded indebtedness, which has paid regularly for the five years next preceding the date of
       such investment all interest charges on said funded indebtedness, and which has paid regularly for
       such period dividends of at least four percent per annum upon all its issues of capital stock, (bold
       and italics added) or whose net earnings available for fixed charges during each of any three, including
       the last two, of the five fiscal years next preceding the date of investment, have been for such years not
       less than one and one half times the total of its present fixed charges, or in the bonds, notes or other
       evidences of indebtedness of any corporation which have been, both as to principal and interest,
       assumed or guaranteed by any such corporation primarily engaged in public transportation… 2

The statute above indicates that in order to become an eligible investment for a Massachusetts
insurance company, a firm has to meet a variety of tests, for example by paying dividends or
meeting an interest coverage test. The important point is that a firm's dividend policy has
important implications for the institutions that dominate the corporate bond market. Quite simply
a stable Lintner style dividend policy is taken as one indicator of low risk, such that the bonds are
of sufficient quality to be bought by institutions. This legislation enshrines the risk reduction
strategy theorized by Akerlof (1970) as a debt market clearing mechanism in the presence of
information asymmetries.

However, although a stable and steadily increasing pattern of dividend payments may increase
bond market access, the question remains why is this an advantage? Berlin and Loeys (1988)
argue that the choice between public debt and bank debt hinges on the tradeoff between the
relative efficiency of liquidation and rescheduling costs versus information acquisition and
monitoring costs. In the same vein, Chemmanur and Fulghieri (1994) focus on the longer time
horizon of banks and their reputation incentive to avoid incorrect liquidation decisions. The
implication is that firms with higher liquidity and/or bankruptcy risk find it optimal to finance
with bank debt, where it is easier to reschedule the debt.

The comparative information advantage of private informed debt (banks) in dealing with riskier
borrowers does not come without a cost to the firm, since the bank may exploit this bargaining
power in debt renegotiation. Rajan (1992), for example, points out the hold up problem attached
to bank debt, while the illiquidity of bank debt forces up costs relative to public market debt.
Despite these costs, riskier firms with potentially strong growth prospects, facing significant
informational asymmetries will generally choose bank debt over public market debt. In contrast
lower risk, more profitable firms with fewer growth prospects prefer to issue public (bonds) debt,
where they also have access to long-term funds.

This discussion points to a strong interaction between dividend policy and the type of debt issued
by a firm. We have already hypothesized that profitable, low investment, low market to book
firms with little debt are more likely to pay dividends. This discussion of debt markets indicates
that these firms are also likely to issue public market debt if they are also of low risk and
consequently have low rescheduling risk. Fama and French (1993) have pointed out that two key
risk factors are the market to book ratio and size. We would argue that small firms with high
market to book ratios are high risk. In contrast, large firms with low market to book ratios are
less risky. Fama and French's empirical results confirm that a size variable such as the natural
logarithm of total assets (lnta) is a proxy for both risk as well as public market access.3

One final aspect of risk is important, which is the collateral available to a lender. We argue that
firms seeking bank debt do so because of the lower rescheduling risk. We further argue that these
firms are likely to have a low proportion of net fixed assets to total assets. Put another way, these
firms will likely have either more intangible assets or more current assets or both. The
information asymmetry advantage accruing to private lenders allows them to be cash flow
lenders, while taking the current assets as collateral. In contrast, the public markets with longer
term financing will be more interested in firms with a high proportion of net fixed assets. Greater
tangibility of the firm's assets (Tang), defined as net fixed assets divided by total assets, serves to
increase the accessibility of a firm to public markets and thus increase the likelihood that the firm
adopts a Lintner style dividend policy.

The firms that are most likely to pay dividends (that is large, low risk, profitable firms with low
growth opportunities, low investment demands, tangible assets and relatively little debt) are also
likely to access the public debt markets and follow a Lintner style dividend smoothing policy. In
contrast, small, unprofitable, indebted, risky firms with few long lived assets, but high
investment demands and significant growth options are unwilling to access the public bond
markets and instead seek safety in the lower rescheduling risks attached to informed bank debt.
Consequently, there is little advantage to them in paying a dividend and even if they do so there
is no need for them to smooth their dividend payments in a Lintner style policy. Instead dividend
payments for these firms are more likely to be a genuine residual and to be much more sensitive
to current free cash flow.

We distinguish between the public versus private debt markets by whether or not the firm has a
bond rating. All firms have a banking relationship and borrow from a bank. Even firms who
access the public bond markets and issue commercial paper will have back up lines of credit and
use banks for check clearing etc.4 The key distinction is that firms without public market access
have no need to pay the fees and provide the information necessary to get a bond rating. The only
problem is that some firms with bond ratings are "fallen angels," which had previously qualified
for the public markets, but no longer do so, since they are in default and/or have weak non-

investment grade ratings.5 In this case, there will be some firms with bond ratings who have
characteristics more similar to that of bank debtors.

4      Sample Characteristics

We use annual data from 1981 to 1999 collected from the Research Insight (US Compustat)
database. The use of annual data is motivated by the empirical observation that firms tend to
make major dividend changes on an annual basis. Using quarterly or higher frequency data will
therefore bias the coefficients of the Lintner model.6 All available firm year observations were
collected and deletions were only made if the value for either total assets or sales were zero. The
result is an unbalanced panel, since there was no requirement that data be available for each firm
throughout the entire period. The total number of firm-year observations is 127,516 from all SIC
industry groups. Bond ratings were also collected, but this data is only available from 1985.

Of the 127,526 firm year observations, 49,300 or about 39% involved a firm that made a dividend
payment. Over the period 1985-1999, there are 104,223 observations, of which 18,676 were for firms
with bond ratings. Of these 18,675 firm year observations, 67% or 12,452 paid a dividend. In
contrast to the 85,547 observations for firms without a bond rating 30% or 25,429 paid a dividend.
Further, there are 8,748 observations with investment grade bond ratings of which 94% or 8,221 paid
a dividend. If we think in terms of conditional probabilities, while the overall probability of a firm
paying a dividend is about 39%, conditional on the firm having a bond rating the probability
increases to 67% and conditional on an investment grade bond rating it increases further to 94%.

Clearly, the dividend decision is closely related to the bond rating; however there may also be other
factors are at work, since the sample characteristics of the rated and non-rated firms differ, for
example, with respect to the number of observations and industry breakdown. To address the
sampling issue we created a matched sample by matching each “rated” firm-year observation with
another non-rated observation from the same year for a firm with the same four digit SIC and as
similar a size as possible based on total assets. This matching procedure generates a sample

containing an equal number of firm-year observations for rated and non-rated firms while controlling
holding for these other factors as closely as possible.

In the matched sample of 34,154 observations 62% included a dividend. For the observations with a
bond rating, 67% paid a dividend and for those without 57%. Although the payment of a dividend is
still more likely for a firm with a bond rating, the difference is reduced dramatically. Clearly the
overall sample of non-dividend paying observations consists of smaller firms from different
industries (SICs) and time periods. Moreover, these differences appear to impact the dividend
decision, in addition to the existence of a bond rating. This observation points to the importance of
industry classification and time periods as important control variables.

Table 2 provides summary statistics on dividend policy and other key financial measures for our two
time periods 1981-1999 and 1985-99. The first observation in each row is the mean and the second
the median, since accounting ratios are often highly skewed. We report ratios for total common share
dividends scaled by both earnings before interest and taxes (EBIT), which we term the "overall"
dividend payout as well as net income, which is the normal or "regular" measure of payout. We
chose to scale dividends by both EBIT and Net Income to avoid the potential influence of extra-
ordinary items. We also report total sales and total assets, and measures of operating performance
and risk: current profit as earnings before extra-ordinary items scaled by total assets, investment as
the ratio of capital expenditures to prior net fixed assets, the market to book ratio, the natural
logarithm of total assets, the debt ratio and the tangibility of the firms‟ assets.

For the entire sample the average overall payout was 10.2% and the regular payout 26.1% with
medians of 0% for both indicating the skewed nature of dividend payments. When the non-zero
dividend observations were removed, the average values increase to 26.3% and 67.4% respectively
with medians of 18.2% and 33.2%. If we focus on the median payout we see that for rated firms the
overall and regular payouts increase to 11% and 19.4% and for rated firms, which pay a dividend,
they increase again to 19.9% and 37.5%. In contrast, the median overall and regular payout rates for
non-rated firms were both zero, while for the non-rated, dividend paying, firms the medians

increased to 21.4% and 39.4%. What is clear from this data is that although non-rated firms are much
less likely to pay a dividend, those that do seem to payout only slightly less than the rated, dividend
paying, firms. Similarly, if we look at the differences between investment and non-investment grade
rated firms,7 we see that the differences are less pronounced between investment and non-investment
grade forms that pay a dividend.

Table 2 also provides summary data on our independent variables. Looking at the medians for the
rated firms versus the non-rated firms indicates that rated firms tend to be: much larger in terms
of both sales and assets; more profitable; less indebted; have more tangible assets and lower
investment rates, but similar if marginally lower market to book ratios. A comparison of the
means with the medians suggests that many variables, especially for the non-rated firms, are
highly skewed, so we focus our discussion on the medians. When contrasting investment versus
non-investment grade firms the comparison is not so clear-cut. Investment grade firms are:
larger, more profitable and invest less. However, although they have more tangible assets and
lower debt ratios, the differences are not as pronounced, while their market to book ratios are
about the same. The differences between means and medians are also not quite so dramatic,
although they still exist particularly for profitability and size.

If the same comparisons are made on the sub set of dividend paying observations, the differences
get smaller still. For example, while rated firms are still larger, their overall and regular payouts
are similar to those of non-rated firms with medians of 19.9% and 17.9% and 37.5% and 31.5%
respectively, but they are less profitable. The non-rated firms also tend to have fewer tangible
assets, but greater investment rates and less debt. Similarly, when we compare investment and
non-investment grade rated firms that pay dividends there is very little difference. Again the non-
investment grade firms tend to be smaller, less profitable and have lower market to book ratios,
but the differences are not as pronounced as with all firm year observations. The main insight is
that the "best quality" firms are investment grade firms that pay dividends. The most diverse are
the non-rated firms that do not pay dividends. These firms tend to be smaller, have fewer tangible
assets, and have high investment rates.

Table 3 provides summary statistics for the matched sample of rated and non-rated observations.
While the non-rated firms were chosen to be as close as possible to the rated ones in terms of size,
the rated firms in the sample are much larger than non-rated ones and in many cases the nearest
match was significantly smaller. However, the size differential narrows somewhat if only the positive
dividend paying observations are included.

Two critical observations are apparent from the data in Table 3. First in comparing the firms that pay
dividends to the overall sample, it is clear that the dividend paying firms are larger, more profitable,
have lower debt ratios and slightly more tangible assets, but they also have slightly lower market to
book ratios and lower investment rates. Secondly, rated firms continue to be bigger, but within the
dividend paying group their overall and regular payout rates are similar to those of non-rated
dividend paying firms. In addition, the rated dividend paying firms are less profitable, have lower
investment rates and more debt. Market to book ratios and tangibility of assets are roughly the same
between rated and non-rated firms. What is clear is that the biggest difference between rated and
non-rated firms is not their dividend payout and that what differences there are less pronounced once
we control for industry (SIC), size and the time period (the matching criteria).

Table 4 provides a breakdown of the observations into seven industry groups numbered 1-7
based on SIC codes. Manufacturing, the second group makes up 41.8% of the total sample with
the balance fairly evenly distributed across the other groups. However, if we look at the bottom
part of the table only 16.3% of the manufacturing observations were for rated firms and only
36.3% had positive dividend payments. The industry with the largest percentage of rated
observations is Transport and Public Utilities (the third group) at 40.8%, with 59.1% of their
observations associated with a dividend payment. As a result, this group is disproportionately
represented in the overall rated group at 22.9%. The industry with the lowest percentage of rated
firms is Public Administration and other (group 7) with only 1 observation overall, since only
2.8% of their observations are rated and only 4.8% associated with a dividend. Of the remainder
the share of observations with ratings were most prominent in finance/insurance and real estate

17.3%, and wholesale and retail at 10.3%. Finance/insurance and real estate along with
transportation and utilities also had the highest share of dividend observations at 63.6% dividend
paying. In contrast only 26.6% of resource firm-year observations included a dividend and only
18.2% of the service sector.

What Table 4 indicates is that there are pronounced industry effects in both bond ratings and
dividend payments. We have already included size, market to book and the tangibility of a firm's
assets as risk variables, but there may are clearly other effects at work. To control for these
missing variables we add industry indicator variables (SIC1-7) for the seven industry groups in
Table 4 and year dummies to control for the changing composition of the Compustat data- base.
Together these variables adjust for the differences revealed by the matched sample comparisons.

5.     The Probability of Paying a Dividend

Our principal dividend paying hypotheses are:

        i) profitable firms with low investment needs, low debt and low market to book ratios are
       more likely to pay dividends, that is,

                P( D  0)  F (' profit' , invest, MB, debtta )

       ii) if the firms are profitable with low investment needs, low debt and low market to book
       ratios and in addition are large with more tangible assets then they will access the public debt
       markets rather than use private bank debt. That is,

                P( Bondrating  0)  F (' profit' , invest, MB, debtta, Lnta, ' Tang ' )

That is, if the firms are profitable with low investment needs, low debt and low market to book ratios
and in addition are large with more tangible assets then they will access the public debt markets
rather than use private bank debt.

We first consider the simple linear probability model where the dividend payment decision is
identified by an indicator variable that takes the value 1 if they pay a dividend and 0 if they don‟t.
These results are presented in the first column of Table 5, estimated with robust standard errors
(White‟s sandwich estimator). The model is not a serious model of the probability of paying a
dividend, since the linear model does not impose the constraint that the probability must lie between
0 and 1.0. The model is estimated simply to examine the multi-collinearity between the independent
variables. If there is multi-collinearity, the standard errors can be inflated, biasing any significance
tests. To check for this we can estimate the variance inflation factors (VIF). These are simply defined
as 1/(1-RSQX), where RSQX is the Adjusted R Square of a regression of the variable against the
other independent variables.8 If there is significant multi-colinearity RSQX will be high, inflating the
standard error of the regression coefficients. For our linear model the VIFs are: 1.11; 1.10; 1.04;
1.03; 1.00 and 1.00 respectively with an average of 1.05. Chatterjee et al (2000) indicate that multi-
collinearity is a problem if any of the individual VIFs exceed ten and the average is “considerably
larger” than 1.0. The VIFs for our independent variables do not seem to suffer from multi-
collinearity, which indicates that the estimated standard errors are unbiased.

More realistically we can estimate the probability of paying a dividend using the logistic regression
                     P( D  0)
                Ln(              )  F (' profit, invest, MB, debtta)
                   1  P( D  0)

where the natural logarithm of the odds ratio is estimated as a linear function of the independent
variables. The logit model results are presented in the columns 2-8 of Table 5. In each case the
estimates are based on robust standard errors, where independence is allowed across firms, but not
necessarily within firm observations. This is to recognize the problem that with panel data many of

the firm-year observations are not genuinely independent. As we will show the robust standard errors
are frequently much larger than conventional estimates. Consequently, our significance tests are not
inflated by the large number of firm-year observations.

The simplest model uses only the observation of whether or not the firm has a bond rating. This
model is in the second column of Table 5, where since all the independent variables are indicator
variables there is no intercept. The coefficient on the bond-rating indicator is 1.554 with a “t”
statistic of 35.73, without the robust standard errors the "t" statistic would have been 90.18. Although
the pseudo R square is only 6.4%, there is clearly a relation between the dividend decision and
whether or not the firm has a bond rating.9 However, our matched sample summary statistics indicate
industry effects. To adjust for these we include indicator variables for each of the seven SIC groups,
where the first is automatically dropped. This results in the model in the third column of Table 5.
Note that the pseudo R square increases to 13.31% and that even though all the indicator variables
are significant, the coefficient on the rating indicator is largely unchanged. Including the SIC
indicator variables formalizes the relationship between industry groups and dividend behavior that
was evident in the summary statistics in Table 4. In other words relative to resource firms all other
industry groups are more likely to pay a dividend, except services, and public administration and

Our time period covers 1981-1999, which includes a variety of economic climates. Regardless of its
dividend policy we would expect firms to be more likely to cut their dividend in bad, rather than
good market conditions. To account for this we can either include specific macro-economic variables
to proxy for market conditions or simply include annual indicator variables for each year. For
simplicity we do the latter and the logit estimates are in column four. Note that all of the annual
indicator variables are significant, but the coefficient on the rating indicator variable is still 1.501
with a t statistic of 32.64, while all the SIC indicators remain largely unchanged. Of note is that many
of the later time dummies are significantly negative indicating a trend towards decreasing dividend
payments over time, which is consistent with the evidence of Fama and French (2001). We examine

this by including time as an independent variable. The logit model in Column 5 indicates that the
coefficient on the rating indicator variable is unchanged, but the time trend is significantly negative.

The upshot of the models estimated in columns 2-5 is that the probability of paying a dividend
changes across industries and has also changed over time, both with the business cycle as well as a
temporal decline. Whether this latter effect is simply due to the inclusion of more firms in the
Research Insight database, or reflects a broader change can not be determined from this data alone.
The important point is that no matter how we adjust for time and industry effects, the fact that a firm
has a bond rating is important additional information for the dividend decision.

We now consider the four independent structural variables introduced earlier reflecting the residual
theory of dividends: profitability, investment rates, the market to book ratio and the debt ratio. These
logit regression estimates are presented in columns 6 and 7, where the estimates in column 6 do not
include time and SIC indicator variables. In both models the probability of a firm paying a dividend
increases with its profitability and decreases with its market to book ratio. We argued previously that
profitability reflects ability to pay a dividend, whereas the market to book ratio reflects the nature of
the firm‟s assets and future prospects. We take both of these results as supporting the residual theory
of dividends. However, the signs on the investment rate and the debt ratio for the simple model are
not significant, but as we saw previously this could be due to industry and time effects. In the model
in column 7 where we control for these effects by including time and industry indicator variables, the
signs on both profitability and the market to book ratio are essentially unchanged, while the sign on
the debt ratio is now significantly negative as expected.

The logit models largely support our proxies for the residual theory of dividends: profitable firms
with low market to book ratios and relatively little debt have a higher probability of paying a
dividend. The proxy for the rate of investment is consistent with the hypothesis, but not significant.
Both the models in columns 6 & 7 are conditional on the SIC groupings in Table 4. In the last
column (8) of Table 5 is the logit model conditional on each individual firm, that is a fixed effects
model with individual company indicator variables. Unfortunately a large number of observations

have to be dropped simply because they are fully identified with company indicator variables.10
Consequently, the sample size is reduced to 25,305 firm-year observations. However, the essential
result remains the same: more profitable firms with low debt are more likely to pay a dividend, but
this time the market to book ratio as well as the investment rate are not significant confirming that
these variables are proxies for industry effects.

We now turn to the question of how public market access affects the firm's dividend decision. We
begin by testing the validity of our proxies for market access: size and the tangibility of the firm's
assets. Recall that we argued previously that both these variables would induce the firm to access
public debt markets, rather than bank debt. The first two models in Table 6 simply add these two
variables to the models of columns 7 and 8 in Table 5. In comparing the models it is clear that while
the significance and signs of the coefficients on profits, investment rate, the market to book ratio and
the debt ratio do not change in a material way, the signs on both size (lnta) and the tangibility of the
firm‟s assets are both positive and significant. The only qualification to this is the tangibility of the
firm's assets, since this is only significant at the 20% level in model 2. The tangibility of a firm's
assets is clearly affected by the inclusion of individual firm indicator variables. Overall, we take
these results as consistent with our hypothesis that larger firms with more tangible assets are more
likely to access public debt markets rather than bank debt.

A more explicit test of the public debt market hypothesis is to examine whether these two variables
affect the probability of a firm seeking a bond rating, since this is required to access the public
markets. In columns 3 and 4 are models predicting whether or not the firm has a bond rating using
the four fundamental variables: profit, investment rate, market to book ratio and the debt ratio. The
models in columns 5 and 6 then add the public market access variables: size and tangibility. Since
bond ratings are only available for the period 1985-1999, the number observations is reduced.

The models in columns 3 -6 suggest that more profitable firms with high debt ratios have bond
ratings. Unlike the dividend decision where a high debt ratio reduces the probability of paying a
dividend, a high debt ratio increases the probability of having a bond rating. Neither the investment

rate nor the market to book ratio are significant, except for the investment rate in the restricted fixed
effects model with individual firm indicator variables. When we add size and tangibility the pseudo
R Square of both models jumps dramatically with the size variable overwhelmingly the most
important. In both models the significance of profit is reduced, while the tangibility variable changes
sign when individual fixed effects are added. Overall, it is quite clear that larger firms seek out
public debt markets, while the importance of the tangibility of the firms' assets seems to be
dependent on the nature of the firm's operations and is thus more affected by industry and firm
control variables.

Finally columns 7 and 8 present the results for two ordered logit models that contain more
information than simply whether or not the firm has a bond rating. In this case the dependent
variables is the rating itself converted to a numeric scale, for example, with AAA converted to 28,
AAA- to 27 etc. This information helps us counter the "fallen angel" problem, that many lower rated
firms may not be able to access the public markets with their current ratings, since their ratings may
be C or D. An ordered logit model takes into account the cardinal ordering from 28 to1, as well as
the discrete nature of the dependent variable. The results in columns 7 and 8 are consistent with the
previous results. Firms are more likely to have a higher bond rating if they are more profitable, have
less debt, are larger and have more tangible assets. Again the probability of having a higher rating
has dropped over time. The fact that the pseudo R Square jumps significantly when the market
access variables of size and tangibility are added confirms our prior hypothesis that firms with these
characteristics are more likely to seek out public market debt than private market bank debt.11

6       Dividend Smoothing

The estimated models and summary data presented in Tables 2 to 6 indicate that the debt and
dividend decisions are affected by similar underlying fundamental variables, but that the type of debt
is also important. What remains to be seen is whether the decision to pay a dividend, is implemented
differently in terms of a dividend policy by firms that issue public market debt. We examine this
question by estimating the standard Lintner model. Our third hypothesis is

           iii)   Firms with bond ratings smooth their dividends more than those without

That is, the dividend policy of firms with bond ratings will follow a Lintner style adaptive process
with more dividend smoothing and less sensitivity to current earnings than firms without bond

The Lintner model estimates are presented in Table 7. The model in the first row includes all firm
year observations and we get the now standard result that the coefficient on the lagged dividend has
increased to about 0.90, while that on earnings has dropped marginally to about 0.14, relative to
Lintner's estimates. We get this result whether we include all observations or restrict the sample to
positive dividend paying observations only. The second model includes the same time and industry
indicator variables that we used in the Logit models. However, unlike those models the addition of
the SIC and time indicator variables does not change the results. Overall these results confirm the
prior work of, for example of Dewenter and Warther (1999). However, they are puzzling, since
Lintner used his model to estimate the equilibrium payout; this occurs when the dividend adjustment
is finished and the lagged and current dividends are the same. If we assume a zero intercept the
equilibrium payout is the coefficient on current earnings dividend by one minus the coefficient on the
lagged dividend.12 Similar to Dewenter and Warther the estimates of the models in the first two rows
of Table 7 imply equilibrium payouts over 100%. Clearly an equilibrium dividend payout of this
magnitude is a puzzle.

The problem is that these results, like those existing in the literature, are based on panel data
estimates of all available data with no prior screens, whereas Lintner's estimates were for 28
carefully chosen firms. The panel data estimates have at least two econometric weaknesses: they
do not adjust for dependence of the firm-year observations within groups, or possible
autocorrelation in the residuals. In the third model are the estimates where we allow for
individual firm effects. In this case unique intercepts for each firm allows for missing variables.
The estimates from this third model are closer to those of Lintner, in that the coefficient on the

lagged dividend is smaller, while that on current earnings increases. The overall model implies
an equilibrium payout of 50%, while the model estimated over positive dividend paying
observations still has an almost 100% payout. The estimates from the fourth model allow
constant autocorrelation across the individual firm observations, as well as fixed firm effects.13
For this model, the coefficient on the lagged dividend is smaller still and the payout starts to look
reasonable. For all observations the equilibrium payout is 31%, which increases, as we would
expect, once the sample is restricted to positive dividend paying observations only. In this case
the equilibrium payout is 50.0%, which is between the average and median payouts in Table 2.
We argue that the results of the fourth model are the most reasonable, since they adjust for the
econometric problems involved with using unbalanced panel data and time series variables that
may be auto-correlated. Fortuitously the results also make sense, which they didn't when these
econometric problems were not recognized.14

The fifth model extends the Lintner estimates to include an interaction term for whether or not
the firm is rated.15 The interaction term is equal to the lagged dividend times an indicator
variable, which is one if the firm is rated and zero otherwise. Conceptually this is equivalent to
estimating two separate regression models, while pooling the observations to maximize the
power of the tests and forcing a common intercept. The simple null hypothesis is that the
indicator variable for a bond rating is insignificant and all firms smooth their dividends to the
same degree. In this case the interaction term should be insignificantly different from zero.
However, this is not the case.

For all observations the interaction term is highly significant at 0.85 and the coefficient on the
lagged dividend drops to -0.136. What this means is that non-rated observations have a negative
coefficient of -.136, while rated firms have a positive coefficient of .714 (.850-.136). This
indicates that rated firms smooth their dividends, while non-rated ones do not. For the rated firms
the equilibrium payout is 40.9%, which is an increase from the 31% of the fourth model. For the
positive dividend observations the results are very much the same. The coefficient on the
interaction term is highly significant (0.784), while that on the lagged dividend is marginally

negative (-.031). Again this indicates the absence of dividend smoothing by non-rated firms and
very significant smoothing by rated firms with an equilibrium dividend payout of 62.3%.

If only rated firms smooth their dividends, what happens to their payout from current operations,
(that is the adjustment coefficient on current earnings)? To answer this question we also include
an interaction term for earnings equal to the earnings per share times the rating indicator variable.
This allows both the coefficient on the lagged dividend and the earnings to vary depending on
whether the firm is rated or not. This is the sixth model in Table 7.

First, notice that the impact on dividend smoothing is even more dramatic than for the fifth
model. For all observations the coefficient on the lagged dividend is -0.316, while the interaction
indicator for the dividend increases to 1.064, indicating an even larger impact on dividend
smoothing between rated and non-rated firms. For earnings the results are also dramatic. The
coefficient on earnings almost doubles from 0.117 to 0.248 indicating a much greater adjustment
of the dividend to current earnings. However, the interaction term is significantly negative at
-0.168, indicating that rated firms adjust their dividend much more slowly in response to
increased earnings. When the model is run on positive dividend payments only, the results are the
same: highly significant interaction terms indicating that rated firms smooth their dividends more
and adjust their dividends more slowly to increased earnings.

The results for the sixth model are impressive; they are also entirely consistent with Lintner's
classic results, and differ from more recent results. In particular, we show in Table 2 that firms
with bond ratings tend to be larger, more profitable, have more tangible assets and debt, and
lower market to book ratios than the typical firm-year observation in our sample. By all objective
yardsticks these firms are more similar to the limited sample of 28 carefully chosen firms used by
Lintner in his original tests than the overall sample. For these rated, dividend paying, firms the
coefficient on lagged dividends is 0.763 and that on earnings 0.109 implying an equilibrium
payout of 48%. For these firms, the Lintner model works as well as it did for Lintner. In contrast
for non-rated firms, the coefficient on lagged dividends is -0.238 and that on earnings 0.265.

These non-rated firms would seem to be following a residual dividend policy in flowing through
a greater share of current earnings in dividends, as well as ignoring the prior period's dividend
payment. Clearly these non-rated firms are not using the dividend to reduce signalling and agency
problems and have no equilibrium dividend payout.

If the Lintner model works better after we control for the existence of a bond rating, what about
the level of the rating itself? It is impractical to include 28 indicators, one for each of the 28
rating classes, instead we distinguish between investment and non-investment grade debt. The
estimates in the seventh model include an interaction term for investment grade observations
only. In both cases the coefficient on the lagged dividend increases marginally, while the size of
the interaction indicator falls slightly. Similarly, the coefficient on earnings is marginally smaller
with a slightly larger coefficient on the earnings interaction term. Overall, however, the
implications of the results in the seventh model are the same as those in the sixth: investment
grade firms smooth their dividends more than other firms and adjust them more slowly to current
earnings. The fact that the results are less dramatic is simply because non-investment grade rated
firms have more in common with investment grade firms than non-rated firms.

Overall, the results including the interaction terms within panel data estimates that account for
both autoregressive residuals and dependence across firm observations are consistent both with
Lintner's classic results and an equilibrium payout that makes sense. Note, that the equilibrium
payout rate for dividend paying firms is marginally below 50% for rated firms, regardless of the
rating. Over all observations, including the non-dividend paying observations, the equilibrium
payout is of course lower.

These results are also much broader than Lintner's, which were based on a very small sample of
US companies. What is quite clear from the data is that the extent of dividend smoothing is
different between firms with bond ratings and firms without. In fact, there is very little evidence
that firms that do not have a bond rating smooth their dividends at all. In fact quite the opposite,
larger dividends tend to be followed by smaller dividends and vice versa, as firms flow through a

much larger share of their current earnings (about 25%) into dividends. In contrast there is very
strong evidence that firms with bond ratings smooth their dividends and adopt radically different
dividend policies than other firms, since their dividends are adjusted much more slowly in
response to current earnings. This is consistent with our prediction that firms with bond ratings
smooth their dividends as part of a strategy to maintain access to the public bond markets.

7. Conclusions

We examined the dividend decisions of US firms over the period 1981 to 1999 and arrived at
several major conclusions.

1)        Consistent with the residual theory of dividends, the probability of a firm paying a
dividend increases with the firm's profitability and decreases with the firm's debt level and the
existence of intangible growth options (high market to book ratios). These results were robust to
various econometric specifications. We also find that the probability of a firm paying a dividend
has declined over time. Whether this is due to sample selection problems or an economic trend is
difficult to determine from our data, but the results support the recent work of Fama and French

2)        When we added two proxies for public debt market access we consistently found that size
was an overwhelmingly important factor in paying a dividend and that the tangibility of a firm's
assets was also important. However, the problem with tangibility is that, like the investment rate,
it is closely associated with industry characteristics, and consequently was affected by including
indicator variables for SIC industry classification versus firm specific indicator variables.

3)        The probability of a firm having a bond rating was also found to be strongly associated
with the same set of fundamental factors, with the public market access proxies again the most
important. An ordered logit model also confirmed the results for specific bond ratings. Large
profitable firms with tangible assets and low market to book ratios tend to have high bond

ratings. The only obvious difference with the dividend models is that firms with high debt tend to
have bond ratings, where high debt firms have a lower probability of paying a dividend. Clearly
if firms do not have much debt outstanding it makes little sense to access the public markets and
get a bond rating. These results are robust to the inclusion of industry controls that account, for
example, for the special circumstances of regulated industries.

4)     Our analysis of the classic Lintner model indicates that the results of recent papers that
show the coefficient on the lagged dividend has increased while that on earnings has decreased
are "correct," but probably result from poor econometric specification. When we correct for
autocorrelation and dependence of observations in panel data we get results consistent with
Lintner's original results and opposite to those recently reported.

5)     We also find that including interaction terms for whether or not the firm is rated
significantly affects the estimated coefficients in the Lintner model. Firms with bond ratings
smooth their dividends and payout less from current earning than firms that are not rated. The
signs on these interaction terms are very, very, large and overwhelmingly significant. Our
interpretation of these results are that firms with bond ratings smooth their dividends, whereas
non-rated firms follow a policy closer to the residual theory of dividends. Moreover the
equilibrium payout from these augmented Lintner models is consistent with common sense and
traditional estimates of an overall dividend payout of 50%.

6)     Finally, when we include an interaction term for whether or not the firm has a bond rating
or whether it is investment grade we get the same results as Lintner and show that dividend
smoothing is primarily a feature of firms with bond ratings. Whether or not that rating is
investment grade has relatively little impact. Firms without a bond rating do not smooth their
dividends; instead they follow a residual policy.

Our overall conclusion is that there clearly is an interaction between debt and dividend policy.
The same fundamental factors that affect the dividend decision are also likely to affect the firms'

capital structure decision. However, we argue that the critical fact for dividend policy is not the
amount of debt, but the type of debt. Firms can raise their debt through private informed bank
debt, and we argue that they do so when they are risky. In particular, when the firm's investment
opportunity set consists of a large amount of growth options, then there is increased rescheduling
risk: the possibility of having to return to the lender to restructure the debt increases. We also
argue that risk is higher for small firms, with high market to book ratios and a large proportion of
intangible, that is, non-net fixed assets in their asset structure.

Our results confirm the importance of fundamental firm characteristics in determining whether or
not the firm is likely to pay a dividend, and that the same characteristics affect the probability of
the firm accessing the public debt markets. The bond rating in turn is simply a way of
aggregating this information into a single variable, which serves to differentiate a firms' dividend
policy. Lintner style dividend smoothing only seems to be the solution to agency and signaling
problems for firms with bond ratings, since it is only these firms that have to deal with a
dispersed public investor in their bonds. In contrast, firms without bond ratings raise their debt
from the private bank market, where there is no reason to use dividend policy to solve these
problems. Hence, these firms follow a residual theory of dividend policy.


Akerlof, G, "The Market for „Lemons‟: Quality Uncertainty and the Market Mechanism,"
      Quarterly Journal of Economics (August l970).

Berlin, M., and J. Loeys, 1988, “Bond covenants and delegated monitoring,” Journal of Finance
        43, 397-412.

Bhattacharya, S., 1979, “Imperfect information, dividend policy, and 'the bird in the hand'
       fallacy,” Bell Journal of Economics 10, 259-270.

Chatterjee, S, S. Hadi and B. Price, 2000, Regression Analysis by Example, (3rd ed), John Wiley
       and Sons, New York.

Chemmanur, T. J., and P. Fulghieri, 1994, “Reputation, renegotiation, and the choice
     between bank loans and publicly traded debt,” The Review of Financial Studies. 7, 475-

Dewenter, K., and V. Warther, 1998, “Dividends, asymmetric information and agency conflicts:
      Evidence from a comparison of dividend policies of Japanese and U.S. firms,” Journal of
      Finance 53, 879-904.

Diamond, D. W., 1984, “Financial intermediation and delegated monitoring,” Review of
      Economic Studies 51, 393-414.

Easterbrook, F. H., 1984, “Two agency-cost explanations of dividends,” American Economic
       Review 74, 650-659.

Fama, E., 1985, “What‟s different about banks?” Journal of Monetary Economics 15, 29-36.

Fama, E., and Babiak, H., 1968, “Dividend policy: An empirical analysis,”, Journal of the
       American Statistical Association 63, 1132-1161.

Fama, E and K. French, 1993, "Common stock factors in the returns on bonds and stocks,"
      Journal of Finance, February, 3-56.

Fama, E., and K. French, 2001, “Disappearing dividends: Changing firm characteristics or lower
      propensity to pay?” Journal of Financial Economics 60, 3-43.

James, C., 1987, “Some evidence on the uniqueness of bank loans,” Journal of Financial
       Economics 19, 217-235.

John, K., and J. Williams, 1985, “Dividends, dilution, and taxes: A signaling equilibrium,”
       Journal of Finance 40, 1053-1070.

La Porta, R., F. Lopez de Silanes, A. Shleifer and R. Vishny, “Agency problems and dividend
       policies around the world”, The Journal of Finance, 2000.

Lintner, J., 1956, “Distribution of incomes of corporations among dividends, retained earnings
       and taxes,” American Economic Review 46, 97-113.

Leland, H., and D. Pyle, 1977, “Information asymmetries, financial structure and financial
       intermediaries,” Journal of Finance 32, 371-387.

Lummer, S., and J. McConnell, 1989, “Further evidence on the bank lending process and the
     capital market response to bank loan announcements,” Journal of Economics 25, 99-122.

Miller, M., and F. Modigliani, 1961, “Dividend policy growth and the valuation of shares,”
        Journal of Business 34, 411-433.

Miller, M. H., and K. Rock, 1985, “Dividend policy under asymmetric information,” Journal of
        Finance 40, 1031-1051.

Rajan, R., 1992, “Insiders and outsiders: The choice between informed and arm‟s length debt,”
       Journal of Finance 47, 1367-1400.

Ravid, S. A., and O. H. Sarig, 1991, “Financial signaling by committing to cash outflows,”
       Journal of Financial and Quantitative Analysis 26, 165-180.

Ross S, "The determination of financial structure: the incentive signalling approach," Bell
       Journal of Economics, November, 629-650.

Sharpe, S., 1990, “Asymmetric information, bank lending and implicit contracts: A stylized
       model of customer relationships,” Journal of Finance 45, 1069-1087.

Smith C Jr, 1977, Substitute methods for raising additional capital: rights offerings versus
       underwritten offerings," Journal of Financial Economics, December 1977, 273-307.

                                                    TABLE 1

                     Corporate & Foreign Bonds – Holdings and Net Purchases

                            December 31 1990 and 1999 (- indicates net sales)

                                                       $ billions

                                                 Holdings                                     Net Purchases

    Type of Investor                1990         %           1999       %          1990         %        1999   %
           Total1                1,706           100      4,551         100     125.2           100    452.5    100
       Households2                192            11.3      596          13.1    44.9            35.9   14.8     3.3

      Rest of world3              217            12.7      817          18.0     5.3            4.2    157.5    34.8

          Banking                  89            5.2       220          4.8      4.6            3.7    38.9     8.6

   Savings institutions            76            4.5       112          2.5     -19.3          -15.4   23.2     5.1

      Life insurance              567            33.2     1,202         26.4    56.5            45.1   71.1     15.7

     Other insurance               89            5.2       174          3.8     10.4            8.3    -1.9      -

  Private pension funds           146            8.6       336          7.4     14.9            11.9   34.6     7.6

 State/ local retirement          180            10.6      305          6.7      8.5            6.8    25.6     5.7

  Money market funds                2            0.01      124          2.7      -1.7          -0.01   42.5     9.4

   Other mutual funds              59            3.5       369          8.1      4.7            3.8    29.3     6.5

Source: US Census Bureau, Statistical Abstract of the United States: 120th Edition, 2000 (page 523).

1 Includes other types not shown separately
2 Includes non-profit organizations
3 Holdings of U.S. issues by foreign residents

                                                           TABLE 2

                                         Summary Statistics: Overall Sample
                       First observation is the mean, second is the median. All observations
                       are 1985-1999, except for the first two rows which are for 1981-1999.

                         Div/                   Total      Total       Profit     Market to     Invest      Debt     Tangibility
                         EBIT      DPS/EPS      Sales      Assets       %          Book                     Ratio
                                                $mm        $mm

    Total Sample        0.102       0.261      1,008       1,954       -14.2        2.98         1.69      0.555       0.305
    n = 127,516         0.000       0.000        65         76         2.34         1.60        0.250      0.228       0.239

    Total Sample        0.263       0.674      2,251       4,582       5.12         2.76        0.254      0.240       0.336
   (Positive DPS)       0.182       0.332       326         461        4.35         1.51        0.203      0.217       0.288
     n = 49,300
    Rated Firms         0.179       0.429      3,923       8,458       1.70         2.63        0.995      0.373       0.384
    n = 18,675          0.110       0.194      1,175       1,778       2.87         1.69        0.195      0.332       0.344

    Rated Firms         0.268       0.644      5,158      11,613       3.34         2.40        0.295      0.332       0.399
   (Positive DPS)       0.199       0.375      1,763       2,948       3.67         1.73        0.182      0.310       0.366
     n = 12,452
  Non-Rated Firms       0.086       0.207       450         792        -20.3        3.28         2.01      0.601       0.275
    n = 85,547          0.000       0.000        39          47        1.70         1.69         .282      0.189       0.202

  Non-Rated Firms       0.288       0.697      1,228       2,324       6.32         3.43         1.17      0.208       0.291
   (Positive DPS)       0.179       0.315       158         250        4.38         1.57        0.223      0.165       0.237
     n = 25,429
  Investment Grade      0.254       0.519      6,523      15,365       4.59         2.60        0.276      0.279       0.398
        Firms           0.214       0.394      2,359       4,127       4.23         1.87        0.186      0.161       0.361
      n = 8,748
  Investment Grade      0.271       0.552      6,681      15,760       4.63         2.56        0.261      0.276       0.404
        Firms           0.224       0.416      2,410       4,165       4.31         1.85        0.181      0.261       0.369
   (Positive DPS)
      n = 8,221
Non-Investment Grade    0.112       0.351      1,632       2,366       -0.09        2.64         1.57      0.457       0.371
        Firms           0.000        0.00       604         787        1.71         1.51        0.208      0.417       0.331
      n = 9,924
Non-Investment Grade    0.263       0.823      2,201       3,558       2.04         2.10        0.358      0.378       0.389
        Firms           0.135       0.253       897        1,302       2.60         1.51        0.182      0.254       0.357
   (Positive DPS)
      n = 4,227
    Div/EBIT is dividends divided by earnings before interest and taxes; DPS/EPS is the regular dividend payout; profit
    is net income before extraordinary items dividend by total assets; invest is capital expenditures divided by lagged net
    fixed assets; debt ratio is total debt divided by total assets and tangibility is net fixed assets divided by total assets.

                                                           TABLE 3

                                        Summary Statistics: Matched Sample
                       First observation is the mean, second is the median. All observations
                          are 1985-1999, except the first two rows which are for 1981-99.

                         Div./      DPS/EPS      Total      Total       Profit    Market to     Invest     Debt      Tangibility
                         EBIT                    Sales      Assets       %         Book                    Ratio
                                                 $mm        $mm

  Matched Sample         0.190       0.359       2,751      5,690        1.81        2.55        1.63      0.328       0.350
    n = 34,154           0.078        .126        559        802         3.09        1.68       0.293      0.293       0.320

  Matched Sample         0.308       0.582       3,903      8,424        4.32        2.34       0.349      0.276       0.380
  (Positive DPS)         0.193       0.176        928       1,618        3.94        1.70       0.196      0.261       0.339
    n = 21,033

Matched Sample Rated     0.174       0.425       3,999      8,621        1.71        2.72        1.07      0.373       0.372
     n = 17,077          0.104       0.185       1,191      1,736        2.87        1.74       0.205      0.328       0.329

Matched Sample Rated     0.263       0.643       5,304     11,957        3.81        2.48       0.306      0.308       0.386
   (Positive DPS)        0.195       0.367       1,824      2,980        3.72        1.78       0.190      0.286       0.347
     n = 11,278

Matched Sample Non-      0.206       0.291       1,503      2,758        1.91        2.37        2.19      0.282       0.359
        rated            0.050       0.074        293        366         3.33        1.62       0.237      0.245       0.311
     n = 17,077

Matched Sample Non-      0.360       0.511       2,284      4,339        4.91        2.18       0.398      0.240       0.373
       Rated             0.191       0.465        440        635         4.22        1.61       0.203      0.216       0.331
   (Positive DPS)
     n = 9,755
    Div/EBIT is dividends divided by earnings before interest and taxes; DPS/EPS is the regular dividend payout; profit
    is net income before extraordinary items dividend by total assets; invest is capital expenditures divided by lagged net
    fixed assets; debt ratio is total debt divided by total assets and tangibility is net fixed assets divided by total assets.

                                                TABLE 4

                                         Industry Breakdown

            The first numbers in each row are for all observations; the second are for positive
            dividends only.

                 SIC<2000   SIC<4000)     <SIC<5000        SIC<6000    SIC<7000      SIC<9000     SIC<10000

                                          Transport                    Finance/Ins                Public
                 Resource   Manufactur    & Public         Wholesale   urance/       Service      Admin/other
                            ing           Utilities        & Retail    Real Estate
127,516             7.4%       41.8%          9.3%            11.0       15.5%         14.4%         0.7%

49,300              5.1%       39.2%         14.2%            9.1%       25.4%          6.8%         0.1%
18,676              5.9%       37.6%         21.0%           10.3%       17.3%          7.8%)        0.1%

12,452              4.8%       37.2%         22.9%            8.5%       22.3%          4.2%         0.0%

85,547             7.2%        42.1%          6.6%           11.0%       15.6%         16.7%         0.7%

25,429              5.4%       37.8%          9.8%            8.4%       30.7%          7.9%         0.1%

                                            Industry Breakdown

Zero dividends      73.4%      63.7%         40.9%            67.8%       36.4%          81.8%        95.2%
Positive DPS        26.6%      36.3%         59.1%            32.2%       63.6%          18.2%        4.8%
Rated               15.2%      16.3%         40.8%            17.0%       19.5%          9.2%         2.8%
Not Rated           84.8%      83.7%         59.2%            83.0%       80.5%          90.8%        97.2%

Rated               30.5%      32.5%         53.3%            33.2%       26.2%          20.9%        5.9%
Positive DPS
Non-Rated           69.5%      67.5%         46.7%            66.8%       73.8%          79.1%        94.1%
Positive DPS

                                                      TABLE 5

                                    Predicting Dividend Payments
The dependent variable equals one if it is a dividend paying observation and 0 otherwise. The first value in each row
is the estimated coefficient and the second the t-statistic. The last row reports the adjusted R square or the pseudo R
square in the case of logit estimates. All estimates include robust standard errors where independence is assumed
across firms (groups) but not within firm-year observations. The number of observations is after the R Square.

                                                      Dividend Models (Dividend indicator=1)
                                          All Logit except model 1. Model 8 is a fixed effects Logit model.

 Independent             #1          #2             #3                #4            #5           #6       #7               #8
   Variable          (Linear)
   Constant            -.160       -0.864         -1.261             -1.659       101.306      -0.570   106.7
                      22.32        -41.55         -16.76             -17.62        11.55       -19.91   14.31
Rating indicator                    1.554          1.489              1.501        1.500
  (= 1 rated)                       35.73          32.53              32.64        32.64
 Profitability        -0.002                                                                    6.370   6.494         2.479
                       -2.18                                                                    26.35   23.92         13.97
Investment rate       -0000                                                                    -0.002   -0.001         .000
                       -1.41                                                                    -0.32    -0.32         0.70
Market to Book        -0.000                                                                   -0.000   -0.000         .001
                       -1.61                                                                    -3.22    -3.82         1.68
   Debt ratio         -0.037                                                                    -.002    -.312        -2.033
                       -2.19                                                                    -0.03    -3.77        -16.58
  Log of total          .112
    assets             95.49
  Tangibility          0.134
      Time                                                                         -0.051               -0.054        -0.101
                                                                                    11.67               -14.37        -10.98
SIC indicators                                      5/6                5/6           5/6                  5/6
                                                significant        significant   significant
Year indicators                                                        All           All                11/18             13/17
                                                                   significant   significant
   (Pseudo) R          30.33        6.39          13.31               13.61         13.61      11.09    16.27             7.66
   Squared %
 # observations       105425       104223         104223            104223        104223       105425   105425        25305

                                                        Table 6

                                    Capital Market Access Models

The dependent variables are: for models 1 & 2 a dividend indicator variable (one for a dividend); models 3, 4, 5 and
6 a rating indicator variable (one for a rating); and models 7 and 8 a rating group ordered 28-1 with 28 equivalent to
AAA. The first value in each row is the estimated coefficient and the second the t-statistic. The last row reports the
adjusted R square or the pseudo R square in the case of logit estimates. All estimates include robust standard errors
where independence is assumed across firms (groups) but not within firm-year observations. The number of
observations is after the R Square. For each set of independent variables there are standard Logit model estimates
and conditional or fixed effects (FE) Logit model estimates.

                   Dividend:                   Rating Access             Rating Access                  Rating
                  Market Access                   Model                     Model                    Classification
                    #1             #2           #3             #4          #5           #6           #7         #8

                   Logit      Conditional      Logit      Conditional     Logit    Conditional       Logit      Conditional
                              Logit (FE)                  Logit (FE)               Logit (FE)                   Logit (FE)
 Constant          213.9                       -14.77                     49.56
                   22.20                        -1.33                      3.64
Profitability      4.753         1.851          2.930       1.370         0.619       0.217          7.426        6.828
                   12.50          9.28          13.11       6.870         1.700       1.040          11.89        10.37
Investment         0.000         0.000         -0.000       -0.038        0.000       -0.087         0.000        0.002
   rate             0.06          0.86          -0.31        -2.43         0.42        -3.67          0.29         4.43
 Market to        -0.000         0.002         -0.000       0.000         0.003       0.004         -0.000        0.002
   Book            -2.75          1.88          -0.44        0.66          1.38        0.70           -.24         1.17
Debt ratio        -2.139         -3.548         1.716       4.021         1.451       3.454         -3.891        -3.354
                  -18.16         -24.10         10.11       23.67          5.85       19.44         -15.57        -14.67
Log of total       0.731         1.544                                    1.049       2.095                       0.848
  assets           47.91         37.85                                    39.23       32.45                       24.22
Tangibility        1.431         0.267                                    0.216       -1.701                      1.128
                   13.01          1.32                                    1.450       -5.140                       6.35
    Time          -0.109         -0.223        0.006        0.205        -0.029       0.008         -0.022        -0.060
                  -22.54         -21.48         1.14        14.82         -4.19        0.53          -2.79         -7.40
    SIC             4/6                         2/6                        2/6                        4/6           3/6
    Year           16/17         10/17          5/13        10 /13        4/13        10/13          7/13          5/13
 (Pseudo) R        37.50         16.75          8.97        18.75        45.48         31.37         8.18          15.29
 Squared %
observations      105425         25305         86280        15155        86280        15155         14585          14585

                                                 TABLE 7

                                 Lintner Model Regression Estimates

 The dividend per share at time „t‟ (DPSt) is regressed against the lagged dividend and earnings per share
 with and without an interaction indicator variable constructed as the rating indicator variable times the
 lagged dividend. For each regression the first is over all observations including zero dividend observations
 and the second over positive dividend payments only. In each case the first row is the coefficient on the
 independent variable and the second the t statistic. Time period is 1981-1999 for all observations and
 1985-1999 for the subset with bond ratings.

                                                                       DPS            EPS
                  Observa      Constant      DPSt-1        EPSt     indicator      indicator                    Adj.R2
                   tions                                             for Debt       for Debt    Optimal          (%)
                                                                      Rating         Rating     payout

       1.          127,516       14.55        0.894         0.138                                                81.4
 Total Sample                     1.18        12.29          3.95                                >100%
  (1981-99)        46,707        13.86        0.892         0.176                                                87.9
                                 20.22        10.68          3.05
      2.           127516        40.69       0.8936         0.138                                                81.4
 SIC & Year                       0.35        12.30          3.95                                >100%
  indicators        46706        38.79        0.892         0.176                                                87.9
                                  0.01        10.69          3.05
       3.          127516                     0.761         0.120                                 50.2%          81.4
 Fixed Effects:                              306.10        108.29
Firm indicators     46707                     0.839         0.160                                 99.0%          87.9
                                             232.30         88.18
       4.          110092       131.07         .619         0.124                                 31.0%          82.4
  Panel data                     6.13        204.08        104.19
Fixed effects &     40234       301.79         .678         0.161                                 50.0%         88.25
Auto-regression                  6.14        133.22         82.29
       5.          110092        153.9       -0.136         0.117      0.850                      40.9%          74.0
  Rated DPS                       7.93        -20.6        108.83     131.77
  interaction       40234        349.1       -0.031         0.154      0.784                      62.3%          82.9
    (FE&A)                        7.84        -3.02         86.61      83.11
       6.          110092        159.2       -0.316         0.248      1.064         -0.168        32%           74.8
  Rated DPS                       8.46       -43.89        117.99     152.68         -71.58
     &EPS           40234        384.6       -0.238         0.265       1.01         -0156         48%           82.2
  Interaction                     8.86       -21.35         85.14      95.18          -42.5
       7.          110092       169.68       -0.233        0.243       1.023         -0.167        36%          75.56
  Investment                      9.38        -43.3        122.5       187.8         -75.08
     grade          40234       450.10       -0.030        0.241       0.793         -0.129        47%          83.73
  interaction                    10.02        -3.31        77.43        88.2         -35.71
  Ravid and Sarig (1991) argue that the optimal commitment mix is such that the marginal cost of committing with
either policy is the same, so that higher quality firms will be more leveraged and pay higher dividends .
  Source: http://insurance.about.com/industry/insurance/gi/dynamic/offsite.htm.
  Smith (1977) showed that equity issue costs were about 7.0% and increased as the size of firm got smaller. Similar
effects are at work in the public debt markets, where there is a minimum issue size.
   Back up lines of credit are usually required for the commercial paper market.
   During the late 1980s the original issue non-investment grade (Junk) bond market came into existence pioneered
(partly using illegal techniques) by Michael Millkin, which then withered during the recession years 1990-1993. We
therefore expect cyclical effects in our models.
  Note that only the annual statements are audited, so an argument could be made that signaling is more important on
a quarterly basis. However, the dividend decision tends to be made on an annual basis in line with the firms‟
planning process for the forthcoming year.
   Non-investment grade debt normally starts at BB+. To maintain slightly more equal samples we divide the sample
based on BBB recognizing that many institutions will not hold lower rated investment grade debt for fear that a
rating cut will force them to sell BBB rated debt. This is mainly a concern for non-utility debt.
  In the case of complete dependence RSQX will be close to 1.0, indicating that the standard error of that coefficient
in the regression equation is seriously inflated leading to biased significance tests.
   The pseudo R square is estimated as 1  L1 / L0 , where L is the log Likelihood of the full model (L1) and a
constant only model (L0) respectively.
   Since the dependent variable is either 0, 1, having a company indicator variable fully identifies the dependent
variable unless the firm has periods of both positive and zero dividend payments.
   The analysis was repeated with the matched sample of firms, but the results were essentially identical, since the
matches themselves were created using the independent variables of size, SIC and time.
   Conceptually the intercept has to be zero, but regression results with suppressed intercepts are problematic.
   The amount of autocorrelation is not high since the Lintner model already includes the lagged dependent variable
as an independent variable. For all observations the autocorrelation coefficient is 0.2 and for dividend paying
observations 0.3.
   We tested a generalized least squares random components model, but there is significant correlation between the
independent variables and the random error terms that biases the model's coefficient estimates. The fixed effects
model is robust to this specification.
   Note again Lintner did not estimate his model over thousands of firms, he estimated it for 28 carefully screened
firms. These firms would more closely match our sample of rated firms, for these firms the regression model has
direct coefficient estimates on lagged dividends of 0.946 and on current earnings of 0.12.