Dividend Persistence and Equity Agency Costs in Banking: Evidence from the

Financial Crisis

Benoit d'Udekem By persisting in paying dividends during crises, banks not only weaken their liquidity and solvency positions; they also exacerbate systemic risks. However,

the drivers of such persistence remain elusive. In this paper, we analyse the propensity of US banks to omit or cut dividends during the 2007-09 financial

crisis. We observe that banks pay dividends to mitigate agency costs, maintain their reputation, and preserve market access. We conclude that agency costs, rather than size, induce banks to continue paying dividends during periods of

greater cash flow uncertainty. We also conclude that bank managers trade off solvency against favourable funding.

Keywords: banks; dividends; agency costs; market access.

JEL Classifications: G21, G28, G35.

CEB Working Paper N° 14/013 2014

Université Libre de Bruxelles - Solvay Brussels School of Economics and Management

Centre Emile Bernheim

ULB CP114/03 50, avenue F.D. Roosevelt 1050 Brussels BELGIUM

e-mail: ceb@admin.ulb.ac.be Tel. : +32 (0)2/650.48.64 Fax: +32 (0)2/650.41.88

Dividend Persistence and Equity Agency Costsin Banking: Evidence from the Financial Crisis

Benoit d’Udekema,1,∗

aUniversite Libre de Bruxelles, Solvay Brussels School of Economics and Management,Centre Emile Bernheim, Avenue Franklin Roosevelt 42, B-1050 Brussels, Belgium

Abstract

By persisting in paying dividends during crises, banks not only weaken their

liquidity and solvency positions; they also exacerbate systemic risks. However,

the drivers of such persistence remain elusive. In this paper, we analyse the

propensity of US banks to omit or cut dividends during the 2007-09 financial

crisis. We observe that banks pay dividends to mitigate agency costs, maintain

their reputation, and preserve market access. We conclude that agency costs,

rather than size, induce banks to continue paying dividends during periods of

greater cash flow uncertainty. We also conclude that bank managers trade off

solvency against favourable funding.

Keywords: banks; dividends; agency costs; market access.

JEL Classification Codes: G21, G28, G35.

∗This version: June 19, 2014. First version: May 3, 2014.Email address: bdudekem@ulb.ac.be (Benoit d’Udekem)

1I thank Ariane Szafarz for her invaluable guidance. I also thank Jose Filipe Abreu, BenoıtDewaele, Marek Hudon, Pierre-Guillaume Meon, Kim Oosterlinck, Lev Ratnovski and EmreUnlu for their helpful suggestions and comments on prior versions of this paper. I am gratefulto the members of Centre Emile Bernheim for their support.

1

1. Introduction

Despite an immediate debilitating effect on their solvency and cash positions,

banks paid dividends well into the financial crisis of 2007-09. Rather than

cutting or omitting dividends in the long-term interest of their shareholders,

lenders and deposit insurers, banks persisted in paying out dividends. By doing

so, they brought themselves, and the system they were part of, ever closer to

collapse.

This apparent absence of foresight has raised concern among scholars and

regulators alike. Scholars have attributed dividend persistence in times of crisis

to the implicit government guarantees given to too-big-to-fail banks, to short-

term debt roll-over risks (Acharya et al., 2011), or to risk-taking and risk-shifting

(Kanas, 2013). They have also attributed dividend persistence to agency costs

between shareholders and debt holders, to high leverage, or to the inability

for interconnected banks to coordinate their dividend policies (Acharya et al.,

2013). Regulators have ruled that banks should be prevented from paying out

dividends in case they do not meet stringent capital and liquidity requirements

(Rosengren, 2010), and have amended the regulatory framework accordingly.

However, as early as the 1970s, scholars noted that dividend cuts by banks

were an exception rather than the rule (Keen, 1978, 1983). Since then, they

have rarely studied the issue of dividend persistence in the banking industry,

possibly hampered by the few occurrences of bank dividend omissions in recent

times.2

In particular, the “substitute model” of La Porta et al. (2000) has not been

formally tested. This model posits that the payment of dividends substitutes for

shareholder protection. According to the model, managers who need to maintain

market access seek to uphold a reputation of moderation in shareholder wealth

appropriation through the payment of dividends, with the consequent reduction

2Empirical work conducted on bank dividend policies in the 1980s and 1990s is charac-

terised by small sample sizes. See for example Keen (1983), Bessler and Nohel (1996), Slovin

et al. (1999), Bessler and Nohel (2000), and Casey and Dickens (2000).

2

of agency costs. In the authors’ own words, the substitute model should hold

for a firm “if, for example, there is enough uncertainty about its future cash

flows that the option of going back to the capital market is always valuable”.

(La Porta et al., 2000, p. 7). Because the value of this option is substantial

for bank managers during a crisis, the substitute model is a prime contender to

explain their reluctance to cut dividends.

In this paper, we take advantage of the unusual volatility of bank dividend

policies during and after the 2007-09 financial crisis to test the substitute model.

Following Benito and Young (2003), we use random effects probit regressions

to identify the characteristics of US banks that either omitted or cut dividends

during this period. We regress the propensity to either omit or cut dividends

against lagged measurements of institutional ownership dispersion and concen-

tration in order to assess the influence of agency costs on bank dividend policies.

We also incorporate accounting variables to gauge the influence of short-term

borrowings and deposits on these policies. Because of the fast-evolving nature

of the crisis, we adopt a more granular approach than that documented in the

literature to analyse financial statements and joint dividend decisions.

We find that banks whose institutional ownership is widely dispersed are

less likely to omit dividends, and that banks whose institutional ownership is

highly concentrated are more likely to cut them. We also observe that greater

dependency on short-term funding, and on markets in general, is associated

with a lower propensity to both omit and cut dividends. Finally, we note that,

after controlling for agency costs, larger banks do not exhibit greater dividend

persistence than smaller ones.

Our findings suggest that bank managers consider dividends as crucial to

mitigate agency costs and maintain market access. For that reason, during pe-

riods of stress, managers of banks with a dispersed shareholding may be more

inclined to pay, or even increase, dividends rather than cut or omit them. By

contrast, managers of banks benefiting from the oversight of concentrated insti-

tutional shareholders are more willing to cut dividends. However, they would

not go as far as omitting them for fear of potential immediate repercussions

3

on their access to markets. When they pay dividends, bank managers thus at-

tribute greater weight to their reputation, and the liquid position they consider

it implies, than to longer-term solvency issues. Because of this, agency costs in

the banking industry have systemic ramifications.

The paper challenges recent thinking in two respects. First, it concludes

that size is not a motive for dividend persistence in banking. Agency costs

and market access, rather than too-big-to-fail considerations, drive managers

to favour paying dividends over preserving solvency. Second, in contrast with

Laeven and Levine (2009), the paper finds that when crises strike, banks with

more powerful owners have a greater propensity to reduce risk-taking by cutting

dividends. Risk-taking incentives are thus not one-sided, but multi-faceted or

time-dependent.

The remainder of the paper is structured as follows. Section 2 reviews prior

literature and details the hypothesis. Section 3 describes the data set and the

empirical approach. Section 4 examines whether agency costs are associated

with a lower propensity to either cut or omit dividends, in order to test the hy-

potheses; it is backed by the robustness tests presented in Appendix A. Section

5 presents the conclusions.

2. Related Literature

Ever since Miller and Modigliani’s (1961) dividend irrelevance theorem, fi-

nancial economists have been puzzled by dividends. The theories they have

proposed to explain dividend policies hypothesise that dividends send credible

signals of firm value to investors, that they induce tax- or cash flow-driven clien-

teles or that they help mitigate agency costs between managers and investors.

Over the years, scholars have tested these theories on different cross-industry

samples and have found limited empirical support for signalling theory and

greater success with clientele theory.

Nonetheless, agency cost explanations, as suggested by Rozeff (1982), appear

to have taken empirical prominence. In particular, the two models introduced

4

by La Porta et al. (2000) have given rise to a substantial amount of literature

evidencing the influence of agency costs on dividend policies. According to the

first model, the “outcome” model, dividends help mitigate agency costs by giv-

ing minority shareholders legal powers to compel managers to paying out cash.

Managers have essentially no choice but to disburse dividends when there is

excess cash at hand and minority shareholder rights are enforceable. According

to the second model, the “substitute” model, dividends help managers estab-

lish a reputation for moderation in appropriating shareholder wealth for their

own use. By doing so, they preserve their option to access markets, which is

particularly relevant when cash flows are uncertain.

In general, La Porta et al. (2000) find empirical support for the outcome

model. They conclude that “firms appear to pay out cash to investors because

the opportunities to steal or misinvest it are in part limited by law, and because

minority shareholders have enough power to extract it” (La Porta et al., 2000,

p. 27). Similarly, Faccio et al. (2001) find that East Asian groups in which large

family shareholders can collude and expropriate minority shareholders pay lower

dividends than other firms in the same region. By contrast, European groups

with several large shareholders pay higher dividends compared with their peers.

The authors reckon that in European legal systems, contrary to East Asian ones,

dividends contribute to reducing agency costs. Brockman and Unlu (2011) relate

agency costs to transparency. They show that the substitute model dominates

in opaque environments, whereas the outcome model dominates in transparent

ones. Transparency increases the pressure on managers to pay out cash to

shareholders; opacity drives them to establish a reputation in the market.

The banking industry is notably absent from cross-industry samples, be-

cause it is regulated and responds to different rules. Gupta and Walker (1975,

p. 515) note that “the banking firms [...] have unique characteristics which

suggest that their dividend disbursal practices may differ from those of other

industries”. They suggest that capital and liquidity requirements as well as the

presence of deposit insurance are sources of differences with non-regulated firms.

Subsequent research confirms that bank dividend policies are different but share

5

similarities with those of non-regulated firms.

First, the signalling hypothesis holds firmly in banking, unlike in other in-

dustries. According to Keen’s (1978) “traditional view,” banks are reluctant to

cut dividends because of the potential deposit flight this might cause and re-

sulting rise in funding costs. Keen (1983) and Bessler and Nohel (1996) provide

evidence in support of this hypothesis. The latter two authors also find that

larger banks experience a steeper decrease in value when they cut dividends,

as if the market were expecting fewer adverse consequences for smaller banks

after a dividend cut. Slovin et al. (1999) and Bessler and Nohel (2000) further

observe that the signal sent by dividend cuts is contagious and affects the stock

prices of competing banks in a similar way to those of the cutting banks. More

recently, Abreu and Gulamhussen (2013) find empirical support for signalling

during the financial crisis. Hirtle (2014) hypothesises that, during the same pe-

riod, bank managers were reluctant to reduce dividends because of the negative

signal this would send in highly uncertain times.

Second, there is evidence that agency costs influence bank dividend policies.

Casey and Dickens (2000) find a relationship between dividend payout ratios

and shareholder dispersion, in support for Rozeff’s (1982) theory, although with

significant differences compared with other firms. Dickens et al. (2002) observe

that dividend yields are inversely related to insider ownership and positively

related to bank size. Abreu and Gulamhussen (2013) also find that the degree

of shareholder independence, together with size, profitability and growth, drove

the dividend payout ratios of US banks before and during the financial crisis.

They conclude that the Fama and French (2001) characteristics hold across both

periods and that “dividends compensate for the need for monitoring” (Abreu

and Gulamhussen, 2013, p. 63). In the Norwegian banking industry, Bøhren

et al. (2012) uphold the substitute model after comparing the dividend policies

of commercial banks, which are fully shareholder-owned, with those of savings

banks, in which ownership is dispersed between shareholders, depositors, em-

ployees, and community citizens.

In fact, dividend policies consistent with the substitute model were suggested

6

in the 1940s (Robinson, 1948, p. 407): “Those [banks] with close-knit ownership

could follow [dividend] policies best adapted to the advantage of both the bank

and its owners. Banks with wider distribution of stock ownership could afford,

other things being equal, to consider a more generous distribution of earnings”.

In this paper, we formally test whether US bank managers have been following

Robinson’s (1948) guidelines, and whether they have been driven to do so by

market access considerations.

3. Data and Empirical Methodology

3.1. Sample

Our initial sample includes all listed US firms categorised as banks in the

Industry Classification Benchmark (ICB) with over USD 1 billion in total assets

as at 31 December 2006.3 The sample covers the period between 1 January 2004

and 31 December 2012 so that bank dividend policies can be observed before,

during and after the financial crisis of 2007-09.

We collected the histories of regular cash dividends, stock prices and quar-

terly accounting variables from Bloomberg. We obtained quarterly institutional

holdings from the Thomson-Reuters Institutional Holdings Database (13F).

This database aggregates the quarterly holdings reported to the Securities and

Exchange Commission (SEC) by banks, brokers-dealers, insurance companies,

pension funds, investment companies, not-for-profit institutions, colleges and

foundations, under Section 13F of the Securities Exchange Act of 1934. We also

sourced analyst coverage data from the Thomson-Reuters I/B/E/S Detailed

US File, a database of individual earnings estimates from a majority of sell-side

analysts covering listed US firms.

3The sample includes bank holding companies categorised as National Commercial Banks

(SIC Code 6021), State Commercial Banks (SIC Code 6022), Commercial Banks Not Else-

where Classified (SIC Code 6029), Federal Savings Institutions (SIC Code 6035) and Savings

Institutions (SIC Code 6036).

7

During the crisis, many US banks received capital support from the US Trea-

sury through the Capital Purchase Program (CPP), itself part of the Troubled

Asset Relief Program (TARP). Such support is likely to have influenced bank

dividend policies and must therefore be controlled for in regressions. We obtain

transaction-level CPP data from the US Treasury.4

We reconcile data sets by CUSIP (Thomson-Reuters and Bloomberg data),

by name and by US state (US Treasury data). After exclusion of observations

with missing data or negative equity, and of those banks that never pay divi-

dends (19 banks) or pay them irregularly (6), the panel comprises 287 banks.

It is unbalanced either because some bank holdings have become listed, been

acquired, or failed during the period, or because of exclusions.

Prior literature has mostly related dividends to the financial statements of

the prior accounting year (e.g. Fama and French (2001)). Our analysis requires a

more granular approach to match dividend decisions with financial statements

in a period during which financial statements evolved rapidly. Nonetheless,

banks, like other firms, synchronise their dividend declaration and accounting

cycles (e.g. Aharony and Swary (1980)). In our sample, most banks declare

dividends together, shortly before or after they disclose their quarterly financial

statements, within a period of 31 days from the end of an accounting quarter.

We assume that dividend declarations made during the period starting 60

days before and ending 31 days after an accounting quarter reflect a bank’s

situation as reported in the financial statements of the previous quarter. We

relate dividends to the institutional ownership records at the end of the quarter

preceding dividend declarations in order to avoid issues of endogeneity. And,

when no dividend is declared and there is no corresponding announcement, we

4We obtain the amount and type of capital support received by each bank, as well

as the dates on which they received it and when it was repaid or disposed of on the

market, from the Transactions Report - Investment Programs dated 26 July 2013, as

available from http://www.treasury.gov/initiatives/financial-stability/reports/Pages/TARP-

Investment-Program-Transaction-Reports.aspx.

8

assume that a dividend declaration takes place 91 days after the previous one.

Because of this approach, combined with the availability of dividend data in

our sample until 31 December 2012, the end of sample period must be brought

back by one quarter to the accounting quarter ending 30 September 2012. Also,

because of the necessity to account for dividend persistence based on past quar-

ter dividend declarations, as detailed in Section 3.2, the start of the sample

period must be pushed forward by one quarter to start with the accounting

quarter ending 30 June 2004. After these eliminations, the sample comprises

7,589 bank-quarter observations.

Figure 1 plots the proportion of banks that paid or omitted (top) and cut or

maintained/increased dividends (bottom), quarter by quarter over the sample

period. The figure suggests that banks reacted belatedly to the outbreak of the

financial crisis. They did not start cutting dividends before the last quarter of

2007. Before that, they persistently paid dividends. The number of banks that

maintained or increased dividends reached its lowest point in the first quarter of

2009, whereas the number that omitted dividends peaked in the last quarter of

2010. Many banks repeatedly reduced dividends in the heat of the crisis before

omitting them, consistent with prior evidence that banks omit dividends as a

last resort. After reaching the peak of dividend omissions in the last quarter of

2010, banks slowly started paying dividends again.

Regression variables are defined in Table 1. The dispersion and concen-

tration of institutional shareholders, as well as proxies for reliance on markets

and for favourable funding, are our main variables of interest. If agency costs

drive dividend persistence, we should observe a positive sign for shareholder

dispersion (NUM INSTIT OWNERS) and/or a negative sign for shareholder

concentration (INSTIT OWN HHI). If market access dictates dividend poli-

cies, we expect differentiated results between dividend omission and dividend

cut regressions, such that the severity of omissions appears more significant

than that of cuts. Also, we may find a positive sign for short-term funding

(ST BORROWINGS TO ASSETS) and/or deposits (DEPOSITS TO ASSETS),

if banks are less willing to cut when they are more reliant on those funding

9

Figure 1: Evolution of the proportion of banks in the sample paying, omitting or never paying

(top) and maintaining/increasing or cutting (bottom), quarter by quarter over the period

between 30 June 2004 and 30 September 2012.

0.0

0.2

0.4

0.6

0.8

1.0

Time

Per

cent

age

of B

ank

Div

iden

d D

ecis

ions

in S

ampl

e

● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●● ●

●●

● ●● ● ● ● ●

●

2005 2006 2007 2008 2009 2010 2011 20122005 2006 2007 2008 2009 2010 2011 2012

Dividend Paid

Dividend Omitted

0.0

0.2

0.4

0.6

0.8

1.0

Time

Per

cent

age

of B

ank

Div

iden

d D

ecis

ions

in S

ampl

e

● ● ● ● ●● ● ●

●● ● ● ● ●

●●

● ●

●

●

●

● ●● ●

● ● ● ●●

● ● ● ●

2005 2006 2007 2008 2009 2010 2011 20122005 2006 2007 2008 2009 2010 2011 2012

Dividend Maintained or Increased

Dividend Cut

10

sources. However, we may also find a negative sign for short-term funding and

deposits if banks with ample short-term funding are less concerned about the

impact of cutting or omitting dividends. Alternatively, if favourable access to

equity markets drives dividend policies (as shown in robustness tests), we should

find a positive sign for the intensity of analyst coverage (NUM ANALYSTS). By

contrast, if too-big-to-fail considerations dictate dividend persistence, we should

observe a positive sign for bank size (LOG ASSETS) and limited statistical sig-

nificance for our main variables of interest.5

Summary statistics are provided in Table 2 for subsamples partitioned by

quarterly dividend decision. The statistics suggest that banks that omit divi-

dends differ significantly from those that pay them. First, the omitting banks

are owned by a less dispersed institutional investor base, consistent with the

substitute hypothesis; also, their largest institutional shareholder has a larger

equity stake. Second, the omitting banks are reliant more on deposit funding

and less on short-term borrowings. Third, the omitting banks are smaller on av-

erage than dividend-paying banks. Larger banks may have more to lose or may

feel protected because they are too big to fail. However, shareholder dispersion

and bank size are significantly correlated. Fourth, the omitting banks are cov-

ered by fewer sell-side analysts, with the incidental higher financing costs that

this entails (Derrien and Kecskes, 2013). Fifth, the omitting banks hold more

cash than the paying banks. They may be more prudent; equally, they may be

facing a cash crisis. Sixth, the omitting banks report fewer retained earnings on

their balance sheet, consistent either with the lifecycle hypothesis of dividends

(DeAngelo et al., 2006) or with the fact that they may have exhausted their

capital buffers.

5While the term too-big-to-fail bank sometimes refers to a bank of systemic importance,

independently of size, we use the expression in the narrower meaning provided by Stern and

Feldman (2004, p. 13): “A TBTF regime is a policy environment in which uninsured creditors

expect the government to protect them from prospective losses from the failure of a big bank;

big banks are said to be too big to fail in countries following such a regime.” See also Brewer

and Jagtiani (2013) and Laeven et al. (2014).

11

Table 1: Definition of variables used in regressions. Dividend declarations made on date d

where Q− 60 ≤ d ≤ Q+ 31 and Q a quarter end date, are related to the financial statements

of the quarter ending on Q. Accounting variables are those of the quarter ending on Q.

Institutional ownership is measured on Q′ where Q′ is a quarter end date and Q′ < d.

Market measurements are also made during the period [Q−60, Q+31]. Sources: institutional

ownership data from Thomson-Reuters Institutional Holdings Database (13F), accounting

variables from Bloomberg, analyst coverage based on the I/B/E/S US Detail File, Troubled

Asset Relief Program (TARP) Capital Purchase Program (CPP) from the US Treasury, TED

spread based on data from the Federal Reserve Bank of St. Louis.

Variable Description

Institutional ownership

INSTIT OWN HHI Concentration of institutional shareholders, mea-

sured as a Herfindahl-Hirschman Index (HHI)

NUM INSTIT OWNERS Number of institutional shareholders

TOP OWNER PERC Equity stake held by the largest institutional

shareholder, expressed in percentage of total

shares outstanding

Accounting and related measurements

CAP TO RBC Risk-based capital ratios

CASH MKT SEC TO ASSETS Ratio of cash holdings and marketable securities

to total assets

DEPOSITS TO ASSETS Ratio of total deposits to total assets

EQUITY TO ASSETS Ratio of total equity to total assets

LOG ASSETS Natural logarithm of total assets, used as a mea-

surement of bank size

QCH ASSETS Difference between total assets on Q and on Q−1,

divided by total assets on Q (following Fama and

French (2001))

QUARTER ROA Quarterly return on assets, defined as quarter

earnings divided by total assets

RET EARN TO ASSETS Ratio of retained earnings to total assets

ST BORROWINGS TO ASSETS Ratio of short-term borrowings to total assets

Sell-side analyst coverage

NUM ANALYSTS Number of analysts having issued at least one rec-

ommendation on a bank over the prior 90-day pe-

riod

12

Time effects

AVG TED Average TED spread measured over the period

[Q − 60, Q + 31]. The TED spread is the differ-

ence between the 3-month LIBOR and secondary

market rates on 3-month US Treasury Bills.

Dit−1 Binary variable taking the value 1 if a bank de-

clared a dividend during the quarter ending on

Q− 1 and 0 otherwise.

UNDER TARP Binary variable taking the value 1 during the pe-

riod when a bank receives financial support from

the US Treasury under TARP and 0 outside of

this period (or if the bank does not receive such

support).

Banks that cut dividends have a greater resemblance to those that pay div-

idends than to those that omit them. However, consistent with prior academic

evidence, the cutting banks are unprofitable and larger. Besides, in common

with the dividend omitting banks, the cutting banks have higher leverage and

lower capital ratios. Both the cutting and omitting banks also experience low as-

set growth or even contraction and are much more likely to have received support

from the US Treasury through the Capital Purchase Program. Interestingly, the

TED spread suggests that unlike dividend cutting decisions, omission decisions

are made during periods of lower stress.

3.2. Empirical Approach

Following Benito and Young (2003), the empirical approach models dividend

decisions as binary outcomes. For dividend omissions, outcomes take the value

1 when banks pay a dividend, regardless of its level, and 0 when they omit it.

For dividend cuts, outcomes take the value 1 when banks maintain or increase

the nominal value of a dividend compared to the last paid dividend, and 0 when

they pay a lower (but non-zero) nominal dividend. Distinguishing omissions

from cuts makes it possible to compare the drivers and implicit severity of the

two dividend decisions.

13

Table 2: Summary statistics: means and (standard deviations) of dependent and control

variables, clustered by the dividend status of sample banks in each quarter. Dividend Paid

clusters bank-quarters during which a dividend was declared. Dividend Omitted clusters

bank-quarters during which no dividend was paid despite a prior history of dividend payment.

Dividend Maintained or Increased clusters bank-quarters during which banks maintained or

increased their nominal quarterly dividend. Dividend Cut clusters bank-quarters during which

banks reduced their nominal quarterly dividend. ***, ** and * denote statistical significance

levels of 1%, 5%, and 10% of a t-test between the mean of each subsample and that of the

Dividend Paid subsample.

VariableDividend

Paid

Dividend

Omitted

Dividend

Maintained

or Increased

Dividend

Cut

N 6,836 799 6,593 243

NUM INSTIT OWNERS 137.828 56.538*** 137.047 159.016

(197.390) (100.190) (196.083) (229.506)

INSTIT OWN HHI 0.103 0.194*** 0.102 0.120

(0.121) (0.191) (0.119) (0.182)

TOP OWNER PERC 0.070 0.078** 0.070 0.071

(0.051) (0.099) (0.051) (0.046)

DEPOSITS TO ASSETS 0.721 0.778*** 0.721 0.714

(0.098) (0.096) (0.098) (0.098)

ST BORROWINGS

TO ASSETS

0.064 0.036*** 0.064 0.064

(0.062) (0.049) (0.062) (0.061)

EQUITY TO ASSETS 0.097 0.060*** 0.097 0.087***

(0.031) (0.040) (0.031) (0.023)

LOG ASSETS 8.600 7.986*** 8.588 8.902***

(1.516) (1.076) (1.505) (1.764)

NUM ANALYSTS 6.650 2.323*** 6.654 6.543

(6.921) (3.393) (6.935) (6.546)

CASH MKT SEC

TO ASSETS

0.046 0.078*** 0.046 0.045

(0.047) (0.053) (0.047) (0.044)

QUARTER ROA 0.002 -0.004*** 0.002*** -0.002***

(0.003) (0.023) (0.003) (0.008)

14

CAP TO RBC 0.140 0.128*** 0.140 0.135**

(0.038) (0.049) (0.038) (0.032)

RET EARN TO ASSETS 0.038 -0.012*** 0.038 0.028***

(0.032) (0.054) (0.032) (0.031)

Dit−1 0.996 0.108*** 0.995 1.000***

(0.066) (0.310) (0.067) (0.000)

UNDER TARP 0.118 0.526*** 0.106** 0.444***

(0.322) (0.500) (0.307) (0.498)

QCH ASSETS 0.019 -0.028*** 0.019 0.007***

(0.052) (0.249) (0.051) (0.061)

AVG TED 57.070 48.705*** 55.456* 100.849***

(50.105) (49.394) (48.837) (62.784)

We estimate random effects probit regressions with maximum likelihood and

controlling for time effects:

Prob (Dit = 1 | Dit−1, Xit, Tt) = Φ (αi + τT ′t + βX ′it + δDit−1 + εit) (1)

where Prob is the probability operator, i indexes banks and t, quarters, Dit is

the binary outcome for bank i at time t as described above, Xit is a vector with

the characteristics of bank i at time t, Tt is a vector of bank-independent control

variables at time t, Φ is the cumulative distribution function of the standard

normal distribution, τ , β, and δ are the regression parameters to be estimated

and εit is an error term.

We account for dividend persistence via the autoregression term Dit−1.

We include random effects, that is, company-specific, time-independent fac-

tors that are assumed to be randomly distributed, via the variable αi, with

αi ∼ N(0, s2α).6 Unlike in Benito and Young (2003), we control explicitly for

time effects through Tt and through bank-specific time effects included in Xit

rather than through fixed effects.7 We measure bank-independent time effects

through the average TED spread during the dividend quarter (AVG TED),

6By contrast, the error term εit is both time and bank dependent, with εit ∼ N(0, s2ε ).7Unreported regressions show that replacing AVG TED by time fixed effects leads to similar

15

which reflects the intensity of the liquidity stress experienced by banks during

the crisis. Bank-specific time effects consist of the financial support provided by

the US Treasury in the context of the TARP Capital Purchase Program (UN-

DER TARP), often through the issuance of preferred stock or subordinated

debentures to the US Treasury, together with warrants.

Other controls are consistent with those in prior literature; they include

quarterly change in assets (QCH ASSETS), leverage (EQUITY TO ASSETS),

and retained earnings to assets (RET EARN TO ASSETS). Controls specific

to the banking industry address prior claims made in the literature. First,

cash and short-term securities holdings (CASH MKT SEC TO ASSETS) and

quarterly profits deflated by assets (QUARTER ROA) control for Keen’s (1978,

p. 5) claim that “no banker would cut dividends unless his bank were in a

severe earnings or liquidity crunch”. Second, bank size (LOG ASSETS) and

regulatory capital (CAP TO RBC) control for scholarly claims that banks may

have continued paying dividends well into the crisis because of the implicit

government guarantees from which larger banks benefit, and because of the

distortions generated by accounting measurements of capital that “[make] even

a distressed bank appear healthy” (Acharya et al., 2011, p. 4).

4. Empirical Results

Tables 3 and 4 report estimates of the random effects probit models for

the likelihood of dividend omissions and cuts. The tables report goodness of fit

statistics appropriate for generalised linear mixed-effects models (Nakagawa and

Schielzeth, 2013). More specifically, the tables report the marginal R2, which is

concerned with the variance explained only by the tested variables (and fixed

effects, if any). The tables also report the conditional R2, also concerned with

the variance explained by random effects.

results. They suggest that Dit−1, AVG TED, and UNDER TARP adequately capture time

effects.

16

Regression results in both tables are consistent with the substitute model.

The results suggest that dividends and agency costs are intertwined and the

agency costs themselves are intimately connected to market access. The results

also indicate that dividend omissions are more severe decisions. This is because

liquidity (cash and short-term funding) influences dividend omissions but not

cuts.

Tables 3 and 4 show a connection between agency costs and dividends, but

the difference between the models is subtle. On the one hand, banks with a more

widely dispersed shareholder base are less likely to omit, but not cut, dividends.

As observed by Bøhren et al. (2012) in Norway, banks may use dividends to mit-

igate agency conflicts, which are all the more significant since shareholders are

“weak” in comparison to other stakeholders. By paying dividends, banks seek

to establish or uphold a reputation for restraint in expropriating shareholders

that cannot exert control over management actions.

On the other hand, banks with a more concentrated shareholder base have a

greater propensity to cut, but not omit, dividends. As evidenced by Faccio et al.

(2001), dividends represent a commitment mechanism for managers to dampen

agency problems, and, indirectly, sustain the market value of their firms and

preserve their access to markets. Lower dividends have a weaker dampening

effect on expropriation risks but are better than no dividends at all. Also, lower

dividends, representing changes in dividend policies, are easier to introduce

when the shareholder base is tighter, closer to management and better informed

of a bank’s situation.

Our estimations also provide clear evidence that the propensity of banks to

omit or cut dividends is connected to market access. In both the dividend and

the omission models, deposits and short-term funding relative to assets carry a

sign consistent with banks fearing that “individual depositors [may] start shying

away from those banks, [or] they [may] find it hard to sell certificates of deposit

to corporate or municipal investors” (Keen, 1978, p. 7). While only short-

term borrowings appear to be statistically significant, the regressions shown in

Appendix A.2 show that short-term borrowings and deposits have a complex

17

Table 3: Random effects probit regressions to explain dividend omissions by US banks during

the period between 30 June 2004 and 30 September 2012. The binary dependent variable

takes the value 0 if a bank omitted a dividend in a quarter and 1 otherwise. The sample

includes 287 banks and 7,589 bank quarters. Dependent variables are described in Table 1.

***, ** and * denote statistical significance levels of 1%, 5%, and 10%, respectively, and z

statistics are reported in parentheses below parameter estimates.

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

NUM INSTIT OWNERS 0.003∗∗ 0.003∗∗

(0.001) (0.001)

INSTIT OWN HHI −0.239 −0.234

(0.463) (0.498)

DEPOSITS TO ASSETS 0.370 0.537

(0.992) (1.054)

ST BORROWINGS TO ASSETS 2.547∗ 2.308

(1.509) (1.568)

LOG ASSETS −0.051 0.206∗∗∗ 0.179∗∗∗ −0.098

(0.127) (0.062) (0.067) (0.133)

EQUITY TO ASSETS 19.737∗∗∗ 19.748∗∗∗ 20.732∗∗∗ 21.389∗∗∗

(3.702) (3.596) (3.659) (3.933)

CASH MKT SEC TO ASSETS −5.370∗∗∗ −4.459∗∗∗ −4.258∗∗∗ −5.251∗∗∗

(1.366) (1.286) (1.367) (1.507)

QUARTER ROA 39.266∗∗∗ 39.952∗∗∗ 38.954∗∗∗ 38.807∗∗∗

(6.397) (5.939) (5.931) (6.832)

CAP TO RBC −1.081 −0.315 −0.383 0.088

(2.795) (2.727) (2.756) (2.955)

Dit−1 3.884∗∗∗ 3.940∗∗∗ 3.932∗∗∗ 3.952∗∗∗

(0.192) (0.189) (0.188) (0.203)

UNDER TARP −0.737∗∗∗ −0.703∗∗∗ −0.679∗∗∗ −0.716∗∗∗

(0.148) (0.142) (0.143) (0.154)

AVG TED −0.005∗∗∗ −0.005∗∗∗ −0.005∗∗∗ −0.005∗∗∗

(0.001) (0.001) (0.001) (0.001)

QCH ASSETS 4.677∗∗∗ 4.570∗∗∗ 4.484∗∗∗ 5.127∗∗∗

(1.005) (0.694) (0.653) (1.220)

RET EARN TO ASSETS 4.654∗∗ 4.344∗∗ 4.108∗∗ 4.087∗

(2.162) (2.049) (2.042) (2.230)

Constant −2.056∗∗ −4.119∗∗∗ −4.409∗∗∗ −2.544∗

(1.028) (0.675) (1.155) (1.450)

Observations 7,589 7,589 7,589 7,589

Akaike Inf. Crit. 809.358 816.555 815.887 811.970

Marginal R2 72.0% 72.0% 72.1% 72.7%

Conditional R2 81.0% 79.7% 79.7% 81.9%

18

Table 4: Random effects probit regressions to explain dividend cuts by US banks during the

period between 30 June 2004 and 30 September 2012. The binary dependent variable takes

the value 0 if a bank cut its nominal dividend in a quarter and 1 otherwise. The sample

includes 286 banks and 6,836 bank quarters. Dependent variables are described in Table 1.

***, ** and * denote statistical significance levels of 1%, 5%, and 10%, respectively, and z

statistics are reported in parentheses below parameter estimates.

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

NUM INSTIT OWNERS 0.0004 0.0005

(0.0004) (0.0004)

INSTIT OWN HHI −0.939∗∗∗ −0.946∗∗∗

(0.214) (0.215)

DEPOSITS TO ASSETS 0.679 0.774

(0.502) (0.491)

ST BORROWINGS TO ASSETS 1.274∗ 1.112

(0.731) (0.709)

LOG ASSETS −0.072 −0.055∗∗ −0.024 −0.100∗∗

(0.049) (0.022) (0.026) (0.047)

EQUITY TO ASSETS 3.169∗∗ 2.415∗ 3.520∗∗ 2.910∗

(1.524) (1.460) (1.548) (1.489)

CASH MKT SEC TO ASSETS 0.046 0.278 −0.020 −0.204

(0.796) (0.759) (0.815) (0.806)

QUARTER ROA 47.423∗∗∗ 48.421∗∗∗ 47.697∗∗∗ 47.327∗∗∗

(6.816) (6.757) (6.804) (6.781)

CAP TO RBC −0.368 −0.319 −0.048 0.161

(1.293) (1.235) (1.344) (1.293)

UNDER TARP −0.768∗∗∗ −0.770∗∗∗ −0.770∗∗∗ −0.768∗∗∗

(0.078) (0.077) (0.080) (0.079)

AVG TED −0.005∗∗∗ −0.005∗∗∗ −0.005∗∗∗ −0.005∗∗∗

(0.001) (0.001) (0.001) (0.001)

QCH ASSETS 0.791 0.933 0.798 0.942

(0.649) (0.648) (0.649) (0.650)

RET EARN TO ASSETS 1.128 1.294 0.973 1.176

(1.141) (1.090) (1.144) (1.097)

Constant 2.612∗∗∗ 2.661∗∗∗ 1.616∗∗∗ 2.275∗∗∗

(0.409) (0.258) (0.571) (0.600)

Observations 6,836 6,836 6,836 6,836

Akaike Inf. Crit. 1,716.489 1,700.517 1,716.115 1,701.816

Marginal R2 18.3% 19.2% 18.4% 19.7%

Conditional R2 20.1% 19.2% 20.2% 19.7%

19

influence on the propensity to cut or omit dividends, consistent with Hirtle’s

(2014) signalling hypothesis. These regressions suggest that banks fear the con-

sequences of dividend cuts on short-term borrowings, and of dividend omissions

on deposits. They hint at a gradation in signal intensity and investor percep-

tion. They also show that banks that maintained ample short-term borrowings

after cutting dividends appear less concerned about omitting dividends. Also,

the regressions show that favourable access to equity markets ranks high on the

agenda of bank managers and drives their reluctance to cut or omit dividends.

Finally, the regressions in Appendix A.4 show that the propensity to omit or

cut dividends decreases as the intensity of the crisis increases, again consistent

with the importance of maintaining market access during periods of acute cash

flow uncertainty.

Strikingly, the incorporation of shareholder dispersion (NUM INSTIT OWN-

ERS) in the dividend omission models (regressions (1) and (4) in Table 3) causes

bank size (LOG ASSETS) to become statistically insignificant. Banks are not

likelier to persist in paying dividends because they are larger but because of

agency costs and of their concern to maintain market access. This finding is

robust to alternative measurements of size and shareholder dispersion and to

the omission of TARP controls, as shown in Appendix A.3.

Consistent with Laeven and Levine’s (2009), the risk-taking induced by div-

idend policies appears to be intimately linked to bank ownership structures.

However, in contrast to the authors’ findings, we find that banks with more

powerful owners favour a low risk-taking stance corresponding to lower divi-

dends. We conclude in the robustness tests presented in Appendix A.1 that

bank managers lean towards dividend omissions only if institutional ownership

is not too dispersed, consistent with the substitute model and with Robinson’s

(1948) guidelines.

The controls in Table 3 provide additional insights on dividend omissions.

First, banks are less likely to omit dividends if their retained earnings (RET

EARN TO ASSETS) are higher, consistent with the lifecycle theory of divi-

dends but also with their preparedness to exhaust capital buffers before omit-

20

ting dividends. Second, banks have a higher propensity to omit dividends

when they hold more cash and short-term market securities (CASH MKT SEC

TO ASSETS). They thus omit dividends to preserve cash, either in crisis or

in precautionary mode, the former consistent with Keen’s (1978) “traditional

view” and the latter with the evidence provided by Acharya et al. (2012). By

contrast, neither retained earnings nor liquid assets explain the propensity to

cut dividends.

Tables 3 and 4 also show that dividend omissions and cuts share similar char-

acteristics. Banks are more likely to omit or cut dividends when profitability

(QUARTER ROA) drops, when their balance sheet contracts (QCH ASSETS),

when they receive capital support from authorities (UNDER TARP) or in times

of stress when liquidity is tight (AVG TED). By contrast, banks are not influ-

enced in their dividend policies by their regulatory capital ratios (CAP TO

RBC).

The evidence above provides substantive empirical support for the substitute

model of La Porta et al. (2000). In contrast to prior literature, this evidence

suggests that bank dividend policies are driven by agency costs and the regular

need to access capital markets rather than by too-big-to-fail considerations.

The evidence suggests that, knowingly or not, US bank managers have adopted

Robinson’s (1948) recommended dividend policy, and that they have been driven

to do so in order to maintain favourable market access.

5. Conclusion

The extraordinary persistence of bank dividends is a phenomenon identified

long ago. During the 2007-09 financial crisis, this persistence contributed to

dwindling bank solvency and exacerbated overall systemic risks. Nonetheless,

the persistence of bank dividends has not frequently been studied and remains

poorly understood, one of the reasons being the lack of opportunities to observe

changes in dividend policies before the financial crisis.

In this paper, we take advantage of the unusual volatility of bank dividend

21

policies during and after the financial crisis to test an agency cost explanation

of dividend persistence. We establish that managers persist in paying dividends

to mitigate agency costs, maintain a reputation for reasonableness towards dis-

persed shareholders and preserve their favourable access to financial markets.

We uphold the substitute model of La Porta et al. (2000). We find that agency

costs, rather than size, induce banks to persist in paying dividends during pe-

riods of greater cash flow uncertainty, consistent with what has been observed

during banking crises. We also find that, in contrast with Laeven and Levine

(2009), more powerful owners may induce bank managers to reduce risk-taking

by paying lower dividends.

On the one hand, we conclude that bank managers and large shareholders

attribute more, or more immediate, weight to the short-term loss of reputation

associated with dividend omissions, and the potential immediate repercussions

on their market access, rather than to longer-term solvency issues. On the

other hand, prior literature shows that bank dividends have signalling power.

The payment of dividends thus appears as a costly substitute for mechanisms

by which shareholders monitor managers (Jensen and Meckling, 1976), and of

which managers are acutely aware.

Since the financial crisis of 2007-09, regulators have secured the right to

restrict the dividend policies of undercapitalised or cash-stricken banks. By re-

stricting these policies, the regulators may harm the effectiveness of dividend

policies as substitutes for shareholder monitoring mechanisms. Indirectly, re-

stricting bank dividend policies may affect the capacity of certain banks to

attract external capital, or increase its cost. Just as importantly, restricting

bank dividend policies may trigger market reactions mitigated by persistent

dividends. When considering dividend restrictions, regulators should be pre-

pared to intervene in capital markets, possibly going as far as providing a liquid

backstop in case their decisions provoke the sudden drying up of bank funding.

22

Appendix A. Robustness Tests

Appendix A.1. Agency Costs

There are several reasons why the results relating to bank ownership pre-

sented in Section 4 may differ from those of (Laeven and Levine, 2009). First,

the ownership dispersion-concentration metric adopted in the regressions is not

the same. Second, consistent with the hypotheses and with what the dichotomy

between the dividend omission and dividend cut models suggests, large owners

may have incentives in presence of a dispersed shareholder base (maintaining

reputation and market access) than otherwise (preserving solvency). Third,

owners may have different risk preferences, other than solvency, relating to

business activities, or they may be opportunistic and favour risk-taking in good

times while merely taking a cautious stance when the worst is in sight.

To test the first explanation, we incorporate the equity stake percentage

held by the largest institutional shareholder (TOP OWNER PERC) into the

regressions presented in Table A.1.8 In order to test the second explanation, we

interact this equity stake with the number of institutional shareholders. The

third explanation is beyond the scope of the present paper. For brevity, we do

not report the control variables or the intercept.

Regression (1) in Table A.1 shows that the first explanation must be re-

jected. Prior results are not only robust to the incorporation of the holding of

the largest institutional shareholder, but strengthened by this additional con-

trol (both the statistical significance of institutional shareholding dispersion and

the explanatory power of the regressions increase). Regression (2) shows that

the propensity to omit dividends depends both on how dispersed institutional

shareholding is, but also on the equity stake held by the largest institutional

shareholder, consistent with the second explanation. More specifically, the like-

lihood of a dividend cut is lower in presence of dispersed shareholders and a large

8Conclusions are identical if the aggregate equity stake of the top 5 shareholders is retained

instead.

23

Table A.1: Random effects probit regressions to show the robustness of agency costs in ex-

plaining dividend omissions and cuts by US banks during the period between 30 June 2004

and 30 September 2012. The binary dependent variable takes the value 0 if a bank omitted

(regressions 1 and 2) or cut (regressions 3 and 4) a dividend in a quarter and 1 otherwise.

The sample includes 287 banks and 7,587 bank quarters for dividend omissions and 286 banks

and 6,835 bank quarters for dividend cuts. Dependent variables are described in Table 1.

***, ** and * denote statistical significance levels of 1%, 5%, and 10%, respectively, and z

statistics are reported in parentheses below parameter estimates. The control variables and

the intercept are not reported.

Dividend Omission Dividend Cut

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

NUM INSTIT OWNERS 0.003∗∗ 0.002 0.001 0.0004

(0.001) (0.002) (0.0004) (0.0005)

TOP OWNER PERC 1.054 −3.583 0.900 0.529

(1.315) (2.422) (0.667) (0.980)

INSTIT OWN HHI −0.237 −0.384 −0.971∗∗∗ −0.989∗∗∗

(0.484) (0.496) (0.213) (0.216)

NUM INSTIT OWNERS*

TOP OWNER PERC

0.058∗∗ 0.003

(0.027) (0.007)

DEPOSITS TO ASSETS 0.346 0.400 0.594 0.599

(1.042) (1.047) (0.502) (0.502)

ST BORROWINGS

TO ASSETS

2.249 2.367 0.959 1.005

(1.560) (1.579) (0.716) (0.723)

LOG ASSETS −0.085 −0.252∗ −0.114∗∗ −0.122∗∗

(0.134) (0.143) (0.049) (0.051)

Observations 7,587 7,587 6,835 6,835

Akaike Inf. Crit. 813.324 810.175 1,691.447 1,693.175

Marginal R272.6% 74.3% 20.2% 20.3%

Conditional R281.5% 82.6% 20.2% 20.3%

24

institutional owner, whereas it is higher when a large institutional owner is rel-

atively “isolated”.9 Also, the association between the propensity of dividend

omissions and short-term funding disappears when controlling for the level of

ownership of the largest institutional shareholder. The presence of large owners

may thus be associated with preferences in terms of a bank’s funding mix.

Regressions (3) and (4) show that those effects are limited to dividend cuts,

but also that the conclusions reached previously are robust to incorporating the

holding of the largest institutional shareholder.

Appendix A.2. Market access

As shown in the previous subsection, the simple, direct evidence that bank

dividend decisions are dictated by the imperative to maintain market access

does not survive the incorporation of the equity stake held by large institu-

tional shareholders. The reasons for this do not seem trivial but may relate to

preferences relating to the short-term funding mix associated with larger institu-

tional ownership, which are beyond the scope of this paper. Alternatively, they

may relate to a complex, non-linear influence of funding on dividend policies.

To test the latter explanation, we incorporate an interaction term between

deposits and short-term funding into the regressions. Also, we add to the re-

gressions a measurement of the intensity of analyst coverage, measured as the

number of analysts reported by I/B/E/S as having published EPS estimates

in the last quarter. We use the intensity of analyst coverage as a proxy for

favourable access to equity markets. Derrien and Kecskes (2013) find that firms

which lost the coverage of one sell-side analyst experienced an increase in the

cost of capital and a decrease in overall financing. Similarly, Lang et al. (2012)

find that firms subject to greater analyst coverage benefit from lower trans-

action costs and greater liquidity, a factor that appears especially important

during more volatile periods.

9In Regression (3) of Table A.1, TOP OWNER PERC is borderline statistically significant,

with a p-value of 0.139.

25

Table A.2: Random effects probit regressions to show the robustness of maintaining market

access in explaining dividend omissions and cuts by US banks during the period between 30

June 2004 and 30 September 2012. The binary dependent variable takes the value 0 if a

bank omitted (regressions 1 and 2) or cut (regressions 3 and 4) a dividend in a quarter and 1

otherwise. The sample includes 287 banks and 7,589 bank quarters for dividend omissions and

286 banks and 6,836 bank quarters for dividend cuts. Dependent variables are described in

Table 1. ***, ** and * denote statistical significance levels of 1%, 5%, and 10%, respectively,

and z statistics are reported in parentheses below parameter estimates. The control variables

and the intercept are not reported.

Dependent variable:

Dividend Omission Dividend Cut

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

DEPOSITS TO ASSETS*

ST BORROWINGS TO ASSETS

23.433∗∗∗ −0.263

(7.063) (4.301)

NUM ANALYSTS 0.081∗∗∗ 0.025∗∗∗

(0.020) (0.008)

NUM INSTIT OWNERS 0.004∗∗∗ 0.003∗∗∗ 0.0005 0.001

(0.001) (0.001) (0.0004) (0.0004)

INSTIT OWN HHI −0.222 −0.048 −0.946∗∗∗ −0.860∗∗∗

(0.487) (0.466) (0.215) (0.219)

DEPOSITS TO ASSETS −1.054 0.027 0.790 0.610

(1.132) (1.002) (0.560) (0.494)

ST BORROWINGS

TO ASSETS

−12.425∗∗∗ 1.972 1.275 1.414∗

(4.350) (1.530) (2.751) (0.726)

LOG ASSETS −0.143 −0.298∗∗ −0.099∗∗ −0.200∗∗∗

(0.130) (0.133) (0.048) (0.057)

Observations 7,589 7,589 6,836 6,836

Akaike Inf. Crit. 804.008 792.378 1,703.813 1,695.292

Marginal R273.9% 75.6% 19.7% 20.5%

Conditional R281.9% 82.2% 19.7% 20.5%

26

Regression (1) in Table A.2 shows a differentiated, composite effect of de-

posits and short-term borrowings on the propensity to omit dividends. Banks

that are dependent both on deposits and short-term borrowings are less likely to

omit dividends, whereas those that are dependent on short-term borrowings only

are more likely to do so. Banks appear to be especially wary of runs of deposits

when omitting dividends. Regression (3) shows no such effect for dividend cuts.

Interestingly, regression (4) (like regression (3) in Table 4), suggest that banks

are less likely to cut when they are more reliant on short-term borrowings.

Regressions (2) and (4) highlight that greater coverage is associated with

a lower likelihood of cuts and omissions, consistent with the hypothesis that

market access is at stake and suggest that bank managers consciously seek to

preserve their banks’ reputation, with the attendant lower cost of capital and

liquidity risk.10

Appendix A.3. Bank size and TARP side-effects

In all regressions shown above, agency costs dominate bank size effects.

However, Fama and French (2001) argue that growth in firm size over time may

bias regression results, so they use size percentiles instead. Similarly, regres-

sions (1) and (3) in Table A.3 incorporate both size and institutional ownership

percentiles to demonstrate that the conclusions on bank size are robust. Unre-

ported tests demonstrate that replacing log assets with the less correlated total

assets leads to the same conclusion, and that an interaction term between the

number of institutional shareholders and log assets takes a negative sign. This

also supports the hypothesis.

Also, Black and Hazelwood (2013) find that, after receiving TARP funds,

larger banks increased their risk-taking, contrary to smaller ones. In order to

confirm that prior results and that larger bank size effects are not connected

10Unreported regressions show that the effect goes beyond having analyst coverage (binary

variable) and is linear rather than logarithmic, consistent with Derrien and Kecskes’s (2013)

results.

27

to TARP-related risk taking, regressions (2) and (4) omit the TARP control to

provide further evidence of the absence of a too-big-to-fail syndrome in bank

dividend decisions.11

Appendix A.4. Financial Crisis

Finally, we control in two different ways for the potential changes in be-

haviour induced by the financial crisis. First, we determine whether the in-

tensity of the crisis accentuates dividend persistence by interacting shareholder

dispersion and concentration with the TED spread. Second, to assess whether

or not our conclusions are influenced by banks that failed during the crisis, we

excluded all banks that disappeared before 31 December 2009 from the sample.

We show regression estimates in Table A.4.

Regressions (1) and (3) in Table A.4 show that the propensity to omit and

cut dividends decreases as the intensity of the crisis increases. This is consistent

with the hypothesis that bank dividend policy is driven by market access con-

siderations, given that maintaining funding during these periods is particularly

critical. Regressions (2) and (4) show that our previous conclusions are not

affected by the behaviour of banks that failed during the financial crisis.

11TARP capital injections almost systematically consisted of a combination of capital sup-

port in different forms (preferred stock, subordinated debentures, trust-preferred securities or

common equity) together with the issuance of warrants. This combination makes it difficult

to single out the effect of warrants and of different forms of capital. Without being fully

conclusive, unreported regressions suggest that the issuance of common stock, followed by

preferred stock, was associated with a greater likelihood of dividend cuts and omissions.

28

Table A.3: Random effects probit regressions to show the robustness of bank size in explaining

dividend omissions and cuts by US banks during the period between 30 June 2004 and 30

September 2012. The binary dependent variable takes the value 0 if a bank omitted (regres-

sions 1 and 2) or cut (regressions 3 and 4) a dividend in a quarter and 1 otherwise. The sample

includes 287 banks and 7,589 bank quarters for dividend omissions and 286 banks and 6,836

bank quarters for dividend cuts. Dependent variables are described in Table 1. ***, ** and

* denote statistical significance levels of 1%, 5%, and 10%, respectively, and z statistics are

reported in parentheses below parameter estimates. The control variables and the intercept

are not reported.

Dependent variable:

Dividend Omission Dividend Cut

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

NIO PERCENTILE 1.740∗∗ −0.273

(0.690) (0.328)

NUM INSTIT OWNERS 0.003∗∗∗ 0.001

(0.001) (0.0004)

TA PERCENTILE −0.243 0.034

(0.571) (0.293)

LOG ASSETS −0.155 −0.140∗∗∗

(0.120) (0.045)

INSTIT OWN HHI 0.397 −0.111 −0.997∗∗∗ −0.794∗∗∗

(0.529) (0.461) (0.233) (0.210)

DEPOSITS TO ASSETS 0.143 0.099 0.835∗ 0.209

(1.003) (0.966) (0.465) (0.459)

ST BORROWINGS TO ASSETS 1.914 2.503∗ 1.132 1.533∗∗

(1.534) (1.450) (0.712) (0.677)

EQUITY TO ASSETS 19.666∗∗∗ 24.323∗∗∗ 3.051∗∗ 5.618∗∗∗

(3.818) (3.643) (1.505) (1.475)

UNDER TARP −0.690∗∗∗ −0.777∗∗∗

(0.147) (0.079)

Observations 7,589 7,589 6,836 6,836

Akaike Inf. Crit. 808.305 830.154 1,704.572 1,794.352

Marginal R2 72.7% 73.0% 19.4% 15.7%

Conditional R2 80.7% 80.6% 19.4% 15.7%

29

Table A.4: Random effects probit regressions to show the robustness of agency costs and

market access in explaining dividend omissions and cuts by US banks during the period be-

tween 30 June 2004 and 30 September 2012, despite the financial crisis. The binary dependent

variable takes the value 0 if a bank omitted (regressions 1 and 2) or cut (regressions 3 and

4) a dividend in a quarter and 1 otherwise. Regressions (1) and (3) are estimated using sam-

ples including respectively 287 banks and 7,589 bank quarters and 286 banks and 6,836 bank

quarters. Regressions (2) and (4) are estimated on the same samples from which banks that

had not survived by 31 December 2009 are excluded, including 217 banks and respectively

6,651 and 5,965 bank-quarters. Dependent variables are described in Table 1. ***, ** and

* denote statistical significance levels of 1%, 5%, and 10%, respectively, and z statistics are

reported in parentheses below parameter estimates. The control variables and the intercept

are not reported.

Dependent variable:

Dividend Omission Dividend Cut

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

NUM INSTIT OWNERS 0.003∗∗ 0.003∗∗ 0.001

(0.001) (0.001) (0.0004)

INSTIT OWN HHI −0.495 −1.696∗∗∗ −0.577∗∗

(0.570) (0.321) (0.282)

NUM INSTIT OWNERS:AVG TED 0.00002

(0.00001)

INSTIT OWN HHI:AVG TED 0.013∗∗∗

(0.005)

DEPOSITS TO ASSETS 1.049 0.782

(1.289) (0.547)

ST BORROWINGS TO ASSETS 3.289 0.715

(2.002) (0.776)

AVG TED −0.006∗∗∗ −0.002∗∗ −0.007∗∗∗ −0.005∗∗∗

(0.001) (0.001) (0.001) (0.001)

LOG ASSETS −0.107 −0.104 −0.049∗∗ −0.089∗

(0.134) (0.156) (0.022) (0.052)

Observations 7,589 6,651 6,836 5,965

Akaike Inf. Crit. 806.881 666.391 1,692.893 1,536.420

Marginal R2 72.7% 72.4% 19.7% 20.4%

Conditional R2 82.0% 82.5% 19.7% 20.4%

30

References

Abreu, J. F., Gulamhussen, M. A., 2013. Dividend payouts: Evidence from

US bank holding companies in the context of the financial crisis. Journal of

Corporate Finance 22, 54–65.

Acharya, V. V., Davydenko, S. A., Strebulaev, I. A., 2012. Cash holdings and

credit risk. Review of Financial Studies 25 (12), 3572–3609.

Acharya, V. V., Gujral, I., Kulkarni, N., Shin, H. S., 2011. Dividends and bank

capital in the financial crisis of 2007-2009. Working Paper 16896, National

Bureau of Economic Research.

Acharya, V. V., Le, H., Shin, H. S., 2013. Bank capital and dividend externali-

ties. Working Paper 19707, National Bureau of Economic Research.

Aharony, J., Swary, I., 1980. Quarterly dividend and earnings announcements

and stockholders’ returns: An empirical analysis. Journal of Finance 35 (1),

1–12.

Benito, A., Young, G., 2003. Hard times or great expectations? Dividend omis-

sions and dividend cuts by UK firms. Oxford Bulletin of Economics and Statis-

tics 65 (5), 531–555.

Bessler, W., Nohel, T., 1996. The stock-market reaction to dividend cuts and

omissions by commercial banks. Journal of Banking & Finance 20 (9), 1485–

1508.

Bessler, W., Nohel, T., 2000. Asymmetric information, dividend reductions, and

contagion effects in bank stock returns. Journal of Banking & Finance 24 (11),

1831–1848.

Black, L. K., Hazelwood, L. N., 2013. The effect of TARP on bank risk-taking.

Journal of Financial Stability 9 (4), 790–803.

Bøhren, Ø., Josefsen, M. G., Steen, P. E., 2012. Stakeholder conflicts and divi-

dend policy. Journal of Banking & Finance 36 (10), 2852–2864.

31

Brewer, III, E., Jagtiani, J., 2013. How much did banks pay to become too-big-

to-fail and to become systemically important? Journal of Financial Services

Research 43 (1), 1–35.

Brockman, P., Unlu, E., 2011. Earned/contributed capital, dividend policy, and

disclosure quality: an international study. Journal of Banking & Finance

35 (7), 1610–1625.

Casey, K. M., Dickens, R. N., 2000. The effects of tax and regulatory changes

on commercial bank dividend policy. Quarterly Review of Economics and

Finance 40 (2), 279–293.

DeAngelo, H., DeAngelo, L., Stulz, R. M., 2006. Dividend policy and the

earned/contributed capital mix: a test of the life-cycle theory. Journal of

Financial Economics 81 (2), 227–254.

Derrien, F., Kecskes, A., 2013. The real effects of financial shocks: Evidence

from exogenous changes in analyst coverage. Journal of Finance 68 (4), 1407–

1440.

Dickens, R. N., Casey, K. M., Newman, J. A., 2002. Bank dividend policy:

Explanatory factors. Quarterly Journal of Business & Economics 41 (1/2),

3–12.

Faccio, M., Lang, L. H., Young, L., 2001. Dividends and expropriation. Ameri-

can Economic Review 91 (1), 54–78.

Fama, E. F., French, K. R., 2001. Disappearing dividends: changing firm char-

acteristics or lower propensity to pay? Journal of Financial Economics 60 (1),

3–43.

Gupta, M. C., Walker, D. A., 1975. Dividend disbursal practices in commercial

banking. Journal of Financial and Quantitative Analysis 10 (03), 515–529.

Hirtle, B., 2014. Bank holding company dividends and repurchases during the

financial crisis. Staff Report 666, Federal Reserve Bank of New York.

32

Jensen, M. C., Meckling, W. H., 1976. Theory of the firm: Managerial behavior,

agency costs and ownership structure. Journal of Financial Economics 3 (4),

305–360.

Kanas, A., 2013. Bank dividends, risk, and regulatory regimes. Journal of Bank-

ing & Finance 37 (1), 1–10.

Keen, Jr., H., 1978. Bank dividend cuts: recent experience and the traditional

view. Business Review, Federal Reserve Bank of Philadelphia (Nov./Dec.),

5–13.

Keen, Jr., H., 1983. The impact of a dividend cut announcement on bank share

prices. Journal of Bank Research (Winter), 274–281.

La Porta, R., Lopez de Silanes, F., Vishny, R., Shleifer, A., 2000. Agency prob-

lems and dividend policies around the world. Journal of Finance 60 (1), 1–33.

Laeven, L., Levine, R., 2009. Bank governance, regulation and risk taking. Jour-

nal of Financial Economics 93 (2), 259–275.

Laeven, L., Ratnovski, L., Tong, H., 2014. Bank size and systemic risk. Staff

discussion note, IMF.

Lang, M., Lins, K. V., Maffett, M., 2012. Transparency, liquidity, and valua-

tion: International evidence on when transparency matters most. Journal of

Accounting Research 50 (3), 729–774.

Miller, M. H., Modigliani, F., 1961. Dividend policy, growth, and the valuation

of shares. Journal of Business 34 (4), 411–433.

Nakagawa, S., Schielzeth, H., 2013. A general and simple method for obtaining

R2 from generalized linear mixed-effects models. Methods in Ecology and

Evolution 4 (2), 133–142.

Robinson, R. I., 1948. Bank capital and dividend policies. Harvard Business

Review 26 (4), 398–409.

33

Rosengren, E. S., 2010. Dividend policy and capital retention: A systemic ‘first

response’. In: Rethinking Central Banking Conference, Washington, DC, Oc-

tober 10.

Rozeff, M., 1982. Growth, beta and agency costs as determinants of dividend

payout ratios. Journal of Financial Research 5 (3), 249–259.

Slovin, M. B., Sushka, M. E., Polonchek, J. A., 1999. An analysis of contagion

and competitive effects at commercial banks. Journal of Financial Economics

54 (2), 197–225.

Stern, G. H., Feldman, R. J., 2004. Too big to fail: The hazards of bank bailouts.

Brookings Institution Press.

34