J. Finan.

Intermediation 28 (2016) 4–21

Contents lists available at ScienceDirect

J. Finan. Intermediation
journal homepage: www.elsevier.com/locate/jfi

The dark side of liquidity creation: Leverage and systemic risk

Viral V. Acharya a,∗, Anjan V. Thakor b
a
C.V. Starr Professor of Economics, Stern School of Business, New York University, NBER Research Associate and CEPR Research Associate, 44 West Fourth
Street, p-84, New York, NY, 10012 USA
b
John E. Simon Professor of Finance, Olin Business School, Washington University in St. Louis, and European Corporate Governance Institute Research
Associate, One Brookings Drive, St. Louis, MO 63130, USA

a r t i c l e i n f o a b s t r a c t
JEL: We consider a model in which the threat of bank liquidations by creditors as well as equity-based com-
G21 pensation incentives both discipline bankers, but with different consequences. Greater use of equity leads
G28
to lower ex-ante bank liquidity, whereas greater use of debt leads to a higher probability of inefficient
G32
bank liquidation. The bank’s privately-optimal capital structure trades off these two costs. With uncer-
G35
G38 tainty about aggregate risk, bank creditors learn from other banks’ liquidation decisions. Such inference
can lead to contagious liquidations, some of which are inefficient; this is a negative externality that is
Keywords: ignored in privately-optimal bank capital structures. Thus, under plausible conditions, banks choose ex-
Micro-prudential regulation cessive leverage relative to the socially optimal level, providing a rationale for bank capital regulation.
Macro-prudential regulation While a blanket regulatory forbearance policy can eliminate contagion, it also eliminates all market dis-
Market discipline
cipline. However, a regulator generating its own information about aggregate risk, rather than relying on
Contagion
market signals, can restore efficiency and market discipline by intervening selectively.
Lender of last resort
Bailout © 2016 Published by Elsevier Inc.
Capital requirements

“Any observed statistical regularity will tend to collapse once on containing the risk of events like systemic capital and liquid-
pressure is placed upon it for control purposes”. ity shortages, manifesting as fire sales and the freezing up of asset
Goodhart (1975) markets. Macro-prudential regulation also examines ways in which
regulatory interventions like bank bailouts can prevent (or engen-
1. Introduction der) such occurrences and contain (or aggravate) their adverse im-
pact. But since both forms of regulation ultimately seek to enhance
In ensuring that the risk of the financial system as a whole financial system stability, a natural question that arises is: what are
stays at “prudent” levels, regulators are tasked to meet two forms the micro foundations that possibly link these two forms of regu-
of regulatory challenges. One is micro-prudential regulation, which lation? In this paper, we show that not only micro-prudential and
needs to ensure that risk-taking at the individual bank level is not macro-prudential regulation affect each other, but that in fact there
excessive. The other is macro-prudential regulation, which seeks to is a fundamental tension between the two.
contain the systemic risk that banks may be excessively exposed to Let us explain. Previous papers have noted that uninsured bank
collective failure. To date, these two forms of regulation have been debt can increase market discipline and thereby enhance bank loan
typically dealt with in isolation of each other, especially in policy quality and/or liquidity creation.2 This notion is also codified in
debates. Micro-prudential regulation aims to contain the distorted
incentives of banks to make choices that maximize the value of
bank shareholders’ risk-shifting (or asset-substitution) options, es- deposit insurance, a bank has an economic incentive to invest in riskier assets and
choose relatively low amounts of capital in its capital structure. This means regu-
pecially in the presence of regulatory put options like deposit in-
latory monitoring of individual banks is necessary to control excessive risk taking
surance.1 Macro-prudential regulation, on the other hand, focuses designed to exploit deposit insurance.
2
Calomiris and Kahn (1991) were the first to formally argue that monitoring by
uninsured depositors can result in a bank manager who is making imprudent as-
∗
Corresponding author. set choices being exposed to the threat of a bank run, and that this can induce
E-mail addresses: [email protected] (V.V. Acharya), [email protected] the manager to shy away from such asset choices. Diamond and Rajan (2001) note
(A.V. Thakor). that banks invest in assets that are inherently illiquid due to the inability of bank
1
There is a long history of academic research on micro-prudential regulation. managers to credibly pre-commit to certain actions, and that the threat of a run by
Merton (1977) aptly recognized the isomorphic correspondence between deposit in- uninsured creditors can make these pre-commitments credible, thereby improving
surance and common stock put options. An important implication was that, given liquidity creation by banks. Acharya and Viswanathan (2011) develop this point in

http://dx.doi.org/10.1016/j.jfi.2016.08.004
1042-9573/© 2016 Published by Elsevier Inc.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 5

bank regulation with market discipline being one of the three pil- count rate exceeds that of the firm, equity financing reduces the
lars of Basel II (the other two being regulatory monitoring and cap- ex-ante liquidity of the bank relative to debt financing which can
ital requirements). This argument about the market discipline of impose discipline without managerial cash payments. Offsetting
debt is concerned primarily with the attenuation of bank-specific this ex-ante advantage of bank leverage is that it leads to liquida-
risks, and thus it can be viewed as a tool of micro-prudential reg- tion of the bank in some states, and this liquidation can be ex-post
ulation. inefficient. The bank’s privately optimal capital structure is deter-
However, high bank leverage has also been held culpable as a mined by the tradeoff between the ex-ante efficiency of leverage
contributor to the recent financial crisis. Many have argued that relative to equity in the provision of incentives to bankers and the
very high financial leverage, especially short-term leverage, in- expected ex-post cost of inefficient liquidations induced by lever-
duced banks to engage in illiquid and risky lending as well as secu- age.
rities activities that resulted in the widespread failures of these in- Bank asset portfolios are then allowed to suffer systematic
stitutions (see e.g., Acharya, Schnabl and Suarez (2013), Adrian and shocks to value that are observed by some of each bank’s creditors
Shin (2010), Goel, et al. (2014), Mian and Sufi (2011), and Shleifer but not commonly observed by creditors across banks. This means
and Vishny (2010)). There appears to be an emerging acceptance that the (interim) liquidation decision made by the creditors of a
of the fact that increases in leverage seem to increase the sys- bank can be due to either bank-specific information or informa-
temic risk, or the collective fragility, of financial institutions. Finan- tion about the systematic shock. Since not all creditors of a bank
cial crises are typically associated with a few highly-levered banks, receive information about the systematic shock, but they can ob-
initially suffering portfolio shocks that engender capital or liquid- serve the liquidation decisions of other creditors, they learn from
ity shortages for those banks, with the malaise quickly ensnaring each other’s decisions and update their beliefs about the system-
other banks as the crisis deepens. atic shock.3 Their learning is noisy, however, because of the com-
As a result, bank-specific and systemic risks, and in turn, micro- mingling of information about idiosyncratic and systematic risks in
prudential and macro-prudential regulation, become difficult to any bank’s observed liquidation. This can give rise to contagion ef-
separate. In particular, there emerges a somewhat schizophrenic fects as those creditors of a bank that possess no adverse idiosyn-
view of the role of leverage. On the one hand, higher leverage cratic or systematic risk information about the bank, may choose
may mean better asset-choices by bank managers and more liq- nonetheless to liquidate their bank at the interim date based solely
uidity when banks are viewed individually. On the other hand, on observing the liquidations of other banks.
higher leverage also means that the system is more fragile. Faced We assume that deadweight costs of individual failures are
with circumstances of possible systemic failure, regulatory inter- lower than those from joint failures –such as those observed when
ventions can play a role in the reduction of ex-post fragility. How- the whole system or a large portion of it collapses — due to lim-
ever, it is also precisely in these circumstances that the disciplin- ited re-intermediation of bank activities and failure of payments
ing effect of the bank’s capital structure on ex-ante asset choices is and settlement systems in such cases. Contagion can then lead
compromised and the lines between micro-prudential and macro- to ex-post inefficient liquidations in some instances because the
prudential regulation begin to become blurred. creditors of a bank may liquidate their bank based on the mis-
The underlying linkage between leverage, ex-ante liquidity cre- taken inference that the observed liquidations of other banks are
ation, and ex-post systemic risk raise some fundamental questions due to a common asset-value shock even when they are due to
that we address in this paper. bank-specific shocks.4 Thus, one dark side of leverage-based liquid-
First, what is the role of bank leverage vis a vis equity capital ity creation is the attendant systemic risk arising from inefficient
in affecting the bank’s ex-ante liquidity and portfolio risk? Second, contagious liquidations, and the higher the leverage of banks, the
how does maximizing individual bank liquidity (a micro-prudential greater the systemic risk.
regulation concern) affect systemic risk (a macro-prudential regula- We solve for the bank’s privately-optimal capital structure in
tion concern)? Third, is there a rationale for regulatory interven- the presence of the systematic asset-value shock, and the regula-
tion, and if yes, under what circumstances? Fourth, how does the tor’s optimal level of leverage, assuming that the regulator’s objec-
regulator affect bank leverage, and what are the implications of tive is to maximize the value of the entire banking industry.5 A
this for micro-prudential regulation? That is, when does the regu- divergence between the regulatory and private optima arises be-
lator interfere with the market discipline role of leverage and what cause, in choosing its own capital structure, an individual bank
are its (unintended) consequences? internalizes neither the valuable information about the systematic
To address these questions, we develop a model of an unin- shock conveyed to other banks by its own leverage and creditor-led
sured bank whose manager has asset-choice flexibility. The bank is liquidation (a positive externality) nor the higher likelihood that its
a priori illiquid because the manager cannot credibly pre-commit liquidation may trigger the inefficient liquidation of another bank
to the right asset choices given his personal preference for a (a negative externality). We establish conditions under which the
private-benefit project. The bank’s ex-ante liquidity is measured privately-optimal bank leverage will be too high relative to the reg-
by the financing it can raise by issuing claims against its termi-
nal cash flows. This financing can be any mix of debt and eq-
uity. We permit both debt and equity to discipline the bank man- 3
For instance, sale and repurchase agreements (repos) are rolled over each morn-
ager to create ex-ante liquidity, but this discipline is different de- ing for dealer banks by financiers such as money market funds. Though a money
pending upon whether it is imposed by debt or equity. Debt dis- market fund rolling over a mortgage-backed securities (MBS) repo may not have
ciplines the bank manager by the credible threat that there will precise information about the overall quality revision in the housing market for to-
be liquidation in some interim states, conditional on interim cash- day, they may see (or hear through the grapevine about) other money market funds
having not rolled over their repos for some dealer, say Bear Stearns or Lehman
flow realizations. Equity disciplines the bank manager by provid- Brothers, and, in turn, consider this information while rolling over repos for other
ing compensation-based incentives to the manager to select the dealers.
efficient project. However, since the incentives provided by equity 4
Note that with the systematic asset-value shock, liquidations are not always ex-
involve payments from ex-post cash flows and the managerial dis- post inefficient since they are sometimes in response to creditors observing a neg-
ative shock to asset value of a bank that falls below liquidation value due to the
shock, and this negative shock contains relevant information for the asset values of
a model where financial intermediaries can switch to riskier assets after borrowing, other banks too.
5
and short-term debt with strong control rights ensures ex-ante liquidity by contain- In our model, this objective is equivalent to maximizing the banking industry’s
ing this agency problem. aggregate liquidity.
6 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

ulatory optimum.6 We argue that in order to cope with this, the ers. The policy is consistent with one of Bagehot’s (1873) principles
regulator, may wish to impose ex-ante (countercyclical) capital re- for a lender of last resort that should, according to Bagehot (1873),
quirements on banks that are binding during high-asset-valuation intervene to prevent failures when they threaten the payment sys-
periods. tem, but allow individual banks to fail when the payment system
Faced with the prospect of contagion arising from a bank’s liq- is not threatened.
uidation, the government regulator can step in with a liquidity in- Finally, we show each of these regulatory closure or bailout
fusion that effectively bails out the bank, prevents liquidations, and policies affects the design of ex-ante capital requirement. Since
forestalls a contagion in the form of a system-wide liquidation of the unconditional policy of bailing out all banks that could fail
banks. We consider the unintended consequences of such interven- makes leverage have no economic function other than to induce
tion under two information regimes. bailouts, the optimal capital requirement in this case is that banks
First, we assume that the regulator can observe neither the id- be funded entirely with equity. This is, however, costly in terms
iosyncratic interim financial condition of the bank nor the system- of ex-ante bank liquidity. At the other extreme, the policy of no
atic asset-value-impairment shock, and adopts an unconditional bailouts in any state of the world leads to failure externalities
bailout policy. We show in this case that the presence of the reg- not being internalized by banks so that there is a room for lim-
ulator can destroy all ex-ante market discipline of debt, as cred- iting bank leverage through an interior capital requirement in this
itors, who anticipate ex-post bailouts, and have no incentives to case. In between these two cases lies the more efficient conditional
engage in privately-costly monitoring. This means that there is no bailout policy in which the regulator uses its own information to
asset-value information generated by creditor liquidations, and the selectively bail out banks when there is a common asset impair-
absence of the threat of creditor liquidations reduces (debt) mar- ment shock. In this case, banks are bailed out only selectively so
ket discipline on banks, making micro-prudential regulation more that social costs of bank failures are limited, and in turn, the cap-
challenging. Somewhat paradoxically, capital structures of banks ital requirement can allow greater bank leverage (and greater ex-
become irrelevant, as both bank debt and equity need to rely on ante bank liquidity) compared to the case of no bailouts. Our anal-
compensation incentives for the manager. Consequently, the sole ysis makes it clear that the design of ex-ante capital requirements
reliance on compensation incentives causes the ex-ante liquidity is intimately tied to the regulatory choice of bailout policy. In par-
of banks to decline relative to that available in the absence of ticular, if the bailout policy can be based on information about sys-
the regulator, as the pursuit of systemic safety through uncondi- temic risk that is generated by the regulator on top of the informa-
tional bailouts (a macro-prudential regulation goal) compromises tion generated by bank debt (through the threat of runs), then the
the micro-prudential regulation goal of better discipline at the in- bailout policy can be partly macro-prudential in nature. This is ad-
dividual bank level. vantageous as this frees up the capital requirements from being ex-
In one sense, our analysis highlights Goodhart’s (1975) Law in cessively conservative and hence bank liquidity is not inefficiently
that the market discipline of debt collapses once regulators rely compromised.
upon the manifestation of this discipline in the form of bank liq- While we refer to the “regulator” throughout as a government
uidations to undertake regulatory interventions. This would – at or quasi-government agency that has the full faith and backing of
least to some extent – not be the case if regulatory interventions the government when it comes to committing resources to bail out
were based on regulatory intelligence about bank solvency over banks, we could also think of the regulator here as a lender of last
and above market signals, rather than just based on market signals resort (LOLR), if we take a broader view of the LOLR than as just
(such as creditor-led liquidations). This suggests that the regulator a collateralized lender. See Calomiris et al. (2016) for such a view
may be more effective if accompanied by information generation from a historical perspective. The idea is that if the banking system
by central banks or bank regulators that is independent of the in- suffers a large enough shock, collateralized lending by the LOLR
formation generated by bank creditors.7 Our analysis can also be may not be sufficient to deal with the shock, so bailouts may be-
viewed as an illustration of the Lucas Critique (1976) in that the come a necessary part of the LOLR’s policy toolkit.
macro-prudential effects of the regulator cannot be predicted with- The rest of the paper is organized as follows. Section 2 pro-
out accounting for how individual institutions and investors will vides a brief summary of related literature. Section 3 develops
change their behavior (monitoring by creditors and recapitalization the model. Section 4 contains the analysis of the basic model and
by bank owners, in our model) in response to the change in policy shows how leverage helps create both liquidity and systemic risk.
resulting from the introduction of the regulator. Section 5 examines the role of the regulator in controlling or ag-
We then examine a second information regime in which the gravating systemic risk with its choice of ex-ante capital require-
regulator continues to be unable to observe any bank’s interim (id- ments and ex-post bailout policy. Section 6 concludes. All proofs
iosyncratic) financial condition (which its creditors can observe), are in Appendix A.
but can observe whether the systematic asset-value impairment
shock has occurred. In this case, a selective intervention policy, 2. Related literature
wherein the regulator bails out banks only if asset values are sys-
tematically impaired, eliminates liquidation contagion, but toler- Our paper is related to many strands of the literature. On the
ates liquidations triggered by idiosyncratic, bank-specific informa- topic of micro-prudential regulation of banks, the role of leverage
tion possessed by creditors. This serves the twin goals of pre- in imposing market discipline on banks has been recognized in nu-
serving market discipline and avoiding liquidation contagion, and merous papers, as mentioned earlier. See, for example, Calomiris
hence stems from two aspects of our model: first, that joint fail- and Kahn (1991), Dewatripont and Tirole (1994),8 Diamond and
ures are more socially costly than individual bank failures, and sec- Rajan (2001), and Acharya and Viswanathan (2011). Like these pa-
ond, that bank creditors provide valuable discipline on bank own- pers, we also show how uninsured bank leverage can play a dis-
ciplining role. However, in contrast to these papers, we also allow

6
We also discuss later the conditions under which the bank’s privately-optimal
leverage is too low compared to the regulatory optimum, although that case is not 8
Dewatripont and Tirole (1994) also discuss how shareholders can affect man-
the focus of our analysis. agerial incentives in banks. Issues of managerial incentives in banks do not arise
7
An important exception is if the regulators require bank creditors to be segre- in the original theories of why banks exist (e.g. Ramakrishnan and Thakor (1984))
gated by priority and rely on information signals generated by non-contagious or because the bank is essentially a cooperative jointly owned by the employees who
run-prone liabilities of banks, as analyzed by Hart and Zingales (2009). can monitor each other.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 7

bank equity to discipline the manager, albeit through a different a given bank’s leverage has a negative information externality for
channel, namely incentive compensation that is costly for share- other banks in the sense that it elevates their liquidation probabil-
holders to provide as managerial discount rates are greater than ities holding fixed their leverage levels.
those of shareholders. This allows us to examine the tradeoff be- Yet another strand of related research has highlighted that
tween disciplining the bank through leverage and disciplining it the presence of the LOLR can make more likely the very state
through equity. Another key difference is that the key force that of systemic risk (correlated capital and liquidity shortages) that
makes bank leverage socially costly in our model is an information the LOLR is trying to avoid ex post (see Acharya and Yorulmazer
externality. (20 07), Acharya (20 09), Acharya, Mehran and Thakor (2016), Buser
In this respect, the manner in which equity discipline works et al. (1981), Kane (2010), and Farhi and Tirole (2012)). In con-
in our model is different from the way it works in various other trast to these papers, our key point, however, is that the safety
papers where high equity capital either induces strong monitor- nets designed to address leverage-induced systemic risk can de-
ing incentives (e.g., Mehran and Thakor (2011)) or deters asset- stroy the efficiency-based economic rationale, namely market dis-
substitution moral hazard (see, for example, Bhattacharya et al. cipline through creditor interventions, for banks to have leverage
(1998) for a review of this literature). Acharya, Mehran and Thakor in the first place. Diamond and Rajan (2012) also highlight that
(2016) have recently pointed out that this role of equity in deter- ex-post forbearance reduces ex-ante liquidity creation by banks
ring asset-substitution incentives also produces a tension between by eliminating the threat of runs, and recommend that the cen-
having leverage and equity in a bank. tral bank adopt a policy that raises interest rates in good times
There are also various papers on systemic risk in banking that to counter the effect of forbearance. What our analysis contributes
are related to our work. One strand of research focuses on the additionally on this issue are two important observations. First,
effects of risk-sharing on systemic risk. Winton (1997) illustrated the elimination of systemic crises through ex-post bailouts neces-
the point that “pooling (diversification) elevates joint failure risk”.9 sarily means eliminating the communication of information about
Wagner (2010) shows that while diversification reduces the risk of asset-value impairment across banks, a potentially valuable mar-
an individual bank, it increases systemic risk. ket mechanism for information aggregation. Second, this undesir-
Another strand of literature focuses on contagion that arises for able aspect of regulatory intervention is encountered only when
reasons other than the kind of information spillover that we exam- the intervention decision is based solely on market signals, such as
ine in this paper.10 For instance, contagion can also arise due to in- observed bank liquidations. We show that if the regulator gener-
terconnectedness rather than systematic risk exposures. Caballero ates its own information and intervenes selectively based on that
and Simsek (2013), for instance, argue that increased complex- information, it is possible to eliminate contagion without destroy-
ity due to greater interconnectedness among individual banks and ing market discipline.
ambiguity aversion to such complexity-can generate endogenous
risk and crises. Allen et al. (2012) analyze how asset commonal- 3. The model
ity interacts with debt maturity, so that it is short-term debt per
se that leads to contagion across banks.11 Further, there may be This section develops the basic model and the contracting op-
a fire-sales related externality as liquidating banks’ creditors seek portunities. Consider an economy in which all agents are risk-
out scarce liquidity from depositors, raising funding costs for other neutral and the riskless rate is zero. There are three dates t = 0, 1,
banks, as in the general equilibrium effect of bank liquidations in 2. There are banks run by managers that choose loan portfolios (or
Diamond and Rajan (2005). Thakor (2015) develops a model in investments) at t = 0 that generate payoffs at t = 1 and t = 2. There
which the combination of the “availability heuristic” and a long are two types of loans (or projects), both of which require liquidity
sequence of good outcomes causes all agents to become exces- (investment) I at t = 0. If this liquidity need is met at t = 0, then the
sively sanguine about risk taking, leading to episodes of (corre- loan generates a random cash flow x at t = 1, with density function
lated) high risk taking by banks, whereas Thakor (2016) focuses on g(x) and cumulative distribution G(x). The support of g is [0, xmax ].
how the assessment of risk by financial institutions and its pric- Each bank is initially owned by shareholders who have no liquid-
ing by investors changes with the business cycle, thereby inducing ity of their own, so they seek to sell debt and equity claims against
correlations across banks in asset choices at different points in the the bank in a competitive capital market as to maximize the total
business cycle. The information contagion we examine is comple- value of the bank and hence the revenue raised. Out of this rev-
mentary to these channels, but distinct in three respects. First, our enue, the manager’s initial wage is paid and the chosen loan port-
model also permits beneficial information contagion. Second, the folio (“loan” henceforth) is financed. The rest is consumed by the
independently-made endogenous capital structure choices of banks initial owners.
act as a crucial amplification mechanism for contagion. And third,
3.1. Types of loans

9 The two types of mutually-exclusive loans available to the bank

Brunnermeier and Sannikov (2016) develop a model in which banks diversify
the idiosyncratic risks of households (with resulting correlations in their equity) are the “good” loan and the “private-benefit” loan.12 The good loan
and their capacity to act as “diversifies” depends on their capital. produces a cash flow at t = 2 of Hx with probability q ∈ (0, 1 ), and
10
We note that while information externality from bank runs has been exten- zero with probability 1 − q. The private-benefit loan produces a
sively modeled in the literature, see, e.g. Chari and Jagannathan (1988), and recently cash flow at t = 2 of Hx with probability p ∈ (0, q ), and zero with
Acharya and Yorulmazer (2008), in our model information spillover can also be ef-
ficient as (some) relevant asset value shocks relevant for a bank’s solvency assess-
probability 1 − p at t = 2. Both loans have the same density func-
ment are obtained only through creditor of other banks (when they “run”). tion of the date-1 cash flow, x, accruing to the bank. The bank’s
11
Moreover, financial innovation incentives can also generate systemic risk as in manager gets a private benefit of B per unit of x, i.e., a total benefit
Thakor (2012). In that model, some institutions take the lead in coming up with of Bx at t = 2. We assume that the private-benefit loan is socially
innovative products because these generate higher expected rents than “standard” inefficient relative to the good loan:
products. The reason is that there may be disagreement about whether the innova-
tion will succeed, and this deters competition. But there is a positive probability of qH > [ pH + B]. (1)
many follower institutions imitating the innovation leader, and when this happens,
a systemic risk is generated due to the likelihood that the creditors of these insti-
12
tutions may disagree at a future date that the innovation is good and therefore cut The model could also potentially be written in terms of a risk-shifting or asset-
off funding. Huang and Ratnovski (2011) also observe that the possibility of funding substitution moral hazard, instead of a private benefit project, as in Acharya and
being cut off is an aspect of wholesale funding. Viswanathan (2011) or Acharya, Mehran and Thakor (2016).
8 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

Moreover, the good loan is socially efficient: We assume that a monitoring technology is available to cred-
itors that would enable them to discover the bank’s loan choice
qHE (x ) > I, (2)
at a cost κ > 0. If they dislike this loan choice, they can ask the
where E(x) is the expected value of x. The good loan and the bank’s manager to change it. If the manager refuses, and financiers
private-benefit loan are mutually exclusive. Moreover, the bank are creditors, they can deny renewal of funding at t = 1 for the
manager makes an initial loan choice at t = 0 but can costlessly second period. This lack of funding renewal may force the bank
switch projects at t = 1. to liquidate in order to meet its repayment obligation. Faced with
We assume that all cash flows are pledgeable but we will show this prospect, the bank manager may agree to switch loan choice.
that they will not always be pledged in their entirety in equilib- It follows immediately that the bank’s creditors will refuse to re-
rium due to the provision of incentives to bankers. The bank can new funding at t = 1 if they discover the manager has not chosen
raise the required initial liquidity via debt or equity whose features the loan they desire. If the bank can fully pay off creditors from its
we describe next. The debt is uninsured. Moreover, the incumbent cash flow at t = 1 and still continue, it will choose to do so. But if
bank manager possesses unique human capital to manage either of this cash flow is insufficient, then the creditors’ threat of liquida-
the two loans. Transferring the management of the loan to another tion will have bite and a loan change will occur.14
manager substantially reduces loan value. In contrast, if shareholders were to expend κ in monitoring ef-
fort and discover a loan choice they did not like, they have no
3.2. Debt and equity contracts and assets portfolio liquidity credible threat that they can make to force a change. This is be-
cause equity has no finite maturity and shareholders cannot ask
If the bank raises debt financing at t = 0, it is assumed that for their investment to be returned at t = 1.15
the debt contract contains a covenant whose violation at t = 1 This disciplining role of debt (but not equity) via a withdrawal
permits the creditors to take control and demand full repayment threat is familiar from the previous work of Calomiris and Kahn
at t = 1. Asking for full repayment could result in the bank being (1991), and Diamond and Rajan (2001). The very nature of the debt
liquidated at t = 1 if its interim cash flow is insufficient to make contract facilitates monitoring-based market discipline that is not
the repayment. Such “accelerated repayment” clauses are standard possible with equity (see also Hart and Moore (1995)). The fact
in debt contracts. If the bank is not liquidated at t = 1, then it will that creditors can deny renewal of funding at t = 1—through ei-
continue until t = 2. Another way to think about this is to view the ther a covenant trigger that leads to a demand for accelerated re-
debt contract as short-term, i.e., having a one-period maturity, so payment on two-period debt or a denial of renewal of one-period
that the creditors decide at t = 1 whether to renew funding for an debt—is also similar to repo market funding drying up for financial
additional period after the debt matures or simply collect whatever institutions during the recent crises.16 While the role of the bank’s
the bank can repay at t = 1 and deny renewal of funding. creditors as monitors is sometimes questioned on the grounds that
The face value of debt (promised repayment to creditors) is F. these creditors are “outsiders” and therefore should have access to
The date-1 cash flow is pledged to the creditors up to this face less information than say bank regulators, it should be remem-
value of F, so the remaining repayment to them at date 2 (in case bered that these are uninsured bondholders. Thus, they are more
the bank is not liquidated at date 1) is max(F−x, 0). It is assumed like the subordinated debtholders in banks than (insured) deposi-
that the liquidation value of the bank at t=1 is dependent on the tors. The market discipline potential of subordinated debt has been
realized date-1 cash flow x. This liquidation value is Lx. We assume recognized even in prudential bank regulation, e.g., it is one of the
that pillars of Basel II.
qH > L > pH + B. (3)
That is, liquidation is better than the private-benefit loan under
the manager, but worse than the good loan. One way to interpret
3.3. Lack of pre-commitment
the liquidation value is to view it as the value of the loan if its
management is transferred to another manager.
The first-best is achieved if the manager could make a pre-
If equity is used instead of debt, it is assumed that sharehold-
commitment at t = 0 that he will invest in the good loan and not
ers have no control rights at t = 1 other than over the cash flow
switch at t = 1 to the private-benefit loan. However, we assume
x. That is, shareholders cannot “withdraw” their equity investment
that such commitment is not credible other than through incen-
in the firm at t = 1 by liquidating the bank. We just take this fea-
tives provided by debt and equity.
ture of the equity contract relative to debt as a given. That is, our
purpose is not to endogenously derive debt and equity as (con-
strained) efficient contracts in a security-design setting. However,
if the shareholders wish, they can provide incentives for the man-
also entails costs to the bank. For simplicity, we do not model these costs. We are
ager to choose the value-maximizing loan by giving him a suitably implicitly assuming that even when assets are liquidated by creditors, managerial
chosen share θ ∈ (0, 1 ) of the shareholders’ payoffs (that is, resid- landing is not as hard as that for shareholders as managers may have alternative
ual payoffs at date-1 and date-2 after payments to creditors). It labor-market options and/or a role to play in the bank’s workout possibly due to
asset-specific knowledge not available to creditors, workout specialists or regulators.
is assumed that the manager values this ownership, which repre-
The primary purpose of ρ , the divergence between managerial and firm’s discount
sents a claim on future cash flows, at a discount factor ρ ∈ (0, 1 ) rates on future cash flows, is to introduce a cost for equity-based governance (in
of the value assigned by the shareholders. The valuation divergence the form of incentive compensation) relative to debt-based governance (in the form
between the shareholders and the manager could arise from fun- of monitoring and contingent liquidations).
14
damental disagreement (e.g., Boot and Thakor (2011), Boot et al. Whenever creditors liquidate a bank, it is in their collective best interest to do
so. In that sense, a “run” on the bank here is different from a run caused by a
(20 06, 20 08), and Van den Steen (2010)) or simply risk-aversion
coordination failure among depositors as in the usual bank runs story.
or managerial myopia. It reflects the fact that the manager values 15
Moreover, because of the nature of the equity contract, shareholders have in-
his future equity-based compensation less than his up-front fixed herently weaker incentives than creditors to liquidate the bank. Continuing (rather
wage.13 than liquidating) is always more attractive for equity than for debt.
16
When short-term financiers in the repo market became aware of adverse infor-
mation about financial institutions, the short-term funding they had provided dried
13
It might well be the case that managers also discount more than shareholders up. This aspect of wholesale funding of banks has also recently been examined by
the states in which the firm is liquidated, so that debt-based incentive provision Huang and Ratnovski (2011), though they focus only on its negative side-effects.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 9

3.4. Managerial compensation is that if one bank fails, some of its services can be provided by
the remaining banks, but less and less of this re-intermediation is
The manager’s reservation wage is W̄ > 0. This can be paid with done as more banks fail, so social costs of bank failures increase at
any combination of a fixed wage and incentive compensation. The an increasing rate.
fixed wage is W f and the equity-based compensation is We . The
fixed wage is paid up-front at t = 0 and We is paid at t = 2. 3.8. Observability

3.5. Interim looting by shareholders We assume that the interim cash flow x of each bank is not
observable to any party other than the bank’s shareholders, credi-
If shareholders are permitted to pay themselves a dividend at tors and the manager. Moreover, the manager’s loan choice is not
t = 1, they can essentially “take their money and run” by pay- directly observable to anyone but the manager himself, unless fi-
ing themselves something analogous to a liquidating dividend. This nanciers intervene and are able to enforce a different loan choice.
payment leaves the bank with no cash flow at t = 2, but the cash The realized value of the systematic shock to a bank’s asset
flow it generates at t = 1 is lower by an amount  > 0 than the value is not observed with certainty by the bank’s creditors. The
expected cash flow of the bank at t = 2 had this “looting” not oc- probability that the bank’s creditors will observe the systematic-
curred at t = 1, i.e., the looting is socially inefficient. This dividend shock signal ξ is γ ∈ (0, 1 ). Conditional on a particular ξ , the ran-
extraction can occur at t = 1 after the bank’s creditors have en- dom variable γ is independent and identically distributed across
forced the choice of the preferred loan portfolio and decided to let banks in the economy. The realization of the systematic shock ξ
the bank continue, and it is possible only if a dividend payment is affecting a bank and the bank’s cash flow x are privately observed
permitted at t = 1. (if at all) only by that bank’s manager, creditors and shareholders.
There is also a free-cash-flow problem (see Jensen (1986)) such However, the liquidation of a bank by its creditors is commonly
that any cash x generated at t = 1 that is left in the bank until observed by all agents.
t = 2 is worth only mx, with m ∈ (0, 1 ). The sequence of events in the model is summarized in Fig. 1.
Given these assumptions, it is clear that it is efficient for the
bank to pledge its entire interim cash flow x to creditors. Since
4. Analysis of the model
paying it out to the shareholders requires permitting a dividend
at t = 1 and this can enable the shareholders to inefficiently loot
First, we present an analysis of the model developed in the pre-
the bank at t = 1 by paying themselves a liquidating dividend (in
vious section by considering the relative advantages of debt and
which the manager will participate due to the equity ownership
equity financing when there is no systematic component to the as-
in his compensation), it will be efficient for the bank to pay it out
set value shock. We begin with an analysis of the events at t = 1,
to the bank’s creditors at t = 1 and impose a ban on any dividend
and then we turn to the events at t = 0.
payments by the bank at t = 1.17

4.1. Analysis of the model without the systematic asset value shock:
3.6. Systematic shock to asset values
managerial incentives and creditors’ liquidation decision at t = 1
We assume that asset values are subject to a systematic shock,
Consider first the case in which the bank finances with all eq-
represented by the realized value of an underlying state variable, ξ ,
uity at t = 0. To induce the manager to select the good loan at t = 1,
that is experienced at t = 1. With probability β ∈ (0, 1 ), a system-
shareholders must provide the manager with equity that repre-
atic asset-value impairment shock is realized as ξ = ξ . This shock
sents a fraction θ ∈ (0, 1 ) of ownership in the bank. The value of θ
lowers the value of H (for both types of loans) to H − < H, with
must satisfy the incentive compatibility (IC) constraint point-wise
qH − < L. With probability 1 − β , the state variable is ξ = ξ h , and
at t = 1 for every realized x.19
asset value is now higher at H + > H, with β H − + [1 − β ]H + = H.
ρθ qHx ≥ ρθ pHx + Bx.
3.7. The social cost of bank failures
Since this IC constraint is binding in equilibrium, we can write the
optimal value of θ , call it θ ∗ , as:
We assume that the failure of a bank creates a social cost
that is not directly reflected in the payoffs of the bank’s share- θ ∗ ≡ B[H ρ
]−1 , (4)
holders, creditors or borrowers. There are many ways to interpret
these costs. Perhaps the most direct is that bank lending gen- where
≡ [q − p]. That is, equity provides incentives to the man-
erates positive externalities for the economy by elevating invest- ager via a payoff-contingent compensation contract. To meet the
ments by small and medium-sized businesses—the credit supply manager’s reservation wage, a fixed wage component may also be
effect—which then boosts employment, tax revenues and so on, necessary. The following result is immediate from our assump-
with attendant social benefits. Bank failure will therefore impose tions:
a cost by creating a loss of these benefits. Another relevant cost
Lemma 1 (Optimality of up-front fixed wage for managers). If the
would be that payments and settlement systems could be crippled
manager is given a fixed wage, W f , it is optimal for the shareholders
due to failures of banks and contagious effects on inter-connected
to pay the fixed wage up-front at t = 0 rather than at a future date.
banks.18 Let Nn > 0 be the social cost of failure associated with
the failures of n banks, with ∂ Nn /∂ n > 0, ∂ 2 Nn /∂ n2 > 0. The idea The intuition is straightforward. A dollar of compensation at t =
0 is valued by the manager at a dollar, but a claim on a dollar
17
of future compensation is valued by the manager at ρ . Thus it is
As we saw driving the financial crisis, when capital conservation is a concern,
not restricting/banning dividend payments can induce banks to pay out large divi-
cheaper for the shareholders to pay the manager his fixed wage
dends to expropriate wealth from the creditors. See Acharya et al. (2013). up-front.
18
A less direct benefit of bank lending may be that the higher employment also
leads to less crime. Garmaise and Moskowitz (2006) document that a reduction in
19
bank lending in a community leads to an increase in crime. Thus, another cost of The reason why the IC constraint must hold point-wise for every x realized at
bank failure may be higher crime. t = 1 is that the manager makes his project choice at t = 1 after observing x.
10 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

Fig. 1. Sequence of Events.

Next we turn to the case in which the bank has been financed creditors do not expend κ to investigate, and unconditionally allow
with debt at t = 0. Having collected x from the date-1 payoff, cred- the manager to continue with the loan. Conditional on having ex-
itors will assess their payoffs from liquidation and continuation pended κ , creditors adopt the following liquidation/continuation pol-
(when they enforce the efficient loan choice) as follows: icy:
- Continue if x ≤ λF ;
Liquidate: min(Lx, F − x ) (5)
κ
- Liquidate if x ∈ (λF , F − );

Continue : qmin(Hx, F − x ) (6)

where λ ≡ q[q + L] −1 ; All liquidations by creditors are (ex-post) inef-
We now have the following result: ficient. When x ≥ F − κ [
] −1 , shareholders provide incentives to the
manager by giving him ownership θ ∗ of the bank’s terminal payoff,
Lemma 2 (Creditors’ monitoring-contingent liquidation policy
where θ ∗ is given by (4).
without systematic shock). In absence of the systematic shock ξ ,
the bank will adopt a debt contract in which the creditors set the The different cases are pictorially depicted in Fig. 2. The intu-
covenant violation trigger at F − κ [
]−1 , so that if x < F − κ [
]−1 ition is that when x is very high, unconditional continuation is op-
creditors expend κ to discover the manager’s loan choice at t = 1 and timal for the creditors at t = 1 because their claim is covered out
enforce a different loan choice if they so desire. If x ≥ F − κ [
]−1 , of just the date-1 cash flow, and it does not matter to them what
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 11

Fig. 2. The Continuation/Liquidation Decision of Creditors.

project the bank invests in. In these states in which creditors are the manager’s fixed wage, which then reduces the initial maximum
paid off out of the date-1 cash flow, shareholders provide the nec- liquidity the firm can raise via external financing at t = 0. This ex-
essary project-choice incentives for the continuation by awarding ante liquidity reduction is reduced with debt financing. The dis-
the manager an ownership share of the bank. That is, when x is advantage of debt financing is that it creates a region [λF , F −
κ]
high, compensation incentives provided by equity replace the mon- over which the bank is (ex-post) inefficiently liquidated. In what
itoring discipline provided by debt. When x is very low, creditors follows, we shall assume that κ is arbitrarily small by letting κ =
cannot be paid off fully from the date-1 cash flow, but they pre- 0 because doing so reduces notation without qualitatively affecting
fer to continue rather than liquidate the bank. This is because the the analysis.
liquidation payoff Lx is so low that it is better for creditors to take
a chance on a higher payoff qHx by continuation. For intermediate 4.2. The bank’s optimal capital structure without the systematic asset
values of x, liquidation is optimal for creditors because they get a value shock
sufficiently high certain payoff of Lx so that given the “risk aver-
sion” induced by concavity of the debt contract, it does not pay for For simplicity, it will be assumed now that g(· ) is uniform on
the creditors to gamble on a risky continuation. [0, xmax ]. Let We (F , θ ) be the manager’s equity compensation as as-
This analysis reveals the pros and cons of debt financing. With sessed by the shareholders, and let W f (F , θ ) be the fixed wage.
debt financing, the manager will not have to be provided a com- Both are functions of the face value of the bank’s debt and the
pensation incentive over the entire range of x, as is the case when fractional ownership θ ∗ given to the manager. Thus,
only equity is used. These incentives need to be provided with 
κ . Thus, debt reduces the region over F
debt only when x > F −
We (F , θ ∗ ) = θ ∗ [(x − F ) + qHx]g(x )dx, (7)
which compensation-based incentives must be provided from [0, x
κ ,x
xmax ] to [F −
20 Since
max ]. This is the benefit of debt financing. where E (· ) is the expectation operator and θ ∗ is given in (4).
performance-based compensation is disliked by the bank manager, The manager’s valuation of his equity compensation, denoted
it induces a concomitant and compensating upward adjustment in Wm (F , θ ∗ ), is:
 F

20
Wm (F , θ ∗ ) = ρθ ∗ [(x − F ) + qHx]g(x )dx. (8)
One implication of this feature of our model is that in banks with (exogenously) x
greater leverage, managerial compensation is less important for incentive purposes
and thus a manager’s compensation package would feature a greater share of fixed Then, assuming a competitive market among banks for hiring
or “pay without performance” component. managers, it can be shown that the manager’s reservation utility
12 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

constraint will be binding in equilibrium, so that the manager’s the first stage). In particular, bank i’s creditors have no incentive
fixed wage is: to decide not to liquidate in stage 2 if they decided to liquidate
in stage 1.21 But, after observing the creditors of the other bank
W f (F , θ ∗ ) = W̄ − Wm (F , θ ). (9)
liquidate, they may decide to liquidate in stage 2 even if they an-
Given the liquidation and continuation regions identified in nounced in stage 1 that they would not liquidate. We assume (and
Lemma 2, we can write the bank’s optimal capital structure as the derive a condition below) such that if a bank is liquidated, the in-
solution to the following problem of maximizing the total value, ference about ξ for the creditors of the other bank is sufficiently
V (F ), of the bank: adverse that they liquidate too.
 λF    F  Note that conditional on the creditors observing ξ = ξ  , the
Max x Lx
V (F ) = E (x ) + qH dx + dx creditors will liquidate if
F o xmax λF xmax
 xmax   qH − < L, (13)
x
+qH dx − We (F , θ ∗ ) − W f (F , θ ∗ ) (10)
F xmax which we assumed earlier. To figure out the sufficient condition for
It is convenient to define an upper bound on the manager’s pri- our assumption that the creditors of bank i will liquidate in stage
vate benefit as: 2 if they observe the creditors of bank j liquidating, let us calculate
  the posterior belief of creditors of bank i (if they observe the cred-

1 − λ2 [qH − L] itors of bank j liquidating) concerning the aggregate shock being
B̄ ≡ . (11)
q [1 − ρ ] adverse. At this stage, we need to introduce some notation that is
We now obtain the following result. useful for examining the inference problem from the perspective
of bank i. Define for bank j:
Proposition 1 (Bank’s optimal capital structure). Assume B < B. 
 F − λFj [1 − λ]Fj
Then the bank’s optimal (value-maximizing) capital structure includes δ j ≡ Pr x ∈ λFj , Fj = j = , (14)
an amount of debt financing F ∗ ∈ (0, xmax ) where:   xmax xmax
δ1 ≡ β [1 − γ ] 1 − δ , and
j 2 j
(15)
xmax   
F∗=  . (12)
[1 − λ2 ][qH − L]{[1 − ρ ]θ ∗ }
−1
− [qH − 1] δ2j ≡ [1 − β ] γ + [1 − γ ] 1 − δ j . (16)

The condition that B not be too large is needed because other- Note that δ j is simply the probability that bank j’s x ∈ (λFj , Fj ) and
wise the need for debt discipline is so large that the bank goes to hence it is liquidated by its creditors solely due to the realization
a corner optimum of all debt. As long as this is not the case, there of its own cash flow.
Next, δ1 is the joint probability that the adverse asset-value
j
is an interior value of leverage, F ∗ , that maximizes the value of the
bank. This optimal value balances the cost of inefficient liquidation shock was realized (the probability that ξ = ξ is β ), neither bank
due to leverage against the benefit of leverage in disciplining the observed the shock (probability (1 − y )2 ), and bank j did not liq-
manager without the dissipative cost of incentivizing him with uidate due to a cash-flow realization x ∈ (λFj , Fj ) (the probability
equity. that x ∈/ (λFj , Fj ) for bank j is 1 − δ j ).
The next result provides comparative statics on F ∗ . Similarly, δ2 is the joint probability that the favorable asset-
j

value shock was realized (probability 1−β ), and either (i) it was
Corollary 1 (Comparative statics on the bank’s optimal lever-
observed by the bank j (probability γ ), or (ii) it was not observed
age). F ∗ is increasing in the manager’s private benefit, B, and de-
by bank j (probability 1 - γ ) and bank j did not liquidate due to
creasing in the manager’s discount factor for equity valuation, ρ .
its own realization of x being in (λFj , Fj ) (which has probability
1 − δ j ). Thus, δ1 + δ2 is the probability that bank i is not liquidated
This result is intuitive. As B increases, the need for debt disci- j j

pline grows and leads to higher optimal leverage. As the manager’s when its own x ∈ (0, λFi ), i.e., the probability that it is not liqui-
discount factor, ρ , increases, using equity compensation to incen- dated when its own interim loan cash flow is low enough to not
tivize him becomes less expensive, so optimal leverage declines. warrant liquidation on the basis of that cash flow.
In calculating the posterior belief that the aggregate shock is
4.3. Analysis of the model with the systematic asset-value shock: the adverse, what we are assuming is that the bank in question (bank
creditors’ liquidation decision at t = 1 i) did not receive the signal about the shock, decided not to liqui-
date in stage 1 but observed the other bank (bank j) liquidating in
Thus far, we have examined the monitoring and liquidation de- stage 1. Then bank i forms the following posterior belief about ξ :
cisions of creditors and the contracting decision of banks when as-
set values are not subject to the systematic asset-value shock. We Pr (ξ = ξ |bank j announced liquidation in stage 1 )
now include this systematic shock, so that one bank’s liquidation Pr (bank j liquidated in stage 1|ξ = ξ ) Pr (ξ = ξ )
=
can convey information about the shock to another bank. Pr (bank j liquidated in stage 1)
Imagine there are two banks in the economy and conditional    
on the systematic shock ξ , each bank is faced with identically and
γ λFj [xmax ] −1 + δ j β
=    
independently distributed (i.i.d.) shocks. In this two-bank economy, γ λFj [xmax ] −1 + δ j β + [1 − β ] δ j
we also need to specify how the banks observe each other’s liqui- 
dation outcomes and how this affects the liquidation decisions of [Fj − [1 − γ ]λFj ]β
creditors of each bank. We model this as a two-stage game.
=  
βγ λFj + Fj [1 − λ]
First, each bank’s creditors announce, simultaneously with the
creditors of the other bank, whether they will liquidate the bank [1 − λ[1 − γ ]]β
= ≡ βˆ . (17)
based solely on their own information about the bank. In this state, {1 − λ[1 − γ ] − (1 − λ )}β + 1 − λ
neither bank’s creditors have access to the decision of the other
bank’s creditors. 21
There are no strategic manipulation incentives in the model. In particular, there
Second, after having observed each other’s first-stage liquida- is nothing to be gained for the creditors in one bank from liquidating or not liq-
tion outcomes, each bank’s creditors decide whether they wish to uidating a bank in order to strategically manipulate the behavior of creditors in
liquidate in the second stage (provided they did not liquidate in another bank.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 13

Note that this probability is positive only if bank i did not observe a bank’s creditors was ex-post inefficient (see Lemma 2). That is
ξ = ξh in the first stage. Further βˆ > β . no longer true, however, when the systematic asset-value shock ξ
To understand the numerator of (17), note that, conditional on is introduced. Now there can be an ex-post efficiency gain from
ξ = ξ , bank j can be liquidated in one of two cases: (i) if its cred- liquidation in the case where the creditors of bank i do not ob-
itors observe ξ (which has probability γ ) and bank j’s cash flow serve ξ but liquidate based upon observing the liquidation of bank
x ≤ λFj (which has probability λFj [xmax ]−1 ), or (ii) bank j’s cash j whose creditors observe ξ = ξ . Proposition 2 shows that bank
flow realization x ∈ (λFj , Fj ) (which has probability δ j ). Hence, the i’s own leverage increases the probability of liquidation of bank i.
probability of being in one of these two states is λF [xmax ] −1 + δ j . Thus, an increase in a bank’s leverage contributes to ex-post ef-
This probability is multiplied with β , the probability that ξ = ξ . ficiency in such states. Hence, with the introduction of the sys-
In the denominator, there is an additional term, [1 − β ], which is tematic asset-value shock, leverage-induced liquidity creation has
the probability that bank j is liquidated due to its x ∈ (λFj , Fj ) even both a dark side (liquidations that are sometimes inefficient) and
when ξ = ξh (which has probability [1 − β ]). a bright side from an ex-post standpoint.22 It is the dark side that
Now, for bank i’s creditors to decide to liquidate (only) after contributes to a leverage-engendered increase in systemic risk.
having observed bank j’s creditors liquidations, we need two con- Note also that we have assumed that creditors in different
ditions: banks are distinct. What if say a single creditor could take diversi-
fied positions in all banks? On the one hand, this will eliminate in-
Condition 1. Creditors will not unconditionally liquidate before ob- efficient liquidations because there is no contagion to worry about.
serving ξ or a liquidation of another bank: On the other hand, efficient liquidations will also decline because
 
q β H − + [1 − β ]H + > L, the probability of receiving information about the systematic shock
will be limited to the probability that one bank will receive this
which implies signal.
qH > L, (18)
which we have assumed throughout. 4.4. The value of the levered bank and optimal leverage at t = 0

Condition 2. It pays for the creditors of bank i to liquidate if they We can now move to t = 0 and examine the bank’s optimal cap-
observe liquidation by the creditors of bank j: ital structure decision with the systematic asset-value shock. Now
    bank i’s value function can be written as:
q βˆ H − + 1 − βˆ H + < L (19)  Fi  Lx   xmax  x 
V (Fi ) = E (x ) + dx + qH dx
where βˆ is defined in (17). Define L̄ as the value of L at which (19) λFi xmax Fi xmax
holds as an equality, so that the inequality in (19) will hold for all    λFi  x 
L > L̄. + δ1j H − + δ2j H + q dx
0 xmax
Condition 3. When both banks are liquidated, the liquidations are    λFi  x 
socially inefficient, although liquidation of only one bank based on ob- + 1 − δ1j − δ2j L dx
0xmax
serving ξ = ξ is efficient.
− We (Fi , θ ∗ ) − W f (Fi , θ ∗ ). (23)
2qH − + N2 > 2L (20)
To understand (23), note that its structure is similar to (10). The
first term, E(x), is the expected value of the interim cash flow.
qH − + N1 < L (21) The second term is the liquidation by creditors of bank i that
occurs when x ∈ [λFi , Fi ], and the value is Lx. Note that, given
Given conditions under which the creditors of bank j liquidate if
condition 2, bank i knows that if it liquidates, so will bank j.
they observe liquidation by the creditors of bank i, the probability
The third term refers to the state x ∈ (Fi , xmax ]. In this state,
that bank j will not be liquidated is
 2   there is no liquidation or creditor monitoring of bank i, so the
δ0 = 1 − δ j β [1 − γ ]2 + [1 − β ] (22) manager is incentivized by shareholders with an equity compen-
sation contract.
since not being liquidated requires that no bank’s creditors liqui-
To understand the fourth term, note that 1 − δ1 − δ2 is the
j j
date, due to adverse cash flow shocks (probability [1 − δ j ]2 ) and
probability of liquidation of bank i that is not based on bank i’s
either that neither bank’s creditors observe the adverse systematic
realization of x when x ∈ [0, λFi ], but based on that of bank j,
shock if it occurs (probability β [1 − γ ]2 ) or that the adverse sys-
and δ1 + δ2 is the probability of no liquidation of bank i, when
j j
tematic shock did not occur (probability 1 − β ).
x ∈ [0, λFi ]. As explained earlier, δ1 is the joint probability of ξ = ξ
j
We now have:
and no liquidation of bank j, and δ2 is the joint probability of
j
Proposition 2 (Dependence of a bank’s liquidation probability on
ξ = ξh and no liquidation of bank j. Note also that if ξ = ξ is re-
its own leverage and the other bank’s leverage). For a given lever-
alized, the expected value of date-2 cash flows is qH − per unit of
age of bank j, Fj , the (unconditional) ex-ante probability of liquidation
x and if ξ = ξh is realized, this value is qH + per unit of x. This
of bank i is increasing in bank i’s leverage, Fi . Moreover, fixing its own
explains the fourth term.
leverage, Fi , the unconditional probability of liquidation of bank i is
The fifth term refers to the liquidation of bank i that occurs
increasing in bank j’s leverage, Fj .
because creditors observe the liquidation of the other bank j, i.e.,
The intuition can be seen by observing that higher leverage of this is liquidation that is not based on bank i’s own x. This occurs
with probability 1 − δ1 − δ2 when for bank i we have x ∈ [0, λFi ],
j j
bank j makes it more likely that bank j will be liquidated. This is
because higher leverage of bank j reduces the probability that bank and the liquidation payoff is Lx.
j’s interim cash flow, x, will exceed its leverage Fj . Wherever bank
j is liquidated, bank i’s creditors noisily infer that state ξ = ξ is 22
We have conducted this analysis for the two-bank case to convey these ideas
likely to have occurred, and liquidate bank i. as transparently as possible. The n-bank case, with n > 2, is qualitatively similar,
Recall that in the previous analysis that was conducted in the although it permits a weakening of the restrictions on the exogenous parameters
absence of the systematic asset-value shock ξ , any liquidation by that are needed for the dark side of leverage, i.e., the liquidation contagion.
14 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

The equity compensation shows up in the sixth term (recall The proposition characterizes the best response of a bank in terms
that the seventh term is the fixed compensation). of its choice of leverage (given the leverage of the other bank) that
Now, note that the leverage of the other bank, j, is contained in maximizes the value of the bank and hence its ex-ante liquidity.
the probabilities δ1 and δ2 . Formally, let Vi (Fi , Fj ) denote the value
j j
The proposition asserts that F 0 is a unique symmetric Nash equi-
of bank i given the leverage of the two banks Fi and Fj . Then the librium outcome. Note that each bank’s leverage choice has to be
best-response of bank i in terms of its leverage choice Fi0 (Fj0 ) max- examined as part of a Nash equilibrium because each bank’s lever-
age affects the other bank’s expected payoff (formally through δ1
j
imizes Vi (Fi , Fj0 ) for a given leverage of bank j, Fj0 . The Nash equi-
and δ2 ), and each bank’s best response in any equilibrium must
j
librium of private leverage choices (Fi , Fj ) satisfies the fixed-point
satisfy (27). While (27) expresses each bank’s leverage as a func-
property Fi = Fi0 (Fj ) and Fj = Fj0 (F i ).
tion of the other bank’s leverage, (28) expresses the symmetric
Define:
Nash equilibrium choice purely as a function of exogenous param-
[1 − λ]qH xmax eters (recall θ ∗ is expressed purely as a function of exogenous pa-
δ≡ = [1 − λ]qH (24)
xmax rameters via (4)).
and assume that the symmetry condition L − H − = H + − L holds The following corollary establishes comparative statics on F 0 .
j
and that β ≤ 0.5. Note that δ takes when the face value of debt Corollary 2 (Comparative statics on optimal capital structure). The
is at its maximum possible value, qHmax , ignoring the managerial privately optimal leverage F 0 is increasing in the manager’s private
wage. Further, define: benefit, B, and the likelihood, γ , that the systematic state is visible to
   bank creditors, and decreasing in the probability, β , of the systematic

qH − L − λ2 δ̄2 [H + − L] − δ̄1 [L − H − ] asset-value impairment shock, and the managerial discount factor, ρ .
Bˆ ≡ (25)
[1 − ρ ]q
The intuition for the impact of B on F 0 is similar to that in our
where earlier discussion. As (22) clarifies, an increase in B increases θ ∗
 
δ̄1 ≡ β [1 − γ ]2 1 − δ̄ and the illiquidity arising due to equity-based compensation, and
debt becomes more attractive. The intuition for γ is straightfor-
    ward in that as debt becomes more informed about the nature of
δ̄2 ≡ 1 − β γ + [1 − γ ] 1 − δ̄ the systematic shock, inefficient liquidations by creditors are fewer
and the deadweight costs due to debt become lower.
Bˆ is similar to the B̄ defined in (11). The following result is use- Finally, the intuition for the impact of β is as follows. The main
ful. role of leverage is to discipline the manager via the threat of bank
Lemma 3. Define liquidation. While bank liquidation by creditors, based on having
   observed a signal that asset value is impaired, also requires the


qH − L − λ2 δ2j [H + − L] − δ1j [L − H − ] presence of some leverage, the probability that the creditors of
B˜ Fj ≡ (26)
[1 − ρ ]q bank i will observe this signal is unaffected by the leverage of bank
i. When β increases, it becomes more likely, ceteris paribus, that
Then ∂ B˜(Fj )/∂ δ j < 0. the creditors of bank i will liquidate based on observing a liqui-
dation of bank j. This contagion effect makes debt less attractive
This means that B˜(Fj ) is minimized when δ j takes its maximum
ex ante. Finally, as the manager’s discount for equity-based com-
value, which occurs when Fj is at its maximum possible value.
pensation decreases (that is, the value the manager assigns to risk-
Thus, where the infimum is over all possible values of Fj : Bˆ ≤
based compensation goes up), the attractiveness of debt declines
inf B˜(Fj ). Defining Bˆ this way allows us to state the following char- relative to equity financing.
acterization of the best responses and the symmetric Nash equi-
librium, without having Bˆ depend on the actual leverage choice of 5. The effect of a regulator
any bank.
We now examine the role of the regulator. We think of the reg-
Proposition 3 (Optimal capital structure of each bank in symmet-
ulator as a government or a central bank that has two policy in-
ric Nash equilibrium). Assume B < Bˆ. Then, in any Nash equilibrium,
struments at its disposal: (i) ex-ante capital requirements and (ii)
bank i’s best response, Fi0 (Fj0 ), at t=0 to bank j’s (privately-optimal)
bailing out failed banks ex post. We assume that the regulator’s
leverage, Fj0 , is: ex-ante objective is to maximize

Fi0 Fj0 Vi + V j − expected social cost of bank failures

[1 − ρ ]θ ∗ xmax − expected costs of regulatory actions (29)

=     , i.e., the sum of the values of the two banks minus the expected
qH −[1 − ρ ]θ ∗ [qH −1] − L + λ2 δ2j [H + − L] − δ1j [L − H − ]
social cost of bank failures but also subtracting the cost of any reg-
(27) ulatory interventions.
where andδ2j δ1j
are functions of Fj0 . In a symmetric Nash equilibrium, We will examine three cases. The first case is when the regula-
each bank’s privately-optimal leverage is: tor pre-commits to not bailing out any failed bank. The next two
√ cases capture less pre-commitment of this sort. The second case


a2 + 4 b − a is when the regulator adopts an unconditional bailout policy (bail
F̄i = Fi Fj0 = Fj0 Fi0 = F̄j = F 0 = (28)
2b out all failing banks). The final case is that of a conditional bailout
{qH −[1−ρ ]θ ∗ [qH −1]L−λ2 [H −L][1−2β +2βγ −βγ 2 ]} policy (bail out banks only in some circumstances as we elaborate
a≡ [1−ρ ]θ ∗ xmax
,
where below). The bailout policy will affect the private sector’s choice of
{λ2 [H−L][1−γ ][1−β{2−γ }][1−λ]}
b≡ . capital structure. In turn, the capital requirement that the regula-
[1−ρ ]θ ∗ [xmax ]2
tor designs ex ante will take account of, and depend on, the bailout
Moreover, F 0 ∈ (0, xmax ).
policy in place.
The condition that the manager’s private benefit not be too The regulatory costs of bailouts will in general depend on the
large is required for the same reasons that were discussed earlier. extent of leverage that is being supported with taxpayer funds in
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 15

the bailouts. For tractability, we will assume that this is not the regulator cares about but individual banks do not—also increases
case and instead assume simply that the cost of bailing out one with bank leverage. Thus, the regulatory optimum drops below the
bank is C1 and the cost of bailout out both banks is C2 . private optimum.
We further assume that C1 > N1 and C2 < N2 . These assump- We now interpret the two conditions in the proposition. When
tions imply that from an ex-post standpoint it is never optimal β is low, (i.e., β ≤ 0.5), the social value of creditors learning about
to bail out an individual bank if the social costs associated with the asset-value impairment shock by observing the liquidations of
bank failure are limited to that one bank only, whereas it is always other banks is also low; since the likelihood of the asset-value im-
optimal ex post to bail out both banks if they are both going to pairment shock is low, so is the value of learning about it. Since
fail (either due to adverse shocks for both banks or due to adverse higher bank leverage leads to a higher liquidation probability, the
shock to one bank and contagion from that failure to the other social value of higher leverage declines as the value of learning
bank). This is meant to capture Bagehot’s (1873) idea that the LOLR about the systematic shock through liquidations declines due to a
should not bail out (or assist) individual banks that are insolvent, reduction in β . The inefficiency associated with contagious liqui-
but should intervene when the payment system is threatened due dations remains unaffected, however. That is, since the privately-
to a meltdown of the whole banking system.23 optimal leverage choice of a bank does not internalize the inef-
Case 1. Regulator never bails out banks ficient contagious liquidations of other banks induced by its own
In this case, the regulators seek to set a minimum capital re- leverage, the bank levers up more than is desirable from the reg-
quirement for each bank that maximizes (29) assuming there are ulator’s (social efficiency) viewpoint, regardless of N2 , the social
no ex-post bailouts. cost of bank failures. When N2 is high enough, however, the so-
Next, we examine how the Nash equilibrium of privately- cially optimal leverage falls below the private optimum regardless
optimal leverage choices compares with the regulatory optimum. of β . That is, the value of creditors learning about asset-value im-
The regulator’s optimal leverage choice ( Fi , Fj ) takes account of pairment through the actions of other creditors is swamped by the
the externalities of bank leverage on other banks and in particu- social costs of bank failures.
lar the social cost of joint bank failures. That is, recognizing that This is the “dark” side of leverage-based liquidity creation by in-
if bank i fails, so will bank j, the regulatory problem is to maxi- dividual banks. Having leverage at the bank level is a desirable ob-
mize Vi (Fi , Fj ) + V j (Fi , Fj ) − [1 − δ0 ]N2 , whose first-order conditions jective from a micro-prudential standpoint but socially costly due
are given by to its perverse implications for macro-prudential outcomes. For-
mally, it is the case that socially-optimal leverage Fˆ has the same
∂ Vi ∂ V j
+ − N2 ∂ [1 − δ0 ]/∂ Fi = 0 (30) comparative statics properties as for the privately-optimal leverage
∂ Fi ∂ Fi F 0 in Corollary 2. Nevertheless, what matters for efficiency is the
difference between the two (F 0 − Fˆ ) and this difference has com-
∂ Vi ∂ V j
+ − N2 ∂ [1 − δ0 ]/∂ F j = 0 (31) parative statics similar to those in Corollary 2. That is, the differ-
∂ Fj ∂ Fj ence between the privately optimal leverage F 0 and the socially
optimal leverage Fˆ is increasing in the manager’s private benefit,
where we recall that δ0 = [1 − δ i ][1 − δ j ]{β [1 − γ ] + 1 − β}. In
2

∂V ∂V B, and the likelihood, γ , that the systematic state is visible to bank

contrast, the private optimum satisfies ∂ Fi = 0 and ∂ F j = 0. creditors, and decreasing in the probability, β , of systematic asset-
i j
Recall that we are assuming the symmetry condition L − H − = value impairment shock, and the managerial discount factor, ρ .
H+ − L. We now discuss the comparative statics generated by the ex-
Then, we obtain the following intuitive result (focusing on the ample above.24 Consider, for instance, the effect of private bene-
symmetric case as before): fit parameter, B. This makes leverage privately more attractive, but
as each bank increases its leverage to minimize on compensation
Proposition 4 (Dark side of liquidity creation: comparison of costs due to equity financing (which are higher for higher B), it ig-
privately-optimal and socially-optimal capital structures with no nores the externality through contagion of increasing leverage on
regulatory bailouts). The regulator’s problem of finding the regula- the other bank. A decrease in the discount factor of the manager
tory optimum is a convex optimization problem with a unique sym- on incentive-based compensation produces a similar effect. And,
metric global optimum. In the (symmetric) Nash equilibrium involving an increase in the visibility of the systematic asset shock to cred-
banks i and j, the privately-optimal leverage F 0 exceeds the socially itors – that is, an increase in the information generation capacity
optimal leverage Fˆ , regardless of N2 if β ≤ 0.5, and regardless of β if of creditors – makes debt privately more attractive, but the private
N2 is sufficiently large. choice again ignores the contagion-based externality imposed on
This proposition captures the essence of our main point high- the other bank.
lighting the systemic risk induced by leverage-based creation of Capital requirements with no bailouts: Since private incen-
liquidity. As discussed earlier, the main role of leverage is to disci- tives to lever up banks are in general inefficient, there is justifi-
pline the manager. However, as bank i’s privately-optimal leverage cation for ex-ante regulatory capital requirements that limit bank
increases, its negative impact on the value of bank j—which derives leverage to a maximum of Fˆ . It is important to point out that
no benefit from better discipline at bank i—also increases; recall absent bailouts, the optimal capital level of the banks in this
that this negative impact arises from bank j being inefficiently liq- case is not 100%, or conversely leverage is not zero. This is be-
uidated when bank i is liquidated only because of an adverse id- cause just as leverage financing of banks in the model has dead-
iosyncratic cash flow realization. Moreover, because higher lever- weight costs for the system due to inefficient contagion from one
age by bank i results in higher liquidation probabilities for both bank’s liquidation to the other bank’s liquidation and the social
banks i and j, the expected social cost of bank failures—that the costs of bank failures, there is also the deadweight cost of equity-
based financing because equity financing requires incentivizing
managers through performance-based compensation that entails
23
Bagehot’s (1873) also recommended that the LOLR should not lend to an in- a higher overall compensation cost to the firm as managers dis-
solvent bank, i.e., it should be concerned with resolving illiquidity problems rather
than insolvency problems. In our model, when the regulator bails out a bank that
24
has been hit with an asset-value-impairment shock, it is assisting a potentially in- These comparative statics on inefficiency of privately optimal leverage are dif-
solvent bank. However, it is an action that prevents contagion and protects the pay- ficult to prove analytically but illustrated graphically in Fig. 3 in Appendix B for a
ment system, something that Bagehot’s (1873) endorsed. numerical example.
16 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

count incentive-based compensation more than cash. Another way uidations impossible, then we have neither a benefit nor a cost
of seeing this is that in the off-equilibrium calculation where the associated with leverage.26 The reason why the regulator causes
bank is entirely financed by outside equity, the compensation re- the bank to have lower ex-ante liquidity is that the disciplining
quired to be paid to the bank manager to ensure correct incen- role of leverage is important for enhancing the bank’s liquidity
tives may be so high that the bank may have little ex-ante liquid- ex ante. Note, however, that this is not a total loss of liquidity
ity in the first place, and it may not be economically viable. Bank as in some models where the absence of the disciplining role of
leverage, in contrast, has the benefit of disciplining the banker in a leverage means that the loan financed by the bank becomes com-
more incentive-efficient manner, implying that there is some ben- pletely illiquid. The equity compensation-based incentives do pro-
efit to bank leverage even in the socially-optimal case. vide ex-ante liquidity, but this is not as much as is available with
Case 2. Unconditional Bailouts: Loss of market discipline role the optimal combination of debt and equity in the absence of the
of debt regulator.27
Given that contagious liquidations are inefficient, there may This is somewhat of a paradox. The role of the regulator is to
also be a role for regulatory intervention in the form of a bailout increase the liquidity in the system. And yet the presence of the
of banks threatened by liquidation. In particular, if the regulator regulator reduces the ex-ante liquidity of banks. This exposes a
observes a threatened liquidation that may trigger a contagion of fundamental tension between the macro-prudential goal of elim-
liquidation on the other banks, it steps in and buys out the cred- inating large-scale failures via contagion and the micro-prudential
itors of threatened banks. The purpose of the following discussion goal of ensuring market discipline as a complement to regulatory
is to examine the consequences of such intervention. We begin by oversight. In effect, the “price” of having a system flush with ex-
assuming that the regulator operates under a serious informational post liquidity provided by the regulator is lower ex-ante liquid-
constraint – it does not observe the x of any bank or the realiza- ity. Thus, our analysis supports Bagehot’s (1873) assertion that the
tion of ξ . Hence, it cannot distinguish between liquidations based LOLR should not engage in unconditional bailouts, which include
on the realization of x observed by the bank’s own creditors and those that occur even when the payment system is not threatened.
those based on the observed liquidations of other banks. In this Capital requirement with unconditional bailouts: Interestingly,
case, if N2 is large enough (and as assumed greater than the cost in the case of unconditional bailouts, the optimal capital require-
of bailing out both banks C2 ), the regulator will find it ex-post op- ment is to require that banks have no leverage whatsoever. To see
timal to pursue unconditional bailouts of all banks threatened with this, note that the regulatory objective (29) in this case takes the
liquidations. form ViII (Fi , Fj ) + V jII (Fi , Fj ) − [1 − δ0 ]C2 .
The problem with a more limited regulatory resolution of As we discussed above, V II does not depend on the leverage of
threatened banks—one in which only some banks are bailed out— either bank. However, the likelihood of joint failures is increasing
is that unless the regulator backstops creditors, information about in the leverage of each bank. It follows then that the optimal cap-
there being potentially an adverse asset-value shock is likely to ital requirement is to set F to zero for both banks, i.e., the optimal
leak out. In this case, the bailout by the regulator serves no pur- capital requirement is 100% equity financing of banks. Such a re-
pose at all since all other banks’ creditors will infer that a liqui- quirement has recently been proposed by Admati et al. (2013) as
dation was imminent, and contagion will set in.25 Thus, uncon- a possible way of regulating bank capital. Our result clarifies that
ditional bailouts of all banks become attractive for the regulator. the case for 100% equity financing of banks as a capital require-
Knowing this, creditors who can threaten liquidation at t = 1 will ment relies crucially on leverage playing no economically useful
demand that the regulator pay them F − x in full after they have role, which in turn, arises when the bailout policy is to bail out
collected the date-1 cash flow x. This means that creditors can banks in all states of the world so that all market discipline role
threaten liquidation at t = 1 in every state and receive F − x in ad- of debt is destroyed and incentive provision for bankers must oc-
dition to x. Anticipating this, they would have no reason to en- cur all through equity incentives. The capital requirement, how-
gage in privately-costly monitoring, and all the market discipline ever, cannot restore the loss of bank liquidity arising from the lack
of leverage is lost. Of course, the ex-ante pricing of leverage will of market discipline from debt.
reflect the fact that it is riskless, so creditors will not ex ante Case 3. The regulator with expanded information access:
get a “free lunch”. However, the entire burden of disciplining the Conditional liquidations
manager now falls on the equity-based compensation contract and We now consider the case in which the regulator can, at a cost
the amount of bank leverage is irrelevant. The value of the bank ψ > 0, obtain with probability 1 information about the systematic
then is: asset-value-impairment shock at t = 1. One can interpret this as
 xmax  x  the regulator setting up a new regulatory body like the Financial
V II = E (x ) + qH dx − W − [1 − ρ ]θ ∗ Stability Oversight Council (FSOC) and the Office for Financial Re-
xmax
 xmax  x 
o
 xmax  x   search (OFR), as required in the United States following the Dodd-
dx + qH dx (32) Frank Act, in order to gather information germane to systemic risk
o xmax o xmax and monitor risk across different parts of the economy. We con-
We now have the following result: tinue to assume that the regulator cannot observe the x of any
bank. The following result indicates that a regulatory policy of se-
Proposition 5 (Capital structure and liquidity creation with un- lective intervention that tolerates some bank failures can achieve
conditional regulatory bailouts). When the regulator unconditionally the goal of eliminating contagion without distorting ex-ante asset
bails out banks at t = 1 and prevents liquidations, the bank’s capi-
tal structure becomes irrelevant, and for κ > 0 sufficiently small, the
bank’s ex-ante liquidity is lower with the regulatory intervention than 26
Now, any distortion in favor of debt such as tax deductibility of interest rate
without. payments can lead the financial sector to lever up in a significant manner. Also,
as argued by Acharya and Yorulmazer (2007), Acharya, Mehran and Thakor (2016),
The intuition for capital structure irrelevance is straightforward. and Farhi and Tirole (2012), there would be an incentive to increase the systematic
If we remove the disciplining role of debt and make interim liq- risk of projects, if that were a choice variable in our model, and banks would pre-
fer to fund these correlated projects with leverage to “loot” the regulator/taxpayers
(assuming bank equity is not bailed out in case of joint failures).
25 27
With dispersed creditors, it may be difficult to prevent this information from Because equity provides some discipline, banks do not fail with probability 1.
leaking out. Hence, the probability of an unconditional regulatory bailout is positive, but not 1.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 17

and leverage decisions away from the equilibrium without LOLR The regulator thus avoids contagion by giving banks a put op-
intervention. Such a policy may represent an optimal policy for the tion on the value of their assets, but this option has value only if
regulator if N1 , the cost of a single bank failing is relatively low, it is exercised at t = 1. If bank creditors do not threaten liquida-
but N2 , the social cost of both banks failing is very high. In general, tion at t = 1 (because they do not receive a signal that asset val-
such a policy will be optimal when the regulator perceives a low ues have declined), but there has actually been a decline in asset
social cost associated with a few idiosyncratic bank failures, but a values, then their loss at t = 2 will not be covered by the regu-
high social cost associated with a meltdown of the whole system. lator. This suggests that it might pay for the creditors to threaten
This is indeed our working assumption in assuming that C2 < N2 . liquidation whenever x ∈ [xmin , λF ], even when they have not re-
ceived an asset-value impairment shock. If the regulator agrees to
Proposition 6 (Regulatory Bailout Policy that Preserves Market bail out the bank (because the systematic shock has, in fact, oc-
Discipline). Suppose the regulator pre-commits to an intervention curred), then the liquidation threat succeeds from the standpoint
policy such that of creditors. If not (because asset values are not truly impaired),
(i) only if the regulator has learned at t = 1 that bank asset val- then creditors can withdraw their liquidation threat. That is, there
ues are systematically impaired, then any bank whose creditors may be “frivolous” liquidation threats in order to “game” the reg-
threaten liquidation is bailed out by the regulator promising the ulator. To avoid this, the regulator can stipulate penalties on cred-
bank’s creditors in private negotiations that they will be paid in itors for liquidation threats that are not carried out in the event
full at t = 2 if they allow the bank to continue; and, the regulator refuses to bail out. Another way that frivolous threats
(ii) if the regulator has learned that there is no asset value impair- may be ruled out is if there are fixed costs to liquidation that are
ment, then any bank whose creditors threaten liquidation is al- incurred ex post by creditors.
lowed to be liquidated at t = 1. Avoiding contagion in this manner and simultaneously restoring
market discipline comes at an ex-post cost as in the benchmark
Under this intervention policy, and assuming that the bailout ne- model. Specifically, in order to avoid contagion but preserve mar-
gotiation in (i) is secret so that the creditors of no other banks learn ket discipline, the regulator prevents bailouts banks when the sys-
that particular bank was bailed out, the probability of contagion is tematic shock indicates asset-value impairment, and permits ineffi-
driven to zero, and no bank’s creditors liquidate the bank when they cient liquidations based on idiosyncratic (x) risk to proceed. There
observe another bank being liquidated. Moreover, each bank chooses may be time-inconsistency issues with the regulator allowing such
leverage inefficient liquidations (Acharya and Yorulmazer (2007), Farhi and
xmax Tirole (2012)), unless its intervention policy such as in Proposition
Fˆ ∗ = −1
(33) 6 is based on rules that have commitment. Hence, our result in
{[1 − λ
2 ][qH o − L] {[1 − ρ ]θ ∗ } − [qH o − 1]} Proposition 6 can be viewed as merely characterizing a minimum
where set of information requirements for the regulator to have a possi-
ble chance of restoring ex-ante efficiency.
H o ≡ β H + + [1 − β ]H − (34) Capital requirement with conditional bailout policy: In
and also chooses the good loan (unlike in Proposition 5). The mar- Proposition 6, we have characterized the bank’s private optimum
ket discipline of both debt and equity are preserved as in the case in response to the regulator’s bailout policy. We could also analyze
in which the regulator does not bail out any banks and there is no the regulator’s (socially-optimal) capital requirement. Note, how-
asset-value impairment shock. ever, that a bank’s leverage no longer has any contagion effects.
Hence, as explained in the proof of Proposition 6 in Appendix A,
This proposition says that the regulator can eliminate systemic each bank’s valuation is given by V III , which is the same as V in
risk by guaranteeing that any bank that is on the brink of being (10) with H replaced by H 0 . In particular, ViIII depends only on the
liquidated due to the systematic shock will be bailed out by the leverage of bank i and not on the leverage of bank j. Then, the ex-
regulator. This can either be a standard bailout or something akin ante social objective function is given by
to the Troubled Asset Relief Program (TARP), wherein the govern-  
ViIII (Fi ) + V jIII (Fj ) − δ j (1 − δ i ) + δ i (1 − δ j ) N1 − δ i δ j N2
ment steps in and agrees to purchase the (devalued) assets (or un-
dervalued equity) of banks at a price exceeding market value. Of −2βγ (1 − γ )C1 − βγ 2C2 − ψ (35)
course, implementing such a scheme requires that the regulator be
which captures the fact that (i) idiosyncratic bank failures do oc-
able to have access to information about the systematic shock, and
cur with this conditional bailout policy; (ii) joint failures may also
this may prove to be quite expensive in practice (high ψ ). It is also
arise due to idiosyncratic adverse shocks hitting both banks simul-
crucial that the regulator’s bailout is a secret negotiation with the
taneously; (iii) bank failures due to systematic asset-impairment
bank in question. This means that even though the systemic as-
shocks are avoided by the regulator incurring bailout costs; and, fi-
set value impairment shock may have been realized, the regulator
nally (iv) the regulator incurs costs of generating information about
will permit unassisted continuation of banks whose creditors do
the systematic asset shock.
not threaten liquidation.
The social optimum of bank leverage is thus lower than the
The other part of this intervention is that the regulator avoids
privately-optimal leverage level, Fˆ ∗ , because banks do not inter-
any bailout if the creditors’ liquidation threat does not coincide
nalize the social costs of their own failures, represented by Nn , and
with a systematic impairment in bank asset values. That is, any
the bailout costs, Cn . It can also be shown (proof available upon re-
liquidation due to an x realization for individual bank being in
quest) that as long as the joint-failures cost N2 is sufficiently large,
the liquidation range is allowed to proceed. Note, however, that
the socially optimal leverage in the case of conditional bailouts
common knowledge of such intervention policy means that any
exceeds that under the first case of no bailouts.28 Intuitively, by
liquidation that is publicly observed noiselessly reveals that asset
generating information on the systematic asset shocks and avoid-
values are not systematically impaired, so there is no contagion.
ing costly joint failures, the conditional bailout policy allows bank
Moreover, because such liquidations are allowed to occur, all the
leverage, and in turn, bank liquidity, to be higher.
market discipline of debt and equity that was present in the case
without the regulator and the systematic shock is resurrected. This
is why Fˆ ∗ in (33) has the same functional form as the F ∗ in (12); 28
Formally, for given choice of bank leverage (F’s), the social costs of failure under
the only difference is that H is replaced by H o. the case of no bailouts exceed those under the case of conditional bailouts.
18 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

In the context of the current regulatory landscape in the U.S., create less liquidity in markets in which regulators engage in more
the regulator here is a combination of the Financial Stability Over- unconditional bailouts.
sight Council (FSOC) and a “Resolution Authority” that can deter- In terms of future research, an interesting extension of our anal-
mine how to intervene in the event of bank failures. Indeed, an ysis will be to examine the role of the information generated by
interesting implication of the analysis is that there is an important stock prices or CDs spreads. While the majority of U.S. banks are
interaction between these two regulatory functions. The informa- not traded, the issue is germane for large, traded banks. With these
tion gathered by the FSOC can be useful to the Resolution Author- additional market signals, information about systematic shocks is
ity in permitting a larger number of bank liquidations via selec- likely to be more readily available to market participants than in
tive intervention, and by doing so it may prevent systemic crises our model. Our analysis implies, however, that a resolute com-
and allow ex-ante bank leverage and liquidity to be regulated at mitment by the regulator to bail out banks threatened by a sys-
higher. tematic shock can stanch the likelihood of contagion even in this
case.30
6. Conclusion
Acknowledgment
In this paper, we re-examined the role of bank leverage as an
instrument of liquidity creation. We exposed a fundamental ten- We gratefully acknowledge the helpful comments of Charlie
sion between the micro-prudential goal of encouraging the mar- Kahn (discussant) and participants at the Fourth Banco Portugal
ket discipline of banks via greater uninsured leverage in their cap- conference (July 2011), Paolo Fulghieri (discussant) and participants
ital structure and the macro-prudential goal of containing systemic at the AEA/AFE meetings in Denver (Jan. 2011), participants at the
risk. While higher bank leverage creates stronger creditor disci- Federal Reserve Bank of Chicago (March 2010), Florian Heider (dis-
pline at the individual bank level, it leads to greater systemic risk cussant) and participants at The Financial Intermediation Research
induced by contagious runs when creditors liquidate banks. To the Society Meeting in Dubrovnik (June 2013), and Doug Diamond (dis-
extent that such contagion threatens the entire financial system cussant), Charlie Calomiris, Allan Meltzer, Jeff Lacker and other par-
and hence jeopardizes the payment system, it invites ex-post regu- ticipants at the JFI-Hoover Institute Conferences at Stanford and in
latory intervention to bail out “failing” banks from being liquidated Washington D.C. (2015), we thank Alok Shashwat and Miguel Cas-
by creditors and thereby to maintain the interim liquidity and con- tro for research assistance.
tinuity of banks.
The consequences of this regulatory intervention depend on the Appendix A
regulator’s information and the nature of the intervention. When
the regulator has no information about whether systematic asset-
value impairment has occurred, it has two choices. It can precom- Proof of Lemma 1. Obvious from the discussion in the text. 
mit to never bail out any bank. But even in this case, the socially
Proof of Lemma 2. First, let us consider the intervention policy of
optimal bank capital is above the private optimum under plausi-
creditors. If the creditors intervene, they can force the bank man-
ble conditions and a regulatory capital requirement is needed. The
ager to invest in the good loan and the creditors’ date-2 payoff,
regulator’s other choice is to bail out banks. Due to the regulator’s
after having been paid x at date-1, is q[F − x] − κ , taking into ac-
information constraints, the intervention is unconditional, so the
count the investigation cost. If creditors do not intervene, the man-
disciplining role of bank leverage is lost, and ex-ante bank liquidity
ager will select the private-benefit loan, and the creditors’ date-2
actually declines. In fact, the more vigilant the regulator is in re-
payoff will be p[F − x]. Thus, intervention is optimal for creditors
acting to early warnings delivered via market signals by interven-
if:
ing with bailouts of failing banks, the worse is the problem. Absent
its disciplining role, all that bank debt offers is the information ex- q[F − x] − κ > p[F − x] or x < F − κ [
] −1 . (A-1)
ternality that leads to contagion. This “dark side” of leverage ne-
Now, we turn to the liquidation/continuation decision after hav-
cessitates arbitrarily high regulatory capital requirements—indeed,
ing invested κ . We compare (5) and (6). Since L < qH, there are
bank debt has no social value at all.29
three cases to consider: (i) F − x < Lx; (ii) Lx < F − x < Hx; and (iii)
We show that when the regulator generates its own informa-
F − x > Hx.
tion about the systematic shock—rather than relying on market
signals—and conditions bailouts on this information, contagion can Case. (i): F − x < Lx: In this region, x > F /[1 + L]. Thus, the liq-
be eliminated without sacrificing debt discipline. There is more ex uidation payoff to creditors is F − x. The continuation payoff is
ante liquidity, and the regulatory capital requirement is lower than q[F − x] < F − x. Hence, the creditors liquidate the bank.
even in the case in which the regulator credibly pre-commits to no
Case. (ii): Lx < F − x < Hx: In this case, Lx + x < F , which means
bailouts.
x < F /[1 + L]. Moreover, F < Hx + x, which means x > F /[1 + H ].
In addition to addressing policy-relevant issues, our analysis
So, F /[1 + H ] < x < F /[1 + L]. The liquidation payoff to creditors is
also generates a few empirical predictions. First, the probability
that a bank will be liquidated is higher not only when the bank
is more highly levered, but also when other banks holding sim- 30
In our analysis, systemic risk issues arise due to commonality in the asset hold-
ilar asset portfolios are more highly levered. Thus, there will be ings of banks. Such correlated portfolio choices have been observed in many crises.
an “industry-wide” leverage effect that will be felt by individual For example, in the 2007–09 subprime crisis, the commonality of asset portfolios
was in the form of real estate loans. A somewhat cynical political economy view is
banks. Second, banks will be more highly levered when manage- that bailouts of banks in the event of a systemic meltdown was a way for the gov-
rial private benefits are higher and systematic shocks to asset val- ernment to induce banks to correlate their asset choices in real estate lending so
ues are more visible to creditors but less likely. And third, banks that ex post bailouts can occur, thereby providing a subsidy for real estate lending,
something that serves political goals. We thank Charlie Calomiris for this insight.
We do not mean to suggest, however, that real-estate lending should not be encour-
29
All of this assumes that the bank’s uninsured creditors generate monitoring- aged by the government. There may be many economic benefits from this lending
relevant information that regulators do not. One reason for this may be that these that are not reflected in our analysis. And of course, real estate lending is not the
creditors have “skin in the game” and are actively trading in a variety of securities only example of correlated lending by banks. There are numerous other examples,
and markets, so they obtain market signals that may be unavailable to regulators such as lending to oil companies. These examples often occur when lending to a
who focus primarily on banks. particular sector has done well for a long time (see, for example, Thakor (2015)).
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 19

Lx, whereas the continuation payoff is q[F − x]. Creditors liquidate Proof of Proposition 2. The (unconditional) ex-ante probability of
whenever: Lx > q[F − x] or liquidation by bank i’s creditors (see (23)) is
x > λF , where λ ≡ q+
q    
L. P RBi ≡ 1 − δ0i = 1 − 1 − δ j 1 − δi β [ 1 − γ ]2 + 1 − β .
Creditors continue if x ≤ qF [q + L] −1 .
where we have explicitly distinguished between δ j and δ i . thus,
Case. (iii): F − x > Hx: In this region, x < F /[1 + H ]. The liquidation  
payoff of creditors is min(Lx, F − x ) = Lx. Since x + xH < F implies ∂ PRBi /∂ Fi = − 1 − δ j [−1]
that x + xL < F . The continuation payoff of creditors is qHx. Since >0
qH > L, the creditors continue.
Moreover, ∂ δ j /∂ Fj = [1 − λ]/xmax > 0. Thus, it follows that
Finally, since qH > L, all liquidation is inefficient. 
∂ PBRi /∂ Fi > 0. Now ∂ PRBi /∂ δ j = −[1 − δ i ][−1] > 0.
Proof of Proposition 1. Using (7), (8) and (9), we can write (10) Since ∂ δ j /∂ Fj = [1 − λ]/xmax > 0, it follows that ∂ P RBi /∂ Fj >
as: 0. 
 F  x   F  Lx  Proof of Lemma 3. Differentiating B˜(Fj ) in (26).
V (F ) = E (x ) + qHE (x ) − qH dx + dx  
λF xmax λF xmax
 ∂ δ2 + ∂ δ1
F ∂ B˜(Fj )/∂ δ j = −
λ2 [ H − L] − −
[L − H ]
−θ ∗ [(x − F ) + qHx]g(x )dx − W ∂δ j ∂δ j
x  
 −β [1 − γ ] [H + − L]
2
= −
λ
F 2
+ ρθ ∗ [(x − F ) + qHx]g(x )dx +[1 − β ][1 − γ ][L − H − ]
x
 F  x   F  Lx  = −
λ2 [H + − L][1 − γ ][−β [1 − γ ] + 1 − β ]
= [1 + qH ]E (x ) − qH dx + dx
 λF

xmax
 λF xmax where the last step follows from H + − L = L − H − < 0 since β ≤
xmax
[x − F ] 0.5. 
− W − [1 − ρ ]θ ∗ dx
F xmax Proof of Proposition 3. Using (7), (8) and (9), we can write (23)
 xmax  x   as:
+ qH
xmax
dx    λF  x 
F
  V (F ) = E (x ) + δ1 H − + δ2 H + q dx
xmax
[qH − L] 1 − λ 2
F2  F 
o
= [1 + qH ]E (x ) − −W Lx
 λF  x 

2xmax
 + dx + [1 − δ1 − δ2 ]L dx
  λF xmax 0 xmax
[1 + qH ] xmax
2
− F2 F [xmax − F ]  xmax  
− [1 − ρ ]θ ∗ − . (A-2) +qH
x
dx − W
2xmax xmax xmax
F
 x    xmax   
The optimal F, say F ∗ , must satisfy the first-order condition:
max
x−F x
− [ 1 − ρ ]θ ∗
dx + qH dx .
xmax xmax
  F F
−[qH − L]F ∗ 1 − λ2 + [1 − ρ ]θ ∗ {[1 + qH ]F ∗ + xmax − 2F ∗ } = 0. (A-7)
(A-3) Simplifying and rearranging terms, we can write (A-7) as:
 
Rearranging (A-3) yields (12). V (F ) = −F 2 /2xmax {qH − [1 − ρ ]θ ∗ [qH − 1]
Next, we verify the second-order condition for a unique global
maximum: −[L + λ2 {δ2 [H + − L] − δ1 [L − H − ]}]}
  +[1 − ρ ]θ ∗ F + constant, (A-8)
−[qH − L] 1 − λ2 + [1 − ρ ]θ ∗ [qH − 1] < 0, (A-4)
where the constant includes all the terms that do not contain F.
that is satisfied under our maintained assumptions. The first-order condition for the optimal best response of bank i,
Now let us see what conditions are needed for F ∗ ∈ (0, xmax ). Fi0 (Fj0 ), is (suppressing the argument Fj0 ):
For F ∗ < xmax , we need:
  −Fi0 {qH − [1 − ρ ]θ ∗ [qH − 1]
1 − λ2 [qH − L]
− [qH − 1] > 1. (A-5) −[L + λ2 {δ2j [H + − L] − δ1j [L − H − ]}]}
[1 − ρ ]θ ∗
+[1 − ρ ]θ ∗ xmax = 0. (A-9)
Substituting for θ ∗ from (4) and rearranging yields that we re-
quire that Rearranging (A-9) and simplifying yields (27).
  The second-order condition for a maximum is:

1 − λ2 [qH − L]
B< ≡ B. (A-6) −{qH −[1 − ρ ]θ [qH −1] −[L + λ2 {δ2j [H + − L] − δ1j [L − H − ]}]} < 0.
q [1 − ρ ]
(A-10)
[1−λ2 ][qH−L]
To ensure that F ∗ > 0, we need [1−ρ ]θ ∗
− [qH − 1] > 0,
To ensure Fi0 < xmax , (27) indicates that we need
which is satisfied for B < B (because then (A-6) holds). Moreover,
it is also clear that since (A-6) holds, the inequality in (A-5) also  j +   
qH > [1 − ρ ]θ ∗ qH + L + λ2 δ2 H − L − δ1j L − H − . (A-11)
holds. 
Using (4) to substitute for θ∗ and rearranging gives us:
Proof of Corollary 1. Note that (12) shows that ∂ F ∗ /∂ θ ∗ > 0.  
Moreover, from (4) we know that ∂ θ ∗ /∂ B > 0. Thus, ∂ F ∗ /∂ B > 0.
qH − L − λ {δ2 [H − L] − δ1 [L − H − ]}
2 +
B< (A-12)
From (12) it also follows immediately that ∂ F ∗ /∂ ρ < 0.  [1 − ρ ]q
20 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21

The right-hand side (RHS) of (A-12) is Bˆ if we set δ1 = δ̄1 and

δ2 = δ̄2 . Moreover, Bˆ ≤ RHS of (A-12). Thus, setting B < Bˆ ensures
that (A-12) will hold.
Note that as long as B < Bˆ, the second-order condition (A-10)
will also be satisfied. Moreover, satisfaction of (A-10) also guaran-
tees that Fi0 > 0. Thus, Fi0 ∈ (0, xmax ).
To obtain (28), note that with symmetry across banks i and j,
both banks have exactly the same best response conditional on
a given leverage choice by the other bank. Substituting Fj0 (Fi0 ) =
Fi0 (Fj0 ) in (27), we can write

1
F0 = , (A-13)
a + bF 0
where a and b are defined in Proposition 3. Solving (A-13) as a
quadratic equation yields (28). Note that we need b > 0, which is
true if β < [2 − γ ]−1 . This is guaranteed by β ≤ 0.5. 

Proof of Corollary 2. Write Fi0 (Fj0 ) as

xmax
Fi0 Fj0 =   . (A-14)
{ }] − [qH − 1]
qH−[L+λ2 δ2j [H + −L]−δ1j [L−H − ]
[1−ρ ]θ ∗

It is clear that, holding Fj0 fixed, ∂ Fi0 (Fj0 )/∂ θ ∗ > 0. Since
∂θ ∗ /∂ B> 0, it follows that ∂ Fi0 (Fj0 )/∂ B > 0. That is, each bank’s
best response is increasing in its own manager’s B. In a symmetric
Nash equilibrium therefore, ∂ F 0 /∂ θ ∗ > 0. To establish the sign of
∂ F o/∂ β , note that the sign of ∂ Fi0 (Fj0 )/∂ β depends solely on the
sign of ∂ A/∂ β where A ≡ δ2 [H + − L] − δ1 [L − H − ]. Now, ∂ A/∂ β =
−[H + − L][γ + {1 − γ }{1 − δ}] − [1 − δ ][1 − γ ]2 [L − H − ] < 0.
Since ∂ Fi0 (Fj0 )/∂ A > 0, we have ∂ Fi0 (Fj0 )/∂ β < 0. In a symmetric
Nash equilibrium therefore, ∂ F 0 /∂β < 0. Moreover, it can also be
verified that ∂ Fi0 (Fj0 )/∂ρ < 0 and ∂ F 0 /∂ρ < 0. 

Proof of Proposition 4. The 

Fi that maximizes Vi + V j is the one
that maximizes:

   
− Fi2 D Fj /2xmax + [1 − ρ ]θi∗ Fi − 1 − δ j 1 − δ i N2 + constanti
    
− Fj2 D(Fi )/2xmax + [1 − ρ ]θ j∗ Fj − 1 − δ j
1 − δ N2 + constant j ,
i

(A-15)
where

D Fj ≡ qH − [1 − ρ ]θi∗ [qH − 1]
 
 + 
 
− L + λ2
δ2 Fj − H − L − δ1 Fj L − H − ,
D(Fi ) ≡ qH − [1 − ρ ]θ j∗ [qH − 1]
     
− L + λ2 δ2 (Fi ) H + − L − δ1 (Fi ) L − H − .

The first-order condition for the regulator’s optimal choice of 

Fi ,
given Fj , is:

   2  
− 
Fi D Fj /xmax + [1 − ρ ]θi∗ + Fj2 /2xmax λ ∂ Ai /∂ 
Fi

−J Fj N2 = 0 (A-16)
where

 
 
Ai ≡ δ2 
Fi H + − L − δ1 
Fi L − H − (A-17)

     
J Fj ≡ − 1 − δ j [1 − λ]/xmax β 1 − γ 2 + [1 − β ] (A-18)
Rearranging (A-16) yields:
  2  

[1 − ρ ]θ ∗ xmax + Fj2 /2 λ ∂ Ai /∂ 
Fi − J Fj N2

Fi =
. (A-19)
D Fj

Comparing (A-20) with (27), we see that ∂ Ai /∂ Fi < 0 is suffi-

cient for Fi < Fi0 . Now, using H + − L = L − H − , we can write Ai as: Fig. 3. The Difference Between the Privately Optimal Leverage and the Socially Op-
timal Leverage.
 V.V. Acharya, A.V. Thakor / J. Finan. Intermediation 28 (2016) 4–21 21

 
Ai = H + − L [δ2 (Fi ) − δ1 (Fi )] Adrian, Tobias, Shin, Hyun, 2010. Liquidity and leverage. J. Financ. Intermed. 19 (3),
418–437.
= [1 − δi ][1 − γ ][1 − β − β{1 − γ }] + [1 − β ]γ . Admati, Anat, DeMarzo, Peter, Hellwig, Martin, Pfleiderer, Paul, 2013. Fallacies, Ir-
relevant Facts, and Myths in the Discussion of Capital Regulation: Why Bank
Thus, Equity is Not Expensive. Stanford Graduate School of Business Working Paper.
Allen, Franklin, Babus, Ana, Carletti, Elena, 2012. Asset commonality, debt maturity,
∂ δi
∂ Ai /∂ Fi = −[1 − γ ][1 − β − β{1 − γ }] and systemic risk. J. Financ. Econ. 104 (3), 519–534.
∂ Fi Bagehot, Walter, 1873. Lombard Street. Richard D. Irwin, Homewood, Ill Reprint:
1962.
[1 − λ] Bhattacharya, Sudipto, Boot, Arnoud, Thakor, Anjan V., 1998. The economics of bank
= −[1 − γ ][1 − β − β{1 − γ }] regulation. J. Money, Credit Bank. 30 (4), 745–770.
xmax
Boot, Arnoud, Gopalan, Radhakrishnan, Thakor, Anjan V., 2006. The entrepreneur’s
< 0 since 1 − β > β{1 − γ } due to β ≤ 0.5. choice between private and public ownership. J. Finance 63 (4), 803–836.
Boot, Arnoud, Gopalan, Radhakrishnan, Thakor, Anjan V., 2008. Market liquidity, in-
Thus, 
Fi < Fi0 regardless of N2 if β ≤ 0.5. However, even if N2 is vestor participation and managerial autonomy: why do firms go private? J. Fi-
sufficiently large, then clearly Fˆi < Fi0 regardless of β . nance 63 (4), 2013–2059.
Boot, Arnoud, Thakor, Anjan V., 2011. Managerial autonomy, allocation of control
To verify that  F is a unique global maximum, let us examine
i rights, and optimal capital structure. Rev. Financ. Stud. 24 (10), 3434–3485.
the second-order condition: Brunnermeier, Markus, Sannikov, Yuliy, 2016. The I Theory of Money. Princeton Uni-
versity Working Paper January.
−D(Fj )/xmax , which is clearly negative. 
Buser, Stephen, Chen, Andrew, Kane, Edward, 1981. Federal deposit insurance, regu-
latory policy, and optimal bank capital. J. Finance 36, 51–60.
Proof of Proposition 5. The proof of capital structure irrelevance Caballero, Ricardo, Simsek, Alp, 2013. Fire sales in a model of complexity. J. Finance
is obvious (note that (32) is independent of F). Comparing the ob- 68 (6), 2549–2587.
jective functions in (23) and (32), we see that if we ignore κ , then Calomiris, Charles, Flandreau, Marc, Laeven, Luc, 2016. Political foundations of the
lender of last resort: a global historical narrative. J. Financ. Intermed. 28, 48–65.
(23) can be made equivalent to (32) by setting F = 0. But the F that
Calomiris, Charles, Kahn, Charles, 1991. The role of demandable debt in structuring
maximizes ex-ante liquidity in (23) is strictly positive. Hence, for κ optimal banking arrangements. Am. Econ. Rev. 81 (3), 497–513.
> 0 small enough, the optimized value of V (F ) in (23) exceeds the Chari, Varadarajan V., Jagannathan, Ravi, 1988. Banking panics, information, and ra-
tional expectations equilibrium. J. Finance 43 (3), 749–761.
VREG in (32). 
Dewatripont, Mathias, Tirole, Jean, 1994. The Prudential Regulation of Banks. MIT
Press.
Proof of Proposition 6. Since the regulator credibly precommits
Diamond, Douglas W., Rajan, Raghuram G., 2005. Liquidity shortages and banking
to bail out any bank that has suffered a systematic asset-value im- crises. J. Finance 60 (2), 615–647.
pairment shock, no bank can fail due to this shock. Moreover, since Diamond, Douglas W, Rajan, Raghuram G., 2001. Liquidity risk, liquidity creation,
and financial fragility: a theory of banking. J. Polit. Econ. 109 (2), 287–327.
the bailout is secret, no other bank’s creditors learn that the asset-
Diamond, Douglas W., Rajan, Raghuram G., 2012. Illiquid banks, financial stability
value impairment shock has occurred if they did not receive the and interest rate policy. J. Polit. Econ. 120 (3), 552–591.
signal. Finally, since no bank is liquidated due to the asset-value- Farhi, Emmanuel, Tirole, Jean, 2012. Collective moral hazard, maturity mismatch and
impairment shock and no liquidations based on x are prevented, systemic bailouts. Am. Econ. Rev. 102 (1), 60–93.
Garmaise, Mark J., Moskowitz, Tobias J., 2006. Bank mergers and crime: the real and
the bank’s leverage-choice problem is the same as in (10), except social effects of credit market competition. J. Finance 61 (2), 495–538.
that H is replaced by H o. Hence, (33) is the same as (12) except Goodhart, Charles A.E., 1975. Monetary relationships: a view from threadneedle
that H is replaced by H o.  street. Papers in Monetary Economics. Reserve Bank of Australia.
Goel, Anand, Song, Fenghua, Thakor, Anjan V., 2014. Correlated leverage and its ram-
ifications. J. Financ. Intermed. 23 (4), 471–503.
Appendix B. Numerical example Hart, Oliver, Moore, John, 1995. Debt and seniority: an analysis of the role of hard
claims in constraining management. Am. Econ. Rev. 83 (3), 567–589.
Hart, Oliver, Zingales, Luigi, 2009. A New Capital Regulation for Large, Financial In-
The comparative statics on inefficiency of privately op- stitutions. University of Chicago working paper.
timal leverage are difficult to prove analytically but illus- Huang, Rocco, Ratnovski, Lev, 2011. The dark side of bank wholesale funding. J. Fi-
trated graphically in Fig. 3 for a numerical example (and nanc. Intermed. 20 (2), 248–263.
Jensen, Michael, 1986. Agency costs of free cash flow, corporate finance, and
found to be robust across a range of parameter values we
takeovers. American Economic Review 76 (2), 323–329.
have tried). In the numerical example, we choose as base- Kane, Edward, 2010. Redefining and controlling systemic risk. Atlantic Econ. J. 38,
line parameters q = 3/4, p = 1/2, B = 0.4, ρ = 0.8, γ = 0.6, xmax = 251–264.
4, H = 7/4, L = 1, H + = 1.5, H − = 0.5, β = 1/8, N2 = 0. This is one Lucas, Robert, 1976. Econometric policy evaluation: a critique. In: Brunner, K.,
Meltzer, A. (Eds.), The Phillips Curve and Labor Markets. Carnegie-Rochester
parameter constellation that ensures that various restrictions as- Conference Series on Public Policy, 1, New York: American Elsevier, pp. 19–46.
sumed throughout the model are all jointly satisfied, except those Mehran, Hamid, Thakor, Anjan V., 2011. Bank capital and value in the cross-section.
involving Nn (because β is assumed low enough for the social op- Rev. Financ. Stud. 24 (4), 1019–1067.
Merton, Robert C., 1977. An analytic derivation of the cost of deposit insurance and
timum to be below the privately optimal leverage anyway). loan guarantees: an application of modern option pricing theory. J. Bank. Fi-
nance 1, 3–11.
References Mian, Atif, Amir, Sufi, 2011. House prices, home-equity based borrowing, and the U.
S. household leverage crisis. Am. Econ. Rev. 101 (5), 2132–2156.
Acharya, Viral V., 2009. A theory of systemic risk and design of prudential bank Ramakrishnan, R.T.S., Thakor, Anjan V., 1984. Information reliability and a theory of
regulation. J. Financ. Stab. 5 (3), 224–255. financial intermediation”. Rev. Econ. Stud. 51 (3), 415–432.
Acharya, Viral V., Gujral, Arvind, Kulkarni, Nirupama, Shin, Hyun S., 2013. Dividends Shleifer, Andrei, Vishny, Robert, 2010. Unstable banking. J. Financ. Econ. 97, 306–318.
and bank capital in the financial crisis of 2007-09, NBER Working Paper No. Thakor, Anjan V., 2012. Incentives to innovate and financial crises. J. Financ. Econ.
16896. 103 (1), 130–148.
Acharya, Viral V., Mehran, Hamid, Thakor, Anjan V., 2016. Caught between scylla Thakor, Anjan V., 2015. Lending booms, smart bankers, and financial crises. Am.
and charybdis? Regulating bank leverage when there is rent seeking and risk Econ. Rev. 105, 305–309.
Shifting. Rev. Corporate Financ. Stud. 5 (1), 36–75. Thakor, Anjan V., 2016. The highs and the lows: a theory of credit risk assessment
Acharya, Viral V., Schnabl, P., Suarez, G., 2013. Securitization without risk transfer. J. and pricing through the business cycle. J. Financ. Intermed. 25 (1), 1–29.
Financ. Econ. 107, 515–536. Van den Steen, Eric, 2010. Interpersonal authority in a theory of the firm. Am. Econ.
Acharya, Viral V., Viswanathan, S., 2011. Leverage, moral hazard and liquidity. J. Fi- Rev. 100 (1), 466–490.
nance 66 (1), 99–138. Wagner, Wolf, 2010. Diversification at financial intermediaries and systemic crises.
Acharya, Viral V., Yorulmazer, Tanju, 2007. Too many to fail – an analysis of time-in- J. Financ. Intermed. 19 (3), 373–386.
consistency in bank closure policies. J. Financ. Intermed. 16 (1), 1–31. Winton, A., 1997. Competition among financial intermediaries when diversification
Acharya, Viral V., Yorulmazer, Tanju, 2008. Information contagion and bank herding. matters. J. Financ. Intermed. 6 (4), 99–129.
J. Money, Credit Bank. 40 (1), 215–231.