Money, Credit, and Banking

Aleksander Berentsen Gabriele Camera Christopher J. Waller

University of Basel Purdue University University of Notre Dame

May 2004

Abstract

We use a modified version of the Lagos Wright model to introduce an essential role for banks.
Due to preference shocks, agents have excess demand for or supply of money balances. Banks
arise to reallocate excess cash by taking deposits from sellers and making loans to buyers. We
consider two variations of the model: one in which buyers borrow to finance consumption and
another in which they borrow to finance investment. We show that for any positive nominal
interest rate, the existence of banks leads to a higher level of steady state output and welfare.
We also derive conditions under which borrowers voluntarily repay loans. Finally, we examine
how shocks affect the economy and the optimal stabilization response by the central bank to
these shocks.

1
1 Introduction

The objective of this paper is to introduce banking and credit into an essential model of money.
In our framework, money is essential as a medium of exchange but due to idiosyncratic preference
shocks some agents have excess money holdings and others have too little. Banks provide inter-
mediation services by accepting deposits and making loans. All deposits and loans are nominal
as is the interest rate paid to depositors and charged to borrowers. Banks have a record keeping
technology over its financial histories with customers but not on trading histories amongst the
agents themselves. Consequently, the existence of banks does not eliminate the need for money as
a medium of exchange. We demonstrate that for any positive nominal interest rate, banks expand
the set of allocations and thus are essential. Under the Friedman rule, banks are not essential —
the allocation without banks is the same as with banks. This raises an interesting interaction in
the choice of monetary policy and the efficiency gains arising from a banking system.
We also explore how banks amplify or dampen shocks to the real economy and the role of
stabilization policy for the central bank. We demonstrate that banks transmit excess liquidity
shocks into the consumption making it more variable than would be the case without banks. We
also examine the Ramsey problem confronting the central bank with regards to how it adjusts the
money stock within a period in response to these shocks. Our main finding is that it is optimal for
the central bank to provide an elastic supply of currency to the market - it should inject reserves
into the banking system when demand is high and withdraw them when demand is low.
The paper proceeds as follows....

1.1 Literature

Modify: Our model differs substantial from other search models with banking in a number of ways.
In our model, money and goods are perfectly divisible. There is no security motive for depositing
cash in banks as in He, Huang and Wright (2003). Furthermore, unlike Cavalacanti and Wallace
(1999), bank claims on deposits will not circulate as media of exchange (at least in the markets
where money is essential) hence there is no inside money. In a sense, our model mimics the results
of Kocherlakota’s (2003) ‘illiquid’ bonds model but there is nothing preventing bank claims to

2
deposits from circulating as media of exchange — agents holding those claims do not want to buy
goods so they simply hold their deposit claims.

2 The Environment

Time is discrete and in each period there are two markets that open sequentially and are perfectly
competitive.1 There is a [0, 1] continuum of infinitely-lived agents and one perishable good produced
and consumed by all agents.
At the beginning of the first market agents get a preference shock such that they either can
consume or produce. With probability 1 − n an agent can consume but cannot produce while with
probability n the agent can produce but cannot consume. We refer to consumers as buyers and
producers as sellers. Agents get utility u(q) from q > 0 consumption in the first market, where
u(0) = 0, u0 (q) > 0, u0 (0) = ∞, u0 (+∞) = 0, u00 (q) < 0. Furthermore, we impose that the elasticity
qu0 (q)
of utility e (q) = u(q) is bounded. Producers incur utility cost c (q) from producing q units of
output with c0 (q) = 1/a > 0 and c (0) = 0 where a is a measure of productivity. To motivate a
role for fiat money, we assume that all goods trades are anonymous. In particular, trading histories
of agents are private information. Consequently, sellers require immediate compensation so buyers
pay with money.
In the second market all agents consume and produce, getting utility U (x) from x consumption,
with U 0 (x) > 0, U 0 (0) = ∞, U 0 (+∞) = 0 and U 00 (x) ≤ 0.2 Agents can produce one unit of x with
one unit of labor h. Production of x units of output generates disutility h. The discount factor
across dates is β ∈ (0, 1).
We assume a central bank exists that controls the supply of fiat currency. The growth rate of
the money stock is given by Mt = γMt−1 where γ > 0 and Mt denotes the per capita money stock
in t. Agents receive lump sum transfers τ Mt−1 = (γ − 1)Mt−1 over the period. Some of the transfer
is received at the beginning of market 1 and some during market 2. Let τ 1 and τ 2 denote the
1
The basic environment is that of Berentsen, Camera and Waller (2004) which is a combination of Lagos and
Wright (2003) and Aruoba, Waller and Wright (2003).
2
Following Lagos-Wright (2003), the difference in preferences over the good sold in the last market allows us to
impose technical conditions such that the distribution of money holdings is degenerate at the beginning of a period.

3
transfers in market 1 and 2 respectively with (τ 1 + τ 2 ) Mt−1 = τ Mt−1 . We show how the timing
of these transfers affects the equilibrium and how they can be manipulated to affect real variables
when stochastic shocks hit the economy. If τ < 0, we assume the central bank has the authority to
levy taxes in the form of currency to extract cash from the economy.3 For notational ease variables
corresponding to the next period are indexed by +1, and variables corresponding to the previous
period are indexed by −1.
The key contribution of this paper is the introduction of a banking sector. Its purpose is to
reallocate cash from those with excess money balances to those with a need for cash. In the first
market the banking sector opens before trading and agents can borrow and deposit after observing
their production and consumption shocks. Then, buyers and sellers trade. In the second market
the goods market and the banking sector opens simultaneously. We assume net settlements, i.e. all
financial claims are settled at the end of the period. This essentially means that loans and deposits
cannot be rolled over. Consequently, all financial contracts are one-period contracts.
In this paper we think of banking as a record keeping technology for financial transactions but
not good market transactions. This ensures that money is still essential for goods trade. We assume
that banks do not issue tangible objects as deposit claims. Consequently, there are no bank notes
in circulation. So the only role for banks in our paper is to reallocate fiat money within a period.4
Since goods are perishable across markets, loans are unsecured. We assume that a borrower who
fails to repay his loan will be shut out of the banking sector in all future periods. That is, although
banks compete they share information about agents’ repayment histories. Given this punishment,
we need to derive conditions to ensure voluntary repayment.
At the beginning of the first market sellers will hold idle balances since they are not consuming.
In contrast, buyers may have insufficient cash and so are willing to borrow to finance consumption
3
Here, we assume that the central bank has the authority to control entry to the market. The lump-sum tax then
is an entry fee. Note that it does not mean that the central bank can force agents to produce or consume certain
quantities once they have entered the market. Nor does it mean the central bank knows the identity of the agents.
4
This is fundamentally different from Cavalcanti-Wallace (1999) who assume that some agents’ goods market
trades are public information, which allows these agents to issue tangible objects that serve as inside money.
In fact, in our model banks are nothing more than cash machines that post interest rates for injections and
withdrawals. In equilibrium these posted interest rates clear out the excess cash.

4
in the first market. Banks accept nominal deposits and pay the nominal interest rate id and make
nominal loans at nominal rate i. The banking sector is perfectly competitive, so banks take these
rates as given. There are no operating costs but they may face reserve requirements on deposits.
It is straightforward to show that in a symmetric equilibrium all borrowers take out the same size
loan, l, and depositors deposit the same amount d.
The zero profit condition implies that

i (1 − n) l = id nd (1)

where the total loan size is

(1 − n) l = (1 − µ) nd (2)

and µ is the reserve requirement ratio. Consequently (1) and (2) imply

id = (1 − µ) i (3)

To ensure that i > id we impose throughout the paper that µ > 0 where i → id when µ → 0. If
µ = 1, there is no financial intermediation so that the allocation is equivalent to the allocation in
an environment without banks.
The precise sequence of action after agents learn their type is as follows (see Figure 1). First,
the monetary injection τ 1 Mt−1 occurs. Second, sellers deposit their excess cash and buyers borrow
money from the banking sector. Finally, agents move on to the goods market and trade.

Repay Loans ¦ Redeem Deposits

Deposits ¦ Loans
  
 

   
Banking Goods Goods and Banking
t-1 t+1

 
 


Market 1
 
Market 2

t
Figure 1: Trading sequence

5
 In the second market the injection τ 2 Mt−1 occurs and the goods market and banking sector open
where all financial claims are settled. Define v as the ratio of aggregate nominal loan repayments,
(1 − n) (1 + i) l, to the money stock, M . If v ≤ 1, then borrowers can repay their loans with one
trip to the bank only since the nominal demand for cash by borrowers for repayment of loans is
less than M . If v > 1, borrowers cannot acquire sufficient balances in the aggregate to repay loans
at once. This implies that they repay part of their loans which is then used to settle deposit claims
and the cash reenters the goods market as depositors use the cash to acquire more goods. This
recycling of cash occurs until all claims are settled.

3 Symmetric equilibrium

In period t, let φ be the real price of money in the second market. We study equilibria where
end-of-period real money balances are time-invariant

φM = φ−1 M−1 = Ω. (4)

We refer to it as a stationary equilibrium.

Consider a stationary equilibrium. Let V (m1 ) denote the expected value from trading in market
1 with m1 money balances conditional on the aggregate shock. Let W (m2 , l, d) denote the expected
value from entering the second market with m2 units of money, l loans, and d deposits. In what
follows, we look at a representative period t and work backwards from the second to the first market.

3.1 The second market

In the second market the sequence of actions is as follows. First, agents produce and trade good
x and adjust their money balances taking into account cash payments or receipts from the bank.
After that loans are repaid by borrowers and the bank redeems deposits. If an agent has borrowed
l units of money, then he pays (1 + i) l units of money. If he has deposited d units of money, he

6
receives (1 + id ) d. The representative agent’s program is

W (m2 , l, d) = max [U (x) − h + βV (m1,+1 )] (5)

x,h,m1,+1

s.t. x + φm1,+1 = h + φ (m2 + τ 2 M−1 ) + φ (1 + id ) d − φ (1 + i) l

where m1,+1 is the money taken into period t + 1, l is nominal borrowing, and d is nominal deposits
in the first market. Rewriting the budget constraint in terms of h and substituting into (5) yields

W (m2 , l, d) = φ [m2 + τ 2 M−1 − (1 + i) l + (1 + id ) d]

+ max [U (x) − x − φm1,+1 + βV (m1,+1 )] .

x,m1,+1

The first-order conditions are

U 0 (x) = 1
(6)
−φ−1 + βV 0 (m1 ) ≤ 0 (= 0 for m1 > 0)

where the first-order condition for money has been lagged one period. Thus, V 0 (m1 ) is the marginal
value of taking an additional unit of money into the first market open in period t, and φ−1 is the
real price of money in the second market of period t − 1 measured in units of utility.
Notice that the optimal choice of x is the same across time for all agents. Furthermore, the m1
chosen in t is independent of m2 in t − 1, since W (m2 , l, d) is linear in m2 so that the marginal
value of money is the same for all agents:

Wm = φ. (7)

The implication is that regardless of how much money the agent brings into the second market, all
agents enter the following period with the same amount of money. Those who bring too much money
into the second market, spend the excess cash on goods, while those with too little money produce
extra output to sell for money.5 As a result, the distribution of money holdings is degenerate at
the beginning of the following period.
Finally, note that

Wl = −φ (1 + i) (8)

Wd = φ (1 + id ) (9)
5
Conditions need to be imposed to ensure h ≥ 0.

7
3.2 The first market

Let qb and qs respectively denote the quantities consumed by a buyer and produced by a seller
trading in market 1. Let p be the nominal price of goods in market 1. For any value of µ > 0,
it is straightforward to show that agents who are buyers will never deposit funds in the bank and
sellers will never take out loans. Thus, ls = db = 0. In what follows we let l denote loans taken out
by buyers and d deposits of sellers. We also drop these arguments in W (m, l, d) where relevant for
notational simplicity.
An agent who has m1 money at the opening of the first market has expected lifetime utility

V (m1 ) = (1 − n) [u (qb ) + W (m1 + τ 1 M−1 + l − pqb , l)]

(10)
+n [−c (qs ) + W (m1 + τ 1 M−1 − d + pqs , d)]

where pqb is the amount of money spent as a buyer, and pqs the money received as a seller. Once
the preference shock occurs, agents become either a buyer or a seller.

Sellers’ decisions If an agent is a seller in the first market, his problem is

max [−c (qs ) + W (m1 + τ 1 M−1 − d + pqs , d)]

qs ,d

s.t. d ≤ m1 + τ 1 M−1

The first-order conditions are

−c0 (qs ) + pWm = 0

−Wm + Wd − λd = 0.

where λd is the Lagrangian multiplier on the deposit constraint. Using (7), the first condition
reduces to
−c0 (qs ) + pφ = 0. (11)

Note that qs is independent of m1 and d. Consequently, sellers produce the same amount no matter
what financial decisions they make. Using (7) and (9), the first-order conditions for d reduces to

φid = λd . (12)

8
Note that if the deposit constraint is non-binding, then id = 0. If the constraint is binding, then
λd
id = φ > 0.

Buyers’ decisions If an agent is a buyer in the first market, his problem is:

max [u (qb ) + W (m1 + τ 1 M−1 + l − pqb , l)]

qb ,l

s.t. pqb ≤ m1 + τ 1 M−1 + l

Notice that buyers cannot spend more cash than what they bring into the first market, m1 , plus
their borrowing, l. As a result buyers do not face a standard cash constraint. They can borrow
cash to supplement their money holdings at the cost of the nominal interest rate. This is equivalent
of buying on credit. The only difference is that the seller is not involved with the credit transaction
as occurs with a credit card transaction. Another difference is that unlike a credit card transaction
the bank knows nothing about the goods transactions.
These differences are illustrated in Figure 2. Figure 2a describes the flow of goods, credit and
money in our model. Note the absence of links between the seller and the bank. Figure 2b describes
the same flow for a credit card transaction with cash settlement.

Bank Bank

Cash
Redeem
IOU IOU IOU
Cash

Goods Goods
Buyer Buyer
Seller Seller
Cash IOU

Cash only trades Credit card with cash settlement

Figure 2a Figure 2b

The buyers first-order conditions are

u0 (qb ) − pWm − pλ = 0 (13)

9
 Wm + Wl + λ = 0 (14)

where λ is the multiplier on the buyer’s cash constraint. Using (7) and (8), (14) reduces to

φi = λ (15)

If the constraint is not binding, then λ = i = 0 and, using (7) and (11), (13) reduces to

u0 (qb ) = c0 (qs ) . (16)

Hence trades are efficient.6

λ
If the constraint is binding, then i = φ > 0. Then, equations (11), (13) and (15) imply

u0 (qb )
=1+i (17)
c0 (qs )

For i > 0, u0 (qb ) > c0 (qs ) which means trades are inefficient. Note that (17) is a common condition
found in money in the utility function or cash in advance models (see Walsh). It simply means that
a positive nominal interest rate acts as tax on consumption. The nominal interest rate depends on
real production and real consumption of the current period only.
When λ > 0 the buyer spends all of his money, pqb = m1 + τ 1 M−1 + l, and consumes qb =
m1 +τ 1 M−1 +l
p . Given the value for loans, buyer’s money holdings in equilibrium satisfy

(1 − µ) n
(1 + τ 1 ) M−1 + (1 + τ 1 ) M−1 = θzM (18)
1−n
1−µn 1+τ 1
where θ = 1−n and z = 1+τ is the fraction of the end of period nominal balances available in
market 1. Evidently, money holdings are increasing in n and decreasing in µ. Note that if τ 1 = 0,
the central bank provides no additional liquidity to the market than what agents bring into the
period. If τ 1 = τ the central bank injects the entire transfer for the period into market 1.

Marginal value of money The marginal value of money satisfies

u0 (qb )
V 0 (m1 ) = (1 − n) + nφ (1 + id )
p
6
With 1 − n buyers and n sellers, the planner maximizes (1 − n) u (qb ) − nc (qs ) s.t. (1 − n) qb = nqs . Use the
constraint to replace qs in the maximand. The first-order condition for qb is u0 (qb ) = c0 (qs ).

10
In the appendix we show that the value function is concave in m.
1
In a symmetric equilibrium, noting that c0 (qs ) = 1/a, we have that p = aφ . In this case V 0 (m1 )
reduces to
£ ¤
V 0 (m1 ) = φ (1 − n) au0 (qb ) + n (1 + id ) . (19)

3.3 Equilibrium

We now derive the symmetric equilibrium. Use (6) to eliminate V 0 (m1 ) and (3) to eliminate id
from (19) to get
φ−1 © ª
= φ (1 − n) au0 (q) + n [1 + (1 − µ) i] .
β
n
because in any symmetric equilibrium qb = 1−n qs =q .
In a steady state equilibrium the real value of money is constant so φM = φ−1 M−1 . Thus
φ−1
γ = φ implying that the real price of money falls over time when γ > 1. The steady state
inflation rate is given by τ . Thus, we can write the first-order condition as
γ−β £ ¤
= (1 − n) au0 (q) − 1 + n (1 − µ) i (20)
β
Next use c0 (qs ) = 1/a in (17)
au0 (q) = 1 + i (21)

Use this to replace i in (20) to get

γ−β £ ¤
= (1 − nµ) au0 (q) − 1 (22)
β
or use (21) to replace au0 (q) in (20) to get
γ−β
= (1 − nµ) i. (23)
β
The nominal interest rate is increasing in µ, n and in γ and decreasing in β.7
We can now state the following.
7 1
Rewriting (23) and noting that β
= 1 + r, where r is the rate of time preference, and γ = (1 + τ ), where τ is the
steady state inflation rate, we get a modified Fisher equation

(1 + τ ) (1 + r) = 1 + i (1 − nµ) = 1 + ei

Note that ei = i (1 − nµ) is the “after-tax” nominal interest rate.

11
Definition 1 A monetary equilibrium with banking is a quantity q satisfying (22).

Proposition 1 For γ ≥ β, a monetary equilibrium exists. It is unique for γ > β. Equilibrium

consumption satisfies q < q ∗ with q → q ∗ as γ → β and is decreasing in µ.

It is clear from (22) that money is neutral, but not super-neutral. Increasing its stock has no
effect on q, while changing the growth rate γ does. Thus, monetary policy can have real effects in
this model. By holding required reserves, banks restrict lending which makes consumption more
costly. As a result, the required reserve ratio acts like a tax on consumption.
Note that the solution for q does not depend on τ 1 . So varying the transfer across the two
markets has no effect on q and only affects the equilibrium price of money in the last market. To
1
see this, note that in equilibrium since p = φ and pq = θ (1 + τ 1 ) M−1

q (1 + τ )
φ= .
aθ (1 + τ 1 ) M

For a given value of τ , if τ 1 is increased, buyers receive more of their total transfer in market 1.
This transfer provides liquidity insurance so that agents reduce the amount of money they want to
bring into market 1. Thus the demand for money declines in market 2 which reduces the price of
money since the stock is fixed.

Corollary 1 Banks improve the allocation and welfare when µ < 1 and γ > β.

The key result of this section is that banks improve the allocation in the economy away from
the Friedman rule. Most interestingly, banks have their greatest impact on welfare for moderate
values of inflation. The reason is that near the Friedman rule there is little gain from redistributing
excess cash balances while for high inflation rates money is of little value anyway. This is shown in
Figure 2 where the difference in consumption is drawn as a function of the inflation rate γ:

Insert Figure 2 (Comparing q with and without banking)

Finally, note that in equilibrium the ratio of aggregate loan repayments to the money stock is
given by
n (γ − βnµ) (1 − µ) (1 + τ 1 )
v=
βγ (1 − nµ)

12
where v is increasing in the gross rate γ and in τ 1 . Thus, the velocity of money in market 2 is
increasing in the steady state inflation rate.

3.4 Incentive to repay

To do...: In equilibrium sellers work less because they use money and deposits to buy goods, while
buyers work more than they consume to acquire money and to repay loans. Note that the hours
buyers work are decreasing in µ for a given value of γ > β. The reason is that as µ increases, the
total repayment made by buyers falls. As we will see below, an incentive to default on one’s loan
is the fact that one does not have to work as much to repay the loan in the second sub-period.
Consequently, increases in µ reduces the immediate benefit of default on one’s loan. Thus, banks
may choose to set µ > 0 to prevent default.

4 Preference shocks

In this section we will consider two types of preference shocks: shocks to the composition of buyers
and sellers and shocks to the marginal utility of consumption. The difference between the two is
that in the first type of shock deposits will fluctuate since changes in the number of sellers affects
the amount of idle cash across states. This is not the case in the second type of shock. We refer to
the first type of shocks as liquidity shocks.

4.1 Liquidity shocks

When the measures of buyers and sellers are random the total demand for goods is also random.
Although the measure of sellers is random, with linear production costs the supply of output is
horizontal, so equilibrium output is entirely demand determined. By varying the number of sellers
we also change the amount of idle cash balances in the economy. When banks intermediate deposits,
this implies deposits and loans will be random. This suggests that bank may amplify shocks to the
economy. In fact, we show below that although banks increase consumption they also create more
consumption volatility for individual agents.

13
 With probability 1 − nt an agent wants to consume but cannot produce while with probability
nt the agent can produce but does not want to consume with
⎧
⎨ nH with probability π
nt =
⎩ nL with probability 1 − π

where H denotes a high seller state and L denotes a low seller state. This means when state
H occurs, there are many sellers and few buyers so aggregate deposits are high and aggregate
loan demand is low. Thus, n also represents an aggregate deposit shock. We also assume that
n = πnH + (1 − π) nL where n is the number of sellers in the deterministic model. This allows us
to compare across models.
Because there is randomness in deposits, the central bank may choose to make the market 1
injections contingent on the aggregate state of the economy. Thus, let τ H
1 M−1 be the transfer in
j j
state H and τ L
1 M−1 be the transfer in state L where τ 1 + τ 2 = τ = γ − 1 for j = H, L.

The agents’ optimization choices for how much to buy and sell are unchanged with the only
difference being that the choices are state contingent. The value function in market 1 is

V (m1 ) = πV H (m1 ) + (1 − π) V L (m1 ) (24)

where for j = H, L
¡ ¢h ³ ´ ³ ´i
V j (m1 ) = π 1 − nj u qbj + W m1 + τ j1 M−1 + lj − pj qbj , lj
h ³ ´ ³ ´i
+ πnj −c qsj + W m1 + τ j1 M−1 − dj + pj qsj , dj

The sellers choose to produce

¡ ¢
φj pj = c0 qsj = 1/a

while buyers set ³ ´

u0 qbj = 1 if λj = 0
(25)
qbj = aθj z j φj M if λj > 0
1+τ j1 1−µnj
for j = H, L, where z j = 1+τ , θj = 1−nj
and θH > θL . This last inequality follows from the
fact that in the high state deposits are also high thus each borrower has more cash. Finally, let
³ j ´
qbj = 1−n
n
j qsj = qj for j = H, L.
We can now state the following

14
Proposition 2 Assume z H θH ≥ z L θL . For γ ≥ β, a monetary equilibrium exists with q H = qL =
q ∗ at γ = β. For γ ∈ (β, γ
e] there is a unique monetary equilibrium with q L < q H = q ∗ . For γ > γ
e,
there is a unique monetary equilibrium with q L < q H < q ∗ .

Once again the Friedman rule replicates the first-best allocation. At the Friedman rule there is
no value of allocating excess cash. Obviously there is no role for banks and no role for stabilization.
Away from the Friedman rule when inflation is low, Proposition (2) states that buyers are
constrained in the low deposit state but not in the high state. Thus, the nominal interest rate is
zero in the high state. In the low state it is positive and satisfies
γ−β
iL =
β (1 − π) (1 − nL µ)
From these equations it is clear that the nominal interest rate is increasing in γ and the reserve ratio
µ. Interestingly, since the nominal interest rate is independent of the injection z L consumption q L
is also independent. Consequently, the central bank cannot affect consumption in market 1 in the
low inflation economy given γ.
In contrast, when inflation is high the central bank can affect consumption in two ways. First,
it can change the steady state growth rate of the money supply γ. Second, by making the first
market injections state contingent. The first effect is standard. The second effect can be seen by
looking at the consumption ratio which satisfies
qH θH z H
=
qL θL z L
in the high inflation economy. From this condition it is clear that the central bank can affect the
relative magnitudes of consumption in the two states. If z H θH ≥ z L θL , then qH > q L .
Finally, it should be noted that if z H θH < z L θL then Proposition 2 must be modified such that
when inflation is low agents are constrained in the high state and not in the low state and q L > q H
for all γ.

4.1.1 Role of banks

We want to compare the equilibrium outcome in the banking economy to the no-banking case
when these shocks occur. For this purpose we eliminate for now policy responses to these shocks.

15
Consequently, assume the central bank acts ‘passively’ and sets τ 1 such that z H = z L = z.
We can replicate the no-banking equilibrium simply by setting µ = 1. In this case the quantities
consumed in the no-banking equilibrium are

q H = q L = q = azΩ

where Ω solves
γ−β £ ¤
= (1 − n) au0 (azΩ) − 1 .
β
Since we have assumed n = πnH + (1 − π) nL , Ω has the same value as in the deterministic case.
Hence, in the no banking equilibrium, buyers consume the same quantity across states since they
can only spend the cash they bring into market 1 which is independent of the state that is realized.
This occurs for all inflation rates. The problem is that while there are many sellers in state H
holding idle cash, it cannot be reallocated to buyers. Aggregate consumption in the economy in
state H is lower however since there are few buyers in this state.
Now consider what happens when banks exist, i.e. θH > θL > 1. Now the idle cash from
sellers is deposited into the banking system and lent back out to buyers. In state H, there is a
large number of depositors and relatively few borrowers. Hence, the banks must lower the nominal
interest rate to get rid of the excess cash. The lower nominal interest rate expands borrowing and
leads to higher quantities of goods consumed by buyers. In state L, the opposite occurs - there
are relatively few depositors and a large number of borrowers. This forces the nominal interest
rate to rise and reduces the quantity of goods per buyer that can be purchased. Thus, qH > q L
in equilibrium. Nevertheless, the quantities consumed in both states are higher with banks than
without banks.
The interesting aspect of this result is that while banks raise average consumption across states,
they transmit deposit shocks to buyers making individual consumption more volatile. Nevertheless,
it is straightforward to show that aggregate consumption is more stable when measured by the
coefficient of variation (See appendix). The next question to address is whether or not a passive
monetary policy is optimal.

16
4.1.2 Optimal stabilization policy

What is the optimal response of the central bank to these shocks? We assume the central bank’s
stabilization policy maximizes the welfare of the representative agent for a given a steady-state
inflation rate. In short, suppose the central bank could pick the quantities consumed and produced
in each state subject to the constraint that the chosen quantities satisfy the conditions of a compet-
itive equilibrium. This is a standard Ramsey problem. We now assume that the growth rate of the
money supply is determined exogenously to finance a fixed amount of government consumption in
market 2. This implies τ > 0. However, the central bank has the power to set τ j1 and τ j2 such that
the growth rate is maintained at the exogenously given value. By choosing τ j1 and τ j2 the central
bank determines the quantities consumed in each state.
As shown in the appendix, the steady state welfare of the representative agent at the beginning
of period t is given by
∙ ¸ ∙ ¸
¡ ¢ ¡ ¢ qH ¡ ¢ ¡ ¢ qL
(1 − β) V (M−1 ) = U (x) − x − g + π 1 − nH u q H − + (1 − π) 1 − nL u qL −
a a
where g is the real amount of government spending that is financed in market 2. It is obvious that
x = x∗ so all that remains is to choose q H and q L .
The Ramsey problem facing the central bank is
µ ¶ µ ¶
¡ H
¢ ¡ H ¢ qH ¡ L
¢ ¡ L¢ qL
M ax π 1 − n u q − + (1 − π) 1 − n u q −
qH ,q L a a
γ−β ¡ ¢ £ ¡ ¢ ¤ ¡ ¢ £ ¡ ¢ ¤
s.t. = π 1 − nH θH au0 q H − 1 + (1 − π) 1 − nL θL au0 q L − 1 (26)
β
where the constraint is derived in the appendix. From now on we consider high enough levels of g
and γ such that buyers are liquidity constrained in both states.
The conditions for a maximum (impose au00 (q) u00 (q) − [au0 (q) − 1] u000 (q) > 0) are
¡ ¢ 1 ¡ ¢
u0 q H − + λθH au00 qH = 0
a
¡ ¢ 1 ¡ ¢
u0 qL − + λθL au00 q L = 0.
a
for agents to want to hold a finite amount of money, λ > 0 so q H and q L are inefficiently low. This
is due to the fact that γ > 1 in this problem. The first-order conditions can be rewritten as
¡ ¢ ¡ ¢
au0 qL − 1 u00 q H θL
= (27)
u00 (q L ) au0 (qH ) − 1 θH

17
We can now state the following:

Proposition 3 The constrained planner’s choice of quantities yields q L = q H when there are no
banks and q L > q H with banks.

Without banks it is clear from (27) that the planner chooses q H = qL which implies z H = z L = z.
The reason is that although the central bank can alter q H and q L via its choice of τ H L
1 and τ 1 , doing

so would only create consumption variability which is welfare reducing. Consequently, the best
stabilization policy is to keep quantities constant across states. Note that this implies aggregate
consumption still varies across states.
Now suppose banks exist implying θH > θL > 1. With banks the ordering on the quantities is
exactly reverse from what happens when the central bank is passive. With a passive policy, buyers
consume more in state H since loans are plentiful and cheap in that state. However, this is just
the opposite of what is optimal. From the viewpoint of the representative agent looking into the
future, he wants to consume more when he is more likely to desire consumption, which is when he
is a buyer. This corresponds to state L. However this is when there is little liquidity in the banking
system since the number of depositors is low. Thus, what the planner would like to do is put more
liquidity in the system when aggregate demand is high and deposits are low then take it out when
aggregate demand is low and deposits are high. To accomplish this, the central bank injects more
cash into the system in state L and less cash in state H. Hence, τ L H
1 > τ 1 and these transfers are

chosen to generate the quantities solving (26) and (27).

An interesting implication of this policy is that the central bank is essentially providing an elastic
supply of currency to the economy — when demand for liquidity is high, the central bank provides
additional currency and withdraws it when the demand for liquidity is low. Note what this does
NOT imply — a constant nominal interest rate across states, which would imply that consumption
is the same in both states, i.e. q H = q L . It is easy to show that stabilizing interest rates would
be welfare improving compared to the passive monetary policy but having them fluctuate in the
correct fashion is optimal.

18
 qLHzL
qL,qH
1

qHHzL

qLHzL L
0.95

0.9 qHHzH L

0.85

0.8

µ
0.2 0.4 0.6 0.8 1

Figure 3. Equilibrium vs optimal consumption

Figure 3 shows equilibrium and optimal consumption as functions of the reserve ratio µ for a
given inflation rate γ. Note that for low reserve ratios equilibrium consumption is efficient in state
H and inefficiently low in state L. Interestingly, increasing the reserve ratio decreases consumption
in the low state as long as the economy is in the first equilibrium. The opposite is true in the
second equilibrium. If we compare the central banks solution with the equilibrium quantities we
see two things. First, the ordering of the quanties are reversed across states. Second, the central
bank smooth consumption across states.

4.2 Aggregate marginal utility shocks

Now we assume that ε is a random variable where

⎧
⎨ εH with probability π
εt =
⎩ εL with probability 1 − π

Thus, ε represents aggregate marginal utility shocks with εH > εL . To study these shocks, we
assume that nH = nL = n. So deposits are no longer random.
The agents’ optimization choices for how much to buy and sell are unchanged with the only
difference being that the choices are state contingent. The value function is still given by (24) with

19
nH = nL = n. The sellers choose to produce
¡ ¢
φj pj = c0 qsj = 1/a

while buyers set ³ ´

εj u0 qbj = 1/a if λj = 0
qbj = az j θΩ if λj > 0
³ ´
where qbj = qsj = q j for j = H, L.
n
1−n
³ ´
The optimal quantities solve εj u0 qbj∗ = 1/a.
We can now state the following

q H∗ q L∗
Proposition 4 Assume zH
> zL
. For γ ≥ β, a monetary equilibrium exists with q H = q L = q ∗
b] there is a unique monetary equilibrium with q H < q H∗ and q L = q L∗ . For
at γ = β. For γ ∈ (β, γ
b, there is a unique monetary equilibrium with q L < q L∗ and q H < qH∗ .
γ>γ

Once again the Friedman rule replicates the first-best allocation. At the Friedman rule there is
no value of allocating excess cash. Obviously there is no role for banks and no role for stabilization.
Away from the Friedman rule when inflation is low, Proposition 4 states that buyers are con-
strained in the high marginal utility state but not in the low state. Thus, the nominal interest rate
is zero in the low state. In the high state it is positive and satisfies
γ−β
iH =
βπ (1 − nµ)
From these equations it is clear that the nominal interest rate is increasing in γ and the reserve ratio
µ. Interestingly, since the nominal interest rate is independent of the injection z H consumption q H
is also independent. Consequently, the central bank cannot affect consumption in market 1 in the
low inflation economy given γ.
In contrast, when inflation is high the central bank can affect consumption in two ways. First,
it can change the steady state growth rate of the money supply γ. Second, by making the first
market injections state contingent. The first effect is standard. The second effect can be seen by
looking at the consumption ratio which satisfies
qH zH
=
qL zL

20
in the high inflation economy. From this condition it is clear that the central bank can affect the
relative magnitudes of consumption in the two states.
q H∗ q L∗
Finally, it should be noted that if zH
< zL
then Proposition 4 must be modified such that
agents are constrained in the low state and not in the high state when inflation is low.
In contrast to the case of liquidity shocks, banks neither amplify nor dampen the marginal
utility shocks. The intuition is that they do not affect the nominal amount of borrowing across
states. All what banks do is to move idle cash from sellers to borrowers and consequently the
amount deposited is the same in both states.

4.3 Optimal stabilization policy

What is the optimal response of the central bank to these utility shocks? With these shocks the
Ramsey problem facing the central bank is
∙ ¸ ∙ ¸
H
¡ H ¢ qH L
¡ L¢ qL
M ax π (1 − n) ε u q − + (1 − π) (1 − n) ε u q −
q H ,q L a a
γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
s.t. = (1 − n) θ π aεH u0 qH − 1 + (1 − π) aεL u0 qL − 1 (28)
β
The conditions for a maximum (impose aεu00 (q) u00 (q) − [aεu0 (q) − 1] u000 (q) > 0) are
¡ ¢ 1 ¡ ¢
εH u0 qH − + λθaεH u00 q H = 0
a
¡ ¢ 1 ¡ ¢
εL u0 q L − + λθaεL u00 q L = 0
a
for agents to want to hold a finite amount of money, λ > 0 so q H and q L are inefficiently low. This
is due to the fact that γ > 1 in this problem. The first-order conditions can be rewritten as
¡ ¢ ¡ ¢
aεL u0 q L − 1 aεH u0 q H − 1
= . (29)
εL u00 (qL ) εH u00 (q H )
We can now state the following:

Proposition 5 The constrained planner’s choice of quantities yields qH > q L .

There are several possibilities to implement the optimal policy. However, we focus on z L +z H = z
to compare with the passive policy. Note that any other restriction on the z’s only affects Ω but
not the quantities q L and q H .

21
5 Aggregate productivity shocks

Now we assume that a is a random variable where

⎧
⎨ aH with probability π
at =
⎩ aL with probability 1 − π

Thus, a represents real aggregate productivity shocks with aH > aL . To study these productivity
shocks, we assume that nH = nL = n. So deposits are no longer random.
The agents’ optimization choices for how much to buy and sell are unchanged with the only
difference being that the choices are state contingent. The value function is still given by (24) with
nH = nL = n. The sellers choose to produce

¡ ¢
φj pj = c0 qsj = 1/aj

while buyers set ³ ´

u0 qbj = 1 if λj = 0
qbj = aj z j θΩ if λj > 0
³ ´
where qbj = n
1−n qsj = q j for j = H, L.
In what follows we assume that
¡ ¢ ¡ ¢
qH∗ u0 qH∗ q L∗ u0 q L∗
> . (30)
zH zL

We can now state the following

e)
Proposition 6 Assume that (30) holds. For γ ≥ β, a monetary equilibrium exists. For γ ∈ [β, γ
there is a unique monetary equilibrium in which the buyers are constrained in the high productivity
e, buyers are constrained in both states.
state but not in the low productivity state. For γ ≥ γ

Proposition 6 states that when inflation is low and (30) holds, agents are constrained in the
high productivity state but not in the low state. In the high productivity state prices fall which
raises the real value of money balances. This allows agents to buy more. However, the efficient
quantity q H∗ goes up even more. The reverse is true in the low productivity state. Even though

22
the real balances fall, agents have sufficient real balances to buy q L∗ . Note that if inequality (30)
is reversed, the opposite happens — buyers are constrained in the low productivity state and not in
the high.
When inflation is low and (30) holds, the nominal interest rate is zero. In the high productivity
state the nominal interest rate is positive and satisfies

γ−β
iH =
β (1 − nµ)

From these equations it is clear that the nominal interest rate is increasing in γ and the reserve
ratio µ since θ is decreasing in µ. Interestingly, the nominal interest rate and hence consumption
is independent of the injection z H . So the central bank cannot affect consumption in market 1 in
the low inflation economy given γ.
When inflation is high, buyers are constrained in both states which implies that the consumption
qH aH z H aH zL
ratio satisfies qL
= aL z L
. Thus q H > qL if aL
> zH
. The key result here is that the banking
system has no impact on the variability of consumption across states since θ does not appear in
qH aH z H
the ratio qL
= aL z L
. However, banks still raise the level of consumption in each state.
In contrast to the case of liquidity shocks, banks do neither amplify nor dampen the productivity
shocks. The intuition is that the productivity shocks do not affect the nominal amount of borrowing
across states. All what banks do is to move idle cash from sellers to borrowers and consequently
the amount deposited is the same in both states.

5.1 Optimal stabilization policy

What is the optimal response of the central bank to aggregate productivity shocks? With produc-
tivity shocks the Ramsey problem facing the central bank is
∙ ¸ ∙ ¸
¡ H ¢ qH ¡ L
¢ ¡ L¢ qL
M ax π (1 − n) u q − H + (1 − π) 1 − n u q − L
q H ,q L a a
γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
s.t. = (1 − n) θ π aH u0 q H − 1 + (1 − π) aL u0 q L − 1 (31)
β

23
The conditions for a maximum (impose au00 (q) u00 (q) − [au0 (q) − 1] u000 (q) > 0) are

¡ ¢ 1 ¡ ¢
u0 q H − H + λθaH u00 q H = 0
a
¡ ¢ 1 ¡ ¢
u0 q L − L + λθaL u00 q L = 0
a

for agents to want to hold a finite amount of money, λ > 0 so q H and q L are inefficiently low. This
is due to the fact that γ > 1 in this problem. The first-order conditions can be rewritten as
¡ ¢ ¡ ¢
u0 q L − a1L u0 qH − a1H
= H 00 H . (32)
aL u00 (q L ) a u (q )

We can now state the following:

Proposition 7 The constrained planner’s choice of quantities yields qH > qL iff γ < γ c and
q H < qL otherwise.

There are several possibilities to implement the optimal policy. However, we focus on z L +z H = z
to compare with the passive policy. Note that any other restriction on the z’s only affects Ω but
not the quantities q L and q H .

6 Conclusion

24
References (to be edited)
Alvarez, F. A. Atkeson and P. Kehoe (2002). Money, Interest Rates and Exchange Rates with
Endogenously Segmented Markets. Journal of Political Economy 110, 1, 73-112.
Aruoba, B., C. Waller and R. Wright, (2003). Money and Capital. Working paper, University
of Pennsylvania.
Berentsen, A., G. Camera and C. Waller (2003). Short Run and Long Run Effects of Money
Creation in an Essential Model of Money.
Cavalcanti, R. and Wallace, N., (1999). A Model of Private Bank-Note Issue. Review of
Economic Dynamics 2, 1, 104-136.
Christiano, L. and M. Eichenbaum (1992). Liquidity Effects and the Monetary Transmission
Mechanism. American Economic Review 82, 2, 346-353.
Eden, Ben (1994). The Adjustment of Prices to Monetary Shocks When Trade is Uncertain
and Sequential. Journal of Political Economy 102, 3, 493-509.
Fuerst, T. (1992). Liquidity, Loanable Funds and Real Activity. Journal of Monetary Economics
29, 1, 3-24.
Lagos, R. and R. Wright, (2002). A Unified Framework for Monetary Theory and Policy
Evaluation. Working paper, University of Pennsylvania.
Kocherlakota, N. (2003). Societal Benefits of Illiquid Bonds. Journal of Economic Theory, 108,
179-193.
He, P., L. Huang and R. Wright (2003). Money and Banking in Search Equilibrium. Unpub-
lished working paper, University of Pennsylvania.
Lucas, R. (1990). Liquidity and Interest Rates. Journal of Economic Theory 50, 2, 237-264.
Stockman, A. (1981). Anticipated Inflation and the Capital Stock in a Cash-in-advance Econ-
omy. Journal of Monetary Economics, 8, 387-393.

25
 Appendix

Marginal value of money Differentiating (10) with respect to m1

h ³ ´ i
1 ∂qb ∂qb ∂lb ∂lb
V 0 (m1 ) = 2 u0 (qb ) ∂m 1
+ Wm 1 − p ∂m1 + ∂m1 + Wl ∂m1
h ³ ´ i
1 0 ∂qs ∂qs ∂ds ∂ds
+ 2 −c (qs ) ∂m1 + Wm 1 + p ∂m1 − ∂m1 + Wd ∂m 1

Recall from (7), (8), and (9) that Wm = φ, Wl = −φ (1 + i) and Wd = φ (1 + id ) ∀m2 . Furthermore,
∂qs
∂m1 = 0 because the quantity a seller produces is independent of his money holdings. We also know
∂ds
that ∂m1 = 1 since a seller deposits all his cash when i > 0. Hence,
h ³ ´ i
1 ∂qb ∂qb ∂lb ∂lb
V 0 (m1 ) = 2 u0 (qb ) ∂m 1
+ φ 1 − p ∂m 1
+ ∂m1 − φ (1 + i) ∂m 1
+ 12 [φ (1 + id )]

³ ´
∂qb ∂lb
Since i > 0 implies pqb = m1 + lb we have 1 − p ∂m 1
+ ∂m1 = 0. Hence8

1 u0 (qb ) φ
V 0 (m1 ) = + (1 + id ) (33)
2 p 2

In a symmetric equilibrium with equal number of buyers and sellers qb = qs = q. Define m∗ = pq ∗

∂qb
where q ∗ solves u0 (q) = c0 (q). Then if m1 < m∗ , 0 < qb < q ∗ , implying ∂m1 > 0 so that
∂qb
V 00 (m1 ) < 0. If m1 ≥ m∗ , qb = q∗ implying ∂m1 = 0, so that V 00 (m1 ) = 0. Thus, V (m1 ) is concave
∀m.¥
h i
8 ∂qb ∂lb ∂qb ∂qb
Note that u0 (qb ) ∂m1
− Aφ (1 + i) ∂m 1
= u0
(qb ) ∂m − Aφ (1 + i) p ∂m − 1
1 1
∂qb u0 (qb )
= ∂m1
(u0 (qb ) − Aφ (1 + i) p) + Aφ (1 + i) = Aφ (1 + i) = p
.

26
Proof of Proposition 1: Because u(q) is strictly concave there is a unique value q that solves
(17), and for γ > β, q < q ∗ where q∗ is the efficient quantity solving u0 (q ∗ ) = 1/a. As γ → β,
u0 (q) → 1/a, q → q ∗ , and from (23) i → 0. In this equilibrium, the Friedman rule sustains efficient
trades in the first market. Furthermore, from (17) q is decreasing in µ. Since there is no lending
when µ = 1 the allocation corresponds to the model without banking.
Recall that, due to idiosyncratic trade shocks and financial transactions, money holdings are
heterogeneous after the first market closes. Therefore, if we set m1 = M−1 , the money holdings of
¡ ¢
agents at the opening of the second market are m2 = 0 for buyers and m2 = 1−n n θ (1 + τ 1 ) M−1
for sellers.
We can now solve for equilibrium consumption and production in the second market. Equation
(6) gives us x∗ = U 0−1 (1). Production is given in the following table

Trading history: Production in the last market:

q ¡ ¢
Buy hb = x∗ + aθ + θ−1 θ e (q) u (q)
¡ ¢ £ q ¡ θ−1 ¢ ¤
Sell hs = x∗ − 1−nn aθ + θ e (q) u (q)

Since we assumed that the elasticity of utility e (q) is bounded, we can scale U (x) such that
there is a value x∗ = U 0−1 (1) greater than the last term for all q ∈ [0, q ∗ ]. Hence, hs is positive for
for all q ∈ [0, q∗ ] ensuring that the equilibrium exists.
¥

27
Proof of Proposition that borrowers repay
Proof: Equation (22) and (23) immediately imply that if the banking equilibrium exists it is unique.
Solving the model without banking yields the same q that satisfies (22) when µ = 1. Since q is
decreasing in µ it follows that aggregate production q in the banking equilibrium is larger than in
the non-banking equilibrium.
Existence: We now derive the conditions under which borrowers repay their loans to ensure
that the equilibrium exists. For buyers entering the second market with no money, who repay their
loans, the expected discounted utility in equilibrium is
X∞ ∙ ¸
1 1
U = U (x∗ ) − hb + β t u (q) − c (q) + U (x∗ ) − h
t=1
2 2

and h = 12 hb + 12 hs where all variables are evaluated at their equilibrium values.

Consider the case of a buyer who reneges on his loan. The benefit of reneging is that he has
more leisure in the second market because he does not work to repay the loan. The cost is that
he is out of the banking system, meaning that he cannot borrow or lend for the rest of his life.
He cannot lend because the bank would confiscate his deposits to settle his loan arrears. Thus, a
deviating buyer’s expected discounted utility is
X∞ ∙ ¸
b b t 1 1 b
U = U (bx) − hb + β u (b
qb ) − c (b
qs ) + U (b
x) − h
t=1
2 2

b, b
where the hat indicates the optimal choice by a deviator. We now derive x h, qbb , and qbs . In the
last market the deviating buyer’s program is
h i
b 2) =
W (m max x) − b
U (b b 1,+1 )
h + βV (m (34)
b,b
x b 1,+1
h,m

s.t. x b 1,+1 = b
b + φm b 2 + τM)
h + φ (m

As before, the first-order conditions are

U 0 (b
x) =
(35)
−φ−1 + βV 0 (m
b 1) ≤ 0 b 1 > 0)
(= 0 for m

b = x∗ . Note that Wm (m2 ) = φ, which is the same marginal value for non-deviators.
Thus, x
The production and consumption decisions are affected as follows. The first-order condition if the

28
deviator becomes a seller in the first market is

−c0 (b b 1 + pb
qs ) + pWm (m qs ) = 0.

Since Wm (m2 ) = φ and agents take prices as given we have qbs = q. The first-order condition if the
deviator becomes a buyer is

u0 (b b 1 − pb
qb ) − pWm (m b=0
qb ) − pλ

Finally, the marginal value of the money satisfies

∙ ¸
0 φ u0 (b
qb )
V (mb 1) = +1
2 c0 (q)

which means that (35) can be written as

∙ ¸
γ−β 1 u0 (b
qb )
= −1 (36)
β 2 c0 (q)

Now if we compare (36) with (22) we find that

£ ¤
u0 (b
q ) − u0 (q) = (1 − µ) u0 (q) − c0 (q)

which implies that qbb < q since u0 (q) − c0 (q) > 0 because q < q∗ . Therefore u (q) − c (q) >
qb ) − c (b
u (b qb ) . In the second market we get the following values for b
h:

Trading history for deviator Production in the last market:

Buy b
hb = x∗ + φm
b1
Sell b
hs = x∗ − θφM

The key differences are b

hb < hb and b
hs > hs . Since the deviator cannot borrow when he is a
buyer, he has to work less since he doesn’t have to repay a loan. However, because he has no interest
income he has to work more in the second market when he sells in the first market compared to a
non-deviator. Finally, let us consider the difference D = U−Ub where
h i β β h i
D= b
hb − hb + [u (q) − u (b
qb )] + b
h−h
2 (1 − β) 2 (1 − β)

29
Substituting for production at night we get

β φβ
b − θM ] +
D = −i (1 − µ) φM + φ [m [u (q) − u (b
qb )] + b − θM ]
[m
2 (1 − β) 2 (1 − β)

b = pb
Note that in equilibrium m qb and θM = pq. Consequently, we have

i (1 − µ) c0 (q) q β © ª
D=− − c0 (q) [q − qbb ] + qb ) + c0 (q) (b
u (q) − u (b qb − q)
θ 2 (1 − β)

Consider the case where c(q) = q. Then c0 (q) = 1. So

2 (γ − β) (1 − µ) q β
D=− 2 − (q − qbb ) + qb ) − qbb ]}
{[u (q) − q] − [u (b
βθ 2 (1 − β)

The first two terms are negative and bounded as β → 1.9 Since qbb < q < q∗ , it follows that
qb ) − qbb . Consequently, the last term is positive. As β → 1, the last term goes to
u (q) − q > u (b
infinity. Hence for a sufficiently high value of β we will have D < 0. So, the banking equilibrium
will exist and all loans will be repaid. The intuition for this is clear - not being in the banking
system imposes future utility losses and with sufficiently patient agents, the present discounted
value of these losses dominates any short-term gain from defaulting on the loan. Most importantly,
for sufficiently high values of β, the value of µ is irrelevant as long as µ < 1. Thus, banks can lend
out all deposits, µ = 0, which maximizes the value of q in equilibrium.
Alternatively, consider a value of β such that D = 0 when µ = 0.Then for a value β slightly
below this critical value, agents would default if µ = 0 and the benefits of the banking system
would be lost. However, by increasing µ above zero, banks can reduce the size of loan repayments
such that the incentive to renege falls enough that agents choose to repay their loans. Thus, idle
reserves may be necessary to improve the allocation when agents are not sufficiently patient.
9
Obviously this requires changing γ if need be to ensure γ > β.

30
Proof of Proposition 2: Taking the expectation of (19) with respect to the liquidity shocks and
using (6) yields

φ−1 £¡ ¢ ¡ ¢ ¡ ¢¤
= πφH 1 − nH au0 q H + nH 1 + iH
d
β
£¡ ¢ ¡ ¢ ¡ ¢¤
+ (1 − π) φL 1 − nL au0 qL + nL 1 + iLd .
³ ´ ³ ´
nH nL
where qbH = 1−nH
qsH = qH and qbL = 1−nL
qsL = q L .
In a steady state equilibrium the real value of money is constant so φH M = φL M = Ω which
implies φH = φL = φ = 1
γ φ−1 . This implies pH = pL = 1/φ. Thus, using (3) we can write the
first-order condition as

γ £¡ ¢ ¡ ¢ ¡ ¢¤ £¡ ¢ ¡ ¢ ¡ ¢¤
= π 1 − nH au0 q H + nH 1 + iH
d + (1 − π) 1 − nL au0 qL + nL 1 + iL
d (37)
β
¡ ¢ 1−µnH 1−µnL
In any equilibrium pj q j ≤ θj 1 + τ j M−1 where θH = 1−nH
> θL = 1−nL
. Then, the seller’s
1
first-order condition p = aφ implies
q j ≤ aθj z j Ω

1+τ j
where z j = 1+τ is greater in one state than the other state. The efficient quantity is defined to
be u0 (q∗ ) = 1/a. If trades are efficient, q = q ∗ ≤ aθj z j Ω. Therefore, two types of equilibria are
feasible.

Low Inflation Economy: For the conjectured equilibrium in the low inflation economy,

q H = q∗ < aθH z H Ω and q L = aθL z L Ω < q ∗ .

Using these expressions we obtain aθL z L Ω < q∗ < aθH z H Ω respectively

q∗ q∗
< Ω < (38)
aθH z H aθL z L

which is a non-empty interval.

In the low inflation economy in equilibrium, q H = q ∗ . This implies from (17) and (3) that
iH = iH
d = 0 and
¡ ¢ 1 L
au0 qL = 1 + iL = 1 + i . (39)
1−µ d

31
Thus, (37) can be solved to obtain a single equation in Ω

γ−β ¡ ¢ £ ¡ ¢ ¤
= (1 − π) 1 − nL θL au0 aθL z L Ω − 1
β

or a single equation in iL
γ−β
iL = (40)
β (1 − π) (1 − nL µ)
It is clear that the value of iL solving (40) is greater than zero for γ > β and approaches zero as
γ → β. Furthermore, we have q L → q ∗ as γ → β. Given this solution for iL we can solve for q L , iL
D

and Ω. The right-hand side of (40) is monotonically increasing in γ. Thus, an increase in the money
growth rate, raises the nominal borrowing rate, increasing the cost of consumption and reducing
q. Since q decreases in γ, Ω decreases as well. Given this behavior on iL it is straightforward to
q∗
show that as γ → β, Ω → aθL z L
so that the right-hand inequality is binding in (38). Thus, as γ
increases from the Friedman rule, Ω declines and satisfies (38). For a sufficiently high value of γ,
e the value of γ that solves q ∗ = aθH z H Ω. Then, a
the left-hand inequality is violated. Denote γ
e).
unique monetary equilibrium exists for γ ∈ [β, γ

High inflation economy: For the conjectured equilibrium in the high inflation economy,

q H = aθH z H Ω < q ∗ and q L = aθL z L Ω < q ∗ . (41)

θH z H L
Note that q H = θL z L
q > qL because θH z H > θL z L .
From (3) and (17) we have

¡ ¢ 1 j
au0 q j = 1 + ij = 1 + i j = L, H (42)
1−µ d

Thus, (37) can be solved to obtain a single equation in Ω

γ−β ¡ ¢ £ ¡ ¢ ¤ ¡ ¢ £ ¡ ¢ ¤
= π 1 − nH θH au0 aθH z H Ω − 1 + (1 − π) 1 − nL θL au0 aθL z L Ω − 1 . (43)
β
e −β
γ q∗
The right-hand side expression is monotonically decreasing in Ω and is equal to β at Ω = aθH z H
.
e, a unique value of Ω exists such that (41) holds.¥
Thus for γ > γ

32
Proof that banks lower aggregate consumption: Consider the no bank case. Aggregate
consumption is given by:

¡ ¢
QH = 1 − nH q when there are few buyers
¡ ¢
QL = 1 − nL q when there are many buyers

where q is the steady state value of individual consumption. Expected aggregate consumption is:

Q̄n = πQH + (1 − π) QL
¡ ¢ ¡ ¢
= π 1 − nH q + (1 − π) 1 − nL q
£ ¡ ¢ ¡ ¢¤
= π 1 − nH + (1 − π) 1 − nL q (44)

The variance of aggregate consumption is:

£ ¤2 £ ¤2
Vn (Q) = π QH − Q̄n + (1 − π) QL − Q̄n
£¡ ¢ £ ¡ ¢ ¡ ¢¤ ¤2
= π 1 − nH q − π 1 − nH + (1 − π) 1 − nL q
£¡ ¢ £ ¡ ¢ ¡ ¢¤ ¤2
+ (1 − π) 1 − nL q − π 1 − nH + (1 − π) 1 − nL q
£¡ ¢ ¡ ¢¤
= 2π (1 − π) 1 − nL − 1 − nH q 2

Using (44) the coefficient of variation is

p " ¡ ¢ ¡ ¢ #
Vn (Q) p 1 − nL − 1 − nH
ρn ≡ = 2π (1 − π) (45)
Q̄n π (1 − nH ) + (1 − π) (1 − nL )

Now consider the equilibrium with banks. Aggregate consumption is given by:

¡ ¢
QH = 1 − nH q H when there are few buyers
¡ ¢
QL = 1 − nL qL when there are many buyers

Expected aggregate consumption is:

Q̄b = πQH + (1 − π) QL
¡ ¢ ¡ ¢
= π 1 − nH qH + (1 − π) 1 − nL qL

33
We know that in this equilibrium q H = aθH Ω and q L = aθL Ω with θH > θL implying

θH L
qH = q
θL

Substitute in expected aggregate consumption to get

∙ ¸
¡ ¢ H
H θ
¡ L
¢ L
Q̄b = π 1 − n + (1 − π) 1 − n q (46)
θL

The variance of consumption is given by

£ ¤2 £ ¤2
Vb (Q) = π QH − Q̄n + (1 − π) QL − Q̄n
∙ ∙ ¸ ¸
¡ ¢ H L
H θ
¡ H θ
¢ H ¡ L
¢ L 2
= 1−n q − π 1−n + (1 − π) 1 − n q
θL θL
∙ ∙ ¸ ¸
¡ L
¢ L ¡ ¢ H
H θ
¡ L
¢ L 2
+ (1 − π) 1 − n q − π 1 − n + (1 − π) 1 − n q
θL
∙ ¸
¡ L
¢ ¡ ¢ H ¡ L ¢2
H θ
= 2π (1 − π) 1 − n − 1 − n q
θL

Using (46) the coefficient of variation is

p " ¡ ¢ ¡ ¢ H #
Vb (Q) p 1 − nL − 1 − nH θθL
ρb ≡ = 2π (1 − π) H (47)
Q̄b π (1 − nH ) θ L + (1 − π) (1 − nL )
θ

It is clear that the numerator of (47) is smaller than the numerator in (45) while the denominator of
(47) is larger than the denominator in (45). Thus, ρb < ρn . In this sense, the bank make aggregate
consumption less disperse than when there are no banks.

34
 Ramsey Problem: Suppose that there is a government that must finance a given level of gov-
ernment spending per capita in market 2 via the use of money creation only. So the government gives
the agents cash and requires them to buy goods and then turn them over. So g = φ (γ − 1) M > 0
which requires γ > 1.
In market two the agents’ problem is

W (m2 , l, d) = max [U (x) − h + βEV+1 (m1,+1 )] (48)

x,h,m1,+1

s.t. x + φm1,+1 + g = h + φm2 + φ (1 + id ) d − φ (1 + i) l + φ (γ − 1) M

Rewriting the budget constraint yields

W (m2 , l, d) = φm2 + φ (γ − 1) M + φ (1 + id ) d − φ (1 + i) l − g

+ max [U (x) − x − φm1,+1 + βEV+1 (m1,+1 )]

x,m1,+1

The first-order conditions are the same as in (6). Furthermore, the agents problem in market 1
is unaffected by these actions. Consequently, (37) is the constraint for a competitive equilibrium.
Also welfare is then given by
∙ ¸ ∙ ¸
¡ H
¢ ¡ H ¢ qH ¡ L
¢ ¡ L¢ qL
(1 − β) EV (M ) = U (x)−x−g+π 1 − n u q − +(1 − π) 1 − n u q − (49)
a a

We consider high enough levels of g and thus γ such that buyers are always liquidity constrained
in any state. This reduces (37) to

γ−β ¡ ¢ £ ¡ ¢ ¤ ¡ ¢ £ ¡ ¢ ¤
= π 1 − nH θH au0 q H − 1 + (1 − π) 1 − nL θL au0 q L − 1 (50)
β

Thus, the Ramsey problem is to choose x, q H and q L to maximize (49) subject to (50). It is
obvious that x = x∗ so the remaining problem is

∙ ¸ ∙ ¸
¡ ¡ H ¢ qH
H
¢ ¡ L
¢ ¡ L¢ qL
max π 1 − n u q − + (1 − π) 1 − n u q −
q H ,q L a a
γ−β ¡ ¢ £ ¡ ¢ ¤ ¡ ¢ £ ¡ ¢ ¤
s.t. = π 1 − nH θH au0 q H − 1 + (1 − π) 1 − nL θL au0 qL − 1
β

which is what we have in the text.

35
Proof of Proposition 3: Since this is a well-defined maximization problem over a compact set
¡ ¢
for q ∈ [0, q ∗ ], there is a unique pair q L , q H in the interval [0, q ∗ ]2 that solves it.
First consider the problem when there are no banks. In this case we can simply set µ = 1 which
makes θH = θL = 1. It is clear from (27) that the planner chooses qH = q L .
Now consider the problem with banks. The first-order conditions must satisfy (27). Let qL =
0
θL
q H < q ∗ . Then the left-hand side of (27) is one which is larger than θH
. Let f (q) = auu00(q)−1
(q) . We
have f (q) < 0 and f 0 (q) > 0 because au00 (q) u00 (q) − [au0 (q) − 1] u000 (q) > 0. Fix q H then the
left-hand side is monotonically decreasing to zero in q L between [0, q∗ ]. Thus there is a unique pair
q ∗ > q L > q H such that (27) is satisfied.¥

36
Proof of Proposition 4: Taking the expectation of (19) with respect to the marginal utility and
using (6) yields

φ−1 £ ¡ ¢ ¡ ¢¤ £ ¡ ¢ ¡ ¢¤
= πφH (1 − n) aεH u0 q H + n 1 + iH d +(1 − π) φL (1 − n) aεL u0 qL + n 1 + iL
d . (51)
β
³ ´ ³ ´
n n
where qbH = 1−n qsH = q H and qbL = 1−n qsL = q L .
In a steady state equilibrium the real value of money is constant so φH M = φL M = φ−1 M−1 =
Ω. Thus, using (3) and (17) we can write (51) as

γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
= (1 − n) θ π aεH u0 q H − 1 + (1 − π) aεL u0 q L − 1 (52)
β
1
In any equilibrium pq j ≤ z j θM , j = H, L. Then, the seller’s first-order condition p = aφ implies

qj ≤ az j θΩ
¡ ¢
The efficient quantity q j∗ satisfies aεj u0 q j∗ = 1. Thus, if trades are efficient we have qj = qj∗ ≤
az j θΩ which requires that
q j∗
≤Ω (53)
aθz j
Consequently, since - as we show below - Ω is strictly decreasing in γ, for a given distribution of
shocks, the monetary equilibrium can be characterized by the rate of inflation. First, in the low
inflation economy we have q H = qH∗ and q L < qL∗ . In the high inflation economy we have q L < q L∗
and q H < q H∗ .

Low Inflation Economy: For the conjectured low inflation economy,

q H = az H θΩ < q H∗ and q L = q L∗ < az L θΩ.

where qH∗ > qL∗ .

Using these expressions we obtain

q L∗ q H∗
L
<Ω< (54)
aθz aθz H
qH∗ zH
which is a non-empty interval if q L∗
> zL
.
¡ H¢
In the low inflation economy, q L = q L∗ . This implies that iL = iL H H 0
d = 0 and i = aε u q − 1.

37
 Then, (52) can be solved to obtain a single equation in Ω

γ−β £ ¡ ¢ ¤
= (1 − n) θ aεH u0 az H θΩ − 1
β

or a single equation in iH
γ−β
iH =
β (1 − nµ)
It is clear from this expression that iH is greater than zero for γ > β and approaches zero as γ → β.
Furthermore, we have qH → q∗ as γ → β. Given the solution for iH and qH we can solve for iH
d and

Ω. The right-hand side of (61) is monotonically increasing in γ. Thus, an increase in the money
growth rate, raises the nominal borrowing rate, increasing the cost of consumption and reducing
q H . Since q H decreases in γ, Ω decreases as well.
qH∗
Given this behavior on iH it is straightforward to show that as γ → β, Ω → aθz H
so that the
right-hand inequality is binding in (??). Thus, as γ increases from the Friedman rule, Ω declines
and satisfies (??). For a sufficiently high value of γ, the left-hand inequality is violated. Thus,
qL∗
b that solves
there exists a critical value of γ aθz L
b] a unique monetary
= Ω. Hence, for γ ∈ [β, γ
equilibrium exists such that (??) holds.

High inflation economy: For the conjectured high inflation economy, we have

q H = az H θΩ < q H∗ and q L = az L θΩ < qL∗ . (55)

In the high inflation economy, q L < q L∗ and q H < q H∗ . This implies from (17) and (3) that

¡ ¢ 1 j
aεj u0 qj = 1 + ij = 1 + i j = L, H
1−µ d

Using this expression we can obtain a single expression in Ω

γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
= (1 − n) θ π aεH u0 az H θΩ − 1 + (1 − π) aεL u0 az L θΩ − 1
β
b −β
γ q L∗
The right-hand side expression is monotonically decreasing in Ω and is equal to β at Ω = aθz L
.
b, a unique value of Ω exists such that (55) holds.¥
Thus for γ > γ

38
Proof of Proposition 5: Since this is a well-defined maximization problem over a compact set
¡ ¢
for q ∈ [0, q ∗ ], there is a unique pair qL , qH in the interval [0, q ∗ ]2 that solves it. The first-order
εu0 (q)−1
conditions must satisfy (29). Define f (q, ε) = εu00 (q) . We have f (q, ε) < 0 and fq (q, ε) > 0 because
au00 (q) u00 (q) − [au0 (q) − 1] u000 (q) > 0. Moreover, we have f (q ∗ , ε) = 0. Note that

εL u0 (q) − 1 εH u0 (q) − 1
f (q, εL ) − f (q, εH ) = − >0
εL u00 (q) εH u00 (q)
¡ ¢
This is true for all q ∈ 0, qL∗ since εH > εL . This implies that 0 > f (q, εL ) > f (q, εH ) which
from (29) implies that q H > q L .¥

39
Proof of Proposition 6: Taking the expectation of (19) with respect to the real shocks and using
(6) yields

φ−1 £ ¡ ¢ ¡ ¢¤ £ ¡ ¢ ¡ ¢¤
= πφH (1 − n) aH u0 qH + n 1 + iH d + (1 − π) φL (1 − n) aL u0 q L + n 1 + iL
d . (56)
β
³ ´ ³ ´
n n
where qbH = 1−n qsH = q H and qbL = 1−n qsL = q L .
In a steady state equilibrium the real value of money is constant so φH M = φL M = φ−1 M−1 =
Ω. Thus, using (3) and (17) we can write (56) as

γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
= (1 − n) θ π aH u0 q H − 1 + (1 − π) aL u0 q L − 1 (57)
β
1
In any equilibrium pj q j ≤ z j θM , j = H, L. Then, the seller’s first-order condition pj = aj φ

implies
q j ≤ aj z j θΩ
¡ ¢
The efficient quantity q j∗ satisfies aj u0 q j∗ = 1. Thus, if trades are efficient we have qj = q j∗ ≤
aj z j θΩ which requires that ¡ ¢
qj∗ q j∗ u0 qj∗
= ≤ θΩ. (58)
aj z j zj
Consequently, since - as we show below - θΩ is strictly decreasing in γ, for a given distribution of
productivity shocks, the monetary equilibrium can be characterized by the rate of inflation. First,
in the low inflation economy we have qH < qH∗ and qL = qL∗ . In the high inflation economy we
have q L < q L∗ and q H < qH∗ .

Low Inflation Economy: For the conjectured low inflation economy,

q H = aH z H θΩ < q H∗ and q L = qL∗ < aL z L θΩ.

where qH∗ > qL∗ .

Using these expressions we obtain
¡ ¢ ¡ ¢
q L∗ u0 q L∗ qL∗ q H∗ qH∗ u0 qH∗
= L L < θΩ < H H = (59)
zL a z a z zH

which is a non-empty interval if (30) holds.

40
 In the low inflation economy, q L = q∗ . This implies from (17) that iL = iL
d = 0 and

¡ ¢
aH u0 q H − 1 = iH . (60)

Then, (57) can be solved to obtain a single equation in Ω

γ−β £ ¡ ¢ ¤
= (1 − n) θ aH u0 aH z H θΩ − 1
β

or a single equation in iL
γ−β
iH = (61)
β (1 − n) θ
It is clear that the value of iH solving (61) is greater than zero for γ > β and approaches zero as
γ → β. Furthermore, we have q H → q ∗ as γ → β. Given the solution for iH and q H we can solve
for iH
d and Ω. The right-hand side of (61) is monotonically increasing in γ. Thus, an increase in

the money growth rate, raises the nominal borrowing rate, increasing the cost of consumption and
reducing qH . Since qH decreases in γ, Ω decreases as well.
q H∗
Given this behavior on iH it is straightforward to show that as γ → β, θΩ → aH z H
so that the
right-hand inequality is binding in (59). Thus, as γ increases from the Friedman rule, θΩ declines
and satisfies (59). For a sufficiently high value of γ, the left-hand inequality is violated. Thus,
q L∗
e that solves
there exists a critical value of γ aL z L
e] a unique monetary
= θΩ. Hence, for γ ∈ [β, γ
equilibrium exists such that (59) holds.

High inflation economy: For the conjectured high inflation economy, we have

q H = aH z H θΩ < q H∗ and q L = aL z L θΩ < q L∗ . (62)

In the high inflation economy, q L < q L∗ and q H < q H∗ . This implies from (17) and (3) that

¡ ¢ 1 j
aj u0 q j = 1 + ij = 1 + i j = L, H (63)
1−µ d
aL z L H
Moreover, since qL = aH z H
q (57) can be solved to obtain a single equation in qH

½ ∙ µ L L ¶ ¸¾
γ−β £ H 0 ¡ H¢ ¤ L 0 a z H
= (1 − n) θ π a u q − 1 + (1 − π) a u q −1
β aH z H

41
Substitute to get

γ−β © £ ¡ ¢ ¤ £ ¡ ¢ ¤ª
= (1 − n) θ π aH u0 aH z H θΩ − 1 + (1 − π) aL u0 aL z L θΩ − 1
β

which is one equation in one unknown, Ω. The right-hand side expression is monotonically decreas-
e −β
γ q L∗
ing in Ω and is equal to β at θΩ = aL z L
. e, a unique value of Ω exists such that
Thus for γ > γ
(62) holds.¥

42
Proof of Proposition 7: Since this is a well-defined maximization problem over a compact set
¡ ¢
for q ∈ [0, q ∗ ], there is a unique pair qL , qH in the interval [0, q ∗ ]2 that solves it. The first-order
au0 (q)−1
conditions must satisfy (32). Define f (q) = a2 u00 (q)
. We have f (q) < 0 and f 0 (q) > 0 because
au00 (q) u00 (q)−[au0 (q) − 1] u000 (q) > 0. Moreover, we have f (q∗ ) = 0. Conjecture q L = q H = q < q ∗ .
This implies from (32) that
1 1
u0 (q) = H
+ L.
a a
Since u0 (q) is a monotonic decreasing function there exists a unique q < q ∗ that satisfies the
equation above. Using this and (28) we obtain
∙ H ¸
γ−β a aL
= (1 − n) θ π L + (1 − π) H
β a a

Solving for γ yields the critical value

∙ H ¸
c a aL
γ = β + β (1 − n) θ π L + (1 − π) H
a a

such that f (q L ) = f (q H ) iff γ = γ c holds. For γ < γ c we have 0 > f (qL ) > f (q H ) which from (32)
implies that q H > q L and for γ > γ c qH < q L .¥

43