Dynamic Bank Capital Requirements

Tetiana Davydiuk

Abstract

The Basel III Accord requires countercyclical capital buffers to protect the banking
system against potential losses associated with excessive credit growth and buildups of
systemic risk. In this paper, I provide a rationale for time-varying capital requirements
in a dynamic general equilibrium setting. An optimal policy trades off reduced inefficient
lending with reduced liquidity provision. Quantitatively, I find that the optimal Ramsey
policy requires a capital ratio that mostly varies between 4% and 6% and depends
on economic growth, bank supply of credit, and asset prices. Specifically, a one
standard deviation increase in the bank credit-to-GDP ratio (GDP) translates into
a 0.1% (0.7%) increase in capital requirements, while each standard deviation increase
in the liquidity premium leads to a 0.1% decrease. The welfare gain from implementing
this dynamic policy is large when compared to the gain from having an optimal fixed
capital requirement.

Keywords: Ramsey policy, capital requirements, liquidity provision, countercyclical buffers

1 Introduction

The recent financial crisis renewed attention on the stability of the banking sector and on capital
regulation. Both policymakers and academics have recognized that risk-based capital requirements,
such as those in the 2004 Basel II Accord, tend to exacerbate business-cycle fluctuations, leading
to an overly cyclical credit supply. To protect the banking sector from periods of excess credit
growth and leverage buildup, the 2010 Basel III Accord introduced the countercyclical capital
buffer (CCyB) framework. The proposed adjustment effectively translates into capital ratios that
vary over time. Despite a few recent trials to implement CCyB regimes within countries in the
European Union, there is no consensus at the practical or theoretical level about the implications
of CCyBs on the macroeconomy.1
The goal of this paper is to investigate both the theoretical and quantitative implications of
the optimal state-dependent capital requirements. By comparison, most of the existing literature
has limited itself to examining the costs and benefits of the overall level of capital requirements.2
To this end, I develop a tractable general equilibrium model and use it to characterize the optimal
bank capital requirements. The problem of choosing an optimal capital policy follows from the
work of Ramsey (1927) and builds on the primal approach used in the optimal taxation literature3 .
My model has three key features. First, high-quality production projects are scarce, especially
during economic downturns. Second, banks benefit from government guarantees in the form of
bank bailouts, which lowers their borrowing cost, thereby creating incentives to risk-shift. Third, I
assume in the model that households value money-like assets in the form of bank debt. This implies
that bank debt is priced at a discount and is a cheaper source of funding than equity.4 The last
two features create a trade-off: while tighter capital ratios can help to mitigate a bank’s motives
to risk-shift and rule out inefficient lending, at the same time they can reduce the bank’s supply of
credit and the creation of safe liquid assets.
1
The expectation is that the buffer will be at zero through most of the business cycle and increase to the maximum
2.5% points only at the peak of the credit cycle. Even though under 2010 Basel Accord the CCyB regime will be
phased in globally between 2016 and 2019, individual EU members have already introduced the CCyBs. For instance,
the Bank of England maintained the buffer rate equal to 0% starting from 2014:Q2, while the Norwegian authorities
set a 1% countercyclical capital buffer rate for the period 2015:Q3-2016:Q2 and 1.5% from 30 June, 2016. The
European Systemic Risk Board (ESRB) provides an overview of the national countercyclical capital buffer rates. See:
https://www.esrb.europa.eu/national policy/ccb/applicable/html/index.en.html.
2
Some examples include Diamond and Rajan (2000), Diamond and Rajan (2001), Hellmann, Murdock and Stiglitz
(2000), Repullo (2004), Morrison and White (2005), Van den Heuvel (2008), Admati, DeMarzo, Hellwig and Pfeiderer
(2010), DeAngelo and Stulz (2013), Nguyen (2014), Begenau (2016),
3
See, for example, Lucas and Stokey (1983), Chari, Christiano and Kehoe (1991) and Atkinson and Stiglitz (2015).
4
In this paper, I interchangeably use debt and deposits, even though in the data as of 2014 deposits amounted
only to 77% of bank liabilities.

2
 In the model, procyclical capital requirements emerge endogenously as an optimal policy scheme
in a Ramsey equilibrium – thus providing the best implementable competitive equilibrium.5 In line
with empirical evidence, excessive investment in risky and inefficient projects builds-up during
periods of high economic growth.6 Because of better lending opportunities during expansions,
banks increase the supply of credit. As a result, expected bailout subsidies become larger, thereby
lowering banks borrowing costs and translating into an even higher level of lending. Consequently,
heightened capital regulation can be most beneficial through credit booms. At the same time,
tight capital requirements can have a contractionary effect on bank lending and deposit creation
especially during downturns, when liquidity is more valuable to households and, as a result, equity
financing is more costly for banks.7 This implies that optimal capital charges should be lower
during credit freezes, thereby reinforcing the cyclicality of optimal capital ratios.
My results also show that the optimal Ramsey policy used as a single policy tool is not sufficient
to restore both the socially optimal level of lending and liquidity provision. To provide some
intuition behind the design of optimal capital regulations, I solve separately for the “lending capital
requirement” – restoring the first-best level of investment but admitting a reduced level of deposits
– and the “liquidity capital requirement” – ensuring the first-level of liquidity provision but also
allowing for excessive lending. Such policies focus only on one dimension of the problem, either
dampening incentives of banks to risk-shift or stimulating creation of safe and liquid assets. By
contrast, the Ramsey capital requirement balances reduced inefficient lending with reduced liquidity
provision.
The model highlights important trade-offs one needs to consider when regulating bank capital.
However, for effective policy making, it is also crucial to assess these guidelines quantitatively.
I therefore calibrate the model to best match key macroeconomic quantities, as well as bank
data counterparts. Using this choice of parameters, I solve numerically for the optimal policy
5
This is consistent with the “countercyclical capital buffer” instituted with Basel III. The required buffer is called
countercyclical because the goal is to dampen the impact of capital regulation on the cyclicality of bank lending. The
buffer itself is procyclical.
6
Even though there is no unifying theory of bank risk-taking over the business cycle, empirical evidence suggests
that credit booms are associated with declines in lending standards, greater recourse to riskier contract structure, and
changes in leverage (Mian and Sufi (2010), Demyanyk and Hemert (2011), DellAriccia, Igan and Laeven (2012b)).
Besides, there is indirect evidence that risks tend to build-up during periods of high economic growth: credit booms
do not just proceed but rather cause financial crises (Lorenzoni (2008), Dell’Ariccia, Igan, Laeven and Tong (2012a),
Borio (2014)).
7
It is important to differentiate between the ongoing costs of holding equity on the balance sheet and the costs of
raising new external equity finance. One of the theories why equity is more expensive on ongoing basis has to do with
the liquidity premium on safe “money-like” assets produced by banks. Provided that investors value liquidity most
when the supply of safe assets is low (e.g., Krishnamurthy and Vissing-Jorgensen (2012)), the liquidity premium is
high during recessions when banks reduce their lending activity and deposit creation. This is turn implies that equity
financing is more costly during economic slowdowns.

3
in the implementable Ramsey equilibrium. Specifically, I characterize the optimal state-contingent
capital regulation and the set of allocations for bank lending and liquidity provision that maximize
consumers’ lifetime utility. Ramsey allocations have the property that they can be implemented as
a competitive equilibrium when banks encounter the prescribed capital requirements.
Based on quantitative implementation of the model, I find that the optimal Ramsey policy
requires a cyclical capital ratio that mostly varies between 4% and 6% and depends on key indicators
of economic growth and asset prices. This range of values is roughly comparable with a 2.5% upper
bound of the CCyB proposed by the Basel III Accord. Importantly, the optimal policy rule is not
capped at 6%, and can rise above it during periods of abnormal economic growth. In addition, the
mean Ramsey capital requirement is around 5%. This is one percentage point higher than the level
of the leverage ratio and one percentage point lower than the Tier 1 capital ratio as recommended
by the Basel III Accord.8
Implementing the optimal Ramsey policy delivers a permanent increase in the annual
consumption of households relative to the current capital ratios.9 Although the exact magnitude of
the welfare gain depends on the attitude of households toward risk, more than 60% is attributed to
having state-dependent capital requirements. The remaining welfare improvement is achieved by
setting an optimal fixed capital requirement equal to the mean Ramsey capital ratio. The optimal
policy leads to a reduction in the cyclicality of bank credit – reducing inefficient lending during
expansions, but increasing the supply of credit during economic slowdowns – and, as a result,
achieves on average a higher level of consumption and deposit creation.
The main contribution of the paper is to characterize the optimal policy rule in terms
of measurable macroeconomic and bank aggregates. While the Basel Committee recommends
adjusting the capital requirements based on “credit gap”, the deviation of the credit-to-GDP ratio
with respect to its long-term trend, I show the credit gap alone fails to capture the time variation
in the optimal Ramsey policy. Instead, I find the optimal policy rule takes into account the joint
behavior of the credit gap, GDP, and the liquidity premium. Specifically, the optimal Ramsey
policy is well approximated by the following rule:

5% + 0.1% × credit gap + 0.7% × GDP − 0.1% × liquidity premium,

where the liquidity premium is a discount on the price of safe liquid assets.10 A one standard
8
Note that these two ratios – Tier 1 capital ratio (Tier 1 capital divided by total risk-weighted assets) and the
leverage ratio (Tier 1 capital divided by total assets) – are the same in the model, even though it is not the case in
the data.
9
In the model, utility of households is defined over consumption and deposits according to a CES aggregator
v (C, D). To this extent, I measure the welfare costs in terms of changes in v (C, D) rather than in C.

4
deviation increase in the credit-to-GDP ratio (GDP) translates into a 0.1% (0.7%) increase in the
capital charges, while a one standard deviation increase in the liquidity premium leads to a 0.1%
drop. Growth in bank credit, along with the output growth, act as indicators of banks’ incentives
to risk-shift. By contrast, the liquidity premium serves as an indicator of how expensive equity
financing is for banks and how valuable liquid assets are for investors.
The paper proceeds as follows: In Section 2, I provide the details of bank regulation, as well as
review the related literature. I develop the baseline model in Section 3. I characterize the first-best
allocation in this economy in Section 4, as well as the competitive equilibria with and without
capital regulations in place. In Section 5, I examine the lending and liquidity capital requirements
within the baseline model. I provide the quantitative assessment of the model in Section 6, as well
as present the optimal policy rule. Concluding remarks are give in Section 7.

2 Bank Regulation and Related Literature

2.1 The Basel Accords

The Basel Accords, developed by the Basel Committee on Bank Supervision (BCBS), consolidate
capital requirements as the cornerstone of bank regulation. The 1988 Basel Accord, known as Basel
I (BCBS (1988)), was criticized for the risk-insensitivity of capital charges. To address this critique,
the internal ratings-based (IRB) approach was introduced with the publication of Basel II in 2004.
Under risk-based regulation, the amount of capital that a bank is required to hold against a given
exposure depends on the estimated credit risk of that exposure, which in turn is determined by the
probability of default (PD), loss given default (LGD), exposure at default (EAD), and maturity.
The key implication of the IRB approach is that riskier exposures carry a higher capital charge. The
intention of this approach is to reduce bank failures and the associated systemic costs by holding
the bank probability of default below some fixed target.
Both policymakers and academics have recognized that risk-sensitive capital regulations, such
as that in Basel II, tend to exacerbate the inherent cyclicality of bank lending and, consequently,
distort investment decisions. This is because in economics downturns losses erode bank capital, and
the remaining (non-defaulted) loans are downgraded by the relevant credit risk model, delivering
higher capital charges. To the extent that it is difficult or costly for a bank to raise new capital
during recessions, it will be forced to cut back on its lending activity. Kashyap, Rajan and Stein
10
The three indicators are expressed in log-deviations from their steady states and normalized by their standard
deviations. The GDP growth variable has been orthogonolized with respect to the credit gap variable.

5
(2008) provide empirical evidence that equity-raising was sluggish during the recent financial crisis.
Kashyap and Stein (2004) argue that the IRB approach is largely microprudential in nature
and ignores the importance of the function of bank lending. In line with the literature on capital
crunches in banking, they claim that the shadow value of bank capital increases during recessions
and a capital requirement that is too high when bank capital is scarce may result in reduced
funding of positive net present value projects. So, if the government’s objective is both to protect
the financial system against the costs of bank defaults and sustain bank-lending efficiency, the
capital charges should be adjusted to the state of the business cycle (for any degree of credit-risk
exposure).
An important argument that is sometimes made is that during periods of economic growth
banks may hold capital in excess of the minimum regulatory requirements that could neutralize
potential cyclicality problems. Repullo and Suarez (2012) develop a dynamic model of relationship
lending, in which banks hold voluntary capital buffers as a precaution. They find that the capital
buffers set aside during expansions are typically not sufficient to prevent credit supply shrinkage
during recessions. They also document that the optimal capital requirements are higher and less
cyclically-varying than the requirements of Basel II when the social cost of bank failure is high.
In an attempt to strengthen bank balance sheets against future financial upheavals, Basel III
introduced the countercyclical capital buffers, which range from zero to 2.5% of risk-weighted assets.
As mentioned before, the BCBS proposed to adjust the level of the buffer based on the credit gap,
meanwhile acknowledging that it may not be a good indicator of stress in downturns. For example,
Repullo and Saurina (2011) find that the credit gap for many countries is negatively correlated with
GDP growth. This can be traced to the fact that the supply of credit typically lags the business
cycle, especially in downturns.
Kashyap and Stein (2004), Gordy and Howells (2006), Saurina and Trucharte (2007), and
Kashyap et al. (2008), among others, focus on the correction of risk-based capital requirements in
a macroprudential direction. Using Spanish data, Repullo, Saurina and Trucharte (2010) analyze
different procedures aimed at mitigating the procyclical effects of capital regulation and conclude
that the most appealing one is to use a business cycle multiplier based on GDP growth. The
proposed adjustment maintains the risk sensitivity in the cross-section (i.e., banks with riskier
portfolios would bear a higher capital charge), but a cyclically-varying scaling factor would increase
capital requirements in good times and reduce them in bad times. In line with this study, I propose
a cyclical policy rule that depends positively on indicators of economic growth. In my setting,
state-contingent capital requirements emerge endogenously as an optimal policy scheme. Optimal

6
regulations promote the stability of the banking sector without contracting the supply of bank
credit and deposit creation. Arguments in favor of time-varying capital regulations are also found
in Kashyap et al. (2008), Hanson, Kashyap and Stein (2011), and Malherbe (2015). Note that
I abstract from the cross-sectional dimension of capital regulation and focus on the time-series
dimension.

2.2 Literature Review

This research project is at the intersection of a large literature on optimal banking regulation
theory and dynamic macroeconomic models of financial intermediation. The recent financial crisis
has brought to the forefront the discussion of whether the capital requirements of banks should
be increased (Admati et al. (2010)). On one side, Hellmann et al. (2000), Repullo (2004), and
Morrison and White (2005) argue that stringent capital regulation can induce prudent behavior by
banks. On the other side, Diamond and Rajan (2000), Diamond and Rajan (2001), and DeAngelo
and Stulz (2013) provide theoretical evidence that tightening capital requirements may distort
banks’ provision of liquidity services, while Dewatripont and Tirole (2012) show that stricter capital
requirements may introduce governance problems. In line with the existing literature, I suggest that
reduced moral hazard is the key rationale for imposing restrictive capital regulation with reduced
liquidity provision being the main downside.
This paper is most closely related to other quantitative studies on the welfare impact of capital
requirements (Van den Heuvel (2008) , Nguyen (2014), Begenau (2016), Van den Heuvel (2016))
and leverage constraints (Bigio (2010), Martinez-Miera and Suarez (2012), Corbae and D’Erasmo
(2014), and Christiano and Ikeda (2014) ). Among the most recent studies estimating the optimal
level of fixed capital requirements are Nguyen (2014) and Begenau (2016). Nguyen (2014) finds that
consumers’ lifetime utility is maximized at 8% capital ratio, while the optimal number estimated
by Begenau (2016) is 14%. A common feature in these two papers is that government guarantees
incentivize banks to engage in excessive risk-taking, while capital regulation helps to address these
distortions. Begenau (2016) also extends the banks’ role to providing liquidity services, which are
valued by households. I complement this branch of literature by developing a tractable framework
to quantify the benefits and costs of capital regulation over the business cycle. This allows me to
provide not only qualitative, but also quantitative recommendations on the policy design. To the
best of my knowledge, this paper is the first to solve for the stage-contingent capital requirements
within the Ramsey framework.
There are few empirical contributions on the effects of higher capital requirements on bank

7
lending and bank cost of capital. Looking at the data on large financial institutions, Kashyap,
Stein and Hanson (2010) find that even if the minimum capital ratio is increased by 10%, the
impact on loan rates is likely to be modest, in the range of 25 to 45 bps. In a similar vein,
Baker and Wurgler (2013) calibrate that a ten-percentage point increase in capital requirements
would translate into a higher weighted average cost of capital by 60-90 bps per year. Kisin and
Manela (2013) estimate the perceived costs of capital requirements by employing the data on banks’
participation in a costly loophole that helped them to bypass capital regulations. They document
that a ten-percentage point increase in Tier 1 capital to risk-weighted assets leads to, at most, a 3
bps increase in banks’ cost of capital. Even though these studies shed light on the potential impact
of capital regulations on the real economy, it is difficult to assess the overall welfare implications of
a time-varying capital requirement in their setting.
More broadly, this paper fits a strand of macroeconomic literature on the role of financial
intermediation in the development of economic crises. The transmission mechanism by which the
effects of small shocks persist, amplify, and spread to the macroeconomy is first identified in the
seminal works of Bernanke and Gertler (1989), Kiyotaki and Moore (1995), and Bernanke, Gertler
and Gilchrist (1999). The more recent studies of the balance sheet channel include Gertler and
Kiyotaki (2010), Gertler and Karadi (2011), He and Krishnamurthy (2012), Di Tella (2013), and
Brunnermeier and Sannikov (2014).

3 Model Setup

To focus on the novel mechanisms associated with dynamic bank capital requirements, I first layout
the baseline model, which is kept as parsimonious as possible. I later augment the model to quantify
policy recommendations on what defines optimal time variation in capital requirements of banks.
In the model, time is discrete and runs for an infinite number of periods. Financial intermediaries
own the production technology and the stock of capital in this economy. They are financed either
partially or entirely with deposits. The presence of deposit insurance distorts banks’ optimal choices
of investment. Households consume the final goods produced in the banking sector and invest any
savings in the banks’ deposits. To finance bailout expenditures, the government levies taxes on
households in lump-sum fashion.

8
3.1 Banking Sector

I start by characterizing a banking sector in detail. At any point in time the economy is populated
with a continuum of ex ante identical banks of measure one, indexed by j ∈ Ω = [0, 1]. Each bank
has access to decreasing returns to scale technology and produces a final good yj,t , using capital as
the only input,
yj,t = eωj,t +at lj,t
α
,

where at is an aggregate productivity shock, which follows:

at = (1 − ρa ) ā + ρa at−1 + σa t , t ∼ iid N (0, 1) (1)

and ωj,t is an idiosyncratic disturbance, which is identically distributed across time and across
banks:
1
ωj,t = − σω2 + σω εj,t , εj,t ∼ iid N (0, 1) . (2)
2
In the cross-section, the bank-specific shocks average to zero. This setup is isomorphic to the one,
where a consumption good is produced by penniless firms, who have been credit rationed in the
capital markets due to unmodeled information asymmetries, and whose only source of funding is
bank loans.11 The capital used in banks’ production process can be correspondingly interpreted
as the amount of bank loans issued to these types of borrowers. From this point on, I refer to
lj,t+1 as bank lending. The decreasing returns to scale assumption is crucial to capture the idea
that borrowers are not homogeneous and that there is a finite number of creditworthy borrowers,
or equivalently a finite number of positive net present value (NPV) production projects. This
setup ensures that banks internalize that each extra lending unit (borrower) is not as productive
(creditworthy) as the previous one.
Bank j enters period t with lj,t lending units, financed with dj,t units of debt and nj,t units of
11
Suppose that each bank operates on an island j; one can think of an island as an industry or a state. On each
island, there is a continuum of cashless firms who require a unit investment at time t for production at time t + 1.
Firms on each island are ranked according to their productivity. Specifically, a firm i on an island j (where i denotes
the firm’s ranking) produces αeωj,t +at iα−1 units of a consumption good. At time t, the bank on island j issues loans
to the first lj,t+1 firms with the total amount equal to:
Z lj,t+1
1di = lj,t+1 .
0

The monopolist bank on the island can extract all surplus. In particular, at time t + 1 the bank on island j receives
from the firms: Z lj,t+1
αeωj,t+1 +at+1 iα−1 di = eωj,t+1 +at+1 lj,t+1
α
.
0

9
equity. The balance sheet equates risky assets to bank debt and equity:

lj,t – loans nj,t – net worth

dj,t – debt

The bank’s revenues realized at time t are equal to its earnings on the production net of the interest
payments on its liabilities:

πj,t = eωj,t +at lj,t

α
− lj,t − (Rd,t − 1) dj,t .

Next, I assume that when a financial intermediary does not have sufficient funds to service its
deposit liabilities, depositors are bailed out with probability one.12 In particular, banks will default
∗ , defined
on their credit obligations whenever their idiosyncratic shock ωj,t is below a cutoff level ωj,t
by the expression:
∗
πj,t + nj,t = 0 ⇔ eωj,t +at lj,t
α
= Rd,t dj,t .

The net worth, nj,t+1 , available to banks at the end of period t (going into period t + 1), evolves
according to:

nj,t+1 = {πj,t + nj,t }+ − zj,t (3)

n  o+
α−1
= eωj,t +at lj,t − Rd,t lj,t + Rd,t nj,t − zj,t ,

where {·}+ denotes the maximum operator max {·, 0} and captures that banks are subject to
limited liability and government guarantees. zj,t is the net payouts to the bank’s shareholders. A
positive net transfer, zj,t > 0, means that the equityholders receive dividends, while a negative
one, zj,t < 0, means that there is an equity issuance. Equation (3) demonstrates that any growth
in bank equity above the deposit return depends on the premium that the financial intermediary
earns on its assets, as well as the total amount of lending. To finance the difference between the
capital investment and available net worth, the financial intermediary borrows an amount dj,t+1
from households, given by:
dj,t+1 = lj,t+1 − nj,t+1 .
12
The banking sector is subject to government guarantees for reasons that I do not explicitly model. As has
been acknowledged by the academics and practitioners, without government protection, economies are prone to bank
runs. Bank runs were a recurrent feature during the Great Depression in the U.S. and continue today, including
recent episodes of bank failures in the U.S. (Countrywide, IndyMac), Great Britain (Northern Rock), and India
(ICICI Bank). With many bank failures, there is a disruption of the monetary system and a reduction in production
(Bernanke (1983), Calomiris and Mason (2003), Dell’Ariccia, Detragiache and Raghuram (2008)). With the goal
of protecting the economy against the potential welfare losses associated with bank runs, deposit insurance was
introduced along with a number of implicit government guarantees.

10
 Finally, banks are subject to capital regulations, which require them to have a minimum amount
of equity as a fraction of assets. Since loans are the only type of asset in my model, the capital
requirement I instate is that equity needs be at least a fraction ζt of loans for a bank to be able to
operate:
nj,t+1 ≥ ζt lj,t+1 .

In each period, bank j decides how many loans to issue, lj,t+1 , and makes a leverage choice to
maximize the discounted sum of the equity payouts:

∞
" #
X
t
max E β zj,t (4)
zj,t ,lj,t+1 ,dj,t+1 ,nj,t+1
t=0
+
α
nj,t+1 = eωj,t +at lj,t

s.t. − Rd,t dj,t − zj,t ,
lj,t = nj,t + dj,t ,
nj,t+1 ≥ ζt lj,t+1 ,
nj,0 , dj,0 given.

3.2 Household Sector

The economy is populated by a measure one of identical households. There are two types of
members in each household: savers and bankers (Gertler and Karadi (2011)). Savers hold deposits
at a diversified portfolio of financial institutions and return the interest earned to their household.
Bankers, on the other hand, manage financial intermediaries and similarly return any earnings back
to the household. The savers hold deposits at the banks that its household does not own, otherwise
in the absence of tax frictions, the Modigliani-Miller theorem would hold and the irrelevance of
banks’ capital structure would follow. Savers’ and bankers’ earnings are perfectly shared across the
entire household to secure a representative agent framework.
Let Ct denote family consumption and Dt+1 denote the holdings of bank deposits. Households
are risk-neutral and have a discount rate of β ∈ (0, 1) .13 I also assume that households value the
non-pecuniary services provided by bank deposits and, hence, enjoy the additional flow of utility
v (Dt+1 ), which is a concave non-decreasing function of the supply of deposits Dt+1 .14
In each period, the household chooses a consumption level and deposit holdings to maximize
13
A household’s consumption can be both positive and negative, as in Brunnermeier and Sannikov (2014). Negative
consumption can be interpreted as disutility from labor.
14
See, for example, Nataliya, Pfeil, Rochet and De Nicolo (2016). This particular utility function allows me to
derive the analytical solutions for the baseline model, but it is generalized when I move to the quantitative assessment
of the optimal capital requirements.

11
 Figure 1: Rate of Return on Deposits

Liquidity
premium

1
β
Rd,t+1
0 20 40 60 80 100
Dt+1
This figure depicts the rate of return on deposits Rd,t+1 as a function of the deposits supply Dt+1 and the rate of
1
return on other safe assets β
. The scale of y-axis is omitted, as the numbers in this parameterized example do not
have economic meaning. The parameter values are set to β = 0.98, η = 0.8, α = 0.85, δ = 1, ρa = 0.97, σa = 0.02,
and σω = 0.2.

their utility subject to the budget constraint:

∞
" #
X
t
max E β (Ct + v (Dt+1 )) (5)
Ct ,Dt+1
t=0
1−η
Dt+1
with v (Dt+1 ) = , η<1
1−η
s.t. Ct = Rd,t Dt − Dt+1 + Zt − Tt ,
D0 given,

where Tt is a lump-sum tax levied by the government. Households are the owners of financial
intermediaries and at the end of each period receive the net proceeds of bank activity, Zt . A unit
of deposits issued in period t yields a gross return of Rd,t+1 at time t + 1.
The presence of the government guarantees implies that Rd,t+1 is a return on a riskless asset.
The first-order conditions of (5) impart that the interest rate on deposits is equal to:

1 1 −η
Rd,t+1 = − D . (6)
β β t+1
| {z }
liquidity premium

12
Since households exhibit a preference for bank debt, there is a discount on its interest rate, which
is the amount households are willing to relinquish in exchange for holding a risk-free asset, which
gives a flow of utility compared to a risk-free asset that does not give a flow of utility.15 Figure 1
shows that liquidity premium is a decreasing function in the supply of bank deposits. This implies
that liquid deposits become more valuable to the household when they are scarce.
Diamond and Dybvig (1983) were among the first to analyze household demand for liquidity.
They find that bank deposits provide better risk-sharing possibilities to consumers, justifying
the existence of the liquidity premium. Stein (2011) and Gorton and Metrick (2012) argue that
collateralized short-term debt issued by banks yields a “money-like” convenience premium based on
its relative safety and the transactions services that safe claims provide. Further, Krishnamurthy
and Vissing-Jorgensen (2012) and Krishnamurthy and Vissing-Jorgensen (2015) document that
investors value the money-like features of Treasuries the most, when the supply of Treasuries is
low. The welfare implications of the presence of liquidity premium via its effects on the lending
costs of banks have also been studied by Begenau (2016) and Van den Heuvel (2016).

3.3 Government

The role of the government is to provide deposit insurance. To finance its expenditures on bank
bailouts, the government levies a lump-sump tax on households in accordance with a balanced
budget rule: Z 1
α
max Rd,t dj,t − eωj,t +at lj,t

Tt = , 0 dj. (7)
0

4 Equilibrium Characterization

Definition. A competitive equilibrium is a set of prices {Rd,t+1 }∞ t=0 , government policies

n o∞
∞
{ζt , Tt }t=0 , and allocations Ct , Dt+1 , {zj,t , lj,t+1 , nj,t+1 , dj,t+1 }j∈Ω , such that:
t=0

(i) Given prices {Rd,t+1 }∞ ∞

t=0 , government policies {Tt }t=0 and initial amount of savings, D0 ,
households maximize their life-time utility given by (5);
15
In this setting, the rate of return on any other safe asset other than bank debt is equal to β1 . As long as households
derive utility from bank deposits and the economy is not fully saturated with liquidity, the interest rate on bank debt
is always lower than the interest rate on any other risk-free asset,
1
lim Rd,t+1 = .
Dt+1 →∞ β

13
 (ii) Given prices {Rd,t+1 }∞ ∞
t=0 , government policies {ζt , Tt }t=0 and initial capital structure
{nj,0 , dj,0 }j∈Ω , each bank j ∈ Ω maximizes the discounted sum of equity payouts given by (4);

(iii) The government budget constraint (7) is satisfied;

(iv) Market clearing conditions hold:

– resource constraint Z 1 Z 1
Ct + lj,t+1 dj = yj,t dj,
0 0

– deposits market Z 1
Dt+1 = dj,t+1 dj.
0

The nature of the bank’s problem (4) implies that the equilibrium is symmetric and all banks
make identical decisions. The realization of the bank-specific shock ω induces a bailout transfer
from the government to a subgroup of banks, but does not affect the optimal decisions of banks.
To see this result, it is useful to define the value of the bank to its shareholders as follows:

J (lj,t , nj,t , ωj,t , St ) = max {πj,t + nj,t , 0} + V (St ) ,

where the continuation value V (·) obeys the following Bellman equation:
  Z +∞ 
V (St ) = max −nj,t+1 + Et βMt,t+1 J (lj,t+1 , nj,t+1 , ωj,t+1 , St+1 ) dF (ωj,t+1 ) (8)
nj,t+1 ,lj,t+1 0
( " !#)
Z +∞
= max −nj,t+1 + Et βMt,t+1 (πj,t+1 + nj,t+1 ) dF (ωj,t+1 ) + V (St+1 ) .
nj,t+1 ,lj,t+1 ∗
ωj,t+1

The conditional expectation Et is taken only over the distribution of aggregate productivity shocks
and St denotes the aggregate state of the economy. To the extent that bank-specific shocks are not
persistent and there is no equity issuance costs, all financial intermediaries are ex ante the same and
make identical decisions. In particular, lj,t+1 = Lt+1 , nj,t+1 = Nt+1 , and dj,t+1 = Dt+1 , ∀j ∈ Ω.
To uncover the inefficiencies introduced with the presence of government subsidies, I first solve
for the socially optimal allocation in this economy. Next, I characterize a competitive equilibrium
when banks face no capital regulations and then compare the optimal decisions of banks in the two
equilibria.

14
4.1 First-Best Allocation

A social planner chooses consumption and supply of liquid deposits to maximize households’ lifetime
utility subject to the resource constraint:

∞
" !#
X D1−η
max E β t
Ct + t+1 (9)
Ct ,Lt+1 ,Dt+1 ≤Lt+1 1−η
t=0

s.t. Ct + Lt+1 = eat Lαt ,

where the regulator takes into account that the competitive equilibrium is symmetric and that the
deposit’s supply is bounded by the stock of capital in the banking sector. I characterize the socially
optimal allocation in Proposition 1.

Proposition 1 The first-best allocation is characterized by:

(i) Bank’s optimal finance policy:

FB
Dt+1 = LFt+1
B
, FB
Nt+1 = 0.

(ii) Optimal level of bank lending, LFt+1

B , defined by:

 FB  FB
Et Rl,t+1 = Rd,t+1 (10)

with
FB

α−1 1 1 F B −η
= αeat+1 LFt+1
B FB


Rl,t+1 and Rd,t+1 = − Dt+1 .
β β

Proposition 1 follows directly from the first order conditions to (9). The proofs of all propositions
are given in Appendix A. Because households value money-like securities, it is optimal to finance
all bank production with debt. The expression (10) pins down the first-best level of lending by
equating the marginal productivity of bank capital stock to the marginal cost of lending (Figure 2).
The social marginal cost of funding is equal to the rate of return at which a household is willing to
F B . It seems implausible that the social planner
hold deposits at the financial intermediaries, Rd,t+1
would find it optimal not to keep any equity on the bank’s balance sheet. As an extension, I can
introduce the social cost of bank default. In such a setup, the optimal level of a bank net worth,
F B > 0, would trade off reduced liquidity provision and reduced social cost of bank default.
Nt+1
I interpret the first-best level of lending, LFt+1
B , as an upper limit on the amount of positive NPV

loans (creditworthy borrowers) in the economy. The model captures endogenously that there are

15
 Figure 2: Socially Optimal Allocation

LF B
t+1

LF B
t+1

EtB [Rl,t+1 ]
EtG [Rl,t+1 ]
Rd,t+1
5 10 15 20 25 30
Lt+1
This figure depicts the social marginal cost of bank lending Rd,t+1 (black dash-dotted line) and the social marginal
benefit of lending during recessions EtB [Rl,t+1 ] (red solid line) and during expansions EtG [Rl,t+1 ] (blue solid line)
as a function of bank lending Lt+1 . The optimal level of lending is at the intersection of the social marginal costs
and benefits. The scale of y-axis is omitted, as the numbers in this parameterized example do not have economic
meaning. The parameter values are set to β = 0.98, η = 0.8, α = 0.85, δ = 1, ρa = 0.97, σa = 0.02, and σω = 0.2.

fewer good lending opportunities during economic slowdowns than during expansions, as I state in
Proposition 2.

Proposition 2 The socially optimal level of lending, LFt+1

B , is procyclical:

∂LFt+1
B
> 0.
∂at

Proposition 2 is established using the equilibrium condition (10) and an implicit function
theorem (see Appendix A). Since the marginal productivity of lending is higher in good times,
when the marginal cost is acyclical (Figure 2), it is optimal to maintain a higher level of production
during expansions. This naturally advocates for a less stringent government regulations during
periods of economic growth, when there are more lending opportunities.

4.2 Competitive Equilibrium with No Capital Regulation

To highlight the economic mechanism at the heart of the model, I now provide a detailed
characterization of the bank’s lending and capital structure choice in a competitive equilibrium.
In this section, suppose that financial intermediaries face no capital regulation or, equivalently,

16
 Figure 3: Competitive Equilibrium Allocation

Bailout LFt+1
B
wedge
LC E
t+1
LFt+1
B

LC E
t+1

Rd,t+1
Rd,t+1 − ξ B (Lt+1, Nt+1; at ) B
Rd,t+1 − ξ G (Lt+1, Nt+1; at ) G
5 10 15 20 25 30 5 10 15 20 25 30
Lt+1 Lt+1
Panel A: Panel B:
Panel A of this figure depicts the social marginal cost of bank lending Rd,t+1 (black dash-dotted line) and the private
marginal cost of lending during recessions Rd,t+1 − ξ B (Lt+1 , Nt+1 ; at ) (red dashed line) and during expansions
Rd,t+1 − ξ G (Lt+1 , Nt+1 ; at ) (blue dashed line) as a function of bank lending Lt+1 . Panel B additionally depicts the
marginal cost of lending Et [Rl,t+1 ] during recessions (red solid line) and expansions (blue solid line). The optimal
level of lending at social optimum is at the intersection of the social marginal costs and benefits. The optimal level of
lending in the competitive equilibrium is at the intersection of the private marginal costs and benefits. The social and
private marginal benefits are identical. The scale of y-axis is omitted, as the numbers in this parameterized example
do not have economic meaning. The parameter values are set to β = 0.98, η = 0.8, α = 0.85, δ = 1, ρa = 0.97,
σa = 0.02, and σω = 0.2.

ζt = 0, ∀t.
The banks’ first order conditions with respect to lending and the amount of equity financing
are, respectively, given by:
"Z ∗
#
ωt+1
ω
Et [Rl,t+1 ] = Rd,t+1 −Et (Rd,t+1 − e Rl,t+1 ) dΦ (ω) , (11)
0
"Z ∗ #
  ωt+1
1
− − Rd,t+1 −Et Rd,t+1 dΦ (ω) < 0, (12)
β 0

where the implicit (aggregate) rate of return on loans is equal to Rl,t+1 = αeat+1 Lα−1
t+1 and the
∗
bailout threshold is defined by eωt+1 +at+1 Lαt+1 = Rd,t+1 (Lt+1 − Nt+1 ). Φ (·) denotes the normal
distribution with mean − 21 σω2 and standard deviation σω .
The presence of government guarantees generates a bailout wedge in the bank’s cost of lending,

17
defined by: "Z #
∗
ωt+1
ξ (Lt+1 , Nt+1 ; at ) ≡ Et (Rd,t+1 − eω Rl,t+1 ) dΦ (ω) .
0

This bailout wedge captures the difference between the social and private marginal cost of lending
(see equations (10) and (11)). The bank does not fully internalize the risk costs, since those
are partially borne by the taxpayers. In particular, whenever the bank’s profits are hit by a
∗ , their credit liabilities are covered by the government.
sufficiently low idiosyncratic shock ω < ωt+1
An important property of the bailout wedge is that for a given level of lending it is decreasing in the
aggregate productivity, ∂ξ/∂at < 0 (Panel A of Figure 3). The economy is less likely to transition
to a recession next period if it is currently in an expansion, implying a lower bailout wedge during
periods of economic growth.
The expression (12) demonstrates that equity is relatively a more expensive source of funding
for banks than debt. The reasons for this are twofold. By holding equity on their balance sheets,
banks forgo government subsidies and, at the same time, relinquish the liquidity premium. As a
result, the financial intermediaries will tilt their capital structure towards debt and hold no equity
on their balance sheets.
The funding wedge is strictly positive ξ (Lt+1 , Nt+1 ; at ) > 0, implying that the social marginal
cost of lending is strictly greater than the private one. Therefore, there exists an excessive lending
in a competitive equilibrium when the banks face no capital regulations, LCE F B 16 It is a well
t+1 > Lt+1 .
documented in the banking literature, that government guarantees incentivize banks to risk-shift.
Contrary to the literature, in my model banks risk-shift in terms of quantity, rather than in terms
of quality. By issuing more loans, banks move further into the distribution of the projects’ quality
(borrowers’ creditworthiness), meaning that more loans result in a lower quality loan portfolio. I
lay out a characterization of the competitive equilibrium in Proposition 3.

Proposition 3 The competitive equilibrium with no capital regulation in place is characterized by:

(i) Optimal level of lending, LCE

t+1 , defined by:

 CE  CE
− ξ LCE CE


Et Rl,t+1 = Rd,t+1 t+1 , Nt+1 ; at , (13)

with
CE

α−1 1 1 CE −η
= αeat+1 LCE CE


Rl,t+1 t+1 and Rd,t+1 = − Dt+1 .
β β
16
ξ (Lt+1 , Nt+1 ; at ) > 0 as long as Lt+1 > 1, which ensures a positive rate of return on deposits.

18
 (ii) Bank’s optimal finance policy:

CE
Dt+1 = LCE FB
t+1 > Lt+1 ,
CE
Nt+1 FB
= Nt+1 = 0.

In the competitive equilibrium with the risk-shifting mechanism in place, the optimal level of
lending continues to be pro-cyclical (see Proposition 4). On the one side, lending is more productive
during expansions, calling for a higher level of investment. On the other side, bank cost of lending
goes up in good times because of a reduced bailout wedge, pushing the optimal level of lending
down (Panel B of Figure 3). Under log-normality of the aggregate productivity shock, the former
channel dominates the later and, hence, the procyclicality of the optimal level of lending follows.

Proposition 4 In a competitive equilibrium with no capital regulations in place, the optimal level
of lending is procyclical:
∂LCE
t+1
> 0.
∂at

Using capital regulations, the government will aim to restrict inefficient lending that arises in
a competitive equilibrium. And whether tight capital requirements will be more beneficial during
expansions or recessions will depend on how the level of excessive investment, LCE FB
t+1 /Lt+1 − 1,
behaves over the business cycle. Proposition 5 provides the condition under which the lending level
in a competitive equilibrium is more procyclical than the first-best allocation.

Proposition 5 The level of overinvestment is procyclical if and only if

∂ξ LCE CE


¯ t+1 , Nt+1 ; at
− ξa,t < < 0. (14)
∂at

The threshold value ξ¯a is defined in Appendix A. There are a number of channels in force that
indicate how banks’ risk-shifting incentives change over the business cycle. First, a decreasing
returns to scale production technology generates a scale effect: in proportionate terms banks risk-
shift, by the same amount across the states of the economy, but since in good times they risk-shift on
a larger scale, the overinvestment is higher in absolute terms. This mechanism can be demonstrated
in the setup in which the bailout wedge is proportionate to the deposit rate and the deposit rate is
constant. The second channel arises due to the fact that the bailout wedge is increasing in lending,
∂ξ/∂Lt+1 > 0, as shown in Panel A of Figure 3. The more loans that banks issue, the bigger is
the size of the government transfer in case a bailout happens. This property of the bailout wedge
results from the concave preferences for liquid assets and decreasing returns to scale in production.

19
On the one side, a higher level of investment during periods of economic growth implies a higher
deposit rate, since the economy is saturated with deposits and households are willing to relinquish
a smaller premium on safe assets. On the other side, decreasing returns to scale imply a lower
income that is forgone in a bailout when banks increase lending. A higher deposit rate and a lower
rate of return on loans, in turn, deliver a larger bailout wedge during expansions. Third, assuming
everything else is equal, the bailout wedge is decreasing in the productivity shock at , meaning that
the amount of inefficient lending is reduced in good times. The restriction (14) ensures that in the
equilibrium the bailout wedge or, equivalently, the banks’ risk-shifting motives are procyclical.

4.3 Competitive Equilibrium with Capital Regulation

Next suppose that equity needs be at least a fraction ζt > 0 of loans for banks to be able to operate.
Given that in my model equity is a relatively more expensive form of finance for banks than debt,
the capital constraint is binding in each period of time (see Proposition 6).

Proposition 6 The capital constraint is binding:

CE
Nt+1 = ζt LCE
t+1 ,
CE
Dt+1 = (1 − ζt ) LCE
t+1 .

To demonstrate what are the implications of capital regulation on bank cost of lending, I now
provide the banks’ first-order conditions with respect to lending and the amount of equity financing,
respectively, given by:

 CE  CE 1
− ξ LCE CE


Et Rl,t+1 = Rd,t+1 t+1 , Nt+1 ; at + λ t ζt , (15)
β
| {z }
equity wedge
"Z #
ω∗  
1 CE 1 CE
λt = Et Rd,t+1 dF (ω) + − Rd,t+1 , (16)
β 0 β

where λt is a Lagrange multiplier on the capital constraint. Imposing capital regulation introduces
an equity wedge into the bank’s cost of funding. The first term of equation (16) captures that a
better capitalized bank forgoes the government subsidies and, as a result, faces higher lending costs.
A positive level of capital requirement also means that the bank’s supply of deposits is reduced,
implying a higher liquidity premium and, as a result, a higher equity wedge, as captured by the
second term of equation (16).
To better understand the costs and benefits of capital regulation, it is useful to decompose

20
 Figure 4: Effects of Capital Regulations on the Bank’s Lending Costs

Risk-shifting cost of lending: ξ˜ (·)

Rd,t+1
1 2
ζt β1 − Rd,t+1
Liquidity cost of lending: θ (·)
0% 20% 40% 60% 0% 20% 40% 60%
ζt ζt
Panel A Panel B
Panel A of this figure depicts the risk-shifting costs of bank lending as a function of capital requirements ζt (holding
lending and productivity fixed). Panel B depicts the liquidity costs of bank lending (black solid line) and its two
components: (i) the deposits rate (blue dashed line) and (ii) the equity wedge (red dash-dotted line) as a function of
capital requirements ζt . The scale of y-axis is omitted, as the numbers in this parameterized example do not have
economic meaning. The parameter values are set to β = 0.98, η = 0.8, α = 0.85, δ = 1, ρa = 0.97, σa = 0.02, and
σω = 0.2.

the bank’s funding costs into (i) the liquidity channel and (ii) the risk-shifting channel, which are,
correspondingly, defined by:
 
1
θ (Lt+1 , ζt ) ≡ Rd,t+1 + ζt − Rd,t+1 ,
β
"Z ∗ #
ωt+1
ξ˜ (Lt+1 , ζt ; at ) ≡ Et ((1 − ζt ) Rd,t+1 − eω Rl,t+1 ) dΦ (ω) .
0

The obvious benefit of tightened capital requirements is reduced risk-shifting incentives for banks.
As shown in Panel A of Figure 4, the bailout wedge adjusted for the presence of capital regulations
ξ˜ decreases as capital regulations are tightened, implying that less inefficient firms are financed.
∗ , as well as the size
There are two reinforcing mechanisms in place: the bailout threshold ωt+1
of a government transfer in case of a bank failure ((1 − ζt ) Rd,t+1 − eω Rl,t+1 ), falls as capital
requirements become more stringent.
The level of the liquidity costs of lending θ is mainly determined by the amount of deposits
supplied to households. A higher capital requirement means that the supply of liquid assets is

21
reduced and, as a result, a liquidity premium is magnified. From the bank’s perspective, this implies
a lower required rate of return on debt, but a higher equity wedge (Panel B of Figure 4). As long as
η < 1, the liquidity cost of lending is strictly increasing in ζt . Overall, capital regulations translate
into a heightened lending costs for banks, suggesting that excessive lending can be restrained by
imposing a sufficiently high capital requirement.

4.4 Optimal Capital Requirements

The goal of a social planner when choosing an optimal level of capital requirements is to dampen
bank’s risk-shifting incentives, but without restricting the bank’ supply of high-quality loans and
deposits. As a matter of fact, it is not feasible to restore both first-best level of lending and liquidity
provision at the same time with only the help of capital requirements. Suppose for now that the
only goal of the regulator is to restore a socially optimal lending level and to achieve this goal the
capital requirement is set to ζtL > 0, which is the “lending capital requirement”. This translates
ζ L
into a level of liquidity provision Dt+1 that is below the socially optimal lending level:

ζL
= 1 − ζtL LFt+1

B
Dt+1 < LFt+1
B FB
= Dt+1 .

Similarly, consider a social planner who aims to recover a socially optimal level of deposits and
imposes a capital requirement of ζtD , which is the “liquidity capital regulation”. This capital
requirement, however, allows inefficient lending from a social perspective:

ζD
D Dt+1 LFt+1
B
Lζt+1 = = > LFt+1
B
.
1 − ζtD 1 − ζtD

To provide the intuition behind the design of the optimal capital regulations, I first solve for the
lending and liquidity capital requirements in the baseline model. This allows me to demonstrate
the cyclical properties of capital regulation in a fairly simple way.

4.4.1 Lending Capital Requirement

Suppose now that the only goal of the social planner is to rule out excessive bank lending. To do
so, she institutes the lending capital requirement, defined in Proposition 7.

L
Proposition 7 The first-best level of bank lending, Lζt+1 = LFt+1
B , is restored when a capital

22
 Figure 5: Lending Capital Requirement

FB
B: θ (·) − Rd,t+1
FB
G: θ (·) − Rd,t+1
17%
B: ξ˜ (·)
G: ξ˜ (·)
16%

ζtL ζ L 15%
t

14%

13%
ζtL
5% 10% 15% 20% 25% 30% 0.1 0.15 0.2 0.25 0.3 0.35 0.4
ζt at
Panel A Panel B
Panel A in this figure depicts (i) the benefit of capital requirements during recessions (dash-dotted red line) and
during expansions (dash-dotted blue line) and (ii) the cost of capital requirements during recessions (solid red line)
and during expansions (solid blue line) as a function of ζt . Panel B depicts the liquidity capital requirement as a
function of the aggregate productivity. The scale of y-axis is omitted, as the numbers in this parameterized example
do not have economic meaning. The parameter values are set to β = 0.98, η = 0.8, α = 0.85, δ = 1, ρa = 0.97,
σa = 0.02, and σω = 0.2.

requirement is set to ζtL , defined by:

θ LFt+1
B L FB
= ξ˜ LFt+1
B L



, ζt − Rd,t+1 , ζt ; at . (17)

The expression (17) equates the bank’s private cost of lending to the social one, ensuring that
the bank chooses the first-best level of lending when it encounters a capital requirement of ζtL .
The right-hand side of the equation (17) captures a key benefit of capital requirements – reduced
risk-shifting incentives – depicted in Panel A of Figure 5 by the dash-dotted lines. The tighter the
capital regulations the lower the expected government subsidies and, hence, the lower the bank’s
overinvestment. As discussed earlier, a capital regulation that is too restrictive can amplify the
liquidity costs of lending via an equity wedge, leading to reduced liquidity provision and potentially
restricting the funding of high-quality projects. Panel A of Figure 5 shows that the spread between
the private liquidity cost of lending and the social one depicted by the solid lines is increasing in ζtL .
The primary question I address in this paper is how optimal bank capital requirements behave over
the business cycle. This essentially means answering whether the risk-shifting incentives for banks

23
are stronger in good or bad economic environments, and whether equity financing is more costly
during expansions or recessions.
˜ Differentiating this
The risk-shifting incentives for banks are defined by the bailout wedge ξ.
wedge with respect to the productivity shock,

dξ˜ LFt+1
B , ζ L; a ∂ ξ˜ LFt+1
B , ζ L; a ∂ ξ˜ LFt+1
B , ζ L; a



FB
t t t t t t ∂Lt+1
= + ≶ 0,
dat ∂at ∂LFt+1 B ∂at
<0 | {z }| {z }
>0 >0

demonstrates that there are two competing effects in place. On the one hand, the probability of a
bank bailout decreases with the productivity shock, implying a smaller wedge in the bank’s lending
costs. On the other hand, during periods of economic growth there are more lending opportunities,
indicating that more projects will be funded in the first best. Because the bailout wedge expands
when lending is higher, the risk-shifting motives become stronger during expansions. Panel A of
Figure 5 shows that under chosen parameter values, the latter channel dominates the former and, as
a result, the bailout wedge is procyclical, suggesting that a higher capital requirement is necessary
during periods of economic growth.
Households’ preference for safe assets implies that equity is costly, especially during recessions.
To the extent a bank’s balance sheet shrinks during a recession, less safe assets are created by banks
and the economy is less saturated with liquidity. This translates into a higher liquidity premium
and, as a result, a higher liquidity cost of lending. Panel A of Figure 5 shows that for a given level
of ζt , the cost of capital requirements is countercyclical, suggesting that capital charges should be
reduced during economic slowdowns.
The two effects - the countercyclical cost and procyclical benefit of capital regulation - reinforce
each other and deliver a capital requirement, ζtL , that is high in times during periods of economic
growth and low during downturns (Panel B of Figure 5).

4.4.2 Liquidity Capital Requirement

Consider now a type of capital regulation designed to restore the first-best level of liquidity provision
(i.e., the first-best level of deposit supply). Proposition 8 provides a characterization of the liquidity
capital requirement.

ζ D
Proposition 8 The first-best level of deposits, Dt+1 F B , is achieved, when the social planner
= Dt+1

24
sets the capital requirement equal to ζtD , defined by:
! !

α−1 LFt+1
B
α−1 LFt+1
B
1− ζtD θ D
, ζtD − FB
Rd,t+1 = 1− ζtD ξ˜ D
, ζtD ; at . (18)
1 − ζt 1 − ζt

As opposed to the lending capital requirement, the liquidity capital requirement allows inefficient
ζD ˜ as well as a
bank lending in the amount of t D LF B . This results in a larger bailout wedge ξ,
1−ζt t+1
higher liquidity cost of lending θ, since the economy is more saturated with safe assets. Importantly,
when a bank’s balance sheet grows, the rate of return on loans drops due to a decreasing returns
to scale. This loss in a bank’s productivity can be viewed as an additional cost that arises with the
liquidity capital requirement, and it is captured by the factor (1 − ζt )α−1 ≥ 1.
When setting the level of liquidity capital requirement, the underlying trade-off remains the
same. On the one side, tight capital regulations restrict banks’ risk-shifting incentives as captured
by the right-hand side of the equation (18). But, on the other side, it increases banks’ lending costs
via the liquidity channel and, as a result, may lead to a reduction in bank loans and deposits as
captured by the left-hand side of the equation (18). Similar to the case with the lending capital
requirement, stringent capital ratios are of most use during periods of economic growth, but are
also less costly, delivering a liquidity capital requirement that is procyclical.
Both lending and liquidity capital requirements focus only on one dimension of the problem –
either dampening banks’ incentives to risk-shift or stimulating banks’ creation of liquid assets. In
order to ensure highest lifetime utility of households in this economy, I solve for an optimal Ramsey
policy that balances reduced inefficient lending and reduced liquidity provision. The Ramsey capital
requirement is characterized in Section 5.

5 Quantitative Assessment

In this section, I quantify policy recommendations on what defines optimal time variation in
bank capital requirements. First, I extend the baseline model and calibrate it to a fixed capital
requirement to match key macroeconomic quantities, as well as their counterparts in the bank data.
Next, I solve for the optimal policy rule using the chosen parameter specification and characterize
it in terms of the relevant macroeconomic aggregates.

25
5.1 Quantitative Model

To best match the data, I generalize the households’ preferences in the quantitative model, as well
as introduce a bank-independent production sector.
Household sector. In the model, households have constant relative risk aversion (CRRA)
preferences with the risk-aversion coefficient γ. The households’ utility is defined over consumption
and deposit holdings according to a constant elasticity of substitution (CES) aggregator:

 η̃−1 η̃−1
 η̃
η̃−1
η̃ η̃
v (Ct , Dt+1 ) = Ct + χDt+1 ,

where χ is a share parameter and η̃ is the elasticity of substitution between consumption and
liquidity services.
The households’ first-order conditions imply that the rate of return on bank debt satisfies:

 − 1
Dt+1 η
Et [Mt,t+1 Rd,t+1 ] = 1 − χ , (19)
Ct

where Mt,t+1 is the stochastic discount factor equal to:

  1 −γ  − 1
v (Ct+1 , Dt+2 ) η̃ Ct+1 η̃
Mt,t+1 = . (20)
v (Ct , Dt+1 ) Ct

Similar to the baseline case, there is a discount on the deposits rate compared to the rate of return
on any other safe asset.17 In the quantitative model, the liquidity premium depends on the ratio
of deposit holdings and consumption, rather than just the level of deposit holdings. This helps to
capture the idea that higher consumption increases the marginal utility of liquidity of households.
Production sector. By and large, the economy’s production sector is reliant either on bank
debt or access to capital markets to undertake production projects. To the extent that some
borrowers lack access to capital markets, typically due to informational asymmetries, agents can
be broadly classified into two groups. The first group is small businesses and households, who have
their financial needs fulfilled by bank loans and mortgages. The second group is borrowers who have
access to public markets and use multiple forms of debt financing. To capture this heterogeneity
in borrowers, the model now includes both a banking sector and a bank-independent sector. The
bank-independent sector includes firms with access to capital markets.
For the most part, the banking sector remains the same in the quantitative as in the baseline
17
In the quantitative model, the rate of return on any other safe asset other than bank debt is equal to 1/Et [Mt,t+1 ].

26
model. Each bank j ∈ Ω operates a sector-specific production technology:

αb
eωj,t +āb +at lj,t ,

where the aggregate productivity shocks follows an AR(1) process:

at = ρa at−1 + σa t , t ∼ iid N (0, 1) .

The constant āb is introduced to better match the fraction of output produced in the banking
sector. The bank-specific shock is iid across time and banks, with the law of motion given by:

1
ωj,t = − σω (at )2 + σω (at ) εj,t , εj,t ∼ iid N (0, 1) .
2

Consistent with Bloom, Floetotto, Jaimovich, Saporta-Eksten and Terry (2012), idiosyncratic
productivity risk varies countercyclically with the aggregate productivity; specifically, I choose
σω (at ) = σω e−νat .
The capital in the banking sector depreciates at rate δ and accumulates according to:

lj,t+1 = (1 − δ) lj,t + ib,j,t .

Further, the financial intermediaries incur an operating cost ob per unit of intermediated assets.
One could think of this cost as the resources spent to monitor bank borrowers. As before, the
banks finance production using a mixture of equity and debt financing. Since there is neither
equity issuance costs nor adjustments costs to capital, the equilibrium continues to be symmetric
and all banks make identical decisions lj,t+1 = Lt+1 , nj,t+1 = Nt+1 , and dj,t+1 = Dt+1 , ∀j ∈ Ω.
Hence, the output produced in the banking sector equals to Yb,t = eāb +at Lαt b .
The bank’s first-order conditions imply that the capital constraint is binding, while the optimal
level of lending satisfies:

Et [Mt,t+1 (Rl,t+1 + (1 − δ − ob ))] = θt − ξ˜t ,

27
where the liquidity and risk-shifting costs of lending are, respectively, defined by:

θt = Et [Mt,t+1 Rd,t+1 ] + ζt (1 − Et [Mt,t+1 Rd,t+1 ]) , (21)

" Z ω∗ !#
ξ˜t = Et Mt,t+1 ω
((1 − ζt ) Rd,t+1 − e Rl,t+1 − (1 − δ − ob )) dΦ (ω) . (22)
0

with the loan rate being equal to

Rl,t = eāb +at Lαt b −1 . (23)

The firms in the bank-independent sector similarly employ a decreasing returns to scale and
α
produce consumption goods, using capital Kf,t as the only input, Yf,t = eat Kf,tf . To form output,
the firms rent out capital from the household sector at the rate of return:

α −1
Rk,t = αf eat Kf,tf . (24)

The households are the owners of the financial intermediaries, as well as of the firms in the
bank-independent sector. Therefore, at the end of each period, they receive proceeds from the
banks’ activity Zt , as well as the firms’ profits Πt . Because the households own the capital used
in production in the bank-independent sector, they are the ones who make the capital investment
choices. To invest in capital, households must pay a small operational cost of . The capital in the
non-banking sector depreciates at the same rate as in the banking sector and accumulates according
to:
Kf,t+1 = (1 − δ) Kf,t + If,t .

5.2 Calibration

Data for the aggregate sector - the banking and non-banking sectors - comes from the Financial
Accounts of the United States Z.1 issued by the Board of Governors of the Federal Reserve18 and
from National Income and Product Accounts (NIPA) from the Bureau of Economic Analysis.19 In
particular, the output produced in the total economy, Y , is measured using the income approach as
the gross value added by all sectors, except Financial Business.20 Similarly, I exclude the financial
18
https://www.federalreserve.gov/releases/z1/.
19
http://www.bea.gov/itable/index.cfm.
20
The sectors included are Households and Nonprofit Institutions Serving Households, Nonfinancial Noncorporate
Business, Nonfinancial Corporate Business and Government. The data is collected from the Financial Accounts, S.2.
Selected Aggregates for Total Economy and Sectors

28
sector when measuring gross fixed capital formation, I, and stock of fixed assets, K.21 Consumption
is defined as the sum of expenditures on non-durable goods and services. All the quantities are
deflated with the Consumer Price Index and normalized by the NIPA population. The data on
CPI inflation are from the Bureau of Labor Statistics, with the price level being normalized to 1
as of December of 2009.
The challenging part of mapping the model to the data is to differentiate the data on output
into the value added by the banking sector and the bank-independent sector. In the model,
the final goods produced by financial intermediaries, Yb , are viewed as the value added by bank
borrowers (predominantly, small businesses and households) and, consequently, is measured as
the the gross value added by Households and Nonprofit Institutions Serving Households and
Nonfinancial Noncorporate Business sectors.22 The output produced by borrowers with access
to capital markets, Yf , is measured as the gross value added by Nonfinancial Corporate Business
and Government sectors.23 As documented in Rauh and Sufi (2010), corporate businesses both use
bonds and utilize bank debt as funding sources. In this sense, I overestimate the value-added by
the bank-independent sector and, as a result, underestimate the costs associated with tight capital
requirements, as reduced lending translates into a smaller loss in the aggregate output. According
to this measure the banking sector accounts for about 28% of the output produced in the economy.
I pursue the same split between borrowers when measuring gross fixed capital formation and stock
of capital in the two-sectors.
The model is calibrated at annual frequency for the 1980-2008 period. I exclude the post Great
Recession period to avoid zero-lower bound concerns. The parameters are calibrated to satisfy the
steady-state conditions of the model, along with the target first and second moments in the data.
Long run behavior. The subjective discount factor is set to β = 0.975, which is a value
traditionally used in macroeconomics. The Cobb-Douglas function parameters αb and αf are
chosen to match a capital-output ratio in the banking and non-banking sectors, respectively. In
the data, a capital-output ratio of the financial intermediaries (or, equivalently, of firms in the
banking sector) is equal to 4.9, which is roughly twice as high as a capital-output ratio of the
bank-independent firms (see Appendix Table B2). As a result, the αb = 0.78 decreasing return to
scale parameter for the banks is larger than the αf = 0.355 corresponding for the firms. The scalar
21
The data on investment are from the Financial Accounts, S.2. Selected Aggregates for Total Economy and
Sectors, while the data on capital stock are collected from the NIPA Fixed Assets tables, specifically, Table 6.1.
Current-Cost Net Stock of Private Fixed Assets by Industry Group and Legal Form of Organization and Table 7.1A.
Current-Cost Net Stock of Government Fixed Assets
22
The data are from the Financial Accounts, S.2. Selected Aggregates for Total Economy and Sectors.
23
The data are from the Financial Accounts, S.2. Selected Aggregates for Total Economy and Sectors.

29
āb is set so that a ratio of loans to a total stock of capital in the model is identical to the one in
the data. By construction, I also match a market fraction of the banking sector in terms of the
produced output, as well as a capital-output ratio of the aggregate sector. The capital depreciation
rate is set to δ = 0.075 to produce a plausible value for the model-implied ratio of investment-to-
capital. The operating cost in the banking sector ob is calibrated to match a bank’s profit-to-loan
ratio. The bank data stems from the aggregated regulatory filings, the so-called call reports, as
reported by the Federal Deposit Insurance Corporation (FDIC).24 A bank’s loans are measured as
the total assets net of the securities, fixed assets,and cash and due to depository institutions, while
a bank’s profits are measured as the total interest and non-interest income net of the securities
interest income and total interest expense. The operating cost in the bank-independent sector is
determined using the standard deviation of firms’ capital investment.
Productivity shocks. Estimates of Solow residuals typically imply a highly persistent AR(1)
process for the log productivity at . I use a standard value of 0.95 for the autoregressive coefficient
√

and choose a σa of 0.02 4 · 0.01 to match the standard deviation of the aggregate output in the
data.
Idiosyncratic shocks. I use the parameter σω to determine both the risk choice of banks
(i.e., level of inefficient lending) and, for the most part, the fraction of banks being bailed out each
period. To the extent that there is no default in the model, I define the bailout rate as a fraction
of failed banks, both assisted and not assisted by the government. The data on bank failures are
from the fail bank list issued by the FDIC.25 On average, 0.76% of all insured institutions fail each
year, delivering σω equal to 0.325. The parameter µω is set to normalize the mean of idiosyncratic
shock ω to 1. The level of ν is chosen to generate the range of the idiosyncratic volatility dispersion
σω2 (at ) consistent with the empirical evidence in Bloom et al. (2012). Specifically, I target a 15%
increase of the volatility dispersion in recessions relative to expansions. This is on the conservative
end of the values reported in Bloom et al. (2012).
Capital adequacy ratios. Even though my aim is to characterize optimal capital regulation
that varies over the time, the model is calibrated with a fixed level of capital requirement in place, ζ̄.
In accordance with Basel III capital standards, FDIC-insured depository institutions are required
to maintain the leverage ratio (Tier 1 capital divided by total assets) of 4% and the Tier 1 capital
ratio (Tier 1 capital divided by total risk-weighted assets) of 6%.26 Obviously, these two ratios
24
https://www5.fdic.gov/idasp/advSearch warp download all.asp?intTab=4.
25
https://www5.fdic.gov/hsob/SelectRpt.asp?EntryTyp=30&Header=1&tab=bankFailures.
26
Core capital elements (Tier 1) consists of common stockholders’ equity capital, noncumulative perpetual preferred
stock, and minority interests in the equity capital accounts of consolidated subsidiaries.

30
are different in the data, but are the same in the model. On average, depository institutions hold
7.26% of total assets and 10.20% of total risk-weighted assets in terms of core capital27 . In the
benchmark calibration, I set ζ̄ to 7.26%.
Liquidity premium. The financial intermediaries play a special role in creating liquidity;
they transform illiquid long-term assets into liquid short-term liabilities that offer non-pecuniary
services to investors. In accordance with this view, when mapping D to the data, I aim to capture
the short-term part of bank debt that is used to finance risky assets. Following Krishnamurthy and
Vissing-Jorgensen (2015), this measure of bank debt equals the sum of all deposits and the other
forms of short-term debt categories net of short-term assets and holdings of Treasuries.28 Further,
I assume that the short-term Treasuries and bank debt are perfect substitutes and, as a result,
require the same rate of return. Under this assumption, the liquidity premium is constructed as
the spread between the rates of return on the 3-month AA Commercial Paper and the 3-month
Treasury bill, similar to Krishnamurthy and Vissing-Jorgensen (2012).29,30
The preference of households for bank debt is governed by a share parameter, χ, and by elasticity
of substitution between consumption and liquidity services, η̃. The level of η̃ is set to match the
volatility of a ratio of the bank debt to consumption. Given the value of η̃, the scalar χ is set so
that the steady-state level of the liquidity premium is equal to 0.57% (annually).
In Table B1 in Appendix B, I summarize the benchmark configuration of the model parameters.
Tables B2 and B3 report the implications of the chosen parameters for the first and second
unconditional moments of the key variables. Overall, the model can successfully match a number
of moments for the main aggregates. The model has a hard time matching the volatility of bank
lending and, because of the binding capital constraint, also the volatility of bank debt. In the
model, bank lending is a stock variable and is quite persistent, therefore having rather modest
27
The data on capital adequacy ratios are collected from the “Ratios by Asset Size Group” table issued by the
FDIC. Due to the availability of the data, I calculate the average leverage ratio for the sample period 1984 - 2008
and the average Tier 1 capital ration for the sample period 1990 - 2008.
28
An alternative approach is to map D to “core” deposits, which is the largest source of bank funding and the
major source of safe and liquid assets for households (see, for example, Drechsler, Savov and Schnabl (2016)). Core
deposits include checking and money market accounts, as well as small time deposits. The cyclical properties of this
measure, however, change over the sample period of 1960-2015: while deposits exhibit strong positive correlation
with output growth in the earlier part of the sample, they become mildly countercyclical in the later one. Yet, over
the sample period I consider in this paper deposits co-move positively with output overall. Nonetheless, I focus on a
broader notion of the liquidity supply as it better captures the role of banks in creation of safe and liquid assets.
29
For the period of 1997-2008, the rate of return on Commercial Paper is calculated as the average of the rate of
return on Financial and Non-Financial Commercial Paper.
30
Other potential measures include (i) the spread between the rates of return on the Resolution Funding Corporation
(Refcorp) bonds and the Treasury bills (Longstaff (2002)), and (ii) the spread between the interest rates on
general collateral repurchase agreement (GC repo) and the Treasury bills (Nagel (2014)). These measures produce
quantitatively similar results.

31
fluctuation over the business cycle. The low volatility of bank debt, in turn, translates into low
volatility of the liquidity premium. As a potential remedy, I can introduce a stochastic process
for one of the parameters governing a demand function for safe assets (χ or η). For parsimony
of the results, for now I consider the calibration without liquidity shocks. As reported in Table
B4, the model generates the correct signs for the business cycle correlations, but overstates their
magnitudes. This is a common characteristic of the standard real business cycle model with one
shock in the economy.
To understand the model dynamics, I now analyze the impulse response functions of key
quantities to a one standard deviation positive total factor productivity (TFP) shock. As shown in
Appendix Figure B1, following a positive shock, lending in the banking sector increases immediately
by 1% and continues to grow for a number of years. This is the case because lending opportunities
improve. An increase in lending, in turn, is accompanied by an increase in deposits (0.03% increase
in the deposits-to-consumption ratio) and, as a result, a reduction in the liquidity premium.
This result reaffirms the conjecture in the baseline model that equity financing is relatively less
expensive during expansions than recessions. Moreover, this is line with the empirical evidence
that the liquidity premium is countercyclical (see Table B4). At the same time, a positive shock to
productivity results in a reduced probability of default in the current period. But in the subsequent
periods, the probability of default spikes up. This is in response to a heightened risk-shifting by
banks, as captured by the impulse response function of the bailout wedge. Note that the bailout
wedge is the expected (not realized) bailout subsidy at time t + 1, conditional on time t information
and, most importantly, on the time t choice of lending Lt+1 .

5.3 Ramsey Equilibrium

The Ramsey problem characterizes a set of allocations that can be implemented as a competitive
equilibrium when banks face the prescribed capital requirements. A Ramsey planner maximizes the

32
lifetime utility of households subject to the resource constraint and implementability conditions:

∞
" #
X
t
argmax E β u (Ct , Dt+1 )
Ct ,Lt+1 ,Nt+1 ,Dt+1 ,Kf,t+1 t=0
v (Ct , Dt+1 )1−γ
with u (Ct , Dt+1 ) = −1
1−γ
α
s.t. Ct = Rd,t Dt − Dt+1 + Rk,t Kf,tf − If,t − of Kf,t + Zt + Πt − Tt ,
Et [Mt,t+1 (Rk,t+1 + (1 − δ − of ))] = 1,
Nt+1 + Dt+1 = Lt+1 ,
Et [Mt,t+1 (Rl,t+1 + (1 − δ − ob ))] = θt − ξ˜t ,
Ct + Ib,t + If,t = Yb,t + Yf,t − ob Lt − of Kf,t ,

where the taxes and net proceeds from banks’ and firms’ activity are, respectively, equal to:
Z 1
max Rd,t Dt − eωj,t +at Lαt b − (1 − δ − ob ) Lt , 0 dj,

Tt =
0
Z 1
zj,t = max eωj,t +at Lαt b + (1 − δ − ob ) Lt − Rd,t Dt , 0 − Nt+1 ,

Zt = zj,t dj, with
0

Πt = Yf,t − Rk,t Kf,t .

The rate of return on deposits, the stochastic discount factor, the liquidity and risk-shifting cost of
lending, the rate of return on loans and capital are defined by (19), (20), (21), (22), (23) and (24),
correspondingly. I provide further details in Appendix C. The optimal Ramsey policy – a capital
n o∞
requirement that implements the Ramsey allocations Ct∗ , L∗t+1 , Nt+1
∗ , D∗ , K ∗
t+1 f,t+1 – requires
t=0
∗ /L∗ . The set of Ramsey allocations and policies constitute
a capital ratio that equals ζt∗ = Nt+1 t+1
a Ramsey equilibrium.
Using the chosen configuration of parameters (Table B1), I solve for the Ramsey equilibrium.
Not surprisingly, the optimal policy rule for bank capital ratios ζt∗ = ζ (St , St−1 ) is a function
of all state variables St = {ζt−1 , Lt , Kf,t , at } in the model, including the previous period capital
requirement, capital stock in the two sectors, and the productivity shock. I find that the Ramsey
capital requirement, for the most part, fluctuates between 4% and 6%, with the mean at 5%. This
range of values is based on one standard deviation of the Ramsey capital requirement above and
below its mean, and is roughly consistent with an upper bound of the CCyB proposed by the Basel
III Accord. Importantly, the optimal policy rule is not capped at 6%, and in periods of abnormal
economic growth can go far above 6%. The steady state value of ζt∗ is substantially lower than the

33
Tier 1 leverage ratio observed in the data, but at the same time 1% higher than the level prescribed
by the Basel III Accord. As shown in Figure B2, the effect of a positive TFP shock on the optimal
policy rule is quite persistent: a one standard deviation positive TFP shock increases the capital
ratio by 0.2% immediately and this effect lasts for a number of periods.
To uncover the implications of the capital regulation that varies over time, I now examine
the impulse response reactions to a one standard deviation positive TFP shock in a Ramsey
equilibrium. Figure B2 demonstrates that the Ramsey capital requirement indeed helps to reduce
the procyclicality of bank lending, by dampening overinvestment during expansions and boosting
bank credit during downturns. This reduced overinvestment, in turn, translates into a lower bailout
wedge and, consequently, a lower probability of default. At the same time, the optimal policy rule
dampens the procyclicality of the deposits, as there are two channels in place. The first channel is
via bank lending directly: to the extent that a bank’s balance sheet may shrinks during recessions,
the supply of deposits contracts, but less so in a Ramsey equilibrium. The second channel is via
capital regulation itself: for a given amount of bank lending, in a Ramsey equilibrium, a larger
fraction of production is financed with bank debt during recessions as compared to a competitive
equilibrium with a fixed capital requirement. In this instance, the optimal policy rule goes down
in a response to a negative productivity shock.

5.4 Optimal Policy Rule

For effective policy recommendations, it is useful to express the Ramsey capital requirement
ζ (St , St−1 ) as a function of key observable macro and bank quantities. To do so, I perform the
following exercise: I simulate the model with the Ramsey optimal policies in place for 1, 000 times.
Then, I estimate a regression of the Ramsey capital requirement on a number of model aggregates
for each simulated sample:
N
ζxj xjt + εt .
X
ζt∗ = ζss +
j=1

Admittedly, this policy rule is forward-looking: ζt∗ is a level of capital requirement set at time t,
but regulating a bank’s choice of capital structure at time t + 1. The results in Panel A of Table B5
demonstrates that credit gap used as the only indicator to adjust the capital charges over the
business cycle does not capture the dynamics of the optimal policy rule (R2 = 14%). I find that
that using the credit gap along with GDP (in log deviations from their steady states) delivers an

34
R2 = 99.98%. In particular, ζt∗ is characterized by the following policy rule:31
 
Lt Lss
4.9% + 3 × log − log + 8 × (log Yt − log Yss ) .
Yt Yss

In particular, a one percentage increase in the credit gap leads to a 0.03% increase in capital charges,
while a one percentage point increase in GDP translates into a 0.08% increase. Equivalently, a one
standard deviation increase in the credit gap leads to a 0.1% increase in capital requirements, while
a one standard deviation increase in GDP translates into a 0.7% increase (Panel B of Table B5).
This finding aligns with my main model prediction: in periods when banks are incentivized to risk-
shift, that is in periods of high GDP and credit growth, the optimal capital requirement is increased.
Interestingly, the liquidity premium plays little role if included as an additional explanatory variable
in a regression (see Panel A of Table B5). However, this result does not continue to hold if the
model includes liquidity shocks.

5.5 Welfare Implications

Following the classic work of Lucas (1987), I calculate the potential welfare benefits (costs)
associated with different types of capital regulation. To do so, I denote the expected lifetime
utility of households in a world with capital regulation ζt by
" +∞ #
1−γ
X
t C̃t −1
Eζt β ,
1−γ
t=0

where the aggregate consumption is defined by C̃t = v (Ct , Dt+1 ), accounting for the value of liquid
assets to the households. The question is how much the households would be willing to pay or have
to be compensated to move from the economy with a fixed level of capital requirement ζt = ζ 0 , ∀t
to the world with a capital ratio of ζ 1 or to the economy where banks face the optimal Ramsey
policy ζt∗ . This compensating variation $ is defined as:
  1−γ 
+∞ (1 + $) C̃ − 1
" +∞ #
X t X C̃ 1−γ − 1
Eζ 0  βt  = Eζ 1 βt t ,
 
1−γ 1−γ
t=0 t=0

31
The GDP growth variable is orthogonalized with respect to the credit gap variable, since the two indicators are
strongly positively correlated.

35
 Figure 6: Welfare Implications
0.04

0.03

0.02

0.01

−0.01

−0.02

−0.03
4% 5% 6% 7% 8%
ζ
The Figure depicts the variation compensation, $, as a function of different levels of fixed capital charges (green
dots), as well as the welfare compensation corresponding to the Ramsey policy (blue square) and the policy solely
based on the credit gap (red diamond). The variation compensation is expressed in percentage points.

or, equivalently:
 hP i 1
+∞ t 1−γ 1−γ
Eζ 1 t=0 β C̃t
$= hP i − 1.
+∞ t 1−γ
Eζ 0 t=0 β C̃t

Unlike the Lucas cost of business cycle, the above measure captures the changes in consumption
instability across different economies, as well as the changes in steady state levels of consumption.
I provide more detains on the calculation of welfare costs in Appendix D.
Figure 6 depicts the variation compensation, $, as a function of different levels of fixed capital
charges, ζ 1 ∈ [0, 1], as well as the welfare result for the Ramsey economy. As the benchmark, I
choose a world with an optimal fixed level capital requirement, which happens to be the stochastic
steady state of the Ramsey policy. Moving from the current level of capital ratio 7.26% to the
Ramsey policy translates into a 0.06% permanent increase in the annual consumption of households,
although the exact magnitude of the gain depends on the attitude of consumers towards risk.32
Two-thirds of this welfare improvement is attributed to a transition from the optimal fixed capital
requirement to the time-varying one. Admittedly, the welfare gain from implementing the policy
rule that is solely based on the credit gap is significantly smaller as compared to the welfare gain
from having the optimal capital ratios. Implementing the optimal Ramsey policy leads to 15.5%
32
This is within the range of values documented by Lucas (1987), Alvarez and Jermann (2004), and Van den Heuvel
(2008).

36
decrease in the cyclicality of the bank credit (relative to a world with an optimal fixed level capital
requirement): the banks engage in less excessive lending during periods of economic growth, but
at the same time avoid contraction of lending activities during recessions. This in turn translates
on average into a higher level of lending (+0.93%) and consumption (+0.11%). In addition, the
increase in bank lending stimulates banks’ creation of liquid assets (+0.88%). The time-varying
capital ratios also lead to a reduction in the volatility of both lending (−0.13%) and deposits
(−0.17%), but to an increase in the consumption volatility (+0.08%).

5.6 Model with Liquidity Shocks

For robustness, I also consider a model with liquidity shocks. Namely, suppose that a share
parameter in the CES aggregator over consumption and deposits follows an AR(1) process:

log (χt ) = (1 − ρχ ) χ̄ + ρχ log (χt−1 ) + σχ ut , ut ∼ iid N (0, 1) .

The χ̄ = 0.01 parameter is set to the value of a share parameter used in the benchmark calibration
of the model without demand chocks. The standard deviation of the demand shock σχ = 0.1 is
calibrated to better match the volatility of the liquidity premium. I choose a persistence parameter
of ρχ = 0.95. As shown in Appendix Figure B4, a one standard deviation liquidity shock leads to
a 0.1% decrease in the optimal capital ratio. A positive liquidity shock means that the households
enjoy holding liquid assets more and, as a result, are willing to accept a lower rate of return on
deposits, as captured by the impulse response function of the liquidity premium. For a given capital
requirement, an increased liquidity premium translates into lower liquidity costs of lending. At the
same time, banks’ incentives to risk-shift become weaker, since the deposit rate is lower and so are
the banks’ credit liabilities. The liquidity channel dominates the risk-shifting channel, therefore
bank lending increases mildly to satisfy the households’ demand for liquidity.
If I continue to use the GDP and bank credit as the only two indicators to guide the dynamics
of capital requirements, the R2 drops to 91% (see Panel A of Table B6). Including the liquidity
premium as an additional indicator increases the R2 to 98%. The Ramsey policy rule can now be
characterized with the following rule:33
 
Lt Lss
4.9% + 2 × log − log + 9 × (log Yt − log Yss ) − 139 × ((rf,t − rd,t ) − (rf,ss − rd,ss )) .
Yt Yss

A one percentage point increase in the liquidity premium leads to a 1.4% decrease in capital charges
33
The GDP growth variable has been orthogonolized with respect to the credit gap variable.

37
or, equivalently, a one standard deviation increase in the liquidity premium generates a 0.1% drop
in capital ratios. The liquidity premium serves as an indicator of how expensive for banks is equity
financing, as well as how valuable for investors are liquid assets. In line with the baseline model,
when the liquidity premium increases, banks’ lending costs go up, potentially leading to a cut in
deposits . This suggests that the capital requirements should be reduced to satisfy the investors’
demand for safe assets.

6 Conclusions

Due to government guarantees, banks have incentives to engage in excessive risk-taking and
undertake inefficient investments from a social point of view. To prevent such risk-shifting,
financial regulators have instituted capital requirements. There is an ongoing debate, however,
about whether they are effective, and if so, what level is most efficient. In this paper, I provide a
rationale for the use of a time-varying capital requirement, employing a dynamic general equilibrium
setting. A sufficiently high capital requirement can be used to rule out excessive lending, but at
the same time it may limit banks from financing high-quality investments and lead to reduced
liquidity provision. In the model, an optimal capital requirement trades off reduced risk-shifting
incentives with reduced liquidity provision. Given that banks are most incentivized to risk-shift
during periods of economic growth and holding equity is most costly during recessions, when the
economy is the least liquid, a procyclical capital regulation emerges endogenously as a solution to
the Ramsey problem.
In this paper, I also quantify policy recommendations on which macroeconomic and bank
aggregates drive the optimal time variation in capital requirements. In line with the predictions
of the baseline model, the optimal Ramsey policy would respond positively to the credit gap
and growth in GDP, where the two serve as indicators of a banks’ incentives to risk-shift. At
the same time, the Ramsey capital requirement would adjust negatively to an increase in the
liquidity premium, which indicates investor’s demand for safe and liquid assets. Two-thirds of the
welfare gain generated by implementing the optimal Ramsey policy is achieved by having state-
dependent capital ratios, while the remaining improvement is attributed to an optimal fixed capital
requirement. This leads me to conclude that implementing the CCyBs, as recommended by the
Basel III Accord, is a crucial step in promoting the stability of the financial sector.

38
References

Admati, A., P.M. DeMarzo, M.F. Hellwig, and M.F. Pfeiderer, “Fallacies, Irrelevant Facts,
and Myths in the Discussion of Capital Regulation: Why Bank Equity is Not Expensive,” 2010.

Alvarez, Fernando and Urban J. Jermann, “Using Asset Prices to Measure the Cost of
Business Cycles,” Journal of Political Economy, 2004, 112 (6), 1223–1256.

Atkinson, Anthony B and Joseph E Stiglitz, “Lectures on public economics,” 2015.

Baker, Malcolm P and Jeffrey Wurgler, “Would stricter capital requirements raise the cost
of capital? Bank capital regulation and the low risk anomaly,” Working Paper No. 19018,
National Bureau of Economic Research 2013.

Begenau, J., “Capital Requirements, Risk Choice, and Liquidity Provision in a Business Cycle
Model,” Job Market Paper, Stanford University, Department of Economics 2016.

Bernanke, B., “Non-Monetary Effects of the Financial Crisis in Propagation of the Great
Depression,” American Economic Review, 1983, 73, 257–276.

Bernanke, Ben and Mark Gertler, “Agency costs, net worth, and business fluctuations,” The
American Economic Review, 1989, pp. 14–31.

Bernanke, Ben S, Mark Gertler, and Simon Gilchrist, “The financial accelerator in a
quantitative business cycle framework,” Handbook of macroeconomics, 1999, 1, 1341–1393.

Bigio, Saki, “Endogenous liquidity and the business cycle,” Technical Report 2010.

Bloom, Nicholas, Max Floetotto, Nir Jaimovich, Itay Saporta-Eksten, and Stephen J
Terry, “Really uncertain business cycles,” Technical Report, National Bureau of Economic
Research 2012.

Borio, Claudio, “The financial cycle and macroeconomics: What have we learnt?,” Journal of
Banking & Finance, 2014, 45 (C), 182–198.

Brunnermeier, Markus K and Yuliy Sannikov, “A macroeconomic model with a financial

sector,” The American Economic Review, 2014, 104 (2), 379–421.

Calomiris, C. and J. Mason, “Consequences of Bank Distress During the Great Depression,”
American Economic Review, 2003, 93, 937–947.

39
Chari, Varadarajan V, Lawrence J Christiano, and Patrick J Kehoe, “Optimal fiscal and
monetary policy: Some recent results,” Journal of Money, Credit and Banking, 1991, 23 (3),
519–539.

Christiano, Lawrence and Daisuke Ikeda, “Leverage Restrictions in a Business Cycle Model,”
Working Paper 18688, National Bureau of Economic Research January 2014.

Corbae, Dean and Pablo D’Erasmo, “Capital requirements in a quantitative model of banking
industry dynamics,” Working Paper No. 14-13, FRB of Philadelphia 2014.

DeAngelo, Harry and Ren M. Stulz, “Why High Leverage is Optimal for Banks,” Working
Paper 19139, National Bureau of Economic Research June 2013.

Dell’Ariccia, G., D. Igan, L. Laeven, and H. Tong, “Policies for Macrofinancial Stability:
How to Deal with Credit Booms,” Discussion note, International Monetary Fund 2012.

, E. Detragiache, and R. Raghuram, “The Real Effect of Banking Crises,” Journal of

Financial Intermediation, 2008.

DellAriccia, Giovanni, Deniz Igan, and Luc Laeven, “Credit Booms and Lending Standards:
Evidence from the Subprime Mortgage Market,” Journal of Money, Credit and Banking, 03
2012, 44, 367–384.

Demyanyk, Yuliya and Otto Van Hemert, “Understanding the Subprime Mortgage Crisis,”
The Review of Finacial Studies, 2011, 24(6), 1848–1880.

den Heuvel, Skander J. Van, “The Welfare Cost of Bank Capital Requirements,” 2008, 55 (2).

, “The Welfare Effects of Bank Liquidity and Capital Requirements,” Working Paper 2016.

Dewatripont, Mathias and Jean Tirole, “Macroeconomic Shocks and Banking Regulation,”
Journal of Money, Credit and Banking, 2012, 44, 237–254.

Diamond, Douglas W and Raghuram G Rajan, “A theory of bank capital,” The Journal of
Finance, 2000, 55 (6), 2431–2465.

and , “Liquidity risk, liquidity creation and financial fragility: A theory of banking,”
Journal of Political Economy, 2001, 109 (2), 287–327.

Diamond, W. D. and P.H. Dybvig, “Bank runs, deposit insurance, and liquidity,” The Journal
of Political Economy, 1983, 91 (3), 401–419.

40
Drechsler, Itamar, Alexi Savov, and Philipp Schnabl, “The deposits channel of monetary
policy,” Technical Report, National Bureau of Economic Research 2016.

Gertler, Mark and Nobuhiro Kiyotaki, “Financial intermediation and credit policy in business
cycle analysis,” Handbook of monetary economics, 2010, 3 (11), 547–599.

and Peter Karadi, “A model of unconventional monetary policy,” Journal of monetary

Economics, 2011, 58 (1), 17–34.

Gordy, Michael B and Bradley Howells, “Procyclicality in Basel II: Can we treat the disease
without killing the patient?,” Journal of Financial Intermediation, 2006, 15 (3), 395–417.

Gorton, Gary and Andrew Metrick, “Securitized banking and the run on repo,” Journal of
Financial Economics, 2012.

Hanson, Samuel G, Anil K Kashyap, and Jeremy C Stein, “A macroprudential approach

to financial regulation,” The Journal of Economic Perspectives, 2011, pp. 3–28.

He, Zhigu and Arvind Krishnamurthy, “A model of capital and crises,” The Review of
Economic Studies, 2012, 79 (2), 735–777.

Hellmann, Thomas F, Kevin C Murdock, and Joseph E Stiglitz, “Liberalization, moral

hazard in banking, and prudential regulation: Are capital requirements enough?,” American
economic review, 2000, pp. 147–165.

Kashyap, Anil K and Jeremy C Stein, “Cyclical implications of the Basel II capital standards,”
Economic Perspectives-Federal Reserve Bank Of Chicago, 2004, 28 (1), 18–33.

, , and Samuel Hanson, “An analysis of the impact of substantially heightenedcapital

requirements on large financial institutions,” Technical Report 2010.

Kashyap, Anil, Raghuram Rajan, and Jeremy Stein, “Rethinking capital regulation,”
Maintaining stability in a changing financial system, 2008, pp. 431–471.

Kisin, Roni and Asaf Manela, “The shadow cost of bank capital requirements,” Working paper
2013.

Kiyotaki, Nobuhiro and John Moore, “Credit Cycles,” Working Paper 5083, National Bureau
of Economic Research April 1995.

41
Krishnamurthy, Arvind and Annette Vissing-Jorgensen, “The Aggregate Demand for
Treasury Debt,” Journal of Political Economy, 2012, 120 (2), 233–267.

and , “The Impact of Treasury Supply on Financial Sector Lending and Stability,”
Working Paper 2015.

Longstaff, Francis A, “The flight-to-liquidity premium in US Treasury bond prices,” Technical

Report, National bureau of economic research 2002.

Lorenzoni, Guido, “Inefficient Credit Booms,” Review of Economic Studies, 2008, 75, 809–833.

Lucas, R.E.Jr., Models of Business Cycles, Basil Blackwell, New York, 1987.

Lucas, Robert and Nancy Stokey, “Optimal fiscal and monetary policy in an economy without
capital,” Journal of Monetary Economics, 1983, 12 (1), 55–93.

Malherbe, F., “Optimal capital requirements over the business and financial cycles,” Working
Paper 2015.

Martinez-Miera, David and Javier Suarez, “A macroeconomic model of endogenous systemic

risk taking,” CEPR Discussion Paper No. DP9134 2012.

Mian, Atif and Amir Sufi, “Household Leverage and the Recession of 200709,” IMF Economic
Review, 2010, 58(1), 74–117.

Morrison, Alan D. and Lucy White, “Crises and Capital Requirements in Banking,” American
Economic Review, 2005, 95 (5), 1548–1572.

Nagel, Stefan, “The liquidity premium of near-money assets,” Technical Report, National Bureau
of Economic Research 2014.

Nataliya, Klimenko, Sebastien Pfeil, Jean-Charles Rochet, and Gianni De Nicolo,

“Aggregate Bank Capital and Credit Dynamics,” Working Paper 2016.

Nguyen, T., “Bank Capital Requirements: A Quantitative Analysis,” Job Market Paper, The
Wharton School, University of Pennsylvania 2014.

on Banking Supervision BCBS, Basel Committee, “Inteenational Convergence of Capital

Measurement and Capital Standards,” Technical Report, Basel: Bank for International
Settlements 1988.

42
Ramsey, F. P., “A Contribution to the Theory of Taxation,” The Economic Journal, 1927, 37
(145), 47–61.

Rauh, Joshua D and Amir Sufi, “Capital structure and debt structure,” Review of Financial
Studies, 2010, p. hhq095.

Repullo, Rafael, “Capital requirements, market power, and risk-taking in banking,” Journal of
financial Intermediation, 2004, 13 (2), 156–182.

and Javier Suarez, “The Procyclical Effects of Bank Capital Regulation,” Review of
Financial Studies, 2012, 26 (2), 452–490.

and Jesús Saurina, The countercyclical capital buffer of Basel III: A critical assessment,
Centre for Economic Policy Research, 2011.

, , and Carlos Trucharte, “Mitigating the Pro-cyclicality of Basel II,” Economic Policy,
2010, 25 (64), 659–702.

Saurina, J. and C. Trucharte, “An Assessment of Basel II Procyclicality in Mortgage Portfolios,”

Journal of Financial Services Research, 2007, (32), 81–101.

Stein, Jeremy C, “Monetary policy as financial-stability regulation,” Technical Report, National

Bureau of Economic Research 2011.

Tella, S.T. Di, Essays on Finance and Macroeconomics, Massachusetts Institute of Technology,
Department of Economics, 2013.

43
Appendices
A Proofs
Proposition 1
The social planner’s problem (9) can be equivalently stated as:

∞
" !#
D1−η
− Lt+1 + t+1 + ϕt (Dt+1 − Lt+1 )
X
t at
L= max E β e Lα
t ,
Lt+1 ,Dt+1
t=0
1−η

where ϕt is the Lagrange multiplier on the deposits supply constraint.

The first-order conditions are given by:
 
∂L FB

−η
: Dt+1 + ϕt < 0, (A.1)
∂Dt+1
FB
= 0 if Dt+1 > 0,

 
∂L h
B α−1
i
−1 − ϕt + βEt αeat+1 LF


: t+1 ≤ 0, (A.2)
∂Lt+1
= 0 if LF B
t+1 > 0,

 
∂L FB
: Dt+1 − LF B
t+1 ≤ 0, (A.3)
∂ϕt
= 0 if ϕt > 0.

The first-order condition (A.1) along with (A.3) delivers a corner solution for the socially optimal level of
FB
bank debt, Dt+1 = LF B FB
t+1 . This in turn implies that Nt+1 = 0. Provided that the following condition holds:

B −η B α−1
n h io
−1 + LF + βEt αeat+1 LF



lim t+1 t+1 = +∞,
LF B
t+1 →0

I have an interior solution for the socially optimal level of lending, LF B

t+1 > 0.
The second-order condition with respect to lending,

∂2L
 
B α−2
h i
βEt α (α − 1) eat+1 LF


: t+1 < 0,
∂L2t+1

ensures that LF B
t+1 is a global maximum of the social planner’s problem. Rearranging equation (A.2) and
B −η
using that ϕt = LF


t+1 completes the proof .

44
Proposition 2
Define:
F B α−1
h i
F LF B at+1 FB



t+1 , at ≡ E t αe Lt+1 − Rd,t+1 = 0. (A.4)

Using an implicit function theorem and condition (A.4), the procyclicality of LF B

t+1 follows from:

h i
∂F (LF
t+1 ,at ) B α−1
B
Et αρeat+1 LF


dLF B
t+1 ∂at t+1
=− =− h FB ,
dat ∂F (LF
t+1 ,at )
∂Rd,t+1
B
i
at+1 LF B α−2 −


∂LF B
Et α (α − 1) e t+1 ∂LF B
t+1 t+1

where I use that:

FB
∂Rd,t+1 1 F B
−η−1 ∂Et [eat+1 ]
=η L >0 and = ρEt [eat+1 ] > 0. 
∂ L̄t+1 β t+1 ∂at

Proposition 3
Under symmetric equilibrium, each bank solves:
∞
" #
 
X
t −
eωt +at Lα
 ωt +at α
L= max E β t − Rd,t Lt + Rd,t Nt − e Lt − Rd,t Lt + Rd,t Nt − Nt+1 .
Lt+1 ,Nt+1 ≤Lt+1
t=0

The first-order conditions are given by:

∗
"Z #
  ωt+1
∂L CE CE
: −1 + βRd,t+1 − βEt Rd,t+1 dΦ (ω) < 0.
∂Nt+1 0

∗
"Z #
  ωt+1
∂L h
at+1 CE α−1
i
CE

CE ω+at+1

α−1 
LCE


: Et αe Lt+1 − Rd,t+1 + Et Rd,t+1 − αe t+1 dΦ (ω) ≤ 0
∂Lt+1 0

= 0 if LCE
t+1 > 0

∗
α
∗
is defined by eωt+1 +at+1 LCE CE
LCE CE


where ωt+1 t+1 = Rd,t+1 t+1 − Dt+1 .
Without loss of generality, I restrict attention to the case when the deposit rate is strictly positive. In the
CE −η
CE
= β1 − β1 LCE


equilibrium, the deposit rate is equal to Rd,t+1 t+1 − Nt+1 , implying that bank’s objective
CE
function is strictly decreasing in the amount of equity financing, delivering a corner solution Nt+1 = 0.
Provided that the following condition holds:
∗
"Z #!
h
α−1 i ωt+1 
α−1 
at+1
lim Et αe LCE
t+1 − CE
Rd,t+1 + Et CE
Rd,t+1 − αe ω+at+1
LCE
t+1 dΦ (ω) = +∞,
LCE
t+1 →0 0

I have an interior solution for the optimal level of lending, LCE

t+1 > 0.
For ease of exposition, define the random variable υt+1 = eω+at+1 , which is, conditionally on time t
information, log-normally distributed with mean ῡt ≡ − 12 σω + (1 − ρa ) ā + ρa at and standard deviation
p
συ ≡ σω2 + σa2 . The change of variables allows me to re-write the first-order conditions with respect to

45
lending as follows: Z +∞
α−1


F OC (Lt+1 ) ≡ αυt+1 Lt+1 − Rd,t+1 dFt (υt+1 ) ,
∗
υt+1

where the bailout threshold equals to

∗ Rd,t+1 (Lt+1 − Nt+1 )

υt+1 = .
Lαt+1

The second-order condition with respect to lending implies that the objective function is neither concave,
nor convex in Lt+1 :
+∞ ∗
∂2L ∂υt+1
  Z
α (α − 1) υt+1 Lα−2 ∗
Lα−1 ∗



: t+1 dFt (υt+1 ) + Rd,t+1 − αυt+1 t+1 ft υt+1 ≶ 0,
∂L2t+1 ∗
υt+1 ∂Lt+1

where
α−1
∗ Rd,t+1 Lα


∂υt+1 t+1 + αLt+1 (Lt+1 − Nt+1 )
= 2α > 0.
∂Lt+1 (Lt+1 )
CE
Given that Nt+1 = 0, ∀Lt+1 ∈ [0, +∞), I can re-write the SOC as follows:
Z +∞
2 −α
α (α − 1) υt+1 Lα−2 2 ∗


SOC (Lt+1 ) ≡ t+1 dFt (υt+1 ) + Rd,t+1 (1 − α) Lt+1 ft υt+1 .
∗
υt+1

∗ α−1
The definition of the bailout threshold delivers that Rd,t+1 = υt+1 Lt+1 , which allows me to further iterate
the SOC:
Z +∞ !
α−2 ∗2 ∗


SOC (Lt+1 ) = (1 − α) Lt+1 −α υt+1 dFt (υt+1 ) + υt+1 (1 − α) ft υt+1 .
∗
υt+1

Note that:
∗2 ∗


∗

υt+1 ft υt+1
SOC (Lt+1 ) < 0 ⇔ α > α υt+1 ≡ R +∞
. (A.5)
∗2 f υ ∗
υt+1 dFt (υt+1 ) + υt+1
∗
υt+1 t t+1

I have that:

1.
∗

{0}
lim α υt+1 = = 0;
∗
υt+1 →0 Et [υt+1 ] + {0}

2.
∗

1
lim α υt+1 = lim R +∞ = 1,
∗
υt+1 →+∞ ∗
υt+1 →+∞ υ∗
υt+1 dFt (υt+1 )
1+ t+1
∗2 f
υt+1 t (υt+1
∗
)
since
R +∞
∗
υt+1
υt+1 dFt (υt+1 ) ∗
−υt+1 ∗
ft υt+1


1
lim ∗2 ∗

= ∗ lim
 logυ ∗ −ῡ
 = ∗ lim logυ∗ −ῡ = 0;
∗
υt+1 →+∞ υt+1 ft υt+1 υ t+1 →+∞ ∗ ∗
υt+1 ft υt+1 1 − t+1 t υ t+1 →+∞ t+1 t
−1
σ2 υ σ2 υ

46
 3.
∗


∂α υt+1
∗ > 0,
∂υt+1
since
(
) ( )
∗ ∗ +∞
∂α υt+1 logυt+1 − ῡt
 Z
∗ ∗

∗3 ∗

2
sign ∗ = sign υt+1 ft υt+1 1− 2
υt+1 dFt (υt+1 ) + υt+1 ft υt+1
∂υt+1 συ ∗
υt+1

( )
∗ +∞
logυt+1 − ῡt
Z
∗2 ∗


= sign 1− 2
υt+1 dFt (υt+1 ) + υt+1 ft υt+1 ,
συ ∗
υt+1

and provided that:

(a)

∗ +∞
logυt+1 − ῡt
 Z
∗2 ∗ ∗
< exp ῡ + συ2 ,


1− 2
υt+1 dFt (υt+1 ) + υt+1 ft υt+1 > 0, υt+1
συ ∗
υt+1

(b) !
∗  Z +∞
logυt+1 − ῡt

∂ ∗2 ∗


∗ 1− υt+1 dFt (υt+1 ) + υt+1 ft υt+1 =
∂υt+1 συ2 ∗
υt+1

+∞
1
Z
=− 2 ∗ υt+1 dFt (υt+1 ) < 0,
συ υt+1 ∗
υt+1

(c) !
∗ +∞
logυt+1 − ῡt
 Z
∗2 ∗


lim 1− υt+1 dFt (υt+1 ) + υt+1 ft υt+1 =
∗
υt+1 →+∞ συ2 ∗
υt+1

!  
+∞ ∗2 ∗2 ∗


logυt+1 − ῡt υt+1 ft υt+1
Z  
= lim υt+1 dFt (υt+1 ) · ∗2lim  1 − + R +∞  = 0.
∗2 →+∞
υt+1 ∗
υt+1 υt+1 →+∞ συ2 ∗ υt+1 dFt (υt+1 ) υt+1

∗ ∗∗


Condition (A.5), along with the properties of α υt+1 , implies that there exists υt+1 , such that:

∗ ∗∗


< 0,

 υt+1 ∈ 0, υt+1
SOC (Lt+1 ) ∗ ∗∗ ,
= 0, υt+1 = υt+1


∗ ∗∗


> 0, υt+1 ∈ υt+1 , +∞


∗∗
where υt+1 is defined by:
∗∗


α = α υt+1 .
∗
Provided that υt+1 is monotonically increasing in Lt+1 , there exists a corresponding L∗∗
t+1 , such that:


Lt+1 ∈ 0, L∗∗


< 0, t+1


SOC (Lt+1 ) = 0, Lt+1 = L∗∗
t+1 . (A.6)


L∗∗


> 0, Lt+1 ∈ t+1 , +∞


47
 Consequently, I have that for Lt+1 ∈ L∗∗


t+1 , +∞

∂
F OC (Lt+1 ) > 0,
∂Lt+1

which together with

lim F OC (Lt+1 ) = 0,
Lt+1 →+∞

delivers that:
∀Lt+1 ∈ L∗∗


F OC (Lt+1 ) < 0, t+1 , +∞ .

This implies that the objective function is strictly decreasing in the convex region, i.e., when Lt+1 ∈
L∗∗


t+1 , +∞ . Finally, provided that the following conditions hold:

lim F OC (Lt+1 ) = +∞,

Lt+1 →0

∂
Lt+1 ∈ 0, L∗∗


F OC (Lt+1 ) < 0, t+1 ,
∂Lt+1
and
F OC L∗∗


t+1 < 0,

there exists a unique LCE ∗∗

t+1 ∈ 0, Lt+1 , such that:

F OC LCE


t+1 = 0.

To sum up, the bank’s objective function is concave in lending in the region Lt+1 ∈ 0, L∗∗


t+1 and strictly
decreasing when Lt+1 ∈ L∗∗ CE


t+1 , +∞ . Hence, Lt+1 is a unique global maximum. 

Proposition 4
Define

CE α−1
h i
F LCE at+1 CE
+ ξ LCE CE




,
t+1 ta = Et αe Lt+1 − Rd,t+1 t+1 , Nt+1 ; at (A.7)
Z +∞ 

α−1 
= αυt+1 LCE
t+1
CE
− Rd,t+1 dFt (υt+1 ) = 0
∗
υt+1

Taking derivative of (A.7) with respect to lending delivers

!
CE

Z +∞
∂F LCE
t+1 , at
α−2 ∂Rd,t+1 ∗
∂υt+1 
CE α−1

1) υt+1 LCE CE ∗ ∗



= α (α − t+1 − dFt (υt+1 ) + R d,t+1 − αυt+1 Lt+1 ft υt+1
∂LCE
t+1 ∗
υt+1 ∂LCE
t+1
CE
∂Lt+1
CE

∂Rd,t+1
= SOC LCE ∗



t+1 − CE
1 − Ft υt+1 < 0,
∂Lt+1

48
since the rate of return on deposits is increasing in bank lending
CE
∂Rd,t+1 1 CE
−η−1
=η L .
∂LCE
t+1 β t+1

Before differentiating (A.7) with respect to productivity shock, it is useful to re-write F LCE


t+1 , at as follows:

  CE
α−1 
F LCE ∗ CE ∗



t+1 , at = αEt υt+1 | υ t+1 ≥ υ t+1 Lt+1 − R d,t+1 1 − Ft υt+1 ,

where  ∗ 
logυt+1 −ῡt
2
1−Φ συ − συ 2 1 − Φ (υ̃t − συ )
∗ ῡt + 12 συ 1
= eῡt + 2 συ
 
Et υt+1 | υt+1 ≥ υt+1 =e  logυ∗ −ῡ
 ,
1−Φ t+1 t 1 − Φ (υ̃t )
συ
∗
logυt+1 −ῡt
where, for ease of notation, I define υ̃t ≡ συ . Now,

1
φ (υ̃t − συ ) (1 − Φ (υ̃t )) − σ1υ (1 − Φ (υ̃t − συ ) φ (υ̃t ))
!
∗
 
∂Et υt+1 | υt+1 ≥ υt+1 1 2 1 − Φ (υ̃t − συ )
= eῡt + 2 συ + συ 2
∂ ῡt 1 − Φ (υ̃t ) (1 − Φ (υ̃t ))
1 ῡt + 1 συ2 1 − Φ (υ̃t − συ ) φ (υ̃t − συ )
 
φ (υ̃t )
= e 2 συ − + .
συ 1 − Φ (υ̃t ) 1 − Φ (υ̃t ) 1 − Φ (υ̃t − συ )

Provided that the following conditions hold:

(i)

φ (υ̃t − συ ) {0} {0}

   
φ (υ̃t )
lim συ − + = συ − + =
υ̃t →+∞ 1 − Φ (υ̃t ) 1 − Φ (υ̃t − συ ) {0} {0}
−υ̃t φ (υ̃t ) − (υ̃t − συ ) φ (υ̃t − συ )
 
lim συ − + = 0,
υ̃t →−∞ −φ (υ̃t ) −φ (υ̃t − συ )

(ii)
φ (υ̃t − συ ) {0} {0}
   
φ (υ̃t )
lim συ − + = συ − + = συ ,
υ̃t →−∞ 1 − Φ (υ̃t ) 1 − Φ (υ̃t − συ ) {1} {1}

(iii)

φ (υ̃t − συ )
 
∂ φ (υ̃t ) ∂
συ − + = (σ − λ (−υ̃t ) + λ (−υ̃t + σ)) = λ0 (−υ̃t )−λ0 (−υ̃t + σ) < 0,
∂ υ̃t 1 − Φ (υ̃t ) 1 − Φ (υ̃t − συ ) ∂ υ̃t

φ(υ̃t )
where λ (υ̃t ) ≡ Φ(υ̃t ) is an inverse of Mill’s ratio, with λ0 (·) < 0 and λ00 (·) > 0,

I have that:
( )
∗

∂Et υt+1 | υt+1 ≥ υt+1 φ (υ̃t − συ )
 
φ (υ̃t )
sign = sign συ − + = ⊕.
∂ ῡt 1 − Φ (υ̃t ) 1 − Φ (υ̃t − συ )

49
This implies that:

∂F LCE ∗

 
t+1 , at ∂Et υt+1 | υt+1 ≥ υt+1 ∂ ῡ
α−1
LCE ∗



=α t+1 1 − Ft υt+1 > 0.
∂at ∂ ῡt ∂at

Using implicit function theorem and equation (A.7), the pro-cyclicality of LCE
t+1

∂F (LCE
t+1 ,at )
dLCE
t+1 ∂at
=− > 0. 
dat t+1 ,at )
∂F (LCE
∂LCE
t+1

Proposition 5
Employing the results from the derivation of Propositions 2 and 4, I have that:

d LCE dLCE dLF B

   
t+1 t+1 F B t+1 CE
sign = sign Lt+1 − L =
dat LF B
t+1 dat dat t+1
 
h
α−1 i ∂ξ(LCEt+1 ,Nt+1 ;at )
CE h
B α−1
i
Et αρeat+1 LCE Et αρeat+1 LF



 t+1 + ∂at t+1


sign LF B
t+1 − FB LCE
t+1 .
CE
∂ξ (LCE
t+1 ,Nt+1 ;at )
∂Rd,t+1
CE
h i
 Et α (1 − α) eat+1 LCE α−2 + ∂Rd,t+1 at+1 LF B α−2 +
h i

−



t+1 ∂L CE − ∂L CE E t α (1 α) e t+1 ∂LF B


t+1 t+1 t+1

This implies that the level of the excessive investment is pro-cyclical if and only if

∂ξ LCE CE


t+1 , Nt+1 ; at
< −ξ¯a,t ,
∂at

where
CE
t+1 ,Nt+1 ;at )
∂ξ (LCE CE
h i
at+1 CE α−2

∂Rd,t+1
E t α (1 − α) e Lt+1 + CE
∂Lt+1
− ∂L CE LCE
B α−1
h i h
α−1 i
ξ¯a,t ≡ Et αρeat+1 LF · t+1 −Et αρeat+1 LCE


t+1 i ∂RF B t+1 F B t+1 .
Lt+1
h
Et α (1 − α) eat+1 LF B α−2 +

d,t+1
t+1 ∂LF B t+1

I can also show that the bailout wedge for any level of lending and equity financing is monotonically decreasing
in at . To do this, I re-write the bailout wedge as follows:
∗
"Z #
ωt+1
Rd,t+1 − αeω+at+1 Lα−1


ξ (Lt+1 , Nt+1 ; at ) = Et t+1 dΦ (ω)
0
Z +∞
α−1


= Rd,t+1 − αυt+1 Lt+1 dFt (υt+1 )
∗
υt+1
 ∗
 α−1

= Rd,t+1 αEt υt+1 | υt+1 ≤ υt+1 Lt+1 Φ (υ̃t+1 ) ,

where
1 2 Φ (υ̃t − συ )
∗
= eῡt + 2 συ
 
Et υt+1 | υt+1 ≤ υt+1 .
Φ (υ̃t )

50
 Now,

1 1
!
∗
Φ (υ̃t − συ ) − συ φ (υ̃t − συ ) Φ (υ̃t ) + συ Φ (υ̃t − συ ) φ (υ̃t )
 
∂Et υt+1 | υt+1 ≤ υt+1 1 2
= eῡt + 2 συ + 2
∂ ῡt Φ (υ̃t ) Φ (υ̃t )
1 ῡt + 1 συ2 Φ (υ̃t − συ ) φ (υ̃t − συ )
 
φ (υ̃t )
= e 2 συ + − .
συ Φ (υ̃t ) Φ (υ̃t ) Φ (υ̃t − συ )

Provided that the following conditions hold:

(i)

φ (υ̃t − συ ) {0} {0}

   
φ (υ̃t )
lim −
συ + = συ + − =
υ̃t →−∞ Φ (υ̃t ) Φ (υ̃t − συ ) {0} {0}
−υ̃t φ (υ̃t ) − (υ̃t − συ ) φ (υ̃t − συ )
 
lim συ + − = 0,
υ̃t →−∞ φ (υ̃t ) φ (υ̃t − συ )

(ii)
φ (υ̃t − συ ) {0} {0}
   
φ (υ̃t )
lim συ + − = συ + − = συ ,
υ̃t →+∞ Φ (υ̃t ) Φ (υ̃t − συ ) {1} {1}

(iii)
φ (υ̃t − συ )
 
∂ φ (υ̃t ) ∂
συ + − = (σ − λ (υ̃t ) λ (υ̃t − σ)) = λ0 (υ̃t ) − λ0 (υ̃t − σ) > 0,
∂ υ̃t Φ (υ̃t ) Φ (υ̃t − συ ) ∂ υ̃t

I have that
( )
∗

∂Et υt+1 | υt+1 ≤ υt+1 φ (υ̃t − συ )
 
φ (υ̃t )
sign = sign συ + − = ⊕.
∂ ῡt Φ (υ̃t ) Φ (υ̃t − συ )

This implies that

∗
 
∂ξ (Lt+1 , Nt+1 ; at ) ∂Et υt+1 | υt+1 ≤ υt+1 ∂ ῡt α−1 φ (υ̃t+1 ) ∂ υ̃t ∂ ῡt
= −α L Φ (υ̃t+1 )+ξ (Lt+1 , Nt+1 ; at ) < 0. 
∂at ∂ ῡt ∂at t+1 Φ (υ̃t+1 ) ∂ ῡt ∂at

Proposition 6
In a competitive equilibrium with capital regulation in place, each bank solves:
∞
" #
 
X
t ωt +at +
L= max E β e Lα
t − Rd,t Lt + Rd,t Nt − Nt+1 + λt (Nt+1 − ζt Lt+1 ) ,
Lt+1 ,Nt+1 ≤Lt+1
t=0

where λt is the Lagrange multiplier associated with the capital constraint. The corresponding first-order
conditions are given by:
∗
"Z #
  ωt+1
∂L CE CE
: −1 + λt + βRd,t+1 − βEt Rd,t+1 dΦ (ω) ≤ 0,
∂Nt+1 0
CE
= 0 if Nt+1 > 0,

51
  
∂L h
α−1 i
: −λt ζt + Et αeat+1 LCE CE
+ ξ LCE CE


t+1 − Rd,t+1 t+1 , Nt+1 ; at ≤ 0,
∂Lt+1
= 0 if LCE
t+1 > 0,

 
∂ CE
: Nt+1 − ζt LCE
t+1 ≥ 0,
∂λt
= 0 if λt > 0.

CE
Suppose first that the capital constraint is slack, i.e., Nt+1 > ζt LCE
t+1 . This means that the Lagrange
multiplier is equal to zero, which in turn implies that the bank’s objective function is strictly decreasing in
CE
Nt+1 , implying a corner solution for the level of equity financing Nt+1 = 0. This is a contradiction. 

Proposition 7
Substituting equation (17) into the bank’s first order conditions with respect to lending 15, I obtain that:
  L α−1 
ζL
Et αeat+1 Lζt+1 = Rd,t+1 ,

which implies that Lζt+1

L
= LF B
t+1 . In an equilibrium with lending capital requirement in place, the deposit
rate is equal to:
ζL 1 1
L −η
Rd,t+1 = − 1 − ζtL Lζt+1 ,
β β
and the bailout rate is defined by:
 L α
∗ ζL
L
eωt+1 +at+1 Lζt+1 = Rd,t+1 1 − ζtL Lζt+1 .

Given that the following conditions hold:

(i)

ζL
∂Rd,t+1
 
∂ FB

FB

1 ζL
θ Lt+1 , ζt − Rd,t+1 = (1 − ζt ) + − Rd,t+1
∂ζt ∂ζt β
1 B −η 1 B −η
(1 − ζt ) LF (1 − ζt ) LF



= −η t+1 + t+1
β β
1 −η
(1 − ζt ) LF B


= (1 − η) t+1 > 0,
β

(ii)
   
∗ ζL
∂ ξ˜ LF B

Z ωt+1  ∗
, ζ ; a ∂R ∂ωt+1

L
t+1 t t ζ d,t+1  ζL ∗


= Et   −Rd,t+1 + (1 − ζt ) dΦ (ω) +Et
 (1 − ζt ) (1 − α) Rd,t+1 φ ωt+1 < 0,
∂ζt 0 ∂ζt ∂ζt

52
 with
∗ ζL
∂ωt+1 1 1 ∂Rd,t+1
=− + ζL < 0,
∂ζt 1 − ζt Rd,t+1 ∂ζt

(iii)
θ LF B FB
ξ˜ LF B



t+1 , 0 = Rd,t+1 & t+1 , 0; at > 0,

there is a unique solution to equation (17). 

Proposition 8
Substituting equation (18) into the bank’s first order condition with respect to lending (15), I obtain that:
  D α−1 
ζD
Et αeat+1 Lζt+1 = Rd,t+1 ,

LF B D
which implies that Lζt+1
D
= 1−ζ ζ FB
D and Dt+1 = Dt+1 . In an equilibrium with liquidity capital requirement in
t+1
t
place, the deposit rate is equal to:

ζ D 1 1
D −η
Rd,t+1 = − 1 − ζtD Lζt+1 FB
= Rd,t+1 ,
β β

and the bailout rate is defined by:

 D α
∗ ζD
D
eωt+1 +at+1 Lζt+1 = Rd,t+1 1 − ζtD Lζt+1 ,

or, equivalently:
∗
α ζ D
α
eωt+1 +at+1 LF B
t+1 = Rd,t+1 1 − ζtD LF B
t+1 .

Given that the following conditions hold:

(i)
 FB  FB
Lt+1 Lt+1
     
∂ α−1 FB α−2 α−1 1 ζD
(1 − ζt ) θ ; ζt − Rd,t+1 = (1 − α) (1 − ζt ) θ ; ζt +(1 − ζt ) − Rd,t+1 > 0,
∂ζt 1 − ζt 1 − ζt β

(ii)
"Z ∗ #
 FB ωt+1
Lt+1
 
∂ α−1 ˜ α−1 ζD
(1 − ζt ) ξ ; ζt = −α (1 − ζt ) Et Rd,t+1 dΦ (ω)
∂ζt 1 − ζt 0
 ∗
∂ωt+1

α ζD ∗


+ Et (1 − ζt ) (1 − α) Rd,t+1 φ ωt+1 ,
∂ζt

with ∗
∂ωt+1 α
=− < 0,
∂ζt 1 − ζt

(iii)
θ LF B FB
ξ˜ LF B



t+1 , 0 = Rd,t+1 & t+1 , 0; at > 0,

53
there is a unique solution to equation (18). 

B Tables and Figures

Table B1: Configuration of Model Parameters

Description Symbol Value Source/Target

Subjective Discount Factor β 0.975 Standard
Risk Aversion Coefficient γ 1.000 Standard
Elasticity of Deposits and Consumption η 1.200 St.dev. of debt-consumption ratio
Deposits Weight χ 0.010 Average liquidity premium

Firm Capital Share αf 0.355 Capital-output ratio

Firm Operating Cost of 0.055 St.dev. of investment-capital ratio

Bank Capital Share αb 0.780 Capital-output ratio

Bank Operating Cost ob 0.065 Profit-to-loan ratio
Bank Output Weight āb -1.35 Capital ratio in two sectors
Capital Adequacy Ratio ζ̄ 0.073 Average leverage ratio

Depreciation Rate δ 0.075 Investment-capital ratio

Persistence of Productivity Schock ρa 0.95

Process for Solow residuals
Std of Productivity Schock σa 0.020
Std of Idiosyncratic Shock σω 0.335 Bailout rate
Dispersion of Idiosyncratic Volatility ν 0.500 Idiosyncratic volatility dispersion

This table reports the benchmark configuration of the model parameters calibrated at annual frequency.

54
 Table B2: First Aggregate Moments

Model
Data Mean 2.5% 97.5%

Aggregate Sector
Capital-Output, K/Y 3.03 2.99 2.86 3.13
Investment-Capital, I/K 0.07 0.08 0.07 0.08

Market Fraction
Capital Weight, Kb /K 0.46 0.45 0.40 0.51
Output Weight, Yb /Y 0.28 0.28 0.23 0.33

Banking Sector
Capital-Output, Kb /Yb 4.96 4.87 4.79 4.94
Investment-Capital, Ib /Kb 0.05 0.08 0.07 0.09
Capital Adequacy Ratio, N/L, % 7.26 7.26 7.26 7.26
Profit-Lending, π/L 0.04 0.05 0.05 0.05
Liquidity Premium, Rf − Rd , % 0.57 0.56 0.46 0.65
Bailout Rate, % 0.76 0.79 0.56 1.06

Bank-Independent Sector
Capital-Output, Kf /Yf 2.29 2.28 2.23 2.33
Investment-Capital, If /Kf 0.09 0.08 0.07 0.08

This table reports the first moments computed as averages of the data and model series. The summary statistics in

the data are computed in the annual sample for the 1980-2008 period. The mean, as well as 2.5% and 97.5% values,

capture the model moment distributions across the samples whose size equals the data.

55
 Table B3: Second Aggregate Moments

Model
Data Mean 2.5% 97.5%

Aggregate Sector
Consumption, σ (∆c) 1.28 0.83 0.61 1.14
Output, σ (∆y) 2.00 2.02 1.66 2.41
Investment, σ (∆i) 4.36 7.16 5.96 8.47

Banking Sector
Output, σ (∆yb ) 2.54 2.22 1.71 2.84
Investment, σ (∆ib ) 9.28 12.49 10.40 14.89
Lending, σ (∆l) 2.60 1.53 0.97 2.32
Debt-Consumpion Ratio, σ (∆d − ∆c) 3.67 0.79 0.47 1.25
Profits, σ (∆π) 13.59 10.58 8.44 13.09

Bank-Independent Sector
Output, σ (∆yf ) 2.07 2.00 1.67 2.38
Investment, σ (∆if ) 3.84 3.09 2.54 3.70

Liquidity Premium, σ (Rf − Rd ) 0.35 0.03 0.01 0.07

This table reports the second moments computed as standard deviations of the data and model series. σ (∆x) is
the standard deviation of annual log growth rates of X. The summary statistics in the data are computed in the
annual sample for the 1980-2008 period. The mean, as well as 2.5% and 97.5% values, capture the model moment
distributions across the samples whose size equals the data. The statistics are expressed in percentages.

56
 Table B4: Business Cycle Correlations

Model
Data Mean 2.5% 97.5%

Aggregate Sector
Consumption, ρ (∆c, ∆y) 0.77 0.89 0.86 0.93
Investment, ρ (∆i, ∆y) 0.84 0.97 0.93 0.99

Banking Sector
Output, ρ (∆yb , ∆y) 0.82 0.95 0.93 0.97
Investment, ρ (∆ib , ∆y) 0.70 0.95 0.91 0.97
Lending, ρ (∆l, ∆y) 0.47 0.69 0.64 0.74
Deposits, ρ (∆d − ∆c, ∆y) 0.54 0.37 0.22 0.55
Profits, ρ (∆π, ∆y) 0.15 0.79 0.74 0.84
Liquidity Premium, ρ (Rf − Rd , ∆y) -0.21 0.04 -0.32 0.38

Bank-Independent Sector
Output, ρ (∆yf , ∆y) 0.96 0.99 0.98 1.00
Investment, ρ (∆if , ∆y) 0.59 0.94 0.92 0.95

This table reports the correlations of aggregate variables with the output growth in the data and in the model.

ρ (∆x, ∆y) is the correlation of annual growth rates of X with annual output growth rates; ρ (Rf − Rd , ∆y) is the

correlation of annual liquidity premium (in levels) with annual output growth rates; ρ (p, ∆y) is the correlation of

fraction of bailed out banks with the annual output growth rates. The summary statistics in the data are computed

in the annual sample for the 1980-2008 period. The mean, as well as 2.5% and 97.5% values, capture the model

moment distributions across the samples whose size equals the data.

57
 Table B5: Optimal Bank Capital Policy (Benchmark Quantitative Model)

Panel A: Level-Log Regressions

(1) (2) (3) (4) (5)

Constant 4.93 4.93 4.93 4.93 4.93

Credit gap 2.64 2.64 1.91
( 1.60, 3.48) ( 1.60, 3.48) ( 0.75, 2.93)
Bank credit 2.51
( 2.45, 2.57)
GDP 5.45 8.17 7.91
( 5.24, 5.72) ( 8.16, 8.18) ( 7.84, 7.99)
Liquidity premium -172.77
( -224.16, -117.29)

R2 0.1366 0.5033 0.8175 0.9998 0.9999

( 0.0431, 0.2465) ( 0.4062, 0.5927) ( 0.7877, 0.8469) ( 0.9997, 0.9998) ( 0.9998, 0.9999)

Panel B: Normalized Level-Log Regressions

(1) (2) (3) (4) (5)

Constant 4.93 4.93 4.93 4.93 4.93

Credit gap 0.09 0.09 0.06
( 0.05, 0.11) ( 0.05, 0.11) ( 0.02, 0.09)
Bank credit 0.08
( 0.07, 0.09)
GDP 0.16 0.69 0.67
( 0.15, 0.18) ( 0.63, 0.76) ( 0.61, 0.74)
Liquidity premium -0.01
( -0.02, -0.01)

R2 0.1366 0.5033 0.8175 0.9998 0.9999

( 0.0431, 0.2465) ( 0.4062, 0.5927) ( 0.7877, 0.8469) ( 0.9997, 0.9998) ( 0.9998, 0.9999)

PN
Panel A of this table presents results from level-log regressions of the form ζt∗ = ζss + j=1 ζxj xjt + εt , while Panel B
presents results from normalized level-log regressions, where independent variables are scaled by their standard
deviations. The dependent variable is the Ramsey policy rule in year t expressed in percentages (1 to 100). The
Lt Lss
independent variables include (i) credit gap log Yt
−log Yss
; (ii) bank credit log Lt −log Lss ; (iii) GDP log Yt −log Yss ;
(iv) liquidity premium (rf,t − rd,t ) − (rf,ss − rd,ss ). Reported coefficients are the mean values across the samples,
while the numbers reported in parentheses are the 2.5% and 97.5% values. Reported R2 s are the mean values across
the samples, while the numbers reported in parentheses are the 2.5% and 97.5% values.

58
Table B6: Optimal Bank Capital Policy (Quantitative Model with Liquidity Shocks)

Panel A: Level-Log Regressions

(1) (2) (3) (4) (5)

Constant 4.93 4.93 4.93 4.93 4.92

Credit gap 1.71 1.71 1.67
( 0.41, 2.96) ( 0.41, 2.96) ( 0.61, 2.58)
Bank credit 2.20
( 1.74, 2.64)
GDP 5.26 8.57 7.87
( 4.62, 5.93) ( 8.18, 8.97) ( 7.77, 7.97)
Liquidity premium -139.24
( -145.78, -132.49)

R2 0.0584 0.3435 0.6577 0.9107 0.9766

( 0.0024, 0.1551) ( 0.2161, 0.4717) ( 0.5562, 0.7417) ( 0.8820, 0.9330) ( 0.9702, 0.9820)

Panel B: Normalized Level-Log Regressions

(1) (2) (3) (4) (5)

Constant 4.93 4.93 4.93 4.93 4.92

Credit gap 0.06 0.06 0.06
( 0.01, 0.10) ( 0.01, 0.10) ( 0.02, 0.09)
Bank credit 0.07
( 0.06, 0.09)
GDP 0.16 0.74 0.68
( 0.13, 0.18) ( 0.66, 0.83) ( 0.62, 0.74)
Liquidity premium -0.09
( -0.09, -0.08)

R2 0.0584 0.3435 0.6577 0.9107 0.9766

( 0.0024, 0.1551) ( 0.2161, 0.4717) ( 0.5562, 0.7417) ( 0.8820, 0.9330) ( 0.9702, 0.9820)

PN
Panel A of this table presents results from level-log regressions of the form ζt∗ = ζss + j=1 ζxj xjt + εt , while Panel B
presents results from normalized level-log regressions, where independent variables are scaled by their standard
deviations. The dependent variable is the Ramsey policy rule in year t expressed in percentages (1 to 100). The
Lt Lss
independent variables include (i) credit gap log Yt
−log Yss
; (ii) bank credit log Lt −log Lss ; (iii) GDP log Yt −log Yss ;
(iv) liquidity premium (rf,t − rd,t ) − (rf,ss − rd,ss ). Reported coefficients are the mean values across the samples,
while the numbers reported in parentheses are the 2.5% and 97.5% values. Reported R2 s are the mean values across
the samples, while the numbers reported in parentheses are the 2.5% and 97.5% values.

59
 Figure B1: Exogenous TFP Shock.

Benchmark Quantitative Model with Fixed Capital Requirement

(a) TFP (b) Output (c) Consumption

−3 −3 −3
x 10 x 10 x 10
20
15 15
15
10
10 10
5 5
5
0
0
−5 0
−5
10 20 30 40 10 20 30 40 10 20 30 40

(d) Lending (e) Deposit-Consumption Ratio (f) Liquidity Premium

−3
0.04 x 10
0.02 0
0.03
0.015
0.02 −5
0.01
0.01 −10
0.005
0 −15
0
−0.01
10 20 30 40 10 20 30 40 10 20 30 40

(g) Probability of Bailout (h) Bailout Wedge

0.06
0
0.04
−0.05
0.02
−0.1
0
−0.15
−0.02
−0.2
−0.04
10 20 30 40 10 20 30 40
The figures show the effect of a one standard deviation shock to log productivity on the variables of the benchmark

quantitative model, as captured by the impulse response functions (blue solid lines). Banks are subject to a fixed

capital requirement of 7.26%. All model quantities are in logs.

60
 Figure B2: Exogenous TFP Shock.

Benchmark Quantitative Model with Ramsey Capital Requirement

(a) TFP (b) Output (c) Consumption

−3 −3 −3
x 10 x 10 x 10
20
15 15
15
10
10 10
5 5
5
0
0
−5 ζ = 7.26% ζ = 7.26%
Ramsey policy 0 Ramsey policy
−5
10 20 30 40 10 20 30 40 10 20 30 40

(d) Lending (e) Deposit-Consumption Ratio (f) Liquidity Premium

−3 −3
0.04 x 10 x 10
5 ζ = 7.26%
0.03 20
Ramsey policy
15 0
0.02
10 IRF of −5
lending-to-consumption
0.01 5 ratio in Ramsey
equilibrium −10
0 0
ζ = 7.26% ζ = 7.26% −15
Ramsey policy −5 Ramsey policy
−0.01
10 20 30 40 10 20 30 40 10 20 30 40

(g) Probability of Bailout (h) Bailout Wedge (i) Ramsey Policy

−3
0.05 x 10
0
−0.05 0
2
−0.1 −0.05

−0.15 −0.1 1

−0.2 −0.15
ζ = 7.26% ζ = 7.26% 0
−0.25 Ramsey policy −0.2 Ramsey policy
10 20 30 40 10 20 30 40 10 20 30 40
The figures show the effect of a one standard deviation shock to log productivity on the variables of the benchmark
quantitative model, as captured by the impulse response functions. The figures provide a comparison of the effects
when banks are subject to a fixed capital requirement of 7.26% (blue solid lines) and when banks face the optimal
Ramsey policy (red dash-dotted lines) . All model quantities are in logs.

61
 Figure B3: Exogenous Liquidity Shock.

Quantitative Model with Liquidity Shocks and Fixed Capital Requirement

(a) χ (b) Output (c) Consumption

−4 −3 −3
x 10 x 10 x 10
10 4
2
8 3
6 1
2
4 0
2 1
−1
0 0
−2
10 20 30 40 10 20 30 40 10 20 30 40

(d) Lending (e) Deposit-Consumption Ratio (f) Liquidity Premium

−3 −3
x 10 x 10
0.1
10 8 0.08
6 0.06
5 4 0.04
2 0.02

0 0
0
10 20 30 40 10 20 30 40 10 20 30 40

(g) Probability of Bailout (h) Bailout Wedge

−3 −3
x 10 x 10
5
0

0
−5

−5
−10

−10
−15

10 20 30 40 10 20 30 40
The figures show the effect of a one standard deviation shock to log productivity on the variables of the quantitative

model with liquidity shocks, as captured by the impulse response functions (blue solid lines). Banks are subject to a

fixed capital requirement of 7.26%. All model quantities are in logs.

62
 Figure B4: Exogenous Liquidity Shock.

Quantitative Model with Liquidity Shocks and Ramsey Capital Requirement

(a) TFP (b) Output (c) Consumption

−4 −3 −3
x 10 x 10 x 10
10 4
2
8 3
6 1
2
4 0
2 1
−1
0
ζ = 7.26% ζ = 7.26%
0 Ramsey policy Ramsey policy
−2
10 20 30 40 10 20 30 40 10 20 30 40

(d) Lending (e) Deposit-Consumption Ratio (f) Liquidity Premium

−3 −3
x 10 x 10 ζ = 7.26%
15 ζ = 7.26% 0.1 Ramsey policy
Ramsey policy 0.08
10
10 0.06
0.04
5 IRF of
5 lending-to-consumption 0.02
ratio in Ramsey
ζ = 7.26% equilibrium
Ramsey policy 0
0 0
10 20 30 40 10 20 30 40 10 20 30 40

(g) Probability of Bailout (h) Bailout Wedge (i) Ramsey Policy

−4
0.06 ζ = 7.26% 0.08 ζ = 7.26% x 10
Ramsey policy Ramsey policy 0
0.06
0.04
−2
0.04
−4
0.02
0.02 −6

0 0 −8
−10
10 20 30 40 10 20 30 40 10 20 30 40
The figures show the effect of a one standard deviation shock to log productivity on the variables of the quantitative
model with liquidity shocks, as captured by the impulse response functions. The figures provide a comparison of the
effects when banks are subject to a fixed capital requirement of 7.26% (blue solid lines) and when banks face the
optimal Ramsey policy (red dash-dotted lines) . All model quantities are in logs.

63
C Ramsey Equilibrium
A Ramsey planner maximizes the lifetime utility of households subject to the resource constraint and
implementability conditions:
∞
" #
X
t
max E β u (Ct , Dt+1 )
Ct ,Lt+1 ,Nt+1 ,Dt+1 ,Kf,t+1
t=0
1−γ
v (Ct , Dt+1 )
with u (Ct , Dt+1 ) = −1
1−γ
α
s.t. Ct = Rd,t Dt − Dt+1 + Rk,t Kf,tf − If,t − of Kf,t + Zt + Πt − Tt , (C.1)
Et [Mt,t+1 (Rk,t+1 + (1 − δ − of ))] = 1, (C.2)
Nt+1 + Dt+1 = Lt+1 , (C.3)
Et [Mt,t+1 (Rl,t+1 + (1 − δ − ob ))] = θt − ξ˜t , (C.4)
Ct + Ib,t + If,t = Yb,t + Yf,t − ob Lt − of Kf,t , (C.5)

where the taxes and net proceeds from banks’ and firms’ activity are, respectively, equal to:
Z 1
max Rd,t Dt − eωj,t +at Ltαb − (1 − δ − ob ) Lt , 0 dj,

Tt = (C.6)
0
Z 1
zj,t = max eωj,t +at Lα

Zt = zj,t dj, with t + (1 − δ − ob ) Lt − Rd,t Dt , 0 − Nt+1 , (C.7)
b

Πt = Yf,t − Rk,t Kf,t . (C.8)

The rate of return on deposits, the stochastic discount factor, the liquidity and risk-shifting cost of lending,
the rate of return on loans and capital are, correspondingly, defined by (19), (20), (21), (22), (23) and (24).
The optimal Ramsey policy requires a capital ratio that equals:

Nt+1
ζt = . (C.9)
Lt+1

The Ramsey problem can be further simplified. By substituting expressions (C.6), (C.7) and (C.8) in the
household’s budget constraint (C.1), I obtain the resource constraint (C.5), therefore implying that I can
omit it.
n o∞
Proposition C1 The allocation rules Ct , Dt , {zj,t , lj,t+1 , nj,t+1 , dj,t+1 }j∈Ω , Kf,t+1 and policies
t=0
∞
{ζt , Tt }t=0
in a competitive equilibrium satisfy conditions (C.1)-(C.9). Furthermore, given the allocation
∞
rules and policies that satisfy (C.1)-(C.9), we can construct the price rules {Mt,t+1 , Rd,t+1 , Rl,t+1 , Rk,t+1 }t=0
that together with the given allocations and policies constitute a competitive equilibrium.

Proof. The first part of the proof is quite straightforward and follows from the fact that the competitive
equilibrium is symmetric. For the second part of the proof, I use the Ramsey allocations to construct the

64
prices as follows:
  η̃1 −γ 
− η̃1
v (Ct+1 , Dt+2 ) Ct+1
Mt,t+1 = ,
v (Ct , Dt+1 ) Ct
 − η1 !
1 Dt+1
Rd,t+1 = 1−χ ,
Et [Mt,t+1 ] Ct
αb −1
Rl,t+1 = αb eāb +at+1 Lt+1 ,
α −1
Rk,t+1 = αf eat+1 Kf,t+1
f
.

The expressions for the SDF, deposits rate and rate of return on capital along with the budget constraint
(C.1) and condition (C.2) ensure that the households choose their optimal allocations for consumption and
deposits. Similarly, the expression for the rate of return on loans together with the capital constraint (C.9),
balance sheet constraint (C.3) and condition (C.4) ensures that the banks choose their optimal lending
level and capital structure. The net distributions to the banks’ shareholders (C.7) are defined by the same
expressions in the competitive and Ramsey equilibria. The government budget constraint (C.6) and resource
constraint (C.5) are identical. 

D Welfare Costs of Capital Requirement

The compensating variation $, which is the permanent change in consumption that households are willing
to accept to move from the economy with a fixed level of capital requirement ζ 0 to the world with a capital
ratio of ζ 1 , is defined by:
  1−γ 
∞ (1 + $) C̃ − 1
"∞ #
X t X C̃ 1−γ − 1
Eζ 0  βt  = Eζ 1 βt t . (D.1)
 
t=0
1 − γ t=0
1−γ

or, equivalently:
hP 1
i  1−γ
+∞

Eζ 1 t=0 β t C̃t1−γ
$= hP i − 1.
+∞
Eζ 0 t=0 β t C̃t1−γ

To obtain an estimate of expected lifetime utility, I first solve for agents’ decision rules in a competitive
equilibrium, varying the capital regulation in place. Next, I perform 1, 000 simulations of a Ramsey
equilibrium. Within each simulation, at a random point on the time path, a new regime of capital
charges is introduced. Then, using agents’ decision rules, I generate 10, 000 observations for the households’
consumption and deposits from the period when a new regime has been introduced on. Finally, I evaluate
the realized utility under the new regime and average it over the simulations. This approach allows me to
take into account the transition dynamics between different policy regimes.
It is crucial to take into the account the transition dynamics to get the correct welfare ranking. In a
competitive equilibrium with a lower level of capital requirement, the financial intermediaries accumulate a
higher stock of capital. In the long-run, this delivers a higher level of consumption, but it is costly for the
households in the short run, since they have to reduce consumption in order to boost capital investment. So,

65
if I were to compare the realized utility based on the simulations where I burn initial observations, I would
find that economies with strict capital requirements always generate lower welfare results than economies
with low capital ratios. To understand why the compensation variation is affected by the transition dynamics
and changes in the steady state levels, define consumption as the product of its cycle component, denoted
by C̃t∗ , and level component, denoted by gt :34

C̃t = gt C̃t∗ .

By taking a second-order Taylor expansion of the momentary utility function, I obtain:

 1−γ  1
E C̃t∗ 2
≈ 1 − γ (1 − γ) σC̃ ∗.
2 t

Substituting the above expression in (D.1) delivers:

∞ ∞
  X !   X !
1  0 1−γ 1  1 1−γ
1−γ ζ ζ
(1 + $) 1 − γ (1 − γ) σ 2 ∗ζ0 β t gt = 1 − γ (1 − γ) σ 2 ∗ζ1 β t gt .
2 C̃t
t=0
2 C̃t
t=0

By taking logs and using that log (1 + x) ≈ x, I get:

∞ ∞
  ! !!
1 1  1
1−γ  0
1−γ
gtζ gtζ
X X
$ ≈ − γ σ 2 ∗ζ1 − σ 2 ∗ζ0 + log β t
− log β t
.
2 C̃t C̃t 1−γ t=0 t=0

Recall that the Lucas cost of eliminating business cycle risk in this framework is equal to − 21 γσC̃
2
∗.
t

34
The level component gt captures the transition dynamics (temporary trend) between a steady state of the economy
with a fixed level of capital requirement ζ 0 to a steady state of the world with a capital ratio of ζ 1 .

66