Joint liquidity and capital regulation with a Lender of Last

∗
Resort: The risk-shifting channel
Demian Macedo and Sergio Vicente†
Universidad Carlos III de Madrid

January 19, 2023

Abstract
We analyze the problem of a regulator that sets both capital and liquidity requirements
to maximize social welfare in a framework in which a bank decides its level of solvency risk
facing a risk-return trade-off. Capital requirements reduce risk-shifting through a ”skin-in-the-
game” channel, but substituting deposits for capital is socially costly. Liquidity requirements
mitigate short-term withdrawal risk, but aggravate risk-shifting because they reduce the bank’s
returns. We find that liquidity and capital requirements complement each other when the cost
of capital or the return on loans is high and offset each other otherwise, so that regulators
should set liquidity and capital requirements jointly taking into consideration capital and
liquidity feedback loops.
The reason why liquidity and capital must be set jointly is because the effectiveness of
capital depends on the amount of liquidity and viceversa.
Keywords: Liquidity, Capital, Basel III, Solvency risk, Liquidity risk
JEL Codes: G01, G20, G21, G28, E58
∗
We are greatful to Elena Carletti, Javier Suárez and David Martı́nez-Miera for very detailed comments on
earlier drafts. We are also thankful to Max Bruche, Xavier Freixas, Arie Gozluklu (discussant), Frédéric Malherbe,
Robert Marquez, Rafael Repullo (discussant), Pablo Ruiz-Verdú, Anatoli Segura and Xavier Vives, as well as seminar
audiences at the I MadBar Workshop on Banking and Corporate Finance, the CEPR European Summer Symposium
in Financial Markets (Gerzensee, evening sessions), the Annual Meetings of the Spanish Finance Association, the
Simposio of the Spanish Economic Association, Universitat Pompeu Fabra and Universidad Carlos III de Madrid. We
have also benefited from Matias Cabrera’s input at the inception of the paper. Part of this research was conducted
while Sergio Vicente was visiting Universitat Pompeu Fabra. He is greatful for their hospitality. He also acknowledges
support from the Spanish Ministry of Education and Culture, project ECO2013-42849-P, and from Fundación Ramón
Areces ”Industrial Organization and Regulation” Grant program.
†
dmacedo@emp.uc3m.es, sergio.vicente@uc3m.es.

1
 ”By way of motivation, note that before the financial crisis, we had a highly developed regime of
capital regulation for banks–albeit one that looks inadequate in retrospect–but we did not have formal
regulatory standards for their liquidity.”
Governor Jeremy Stein, Speech at the Federal Reserve Bank of Richmond, Charlotte, North
Carolina, April 19, 2013.

”In the first instance, at least, this was a liquidity crisis. Its fast-moving dynamic was very
different from that of the savings and loan crisis or the Latin American debt crisis of the 1980s.
The phenomenon of runs instead recalled a more distant banking crisis–that of the 1930s. Despite
this defining characteristic of the last crisis, measures to regulate liquidity have by-and-large lagged
other regulatory reforms [...]”.
Governor Daniel K. Tarullo, Speech at the Clearing House 2014 Annual Conference, New York,
New York, November 20, 2014

1 Introduction
Before the last financial crisis, most of the regulatory efforts were concentrated on capital require-
ments. Although concerns regarding illiquidity risk have been pervasive in the regulatory debate
dating back to at least ? celebrated essay, the emphasis on capital had relegated liquidity regu-
lation to the background. However, during the 2007 early ’liquidity phase’ of the financial crisis,
several banking institutions experienced liquidity-driven difficulties despite their adequate capital
levels (?, ?, ?). By September, Northern Rock suffered a bank run after seeking (and obtaining)
liquidity support from the Bank of England. As response to the 2007-2009 financial crisis, the Basel
Committee for Banking Supervision introduced the Liquidity Coverage (LCR) and the Net Stable
Funding (NSFR) ratios within Basel III to reinforce the resilience of banks to illiquidity risk.1
In general, capital and liquidity regulation have been analyzed in isolation. Classical treatments
of bank capital regulation have emphasized the role of capital as a means of mitigating excessive
risk-taking by banks (e.g., ?, ?, ?, ?). On the other hand, liquidity regulation has typically been
thought of as a way of dealing with issues of maturity mismatch or refinancing risk (e.g., ?, ?, ?, ?).
And it has not been until recently that the regulation of capital and liquidity has been addressed
1
The LCR requires that banks hold a buffer of liquid assets so as to cover cash outflows during a 30-day window
under liquidity distress. The NFSR, somehow redundantly, complements the former by requiring a long-term stable
funding structure that protect banks against maturity mismatches. The British FSA issued the ”Turner Review” in
2009, which included a set of guidances to regulate banks’ liquidity in a similar spirit to the Basel Liquidity Coverage
Ratio (see ?).

2
together (e.g., ?, ?, ?).2
In this paper we analyze joint liquidity and capital regulation in a bank risk-taking framework.
Our model features a bank that is partially funded with deposits and protected by limited liability.
Moreover, there is a deposit insurance scheme that insulates depositors against potential bank
failures.3 A consequence of these three features is that the bank does not internalize the losses
that its potential failure inflicts on the deposit insurance scheme, leading to excessive risk-shifting.
This market failure can be partially mitigated by a social welfare maximizer banking regulator, who
can employ two regulatory tools: a capital and a liquidity requirement. The bank takes its capital
structure as given (through capital requirements) and chooses its (unobservable) insolvency risk
profile facing a standard risk-return trade-off.4 Capital requirements increase the bank’s skin in
the game by reducing its deposit liabilities, therefore mitigating the bank’s incentives to risk-shift.
Additionally, the bank is subject to an uncertain (exogenous) level of early deposit withdrawals,
which generates illiquidity risk.5 Liquidity requirements reduce the bank’s exposure to liquidity
crises. We show that the effectiveness of capital (respectively, liquidity) requirements depends on
the bank’s level of liquidity (respectively, capital). Hence, optimal liquidity and capital requirements
need be determined jointly.
We assume that substituting deposits for equity entails a social cost, so that capital requirements
weigh the marginal return of capital–through its impact on reducing insolvency risk–and its social
cost.6 The effect of liquidity on capital requirements is that the share of liquid assets held by
2
Although their focus is on the provision of liquidity aid by a Lender of Last Resort, ? address the convenience
of establishing a liquidity and a capital ratio in combination with LoLR interventions.
3
Since deposits are insured, the rates of return on deposits do not reflect the riskiness of the bank’s portfolio.
4
In our framework, capital is used to mitigate the bank’s risk-shifting incentives. If the bank’s risk profile were
verifiable, the first-best policy intervention would consist of a zero capital requirement.
5
We want to emphasize that we are taking a rather extreme vision of solvency and liquidity risks as independently
generated pheonomena. Arguably, one may expect that the risk of facing a large amount of early withdrawals be
positively correlated with the risk profile of the bank, at least if withdrawing at an early stage may be influenced
to a certain extent by the bank’s fundamentals and not only by depositors’ idiosyncratic motives. For instance,
one may reasonably expect that liquidity shocks may be originated by signals of solvency problems. Moreover,
a situation of liquidity distress may well lead to solvency troubles. However, we want to abstract from potential
feedback loops between the generation of liquidity and solvency risk in the basic model to highlight that capital
and liquidity requirements feed back into each other even if liquidity and solvency risks are generated indepedently.
Nevertheless, in an extension we conduct an exploration of a model in which early withdrawals are relatively more
likely the higher the level of solvency risk. In this setup, reducing solvency risk has a double effect and, consequently,
capital requirements are larger than in the base model. However, the nature of the feedback effects of capital and
liquidity remains.
6
Previous research has emphasized that substituting deposits for equity is socially costly. For instance, ? and
? state that capital requirements are socially costly because substituting capital for deposits entails eliminating a

3
the bank affects the marginal return of capital. For instance, if the bank’s liquidity is low, the
likelihood that the bank survives a liquidity crisis is low. Hence, the expected return of reducing
insolvency risk is low as well, because the bank may nevertheless fail due to liquidity driven troubles.
Consequently, the marginal return of capital is low when liquidity is low. As liquidity increases, so
does the probability of surviving a liquidity crisis and therefore the return of reducing insolvency
risk. Hence, liquidity requirements are a complementary tool to capital regulation when liquidity
requirements are small, for the marginal return of capital increases with liquidity when the liquidity
level is low. However, liquidity reserves reduce the social loss in the event of a bank failure, because
liquidity reserves constitute a cash-in-hand asset that can be used to reduce the deposit insurance
toll in that event. When liquidity levels are sufficiently high, the positive effect of reducing illiquidity
risk is outweighed by the reduction of the social loss in the event of a bank failure. Hence, liquidity
requirements constitute an offsetting tool to capital requirement when these are high.
On the other hand, the rationale for liquidity regulation in this paper is to reduce illiquidity risk
beyond the bank’s will. In a laissez-faire economy, the bank would choose a certain level of liquidity
to balance out the positive effect of reducing illiquidity risk with the opportunity cost of liquidity,
which is given by the foregone investments in more profitable assets. The regulator faces the same
liquidity trade-off as the bank in nature, but liquidity is more valuable for the regulator than for
the bank for two reasons. First, a bank failure is more costly for society than for the bank, because
the bank does not internalize the loss that its failure inflicts on the deposit insurance fund. Second,
liquid assets are valuable in the event of the bank failure. This is a source of value of liquidity that
the bank neglects, because the bank does not take into account the value of its assets in the event
of a failure. Hence, the opportunity cost of liquidity in terms of foregone investment opportunities
is smaller for society than for the bank. As a result, the regulator establishes liquidity requirements
in excess of the level that the bank would choose in the absence of regulation. An immediate
implication of binding liquidity requirements is that they harm the bank’s expected profits. Hence,
liquidity requirements harm the bank’s incentives to reduce insolvency risk.
The effectiveness of liquidity requirements does in turn depend on the level of bank’s equity
capital. On the one hand, increasing capital reduces the negative effect of a binding liquidity
requirement on the bank’s choice of insolvency risk. The reason is that the wedge between the
bank’s and society’s objective function narrows down as capital increase–recall that this wedge is
valuable source of liquidity for consumers. ? argue that issuing new equity could be costly when old shareholders and
managers have access to information that new investors do not have. ? points out that the presence of asymmetric
information makes equity capital more costly than other sources of bank
funding. Capital may also be costlier than deposits due to its relative scarcity (?) or because it carries a higher
return than deposits in segmented markets (?).

4
driven by the fact that the bank is partially funded with non-risk-priced deposits and is insulated
by limited liability. Hence, by reducing the negative effect of liquidity on the bank’s incentives to
curtail insolvency risk, raising capital leads to an increase of the marginal social return of liquidity
requirements. On the other hand, an increase in capital leads to a reduction of insolvency risk.
Hence, as capital increases, the value of the bank’s upside payoff carries a higher weight in the
regulator’s objective function, because the solvent state is more likely to occur. The higher likelihood
that the bank’s loan investments turn out successfully imply that the value of surviving a liquidity
crisis is higher, but also that the opportunity cost of holding liquidity instead of granting loans
increases as well. For a sufficiently high liquidity level, so that the bank is resilient enough to
liquidity shocks, the effect of the increased opportunity cost of holding liquidity dominates. Hence,
capital requirements are a complementary tool to liquidity requirements when they are low, while
they constitute an offsetting tool to liquidity requirements when they are high.
Summing up, we have two regions for liquidity and capital. In the ’complementary tools’ region,
increasing the level of one of the regulatory variables leads to an increase of the effectiveness of the
other. The optimal capital and liquidity requirements fall in that region when the cost of capital
and the opportunity cost of liquidity are large, so that the level of optimal capital and liquidity are
small. In the ’offsetting tools’ region, on the contrary, increasing the level of one regulatory tool
offsets the effectiveness of the other tool. The optimal capital and liquidity requirement fall in that
region when the cost of capital and the opportunity cost of liquidity are small, so that the optimal
requirements are large.
There are several important ingredients that we abstract from in the main analysis in order to
keep the analysis tractable. We extend the analysis to allow for asset liquidation at (exogenous) fire
sales prices in order to overcome a liquidity shock exceeding the bank’s cash reserves. Although the
main feedback effects between liquidity and capital prevail, namely that capital and liquidity are
complementary or offsetting tools depending on the level of both variables, there are some changes
that are worth noting. First, the magnitude of optimal liquidity requirements is unambiguously
smaller than in the absence of asset liquidation. The reason is that the bank is now able to overcome
larger liquidity shocks and therefore the marginal return of liquidity requirements through reducing
illiquidity risk diminishes. It is worth remarking that, although liquidity requirements are reduced,
illiquidity risk unambiguously reduces–the combined effect of lower liquidity and asset liquidation
allows the bank to overcome higher liquidity shocks. On the other hand, capital requirements are
unambiguously larger. Since illiquidity risk is smaller, the value of reducing insolvency risk–and
therefore, of capital requirements–is larger.
Finally, we consider a modified version of our model in which a Lender of Last Resort (LoLR)

5
comes to the rescue of the bank in case it faces a shortage of liquid funds to meet early withdrawals.
We model the LoLR as an agent that is committed to provide liquidity assistance until a certain
(exogenous) level at a given rate of return.7 The introduction of an LoLR reduces–or, if the
LoLR is unconstrained, eliminates–illiquidity risk, leading an unambiguous reduction of liquidity
requirements as compared to our benchmark. The reduction of illiquidity risk makes the reduction
of insolvency risk more valuable. Hence, capital requirements should be higher under an LoLR.
How effective capital requirements are do also depend on the rate at which the bank borrows from
the LoLR, for this rate negatively affects the bank’s payoff and therefore its incentives to reduce
insolvency risk. Hence, since the magnitude of the LoLR’s subsidy to the bank constitutes a mere
transfer among agents in the economy that does not affect social welfare, the bank should borrow
from the LoLR at the minimum possible rate.
The rest of the paper is organized as follows. In the next section we review related research.
Section 3 contains the description of the basic model. We analyze the bank’s asset choices and
the main effects of both liquidity and capital on the bank’s incentives to reduce insolvency risk in
Section 4. The main analysis is contained in Section 5. We conduct several extensions in Section
??, where we analyze the implications of enriching the baseline model. We conclude in Section ??.

2 Related Literature
Several papers analyze the role of equity capital as a way to reduce excessive bank risk-taking.
? shows that capital enhances the incentives of banks to reduce risk-shifting through increasing
shareholders’ losses in case of default. ? notes that although capital may reduce risk-shifting
by putting banks’ equity capital at risk, they may also induce higher risk-taking because they
erode banks’ charter value. Their analysis calls for a combination of capital requirements and caps
on deposit rates to avoid the negative effect of deleveraging on franchise values. ? reexamines
the relationship between capital requirements and risk-taking behavior in a setup in which banks
imperfectly compete for depositors. This paper shows that capital requirements reduces bank’s
risk-taking because the extra cost of capital born by regulated banks is passed onto depositors
through reduced deposit rates, so that capital requirements do not erode banks’ charter values.
Our paper draws from these insights to address the effect of capital on risk-taking. In our model,
capital reduces the bank’s reliance on deposits, therefore increasing the bank’s ’skin in the game’.
7
For instance, a LoLR may derive utility from bailing out an otherwise failing bank but only, as in ?, as long as
the liquidity shortfall lies below a certain threshold related to the level of solvency risk. Moreover, the LoLR may
additionally face limitation of funds or institutional constraints of some sort that may prevent the LoLR to rescue
the bank beyond a certain level of liquidity needs.

6
As a consequence, capital requirements reduce risk-shifting incentives. The main novelty of our
analysis is that we combine a liquidity requirement with a capital requirement. We find that the
effectiveness of capital requirement depends on the level of liquidity requirements.
There is a handful of papers looking into the joint regulation of capital and liquidity require-
ments. In an early treatment focusing on the Lender of Last Resort (LoLR), ? analyze the optimal
LoLR intervention policy in a model in which investors may face a coordination failure to roll-over
credit to a solvent but illiquid bank. In this model, solvency and liquidity requirements could pre-
vent the coordination failure, but they may entail a large cost in terms of foregone returns, which
the LoLR intervention can mitigate. In our model, we show that the LoLR policy should be jointly
established with liquidity and capital requirements. In our model, the role of the LoLR is to alle-
viate (or eliminate) illiquidity risk faced by a solvent bank. ? argues that liquidity, as a verifiable
and riskless asset, can be used as a commitment device to engage in efficient risk management. The
paper shows that in the presence of idiosyncratic liquidity shocks, a liquidity insurance scheme with
liquidity requirements is optimal, as it diversifies away individual shocks while avoiding a common
pool resource problem. Moreover, since liquidity enhances risk management, liquidity requirements
lead to book equity being a more precise indicator of equity market values, thus identifying a chan-
nel by which liquidity enhances the value of equity requirements. In our paper, in contrast, liquidity
leads to an increase in insolvency risk because it erodes the bank upside payoff by foregoing more
profitable investments. As noted by ?, which studies the effect of liquidity regulation on banks’
overall risk, liquidity requirements harm the bank’s upside payoff by reducing the share of assets in-
vested in loans. ? find out that while (mild) capital requirements lead to a higher volume of lending
and a lower default probability than in the case of unregulated banks, these positive effects disap-
pear when combined with liquidity requirements. In our model, in contrast, we find that liquidity
requirements complement capital requirements by increasing the likelihood of surviving a liquidity
crisis and, consequently, increasing the value of capital a tool to reduce insolvency risk. ? analyze a
global game in which investors’ decisions to roll-over debt are strategic complements and solvency
refers to the financial intermediary fundamentals to cope with coordinated actions. The analysis
relates liquidity–to deal with liquidity issues–and solvency ratios–to cope with insolvency–to the
level of transparency, as agents’ information is a main driver of coordinated actions. The paper
results calls for a joint determination of prudential and disclosure policies. In a recent treatment,
? analyzes the problem of a set of banks that have incentives to induce excessive systemic risk
through leverage. While the liquidity ratios in Basel III can be used to reduce the advent of fire
sales, capital requirements can be used to avoid individual bank failures. In our model, in contrast,
although capital and liquidity requirements are set to reduce solvency and illiquidity risk, respec-

7
tively, capital and liquidity feed back into each other because the effectiveness of one instrument
depends on the level of the alternative tool.

3 The baseline model

In this section, we describe the baseline model, which allows us to highlight the main intuition
underlying our analysis. In Section 6 we add a Lender of Last Resort (LoLR) and adress the
implications of having an LoLR in the optimal joint liquidity and capital regulation. Some of the
assumptions of the baseline model are discussed further below.
We consider a four-period economy, t ∈ {−1, 0, 1, 2}, where all agents are risk-neutral and there
is a zero discount factor. In this economy there is a social welfare maximizer regulator that sets
both liquidity and capital requirements (l∗ , k ∗ ) at t = −1. The capital structure of the bank at time
t = 0 consists of an amount d of common deposits, which are available upon demand both at t = 1
and at t = 2; an amount b of long-term deposits, which are cashable at t = 2 only; and an amount k
of equity capital, which is required to either meet or exceed the capital requirement, that is, k ≥ k ∗ .
We normalize the bank funds to 1, so that we have that d + b + k = 1. Consequently, the capital
requirement is both an absolute amount and a fraction of the bank’s assets. We fix the amount of
deposits d exogenously and let the regulator set a minimum capital requirement k ∗ ∈ [0, 1 − d].
We assume that common deposits constitute a substantial share of the bank’s funds. In partic-
ular, we assume that d ≥ d, for some d > 0, whose value we define precisely in Definition 7, laid
out in Appendix A. For simplicity, we assume that all deposits are completely insured by a deposit
insurance scheme, so that we can normalize the (excess) interest rates on common deposits to 0.
Long-term deposits, on the contrary, carry an (excess) interest rate r ≥ 0 which, for simplicity, we
take as exogenous.8 Moreover, we assume that equity capital is costlier than deposits, both from a
social perspective and to the bank. In particular, we assume that each unit of equity capital carry
an excess cost of ρ, with ρ > r. For ease of exposition, we assume that the cost of raising a unit
of equity capital to the bank is the same as the social shadow cost.9 We assume that the bank is
protected by limited liability, signifying that the bank is not required to meet its deposit obligations
in the event that its loan does not perform.
At time t = 0, the bank decides which fraction l of its resources to store as a liquid asset (i.e.,
8
Although we assume that long-term deposits are insured, depositors may require a higher return on term deposits
than on common deposits because common deposits are available upon demand, so that they constitute a source of
liquidity for consumers. Consumers may forgo some liquidity in exchance for an extra return r.
9
Nonetheless, this assumption does not play any role in our analysis, as banks will not be willing to hold equity
capital in excess of capital requirements, that is, capital will effectively not be a choice variable for the bank.

8
cash) and which amount 1 − l to invest in a higher yield asset (i.e., a loan). The loan is used to
finance a long-term project, which delivers its returns at t = 2. The loan returns M units per unit of
investment if it succeeds, yielding 0 if it fails. The probability of success θ is an endogenous choice
for the bank. Following ? and ?, we let the bank choose an unobservable solvency level θ ∈ [0, 1],
which determines the probability of success of its lending portfolio.10 Choosing a solvency level θ
carries a cost 2c θ2 , for some c > 0.11 This assumption gives rise to a standard risk-return trade-off,
as higher risk (i.e., lower θ) leads to a higher return. We refer to insolvency risk as to the value
1 − θ.
We model illiquidity risk assuming that some depositors have liquidity needs at t = 1. As in
?, we assume that a fraction β withdraws its deposits at t = 1, but we add aggregate uncertainty
in the amount β of early withdrawals. We assume that β is drawn from a common knowledge
random variable with c.d.f. F (·) and a well-defined density f (·) in the whole of its support [0, d].
We assume that this distribution is log-concave, and that the density function is continuous and
positive everywhere in the interior of its support. For any given level of liquidity l held by the
bank, the illiquidity risk is given by 1 − F (l). In this baseline model, we assume that the bank
is liquidated if the deposit withdrawals exceeds its cash reserves, that is, if β > l. We relax this
assumption in Section 6, where we allow for a Lender of Last Resort intervention to rescue a solvent
but illiquid bank.
Finally, we make technical assumptions, Assumptions 1-4, which we outline and discuss in
Appendix A.

3.1 Discussion of assumptions

In this section, we discuss the implications of some of the assumptions of the baseline model. On the
one hand, the bank is protected by limited liability. The immediate implication of this assumption
is that the bank does not internalize the losses that its default inflicts on society. Limited liability
induces a wedge between the bank’s objective function and social welfare, so that there is room for
welfare-improving regulation.
Also,deposits are insured, so that they carry either a zero interest rate (common deposits) or
an interest rate of r (long-term deposits). The results that we obtain do not strictly depend on
the assumption that deposits are insured, but on the underlying assumption that deposit rates
10
For instance, a bank may invest resources into developing a credit score model to improve the quality of its credit
appraisals.
11
See also ? and ? for analogous arguments of costly actions that enhance banks’ expected returns operating
through a reduction of the probability of a project failure.

9
are independent of the bank’s choice of risk θ and the regulator’s choices of liquidity and capital
requirements (l∗ , k ∗ ). If, on the contrary, deposits were not insured and the choice of risk were
observable, deposit interest rates would be proportional to the inverse of the bank’s choice of risk
1/θ. In this case, the bank would internalize the effect of its choice of insolvency risk on depositors
and the bank would act as if not insulated by limited liability.
A certain fraction of the bank’s liabilities is levied in the form of long-term deposits. The purpose
of this assumption is to eliminate a purely mechanical effect of capital requirements on illiquidity risk
through a reduction of the deposit base. In a model with deposits and capital only, any increase in
the capital requirement would necessarily lead to one of following two situations (or a combination
of both). On the one hand, an increase of capital requirements from k to k ′ keeping deposits d
constant leads to an expansion of the bank’s size from d + k to d + k ′ , which implies a reduction
of the share of deposits over assets from d/ (d + k) to d/ (d + k ′ ). On the other hand, an increase
of capital requirements from k to k ′ keeping the bank’s operating scale constant (at, say, a level of
1) leads to a reduction of the fraction of deposits over assets from d = 1 − k to d′ = 1 − k ′ . We
eliminate the interference of capital requirements with the common deposit base–the one subject
to liquidity risk–or the asset size, by assuming that the bank always substitutes away long-term
deposits by capital whenever the capital requirement increases. Assumption 1 in Appendix A on
the minimum value for the social cost of capital ensures that the optimal capital requirement never
exceeds 1 − d.12
Capital is raised from outside shareholders. Essentially, as in ? and ?, the bank must allocate a
fraction α of its ownership profits to reward its outside equityholders in exchange for their provision
k of outside equity capital at an exogenously given competitive rate ρ, so that:

α · ΠB (l, θ, k) = (1 + ρ) · k,

where ΠB (l, θ, k) stands for the bank’s profits, which we define precisely below, in equation (1) on
Section 4.
Regarding the early withdrawal phase, the amount of early withdrawals is a random variable
with an exogenously given distribution function. Unlike in ?, we need aggregate uncertainty in the
amount of early withdrawals in order to induce illiquidity risk. If the actual withdrawal amount
were known to the bank, it would store exactly as much liquidity as that amount so as to meet its
12
Nonetheless, the results that we obtain generalize to a model in which an elevation of capital requirements lead
to an expansion of the bank’s balance sheet at t = 0, insofar as deposits constitute a sufficiently large share of the
assets–If capital were too large a fraction of the bank’s funds, the bank would choose to hedge against any potential
early withdrawal, so that we would not have liquidity risk. An extended version of this paper with an analysis of
this case is available from the authors upon request.

10
early demand for deposits. The log-concavity assumption implies that the reverse hazard rate g (·),
f (l)
which is given by the ratio g (l) ≡ F (l)
for all l ∈ (0, d), is strictly decreasing and has an asymptote
at l = 0, i.e., liml→0 g (l) = +∞ (?), a property which we will draw from in our main result. This
assumption provides an Inada-type condition for liquidity, since any increase in liquidity at l = 0
yields an infinite return (from 0 to positive profits). Moreover, the log-concavity assumption on the
distribution of liquidity shocks implies that there are decreasing returns to liquidity since, as we show
below, the marginal return of liquidity is decreasing. The family of log-concave distributions with
compact support includes the uniform distribution, all beta distributions with shape parameters at
least as large as one, or power distribution with exponent at least as large as one, among others,
which encompass a large variety of shapes.
Moreover, we assume that the distribution function of early withdrawals is independent of the
(solvency) risk profile adopted by the bank. Arguably, one may expect that the amount of early
withdrawals be positively correlated with the risk profile of the bank, at least if withdrawing at
an early stage may be influenced to a certain extent by the bank’s fundamentals, and not only by
depositors’ idiosyncratic motives. While we acknowledge that this correlation may be very relevant,
the exercise that we carry out here assumes that illiquidity risk is originated independently of
insolvency risk, because we want to highlight that liquidity and insolvency risk are interrelated
through the bank’s decisions even when originated independently.
Finally, notice that our framework does not include any negative externality from a banking
failure. Including an externality in the form of, for instance, a positive shadow cost associated to
the bank liquidation, would reinforce our case for regulation, but would not add further insight.

4 Bank’s choice of solvency risk

Before addressing the optimal regulation of capital and liquidity, we first analyze the problem of
a bank that takes its capital structure as exogenously given and decides how much to invest in
reducing the risk of its loan portfolio. This section allows us to assess the roles of liquidity and
capital on the bank’s choice of risk.
We focus on three aspects. First, in Section 4.1, we find the liquidity level that maximizes
the bank’s profits for any predetermined level of capital. This choice will serve as a benchmark
to assess how the optimal liquidity requirement differs from the value that maximizes the bank’s
profits. Second, in Section 4.2, we analyze the interaction between illiquidity and insolvency risk.
Finally, in Sections 4.3 and 4.4 we assess the role of capital in shaping the bank’s choice of risk.
The bank’s problem consists of choosing a fraction of cash (or liquid assets) lB , and a screening

11
level θB , to solve the following problem:

maxθ∈[0,1],l∈[0,d] ΠB (θ, l, k) 

, (B)
ΠB (θ, l, k) ≥ 0 

s.t.
where the bank’s objective function is given by:
 
Upside payoff
l
Z z {
 }|  c 2
ΠB (θ, l, k) ≡ θ M (1 − l) + l − β − (1 − β + br − k)
 dF (β) − θ − (1 + ρ) k . (1)
0 | {z } |Cash {z } | {z 
} 2
|{z} | {z }
Loan value Deposit liabilities Cost of capital
| {z } Screening cost
Assets

The upper bound of the integral stems from the bank surviving the early withdrawal phase only
if it holds enough cash to meet the demand for deposits at t = 1, that is, as long as β ≤ l. Moreover,
the bank can only make positive profits if its investment succeds, which occurs with probability θ.
Upon successful completion of the project, the bank obtains a loan return of M · (1 − l) and keeps
cash for an amount l − β. Its deposit liabilities amount to 1 − β + b · r − k.13 Finally, we include
the constraint that the bank only operates if it makes non-negative profits.14
We rewrite the bank’s problem defining the bank’s upside payoff as:

π (l, k) ≡ M (1 − l) + l
|{z} − (1 + br − k) . (2)
| {z } | {z }
Loan value Leftover cash Deposit liabilities

This expression stands for the bank’s payoff conditional on a successful project completion once the
cost of equity and the cost of reducing risk have been deducted. Observe that, for any level of equity
capital k, π (l, ·) is a strictly decreasing function of the amount of cash l, reflecting the fact that
each unit of cash reserves carries an opportunity cost in terms of foregone loan opportunities. Also,
for any liquidity level l, π (·, k) is a strictly increasing function of capital k. While capital is costlier
than deposits and, consequently, any additional unit of capital hurts the bank’s profits, equity
capital increases the bank’s upside payoff. This feature reflects a central element of our analysis,
namely that capital increases the bank’s skin-in-the-game by reducing its deposit liabilities.15
13
Notice that the amount β of early withdrawals does not affect bank’s profits, as long as the bank holds enough
cash at time t = 1 to survive the early withdrawal phase. We show that early withdrawals indeed do affect banks’
profits in Section 6, where we allow for an LoLR to provide liqudity to a bank facing a large amount of withdrawals.
14
Below, we solve the problem of a regulator that maximizes social welfare. Hence, the non-negative-profit con-
straint for the bank will be harder to meet. In Appendix A we impose conditions on M , c, and ρ so that the constraint
does not bind in that problem. Consequently, it will not bind for this problem either. We henceforth ignore this
constraint.
15
Observe that the role of capital in this model is simply to reduce the amount of deposit liabilities. Hence, capital
serves the purpose of increasing the bank’s skin-in-the-game. As we shall see below, capital induces the bank to
reduce its risk profile.

12
 Substituting out equation (2) into equation (1), we can write the bank’s objective function as:
c
ΠB (θ, l, k) = θF (l) π (l, k) − θ2 − (1 + ρ) k. (3)
2

4.1 The bank’s profit-maximizing liquidity level

We first analyze the profit-maximizing liquidity level for a predetermined level of capital k, which
serves as a benchmark to assess optimal liquidity requirements. We write lB (k) for the bank’s
profit-maximizing liquidity level to highlight its dependence on the level of equity capital k. The
first order condition for an interior bank’s profit-maximizing level of liquidity is given by:

[l] g (lB (k)) π (lB (k) , k) − (M − 1) =0

| {z } | {z } , (4)
Marginal value of surviving early withdrawals Opportunity cost of liquidity

f (lB (k))
where g (lB (k)) ≡ F (lB (k))
stands for the reverse hazard rate of the distribution of early with-
drawals evaluated at the optimum lB (k).
Observe from equation (3) that the bank’s choice of liquidity lB (k) is given by the unique
liquidity level that maximizes F (l) · π (l, k), which is hump-shaped. The marginal effect of liquidity
on bank’s profits, which is given by equation (4), can be split into two components pushing in
opposite directions. On the one hand, liquidity reduces illiquidity risk. On the other hand, liquidity
carries an opportunity cost in terms of the foregone loan return M per 1 unit of stored cash. The
assumption that the distribution of early withdrawals is log-concave ensures that g (·) is strictly
decreasing with an asymptote at l = 0 (?): When liquidity is very small, the likelihood of surviving
the early withdrawal phase is very small and, consequently, the marginal value of holding liquidity
exceeds the opportunity cost of liquidity. Hence, lB (k) > 0. Also, Assumption 4 guarantees that
the deposit base is sufficiently large so that the bank does not want to keep its whole deposit
base as cash, that is, lB (k) < d. Finally, observe that the profit-maximizing liquidity level lB (k)
is independent of the bank’s choice θB of insolvency risk. The reason for this is that the bank’s
profit-maximizing choices are made conditioning on the event that the loan succeeds, as it obtains
no rents in the event that the loan fails.
We have argued above that capital increases the bank’s upside payoff. Hence, the marginal
value surviving early withdrawals (first term of Equation 4) increases with equity capital. On the
contrary, the opportunity cost of liquidity (second term of Equation 4) is independent of capital.
Hence, the optimal level of liquidity increases with capital. We summarize this discussion in the
following instrumental result, which is straightforward to derive.

Lemma 1 (Bank’s profit-maximizing liquidity level) For any level of capital k, we have that:

13
 (i) There exists a unique level of liquidity lB (k) that maximizes the bank’s profits, which is given
by the liquidity level that maximizes the (hump-shaped) expected bank’s upside payoff F (l) π (l, k).
(ii) The bank’s profit-maximizing liquidity value lB (k) is strictly increasing in capital, that is,
∂lB (k)
∂k
> 0.

Figure ?? depicts the effect of capital on the bank’s choice of liquidity, which is represented
on the horizontal axis. The solid and dashed decreasing lines correspond to the marginal value of
surviving the early withdrawal phase g (l) π (l, k) for two different levels of equity capital k2 > k1 ,
respectively. The marginal value g (l) π (l, k) of surviving the early withdrawal phase corresponds
to the product of two positive strictly decreasing functions, and is therefore strictly decreasing.
The marginal survival value for k2 consists of an upward shift of the marginal survival value for k1 ,
reflecting the fact that π (l, k2 ) > π (l, k1 ). The marginal value of liquidity M − 1 is independent of
both liquidity and capital. It is depicted as a horizontal solid line. The profit-maximizing liquidity
level lB (k) is given by the intersection between g (l) π (l, k) and M − 1. Hence, we have that
lB (k2 ) > lB (k1 ).

itbphFX5.6164in3.2785in0ptFIG Bank liquidityFigure 1: Bank’s profit-maximizing liquidity as a fuct

4.2 The effect of liquidity on insolvency risk

We analyze now the interaction between liquidity and the bank’s choice of insolvency risk. The first
order condition for an interior solution to the bank’s insolvency risk choice θB is given by:16

1
[θ] θB (l, k) = c
· F (l) · π (l, k) , (5)

where θB (l, k) stands for the solvency level chosen by a bank with liquidity l and equity capital k.
Observe that the marginal return of reducing insolvency risk is given by the bank’s upside payoff
F (l) · π (l, k), weighted by the factor c. Intuitively, the bank obtains an amount π (l, k) if it ends up
being solvent. But a precondition for the bank to be solvent is that it survives the early withdrawal
phase, whose likelihood is given by F (l). Consequently, the bank chooses a higher level of solvency
the higher its upside payoff. Since lB (k) maximizes the bank’s upside payoff F (l) · π (l, k), it follows
that θB (l, k) is also maximized at lB (k). The following result is immediate in light of Statement
(i) on Lemma 1.
16
In order to ensure that θB (·, ·) is interior, so that the regulation problem is interesting, we make an assumption
on the range of c (see Assumption 3 in Appendix A).

14
Proposition 1 (Liquidity and insolvency risk) For any given level of equity capital k, the
bank’s choice of solvency θB (l, k) is a hump-shaped function of its liquidity reserves l, and is maxi-
mized at the bank’s profit-maximizing liquidity level lB (k).

Figure ?? illustrates Proposition 1. The dashed line corresponds to the bank’s upside payoff
F (l) · π (l, k) for a given level of capital k. As stated in Lemma 1, this function is hump-shaped
in liquidity. The solid line represents the optimal solvency level, which is a scaled (down, by c)
transformation of the bank’s upside payoff.

itbpF5.7019in3.3067in0inFIG Solvency and liquidityFigure

We shall see below that liquidity requirements exceed the bank’s profit-maximizing level, which
is the level that a bank would choose in the absence of liquidity regulation. Hence, liquidity
requirements induce a higher level of insolvency risk. The following corollary highlights the fact
that an excessive value of liquidity harms solvency.17

Corollary 1 (Binding liquidity requirements increase insolvency risk) Any level of bank’s
liquidity in excess of the bank’s profit-maximizing liquidity level lB (k) induces higher insolvency risk
∂θB (l,k)
than in the absence of liquidity regulation. Formally, ∂l
< 0 for any l > lB (k).

4.3 The effect of equity capital on insolvency risk

We now turn to the analysis of the effect of capital on the bank’s choice of solvency. We first state
the result, which follows immediately from differentiating equation (5), and then provide intuition
for it.

Proposition 2 (Capital and insolvency risk) For any level of liquidity l, solvency θB (l, k) is
a strictly increasing function of capital k.

As argued above, capital increases the bank’s upside payoff π (l, k). Hence, the bank’s marginal
return of increasing its level of solvency, which is given by F (l) · π (l, k), is larger the higher the
amount of capital. As a consequence, an elevation of capital makes the bank more solvent through
increasing its skin-in-the-game.
17
This result is reminiscent of the result found by ?, who shows that liquidity may harm insolvency risk.

15
4.4 Capital and the effect of liquidity on insolvency risk
We have seen above (Corollary 1) that liquidity in excess of the bank’s profit-maximizing level of
liquidity lB (k) leads to a reduction of solvency because it harms the bank’s profits. We shall see
below that liquidity requirements will be set in excess of the bank’s profit-maximizing liquidity
level. Hence, optimal liquidity requirements will harm solvency. The following result states that
capital mitigates this effect. This results plays an essential role in the determination of the joint
optimal capital and liquidity requirements.

Proposition 3 (Impact of capital on the effect of liquidity on solvency) Equity capital re-
∂θB (l,k)
duces the (negative) effect of excessive liquidity on solvency. That is, for any l > lB (k), ∂l
is
strictly decreasing in equity capital k.

In order to provide intuition for this result, consider the effect of liquidity on solvency, which is
proportional to the first order condition of the bank’s problem with respect to liquidity, as stated
in Equation (4):
∂θB (l, k)
∝ g (l) · π (l, k) − (M − 1) < 0 for l > lB (k) .
∂l | {z } | {z }
Marginal value of surviving early withdrawals Opportunity cost of liquidity

Observe that equity capital increases the marginal value of surviving early withdrawals without
affecting the opportunity cost of liquidity. Hence, increasing capital makes the (negative) effect of
excess liquidity on solvency less negative.
Figure ?? depicts the bank’s choice of solvency as a function of its liquidity reserves for two levels
of capital (the upper curve for a higher level of capital than the lower curve, i.e., k2 > k1 ). The
picture illustrates the three effects of equity capital on illiquidity and insolvency risk stated above.
First, an increase in capital shifts the bank’s profit-maximizing level of liquidity to the right (Lemma
1, Statement (ii)), that is, lB (k2 ) > lB (k1 ). Second, an increase in capital leads to an upward shift
of the solvency curve (Proposition 2), that is, θB (l, k2 ) > θB (l, k1 ) for any l. Finally, the upper
solvency curve is flatter than the lower solvency curve to the right of the profit-maximizing level of
∂θB (l,k2 ) ∂θB (l,k1 )
capital (Proposition 3), that is, ∂l
< ∂l
for all l > lB (k1 ).

itbphF5.5799in3.2428in0ptFIG SolvencyFigure

5 Regulation of liquidity and capital

We now proceed to the analysis of the optimal regulation of capital and liquidity. In particular,
we analyze the problem of a regulator that sets capital and liquidity requirements at t = −1 to

16
maximize social welfare. More precisely, the regulator sets up a minimum liquidity and capital
requirement pair (l∗ , k ∗ ) to solve the following problem:

maxl∈[0,d],k∈[0,1−d] ΠR (l, k) 





s.t. b+k =1−d , (R)





ΠB (θB (l, k) , l, k) ≥ 0 

where the regulator’s objective function is given by:

Z l
c 2
ΠR (l, k) ≡ θB (l, k) M (1 − l) dF (β) + l − (1 + br − k) − θB (l, k) − (1 + ρ) k. (6)
0 2
First, observe that the choice of solvency is chosen by the bank and is not verifiable by the regulator.
Hence, the regulator must take the bank’s self-enforcing reaction to liquidity and capital require-
ments θB (l, k) as given. The first constraint follows from our assumption that the regulator cannot
change the scale of the bank’s resources, which we have normalized to 1. Hence, b is determined by
the choice of k given this accounting identity. The second constraint is the participation constraint
for the bank, which has to obtain non-negative profits. Assumption 2, laid out in Appendix A,
guarantees that the bank has non-negative expected profits. Hence, we can prescind of the bank’s
participation constraint, which we omit henceforth.

5.1 The role of regulation

For the remainder of the paper it is useful to write the bank’s (net) deposit liabilities, which play a
central role in the design of the optimal regulatory scheme, as follows:
D (l, k) ≡ 1 + br − k − l. (7)
The bank’s liabilities consist of an amount 1 − b − k of common deposits, as well as an amount
b·(1 + r) of long-term deposits. Hence, D (l, k) stands for the amount of loan earnings that the bank
will have to allocate to paying its deposit liabilities in case it succeeds–observe that the amount
of liquid assets l that the bank uses to meet early withdrawals can also be used at t = 2 and is
therefore fully deducted from its deposit liabilities.
There is a natural alternative interpretation for D (l, k) as the deposit insurance loss in case of
a bank failure: D (l, k) represents the depositors’ claims that the bank cannot meet if it fails. This
expression is core in our analysis. Indeed, combining expressions (1) and (6), the former evaluated
at the bank’s optimal choice of solvency θB (l, k), we can write the regulator’s objective function as:
ΠR (l, k) = ΠB (l, θB (l, k) , k) − [1 − θB (l, k) F (l)] D (l, k) . (8)
| {z } | {z }
Bank’s profits Expected externality

17
 Regulation is needed because the bank does not internalize the harm that its potential failure
inflicts on the deposit insurance scheme, which is captured by the second term in Expression (8).
Put differently, since the bank is protected by limited liability, it does not face the downside of a
failure. On the contrary, society experiences a welfare loss of D (l, k) when the bank cannot meet
its deposit obligations, an event which occurs with probability 1 − θB (l, k) F (l).
In order to see that elevating the level of solvency is welfare-enhancing, we can differentiate
Expression (8) and write the effect of the solvency level on the regulator’s objective function as:
∂ΠR (l, kR (l)) ∂ΠB (l, θB (l, k) , k)
= +F (l) D (l, k) > 0. (9)
∂θB ∂θB
| {z }
Bank′ s FOC:=0

The first term in expression (9) captures the effect of increasing solvency on the bank’s profit.
This term is zero, because it corresponds to the bank’s first order condition for an optimal level
of solvency. Loosely speaking, the bank fully internalizes this portion of the effect when it chooses
its solvency level. The second term in Expression (9) is positive, reflecting the fact that the bank
chooses too low a solvency level because it does not internalize the social loss inflicted on the deposit
insurance scheme when it fails.
In a (first-best) world in which the bank’s choice of insolvency risk were verifiable, the regulator
would set a solvency requirement in excess of the bank’s laissez-faire choice precisely because the
bank does not internalize the social loss following a failure. Capital requirements would not play
any role in this framework and would be set to zero. However, in a (second-best) world in which
the choice of insolvency risk is not verifiable, capital requirements induce a reduction of insolvency
risk through increasing the bank’s skin-in-the-game, as seen above (Proposition 2). The purpose
of capital regulation, which entails the substitution of a valuable part of the depositors’ base, is
therefore to reduce the level of insolvency risk beyond the bank’s will.
Liquidity requirements increase the bank’s resilience against the advent of excessive early with-
drawals. In the absence of regulation, the bank would hold the (the profit-maximizing) liquidity
level lB (k) derived above (Lemma 1 (i)). In trading off illiquidity risk and investment in loans,
the bank’s laissez-faire choice lB (k) ignores the expected (social) cost of a potential bank failure
captured by the second term in expression (8). Liquidity requirements are set taking this addi-
tional effect into account. The reason why liquidity and capital must be set jointly is because the
effectiveness of capital depends on the amount of liquidity and viceversa.
In what follows, we solve Problem (R). In order to get some insight on the effect of liquidity
on the effectiveness of capital, in the following subsection we construct the optimal capital response
curve kR (l), which establishes a scheme of optimal capital requirements for any given level of
liquidity l. In Subsection 5.3 we assess the effect of capital on the effectiveness of liquidity. With

18
this purpose, we construct the optimal liquidity response curve lR (k), which maps each level of
capital k to its corresponding optimal liquidity requirement. In Subsection 5.2 we put them together
to establish the joint optimal liquidity and capital requirements (l∗ , k ∗ ), which is the unique pair
satisfying l∗ = lR (k ∗ ) and k ∗ = kR (l∗ ). We discuss the implications of our analysis in Subsection
5.5, where we argue whether liquidity and capital are complementary or offsetting tools.

5.2 Optimal capital response curve

In this section, we construct the optimal capital response curve kR (l), which maps each liquidity
level l to the optimal capital response. Differentiating the regulator’s objective function with respect
to capital, we have that an interior optimal capital response curve must satisfy:18

dΠR (l, kR (l)) ∂ΠR (l, kR (l)) ∂θB (l, kR (l))

= − (ρ − r) = 0. (10)
dk ∂θB | ∂k
{z } | {z }
| {z } Direct effect of capital:<0
>0 >0
| {z }
Indirect effect of capital ( IEK) >0

On the one hand, there is a (negative) direct effect of capital of imposing capital requirements,
which is given by the shadow cost difference ρ − r > 0 of substituting deposits by equity capital. In
addition, there is a (positive) indirect effect of capital on social welfare, which follows from the fact
that capital increases the level of solvency beyond the inefficiently low level that the bank would
set in the absence of a capital requirement.
∂ΠR (l,kR (l))
Recall from Expression (9) that ∂θB
> 0. Combining this observation with Proposition 2,
∂θB (l,kR (l))
which states that equity capital induces a higher level of solvency, that is, ∂k
> 0, we have
that the indirect effect of capital on social welfare is positive.
Liquidity does not interfere with the direct effect, but it does have an impact on the indirect
effect. Hence, the optimal capital response depends on the liquidity level. The following proposition
characterizes the optimal capital response curve kR (l), which is illustrated in Figure ???

Proposition 4 (Optimal capital response curve) There exists a cost of capital ρmax (whose
value is determined on Definition 4 in Appendix A) such that:
(i) (Zero capital requirement when capital too costly) For any ρ ≥ ρmax , the optimal capital
response is kR (l) = 0 for any l.
(ii) (Capital response curve hat-shaped) For any ρ < ρmax , there exist liquidity levels 0 < l1 (ρ) <
ˆl (ρ) < l2 (ρ) ≤ d such that the optimal capital response curve is hump-shaped for l ∈ (l1 (ρ) , l2 (ρ)),
18
Below, we derive the conditions under which the optimal capital requirement is not interior.

19
attaining a maximum at ˆl (ρ), and zero otherwise, that is:


 =0 for l ≤ l1 (ρ)

 is increasing for l (ρ) < l < ˆl (ρ)

1
kR (l) : .

 ˆ
is decreasing for l (ρ) < l ≤ l2 (ρ)


=0 for l > l2 (ρ)


(iii) (Capital response independent of M and ”larger” as ρ decreases) As ρ decreases, we have

that l1 (ρ) shifts to the left, l2 (ρ) shifts to the right (as long as l2 (ρ) < d) and kR (l) larger for any
l ∈ (l1 (ρ) , l2 (ρ)). Moreover, kR (l) is invariant in M .

Proof. See Appendix C.

itbpF2.9187in1.8014in0inFigure itbpF2.9179in1.8014in0inFigure

The intuition behind this result is as follows. For notational simplicity, we write the indirect
effect of capital referred to in Expression (10) as follows:

∂ΠR (l, kR (l)) ∂θB (l, kR (l)) 1+r

IEK (l, k) ≡ = F 2 (l) D (l, k) .
∂θB ∂k c
Capital increases social welfare because it makes the bank more solvent which, in turn, increases
social welfare. In particular, the optimal capital requirement must be set at the level that equalizes
the indirect marginal return of capital IEK (l, k) to the shadow cost ρ − r of substituting depositors
by equityholders. When the cost of capital is very high (ρ ≥ ρmax ) the indirect marginal return
of capital lies below the shadow cost of capital, so that the optimal capital response is zero for all
liquidity levels.
Now consider the case in which capital is not too costly, that is, consider a value of the cost
of capital such that ρ < ρmax . First, notice that D (l, k) is strictly decreasing in capital: Substi-
tuting depositholders by equityholders reduces the bank’s net deposit liabilities.19 Consequently,
IEK (l, k) is strictly decreasing in capital as well. Therefore, the marginal return of capital is
maximized when k = 0. Now, observe that when liquidity is low, IEK (l, 0) is small: When the
probability of surviving the early withdrawal phase F (l) is small, the expected return of increasing
solvency is small and therefore the return of capital is also small. On the other extreme, when
liquidity is high, the term IEK (l, 0) is also small: When the value of the deposit insurance loss in
case of failure D (l, 0) is small, the wedge between the bank and the regulator’s objective functions
is small as well. Consequently, when liquidity is either low or high (formally, either when l ≤ l1 (ρ)
19
Notice that we can write D (l, k) = D (l, 0) − k.

20
or when l ≥ l2 (ρ)), the optimal capital response is zero, because the marginal social return of the
first unit of capital is smaller than the shadow cost of substituting depositors by equityholders.
For intermediate values of liquidity, for which the return of the first unit of capital exceeds its
opportunity cost, it is optimal to set a positive capital response. In this case, in the liquidity range
l ∈ (l1 (ρ) , l2 (ρ)), the optimal capital requirement is given by:

IEK (l, k) = ρ − r. (11)

The capital response curve is hump-shaped in l ∈ (l1 (ρ) , l2 (ρ)). Intuitively, IEK (l, k) is
hump-shaped because there are two effects of liquidity in the return of capital pushing in oppo-
site directions. As liquidity increases, illiquidity risk reduces. Hence, the probability of reaching
the loan maturity phase–which is the time at which increasing solvency matters– increases. How-
ever, as liquidity increases, the wedge between the bank and the regulator’s objective functions
shrinks, making regulation less effective. Our assumption that F (·) is log-concave ensures that
when liquidity is low the first effect dominates, while the second one dominates when liquidity is
large.
So far, we have conducted this analysis fixing the cost of capital ρ. As ρ decreases, the incre-
mental cost of capital ρ − r is reduced. A glance at equation (11) reveals that cheaper capital leads
to an upward shift of the optimal capital response curve in the range in which is positive: Since the
right-hand-side diminishes, the capital response must be increased so as to reduce the left-hand-side
as well. Moreover, the minimum liquidity level l1 (ρ) for which the marginal return of the first unit of
(ρ−r)·c
capital exceeds the incremental cost of capital, which satisfies IEK (l1 (ρ) , k) = 1+r
decreases as
(ρ−r)·c
well. If l2 (ρ) is interior, then it satisfies the same condition as l1 (ρ), that is, IEK (l2 (ρ) , k) = 1+r
.
Then, a reduction of the cost of capital leads to an increase in the maximum level of liquidity l2 (ρ) for
(ρ−r)·c
which there is a positive capital requirement. If, on the contrary, we have that IEK (d, k) = 1+r
,
then the capital requirement is positive for the largest possible value of liquidity, that is, l2 (ρ) = d.
In this case, l2 (ρ) does not change as ρ decreases. Finally, ˆl (ρ) is decreasing in ρ because the
marginal social return of capital IEK (l, k) is decreasing in capital. Hence, as capital increases, the
maximum of the marginal social return of capital decreases.

5.3 Optimal liquidity response curve

In this section, we construct the optimal liquidity response curve lR (k), which maps each capital
level k to the optimal liquidity response. Differentiating the regulator’s objective function with

21
respect to liquidity, we have that an interior liquidity requirement satisfies:20

dΠR (lR (k) , k) ∂ΠR (lR (k) , k) ∂ΠR (lR (k) , k) ∂θB (lR (k) , k)
= + = 0. (12)
dl | ∂l
{z } ∂θB | ∂l
{z }
| {z }
Direct effect of liquidity ( lD (k))>0 >0 <0
| {z }
Indirect effect of liquidity ( IEL(l,k))<0

We can split the effect of liquidity on social welfare into a direct and an indirect effect. The direct
effect of liquidity captures the effect of liquidity on social welfare ignoring the effect of liquidity
on the bank’s choice of insolvency risk. The indirect effect of liquidity accounts for the (negative)
impact of liquidity on the objective function through its influence on insolvency risk. As we shall
see, capital influences both effects. Hence, the optimal liquidity requirement depends on the level of
capital. The following proposition characterizes the optimal liquidity response curve lR (k), which
is illustrated in Figure ???

Proposition 5 (Optimal liquidity response curve) (i) (Liquidity requirements binding) The
liquidity requirement is binding for the bank for any level of capital, that is, lR (k) > lB (k) > 0 for
all k.
(ii) (Liquidity response either hump-shaped or increasing) There exists M > M such that:
(ii.1) If M < M , the liquidity response lR (k) is a hump-shaped function of capital, that is,
there exists a capital threshold k̂ (M ) < 1 − d such that:
(
is increasing for k < k̂ (M )
lR (k) : .
is decreasing for k > k̂ (M )

(ii.2) If M ≥ M , the liquidity response lR (k) is strictly increasing in capital.

(iii) (Liquidity response independent of ρ and ”smaller” as M increases) For any M ′ > M , we
have that lR (k) |M ′ < lR (k) |M . Moreover, lR (k) is invariant in ρ.

Proof. See Appendix C.

itbpF2.9421in1.791in0inFigure itbpF2.9421in1.791in0inFigure

In order to provide some intuition for this result, we first construct the ”direct-effect liquidity
response” lD (k). The direct-effect liquidity response lD (k) corresponds to the liquidity response
curve that would be set by a regulator that ignored the (negative) indirect effect of liquidity on
20
As we shall see below, the optimal liquidity requirement is always interior.

22
solvency. Hence, for any given value of capital k, the direct-effect liquidity response curve lD (k)
satisfies:
 
+ |{z} 1
∂ΠR (lD (k) , k)  Cash-in-hand

=  = 0. (13)
 
| ∂l
{z }  θ|B (lD (k) , k) · M · F (lD (k))
{z
· [g (lD (k)) (1 − lD (k)) − 1]
}

Direct effect Opportunity cost

The direct effect of liquidity on social welfare is given by two factors. On the one hand, liquidity
constitutes a cash-in-hand asset that reduces the bank’s net deposit liabilities D (l, k) regardless of
the loan performance. The marginal value of cash-in-hand is therefore 1. On the other hand, storing
cash has an opportunity cost in terms of foregone loans. Observe that an ”expected loan maximizer
bank” (ELVB) would choose a liquidity level lELV B satisfying g (lELV B ) (1 − lELV B ) = 1.21 Hence,
any liquidity level in excess of lELV B contributes negatively to the expected loan value.
We now argue how capital affects the direct-effect liquidity response curve lD (k). Capital
increases the probability of a loan success by the skin-in-the-game effect on the bank’s choice of
solvency (first term of Equation (13)). However, it does not affect the cash-in-hand effect (second
term of Equation (13)). Hence, capital increases the relative weight of the expected loan value
versus the cash-in-hand effect: The more resilient the bank, the higher the opportunity cost of
storing liquidity. Hence, as capital increases, lD (k) comes closer to lELV B . Consequently, the
direct-effect liquidity response curve lD (k) is strictly decreasing in capital k.
The direct-effect liquidity requirement lD (k) lies above the bank’s profit-maximizing liquidity
level lB (k), that is, lD (k) > lB (k). The reason is that the bank does only value the cash-in-hand
asset in the event of a success. Hence, the cash-in-hand factor in the bank’s optimality condition is
weighted by θ · F (lB (k)) < 1.
Consider now the indirect effect of liquidity, which we can write as:

∂ΠR (l, k) ∂θB (l, k) ∂θB (l, k)

IEL (l, k) ≡ = [F (l) D (l, k)] < 0 iff l > lB (k) . (14)
∂θB ∂l ∂l
From Lemma 1, insolvency risk is minimized at the bank’s profit-maximizing liquidity level
∂θB (l,k)
lB (k). Hence, we have that ∂l
< 0 for any l > lB (k) and, in particular, for lD (k). Moreover,
recall from Equation (9) that an increase in solvency is welfare enhancing. Hence, the indirect effect
of liquidity on social welfare is negative, reflecting the fact that liquidity requirements above the
bank’s profit-maximizing liquidity level reduce the bank’s loan investments and therefore increases
21
An ”expected loan maximizer bank” (ELVB) would choose l to maximize the expected value of its loan portfolio,
that is: lELV B = arg maxl∈[0,d] θ · F (l) · M · (1 − l). Observe that the bank that we are modelling maximizes its
expected profits which, in addition to loans, include its liabilities, that is: lB (k) = arg maxl∈[0,d] θ · F (l) · π (l, k).

23
insolvency risk. As a consequence, the optimal liquidity requirement must be smaller than the
direct-effect liquidity requirement would be, that is, lR (k) < lD (k).
Nonetheless, the indirect effect does not completely offset the direct effect. In order to see
why, observe that the indirect effect diminishes as liquidity reduces, and it completely vanishes at
∂θB (lB (k),k)
the bank’s profit-maximizing liquidity level lB (k), since ∂l
= 0. Hence, when l = lB (k)
the indirect effect is zero, while the direct effect is positive. Therefore, we have that the optimal
liquidity response does always constitute a binding requirement for the bank, that is:

lB (k) < lR (k) < lD (k) . (15)

Capital does therefore exert two opposing effects over the optimal liquidity response curve lR (k).
As capital increases, the negative effect of liquidity on welfare diminishes. The reason is twofold.
First, because D (l, k) gets reduced, so that the effect of the level of solvency on welfare diminishes.
Second, because the negative effect of liquidity on solvency also reduces (”slope effect” of Proposi-
tion 1, displayed in Figure ??). Hence, as capital increases, liquidity requirements must increase.
However, as these effects vanish out, an aditional effect pushing in the opposite direction emerges.
As capital increases, the bank becomes more solvent (”level effect” of Proposition 2, represented in
Figure ??). Consequently, the opportunity cost of liquidity in terms of foregone loans increases. If
this ”opportunity cost” effect gets to dominate, the optimal liquidity response curve lR (k) eventu-
ally decreases. However, the ”opportunity cost” effect becomes strong enough so as to eventually
dominate only if the project profitability M is not too large.
The effect of the project profitability M on the ”opportunity cost” effect can be seen in Equation
(13). The project profitability M constitutes a weighting factor of the expected loan value versus
cash-in-hand–the higher the project profitability, the higher the opportunity cost of liquidity. As
the project profitability M increases, liquidity requirements approach the liquidity level lELV B that
an ”expected loan maximizer bank” (ELVB) would choose for any level of capital. Hence, the first
term on Equation (13) becomes less responsive to an elevation of capital as M increases. When M
is sufficiently large the optimal liquidity response function lR (k) is always increasing in k .
Figure ?? depicts the relationship between several liquidity measures for M < M̄ . On the
one hand, we have the bank’s profit-maximizing liquidity level lB (k) which, as stated in Lemma
1 is strictly increasing in capital. The upper line corresponds to the direct-effect optimal liquidity
response lD (k), which corresponds to the optimal liquidity response curve that would be set by
a regulator that ignored the indirect effect of liquidity on the solvency level. As argued above,
this liquidity measures is strictly decreasing in capital. As stated in Proposition 5, the regulator’s
liquidity response lR (k) is a hump-shaped function of capital with a maximum at some capital level
k̂ (M ). Also, as specified by the set of inequalities 15, lR (k) lies above lB (k) (i.e., the liquidity

24
requirement will bind) and below lD (k) (i.e., the indirect effect of liquidity pushes the optimal
amount of liquidity requirements down). The indirect effect of liquidity (through its negative effect
on solvency) is simply the difference between the direct effect and the overall effect of liquidity.
From 3 we have that the negative effect of liquidity on solvency is mitigated as capital increases.
Hence, lD (k) − lR (k) is strictly decreasing in capital.

itbpF4.9282in2.7804in0inFIG LiquidityFigure

5.4 Optimal joint liquidity and capital requirements: Do liquidity and

capital complement or offset each other?
We are left with determining how capital and liquidity should be set together. Before proceeding
to the main analysis, we establish two instrumental results in the following two lemmata. The first
one states that there exists a unique regulatory policy pair (l∗ , k ∗ ), which is such that the liquidity
requirement is optimal given the capital requirement, i.e., l∗ = lR (k ∗ ); and, conversely, that the
capital requirement is optimal given the liquidity requirement, that is, k ∗ = kR (l∗ ). The second
lemma asserts that capital and liquidity may be complementary or offsetting tools. We then address
the issue of whether capital and liquidity complement or offset each other in equilibrium, which
depends on the social cost of capital and the project profitability level. We finally analyze the
implications of whether liquidity and capital are complementary or offsetting regulatory tools to
explain how changes in the social cost of capital ρ and the project profitability level M affect the
optimal liquidity and capital requirements. Abusing notation, we let l∗ (ρ, M ) and k ∗ (ρ, M ) stand
for the optimal liquidity and capital requirement, respectively, for a given cost of capital ρ and
return to investment M . The following lemma establishes that the optimal joint regulatory policy
is well-defined and unique.

Lemma 2 (Existence and uniqueness) For any given social cost of capital ρ and project prof-
itability M , there exists a unique pair of liquidity and capital requirements (l∗ , k ∗ ) such that l∗ =
lR (k ∗ ) and k ∗ = kR (l∗ ).

Proof. See Appendix C.

Having established the existence and uniqueness of the optimal regulatory policy, we proceed
now to the analysis of the determinants of the optimal liquidity and capital requirements. At the
heart of the interplay between liquidity and capital requirements is the issue of whether capital and
liquidity complement or offset each other. We formalize the notion of complementary or offsetting
tools in the following definition.

25
Definition 1 (Complementary and offsetting regulatory tools) We say that capital and liq-
∂ ∗
uidity are complementary tools in equilibrium whenever ∂ρ
l (ρ, M ) < 0 and ∂
∂M
k ∗ (ρ, M ) < 0. We
∂ ∗
say that capital and liquidity are offsetting tools whenever ∂ρ l (ρ, M ) > 0 ∂
and ∂M k ∗ (ρ, M ) > 0.22

The concept of complementary tools captures the notion that the optimal requirement of one
of the regulatory tools (e.g., liquidity) decreases when the shadow cost of the other regulatory tool
(e.g., the social cost of capital) increases. As we shall see below, whenever the cost of one regulatory
tool raises (e.g., the social cost of capital), the optimal requirement of that regulatory tool (e.g.,
capital) decreases. If liquidity and capital are complementary, the use of the other regulatory tool
(e.g., liquidity) should therefore decrease as well. On the contrary, liquidity and capital offset each
other whenever an increase of one regulatory tool should be optimally coupled with the decrease of
the other.

Lemma 3 (Liquidity and capital complementarity/offsetting regions) The function 2g (l)·

D (l, k) − 1 = 0 partitions the interior of the space of regulatory tools {(l, k)}(0,d)×(0,1−d) into a Com-
plementary Region (to the left-and-below the curve) and an Offsetting Region (to the right-and-above
the curve). Liquidity and capital are complementary (resp. offsetting) tools if the equilibrium falls
on the Complementary Region (resp. Offsetting Region).23

itbpF4.7314in3.103in0inFigure

The space of...??? explain why not depends on factor costs

Proposition 6 (Liquidity and capital complementary or offsetting tools in equilibrium)

There exists a threshold M̄ , and, for each M < M̄ , there does also exist a threshold ρ̂ (M ) such
that:
(i) Liquidity and capital are offsetting tools in equilibrium if M < M̄ and ρ < ρ̂ (M ).
(ii) Liquidity and capital are complementary tools in equilibrium otherwise (i.e., if either M > M̄
or if both M < M̄ and ρ > ρ̂ (M )).

The following proposition establishes the ranges for which the regulatory tools are complemen-
tary or offset each other.
22 ∂ ∗ ∂ ∗
For the sake of completeness and coherence of this definition, observe that ∂ρ l (ρ, M ) and ∂M k (ρ, M ) > 0 do
∂2 ∂2 ∂
have the same sign, which follows immediately from the fact that ∂l∂k ΠR (l, k) = ∂k∂l ΠR (l, k) and ∂ρ lR (ρ, M ) =
∂
∂M kR (ρ, M ) = 0.
23
In the boundary case in which k ∗ (ρ, M ) = 0 we have that ∂
∂M k
∗
(ρ, M ) = ∂ ∗
∂ρ l (ρ, M ) = 0.

26
Corollary 2 (Comparative statics on optimal liquidity and capital requirements) (i) A
raise in the social cost of capital ρ leads to a decrease in the optimal capital requirement k ∗ (ρ, M ).
Moreover, it leads to a decrease in the optimal liquidity requirement l∗ (ρ, M ) if and only if liquidity
and capital are complementary tools.
(ii) A raise in the opportunity cost of liquidity M leads to a decrease in the optimal liquidity
requirement l∗ (ρ, M ). Moreover, it leads to a decrease in the optimal capital requirement k ∗ (ρ, M )
if and only if liquidity and capital are complementary tools.

Figure ?? may help understand the intuition behind Proposition ??. In these figures, liquidity is
depicted in the horizontal axis and the optimal capital response curve is drawn against the horizontal
axis (that is, as functions are typically depicted). Capital is represented on the vertical axis. We
draw the optimal liquidity response curve as a function that maps a certain capital level to a unique
optimal liquidity requirement, that is, we invert the axes and plot the curve as a function of the
variable represented on the vertical axis. From the perspective of the horizontal axis, the optimal
liquidity response curve is a correspondence, mapping each liquidity level to the (potentially two)
capital level(s) for which that particular liquidity level constitute an optimal liquidity response.
Consider first the liquidity requirement curve lR (k) |MLow for a given (low) opportunity cost of
liquidity MLow . This curve intersects the capital requirement curves kR (l) |ρ , which represent the
optimal capital response for several values ρ1 > ρ2 > ρ̂ (MLow ) > ρ3 of the cost of capital. The
jointly determined capital and liquidity requirement for a given cost of capital ρ is given by the
intersection of the liquidity requirement curve lR (k) |Mlow and the corresponding capital requirement
curve kR (l) |ρ . Notice that an increase in the cost of capital ρ would shift the capital requirement
curve kR (l) |ρ down-and-rightwards, as seen in Proposition 4. However, the cost of capital does not
affect the liquidity requirement curve lR (k) |M . On the contrary, an increase in the opportunity
cost of liquidity M leads to a downward (leftward, in the picture) shift of the liquidity requirement
curve lR (k) |M , leaving the capital requirement curve kR (l) |ρ unchanged. Consequently, increasing
the cost of capital leads to a reduction of a positive capital requirement. Also, an increase in the
opportunity cost of liquidity leads to a reduction of the liquidity requirement.
When the cost of capital is large (ρ1 > ρ̄), the curve lR (k) |Mlow intersects kR (l) |ρ1 at its (zero)
flat portion. Hence, the capital requirement k ∗ (ρ1 , MLow ) is zero. Moreover, an increase of the
cost of capital ρ would leave capital requirements unchanged at zero, as the liquidity and capital
requirement curves would continue to intersect at the flat segment of the capital response curve.
The next pair of capital requirements corresponds to a cost of capital ρ2 ∈ (ρ̂ (MLow ) , ρ̄), which is

27
smaller than ρ1 . The optimal capital an liquidity response curves intersect at a point where both
curves are upward slopping. In this range, capital and liquidity are complements: an increase of
either the cost of capital ρ or of the opportunity cost of liquidity M would lead to a decrease of
both the capital and the liquidity requirement. The cost of capital that determines whether capital
and liquidity are complements or substitutes is ρ̂ (MLow ), which corresponds to the cost of capital
for which both curves intersect at their respective peaks. The last capital response curve, for ρ3 <
ρ̂ (MLow ), depicts a case in which capital and liquidity are substitutes: any change in the cost of
one of the factors leads to opposite movements of the optimal capital and liquidity requirements.
The curve lR (MHigh ) in Figure ??.BIS corresponds to the optimal liquidity response curve for a
higher return to investment. This curve lies below the curve lR (MLow )–to the left, in the picture–
representing the fact that the optimal liquidity response is strictly smaller for any level of capital
the higher the return to investment, which represents the opportunity cost of liquidity. Moreover,


we have depicted this curve for a value of M exceeding M̄ , which is defined as ρ̂ M̄ = ρmin . In this
case, the optimal liquidity response is increasing for all capital levels. Formally, we have that the
capital level for which the optimal liquidity response achieves a maximum is beyond the maximum
possible capital requirement, that is, k̂ (MHigh ) > 1−d. This capital level is never an optimal capital
requirement due to our assumption that the minimum cost of capital is at least as large as ρmin ,
which is precisely set at the value for which the optimal capital requirement hits the boundary,
that is, k ∗ |ρmin = 1 − d. When the return to investment is sufficiently large (M > M̄ ), capital
and liquidity are always complements, as long as the cost of capital is low enough so as to have
a positive capital requirement (i.e., ρ < ρ̄ (M )). The reason is that the optimal liquidity response
for any given capital level is lower the higher the opportunity cost of liquidity. Therefore, as the
opportunity cost of liquidity increases, the optimal liquidity response gets closer to the liquidity
level value that maximizes the expected loan value, so that the weight of the first term of expression
(13)–which measures the effect of the loan value on social welfare and drives liquidity down as it
increases–, gets smaller. Hence, the capital level k̂ (M ) for which the optimal liquidity response
attains a maximum is increasing in M .

itbpF2.9456in1.7746in0inFIG Eq5Figure

Finally, the return to investment does also affect the range for which the optimal capital re-
quirement is positive. In order to see why, consider a given opportunity cost of liquidity M . The
range for which the optimal capital response curve is positive is given by the liquidity segment
(l1 (ρ) , l2 (ρ)), which depends on the cost of capital ρ. The capital requirement is positive if and
only if the optimal liquidity response when capital is zero exceeds the minimum liquidity level for

28
which the optimal capital response is positive, that is, if and only if lR (0) |M > l1 (ρ). Otherwise,
if lR (0) |M ≤ l1 (ρ), we have that the optimal requirements are given by l∗ = lR (0) |M and k ∗ = 0.
Therefore, since l1 (ρ) decreases in ρ, we have that for each M there exists ρ̄ (M ) such that the
capital requirement is positive if and only if ρ < ρ̄ (M ).

5.5 Liquidity and capital: complementary or offsetting tools?

???
——————————————————————————————————

 Hence, liquidity
 is a complementary tool to capital when liquidity is low (formally, when l ∈
l1 (ρ) , ˆl (ρ) ), so that increasing liquidity leads to an increase of the optimal capital response; but
 
ˆ
liquidity offsets capital when liquidity is high (formally, for l ∈ l (ρ) , l2 (ρ) ), so that increasing
liquidity leads to an decrease of the optimal capital response.
(Comparative statics with respect to the cost of capital) For any ρ < ρmax , a reduction of the
cost of equity capital ρ leads to:
(a) an upward shift of the capital response curve kR (l) for any given liquidity level l ∈
(l1 (ρ) , l2 (ρ)) for which the capital response curve is positive.
(b) an enlargement of the range of liquidity levels for which capital response curve are
positive, that is, l1 (ρ) is decreasing and l2 (ρ) is non-decreasing in ρ.
(c) a reduction of the liquidity threshold ˆl (ρ) that determines the extent of complementarity
or substitutability of capital and liquidity, that is, ˆl (ρ) is decreasing in ρ.
——————————————————————————————————
???
—————————————————————————————————–
*********************
Hence, capital complements liquidity when capital is low (formally, when k < k̂ (M )), so that
increasing capital leads to an increase of the liquidity requirement; but capital reduces the benefits
of liquidity when capital is high (formally, when k > k̂ (M )), so that increasing capital leads to a
decrease of the liquidity requirement.
*****************************
(ii) (Comparative statics with respect to the return to investment factor) An increase in the
return to investment factor M leads to:
(a) a reduction of the liquidity requirement lR (k) for any capital level k, that is, lR (k) |M
is decreasing in M for all k.

29
 (b) an increase in the threshold value k̂ (M ) that determines the extent of complementarity
or substitutability of capital and liquidity, that is, k̂ (M ) is increasing in M . Moreover, there exists


a threshold M̄ such that k̂ M̄ = 1 − d. Hence, for any M > M̄ , capital is always a complementary
policy tool to liquidity.
—————————————————————————————————–
???
—————————————————————————————————–
Hence, liquidity complements capital requirements when liquidity is low because reducing illiq-
uidity risk makes insolvency risk reduction more effective. On the contrary, when the amount of
liquidity is high, liquidity offsets the effect of capital in reducing the wedge between the bank’s
private and the regulator’s social goals, so that liquidity makes capital requirements less desirable.
As a consequence, increasing liquidity leads to an elevation of the optimal capital response when
liquidity is low, but to a reduction of the optimal capital response when liquidity is high.
—————————————————————————————————
???
—————————————————————-

Proposition 7 (Optimal joint liquidity and capital requirements) For any given cost of cap-
ital ρ and return to investment (opportunity cost of liquidity) M , we have that:
(ii) (Comparative statics)
(a) (Zero capital requirements for high cost of capital) There exists ρ̄ (M ) < ρmax such that
for any ρ > ρ̄ (M ), the capital requirement is zero.
(b) (Raising the cost of one factor reduces the requirement of that factor) The optimal
liquidity requirement l∗ (ρ, M ) is a strictly decreasing function of M . The optimal capital requirement
k ∗ (ρ, M ) is a decreasing function of ρ (strictly decreasing if and only if ρ < ρ̄ (M )).
(c) (Capital and liquidity complementary tools for low cost of capital and offsetting tools for
high cost of capital)There exists a threshold ρ̂ (M ) < ρ̄ (M ) for the cost of capital such that:
If ρ ∈ (ρ̂ (M ) , ρ̄ (M )), then capital and liquidity requirements are complementary tools:
a raise of the cost of capital ρ leads to an elevation of the liquidity requirement l∗ ; and a raise of
the opportunity cost of liquidity M leads to an elevation of the capital requirement k ∗ .
If ρ ∈ (ρmin , ρ̂ (M )), then capital and liquidity requirements are offsetting tools: a raise
of the cost of capital ρ leads to a reduction of the liquidity requirement l∗ ; and a raise of the
opportunity cost of liquidity M leads to a reduction of the capital requirement k ∗ .
(d) (Capital and liquidity are always complements when the return to investment is high)
The threshold ρ̂ (M ) that determines whether capital and liquidity are complementary or offsetting

30
tools is strictly decreasing on the return to investment M . Moreover, there exists a threshold M̄


such that ρ̂ M̄ = ρmin . Consequently, for any M > M̄ , capital and liquidity are complements for
any cost of capital ρ < ρ̄ (M ).
(e) (Zero capital requirements for high return to investment) The capital cost threshold
ρ̄ (M ) that establishes the maximum cost of capital for which capital requirements are positive de-
creases as the return to investment enlarges, that is, ρ̄ (M ) is strictly decreasing in M .

?????
————————————-

Proposition 8 (ii) [Comparative statics]

(a) [Zero capital requirements for high cost of capital] There exists ρ̄ (M ) < ρmax such that
for any ρ > ρ̄ (M ), the capital requirement is zero.
(b) [Raising the cost of one factor reduces the requirement of that factor] The optimal
liquidity requirement l∗ (ρ, M ) is a strictly decreasing function of M . The optimal capital requirement
k ∗ (ρ, M ) is a decreasing function of ρ (strictly decreasing if and only if ρ < ρ̄ (M )).

6 Regulation with a Lender of Last Resort

So far, we have assumed that the bank fails if the withdrawal of deposits at t = 1 exceeds the
amount of liquid reserves stored by the bank. In this section, we assume that there exists a Lender
of Last Resort (LoLR) that is instructed by the regulator to provide liquidity to a solvent bank
facing a liquidity shortage at t = 1 arising from a deposit withdrawal β > l in excess of its liquid
reserves l. We show that in the presence of an LoLR capital and liquidity become offsetting tools
and derive the ensuing regulatory implications.
We model the LoLR intervention as an agent that has access to a liquidity storage facility at a
per unit opportunity cost γ > 1.24 The LoLR is instructed to provide liquidity assistance to banks
in need of liquid funds to meet deposit withdrawals in excess of its liquid reserves. We make two
assumptions regarding the LoLR intervention. First, the LoLR is constrained to not incur private
losses by providing liquidity assistance to an illiquid bank. Second, as in ?, the LoLR can observe
24
The opportunity cost of the the LoLR funds can be though of as the missed investment opportunities arising
from the need to have immediate access to ready-to-use idle liquid resources. Alternatively, one can think of γ
as the shadow cost of having to expand the monetary base to meet the liquidity needs of a bank facing excessive
withdrawals.

31
whether the bank is solvent or not if called upon to intervene at t = 1.25 An immediate implication
of these two assumptions in the context of this model is that the LoLR can only provide liquidity
assistance if the bank is solvent.26 The second implication is that the LoLR must lend at a unit price
τ ≥ γ that compensates for the opportunity cost γ of liquid funds.27 Finally, the third implication
is that the LoLR provides liquidity assistance only if the bank can repay back the funds, that is,
only if the amount withdrawn does not exceed a certain threshold β to be derived below; effectively,
the LoLR seizes the bank’s loan future cash-flows as collateral to secure the short-term liquidity
loan granted to the bank.28
In particular, the timing of the LoLR intervention is as follows. First, at t = −1, the regulator
announces the unit price τ ≥ γ at which an illiquid but solvent bank can borrow funds from the
LoLR. At t = 1, if the bank faces a deposit withdrawal β > l in excess of its liquidity reserves, the
bank demands liquidity assistance from the LoLR. Upon observation of the quality of the bank’s
assets, the LoLR engages in a secured lending transaction with the bank: The LoLR provides an
amount β − l of liquid funds at t = 1 to the bank in exchange for an amount τ (β − l) to be paid
back by the bank at t = 2, taking the bank’s loan cash-flows as collateral to secure the transaction.
We now derive the bank’s profit function in the presence of the LoLR. First, notice that the
bank’s realized profits remain unchanged with respect to the baseline model in the events in which
the LoLR is not called upon to intervene, that is, whenever β ≤ l. If β > l, the LoLR intervenes
as long as the value of the bank’s collateral is sufficient to pay back the cost τ (β − l) of the LoLR
liquidity facility. In order to derive the maximum withdrawal amount that the bank can meet with
the aid of the LoLR, observe that the LoLR’ intervention reduces the bank’s deposit liabilities by
β. Therefore, the LoLR provides liquidity assistance as long as τ (β − l) ≤ M (1 − l) − D (β, k), or
25
We make this assumption for analytical simplicity only. The results obtained here generalize to a setting in
which the LoLR observes a sufficiently informative signal about the bank’s solvency.
26
The classical LoLR doctrine, first formally laid out by ? and ?, establishes that the LoLR should lend to ”illiquid
but solvent” banks. Abiding by these principles, the ECB’s Emergency Liquidity Assistance (ELA) program is set
to assist a ”solvent financial institution that is facing a temporary liquidity problem”. Here, we take this position
and assume that the LoLR does only intervene when the bank is solvent.
27
The Federal Reserve discount window allows banks with short-run liquidity problems to access a liquidity facility
at preferential rates, these being 100 and 50 basis points over the federal interest rate target, depending on whether
the institution qualifies for primary or secondary prime rates, respectively. Banks in the eurozone can use the
Marginal Lending Facility to access liquidity overnight at a small spread over the standard deposit facility rate.
The BoE has a discretionary policy that establishes that the spread over standard operations should depend on the
counterparty. The Bank of Canada charges market interest rates on its LoLR operations.
28
Access to the discount window for short-term liquidity assistance at the world leading central banks (e.g., the
Federal Reserve, the European Central Bank, the Bank of England, or the Bank of Japan, among others) requires
that the bank provides acceptable collateral to secure the loan.

32
as long as:
M (1 − l) − D (l, k)
β ≤ β (l, k) ≡ l + .
τ −1
We can thus write the bank’s objective function with an LoLR in compact form as:

Z β(l,k)
ΠLoLR
B (θ, l, k) ≡ ΠB (θ, l, k) + θ · [π(l, k) − (τ − 1) (β − l)] dF (β) ,
l
where ΠB (θ, l, k) is the bank’s objective function in the baseline model, as stated in Expression (1).
Analogously, we can write the bank’s profit-maximizing level of solvency as:

1 β(l,k)
Z
LoLR
θB (l, k) = θB (l, k) + (π(l, k) − (τ − 1) (β − l)) dF (β).
c l
Observe that the cost τ at which the bank can access the LoLR funds reduces the bank’s profits
in two respects. On the hand, it reduces the maximum withdrawal threshold β(l, k). On the other
hand, it reduces the bank’s post-intervention profits. As a consequence, the maximum level of bak’s
solvency is achieved when τ is mimimum, that is, when τ = γ. Hence, the optimal policy prescribes
that the LoLR breaks-even in its liquidity assistance intervention. In this case, we have that the
presence of the LoLR increases the bank’s solvency level, that is:
LoLR
θB (l, k) > θB (l, k)M .

The intuition for this result is straightforward: The LoLR allows the bank to survive upon a larger
deposit withdrawal, increasing the value of reducing solvency risk.
Analogously to the baseline model, we have that there exists a unique bank’s profit-maximizing
liquidity level, which the bank would choose in the absence of regulation, that does also maximize
the bank’s solvency. The following proposition formalizes this observation.

Proposition 9 (Bank’s profit maximizing liquidity and solvency) (i) There exists a unique
LolR LolR
bank’s profit-maximizing liquidity level lB (k). Moreover, lB (k) > 0 if and only if γ > M .
LolR
(ii) The bank’s solvency is maximized when the bank’s liquidity is set to lB (k). Formally,
LolR LoLR
lB (k) is the unique maximizer of θB (l, k) for any given k.

Also, as in the baseline model, we can write the wedge between the regulator’s and the bank’s
problems as:

ΠLoLR LoLR
(l, k) , l, k − ΠLoLR


LolR

B θB R (l, k) = 1 − F β(l, k) · θB · D (l, k) . (16)
| {z }
Downside upon failure

Equation (16) is the counterpart of expression (8), which stands for the social cost of a bank
failure in the presence of an LoLR. The regulatory intervention is justified on the basis of the

33
existence of this wedge. The analysis of the optimal capital and liquidity requirement with an
LoLR parallels the one carried out for the baseline model. Instead of providing a detailed account
of all the steps, we state the main finding in the following proposition and highlight the most
important differences with the baseline model below.

Proposition 10 (Joint liquidity and capital regulation with

 a Lender of Last Resort) Defin

 M −1 1
γ̄ implicitely as the unique value satisfying g β̄ (0, 0) 1 − γ̄−1 = 2D(0,0) . Then, it follows that
γ̄ > M and:
(i) For any γ < γ̄, liquidity and capital are offsetting tools, that is, an increase in the requirement
of one of the instrument should be coupled with a decrease in the requirement of the other instrument.
∂lLoLR LoLR

∂kLoLR

Formally, we have that: R∂k lR , kRLoLR < 0 and R∂l LoLR
lR , kRLoLR < 0.
LoLR
(ii) For any γ > M , there is a proper liquidity requirement, that is, lR > 0. Moreover, for
any γ, there is a proper capital requirement, that is, kRLoLR > 0.

In the baseline model, liquidity and capital may either be complementary or offsetting tools.
Increasing capital requirements makes loans safer, which affects the value of liquidity in two re-
spects. On the one hand, storing liquidity carries a higher opportunity cost in terms of the foregone
investment in higher yield loans. On the other hand, keeping liquid assets reduces illiquidity risk,
making it more likely to survive a liquidity shock and cashing the return of loans. However, in
the presence of a LoLR, liquidity and capital are offsetting tools, as long as the LoLR’s access to
liquid funds is not too onerous (i.e., as long as γ < γ̄). The reason is twofold. On the one hand, as
in the baseline model, increasing capital requirements induces a higher degree of solvency, making
loans safer, and thus increasing the opportunity cost of storing liquidity. But, on the other hand,
capital requirements reduce illiquidity risk by increasing the value of the bank’s collateral, and thus
augmenting the range for which the LoLR provides liquidity support. Hence, higher capital require-
ments reduce the value of storing liquidity further. Hence, in the presence of a LoLR increasing
capital requirements calls for an unambiguous reduction of liquidity requirements.
The main result of this section, namely that capital and liquidity are offsetting tools, but are
neverthelss required in positive amounts, requires that γ ∈ (M, γ̄). On the one hand, if the cost of
holding liquidity is too small, that is, if γ < M , liquidity requirements should be zero. In this case,
it is efficient that the bank invests all its resources in loans and that all liquidity needs be provided
by the LoLR on demand. On the other hand, for a sufficiently high opportunity cost of liquidity,
that is, for γ > γ̄, the provision of liquidity by the LoLR becomes prohibitely costly. In this case, the
range of liquidity withdrawals for which the LoLR intervens shrinks as γ increases, thus increasing
illiquidity risk. Indeed, the model with the LoLR continuously approaches the baseline model as γ
grows larger.

34
 It is worth remarking that there are several instruments that a bank can use to avoid liquidation
in the event of facing an abnormally high level of deposit withdrawals. We have analyzed here
the role of a lender of last resort. Alternatively, a bank may borrow in the interbank market, or
liquidate assets (albeit facing haircuts or at fire sales prices). And the bank could also combine
these instruments at its best convenience. Although the details of any of these alternatives differ,
a common regularity emerges from all these analysis: As illiquidity risk is reduced by resorting
to alternative means of obtaining liquidity in the interim, capital and liquidity become offsetting
tools. The underlying reason is the same as in the analysis of this section. An increase in capital
requirements enhances the value of the bank’s assets that can be used to borrow from an LoLR, to
borrow in the interbank market, or to liquidate assets. All these instruments reduce illiquidity risk
and thus make it relatively less valuable to store liquid assets.

35
A Technical assumptions
We start off by defining the level of liquidity lΦ (k) that maximizes the marginal return of capital
for any given value of k, which is implicitly defined by:

Definition 2 (Capital return maximizing level of liquidity )

1
g (lΦ (k)) · D(lΦ (k) , k) ≡ .
2
Next, we define thresholds for the maximum and minimum values of the social cost of capital.

Definition 3 (Upper threshold social cost of capital)

1+r
ρmax ≡ r + F 2 (lΦ (0)) · D (lΦ (0) , 0) · .
c
Definition 4 (Lower threshold social cost of capital)
1+r
ρmin ≡ r + F 2 (lΦ (1 − d)) · D (lΦ (1 − d) , 1 − d) · .
c
In the main text we show that for values ρ ≥ ρmax the optimal capital response is zero for any
given value of liquidity. In addition, in order to limit the number of cases to discuss and avoid a
capital requirement in excess of the maximum level of capital that the bank may hold, we assume
that the social cost of capital is sufficiently high so as to have an interior solution. Formally:

Assumption 1
ρ > ρmin .

We now define a (lower) threshold for the project profitability as follows:

Definition 5 (Lower threshold project profitability)

   2 2 1/2 
F (l Φ (0))·(1−lΦ (0))·D(l Φ (0),1−d)
+

 

c

·


 c  2
 

F (l (0))·(1−l (0))
 
2 Φ Φ
· [(1 + ρ ) · (1 − d) + D(l (0) , 1 − d)]
 
D(lΦ (0) , 1 − d) 
c max Φ

M≡ + .
1 − lΦ (0) F 2 (l (0)) · (1 − l (0))2


 Φ Φ 



 


 

 

36
 The following assumption is a sufficient condition that guarantees that the project profitability
is sufficiently high so that the optimal regulatory policy does not entail shutting down the bank.29

Assumption 2
M ≥ M.

In order to ensure an interior solution for the level solvency we need to impose a condition on the
cost c of reducing insolvency risk. Define the (lower) threshold for the cost of reducing insolvency
risk as:

Definition 6 (Lower threshold insolvency risk reduction cost)

c ≡ F (lB (1 − d)) · π (lB (1 − d) , 1 − d) .

The following asssumption guarantees that insolvency risk is not totally eliminated in equilib-
rium.

Assumption 3
c ≥ c.

We also need to impose a condition so as to ensure that the bank is a ”proper bank” in the sense
that deposits constitute a non-negligible source of funds for the bank. Otherwise, if the bank relies
on too large an amount of non-deposit liabilities, a 100% liquidity requirement may be optimal
whenever eliminating illiquidity risk entailed a negligible cost in terms of foregone loans because of
a small deposit base. In this situation, deposits would simply be stored as liquid assets and would
not play affect the bank’s profits. We define the (lower) threshold for the deposit base as:

Definition 7 (Lower threshold deposits)

29
This is a strong sufficient condition to guarantee that the social value of the bank is not negative, but that we
do otherwise not use to prove our results. ??? Given our choice of M , in equilibrium we have that the social value
generated by the bank is positive. We could therefore relax this assumption so that the social value produced by the
bank is exactly zero. In order to do so, on Definition 5 we could replace the maximum possible equilibrium level of
liquidity lΦ (0) by the actual equilibrium value of liquidity l∗ , as well as the maximum level of capital 1 − d by the
equilibrium value k ∗ . This replacement would reduce the lower threshold for the minimum level of profitability M ,
so that the range of profitability would be enlarged. However, l∗ and k ∗ are determined endogenously in equilibrium,
whereas both lΦ (0) and 1 − d are determined exogenously by the model parametric assumptions. By focusing on
these upper thresholds we are able to provide a explicit expression that depends exclusively on the model parameters.

37
 1
g (d) · (1 − d) ≡ .
2 (1 + r)
Notice that log-concavity of F (·) implies that liml→0 g (l) = +∞, so that we can make g (l) · (1 − l)
arbitrarily large as l approaches 0. Moreover, log-concavity requires that g (d) < +∞, so that
g (l) · (1 − l) can be made arbitrarily close to 0 by choosing values of l sufficiently close to d. In
addition, observe that g (l) · (1 − l) is a strictly decreasing function of l for all l < d, since both g (l)
and (1 − l) are positive strictly decreasing functions of liquidity l. Hence, d is well-defined, unique
and such that d < 1. The following asssumption ensures an interior liquidity requirement, so that
liquidity risk is never eliminatd in equilibrium.

Assumption 4
d ≥ d.

Finally, we need a condition on the maximum value that the interest rate r on long-term deposits
can achieve, so that deposit liabilities do not explode.

Definition 8 (Upper threshold long-term deposits interest rate)

1
r̄ ≡ .
1−d
We then have that:

Assumption 5
r ∈ [0, r̄] .

B Additional technical assumptions for the asset liquida-

tion case
As the base model, let be ˜lΦAL (k) the level of liquidity that maximizes the indirect effect of capital
under asset liquidation for a given value of k, which is implicitly defined as:

1
g(γ + ˜lΦAL (k))(D(˜lΦAL (k), k)) = .
2
Then, we set both an upper and a lower bound for capital cost as follows
 R γ+l̃ΦAL (0) 
˜
(β − lΦAL (0))f (β)
l̃Φ (0)  1+r
ρmax = r + F (γ + ˜lΦAL (0))2 D(˜lΦAL (0), 0) − AL ·

F (γ + ˜lΦAL (0)) c

38
and   1+r
˜2 ˜
ρmin = r + F (γ + lΦ (1 − d))D lΦAL (1 − d), 0 · .
c
The lower bound ρmin ensures that capital requirements never exceed 1 − d, while for any ρ ≥ ρmax
the optimal capital response is zero for any l.
According to this, we assume (Assumptions 1-AL):

ρmin < ρ < ρmax .

Moreover, we impose a minimum cost for the cost of solvency to ensure an interior solvency level
denoted by

c1 = M · F (γ + lELVAL ) · (1 − lELVAL ),

where lELVAL is implicitly defined as g(γ + lELVAL )(1 − lELVAL ) = 1.

χ(l)
Moreover, in order to guarantee that g(γ + lRAL (k))(1 − lRAL (k)) − 1 + ψ
< 0, we define

1 + r(1 − d)
c2 = 1+F (lELVAL )
.
F (lELVAL )
+ g(γ + lELVAL )γ − 1

Then, we assume (Assumptions 2-AL):

c ≥ c = M ax[c1 , c2 ].

In order to ensure that the bank always has enough asset in place to liquidate at least a fraction
γ
ψ
, we set

γ1 = (d − ˜lΦAL (0))ψ.

On the other hand, we implicitly define γ2

g(γ2 )π(0, 0) = M − 1

to ensure that θBAL (l, k) is a hump-shaped function of l.

Then, we define γ as
γ = M in[γ1 , γ2 ],

and assume that (Assumptions 3-AL):

γ ≤ γ.

39
 Finally, we implicitly define d as
1
g(γ + lELVAL ) (d − lELVAL ) = .
2
Then, we assume (Assumptions 4-AL):

d > d.

C Omitted Proofs
Before proceeding to the proof of the Propositions in the main body of the text, we show the
following instrumental result, which will be useful in the proofs of Propositions 4 and 5.
f (l)
Fact 1 (Properties of Φ (l, k) curve) Define Φ (l, k) ≡ 2g (l) · D (l, k) − 1, where g (l) ≡ F (l)
and
D (l, k) ≡ 1 + (1 − d) · r − (1 + r) · k − l, as defined in Sections 3.1 and 3, respectively. Let lΦ (k) be
implicitly defined as Φ (lΦ (k) , k) = 0. Then, the functions Φ (l, k) and lΦ (k) satisfy the following
properties:
(i) For all k ∈ [0, 1 − d], liml→0 Φ (l, k) = +∞ and Φ (d, k) < 0.
(ii) For all k ∈ [0, 1 − d], Φ (l, k) is strictly decreasing in l.
(iii) For all l ∈ (0, d], Φ (l, k) is strictly decreasing in k.
(iv) The function lΦ (k) is strictly decreasing. Moreover, lΦ (0) ∈ (0, d).

Proof of Fact 1. (i) The first property, liml→0 Φ (l, k) = +∞, follows from the fact that
F (·) is log-concave, so that liml→0 g (l) = +∞. The second property, namely that Φ (d, k) =
2g (d) · [(1 + r) · (1 − d − k)] − 1 < 0, follows directly from Assumption 4 on Section A.
(ii) Both g (l) and D (l, k) are positive and strictly decreasing functions, so that the product
g (l) · D (l, k) is also positive and strictly decreasing. The result is then immediate.
∂D(l,k)
(iii) This property follows immediately from ∂
= − (1 + r) < 0.
(iv) The locus lΦ (k) satisfies Φ (lΦ (k) , k) = 0. Implicitly differentiating this expression with
∂lΦ (k)
respect to k yields ∂k
= − ∂Φ(l,k)
∂k
/ ∂Φ(l,k)
∂l
< 0, the last inequality following from properties (ii)
and (iii). We now show that lΦ (0) < d. From from Property (ii) in Fact 1, we have that Φ (l, 0)
is strictly decreasing in its first argument. Also, Φ (l, 0) is continuous in its first argument as well.
Moreover, liml→0 Φ (l, 0) = +∞. In addition, it follows from item (i), just proved, that Φ (d, 0) < 0.
Hence, it follows that there exists a unique lΦ (0) ∈ (0, d) that solves Φ (lΦ (k) , k) = 0.

Fact 2 (Properties of the regulator’s objective function) The regulator’s objective function
ΠR (l, k) satisfies:

40
 (i) For any given l ∈ [0, d], ΠR (l, k) is strictly concave in k in all of its domain.
(ii) For any given k ∈ [0, ], ΠR (l, k) is strictly quasiconcave in l in the domain [lB (k) , lΦ (k)].

Remark 1 For a family of log-concave distribution functions, the function ΠR (l, k) is strictly con-
vex in l around zero. Hence, ΠR (l, k) is not concave in l in all of its domain. However, ΠR (l, k) is
quasiconcave in l in all of its domain. We restrict ourselves to showing that ΠR (l, k) is quasiconcave
in l in the domain [lB (k) , lΦ (0)] because this weaker statement suffices for our purposes and the
proof is (much) less involved.

Proof of Fact 2. (i) For any given l ∈ [0, d], we have that
∂ 2 ΠR (l, k) F 2 (l) · (1 + r)2
= − < 0.
∂k 2 c
(ii) We prove this property through a succession of steps through which we show that, for any
dΠR (lR (k),k)
given k, there exists a unique value of liquidity lR (k) ∈ (lB (k) , lΦ (0)) such that dl
= 0.
Consider an arbitrary k ∈ [0, 1 − d] and, for the sake of notational simplicity, rename the first
derivative of the bank and the regulator’s objective functions, respectively, as follows:
dΠB (l, θ, k)
HB (l, θ, k) ≡ ,
dl
and
dΠR (l, k)
HR (l, k) ≡ .
dl
Proof.

Step 1 HR (lB (k) , k) > 0, where lB (k) is the unique value satisfying HB (lB (k) , k) = 0.

We can write the regulator’s objective function in terms of the bank’s as follows:

ΠR (l, k) = ΠB (l, θB (l, k) , k) − (1 − θB (l, k) · F (l)) · D (l, k) .

We can therefore write the regulator’s first derivative w.r.t. liquidity as:
d
HR (l, k) = HB (l, θB (l, k) , k) − [(1 − θB (l, k) · F (l)) · D (l, k)] .
dl
From the bank’s FOC w.r.t. to liquidity, we have that HB (lB (k) , θB (lB (k) , k) , k) = 0. Moreover,
∂
from Proposition 1, it follows that θ
∂l B
(lB (k) , k) = 0. Hence, we can write

HR (lB (k) , k) = 1 + θB (lB (k) , k) · F (lB (k)) · [g (lB (k)) · D (lB (k) , k) − 1]

Notice that g (lB (k)) · D (lB (k) , k) ≥ 0. Hence g (lB (k)) · D (lB (k) , k) − 1 ≥ −1. Moreover,
θB (lB (k) , k) · F (lB (k)) < 1 (both θB (·, ·) and F (·) are smaller or equal to 1, the former always
strictly smaller). Hence, we have that HR (lB (k) , k) > 0.

41
Step 2 HR (lΦ (0) , k) < 0 where, given Φ (l, k) = 2g (l) · D (l, k) − 1 as defined in Fact 1, lΦ (k) is
the unique value satisfying Φ (lΦ (k) , k) = 0.

Straightforward algebra leads to:

!
F (lΦ (0))2 M · π (lΦ (0) , k) · [g(lΦ (0)) · (1 − lΦ (0)) − 1]
HR (lΦ (0) , k) = · + 1.
c +D (lΦ (0) , k) · [g(lΦ (0)) · π (lΦ (0) , k) − M + 1]
∂π(lΦ (0),k)
Recall that π (l, k) ≡ M · (1 − l) − D (l, k). Abusing notation, we have that ∂M
= 1 − lΦ (0).
Recognizing that HR (lΦ (0) , k) depends on M both directly and through π (lΦ (0) , k), we can then
derive the following expression:
dHR (lΦ (0) , k) F (lΦ (0))2
= 2M · · (1 − lΦ (0)) · [g(lΦ (0)) · (1 − lΦ (0)) − 1] .
dM c
By construction of lΦ (0), we have that g (lΦ (0)) · D (lΦ (0) , k) = 1/2 < 1. Hence, it follows that
∂HR (lΦ (0),k)
g(lΦ (0)) · (1 − lΦ (0)) − 1 < 0. Hence, we have that ∂M
< 0. Hence, there exists M such that
HR (lΦ (0) , k) |M < 0 for all M > M . Tedious but straightforward algebra shows that M < M , as
defined in Definition 5.
 
Step 3 HR ˆl, k = 0 for some ˆl ∈ (lB (k) , lΦ (0)).

It follows directly from Steps 1 and 2 and the continuity of HR (l, k) in l.

∂HR (l̂,k)
 
Step 4 For any ˆl ∈ (lB (k) , lΦ (0)) such that HR ˆl, k = 0, it follows that ∂l
< 0.

Define the following function:

χ (l) ≡ M · π (l, k) · [g(l) · (1 − l) − 1] + D (l, k) · [g(l) · π (l, k) − M + 1] .

We can then write the regulator’s first derivative w.r.t. liquidity as follows:
F (l)2
HR (l, k) = · χ (l) + 1.
c
By construction of ˆl (Step 3), we have that:
  F (ˆl)2 
HR ˆl, k = · χ ˆl + 1 = 0.
c

Hence, it follows that χ ˆl < 0.
The first derivative of HR (l, k) w.r.t. l can be written, for l > 0, as:
F (l)2
 
∂HR (l, k) ∂χ (l)
= · 2g (l) · χ (l) + .
∂l c ∂l

42
 ∂χ(l̂)

Since χ ˆl < 0, recognizing that F (l) > 0 and g (l) > 0 for all l, it suffices to show that ∂l
< 0.
We can write:
∂χ (l) ∂g (l)
= M 2 · [1 − 2g (l) · (1 − l)] + M 2 · (1 − l) − D (l, k) · (1 − l) ·
 
∂l ∂l
+ [g (l) · (1 − l) − 1] + [D (l, k) · g (l)] ≤
∂g (l)
M 2 − 1 · (1 − 2g (l) · (1 − l)) + M 2 · (1 − l) − D (l, k) · (1 − l) ·

  
≤
∂l
By definition of lΦ (0), we know that 1−2g (lΦ (0))·(1 − lΦ (0)) = 0. Recognizing that g (l)·(1 − l)
is strictly decreasing in l (recall that both 1−l and g (l) are strictly decreasing and positive
  functions,

the latter because of log-concavity of the distribution function), it follows that 1 − 2g ˆl · 1 − ˆl <
0. Clearly, M 2 − 1 > 0. Moreover, since M > 1, we have that M 2 · (1 − l) − D (l, k) > M · (1 − l) −
∂g(l)
D (l, k) = π (l, k) > 0. From log-concavity of the distribution function, we have that ∂l
< 0 for
∂χ(l̂)
all l. Hence, it follows that ∂l < 0.

Step 5 There exists a unique ˆl ∈ (lB (k) , lΦ (0)), which we dub lR (k), such that HR (lR (k) , k) = 0.

ˆ
  for the sake of contradiction, that there is more than one l ∈(lB (k)
Assume,  , lΦ (0))
 such
 that
HR ˆl, k = 0. Let ˆl1 , ˆl2 , with ˆl1 < ˆl2 be the first two values satisfying HR ˆl1 , k = HR ˆl2 , k = 0.
∂χ(l̂2 )
From continuity of HR (l, k) with respect to l, it follows that ∂l > 0, which contradicts the
statement shown in Step 4.
Proof of Proposition 4. Since, for any given l, ΠR (l, k) is strictly concave in k in all of
its domain, the first order necessary condition for an interior optimal capital requirement is also
sufficient. Consider the first order condition of the social welfare objective function with respect to
capital:
dΠR (l, k) ∂ΠR (l, k) ∂θB (l, k) ∂ΠR (l, k)
= · + ,
dk ∂θ ∂k ∂k
which we can write as follows:
dΠR (l, k) 1+r
= · F 2 (l) · D (l, k) − (ρ − r) .
dk c
dΠR (l,kR (l))
Take any given set of parameters c, r and ρ. Fix any l ∈ [0, d]. Then, kR (l) > 0 only if dk
= 0.
(ρ−r)·c dΠR (l,k)
Hence, if kR (l) > 0 we have that Υ(l, kR (l)) = 1+r
. Moreover, if dk
< 0 for all k ∈ [0, 1 − d],
then kR (l) = 0. The proof consists of comparing Υ(l, k) ≡ F (l) · D (l, k) with (ρ−r)·c
2
1+r
. In order to
do this comparison, we first show that Υ(l, k) satisfies the following properties.

Fact 3 The function Υ(l, k) ≡ F 2 (l) · D (l, k) satisfies the following properties.

43
 (a) For any given k ∈ [0, 1 − d], Υ(l, k) is strictly increasing in l for l < lΦ (k) and strictly
decreasing in l for l > lΦ (k), where lΦ (k) was implicitly defined as Φ (lΦ (k) , k) = 0 in Fact 1.
(b) For any given l ∈ (0, d], Υ(l, k) is strictly decreasing in k.
(c) Υ(lΦ (k) , k) is strictly decreasing in k.
Proof of Fact 3. (a) Fix k, and take the first derivative of Υ(l, k) with respect to l to get
∂Υ(l,k) ∂Υ(l,k)
∂l
= F (l) · Φ (l, k), where Φ (l, k) is as defined on Fact 1. Since F 2 (l) > 0, the sign of
2
∂l
is determined by Φ (l, k). From Property (i) in Fact 1 and the definition of lΦ (k), we have that
∂Υ(l,k) ∂Υ(d,k)
∂l
> 0 for all l ∈ (0, lΦ (k)). Also, from Property (i) in Fact 1, we have that ∂l
< 0.
∂Υ(l,k)
Moreover, from Property (ii) in Fact 1, it follows that ∂l
< 0 for all l ∈ (lΦ (k) , d). Hence, the
result follows.
(b) This statement follows directly from first differentiating Υ(l, k) with respect to k, to obtain
∂Υ(l,k)
∂k
= −F 2 (l) · (1 + r) < 0.
(c) Take any two k1 < k2 . From (b) it follows that, for any given l ∈ (0, d], Υ(l, k1 ) > Υ(l, k2 ).
Also, by definition of lΦ (k), we have that Υ(lΦ (k1 ) , k1 ) ≥ Υ(l, k1 ) for all l ∈ (0, d]. Combining
these two inequalities, we have that Υ(lΦ (k1 ) , k1 ) > Υ(l, k2 ) for any given l ∈ (0, d]. In particular,
the last inequality holds true for l = lΦ (k2 ). Hence, we can write Υ(lΦ (k1 ) , k1 ) > Υ(lΦ (k2 ) , k2 ).
Now we show the Proposition statements.
(ρmax −r)·c
(i) Let ρmax be the unique value satisfying Υ (lΦ (0) , 0) = 1+r
. Observe that, for any given
k, Υ(l, k) is maximized at lΦ (k) (Property (a)); also, for any given l, Υ(l, k) is maximized at k = 0
(Property (b)). Hence, Υ(l, k) is jointly maximized for (l, k) = (lΦ (0) , 0). Hence, by construction
(ρ−r)·c
of ρmax , it follows that Υ (l, k) < 1+r
for all (l, k), as we wanted to show.
(ii) Now, let ρ < ρmax . We prove this result in a succession of steps.

Step 6 The optimal capital response at l = 0 is zero, i.e., kR (0) = 0.

(ρ−r)·c
Observe that Υ (0, k) = 0 < 1+r
for all k ∈ [0, 1 − d]. Hence, we have that kR (0) = 0.

(ρ−r)·c
Step 7 There exists a value l1 (ρ) ∈ (0, d) such that Υ (l1 (ρ) , 0) = 1+r
. Moreover, Υ (l, 0) <
(ρ−r)·c (ρ−r)·c
1+r
for l < l1 (ρ) and Υ (l, 0) > 1+r
for l > l1 (ρ).

(ρ−r)·c
First, recall from the previous statement that Υ (0, 0) = 0 < 1+r
. Moreover, by construction
(ρmax −r)·c (ρ−r)·c
of ρmax , which is such that Υ (lΦ (0) , 0) > 1+r
, we have that Υ (lΦ (0) , 0) > 1+r
, where
lΦ (0) ∈ (0, d) (From Property (iv) in Fact 1). Moreover, it follows from Property (a) in Fact 3,
that Υ(l, k) is strictly increasing in l for l < lΦ (0). Hence, by continuity of Υ(l, k) in its first
argument, it follows that there exists a unique value for liquidity, which we dub l1 (ρ), such that
(ρ−r)·c
Υ (l1 (ρ) , 0) = 1+r
. It also follows that Υ (l, 0) < 0 for l < l1 (ρ) and Υ (l, 0) > 0 for l > l1 (ρ).

44
Step 8 The optimal response for l ≤ l1 (ρ) is zero, i.e., kR (l) = 0 for any l ≤ l1 (ρ).

We have that Υ (l, k) is strictly decreasing in k for all l > 0 (Property (b) in Fact 3). Also, we
know from Step 7 that Υ (l, 0) < 0 for l < l1 (ρ). Hence, it follows that Υ (l, k) < 0 for any l < l1 (ρ)
and any k ∈ [0, 1 − d].

Step 9 There exists a unique liquidity maximizer ˆl (ρ) ∈ (l1 (ρ) , d) of the capital response func-
ˆ ˆ ˆ ˆ
 that is, kR(l (ρ)) > kR (l) for all l ̸= l (ρ). Moreover, l (ρ) and kR (l (ρ) are defined by
tion,
Φ ˆl (ρ) , kR (ˆl (ρ) = 0.

(ρ−r)·c
By construction of ρmax , we have that Υ (lΦ (0) , 0) > 1+r
. From Property (c) in Fact 3, it
follows that thereexists ′
 a unique k > 0 such that Υ (lΦ (k ′ ) , k ′ ) = (ρ−r)·c
1+r
. Let ˆl (ρ) = lΦ (k ′ ). Hence,
we have that kR ˆl (ρ) = k ′ . Now, consider k ′′ > k . Then, using again Property (c) in Fact 3, it
′

(ρ−r)·c
follows that Υ (lΦ (k ′ ) , k ′′ ) < 1+r
. Now, from Property (iv) in Fact 1, we have that lΦ (k) is strictly
decreasing in k. Hence, since k > k ′ , we have that lΦ (k ′′ ) < lΦ (k ′ ). Also, from Property (a) in
′′

Fact 3, we know that Υ (l, k) is strictly increasing in l for l < lΦ (k). Hence, since lΦ (k ′′ ) < lΦ (k ′ ),
′′ ′′ ′ ′′ ′′ ′′ (ρ−r)·c
it follows
  that Υ (lΦ (k ) , k ) < Υ (lΦ (k ) , k ). Consequently, Υ (lΦ (k ) , k ) < 1+r . Hence,
kR ˆl (ρ) = k ′ is indeed the maximum capital response, which is achieved at ˆl (ρ) = lΦ (k ′ ). Finally,
 
from construction of lΦ (k ′ ), it follows that Φ ˆl (ρ) , kR (ˆl (ρ) = 0.

Step 10 The function Φ (l, kR (l)) is such that Φ (l, kR (l)) > 0 if l < ˆl (ρ).
 
From Step 9, we know that Φ ˆl (ρ) , kR (ˆl (ρ) = 0. Consider l < ˆl (ρ). We also know from Step
9 that kR (ˆl (ρ)) > kR (l). From Properties (ii) and (iii) in Fact 1, we know that  Φ (l, k) is strictly
decreasing in k and in l. Hence, it follows that Φ (l, kR (l)) > Φ ˆl (ρ) , kR (ˆl (ρ) = 0 for l < ˆl (ρ).
 
Step 11 For any l ∈ l1 (ρ) , ˆl (ρ) , we have that the optimal capital response increases in l, that
is, kR (l) strictly increasing in l.

(ρ−r)·c
From Step 7, we have that Υ (l, 0) > 1+r
for l > l1 (ρ). Hence, we have a interior solution,
(ρ−r)·c
that is, kR (l) is such that Υ (l, kR (l)) = 1+r
. From Property (b) in Fact 3, we know that Υ (l, k)
is strictly decreasing in k. Hence, there exists a unique capital level, which we label kR (l), such that
(ρ−r)·c ∂kR (l)
Υ (l, kR (l)) = 1+r
. Now, we show that kR (l) is strictly increasing by showing that ∂l
> 0.
(ρ−r)·c ∂kR (l)
Implicitly differentiating Υ (l, kR (l)) = 1+r
with respect to l it follows that
> 0 if and only ∂l
∂Υ(l,kR (l))
∂l
> 0, that is, if and only if Φ (l, kR (l)) > 0. Hence, it follows from Step 10 that if l < ˆl (ρ)
∂kR (l)
we have ∂l
> 0.

45
  
ˆ
Step 12 For any l ∈ l (ρ) , l2 (ρ) , we have that the optimal capital response decreases in l, that
is, kR (l) strictly decreasing in l.
  
From construction of kR (l), we have that Υ ˆl (ρ) , kR ˆl (ρ) = (ρ−r)·c 1+r
. Moreover, since Υ (l, k)
 
is strictly decreasing in k, it follows that Υ ˆl (ρ) , 0 > (ρ−r)·c 1+r
. By continuity of Υ (l, k) with
respect to its first argument, it follows that there exists ˜ ˆ (ρ−r)·c
  l2 (ρ) > l (ρ) such that Υ (l, 0) > 1+r
for all l ∈ ˆl (ρ) , ˜l2 (ρ) . Hence, for all l ∈ ˆl (ρ) , ˜l2 (ρ) , we have that kR (l) > 0. Hence, for all
   
l ∈ ˆl (ρ) , ˜l2 (ρ) , kR (l) is given by Υ (l, kR (l)) = (ρ−r)·c . Consider l ∈ ˆl (ρ) , ˜l2 (ρ) and let kR (l)
 1+r 
be such that Υ (l, kR (l)) = (ρ−r)·c . Recall that Υ ˆl (ρ) , kR (ˆl (ρ) = (ρ−r)·c . From Property (a) in
1+r 1+r
     
ˆ ˆ ˆ (ρ−r)·c
Fact 3, we have that Υ l, kR (l (ρ) < Υ l (ρ) , kR (l (ρ) = 1+r for all l ∈ l (ρ) , l2 (ρ) . Now, ˆ ˜
 
since Υ (l, k) strictly decreasing in k, it follows that kR (l) < kR ˆl (ρ) . Hence, we have shown that
     
kR (l) < kR ˆl (ρ) for l ∈ ˆl (ρ) , ˜l2 (ρ) . Now, consider any l′ ∈ ˆl (ρ) , ˜l2 (ρ) such that l′ > l.
Then, by the same token, we have that kR (l′ ) < kR (l).
Finally, to complete this statement of the proof, consider the following two cases. First, assume
that Υ (d, 0) ≥ (ρ−r)·c . Then, it follows that d = ˜l2 (ρ). In this case, we have that l2 (ρ) = d. Now
1+r
(ρ−r)·c
suppose, on the contrary, that Υ (d, 0) < 1+r
. In this case, we have that there exists l2 (ρ) < d
(ρ−r)·c (ρ−r)·c
such that Υ (l2 (ρ) , 0) = 1+r
. Then, for any l ∈ (l2 (ρ) , d), we have that Υ (l, 0) < 1+r
, in
which case kR (l) = 0.
(iii) For the remainder of the proof, let ρ′ < ρ. Recall from Step 7 above that l1 (ρ) is defined
(ρ−r)·c (ρ′ −r)·c
as Υ (l1 (ρ) , 0) = 1+r
. Then, we have that Υ (l1 (ρ) , 0) > 1+r
= Υ (l1 (ρ′ ) , 0), where the first
inequality follows from ρ′ < ρ and the last inequality follows from definition of l1 (ρ′ ). Since Υ (l, 0)
is strictly increasing in l for any l < lΦ (0) (Property (a) in Fact 3) and l1 (ρ) < lΦ (0), the result
follows.
Now, if l2 (ρ) = d, we have argued in the last statement of the proof of item (ii) that Υ (d, 0) ≥
(ρ−r)·c (ρ′ −r)·c
1+r
. Hence, it follows that Υ (d, 0) > 1+r
. Hence, l2 (ρ) = d. Now, suppose that l2 (ρ) < d, in
(ρ−r)·c (ρ′ −r)·c
which case Υ (l2 (ρ) , 0) = 1+r
. It then follows that Υ (l2 (ρ) , 0) > 1+r
. Since Υ (l, 0) is strictly
decreasing in l for any l > lΦ (0) (Property (a) in Fact 3) and l2 (ρ) > lΦ (0), the result follows.
(ρ−r)·c (ρ′ −r)·c
Finally, consider l ∈ (l1 (ρ) , l2 (ρ)). Then, we have that Υ (l, kR (l)) = 1+r
> 1+r
. Hence,
since Υ (l, k) is strictly decreasing in k, it follows that kR (l) must increase so as to meet the last
inequality
Proof of Proposition 5. Since, for any given k, ΠR (l, k) is strictly quasiconcave in l in the domain
[lB (k) , lΦ (0)], the first order necessary condition for an interior optimal liquidity requirement is
also sufficient.
(i) Any liquidity requirement such that l ≤ lB (k) is non-binding, so that the bank would choose

46
 dΠR (lB (k),k)
lB (k) if the liquidity requirement is smaller. This requirement is not optimal, since dl
>0
(Step 1 in the Proof of Fact 2). Moreover, we have that there exists a unique value of liquidity
dΠR (lR (k),k)
lR (k) ∈ (lB (k) , lΦ (0)) such that dl
= 0 (Step 5 in the Proof of Fact 2). Hence, the liquidity
requirement lR (k) is such that lR (k) > lB (k).
(ii) ???
In order to prove this result, we show that there exists a value of capital k̂(M ) such that, for
∂lR (k) ∂lR (k)
any k < k̂(M ), it follows that ∂k
> 0, while for any k > k̂(M ), we have that ∂k
< 0. We start
by calculating the slope of lR (k), which is given by:
∂ 2 Π (l,k)
R
∂lR (k)
= − ∂ 2 Π∂k∂l(l,k) .
∂k R
2 ∂l

2
From the regulator’s problem second order condition with respect to l, we know that ∂ Π∂lR2(l,k) < 0,
∂lR (k)
so that the sign of is given by the second order cross derivative, which in turn is given by:
∂k
∂ 2 ΠR (l, k) (1 + r)F (l)2
= [2g(l)D(l, k) − 1] .
∂k∂l c

∂ 2 ΠR (l,k) ∂lR (k)

Note that the term in brackets determines the sign of ∂k∂l
, and therefore the sign of .
∂k
∂lR (k)
Particularly, if the second order cross derivative equals zero, then we have that = 0. Let us
∂k
now define:
Φ(l, k) ≡ [2g(l)D(l, k) − 1] = 0, (17)
∂ 2 ΠR (l,k)
as the loci of k and l such that ∂k∂l
= 0. Note that the term inside brackets decreases with
capital and liquidity. Hence, if the the pair (lR (k), k) lies below the curve Φ(l, k) evaluated at
∂lR (k)
(lR (k), k), then we have that 2g(lR (k))D(lR (k), k) − 1 > 0 and that > 0, as otherwise
∂k
∂lR (k)
< 0. Since, for k = 0, we have that Φ(lR (0) , 0) > 0, while for k = 1 − d we have that
∂k
Φ(tlR (1 − d) , 1 − d) < 0, the continuity
 of lR (k) guarantees that there exists at least a value of
equity capital k̂(M ) such that Φ lR (k̂(M )), k̂(M ) = 0.
We now prove that k̂(M ) is unique. We prove this statement by showing that the curve lR (k)
∂lR (k)
never crosses the curve Φ(l, k) from above. Hence, ∂k
cannot change its sing from negative
to positive. By the sake of contradiction, assume that there exists a value k̂ ′ > k̂(M ) such that
∂lR (k̂′ )
lR (k) crosses Φ(l, k) at (lR (k̂ ′ ), k̂ ′ ) from above. This implies that ∂l
< 0. However, since
∂lR(k̂′ )
Φ(lR (k̂ ′ ), k̂ ′ ) = 0, the slope of lR at k̂ ′ must be ∂k
= 0. Then, there is no value of k > k̂(M ) such
∂lR (k)
that ∂k
> 0. We thus conclude that lR (k) increases for k < k̂(M ) and decreases for k > k̂(M ).
Hence, lR (k) is a hump-shaped function of capital.

47
 ————————————————————————————
——————————————————————————————————————————
-
???
(i) Let lB (k) and lR (k) stand for the bank’s profit-maximizing level of liquidity and the reg-
ulator’s liquidity choice, respectively, as defined in the main body of the text. We can write the
regulator’s FOC w.r.t. to liquidity as follows:
dΠR (lR (k) , k) ∂ΠR (lR (k) , k) ∂ΠR (lR (k) , k) ∂θB (lR (k) , k)
= + · = 0.
dl | ∂l
{z } | ∂θB | ∂l
{z }
{z }
Direct Ef f ect>0 >0 <0
| {z }
Indirect Ef f ect<0

———————————————————————————————————————
————————————————————————————–
???
(i) Let lB (k) and lR (k) stand for the bank’s profit-maximizing level of liquidity and the reg-
ulator’s liquidity choice, respectively, as defined in the main body of the text. We can write the
regulator’s objective function in terms of the bank’s as follows:

ΠR (l, k) = ΠB (l, θB (l, k) , k) − (1 − θB (l, k) · F (l)) · D (l, k) .

We can therefore write the regulator’s first derivative w.r.t. liquidity as folllows:
∂ΠR (l, k) ∂ΠB (l, θB (l, k) , k) ∂
= − [(1 − θB (l, k) · F (l)) · D (l, k)] .
∂l ∂l ∂l
∂ΠB (lB (k),θB (lB (k),k),k)
From the bank’s FOC w.r.t. to liquidity, we have that ∂l
= 0. Moreover, from
∂
Proposition 1, it follows that θ
∂l B
(lB (k) , k) = 0. Hence, we can write

∂ΠR (lB (k) , k)

= 1 + θB (lB (k) , k) · F (lB (k)) · [g (lB (k)) · D (lB (k) , k) − 1]
∂l
Notice that g (lB (k))·D (lB (k) , k) is non-negative. Hence, g (lB (k))·D (lB (k) , k) ≤ −1. Moreover,
θB (lB (k) , k) · F (lB (k)) < 1 (both θB (·, ·) and F (·) are smaller or equal to 1, the former always
∂ΠR (lB (k),k)
strictly smaller). Hence, we have that ∂l
> 0. Moreover, from Assumption 4, it follows
∂ΠR (d,k) ∂ΠR (l,k)
that ∂l
< 0. The result follows from continuity of ∂l
w.r.t. l.
(ii)
∂ΠR (d,k) ∂ΠB (d,θB (d,k),k)
Moreover, we have that ∂l
< 0, which follows from the fact that ∂l
< 0 (Propo-
sition 1) and
∂ΠR (l,k) ∂ΠR (d,k)
Recognizing that ∂l
is continuous in l and that ∂l
< 0, the result follows.

48
 ————————————————————————————————————————-
——————————————————————————————-
???
(ii) Liquidity requirement is binding

We want to show that regulator’s liquidity requirement is binding for the bank, i.e., lR (k) >
lB (k). In order to do this, we assume that regulator only chooses l taking k as given. Then, we are
going to compare liquidity’s F.O.C. for the bank and regulator. Recall that:
∂Πb F (l)π(l, k)
(lB , k) = g(l)π(l, k) + 1 − M = 0, θB (l, k) = .
∂l c
Additionally, we can rearrange the regulator’s maximization problem as follows:
F (l)2 π(l, k)2 F (l)2 π(l, k)
 
max ΠR (l, k) ≡ − 1− D(l, k) − (1 + ρ)k.
0≤l≤d 2c c
Then, the corresponding F.O.C. with respect to l is given by:
F (lR )2
 
∂ΠR π(l R , k) [g(l R )π(lR , k) + 1 − M ] +
(lR , k) =  

2
c  2 =0

∂l F (lR ) π(lR , k) F (lR )
1− + [2g(lR )π(lR , k) + 1 − M ] D(l, k)
c c

Then, we evaluate this expression at the bank’s optimal liquidity, lB (k), and obtain:

F (lB )2
 
∂ΠR π(lB , k) [g(lB )π(lB , k) + 1 − M ] +
(lB , k) =  

2
c  2  > 0.

∂l F (lB ) π(lB ) F (lB )
1− + [2g(l)π(lB , k) + 1 − M ] D(lB , k)
c c

The first term on the Right Hand Side (RHS) equals zero, since the term in brackets corresponds
to the bank’s F.O.C. with respect to liquidity. By the same token, the third term on the RHS is
positive. Finally, using the bank’s F.O.C. for monitoring, we know that the second term on the
∂ΠR
RHS is positive as well. This means that ∂l
(lB ) > 0. This finding, coupled with the fact that
∂ 2 ΠR
∂l2
(lB ) < 0 , implies that lR (k) > lB (k).

—————————————————————————————————————————–
???
(iii) Comparative statics with respect to the profitability factor
∂lR (k)
(a) We now prove that ∂M
< 0.
Using the implicit function theorem, we know that:
∂ 2 Π (M,k)
R
∂lR (M, k)
= − ∂ 2 Π∂M(M,k)
∂l
< 0,
∂M R
2
∂l

49
which follows from the fact that:
∂ 2 ΠR F (lR )2
= (D(lR , k) + M + π(lR , k)) (g(lR )(1 − lR ) − 1) < 0,
∂M ∂l c
because g(lR )(1 − lR ) < 1.30
(b) The fact that k̂(M ) satisfies Φ(lR (k̂(M )), that k̂(M )) = 0 and that Φ is decreasing in both
∂lR (k)
l and k implies that k̂(M ) is an increasing function of M . This is due to the fact that ∂M
< 0.
Hence, as M increases, the optimal level of liquidity decreases. Moreover, k̂(M ) has to increase so
to satisfy Φ(lR (k̂(M )), k̂(M )) = 0. Furthermore, for:
 0.5
(d − l)2 c(d − l)
+
 2 F (l)2

M =  ,
 
2
 (1 − l) 
(1 − l) · (d − l) −
2

we have that k̂(M ) = 1 − d.

——————————————————————————————————
Proof of Proposition ??. (i) (Uniqueness) First, we show that the regulator’s maximization
problem satisfies the condition for the Theorem of the Maximum (?) for quasiconcave objective
functions over convex-valued sets to apply. Let M take a fixed value. Define the set of possible joint
liquidity and capital optimal requirements Φ = [0, d] × [0, 1 − d] and the set of all possible values of
the cost of capital Γ = [ρmin , ρmax ]. Define the real-valued function Π̃R : Φ × Γ → R as Π̃R (l, k, ρ) ≡
ΠR (l, k), as defined in equation (6), where Π̃R (l, k, ρ) is simply a relabeling of the regulator’s
objective function to let it depend not only on k and l, but also on the cost of capital parameter
ρ. Moreover, define the mapping Φ (ρ) : Γ → Φ of possible joint liquidity and capital optimal
requirements for each value of the cost of capital ρ ∈ Γ. That is, we have that Φ (ρ) = [0, d]×[0, 1 − d]
for all ρ ∈ Γ. Let the pair (l∗ (ρ, M ) , k ∗ (ρ, M )) ∈ arg max(l,k)∈Φ(ρ) Π̃R (l, k, ρ) be the optimal
liquidity and capital requirements, for a given fixed M , and let Π̃∗R (ρ) = Π̃R (l∗ (ρ, M ) , k ∗ (ρ, M ) , ρ)
be regulator’s value for each ρ ∈ Γ. First, we have that Π̃R (l, k, ρ) is jointly continuous in all its
arguments. Moreover, Π̃R (l, k, ρ) is strictly quasiconcave in (l, k) for each ρ ∈ Γ (we have shown
above that ΠR (l, k) is strictly concave in k and single-peaked–hence quasiconcave–in l; proving
that ΠR (l, k) is quasiconcave in (l, k) entails computing the bordered Hessian of ΠR (l, k), which
is tedious but straightforward). Also, Φ (ρ) is a compact-valued, convex-valued and continuous
correspondence for all ρ ∈ Γ, because it is an invariant compact and convex set for all ρ ∈ Γ. Hence,
by the Theorem of the Maximum for quasiconcave functions over convex-valued sets, it follows that
30
Recall that the regulator always chooses a level of liquidity higher than lELV for any value of k.

50
Π̃∗R (ρ) is continuous in ρ and (l∗ (ρ, M ) , k ∗ (ρ, M )) is a (single-valued) continuous function for all
ρ ∈ Γ, for any given M .
(ii.a) (Zero capital requirements for high cost of capital) From Proposition 4, we have that for
all ρ ≥ ρmax the optimal capital response curve is zero for all liquidity levels. In this case, we have
that the optimal liquidity and capital requirements are given by l∗ = lR (0) and k ∗ = 0.
For any ρ < ρmax , the optimal capital response curve is positive in a range (l1 (ρ) , l2 (ρ)) and
zero elsewhere, i.e., kR (ρ) > 0 if and only if l ∈ (l1 (ρ) , l2 (ρ)). Moreover, we also know that l1 (ρ)
is continuous and decreasing in ρ. Hence, l1 (ρ) attains a maximum at l1 (ρmax ) and a minimum
at l1 (ρmin ). Also, from Proposition 5, we know that l (k) |M is continuous and decreasing in M
for all k. Hence, l (0) |M attains a maximum at l (0) |M for each k. Moreover, we know that inf
M ∈(M ,∞) {l (0) |M } = lELV and also that l (0) |M > lELV for all M < ∞. Let ρ̄ (M ) be implicitly
defined as l (0) |M = l1 (ρ̄ (M )). We distinguish two cases. If l1 (ρmin ) > lELV , then there exists
M̃ > M such that for any M ≥ M̃ the optimal liquidity and capital response curves do only
intersect in the range in which the optimal capital response curve is flat (case 1). However, if either
l1 (ρmin ) > lELV and M < M̃ , or if l1 (ρmin ) ≤ lELV , then the curves do intersect for some M at some
point in which the optimal capital response is positive (case 2). In case 1, we have that the optimal
liquidity and capital requirements are given by l∗ = lR (0) and k ∗ = 0 . We focus henceforth on
case 2.
Let ρ ≥ ρ̄ (M ). We show that the optimal liquidity and capital requirements are given by
l = lR (0) and k ∗ = 0, respectively, for any ρ ≥ ρ̄ (M ). Fix M and let ρ̃ be supremum of the
∗

set of values of the cost of capital for which the curves intersect at some level for which capital is
positive, that is, ρ̃ = sup {ρ : k ∗ (ρ, M ) > 0}. If ρ̃ ≤ ρ̄ (M ), then we are done. Suppose now that
ρ̃ > ρ̄ (M ). Then, by construction of ρ̃, we have that the curves intersect in the range in which the
optimal capital response is zero for ρ > ρ̃ and for some level of capital bounded away from zero,
k̃ > 0, for ρ = ρ̃. Hence, it follows that the capital requirement would be k ∗ (ρ, M ) = 0 for ρ > ρ̃.
We now show that and k ∗ (ρ̃, M ) = 0 as well. Define the sequences ρn ≡ ρ̃ + n1 and kn∗ ≡ k ∗ (ρn , M )
for all n ∈ N. On the one hand, we have that limn→∞ ρn = ρ̃. On the other hand, we have that
kn∗ = 0 for all n ∈ N, so that it follows that limn→∞ kn∗ = 0. Therefore, since (l∗ (ρ, M ) , k ∗ (ρ, M ))
is continuous, it follows that k ∗ (limn→∞ ρn , M ) = k ∗ (ρ̃, M ) = 0.
From now on, we consider the case ρ < ρ̄ (M ). First, observe that by construction of ρ̄ (M ), we
have that l1 (ρ) < l (0) |M .
(ii.b) (Raising the cost of one factor reduces the requirement of that factor) For any given ρ and
M , we have that kR (l) |ρ is strictly decreasing in ρ for any l ∈ (l1 (ρ) , l2 (ρ)) and lR (M ) is strictly
decreasing in M . The result follows immediately.

51
 (ii.c) (Capital and liquidity complements for low cost of capital and substitutes for high cost of
capital)
First, we fix M and analyze the impact of changing ρ on the optimal liquidity requirements.
Observe that the optimal liquidity response curve lR (k) is increasing in the complementarity region
and decreasing in the substitutability region. Moreover, lR (k) is independent of ρ. Hence, since M is
fixed, lR (k) does not change with changes in ρ. Now, consider the map of optimal capital responses
{kR (l) |ρ }ρ∈(ρmin ,ρ̄(M )) . This map consists of a collection of hat-shaped functions that increase in
the complementarity region and decrease in the substitutability region. Suppose that, for any
given ρ, (l∗ (ρ, M ) , k ∗ (ρ, M )) belongs to the complementarity region. Then, an increase in ρ shifts
the optimal capital response curve down. Since the optimal liquidity response function does not
change, the optimal requirements are given by moving down along the optimal liquidity response
curve, where both curves intersect. Since lR (k) is increasing (as (l∗ (ρ, M ) , k ∗ (ρ, M )) is in the
complementarity region), it follows that the optimal liquidity requirement l∗ (ρ, M ) decreases. By
the same token, suppose that for any given ρ, (l∗ (ρ, M ) , k ∗ (ρ, M )) belongs to the substitutability
region. Then, an increase in ρ also shifts the optimal capital response curve down. Since the optimal
liquidity response function does not change, the optimal requirements are given by moving down
along the optimal liquidity response curve, where both curves intersect. Since lR (k) is decreasing,
it follows that the optimal liquidity requirement l∗ (ρ, M ) increases. Finally, we need to show that
there exists a threshold ρ̂ (M ) such that the optimal liquidity and capital requirements belong to the
complementarity region if ρ > ρ̂ (M ) and to the substitutability region if ρ < ρ̂ (M ). Observe that
for ρ̄ (M ), the optimal liquidity requirement l∗ (ρ̄ (M ) , M ) belongs to the complementarity region.
As ρ decreases, the optimal liquidity requirement l∗ (ρ, M ) moves up along the optimal liquidity
response curve lR (k) until the curve reaches its maximum and crosses to the substitutability region,
which occurs for a given ρ̂ (M ). Then, l∗ (ρ, M ) decreases along the optimal liquidity response curve
lR (k) within the substitutability region for ρ < ρ̂ (M ).
Now, we fix ρ < ρ̄ (M ) and analyze the impact of changing M on the optimal capital require-
ments. Observe that the optimal capital response curve kR (l) is increasing in the complementarity
region and decreasing in the substitutability region. Now, consider the map of optimal liquidity
responses {lR (k) |M }ρ∈(M ,∞) . This map consists of a collection of hump-shaped functions that in-
crease in the complementarity region and decrease in the substitutability region. Moreover, kR (l) is
independent of M . Hence, since ρ is fixed, kR (l) does not change with changes in M . Suppose that,
for any given M , (l∗ (ρ, M ) , k ∗ (ρ, M )) belongs to the substitutability region. Then, an increase
in M shifts the optimal liquidity response curve down (or to the left, if one looks at Figure ??).
Since the optimal capital response function does not change, the optimal requirements are given

52
by moving up along the optimal capital response curve, where both curves intersect. Since kR (l)
is decreasing (as (l∗ (ρ, M ) , k ∗ (ρ, M )) is in the substitutability region), it follows that the optimal
capital requirement k ∗ (ρ, M ) increases. By the same token, if (l∗ (ρ, M ) , k ∗ (ρ, M )) belongs to the
complementarity region, an elevation of M leads to an increase of the optimal capital requirement
k ∗ (ρ, M ). Finally, define M̂ implicitly so that the chosen cost of capital ρ constitutes the threshold
∗ ∗
that establishes whether
 (l
 (ρ, M ) , k (ρ, M )) are in the complementarity or in the susbtitutability
region, that is, ρ = ρ̂ M̂ . Then, for any M > M̂ it follows that ρ > ρ̂ (M ). In this case, the op-
timal liquidity and capital requirements (l∗ (ρ, M ) , k ∗ (ρ, M )) belong to the substitutability region.
On the contrary, if M < M̂ then ρ > ρ̂ (M ), so that the optimal requirements (l∗ (ρ, M ) , k ∗ (ρ, M ))
belong to the complementarity region.
(ii.d) (Capital and liquidity are always complements when the return to investment is high) It
follows directly from Proposition 5, statement (ii.b).
(ii.e) (Zero capital requirements for high return to investment) The cost of capital threshold
ρ̄ (M ) that determines whether capital requirements are positive or not is defined as l (0) |M =
l1 (ρ̄ (M )). Since l1 (ρ) is invariant with M and l (0) |M is strictly decreasing in M , the result follows
immediately.

53