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: [email protected]Tel. : +32 (0)2/650.48.64 Fax: +32 (0)2/650.41.88
35
Embed
Dividend Persistence and Equity Agency Costs in Banking ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
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.
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
∗This version: June 19, 2014. First version: May 3, 2014.Email address: [email protected] (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
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∗∗∗